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> |
57c8a661 | 34 | #include <linux/memblock.h> |
61a58c6c | 35 | #include <linux/pid_namespace.h> |
820e45db | 36 | #include <linux/init_task.h> |
3eb07c8c | 37 | #include <linux/syscalls.h> |
0bb80f24 | 38 | #include <linux/proc_ns.h> |
0a01f2cc | 39 | #include <linux/proc_fs.h> |
29930025 | 40 | #include <linux/sched/task.h> |
95846ecf | 41 | #include <linux/idr.h> |
1da177e4 | 42 | |
e1e871af DH |
43 | struct pid init_struct_pid = { |
44 | .count = ATOMIC_INIT(1), | |
45 | .tasks = { | |
46 | { .first = NULL }, | |
47 | { .first = NULL }, | |
48 | { .first = NULL }, | |
49 | }, | |
50 | .level = 0, | |
51 | .numbers = { { | |
52 | .nr = 0, | |
53 | .ns = &init_pid_ns, | |
54 | }, } | |
55 | }; | |
1da177e4 LT |
56 | |
57 | int pid_max = PID_MAX_DEFAULT; | |
1da177e4 LT |
58 | |
59 | #define RESERVED_PIDS 300 | |
60 | ||
61 | int pid_max_min = RESERVED_PIDS + 1; | |
62 | int pid_max_max = PID_MAX_LIMIT; | |
63 | ||
1da177e4 LT |
64 | /* |
65 | * PID-map pages start out as NULL, they get allocated upon | |
66 | * first use and are never deallocated. This way a low pid_max | |
67 | * value does not cause lots of bitmaps to be allocated, but | |
68 | * the scheme scales to up to 4 million PIDs, runtime. | |
69 | */ | |
61a58c6c | 70 | struct pid_namespace init_pid_ns = { |
1e24edca | 71 | .kref = KREF_INIT(2), |
f6bb2a2c | 72 | .idr = IDR_INIT(init_pid_ns.idr), |
e8cfbc24 | 73 | .pid_allocated = PIDNS_ADDING, |
faacbfd3 PE |
74 | .level = 0, |
75 | .child_reaper = &init_task, | |
49f4d8b9 | 76 | .user_ns = &init_user_ns, |
435d5f4b | 77 | .ns.inum = PROC_PID_INIT_INO, |
33c42940 AV |
78 | #ifdef CONFIG_PID_NS |
79 | .ns.ops = &pidns_operations, | |
80 | #endif | |
3fbc9648 | 81 | }; |
198fe21b | 82 | EXPORT_SYMBOL_GPL(init_pid_ns); |
1da177e4 | 83 | |
92476d7f EB |
84 | /* |
85 | * Note: disable interrupts while the pidmap_lock is held as an | |
86 | * interrupt might come in and do read_lock(&tasklist_lock). | |
87 | * | |
88 | * If we don't disable interrupts there is a nasty deadlock between | |
89 | * detach_pid()->free_pid() and another cpu that does | |
90 | * spin_lock(&pidmap_lock) followed by an interrupt routine that does | |
91 | * read_lock(&tasklist_lock); | |
92 | * | |
93 | * After we clean up the tasklist_lock and know there are no | |
94 | * irq handlers that take it we can leave the interrupts enabled. | |
95 | * For now it is easier to be safe than to prove it can't happen. | |
96 | */ | |
3fbc9648 | 97 | |
1da177e4 LT |
98 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock); |
99 | ||
7ad5b3a5 | 100 | void put_pid(struct pid *pid) |
92476d7f | 101 | { |
baf8f0f8 PE |
102 | struct pid_namespace *ns; |
103 | ||
92476d7f EB |
104 | if (!pid) |
105 | return; | |
baf8f0f8 | 106 | |
8ef047aa | 107 | ns = pid->numbers[pid->level].ns; |
92476d7f | 108 | if ((atomic_read(&pid->count) == 1) || |
8ef047aa | 109 | atomic_dec_and_test(&pid->count)) { |
baf8f0f8 | 110 | kmem_cache_free(ns->pid_cachep, pid); |
b461cc03 | 111 | put_pid_ns(ns); |
8ef047aa | 112 | } |
92476d7f | 113 | } |
bbf73147 | 114 | EXPORT_SYMBOL_GPL(put_pid); |
92476d7f EB |
115 | |
116 | static void delayed_put_pid(struct rcu_head *rhp) | |
117 | { | |
118 | struct pid *pid = container_of(rhp, struct pid, rcu); | |
119 | put_pid(pid); | |
120 | } | |
121 | ||
7ad5b3a5 | 122 | void free_pid(struct pid *pid) |
92476d7f EB |
123 | { |
124 | /* We can be called with write_lock_irq(&tasklist_lock) held */ | |
8ef047aa | 125 | int i; |
92476d7f EB |
126 | unsigned long flags; |
127 | ||
128 | spin_lock_irqsave(&pidmap_lock, flags); | |
0a01f2cc EB |
129 | for (i = 0; i <= pid->level; i++) { |
130 | struct upid *upid = pid->numbers + i; | |
af4b8a83 | 131 | struct pid_namespace *ns = upid->ns; |
e8cfbc24 | 132 | switch (--ns->pid_allocated) { |
a6064885 | 133 | case 2: |
af4b8a83 EB |
134 | case 1: |
135 | /* When all that is left in the pid namespace | |
136 | * is the reaper wake up the reaper. The reaper | |
137 | * may be sleeping in zap_pid_ns_processes(). | |
138 | */ | |
139 | wake_up_process(ns->child_reaper); | |
140 | break; | |
e8cfbc24 | 141 | case PIDNS_ADDING: |
314a8ad0 ON |
142 | /* Handle a fork failure of the first process */ |
143 | WARN_ON(ns->child_reaper); | |
e8cfbc24 | 144 | ns->pid_allocated = 0; |
314a8ad0 | 145 | /* fall through */ |
af4b8a83 | 146 | case 0: |
af4b8a83 EB |
147 | schedule_work(&ns->proc_work); |
148 | break; | |
5e1182de | 149 | } |
95846ecf GS |
150 | |
151 | idr_remove(&ns->idr, upid->nr); | |
0a01f2cc | 152 | } |
92476d7f EB |
153 | spin_unlock_irqrestore(&pidmap_lock, flags); |
154 | ||
92476d7f EB |
155 | call_rcu(&pid->rcu, delayed_put_pid); |
156 | } | |
157 | ||
8ef047aa | 158 | struct pid *alloc_pid(struct pid_namespace *ns) |
92476d7f EB |
159 | { |
160 | struct pid *pid; | |
161 | enum pid_type type; | |
8ef047aa PE |
162 | int i, nr; |
163 | struct pid_namespace *tmp; | |
198fe21b | 164 | struct upid *upid; |
35f71bc0 | 165 | int retval = -ENOMEM; |
92476d7f | 166 | |
baf8f0f8 | 167 | pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL); |
92476d7f | 168 | if (!pid) |
35f71bc0 | 169 | return ERR_PTR(retval); |
92476d7f | 170 | |
8ef047aa | 171 | tmp = ns; |
0a01f2cc | 172 | pid->level = ns->level; |
95846ecf | 173 | |
8ef047aa | 174 | for (i = ns->level; i >= 0; i--) { |
95846ecf GS |
175 | int pid_min = 1; |
176 | ||
177 | idr_preload(GFP_KERNEL); | |
178 | spin_lock_irq(&pidmap_lock); | |
179 | ||
180 | /* | |
181 | * init really needs pid 1, but after reaching the maximum | |
182 | * wrap back to RESERVED_PIDS | |
183 | */ | |
184 | if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS) | |
185 | pid_min = RESERVED_PIDS; | |
186 | ||
187 | /* | |
188 | * Store a null pointer so find_pid_ns does not find | |
189 | * a partially initialized PID (see below). | |
190 | */ | |
191 | nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min, | |
192 | pid_max, GFP_ATOMIC); | |
193 | spin_unlock_irq(&pidmap_lock); | |
194 | idr_preload_end(); | |
195 | ||
287980e4 | 196 | if (nr < 0) { |
f83606f5 | 197 | retval = (nr == -ENOSPC) ? -EAGAIN : nr; |
8ef047aa | 198 | goto out_free; |
35f71bc0 | 199 | } |
92476d7f | 200 | |
8ef047aa PE |
201 | pid->numbers[i].nr = nr; |
202 | pid->numbers[i].ns = tmp; | |
203 | tmp = tmp->parent; | |
204 | } | |
205 | ||
0a01f2cc | 206 | if (unlikely(is_child_reaper(pid))) { |
c0ee5549 | 207 | if (pid_ns_prepare_proc(ns)) |
0a01f2cc EB |
208 | goto out_free; |
209 | } | |
210 | ||
b461cc03 | 211 | get_pid_ns(ns); |
92476d7f | 212 | atomic_set(&pid->count, 1); |
92476d7f EB |
213 | for (type = 0; type < PIDTYPE_MAX; ++type) |
214 | INIT_HLIST_HEAD(&pid->tasks[type]); | |
215 | ||
417e3152 | 216 | upid = pid->numbers + ns->level; |
92476d7f | 217 | spin_lock_irq(&pidmap_lock); |
e8cfbc24 | 218 | if (!(ns->pid_allocated & PIDNS_ADDING)) |
5e1182de | 219 | goto out_unlock; |
0a01f2cc | 220 | for ( ; upid >= pid->numbers; --upid) { |
95846ecf GS |
221 | /* Make the PID visible to find_pid_ns. */ |
222 | idr_replace(&upid->ns->idr, pid, upid->nr); | |
e8cfbc24 | 223 | upid->ns->pid_allocated++; |
0a01f2cc | 224 | } |
92476d7f EB |
225 | spin_unlock_irq(&pidmap_lock); |
226 | ||
92476d7f EB |
227 | return pid; |
228 | ||
5e1182de | 229 | out_unlock: |
6e666884 | 230 | spin_unlock_irq(&pidmap_lock); |
24c037eb ON |
231 | put_pid_ns(ns); |
232 | ||
92476d7f | 233 | out_free: |
95846ecf | 234 | spin_lock_irq(&pidmap_lock); |
1a80dade MW |
235 | while (++i <= ns->level) { |
236 | upid = pid->numbers + i; | |
237 | idr_remove(&upid->ns->idr, upid->nr); | |
238 | } | |
95846ecf | 239 | |
c0ee5549 EB |
240 | /* On failure to allocate the first pid, reset the state */ |
241 | if (ns->pid_allocated == PIDNS_ADDING) | |
242 | idr_set_cursor(&ns->idr, 0); | |
243 | ||
95846ecf | 244 | spin_unlock_irq(&pidmap_lock); |
8ef047aa | 245 | |
baf8f0f8 | 246 | kmem_cache_free(ns->pid_cachep, pid); |
35f71bc0 | 247 | return ERR_PTR(retval); |
92476d7f EB |
248 | } |
249 | ||
c876ad76 EB |
250 | void disable_pid_allocation(struct pid_namespace *ns) |
251 | { | |
252 | spin_lock_irq(&pidmap_lock); | |
e8cfbc24 | 253 | ns->pid_allocated &= ~PIDNS_ADDING; |
c876ad76 EB |
254 | spin_unlock_irq(&pidmap_lock); |
255 | } | |
256 | ||
7ad5b3a5 | 257 | struct pid *find_pid_ns(int nr, struct pid_namespace *ns) |
1da177e4 | 258 | { |
e8cfbc24 | 259 | return idr_find(&ns->idr, nr); |
1da177e4 | 260 | } |
198fe21b | 261 | EXPORT_SYMBOL_GPL(find_pid_ns); |
1da177e4 | 262 | |
8990571e PE |
263 | struct pid *find_vpid(int nr) |
264 | { | |
17cf22c3 | 265 | return find_pid_ns(nr, task_active_pid_ns(current)); |
8990571e PE |
266 | } |
267 | EXPORT_SYMBOL_GPL(find_vpid); | |
268 | ||
2c470475 EB |
269 | static struct pid **task_pid_ptr(struct task_struct *task, enum pid_type type) |
270 | { | |
271 | return (type == PIDTYPE_PID) ? | |
272 | &task->thread_pid : | |
2c470475 EB |
273 | &task->signal->pids[type]; |
274 | } | |
275 | ||
e713d0da SB |
276 | /* |
277 | * attach_pid() must be called with the tasklist_lock write-held. | |
278 | */ | |
81907739 | 279 | void attach_pid(struct task_struct *task, enum pid_type type) |
1da177e4 | 280 | { |
2c470475 EB |
281 | struct pid *pid = *task_pid_ptr(task, type); |
282 | hlist_add_head_rcu(&task->pid_links[type], &pid->tasks[type]); | |
1da177e4 LT |
283 | } |
284 | ||
24336eae ON |
285 | static void __change_pid(struct task_struct *task, enum pid_type type, |
286 | struct pid *new) | |
1da177e4 | 287 | { |
2c470475 | 288 | struct pid **pid_ptr = task_pid_ptr(task, type); |
92476d7f EB |
289 | struct pid *pid; |
290 | int tmp; | |
1da177e4 | 291 | |
2c470475 | 292 | pid = *pid_ptr; |
1da177e4 | 293 | |
2c470475 EB |
294 | hlist_del_rcu(&task->pid_links[type]); |
295 | *pid_ptr = new; | |
1da177e4 | 296 | |
92476d7f EB |
297 | for (tmp = PIDTYPE_MAX; --tmp >= 0; ) |
298 | if (!hlist_empty(&pid->tasks[tmp])) | |
299 | return; | |
1da177e4 | 300 | |
92476d7f | 301 | free_pid(pid); |
1da177e4 LT |
302 | } |
303 | ||
24336eae ON |
304 | void detach_pid(struct task_struct *task, enum pid_type type) |
305 | { | |
306 | __change_pid(task, type, NULL); | |
307 | } | |
308 | ||
309 | void change_pid(struct task_struct *task, enum pid_type type, | |
310 | struct pid *pid) | |
311 | { | |
312 | __change_pid(task, type, pid); | |
81907739 | 313 | attach_pid(task, type); |
24336eae ON |
314 | } |
315 | ||
c18258c6 | 316 | /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */ |
7ad5b3a5 | 317 | void transfer_pid(struct task_struct *old, struct task_struct *new, |
c18258c6 EB |
318 | enum pid_type type) |
319 | { | |
2c470475 EB |
320 | if (type == PIDTYPE_PID) |
321 | new->thread_pid = old->thread_pid; | |
322 | hlist_replace_rcu(&old->pid_links[type], &new->pid_links[type]); | |
c18258c6 EB |
323 | } |
324 | ||
7ad5b3a5 | 325 | struct task_struct *pid_task(struct pid *pid, enum pid_type type) |
1da177e4 | 326 | { |
92476d7f EB |
327 | struct task_struct *result = NULL; |
328 | if (pid) { | |
329 | struct hlist_node *first; | |
67bdbffd | 330 | first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]), |
db1466b3 | 331 | lockdep_tasklist_lock_is_held()); |
92476d7f | 332 | if (first) |
2c470475 | 333 | result = hlist_entry(first, struct task_struct, pid_links[(type)]); |
92476d7f EB |
334 | } |
335 | return result; | |
336 | } | |
eccba068 | 337 | EXPORT_SYMBOL(pid_task); |
1da177e4 | 338 | |
92476d7f | 339 | /* |
9728e5d6 | 340 | * Must be called under rcu_read_lock(). |
92476d7f | 341 | */ |
17f98dcf | 342 | struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns) |
92476d7f | 343 | { |
f78f5b90 PM |
344 | RCU_LOCKDEP_WARN(!rcu_read_lock_held(), |
345 | "find_task_by_pid_ns() needs rcu_read_lock() protection"); | |
17f98dcf | 346 | return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID); |
92476d7f | 347 | } |
1da177e4 | 348 | |
228ebcbe PE |
349 | struct task_struct *find_task_by_vpid(pid_t vnr) |
350 | { | |
17cf22c3 | 351 | return find_task_by_pid_ns(vnr, task_active_pid_ns(current)); |
228ebcbe | 352 | } |
228ebcbe | 353 | |
2ee08260 MR |
354 | struct task_struct *find_get_task_by_vpid(pid_t nr) |
355 | { | |
356 | struct task_struct *task; | |
357 | ||
358 | rcu_read_lock(); | |
359 | task = find_task_by_vpid(nr); | |
360 | if (task) | |
361 | get_task_struct(task); | |
362 | rcu_read_unlock(); | |
363 | ||
364 | return task; | |
365 | } | |
366 | ||
1a657f78 ON |
367 | struct pid *get_task_pid(struct task_struct *task, enum pid_type type) |
368 | { | |
369 | struct pid *pid; | |
370 | rcu_read_lock(); | |
2c470475 | 371 | pid = get_pid(rcu_dereference(*task_pid_ptr(task, type))); |
1a657f78 ON |
372 | rcu_read_unlock(); |
373 | return pid; | |
374 | } | |
77c100c8 | 375 | EXPORT_SYMBOL_GPL(get_task_pid); |
1a657f78 | 376 | |
7ad5b3a5 | 377 | struct task_struct *get_pid_task(struct pid *pid, enum pid_type type) |
92476d7f EB |
378 | { |
379 | struct task_struct *result; | |
380 | rcu_read_lock(); | |
381 | result = pid_task(pid, type); | |
382 | if (result) | |
383 | get_task_struct(result); | |
384 | rcu_read_unlock(); | |
385 | return result; | |
1da177e4 | 386 | } |
77c100c8 | 387 | EXPORT_SYMBOL_GPL(get_pid_task); |
1da177e4 | 388 | |
92476d7f | 389 | struct pid *find_get_pid(pid_t nr) |
1da177e4 LT |
390 | { |
391 | struct pid *pid; | |
392 | ||
92476d7f | 393 | rcu_read_lock(); |
198fe21b | 394 | pid = get_pid(find_vpid(nr)); |
92476d7f | 395 | rcu_read_unlock(); |
1da177e4 | 396 | |
92476d7f | 397 | return pid; |
1da177e4 | 398 | } |
339caf2a | 399 | EXPORT_SYMBOL_GPL(find_get_pid); |
1da177e4 | 400 | |
7af57294 PE |
401 | pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns) |
402 | { | |
403 | struct upid *upid; | |
404 | pid_t nr = 0; | |
405 | ||
406 | if (pid && ns->level <= pid->level) { | |
407 | upid = &pid->numbers[ns->level]; | |
408 | if (upid->ns == ns) | |
409 | nr = upid->nr; | |
410 | } | |
411 | return nr; | |
412 | } | |
4f82f457 | 413 | EXPORT_SYMBOL_GPL(pid_nr_ns); |
7af57294 | 414 | |
44c4e1b2 EB |
415 | pid_t pid_vnr(struct pid *pid) |
416 | { | |
17cf22c3 | 417 | return pid_nr_ns(pid, task_active_pid_ns(current)); |
44c4e1b2 EB |
418 | } |
419 | EXPORT_SYMBOL_GPL(pid_vnr); | |
420 | ||
52ee2dfd ON |
421 | pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type, |
422 | struct pid_namespace *ns) | |
2f2a3a46 | 423 | { |
52ee2dfd ON |
424 | pid_t nr = 0; |
425 | ||
426 | rcu_read_lock(); | |
427 | if (!ns) | |
17cf22c3 | 428 | ns = task_active_pid_ns(current); |
2c470475 EB |
429 | if (likely(pid_alive(task))) |
430 | nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns); | |
52ee2dfd ON |
431 | rcu_read_unlock(); |
432 | ||
433 | return nr; | |
2f2a3a46 | 434 | } |
52ee2dfd | 435 | EXPORT_SYMBOL(__task_pid_nr_ns); |
2f2a3a46 | 436 | |
61bce0f1 EB |
437 | struct pid_namespace *task_active_pid_ns(struct task_struct *tsk) |
438 | { | |
439 | return ns_of_pid(task_pid(tsk)); | |
440 | } | |
441 | EXPORT_SYMBOL_GPL(task_active_pid_ns); | |
442 | ||
0804ef4b | 443 | /* |
025dfdaf | 444 | * Used by proc to find the first pid that is greater than or equal to nr. |
0804ef4b | 445 | * |
e49859e7 | 446 | * If there is a pid at nr this function is exactly the same as find_pid_ns. |
0804ef4b | 447 | */ |
198fe21b | 448 | struct pid *find_ge_pid(int nr, struct pid_namespace *ns) |
0804ef4b | 449 | { |
95846ecf | 450 | return idr_get_next(&ns->idr, &nr); |
0804ef4b EB |
451 | } |
452 | ||
95846ecf | 453 | void __init pid_idr_init(void) |
1da177e4 | 454 | { |
840d6fe7 | 455 | /* Verify no one has done anything silly: */ |
e8cfbc24 | 456 | BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING); |
c876ad76 | 457 | |
72680a19 HB |
458 | /* bump default and minimum pid_max based on number of cpus */ |
459 | pid_max = min(pid_max_max, max_t(int, pid_max, | |
460 | PIDS_PER_CPU_DEFAULT * num_possible_cpus())); | |
461 | pid_max_min = max_t(int, pid_max_min, | |
462 | PIDS_PER_CPU_MIN * num_possible_cpus()); | |
463 | pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min); | |
464 | ||
95846ecf | 465 | idr_init(&init_pid_ns.idr); |
92476d7f | 466 | |
74bd59bb | 467 | init_pid_ns.pid_cachep = KMEM_CACHE(pid, |
5d097056 | 468 | SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT); |
1da177e4 | 469 | } |