[linux-block.git] / io_uring / sqpoll.c

// SPDX-License-Identifier: GPL-2.0
/*
 * Contains the core associated with submission side polling of the SQ
 * ring, offloading submissions from the application to a kernel thread.
 */
#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/file.h>
#include <linux/mm.h>
#include <linux/slab.h>
#include <linux/audit.h>
#include <linux/security.h>
#include <linux/io_uring.h>

#include <uapi/linux/io_uring.h>

#include "io_uring.h"
#include "sqpoll.h"

#define IORING_SQPOLL_CAP_ENTRIES_VALUE 8

enum {
	IO_SQ_THREAD_SHOULD_STOP = 0,
	IO_SQ_THREAD_SHOULD_PARK,
};

void io_sq_thread_unpark(struct io_sq_data *sqd)
	__releases(&sqd->lock)
{
	WARN_ON_ONCE(sqd->thread == current);

	/*
	 * Do the dance but not conditional clear_bit() because it'd race with
	 * other threads incrementing park_pending and setting the bit.
	 */
	clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
	if (atomic_dec_return(&sqd->park_pending))
		set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
	mutex_unlock(&sqd->lock);
}

void io_sq_thread_park(struct io_sq_data *sqd)
	__acquires(&sqd->lock)
{
	WARN_ON_ONCE(sqd->thread == current);

	atomic_inc(&sqd->park_pending);
	set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
	mutex_lock(&sqd->lock);
	if (sqd->thread)
		wake_up_process(sqd->thread);
}

void io_sq_thread_stop(struct io_sq_data *sqd)
{
	WARN_ON_ONCE(sqd->thread == current);
	WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state));

	set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
	mutex_lock(&sqd->lock);
	if (sqd->thread)
		wake_up_process(sqd->thread);
	mutex_unlock(&sqd->lock);
	wait_for_completion(&sqd->exited);
}

void io_put_sq_data(struct io_sq_data *sqd)
{
	if (refcount_dec_and_test(&sqd->refs)) {
		WARN_ON_ONCE(atomic_read(&sqd->park_pending));

		io_sq_thread_stop(sqd);
		kfree(sqd);
	}
}

static __cold void io_sqd_update_thread_idle(struct io_sq_data *sqd)
{
	struct io_ring_ctx *ctx;
	unsigned sq_thread_idle = 0;

	list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
		sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle);
	sqd->sq_thread_idle = sq_thread_idle;
}

void io_sq_thread_finish(struct io_ring_ctx *ctx)
{
	struct io_sq_data *sqd = ctx->sq_data;

	if (sqd) {
		io_sq_thread_park(sqd);
		list_del_init(&ctx->sqd_list);
		io_sqd_update_thread_idle(sqd);
		io_sq_thread_unpark(sqd);

		io_put_sq_data(sqd);
		ctx->sq_data = NULL;
	}
}

static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p)
{
	struct io_ring_ctx *ctx_attach;
	struct io_sq_data *sqd;
	struct fd f;

	f = fdget(p->wq_fd);
	if (!f.file)
		return ERR_PTR(-ENXIO);
	if (!io_is_uring_fops(f.file)) {
		fdput(f);
		return ERR_PTR(-EINVAL);
	}

	ctx_attach = f.file->private_data;
	sqd = ctx_attach->sq_data;
	if (!sqd) {
		fdput(f);
		return ERR_PTR(-EINVAL);
	}
	if (sqd->task_tgid != current->tgid) {
		fdput(f);
		return ERR_PTR(-EPERM);
	}

	refcount_inc(&sqd->refs);
	fdput(f);
	return sqd;
}

static struct io_sq_data *io_get_sq_data(struct io_uring_params *p,
					 bool *attached)
{
	struct io_sq_data *sqd;

	*attached = false;
	if (p->flags & IORING_SETUP_ATTACH_WQ) {
		sqd = io_attach_sq_data(p);
		if (!IS_ERR(sqd)) {
			*attached = true;
			return sqd;
		}
		/* fall through for EPERM case, setup new sqd/task */
		if (PTR_ERR(sqd) != -EPERM)
			return sqd;
	}

	sqd = kzalloc(sizeof(*sqd), GFP_KERNEL);
	if (!sqd)
		return ERR_PTR(-ENOMEM);

	atomic_set(&sqd->park_pending, 0);
	refcount_set(&sqd->refs, 1);
	INIT_LIST_HEAD(&sqd->ctx_list);
	mutex_init(&sqd->lock);
	init_waitqueue_head(&sqd->wait);
	init_completion(&sqd->exited);
	return sqd;
}

static inline bool io_sqd_events_pending(struct io_sq_data *sqd)
{
	return READ_ONCE(sqd->state);
}

static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
{
	unsigned int to_submit;
	int ret = 0;

	to_submit = io_sqring_entries(ctx);
	/* if we're handling multiple rings, cap submit size for fairness */
	if (cap_entries && to_submit > IORING_SQPOLL_CAP_ENTRIES_VALUE)
		to_submit = IORING_SQPOLL_CAP_ENTRIES_VALUE;

	if (!wq_list_empty(&ctx->iopoll_list) || to_submit) {
		const struct cred *creds = NULL;

		if (ctx->sq_creds != current_cred())
			creds = override_creds(ctx->sq_creds);

		mutex_lock(&ctx->uring_lock);
		if (!wq_list_empty(&ctx->iopoll_list))
			io_do_iopoll(ctx, true);

		/*
		 * Don't submit if refs are dying, good for io_uring_register(),
		 * but also it is relied upon by io_ring_exit_work()
		 */
		if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) &&
		    !(ctx->flags & IORING_SETUP_R_DISABLED))
			ret = io_submit_sqes(ctx, to_submit);
		mutex_unlock(&ctx->uring_lock);

		if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait))
			wake_up(&ctx->sqo_sq_wait);
		if (creds)
			revert_creds(creds);
	}

	return ret;
}

static bool io_sqd_handle_event(struct io_sq_data *sqd)
{
	bool did_sig = false;
	struct ksignal ksig;

	if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) ||
	    signal_pending(current)) {
		mutex_unlock(&sqd->lock);
		if (signal_pending(current))
			did_sig = get_signal(&ksig);
		cond_resched();
		mutex_lock(&sqd->lock);
	}
	return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
}

static int io_sq_thread(void *data)
{
	struct io_sq_data *sqd = data;
	struct io_ring_ctx *ctx;
	unsigned long timeout = 0;
	char buf[TASK_COMM_LEN];
	DEFINE_WAIT(wait);

	snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid);
	set_task_comm(current, buf);

	if (sqd->sq_cpu != -1)
		set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu));
	else
		set_cpus_allowed_ptr(current, cpu_online_mask);

	mutex_lock(&sqd->lock);
	while (1) {
		bool cap_entries, sqt_spin = false;

		if (io_sqd_events_pending(sqd) || signal_pending(current)) {
			if (io_sqd_handle_event(sqd))
				break;
			timeout = jiffies + sqd->sq_thread_idle;
		}

		cap_entries = !list_is_singular(&sqd->ctx_list);
		list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
			int ret = __io_sq_thread(ctx, cap_entries);

			if (!sqt_spin && (ret > 0 || !wq_list_empty(&ctx->iopoll_list)))
				sqt_spin = true;
		}
		if (io_run_task_work())
			sqt_spin = true;

		if (sqt_spin || !time_after(jiffies, timeout)) {
			cond_resched();
			if (sqt_spin)
				timeout = jiffies + sqd->sq_thread_idle;
			continue;
		}

		prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE);
		if (!io_sqd_events_pending(sqd) && !task_work_pending(current)) {
			bool needs_sched = true;

			list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
				atomic_or(IORING_SQ_NEED_WAKEUP,
						&ctx->rings->sq_flags);
				if ((ctx->flags & IORING_SETUP_IOPOLL) &&
				    !wq_list_empty(&ctx->iopoll_list)) {
					needs_sched = false;
					break;
				}

				/*
				 * Ensure the store of the wakeup flag is not
				 * reordered with the load of the SQ tail
				 */
				smp_mb__after_atomic();

				if (io_sqring_entries(ctx)) {
					needs_sched = false;
					break;
				}
			}

			if (needs_sched) {
				mutex_unlock(&sqd->lock);
				schedule();
				mutex_lock(&sqd->lock);
			}
			list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
				atomic_andnot(IORING_SQ_NEED_WAKEUP,
						&ctx->rings->sq_flags);
		}

		finish_wait(&sqd->wait, &wait);
		timeout = jiffies + sqd->sq_thread_idle;
	}

	io_uring_cancel_generic(true, sqd);
	sqd->thread = NULL;
	list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
		atomic_or(IORING_SQ_NEED_WAKEUP, &ctx->rings->sq_flags);
	io_run_task_work();
	mutex_unlock(&sqd->lock);

	complete(&sqd->exited);
	do_exit(0);
}

void io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
{
	DEFINE_WAIT(wait);

	do {
		if (!io_sqring_full(ctx))
			break;
		prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE);

		if (!io_sqring_full(ctx))
			break;
		schedule();
	} while (!signal_pending(current));

	finish_wait(&ctx->sqo_sq_wait, &wait);
}

__cold int io_sq_offload_create(struct io_ring_ctx *ctx,
				struct io_uring_params *p)
{
	int ret;

	/* Retain compatibility with failing for an invalid attach attempt */
	if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) ==
				IORING_SETUP_ATTACH_WQ) {
		struct fd f;

		f = fdget(p->wq_fd);
		if (!f.file)
			return -ENXIO;
		if (!io_is_uring_fops(f.file)) {
			fdput(f);
			return -EINVAL;
		}
		fdput(f);
	}
	if (ctx->flags & IORING_SETUP_SQPOLL) {
		struct task_struct *tsk;
		struct io_sq_data *sqd;
		bool attached;

		ret = security_uring_sqpoll();
		if (ret)
			return ret;

		sqd = io_get_sq_data(p, &attached);
		if (IS_ERR(sqd)) {
			ret = PTR_ERR(sqd);
			goto err;
		}

		ctx->sq_creds = get_current_cred();
		ctx->sq_data = sqd;
		ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
		if (!ctx->sq_thread_idle)
			ctx->sq_thread_idle = HZ;

		io_sq_thread_park(sqd);
		list_add(&ctx->sqd_list, &sqd->ctx_list);
		io_sqd_update_thread_idle(sqd);
		/* don't attach to a dying SQPOLL thread, would be racy */
		ret = (attached && !sqd->thread) ? -ENXIO : 0;
		io_sq_thread_unpark(sqd);

		if (ret < 0)
			goto err;
		if (attached)
			return 0;

		if (p->flags & IORING_SETUP_SQ_AFF) {
			int cpu = p->sq_thread_cpu;

			ret = -EINVAL;
			if (cpu >= nr_cpu_ids || !cpu_online(cpu))
				goto err_sqpoll;
			sqd->sq_cpu = cpu;
		} else {
			sqd->sq_cpu = -1;
		}

		sqd->task_pid = current->pid;
		sqd->task_tgid = current->tgid;
		tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE);
		if (IS_ERR(tsk)) {
			ret = PTR_ERR(tsk);
			goto err_sqpoll;
		}

		sqd->thread = tsk;
		ret = io_uring_alloc_task_context(tsk, ctx);
		wake_up_new_task(tsk);
		if (ret)
			goto err;
	} else if (p->flags & IORING_SETUP_SQ_AFF) {
		/* Can't have SQ_AFF without SQPOLL */
		ret = -EINVAL;
		goto err;
	}

	return 0;
err_sqpoll:
	complete(&ctx->sq_data->exited);
err:
	io_sq_thread_finish(ctx);
	return ret;
}
Commit	Line	Data
17437f31 JA	1	// SPDX-License-Identifier: GPL-2.0
	2	/*
	3	* Contains the core associated with submission side polling of the SQ
	4	* ring, offloading submissions from the application to a kernel thread.
	5	*/
	6	#include <linux/kernel.h>
	7	#include <linux/errno.h>
	8	#include <linux/file.h>
	9	#include <linux/mm.h>
	10	#include <linux/slab.h>
	11	#include <linux/audit.h>
	12	#include <linux/security.h>
	13	#include <linux/io_uring.h>
	14
	15	#include <uapi/linux/io_uring.h>
	16
17437f31 JA	17	#include "io_uring.h"
	18	#include "sqpoll.h"
	19
	20	#define IORING_SQPOLL_CAP_ENTRIES_VALUE 8
	21
	22	enum {
	23	IO_SQ_THREAD_SHOULD_STOP = 0,
	24	IO_SQ_THREAD_SHOULD_PARK,
	25	};
	26
	27	void io_sq_thread_unpark(struct io_sq_data *sqd)
	28	__releases(&sqd->lock)
	29	{
	30	WARN_ON_ONCE(sqd->thread == current);
	31
	32	/*
	33	* Do the dance but not conditional clear_bit() because it'd race with
	34	* other threads incrementing park_pending and setting the bit.
	35	*/
	36	clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
	37	if (atomic_dec_return(&sqd->park_pending))
	38	set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
	39	mutex_unlock(&sqd->lock);
	40	}
	41
	42	void io_sq_thread_park(struct io_sq_data *sqd)
	43	__acquires(&sqd->lock)
	44	{
	45	WARN_ON_ONCE(sqd->thread == current);
	46
	47	atomic_inc(&sqd->park_pending);
	48	set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
	49	mutex_lock(&sqd->lock);
	50	if (sqd->thread)
	51	wake_up_process(sqd->thread);
	52	}
	53
	54	void io_sq_thread_stop(struct io_sq_data *sqd)
	55	{
	56	WARN_ON_ONCE(sqd->thread == current);
	57	WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state));
	58
	59	set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
	60	mutex_lock(&sqd->lock);
	61	if (sqd->thread)
	62	wake_up_process(sqd->thread);
	63	mutex_unlock(&sqd->lock);
	64	wait_for_completion(&sqd->exited);
	65	}
	66
	67	void io_put_sq_data(struct io_sq_data *sqd)
	68	{
	69	if (refcount_dec_and_test(&sqd->refs)) {
	70	WARN_ON_ONCE(atomic_read(&sqd->park_pending));
	71
	72	io_sq_thread_stop(sqd);
	73	kfree(sqd);
	74	}
	75	}
	76
	77	static __cold void io_sqd_update_thread_idle(struct io_sq_data *sqd)
	78	{
	79	struct io_ring_ctx *ctx;
	80	unsigned sq_thread_idle = 0;
81
82	list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
83	sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle);
84	sqd->sq_thread_idle = sq_thread_idle;
85	}
86
87	void io_sq_thread_finish(struct io_ring_ctx *ctx)
88	{
89	struct io_sq_data *sqd = ctx->sq_data;
90
91	if (sqd) {
92	io_sq_thread_park(sqd);
93	list_del_init(&ctx->sqd_list);
94	io_sqd_update_thread_idle(sqd);
95	io_sq_thread_unpark(sqd);
96
97	io_put_sq_data(sqd);
98	ctx->sq_data = NULL;
99	}
100	}
101
102	static struct io_sq_data io_attach_sq_data(struct io_uring_params p)
103	{
104	struct io_ring_ctx *ctx_attach;
105	struct io_sq_data *sqd;
106	struct fd f;
107
108	f = fdget(p->wq_fd);
109	if (!f.file)
110	return ERR_PTR(-ENXIO);
111	if (!io_is_uring_fops(f.file)) {
112	fdput(f);
113	return ERR_PTR(-EINVAL);
114	}
115
116	ctx_attach = f.file->private_data;
117	sqd = ctx_attach->sq_data;
118	if (!sqd) {
119	fdput(f);
120	return ERR_PTR(-EINVAL);
121	}
122	if (sqd->task_tgid != current->tgid) {
123	fdput(f);
124	return ERR_PTR(-EPERM);
125	}
126
127	refcount_inc(&sqd->refs);
128	fdput(f);
129	return sqd;
130	}
131
132	static struct io_sq_data io_get_sq_data(struct io_uring_params p,
133	bool *attached)
134	{
135	struct io_sq_data *sqd;
136
137	*attached = false;
138	if (p->flags & IORING_SETUP_ATTACH_WQ) {
139	sqd = io_attach_sq_data(p);
140	if (!IS_ERR(sqd)) {
141	*attached = true;
142	return sqd;
143	}
144	/* fall through for EPERM case, setup new sqd/task */
145	if (PTR_ERR(sqd) != -EPERM)
146	return sqd;
147	}
148
149	sqd = kzalloc(sizeof(*sqd), GFP_KERNEL);
150	if (!sqd)
151	return ERR_PTR(-ENOMEM);
152
153	atomic_set(&sqd->park_pending, 0);
154	refcount_set(&sqd->refs, 1);
155	INIT_LIST_HEAD(&sqd->ctx_list);
156	mutex_init(&sqd->lock);
157	init_waitqueue_head(&sqd->wait);
158	init_completion(&sqd->exited);
159	return sqd;
160	}
161
162	static inline bool io_sqd_events_pending(struct io_sq_data *sqd)
163	{
164	return READ_ONCE(sqd->state);
165	}
166
167	static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
168	{
169	unsigned int to_submit;
170	int ret = 0;
171
172	to_submit = io_sqring_entries(ctx);
173	/* if we're handling multiple rings, cap submit size for fairness */
174	if (cap_entries && to_submit > IORING_SQPOLL_CAP_ENTRIES_VALUE)
175	to_submit = IORING_SQPOLL_CAP_ENTRIES_VALUE;
176
177	if (!wq_list_empty(&ctx->iopoll_list) \|\| to_submit) {
178	const struct cred *creds = NULL;
179
180	if (ctx->sq_creds != current_cred())
181	creds = override_creds(ctx->sq_creds);
182
183	mutex_lock(&ctx->uring_lock);
184	if (!wq_list_empty(&ctx->iopoll_list))
185	io_do_iopoll(ctx, true);
186
187	/*
188	* Don't submit if refs are dying, good for io_uring_register(),
189	* but also it is relied upon by io_ring_exit_work()
190	*/
191	if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) &&
192	!(ctx->flags & IORING_SETUP_R_DISABLED))
193	ret = io_submit_sqes(ctx, to_submit);
194	mutex_unlock(&ctx->uring_lock);
195
196	if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait))
197	wake_up(&ctx->sqo_sq_wait);
198	if (creds)
199	revert_creds(creds);
200	}
201
202	return ret;
203	}
204
205	static bool io_sqd_handle_event(struct io_sq_data *sqd)
206	{
207	bool did_sig = false;
208	struct ksignal ksig;
209
210	if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) \|\|
211	signal_pending(current)) {
212	mutex_unlock(&sqd->lock);
213	if (signal_pending(current))
214	did_sig = get_signal(&ksig);
215	cond_resched();
216	mutex_lock(&sqd->lock);
217	}
218	return did_sig \|\| test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
219	}
220
221	static int io_sq_thread(void *data)
222	{
223	struct io_sq_data *sqd = data;
224	struct io_ring_ctx *ctx;
225	unsigned long timeout = 0;
226	char buf[TASK_COMM_LEN];
227	DEFINE_WAIT(wait);
228
229	snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid);
230	set_task_comm(current, buf);
231
232	if (sqd->sq_cpu != -1)
233	set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu));
234	else
235	set_cpus_allowed_ptr(current, cpu_online_mask);
17437f31	236
17437f31 JA	237	mutex_lock(&sqd->lock);
	238	while (1) {
	239	bool cap_entries, sqt_spin = false;
	240
	241	if (io_sqd_events_pending(sqd) \|\| signal_pending(current)) {
	242	if (io_sqd_handle_event(sqd))
	243	break;
	244	timeout = jiffies + sqd->sq_thread_idle;
	245	}
	246
	247	cap_entries = !list_is_singular(&sqd->ctx_list);
	248	list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
	249	int ret = __io_sq_thread(ctx, cap_entries);
	250
	251	if (!sqt_spin && (ret > 0 \|\| !wq_list_empty(&ctx->iopoll_list)))
	252	sqt_spin = true;
	253	}
	254	if (io_run_task_work())
	255	sqt_spin = true;
	256
	257	if (sqt_spin \|\| !time_after(jiffies, timeout)) {
	258	cond_resched();
	259	if (sqt_spin)
	260	timeout = jiffies + sqd->sq_thread_idle;
	261	continue;
	262	}
	263
	264	prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE);
	265	if (!io_sqd_events_pending(sqd) && !task_work_pending(current)) {
	266	bool needs_sched = true;
	267
	268	list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
	269	atomic_or(IORING_SQ_NEED_WAKEUP,
	270	&ctx->rings->sq_flags);
	271	if ((ctx->flags & IORING_SETUP_IOPOLL) &&
	272	!wq_list_empty(&ctx->iopoll_list)) {
	273	needs_sched = false;
	274	break;
	275	}
	276
	277	/*
	278	* Ensure the store of the wakeup flag is not
	279	* reordered with the load of the SQ tail
	280	*/
	281	smp_mb__after_atomic();
	282
	283	if (io_sqring_entries(ctx)) {
	284	needs_sched = false;
	285	break;
	286	}
	287	}
	288
	289	if (needs_sched) {
	290	mutex_unlock(&sqd->lock);
	291	schedule();
	292	mutex_lock(&sqd->lock);
	293	}
	294	list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
	295	atomic_andnot(IORING_SQ_NEED_WAKEUP,
	296	&ctx->rings->sq_flags);
	297	}
	298
	299	finish_wait(&sqd->wait, &wait);
	300	timeout = jiffies + sqd->sq_thread_idle;
301	}
302
303	io_uring_cancel_generic(true, sqd);
304	sqd->thread = NULL;
305	list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
306	atomic_or(IORING_SQ_NEED_WAKEUP, &ctx->rings->sq_flags);
307	io_run_task_work();
308	mutex_unlock(&sqd->lock);
309
17437f31 JA	310	complete(&sqd->exited);
	311	do_exit(0);
	312	}
	313
88b80534	314	void io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
17437f31 JA	315	{
	316	DEFINE_WAIT(wait);
	317
	318	do {
	319	if (!io_sqring_full(ctx))
	320	break;
	321	prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE);
	322
	323	if (!io_sqring_full(ctx))
	324	break;
	325	schedule();
	326	} while (!signal_pending(current));
	327
	328	finish_wait(&ctx->sqo_sq_wait, &wait);
17437f31 JA	329	}
	330
	331	__cold int io_sq_offload_create(struct io_ring_ctx *ctx,
	332	struct io_uring_params *p)
	333	{
	334	int ret;
	335
	336	/* Retain compatibility with failing for an invalid attach attempt */
	337	if ((ctx->flags & (IORING_SETUP_ATTACH_WQ \| IORING_SETUP_SQPOLL)) ==
	338	IORING_SETUP_ATTACH_WQ) {
	339	struct fd f;
	340
	341	f = fdget(p->wq_fd);
	342	if (!f.file)
	343	return -ENXIO;
	344	if (!io_is_uring_fops(f.file)) {
	345	fdput(f);
	346	return -EINVAL;
	347	}
	348	fdput(f);
	349	}
	350	if (ctx->flags & IORING_SETUP_SQPOLL) {
	351	struct task_struct *tsk;
	352	struct io_sq_data *sqd;
	353	bool attached;
	354
	355	ret = security_uring_sqpoll();
	356	if (ret)
	357	return ret;
	358
	359	sqd = io_get_sq_data(p, &attached);
	360	if (IS_ERR(sqd)) {
	361	ret = PTR_ERR(sqd);
	362	goto err;
	363	}
	364
	365	ctx->sq_creds = get_current_cred();
	366	ctx->sq_data = sqd;
	367	ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
	368	if (!ctx->sq_thread_idle)
	369	ctx->sq_thread_idle = HZ;
	370
	371	io_sq_thread_park(sqd);
	372	list_add(&ctx->sqd_list, &sqd->ctx_list);
	373	io_sqd_update_thread_idle(sqd);
	374	/* don't attach to a dying SQPOLL thread, would be racy */
	375	ret = (attached && !sqd->thread) ? -ENXIO : 0;
	376	io_sq_thread_unpark(sqd);
	377
	378	if (ret < 0)
	379	goto err;
	380	if (attached)
	381	return 0;
	382
	383	if (p->flags & IORING_SETUP_SQ_AFF) {
	384	int cpu = p->sq_thread_cpu;
	385
	386	ret = -EINVAL;
	387	if (cpu >= nr_cpu_ids \|\| !cpu_online(cpu))
	388	goto err_sqpoll;
	389	sqd->sq_cpu = cpu;
	390	} else {
	391	sqd->sq_cpu = -1;
	392	}
393
394	sqd->task_pid = current->pid;
395	sqd->task_tgid = current->tgid;
396	tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE);
397	if (IS_ERR(tsk)) {
398	ret = PTR_ERR(tsk);
399	goto err_sqpoll;
400	}
401
402	sqd->thread = tsk;
403	ret = io_uring_alloc_task_context(tsk, ctx);
404	wake_up_new_task(tsk);
405	if (ret)
406	goto err;
407	} else if (p->flags & IORING_SETUP_SQ_AFF) {
408	/* Can't have SQ_AFF without SQPOLL */
409	ret = -EINVAL;
410	goto err;
411	}
412
413	return 0;
414	err_sqpoll:
415	complete(&ctx->sq_data->exited);
416	err:
417	io_sq_thread_finish(ctx);
418	return ret;
419	}