Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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 /* ->sqe isn't available if no async data */
78 if (!req_has_async_data(req))
79 cmd->sqe = NULL;
80 file->f_op->uring_cmd(cmd, IO_URING_F_CANCEL |
81 IO_URING_F_COMPLETE_DEFER);
82 ret = true;
83 }
84 }
85 io_submit_flush_completions(ctx);
86 return ret;
87}
88
89static void io_uring_cmd_del_cancelable(struct io_uring_cmd *cmd,
90 unsigned int issue_flags)
91{
92 struct io_kiocb *req = cmd_to_io_kiocb(cmd);
93 struct io_ring_ctx *ctx = req->ctx;
94
95 if (!(cmd->flags & IORING_URING_CMD_CANCELABLE))
96 return;
97
98 cmd->flags &= ~IORING_URING_CMD_CANCELABLE;
99 io_ring_submit_lock(ctx, issue_flags);
100 hlist_del(&req->hash_node);
101 io_ring_submit_unlock(ctx, issue_flags);
102}
103
104/*
105 * Mark this command as concelable, then io_uring_try_cancel_uring_cmd()
106 * will try to cancel this issued command by sending ->uring_cmd() with
107 * issue_flags of IO_URING_F_CANCEL.
108 *
109 * The command is guaranteed to not be done when calling ->uring_cmd()
110 * with IO_URING_F_CANCEL, but it is driver's responsibility to deal
111 * with race between io_uring canceling and normal completion.
112 */
113void io_uring_cmd_mark_cancelable(struct io_uring_cmd *cmd,
114 unsigned int issue_flags)
115{
116 struct io_kiocb *req = cmd_to_io_kiocb(cmd);
117 struct io_ring_ctx *ctx = req->ctx;
118
119 if (!(cmd->flags & IORING_URING_CMD_CANCELABLE)) {
120 cmd->flags |= IORING_URING_CMD_CANCELABLE;
121 io_ring_submit_lock(ctx, issue_flags);
122 hlist_add_head(&req->hash_node, &ctx->cancelable_uring_cmd);
123 io_ring_submit_unlock(ctx, issue_flags);
124 }
125}
126EXPORT_SYMBOL_GPL(io_uring_cmd_mark_cancelable);
127
128static void io_uring_cmd_work(struct io_kiocb *req, struct io_tw_state *ts)
129{
130 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
131 unsigned int flags = IO_URING_F_COMPLETE_DEFER;
132
133 if (current->flags & (PF_EXITING | PF_KTHREAD))
134 flags |= IO_URING_F_TASK_DEAD;
135
136 /* task_work executor checks the deffered list completion */
137 ioucmd->task_work_cb(ioucmd, flags);
138}
139
140void __io_uring_cmd_do_in_task(struct io_uring_cmd *ioucmd,
141 void (*task_work_cb)(struct io_uring_cmd *, unsigned),
142 unsigned flags)
143{
144 struct io_kiocb *req = cmd_to_io_kiocb(ioucmd);
145
146 ioucmd->task_work_cb = task_work_cb;
147 req->io_task_work.func = io_uring_cmd_work;
148 __io_req_task_work_add(req, flags);
149}
150EXPORT_SYMBOL_GPL(__io_uring_cmd_do_in_task);
151
152static inline void io_req_set_cqe32_extra(struct io_kiocb *req,
153 u64 extra1, u64 extra2)
154{
155 req->big_cqe.extra1 = extra1;
156 req->big_cqe.extra2 = extra2;
157}
158
159/*
160 * Called by consumers of io_uring_cmd, if they originally returned
161 * -EIOCBQUEUED upon receiving the command.
162 */
163void io_uring_cmd_done(struct io_uring_cmd *ioucmd, ssize_t ret, u64 res2,
164 unsigned issue_flags)
165{
166 struct io_kiocb *req = cmd_to_io_kiocb(ioucmd);
167
168 io_uring_cmd_del_cancelable(ioucmd, issue_flags);
169
170 if (ret < 0)
171 req_set_fail(req);
172
173 io_req_set_res(req, ret, 0);
174 if (req->ctx->flags & IORING_SETUP_CQE32)
175 io_req_set_cqe32_extra(req, res2, 0);
176 io_req_uring_cleanup(req, issue_flags);
177 if (req->ctx->flags & IORING_SETUP_IOPOLL) {
178 /* order with io_iopoll_req_issued() checking ->iopoll_complete */
179 smp_store_release(&req->iopoll_completed, 1);
180 } else if (issue_flags & IO_URING_F_COMPLETE_DEFER) {
181 if (WARN_ON_ONCE(issue_flags & IO_URING_F_UNLOCKED))
182 return;
183 io_req_complete_defer(req);
184 } else {
185 req->io_task_work.func = io_req_task_complete;
186 io_req_task_work_add(req);
187 }
188}
189EXPORT_SYMBOL_GPL(io_uring_cmd_done);
190
191static int io_uring_cmd_prep_setup(struct io_kiocb *req,
192 const struct io_uring_sqe *sqe)
193{
194 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
195 struct io_uring_cmd_data *cache;
196
197 cache = io_uring_async_get(req);
198 if (unlikely(!cache))
199 return -ENOMEM;
200
201 if (!(req->flags & REQ_F_FORCE_ASYNC)) {
202 /* defer memcpy until we need it */
203 ioucmd->sqe = sqe;
204 return 0;
205 }
206
207 memcpy(req->async_data, sqe, uring_sqe_size(req->ctx));
208 ioucmd->sqe = req->async_data;
209 return 0;
210}
211
212int io_uring_cmd_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
213{
214 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
215
216 if (sqe->__pad1)
217 return -EINVAL;
218
219 ioucmd->flags = READ_ONCE(sqe->uring_cmd_flags);
220 if (ioucmd->flags & ~IORING_URING_CMD_MASK)
221 return -EINVAL;
222
223 if (ioucmd->flags & IORING_URING_CMD_FIXED) {
224 struct io_ring_ctx *ctx = req->ctx;
225 struct io_rsrc_node *node;
226 u16 index = READ_ONCE(sqe->buf_index);
227
228 node = io_rsrc_node_lookup(&ctx->buf_table, index);
229 if (unlikely(!node))
230 return -EFAULT;
231 /*
232 * Pi node upfront, prior to io_uring_cmd_import_fixed()
233 * being called. This prevents destruction of the mapped buffer
234 * we'll need at actual import time.
235 */
236 io_req_assign_buf_node(req, node);
237 }
238 ioucmd->cmd_op = READ_ONCE(sqe->cmd_op);
239
240 return io_uring_cmd_prep_setup(req, sqe);
241}
242
243int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags)
244{
245 struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
246 struct io_ring_ctx *ctx = req->ctx;
247 struct file *file = req->file;
248 int ret;
249
250 if (!file->f_op->uring_cmd)
251 return -EOPNOTSUPP;
252
253 ret = security_uring_cmd(ioucmd);
254 if (ret)
255 return ret;
256
257 if (ctx->flags & IORING_SETUP_SQE128)
258 issue_flags |= IO_URING_F_SQE128;
259 if (ctx->flags & IORING_SETUP_CQE32)
260 issue_flags |= IO_URING_F_CQE32;
261 if (ctx->compat)
262 issue_flags |= IO_URING_F_COMPAT;
263 if (ctx->flags & IORING_SETUP_IOPOLL) {
264 if (!file->f_op->uring_cmd_iopoll)
265 return -EOPNOTSUPP;
266 issue_flags |= IO_URING_F_IOPOLL;
267 req->iopoll_completed = 0;
268 }
269
270 ret = file->f_op->uring_cmd(ioucmd, issue_flags);
271 if (ret == -EAGAIN) {
272 struct io_uring_cmd_data *cache = req->async_data;
273
274 if (ioucmd->sqe != (void *) cache)
275 memcpy(cache, ioucmd->sqe, uring_sqe_size(req->ctx));
276 return -EAGAIN;
277 } else if (ret == -EIOCBQUEUED) {
278 return -EIOCBQUEUED;
279 }
280
281 if (ret < 0)
282 req_set_fail(req);
283 io_req_uring_cleanup(req, issue_flags);
284 io_req_set_res(req, ret, 0);
285 return IOU_OK;
286}
287
288int io_uring_cmd_import_fixed(u64 ubuf, unsigned long len, int rw,
289 struct iov_iter *iter, void *ioucmd)
290{
291 struct io_kiocb *req = cmd_to_io_kiocb(ioucmd);
292 struct io_rsrc_node *node = req->buf_node;
293
294 /* Must have had rsrc_node assigned at prep time */
295 if (node)
296 return io_import_fixed(rw, iter, node->buf, ubuf, len);
297
298 return -EFAULT;
299}
300EXPORT_SYMBOL_GPL(io_uring_cmd_import_fixed);
301
302void io_uring_cmd_issue_blocking(struct io_uring_cmd *ioucmd)
303{
304 struct io_kiocb *req = cmd_to_io_kiocb(ioucmd);
305
306 io_req_queue_iowq(req);
307}
308
309static inline int io_uring_cmd_getsockopt(struct socket *sock,
310 struct io_uring_cmd *cmd,
311 unsigned int issue_flags)
312{
313 bool compat = !!(issue_flags & IO_URING_F_COMPAT);
314 int optlen, optname, level, err;
315 void __user *optval;
316
317 level = READ_ONCE(cmd->sqe->level);
318 if (level != SOL_SOCKET)
319 return -EOPNOTSUPP;
320
321 optval = u64_to_user_ptr(READ_ONCE(cmd->sqe->optval));
322 optname = READ_ONCE(cmd->sqe->optname);
323 optlen = READ_ONCE(cmd->sqe->optlen);
324
325 err = do_sock_getsockopt(sock, compat, level, optname,
326 USER_SOCKPTR(optval),
327 KERNEL_SOCKPTR(&optlen));
328 if (err)
329 return err;
330
331 /* On success, return optlen */
332 return optlen;
333}
334
335static inline int io_uring_cmd_setsockopt(struct socket *sock,
336 struct io_uring_cmd *cmd,
337 unsigned int issue_flags)
338{
339 bool compat = !!(issue_flags & IO_URING_F_COMPAT);
340 int optname, optlen, level;
341 void __user *optval;
342 sockptr_t optval_s;
343
344 optval = u64_to_user_ptr(READ_ONCE(cmd->sqe->optval));
345 optname = READ_ONCE(cmd->sqe->optname);
346 optlen = READ_ONCE(cmd->sqe->optlen);
347 level = READ_ONCE(cmd->sqe->level);
348 optval_s = USER_SOCKPTR(optval);
349
350 return do_sock_setsockopt(sock, compat, level, optname, optval_s,
351 optlen);
352}
353
354#if defined(CONFIG_NET)
355int io_uring_cmd_sock(struct io_uring_cmd *cmd, unsigned int issue_flags)
356{
357 struct socket *sock = cmd->file->private_data;
358 struct sock *sk = sock->sk;
359 struct proto *prot = READ_ONCE(sk->sk_prot);
360 int ret, arg = 0;
361
362 if (!prot || !prot->ioctl)
363 return -EOPNOTSUPP;
364
365 switch (cmd->sqe->cmd_op) {
366 case SOCKET_URING_OP_SIOCINQ:
367 ret = prot->ioctl(sk, SIOCINQ, &arg);
368 if (ret)
369 return ret;
370 return arg;
371 case SOCKET_URING_OP_SIOCOUTQ:
372 ret = prot->ioctl(sk, SIOCOUTQ, &arg);
373 if (ret)
374 return ret;
375 return arg;
376 case SOCKET_URING_OP_GETSOCKOPT:
377 return io_uring_cmd_getsockopt(sock, cmd, issue_flags);
378 case SOCKET_URING_OP_SETSOCKOPT:
379 return io_uring_cmd_setsockopt(sock, cmd, issue_flags);
380 default:
381 return -EOPNOTSUPP;
382 }
383}
384EXPORT_SYMBOL_GPL(io_uring_cmd_sock);
385#endif