Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
86db1e29 JA |
2 | /* |
3 | * Functions related to io context handling | |
4 | */ | |
5 | #include <linux/kernel.h> | |
6 | #include <linux/module.h> | |
7 | #include <linux/init.h> | |
8 | #include <linux/bio.h> | |
9 | #include <linux/blkdev.h> | |
5a0e3ad6 | 10 | #include <linux/slab.h> |
a411cd3c | 11 | #include <linux/security.h> |
f719ff9b | 12 | #include <linux/sched/task.h> |
86db1e29 JA |
13 | |
14 | #include "blk.h" | |
2aa7745b | 15 | #include "blk-mq-sched.h" |
86db1e29 JA |
16 | |
17 | /* | |
18 | * For io context allocations | |
19 | */ | |
20 | static struct kmem_cache *iocontext_cachep; | |
21 | ||
5ef16305 | 22 | #ifdef CONFIG_BLK_ICQ |
6e736be7 TH |
23 | /** |
24 | * get_io_context - increment reference count to io_context | |
25 | * @ioc: io_context to get | |
26 | * | |
27 | * Increment reference count to @ioc. | |
28 | */ | |
87dd1d63 | 29 | static void get_io_context(struct io_context *ioc) |
6e736be7 TH |
30 | { |
31 | BUG_ON(atomic_long_read(&ioc->refcount) <= 0); | |
32 | atomic_long_inc(&ioc->refcount); | |
33 | } | |
6e736be7 | 34 | |
7e5a8794 TH |
35 | static void icq_free_icq_rcu(struct rcu_head *head) |
36 | { | |
37 | struct io_cq *icq = container_of(head, struct io_cq, __rcu_head); | |
38 | ||
39 | kmem_cache_free(icq->__rcu_icq_cache, icq); | |
40 | } | |
41 | ||
3d492c2e | 42 | /* |
7b36a718 JA |
43 | * Exit an icq. Called with ioc locked for blk-mq, and with both ioc |
44 | * and queue locked for legacy. | |
3d492c2e | 45 | */ |
7e5a8794 | 46 | static void ioc_exit_icq(struct io_cq *icq) |
621032ad TH |
47 | { |
48 | struct elevator_type *et = icq->q->elevator->type; | |
49 | ||
50 | if (icq->flags & ICQ_EXITED) | |
51 | return; | |
52 | ||
f9cd4bfe JA |
53 | if (et->ops.exit_icq) |
54 | et->ops.exit_icq(icq); | |
621032ad TH |
55 | |
56 | icq->flags |= ICQ_EXITED; | |
57 | } | |
58 | ||
4be8a2ea CH |
59 | static void ioc_exit_icqs(struct io_context *ioc) |
60 | { | |
61 | struct io_cq *icq; | |
62 | ||
63 | spin_lock_irq(&ioc->lock); | |
64 | hlist_for_each_entry(icq, &ioc->icq_list, ioc_node) | |
65 | ioc_exit_icq(icq); | |
66 | spin_unlock_irq(&ioc->lock); | |
67 | } | |
68 | ||
7b36a718 JA |
69 | /* |
70 | * Release an icq. Called with ioc locked for blk-mq, and with both ioc | |
71 | * and queue locked for legacy. | |
72 | */ | |
621032ad | 73 | static void ioc_destroy_icq(struct io_cq *icq) |
7e5a8794 TH |
74 | { |
75 | struct io_context *ioc = icq->ioc; | |
76 | struct request_queue *q = icq->q; | |
77 | struct elevator_type *et = q->elevator->type; | |
78 | ||
79 | lockdep_assert_held(&ioc->lock); | |
7e5a8794 TH |
80 | |
81 | radix_tree_delete(&ioc->icq_tree, icq->q->id); | |
82 | hlist_del_init(&icq->ioc_node); | |
83 | list_del_init(&icq->q_node); | |
84 | ||
85 | /* | |
86 | * Both setting lookup hint to and clearing it from @icq are done | |
87 | * under queue_lock. If it's not pointing to @icq now, it never | |
88 | * will. Hint assignment itself can race safely. | |
89 | */ | |
ec6c676a | 90 | if (rcu_access_pointer(ioc->icq_hint) == icq) |
7e5a8794 TH |
91 | rcu_assign_pointer(ioc->icq_hint, NULL); |
92 | ||
621032ad | 93 | ioc_exit_icq(icq); |
7e5a8794 TH |
94 | |
95 | /* | |
96 | * @icq->q might have gone away by the time RCU callback runs | |
97 | * making it impossible to determine icq_cache. Record it in @icq. | |
98 | */ | |
99 | icq->__rcu_icq_cache = et->icq_cache; | |
30a2da7b | 100 | icq->flags |= ICQ_DESTROYED; |
7e5a8794 TH |
101 | call_rcu(&icq->__rcu_head, icq_free_icq_rcu); |
102 | } | |
103 | ||
b2efa052 TH |
104 | /* |
105 | * Slow path for ioc release in put_io_context(). Performs double-lock | |
c5869807 | 106 | * dancing to unlink all icq's and then frees ioc. |
b2efa052 TH |
107 | */ |
108 | static void ioc_release_fn(struct work_struct *work) | |
86db1e29 | 109 | { |
b2efa052 TH |
110 | struct io_context *ioc = container_of(work, struct io_context, |
111 | release_work); | |
a43f085f | 112 | spin_lock_irq(&ioc->lock); |
b2efa052 | 113 | |
c5869807 TH |
114 | while (!hlist_empty(&ioc->icq_list)) { |
115 | struct io_cq *icq = hlist_entry(ioc->icq_list.first, | |
116 | struct io_cq, ioc_node); | |
2274b029 TH |
117 | struct request_queue *q = icq->q; |
118 | ||
0d945c1f | 119 | if (spin_trylock(&q->queue_lock)) { |
621032ad | 120 | ioc_destroy_icq(icq); |
0d945c1f | 121 | spin_unlock(&q->queue_lock); |
2274b029 | 122 | } else { |
ab96bbab JO |
123 | /* Make sure q and icq cannot be freed. */ |
124 | rcu_read_lock(); | |
125 | ||
126 | /* Re-acquire the locks in the correct order. */ | |
127 | spin_unlock(&ioc->lock); | |
128 | spin_lock(&q->queue_lock); | |
129 | spin_lock(&ioc->lock); | |
130 | ||
131 | /* | |
132 | * The icq may have been destroyed when the ioc lock | |
133 | * was released. | |
134 | */ | |
135 | if (!(icq->flags & ICQ_DESTROYED)) | |
136 | ioc_destroy_icq(icq); | |
137 | ||
138 | spin_unlock(&q->queue_lock); | |
139 | rcu_read_unlock(); | |
b2efa052 | 140 | } |
b2efa052 | 141 | } |
ffc4e759 | 142 | |
a43f085f | 143 | spin_unlock_irq(&ioc->lock); |
b2efa052 TH |
144 | |
145 | kmem_cache_free(iocontext_cachep, ioc); | |
86db1e29 JA |
146 | } |
147 | ||
edf70ff5 CH |
148 | /* |
149 | * Releasing icqs requires reverse order double locking and we may already be | |
150 | * holding a queue_lock. Do it asynchronously from a workqueue. | |
151 | */ | |
152 | static bool ioc_delay_free(struct io_context *ioc) | |
153 | { | |
154 | unsigned long flags; | |
155 | ||
156 | spin_lock_irqsave(&ioc->lock, flags); | |
157 | if (!hlist_empty(&ioc->icq_list)) { | |
158 | queue_work(system_power_efficient_wq, &ioc->release_work); | |
159 | spin_unlock_irqrestore(&ioc->lock, flags); | |
160 | return true; | |
161 | } | |
162 | spin_unlock_irqrestore(&ioc->lock, flags); | |
163 | return false; | |
164 | } | |
165 | ||
5ef16305 CH |
166 | /** |
167 | * ioc_clear_queue - break any ioc association with the specified queue | |
168 | * @q: request_queue being cleared | |
169 | * | |
170 | * Walk @q->icq_list and exit all io_cq's. | |
171 | */ | |
172 | void ioc_clear_queue(struct request_queue *q) | |
173 | { | |
174 | LIST_HEAD(icq_list); | |
175 | ||
176 | spin_lock_irq(&q->queue_lock); | |
177 | list_splice_init(&q->icq_list, &icq_list); | |
178 | spin_unlock_irq(&q->queue_lock); | |
179 | ||
180 | rcu_read_lock(); | |
181 | while (!list_empty(&icq_list)) { | |
182 | struct io_cq *icq = | |
183 | list_entry(icq_list.next, struct io_cq, q_node); | |
184 | ||
185 | spin_lock_irq(&icq->ioc->lock); | |
186 | if (!(icq->flags & ICQ_DESTROYED)) | |
187 | ioc_destroy_icq(icq); | |
188 | spin_unlock_irq(&icq->ioc->lock); | |
189 | } | |
190 | rcu_read_unlock(); | |
191 | } | |
192 | #else /* CONFIG_BLK_ICQ */ | |
193 | static inline void ioc_exit_icqs(struct io_context *ioc) | |
194 | { | |
195 | } | |
196 | static inline bool ioc_delay_free(struct io_context *ioc) | |
197 | { | |
198 | return false; | |
199 | } | |
200 | #endif /* CONFIG_BLK_ICQ */ | |
201 | ||
42ec57a8 TH |
202 | /** |
203 | * put_io_context - put a reference of io_context | |
204 | * @ioc: io_context to put | |
205 | * | |
206 | * Decrement reference count of @ioc and release it if the count reaches | |
11a3122f | 207 | * zero. |
86db1e29 | 208 | */ |
11a3122f | 209 | void put_io_context(struct io_context *ioc) |
86db1e29 | 210 | { |
42ec57a8 | 211 | BUG_ON(atomic_long_read(&ioc->refcount) <= 0); |
edf70ff5 | 212 | if (atomic_long_dec_and_test(&ioc->refcount) && !ioc_delay_free(ioc)) |
ff8c1474 | 213 | kmem_cache_free(iocontext_cachep, ioc); |
86db1e29 | 214 | } |
222ee581 | 215 | EXPORT_SYMBOL_GPL(put_io_context); |
86db1e29 | 216 | |
f6e8d01b TH |
217 | /* Called by the exiting task */ |
218 | void exit_io_context(struct task_struct *task) | |
219 | { | |
220 | struct io_context *ioc; | |
221 | ||
222 | task_lock(task); | |
223 | ioc = task->io_context; | |
224 | task->io_context = NULL; | |
225 | task_unlock(task); | |
226 | ||
4be8a2ea CH |
227 | if (atomic_dec_and_test(&ioc->active_ref)) { |
228 | ioc_exit_icqs(ioc); | |
229 | put_io_context(ioc); | |
230 | } | |
f6e8d01b TH |
231 | } |
232 | ||
a0f14d8b | 233 | static struct io_context *alloc_io_context(gfp_t gfp_flags, int node) |
86db1e29 | 234 | { |
df415656 | 235 | struct io_context *ioc; |
86db1e29 | 236 | |
42ec57a8 TH |
237 | ioc = kmem_cache_alloc_node(iocontext_cachep, gfp_flags | __GFP_ZERO, |
238 | node); | |
239 | if (unlikely(!ioc)) | |
a0f14d8b | 240 | return NULL; |
42ec57a8 | 241 | |
42ec57a8 | 242 | atomic_long_set(&ioc->refcount, 1); |
f6e8d01b | 243 | atomic_set(&ioc->active_ref, 1); |
5ef16305 | 244 | #ifdef CONFIG_BLK_ICQ |
42ec57a8 | 245 | spin_lock_init(&ioc->lock); |
c137969b | 246 | INIT_RADIX_TREE(&ioc->icq_tree, GFP_ATOMIC); |
c5869807 | 247 | INIT_HLIST_HEAD(&ioc->icq_list); |
b2efa052 | 248 | INIT_WORK(&ioc->release_work, ioc_release_fn); |
5ef16305 | 249 | #endif |
a0f14d8b CH |
250 | return ioc; |
251 | } | |
252 | ||
a411cd3c CH |
253 | int set_task_ioprio(struct task_struct *task, int ioprio) |
254 | { | |
255 | int err; | |
a411cd3c CH |
256 | const struct cred *cred = current_cred(), *tcred; |
257 | ||
258 | rcu_read_lock(); | |
259 | tcred = __task_cred(task); | |
260 | if (!uid_eq(tcred->uid, cred->euid) && | |
261 | !uid_eq(tcred->uid, cred->uid) && !capable(CAP_SYS_NICE)) { | |
262 | rcu_read_unlock(); | |
263 | return -EPERM; | |
264 | } | |
265 | rcu_read_unlock(); | |
266 | ||
267 | err = security_task_setioprio(task, ioprio); | |
268 | if (err) | |
269 | return err; | |
270 | ||
8472161b CH |
271 | task_lock(task); |
272 | if (unlikely(!task->io_context)) { | |
273 | struct io_context *ioc; | |
a411cd3c | 274 | |
8472161b | 275 | task_unlock(task); |
5fc11eeb CH |
276 | |
277 | ioc = alloc_io_context(GFP_ATOMIC, NUMA_NO_NODE); | |
278 | if (!ioc) | |
279 | return -ENOMEM; | |
280 | ||
281 | task_lock(task); | |
a957b612 | 282 | if (task->flags & PF_EXITING) { |
a957b612 JA |
283 | kmem_cache_free(iocontext_cachep, ioc); |
284 | goto out; | |
285 | } | |
669a0646 | 286 | if (task->io_context) |
5fc11eeb | 287 | kmem_cache_free(iocontext_cachep, ioc); |
669a0646 | 288 | else |
5fc11eeb | 289 | task->io_context = ioc; |
8472161b CH |
290 | } |
291 | task->io_context->ioprio = ioprio; | |
a957b612 | 292 | out: |
8472161b | 293 | task_unlock(task); |
15583a56 | 294 | return 0; |
a411cd3c CH |
295 | } |
296 | EXPORT_SYMBOL_GPL(set_task_ioprio); | |
297 | ||
88c9a2ce CH |
298 | int __copy_io(unsigned long clone_flags, struct task_struct *tsk) |
299 | { | |
300 | struct io_context *ioc = current->io_context; | |
88c9a2ce CH |
301 | |
302 | /* | |
303 | * Share io context with parent, if CLONE_IO is set | |
304 | */ | |
305 | if (clone_flags & CLONE_IO) { | |
50569c24 | 306 | atomic_inc(&ioc->active_ref); |
88c9a2ce CH |
307 | tsk->io_context = ioc; |
308 | } else if (ioprio_valid(ioc->ioprio)) { | |
8ffc1368 CH |
309 | tsk->io_context = alloc_io_context(GFP_KERNEL, NUMA_NO_NODE); |
310 | if (!tsk->io_context) | |
88c9a2ce | 311 | return -ENOMEM; |
8ffc1368 | 312 | tsk->io_context->ioprio = ioc->ioprio; |
88c9a2ce CH |
313 | } |
314 | ||
315 | return 0; | |
316 | } | |
317 | ||
5ef16305 | 318 | #ifdef CONFIG_BLK_ICQ |
47fdd4ca TH |
319 | /** |
320 | * ioc_lookup_icq - lookup io_cq from ioc | |
47fdd4ca TH |
321 | * @q: the associated request_queue |
322 | * | |
323 | * Look up io_cq associated with @ioc - @q pair from @ioc. Must be called | |
324 | * with @q->queue_lock held. | |
325 | */ | |
eca5892a | 326 | struct io_cq *ioc_lookup_icq(struct request_queue *q) |
47fdd4ca | 327 | { |
eca5892a | 328 | struct io_context *ioc = current->io_context; |
47fdd4ca TH |
329 | struct io_cq *icq; |
330 | ||
0d945c1f | 331 | lockdep_assert_held(&q->queue_lock); |
47fdd4ca TH |
332 | |
333 | /* | |
334 | * icq's are indexed from @ioc using radix tree and hint pointer, | |
335 | * both of which are protected with RCU. All removals are done | |
336 | * holding both q and ioc locks, and we're holding q lock - if we | |
337 | * find a icq which points to us, it's guaranteed to be valid. | |
338 | */ | |
339 | rcu_read_lock(); | |
340 | icq = rcu_dereference(ioc->icq_hint); | |
341 | if (icq && icq->q == q) | |
342 | goto out; | |
343 | ||
344 | icq = radix_tree_lookup(&ioc->icq_tree, q->id); | |
345 | if (icq && icq->q == q) | |
346 | rcu_assign_pointer(ioc->icq_hint, icq); /* allowed to race */ | |
347 | else | |
348 | icq = NULL; | |
349 | out: | |
350 | rcu_read_unlock(); | |
351 | return icq; | |
352 | } | |
353 | EXPORT_SYMBOL(ioc_lookup_icq); | |
354 | ||
f1f8cc94 TH |
355 | /** |
356 | * ioc_create_icq - create and link io_cq | |
357 | * @q: request_queue of interest | |
f1f8cc94 | 358 | * |
24acfc34 TH |
359 | * Make sure io_cq linking @ioc and @q exists. If icq doesn't exist, they |
360 | * will be created using @gfp_mask. | |
f1f8cc94 TH |
361 | * |
362 | * The caller is responsible for ensuring @ioc won't go away and @q is | |
363 | * alive and will stay alive until this function returns. | |
364 | */ | |
18b74c4d | 365 | static struct io_cq *ioc_create_icq(struct request_queue *q) |
f1f8cc94 | 366 | { |
18b74c4d | 367 | struct io_context *ioc = current->io_context; |
f1f8cc94 | 368 | struct elevator_type *et = q->elevator->type; |
f1f8cc94 TH |
369 | struct io_cq *icq; |
370 | ||
371 | /* allocate stuff */ | |
18b74c4d | 372 | icq = kmem_cache_alloc_node(et->icq_cache, GFP_ATOMIC | __GFP_ZERO, |
f1f8cc94 TH |
373 | q->node); |
374 | if (!icq) | |
375 | return NULL; | |
376 | ||
18b74c4d | 377 | if (radix_tree_maybe_preload(GFP_ATOMIC) < 0) { |
f1f8cc94 TH |
378 | kmem_cache_free(et->icq_cache, icq); |
379 | return NULL; | |
380 | } | |
381 | ||
382 | icq->ioc = ioc; | |
383 | icq->q = q; | |
384 | INIT_LIST_HEAD(&icq->q_node); | |
385 | INIT_HLIST_NODE(&icq->ioc_node); | |
386 | ||
387 | /* lock both q and ioc and try to link @icq */ | |
0d945c1f | 388 | spin_lock_irq(&q->queue_lock); |
f1f8cc94 TH |
389 | spin_lock(&ioc->lock); |
390 | ||
391 | if (likely(!radix_tree_insert(&ioc->icq_tree, q->id, icq))) { | |
392 | hlist_add_head(&icq->ioc_node, &ioc->icq_list); | |
393 | list_add(&icq->q_node, &q->icq_list); | |
f9cd4bfe JA |
394 | if (et->ops.init_icq) |
395 | et->ops.init_icq(icq); | |
f1f8cc94 TH |
396 | } else { |
397 | kmem_cache_free(et->icq_cache, icq); | |
eca5892a | 398 | icq = ioc_lookup_icq(q); |
f1f8cc94 TH |
399 | if (!icq) |
400 | printk(KERN_ERR "cfq: icq link failed!\n"); | |
401 | } | |
402 | ||
403 | spin_unlock(&ioc->lock); | |
0d945c1f | 404 | spin_unlock_irq(&q->queue_lock); |
f1f8cc94 TH |
405 | radix_tree_preload_end(); |
406 | return icq; | |
407 | } | |
408 | ||
87dd1d63 CH |
409 | struct io_cq *ioc_find_get_icq(struct request_queue *q) |
410 | { | |
d538ea4c CH |
411 | struct io_context *ioc = current->io_context; |
412 | struct io_cq *icq = NULL; | |
87dd1d63 | 413 | |
d538ea4c | 414 | if (unlikely(!ioc)) { |
90b627f5 | 415 | ioc = alloc_io_context(GFP_ATOMIC, q->node); |
d538ea4c CH |
416 | if (!ioc) |
417 | return NULL; | |
90b627f5 CH |
418 | |
419 | task_lock(current); | |
420 | if (current->io_context) { | |
421 | kmem_cache_free(iocontext_cachep, ioc); | |
422 | ioc = current->io_context; | |
423 | } else { | |
424 | current->io_context = ioc; | |
425 | } | |
426 | ||
427 | get_io_context(ioc); | |
428 | task_unlock(current); | |
d538ea4c CH |
429 | } else { |
430 | get_io_context(ioc); | |
87dd1d63 | 431 | |
d538ea4c | 432 | spin_lock_irq(&q->queue_lock); |
eca5892a | 433 | icq = ioc_lookup_icq(q); |
d538ea4c CH |
434 | spin_unlock_irq(&q->queue_lock); |
435 | } | |
87dd1d63 CH |
436 | |
437 | if (!icq) { | |
18b74c4d | 438 | icq = ioc_create_icq(q); |
d538ea4c CH |
439 | if (!icq) { |
440 | put_io_context(ioc); | |
87dd1d63 | 441 | return NULL; |
d538ea4c | 442 | } |
87dd1d63 | 443 | } |
87dd1d63 CH |
444 | return icq; |
445 | } | |
446 | EXPORT_SYMBOL_GPL(ioc_find_get_icq); | |
5ef16305 | 447 | #endif /* CONFIG_BLK_ICQ */ |
87dd1d63 | 448 | |
13341598 | 449 | static int __init blk_ioc_init(void) |
86db1e29 JA |
450 | { |
451 | iocontext_cachep = kmem_cache_create("blkdev_ioc", | |
452 | sizeof(struct io_context), 0, SLAB_PANIC, NULL); | |
453 | return 0; | |
454 | } | |
455 | subsys_initcall(blk_ioc_init); |