1// SPDX-License-Identifier: GPL-2.0
  2#include <linux/kernel.h>
  3#include <linux/errno.h>
  4#include <linux/file.h>
  5#include <linux/io_uring/cmd.h>
  6#include <linux/io_uring/net.h>
  7#include <linux/security.h>
  8#include <linux/nospec.h>
  9#include <net/sock.h>
 10
 11#include <uapi/linux/io_uring.h>
 12#include <asm/ioctls.h>
 13
 14#include "io_uring.h"
 15#include "alloc_cache.h"
 16#include "rsrc.h"
 17#include "uring_cmd.h"
 18
 19static struct io_uring_cmd_data *io_uring_async_get(struct io_kiocb *req)
 20{
 21	struct io_ring_ctx *ctx = req->ctx;
 22	struct io_uring_cmd_data *cache;
 23
 24	cache = io_alloc_cache_get(&ctx->uring_cache);
 25	if (cache) {
 26		cache->op_data = NULL;
 27		req->flags |= REQ_F_ASYNC_DATA;
 28		req->async_data = cache;
 29		return cache;
 30	}
 31	if (!io_alloc_async_data(req)) {
 32		cache = req->async_data;
 33		cache->op_data = NULL;
 34		return cache;
 35	}
 36	return NULL;
 37}
 38
 39static void io_req_uring_cleanup(struct io_kiocb *req, unsigned int issue_flags)
 40{
 41	struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
 42	struct io_uring_cmd_data *cache = req->async_data;
 43
 44	if (cache->op_data) {
 45		kfree(cache->op_data);
 46		cache->op_data = NULL;
 47	}
 48
 49	if (issue_flags & IO_URING_F_UNLOCKED)
 50		return;
 51	if (io_alloc_cache_put(&req->ctx->uring_cache, cache)) {
 52		ioucmd->sqe = NULL;
 53		req->async_data = NULL;
 54		req->flags &= ~REQ_F_ASYNC_DATA;
 55	}
 56}
 57
 58bool io_uring_try_cancel_uring_cmd(struct io_ring_ctx *ctx,
 59				   struct io_uring_task *tctx, bool cancel_all)
 60{
 61	struct hlist_node *tmp;
 62	struct io_kiocb *req;
 63	bool ret = false;
 64
 65	lockdep_assert_held(&ctx->uring_lock);
 66
 67	hlist_for_each_entry_safe(req, tmp, &ctx->cancelable_uring_cmd,
 68			hash_node) {
 69		struct io_uring_cmd *cmd = io_kiocb_to_cmd(req,
 70				struct io_uring_cmd);
 71		struct file *file = req->file;
 72
 73		if (!cancel_all && req->tctx != tctx)
 74			continue;
 75
 76		if (cmd->flags & IORING_URING_CMD_CANCELABLE) {
 77			file->f_op->uring_cmd(cmd, IO_URING_F_CANCEL |
 78						   IO_URING_F_COMPLETE_DEFER);
 79			ret = true;
 80		}
 81	}
 82	io_submit_flush_completions(ctx);
 83	return ret;
 84}
 85
 86static void io_uring_cmd_del_cancelable(struct io_uring_cmd *cmd,
 87		unsigned int issue_flags)
 88{
 89	struct io_kiocb *req = cmd_to_io_kiocb(cmd);
 90	struct io_ring_ctx *ctx = req->ctx;
 91
 92	if (!(cmd->flags & IORING_URING_CMD_CANCELABLE))
 93		return;
 94
 95	cmd->flags &= ~IORING_URING_CMD_CANCELABLE;
 96	io_ring_submit_lock(ctx, issue_flags);
 97	hlist_del(&req->hash_node);
 98	io_ring_submit_unlock(ctx, issue_flags);
 99}
100
101/*
102 * Mark this command as concelable, then io_uring_try_cancel_uring_cmd()
103 * will try to cancel this issued command by sending ->uring_cmd() with
104 * issue_flags of IO_URING_F_CANCEL.
105 *
106 * The command is guaranteed to not be done when calling ->uring_cmd()
107 * with IO_URING_F_CANCEL, but it is driver's responsibility to deal
108 * with race between io_uring canceling and normal completion.
109 */
110void io_uring_cmd_mark_cancelable(struct io_uring_cmd *cmd,
111		unsigned int issue_flags)
112{
113	struct io_kiocb *req = cmd_to_io_kiocb(cmd);
114	struct io_ring_ctx *ctx = req->ctx;
115
116	if (!(cmd->flags & IORING_URING_CMD_CANCELABLE)) {
117		cmd->flags |= IORING_URING_CMD_CANCELABLE;
118		io_ring_submit_lock(ctx, issue_flags);
119		hlist_add_head(&req->hash_node, &ctx->cancelable_uring_cmd);
120		io_ring_submit_unlock(ctx, issue_flags);
121	}
122}
123EXPORT_SYMBOL_GPL(io_uring_cmd_mark_cancelable);
124
125static void io_uring_cmd_work(struct io_kiocb *req, struct io_tw_state *ts)
126{
127	struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
128	unsigned int flags = IO_URING_F_COMPLETE_DEFER;
129
130	if (current->flags & (PF_EXITING | PF_KTHREAD))
131		flags |= IO_URING_F_TASK_DEAD;
132
133	/* task_work executor checks the deffered list completion */
134	ioucmd->task_work_cb(ioucmd, flags);
135}
136
137void __io_uring_cmd_do_in_task(struct io_uring_cmd *ioucmd,
138			void (*task_work_cb)(struct io_uring_cmd *, unsigned),
139			unsigned flags)
140{
141	struct io_kiocb *req = cmd_to_io_kiocb(ioucmd);
142
143	ioucmd->task_work_cb = task_work_cb;
144	req->io_task_work.func = io_uring_cmd_work;
145	__io_req_task_work_add(req, flags);
146}
147EXPORT_SYMBOL_GPL(__io_uring_cmd_do_in_task);
148
149static inline void io_req_set_cqe32_extra(struct io_kiocb *req,
150					  u64 extra1, u64 extra2)
151{
152	req->big_cqe.extra1 = extra1;
153	req->big_cqe.extra2 = extra2;
154}
155
156/*
157 * Called by consumers of io_uring_cmd, if they originally returned
158 * -EIOCBQUEUED upon receiving the command.
159 */
160void io_uring_cmd_done(struct io_uring_cmd *ioucmd, ssize_t ret, u64 res2,
161		       unsigned issue_flags)
162{
163	struct io_kiocb *req = cmd_to_io_kiocb(ioucmd);
164
165	io_uring_cmd_del_cancelable(ioucmd, issue_flags);
166
167	if (ret < 0)
168		req_set_fail(req);
169
170	io_req_set_res(req, ret, 0);
171	if (req->ctx->flags & IORING_SETUP_CQE32)
172		io_req_set_cqe32_extra(req, res2, 0);
173	io_req_uring_cleanup(req, issue_flags);
174	if (req->ctx->flags & IORING_SETUP_IOPOLL) {
175		/* order with io_iopoll_req_issued() checking ->iopoll_complete */
176		smp_store_release(&req->iopoll_completed, 1);
177	} else if (issue_flags & IO_URING_F_COMPLETE_DEFER) {
178		if (WARN_ON_ONCE(issue_flags & IO_URING_F_UNLOCKED))
179			return;
180		io_req_complete_defer(req);
181	} else {
182		req->io_task_work.func = io_req_task_complete;
183		io_req_task_work_add(req);
184	}
185}
186EXPORT_SYMBOL_GPL(io_uring_cmd_done);
187
188static int io_uring_cmd_prep_setup(struct io_kiocb *req,
189				   const struct io_uring_sqe *sqe)
190{
191	struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
192	struct io_uring_cmd_data *cache;
193
194	cache = io_uring_async_get(req);
195	if (unlikely(!cache))
196		return -ENOMEM;
197
198	/*
199	 * Unconditionally cache the SQE for now - this is only needed for
200	 * requests that go async, but prep handlers must ensure that any
201	 * sqe data is stable beyond prep. Since uring_cmd is special in
202	 * that it doesn't read in per-op data, play it safe and ensure that
203	 * any SQE data is stable beyond prep. This can later get relaxed.
204	 */
205	memcpy(cache->sqes, sqe, uring_sqe_size(req->ctx));
206	ioucmd->sqe = cache->sqes;
207	return 0;
208}
209
210int io_uring_cmd_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
211{
212	struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
213
214	if (sqe->__pad1)
215		return -EINVAL;
216
217	ioucmd->flags = READ_ONCE(sqe->uring_cmd_flags);
218	if (ioucmd->flags & ~IORING_URING_CMD_MASK)
219		return -EINVAL;
220
221	if (ioucmd->flags & IORING_URING_CMD_FIXED) {
222		struct io_ring_ctx *ctx = req->ctx;
223		struct io_rsrc_node *node;
224		u16 index = READ_ONCE(sqe->buf_index);
225
226		node = io_rsrc_node_lookup(&ctx->buf_table, index);
227		if (unlikely(!node))
228			return -EFAULT;
229		/*
230		 * Pi node upfront, prior to io_uring_cmd_import_fixed()
231		 * being called. This prevents destruction of the mapped buffer
232		 * we'll need at actual import time.
233		 */
234		io_req_assign_buf_node(req, node);
235	}
236	ioucmd->cmd_op = READ_ONCE(sqe->cmd_op);
237
238	return io_uring_cmd_prep_setup(req, sqe);
239}
240
241int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags)
242{
243	struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
244	struct io_ring_ctx *ctx = req->ctx;
245	struct file *file = req->file;
246	int ret;
247
248	if (!file->f_op->uring_cmd)
249		return -EOPNOTSUPP;
250
251	ret = security_uring_cmd(ioucmd);
252	if (ret)
253		return ret;
254
255	if (ctx->flags & IORING_SETUP_SQE128)
256		issue_flags |= IO_URING_F_SQE128;
257	if (ctx->flags & IORING_SETUP_CQE32)
258		issue_flags |= IO_URING_F_CQE32;
259	if (ctx->compat)
260		issue_flags |= IO_URING_F_COMPAT;
261	if (ctx->flags & IORING_SETUP_IOPOLL) {
262		if (!file->f_op->uring_cmd_iopoll)
263			return -EOPNOTSUPP;
264		issue_flags |= IO_URING_F_IOPOLL;
265		req->iopoll_completed = 0;
266	}
267
268	ret = file->f_op->uring_cmd(ioucmd, issue_flags);
269	if (ret == -EAGAIN || ret == -EIOCBQUEUED)
270		return ret;
271	if (ret < 0)
272		req_set_fail(req);
273	io_req_uring_cleanup(req, issue_flags);
274	io_req_set_res(req, ret, 0);
275	return IOU_OK;
276}
277
278int io_uring_cmd_import_fixed(u64 ubuf, unsigned long len, int rw,
279			      struct iov_iter *iter, void *ioucmd)
280{
281	struct io_kiocb *req = cmd_to_io_kiocb(ioucmd);
282	struct io_rsrc_node *node = req->buf_node;
283
284	/* Must have had rsrc_node assigned at prep time */
285	if (node)
286		return io_import_fixed(rw, iter, node->buf, ubuf, len);
287
288	return -EFAULT;
289}
290EXPORT_SYMBOL_GPL(io_uring_cmd_import_fixed);
291
292void io_uring_cmd_issue_blocking(struct io_uring_cmd *ioucmd)
293{
294	struct io_kiocb *req = cmd_to_io_kiocb(ioucmd);
295
296	io_req_queue_iowq(req);
297}
298
299static inline int io_uring_cmd_getsockopt(struct socket *sock,
300					  struct io_uring_cmd *cmd,
301					  unsigned int issue_flags)
302{
303	bool compat = !!(issue_flags & IO_URING_F_COMPAT);
304	int optlen, optname, level, err;
305	void __user *optval;
306
307	level = READ_ONCE(cmd->sqe->level);
308	if (level != SOL_SOCKET)
309		return -EOPNOTSUPP;
310
311	optval = u64_to_user_ptr(READ_ONCE(cmd->sqe->optval));
312	optname = READ_ONCE(cmd->sqe->optname);
313	optlen = READ_ONCE(cmd->sqe->optlen);
314
315	err = do_sock_getsockopt(sock, compat, level, optname,
316				 USER_SOCKPTR(optval),
317				 KERNEL_SOCKPTR(&optlen));
318	if (err)
319		return err;
320
321	/* On success, return optlen */
322	return optlen;
323}
324
325static inline int io_uring_cmd_setsockopt(struct socket *sock,
326					  struct io_uring_cmd *cmd,
327					  unsigned int issue_flags)
328{
329	bool compat = !!(issue_flags & IO_URING_F_COMPAT);
330	int optname, optlen, level;
331	void __user *optval;
332	sockptr_t optval_s;
333
334	optval = u64_to_user_ptr(READ_ONCE(cmd->sqe->optval));
335	optname = READ_ONCE(cmd->sqe->optname);
336	optlen = READ_ONCE(cmd->sqe->optlen);
337	level = READ_ONCE(cmd->sqe->level);
338	optval_s = USER_SOCKPTR(optval);
339
340	return do_sock_setsockopt(sock, compat, level, optname, optval_s,
341				  optlen);
342}
343
344#if defined(CONFIG_NET)
345int io_uring_cmd_sock(struct io_uring_cmd *cmd, unsigned int issue_flags)
346{
347	struct socket *sock = cmd->file->private_data;
348	struct sock *sk = sock->sk;
349	struct proto *prot = READ_ONCE(sk->sk_prot);
350	int ret, arg = 0;
351
352	if (!prot || !prot->ioctl)
353		return -EOPNOTSUPP;
354
355	switch (cmd->cmd_op) {
356	case SOCKET_URING_OP_SIOCINQ:
357		ret = prot->ioctl(sk, SIOCINQ, &arg);
358		if (ret)
359			return ret;
360		return arg;
361	case SOCKET_URING_OP_SIOCOUTQ:
362		ret = prot->ioctl(sk, SIOCOUTQ, &arg);
363		if (ret)
364			return ret;
365		return arg;
366	case SOCKET_URING_OP_GETSOCKOPT:
367		return io_uring_cmd_getsockopt(sock, cmd, issue_flags);
368	case SOCKET_URING_OP_SETSOCKOPT:
369		return io_uring_cmd_setsockopt(sock, cmd, issue_flags);
370	default:
371		return -EOPNOTSUPP;
372	}
373}
374EXPORT_SYMBOL_GPL(io_uring_cmd_sock);
375#endif