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