f53c80ecaf076c69367b371a334c15bc76473d10
2 * Functions related to io context handling
4 #include <linux/kernel.h>
5 #include <linux/module.h>
6 #include <linux/init.h>
8 #include <linux/blkdev.h>
9 #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
10 #include <linux/slab.h>
15 * For io context allocations
17 static struct kmem_cache
*iocontext_cachep
;
20 * get_io_context - increment reference count to io_context
21 * @ioc: io_context to get
23 * Increment reference count to @ioc.
25 void get_io_context(struct io_context
*ioc
)
27 BUG_ON(atomic_long_read(&ioc
->refcount
) <= 0);
28 atomic_long_inc(&ioc
->refcount
);
30 EXPORT_SYMBOL(get_io_context
);
32 static void icq_free_icq_rcu(struct rcu_head
*head
)
34 struct io_cq
*icq
= container_of(head
, struct io_cq
, __rcu_head
);
36 kmem_cache_free(icq
->__rcu_icq_cache
, icq
);
40 * Exit and free an icq. Called with both ioc and q locked.
42 static void ioc_exit_icq(struct io_cq
*icq
)
44 struct io_context
*ioc
= icq
->ioc
;
45 struct request_queue
*q
= icq
->q
;
46 struct elevator_type
*et
= q
->elevator
->type
;
48 lockdep_assert_held(&ioc
->lock
);
49 lockdep_assert_held(q
->queue_lock
);
51 radix_tree_delete(&ioc
->icq_tree
, icq
->q
->id
);
52 hlist_del_init(&icq
->ioc_node
);
53 list_del_init(&icq
->q_node
);
56 * Both setting lookup hint to and clearing it from @icq are done
57 * under queue_lock. If it's not pointing to @icq now, it never
58 * will. Hint assignment itself can race safely.
60 if (rcu_dereference_raw(ioc
->icq_hint
) == icq
)
61 rcu_assign_pointer(ioc
->icq_hint
, NULL
);
63 if (et
->ops
.elevator_exit_icq_fn
)
64 et
->ops
.elevator_exit_icq_fn(icq
);
67 * @icq->q might have gone away by the time RCU callback runs
68 * making it impossible to determine icq_cache. Record it in @icq.
70 icq
->__rcu_icq_cache
= et
->icq_cache
;
71 call_rcu(&icq
->__rcu_head
, icq_free_icq_rcu
);
75 * Slow path for ioc release in put_io_context(). Performs double-lock
76 * dancing to unlink all icq's and then frees ioc.
78 static void ioc_release_fn(struct work_struct
*work
)
80 struct io_context
*ioc
= container_of(work
, struct io_context
,
85 * Exiting icq may call into put_io_context() through elevator
86 * which will trigger lockdep warning. The ioc's are guaranteed to
87 * be different, use a different locking subclass here. Use
88 * irqsave variant as there's no spin_lock_irq_nested().
90 spin_lock_irqsave_nested(&ioc
->lock
, flags
, 1);
92 while (!hlist_empty(&ioc
->icq_list
)) {
93 struct io_cq
*icq
= hlist_entry(ioc
->icq_list
.first
,
94 struct io_cq
, ioc_node
);
95 struct request_queue
*q
= icq
->q
;
97 if (spin_trylock(q
->queue_lock
)) {
99 spin_unlock(q
->queue_lock
);
101 spin_unlock_irqrestore(&ioc
->lock
, flags
);
103 spin_lock_irqsave_nested(&ioc
->lock
, flags
, 1);
107 spin_unlock_irqrestore(&ioc
->lock
, flags
);
109 kmem_cache_free(iocontext_cachep
, ioc
);
113 * put_io_context - put a reference of io_context
114 * @ioc: io_context to put
116 * Decrement reference count of @ioc and release it if the count reaches
119 void put_io_context(struct io_context
*ioc
)
126 BUG_ON(atomic_long_read(&ioc
->refcount
) <= 0);
129 * Releasing ioc requires reverse order double locking and we may
130 * already be holding a queue_lock. Do it asynchronously from wq.
132 if (atomic_long_dec_and_test(&ioc
->refcount
)) {
133 spin_lock_irqsave(&ioc
->lock
, flags
);
134 if (!hlist_empty(&ioc
->icq_list
))
135 schedule_work(&ioc
->release_work
);
136 spin_unlock_irqrestore(&ioc
->lock
, flags
);
139 EXPORT_SYMBOL(put_io_context
);
141 /* Called by the exiting task */
142 void exit_io_context(struct task_struct
*task
)
144 struct io_context
*ioc
;
147 ioc
= task
->io_context
;
148 task
->io_context
= NULL
;
151 atomic_dec(&ioc
->nr_tasks
);
156 * ioc_clear_queue - break any ioc association with the specified queue
157 * @q: request_queue being cleared
159 * Walk @q->icq_list and exit all io_cq's. Must be called with @q locked.
161 void ioc_clear_queue(struct request_queue
*q
)
163 lockdep_assert_held(q
->queue_lock
);
165 while (!list_empty(&q
->icq_list
)) {
166 struct io_cq
*icq
= list_entry(q
->icq_list
.next
,
167 struct io_cq
, q_node
);
168 struct io_context
*ioc
= icq
->ioc
;
170 spin_lock(&ioc
->lock
);
172 spin_unlock(&ioc
->lock
);
176 void create_io_context_slowpath(struct task_struct
*task
, gfp_t gfp_flags
,
179 struct io_context
*ioc
;
181 ioc
= kmem_cache_alloc_node(iocontext_cachep
, gfp_flags
| __GFP_ZERO
,
187 atomic_long_set(&ioc
->refcount
, 1);
188 atomic_set(&ioc
->nr_tasks
, 1);
189 spin_lock_init(&ioc
->lock
);
190 INIT_RADIX_TREE(&ioc
->icq_tree
, GFP_ATOMIC
| __GFP_HIGH
);
191 INIT_HLIST_HEAD(&ioc
->icq_list
);
192 INIT_WORK(&ioc
->release_work
, ioc_release_fn
);
195 * Try to install. ioc shouldn't be installed if someone else
196 * already did or @task, which isn't %current, is exiting. Note
197 * that we need to allow ioc creation on exiting %current as exit
198 * path may issue IOs from e.g. exit_files(). The exit path is
199 * responsible for not issuing IO after exit_io_context().
202 if (!task
->io_context
&&
203 (task
== current
|| !(task
->flags
& PF_EXITING
)))
204 task
->io_context
= ioc
;
206 kmem_cache_free(iocontext_cachep
, ioc
);
211 * get_task_io_context - get io_context of a task
212 * @task: task of interest
213 * @gfp_flags: allocation flags, used if allocation is necessary
214 * @node: allocation node, used if allocation is necessary
216 * Return io_context of @task. If it doesn't exist, it is created with
217 * @gfp_flags and @node. The returned io_context has its reference count
220 * This function always goes through task_lock() and it's better to use
221 * %current->io_context + get_io_context() for %current.
223 struct io_context
*get_task_io_context(struct task_struct
*task
,
224 gfp_t gfp_flags
, int node
)
226 struct io_context
*ioc
;
228 might_sleep_if(gfp_flags
& __GFP_WAIT
);
232 ioc
= task
->io_context
;
239 } while (create_io_context(task
, gfp_flags
, node
));
243 EXPORT_SYMBOL(get_task_io_context
);
246 * ioc_lookup_icq - lookup io_cq from ioc
247 * @ioc: the associated io_context
248 * @q: the associated request_queue
250 * Look up io_cq associated with @ioc - @q pair from @ioc. Must be called
251 * with @q->queue_lock held.
253 struct io_cq
*ioc_lookup_icq(struct io_context
*ioc
, struct request_queue
*q
)
257 lockdep_assert_held(q
->queue_lock
);
260 * icq's are indexed from @ioc using radix tree and hint pointer,
261 * both of which are protected with RCU. All removals are done
262 * holding both q and ioc locks, and we're holding q lock - if we
263 * find a icq which points to us, it's guaranteed to be valid.
266 icq
= rcu_dereference(ioc
->icq_hint
);
267 if (icq
&& icq
->q
== q
)
270 icq
= radix_tree_lookup(&ioc
->icq_tree
, q
->id
);
271 if (icq
&& icq
->q
== q
)
272 rcu_assign_pointer(ioc
->icq_hint
, icq
); /* allowed to race */
279 EXPORT_SYMBOL(ioc_lookup_icq
);
282 * ioc_create_icq - create and link io_cq
283 * @q: request_queue of interest
284 * @gfp_mask: allocation mask
286 * Make sure io_cq linking %current->io_context and @q exists. If either
287 * io_context and/or icq don't exist, they will be created using @gfp_mask.
289 * The caller is responsible for ensuring @ioc won't go away and @q is
290 * alive and will stay alive until this function returns.
292 struct io_cq
*ioc_create_icq(struct request_queue
*q
, gfp_t gfp_mask
)
294 struct elevator_type
*et
= q
->elevator
->type
;
295 struct io_context
*ioc
;
299 ioc
= create_io_context(current
, gfp_mask
, q
->node
);
303 icq
= kmem_cache_alloc_node(et
->icq_cache
, gfp_mask
| __GFP_ZERO
,
308 if (radix_tree_preload(gfp_mask
) < 0) {
309 kmem_cache_free(et
->icq_cache
, icq
);
315 INIT_LIST_HEAD(&icq
->q_node
);
316 INIT_HLIST_NODE(&icq
->ioc_node
);
318 /* lock both q and ioc and try to link @icq */
319 spin_lock_irq(q
->queue_lock
);
320 spin_lock(&ioc
->lock
);
322 if (likely(!radix_tree_insert(&ioc
->icq_tree
, q
->id
, icq
))) {
323 hlist_add_head(&icq
->ioc_node
, &ioc
->icq_list
);
324 list_add(&icq
->q_node
, &q
->icq_list
);
325 if (et
->ops
.elevator_init_icq_fn
)
326 et
->ops
.elevator_init_icq_fn(icq
);
328 kmem_cache_free(et
->icq_cache
, icq
);
329 icq
= ioc_lookup_icq(ioc
, q
);
331 printk(KERN_ERR
"cfq: icq link failed!\n");
334 spin_unlock(&ioc
->lock
);
335 spin_unlock_irq(q
->queue_lock
);
336 radix_tree_preload_end();
340 void ioc_set_icq_flags(struct io_context
*ioc
, unsigned int flags
)
343 struct hlist_node
*n
;
345 hlist_for_each_entry(icq
, n
, &ioc
->icq_list
, ioc_node
)
350 * ioc_ioprio_changed - notify ioprio change
351 * @ioc: io_context of interest
352 * @ioprio: new ioprio
354 * @ioc's ioprio has changed to @ioprio. Set %ICQ_IOPRIO_CHANGED for all
355 * icq's. iosched is responsible for checking the bit and applying it on
356 * request issue path.
358 void ioc_ioprio_changed(struct io_context
*ioc
, int ioprio
)
362 spin_lock_irqsave(&ioc
->lock
, flags
);
363 ioc
->ioprio
= ioprio
;
364 ioc_set_icq_flags(ioc
, ICQ_IOPRIO_CHANGED
);
365 spin_unlock_irqrestore(&ioc
->lock
, flags
);
369 * ioc_cgroup_changed - notify cgroup change
370 * @ioc: io_context of interest
372 * @ioc's cgroup has changed. Set %ICQ_CGROUP_CHANGED for all icq's.
373 * iosched is responsible for checking the bit and applying it on request
376 void ioc_cgroup_changed(struct io_context
*ioc
)
380 spin_lock_irqsave(&ioc
->lock
, flags
);
381 ioc_set_icq_flags(ioc
, ICQ_CGROUP_CHANGED
);
382 spin_unlock_irqrestore(&ioc
->lock
, flags
);
384 EXPORT_SYMBOL(ioc_cgroup_changed
);
387 * icq_get_changed - fetch and clear icq changed mask
388 * @icq: icq of interest
390 * Fetch and clear ICQ_*_CHANGED bits from @icq. Grabs and releases
393 unsigned icq_get_changed(struct io_cq
*icq
)
395 unsigned int changed
= 0;
398 if (unlikely(icq
->flags
& ICQ_CHANGED_MASK
)) {
399 spin_lock_irqsave(&icq
->ioc
->lock
, flags
);
400 changed
= icq
->flags
& ICQ_CHANGED_MASK
;
401 icq
->flags
&= ~ICQ_CHANGED_MASK
;
402 spin_unlock_irqrestore(&icq
->ioc
->lock
, flags
);
406 EXPORT_SYMBOL(icq_get_changed
);
408 static int __init
blk_ioc_init(void)
410 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
411 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
);
414 subsys_initcall(blk_ioc_init
);
This page took 0.038823 seconds and 4 git commands to generate.