]>
Commit | Line | Data |
---|---|---|
194bb58c JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | #include <linux/kernel.h> | |
3 | #include <linux/errno.h> | |
4 | #include <linux/fs.h> | |
5 | #include <linux/file.h> | |
6 | #include <linux/io_uring.h> | |
7 | ||
8 | #include <uapi/linux/io_uring.h> | |
9 | ||
10 | #include "../kernel/futex/futex.h" | |
11 | #include "io_uring.h" | |
12 | #include "rsrc.h" | |
13 | #include "futex.h" | |
14 | ||
15 | struct io_futex { | |
16 | struct file *file; | |
8f350194 JA |
17 | union { |
18 | u32 __user *uaddr; | |
19 | struct futex_waitv __user *uwaitv; | |
20 | }; | |
194bb58c JA |
21 | unsigned long futex_val; |
22 | unsigned long futex_mask; | |
8f350194 | 23 | unsigned long futexv_owned; |
194bb58c | 24 | u32 futex_flags; |
8f350194 JA |
25 | unsigned int futex_nr; |
26 | bool futexv_unqueued; | |
194bb58c JA |
27 | }; |
28 | ||
29 | struct io_futex_data { | |
30 | union { | |
31 | struct futex_q q; | |
32 | struct io_cache_entry cache; | |
33 | }; | |
34 | struct io_kiocb *req; | |
35 | }; | |
36 | ||
37 | void io_futex_cache_init(struct io_ring_ctx *ctx) | |
38 | { | |
39 | io_alloc_cache_init(&ctx->futex_cache, IO_NODE_ALLOC_CACHE_MAX, | |
40 | sizeof(struct io_futex_data)); | |
41 | } | |
42 | ||
43 | static void io_futex_cache_entry_free(struct io_cache_entry *entry) | |
44 | { | |
45 | kfree(container_of(entry, struct io_futex_data, cache)); | |
46 | } | |
47 | ||
48 | void io_futex_cache_free(struct io_ring_ctx *ctx) | |
49 | { | |
50 | io_alloc_cache_free(&ctx->futex_cache, io_futex_cache_entry_free); | |
51 | } | |
52 | ||
8f350194 JA |
53 | static void __io_futex_complete(struct io_kiocb *req, struct io_tw_state *ts) |
54 | { | |
55 | req->async_data = NULL; | |
56 | hlist_del_init(&req->hash_node); | |
57 | io_req_task_complete(req, ts); | |
58 | } | |
59 | ||
194bb58c JA |
60 | static void io_futex_complete(struct io_kiocb *req, struct io_tw_state *ts) |
61 | { | |
62 | struct io_futex_data *ifd = req->async_data; | |
63 | struct io_ring_ctx *ctx = req->ctx; | |
64 | ||
65 | io_tw_lock(ctx, ts); | |
66 | if (!io_alloc_cache_put(&ctx->futex_cache, &ifd->cache)) | |
67 | kfree(ifd); | |
8f350194 | 68 | __io_futex_complete(req, ts); |
194bb58c JA |
69 | } |
70 | ||
8f350194 | 71 | static void io_futexv_complete(struct io_kiocb *req, struct io_tw_state *ts) |
194bb58c | 72 | { |
8f350194 JA |
73 | struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex); |
74 | struct futex_vector *futexv = req->async_data; | |
194bb58c | 75 | |
8f350194 JA |
76 | io_tw_lock(req->ctx, ts); |
77 | ||
78 | if (!iof->futexv_unqueued) { | |
79 | int res; | |
80 | ||
81 | res = futex_unqueue_multiple(futexv, iof->futex_nr); | |
82 | if (res != -1) | |
83 | io_req_set_res(req, res, 0); | |
84 | } | |
85 | ||
86 | kfree(req->async_data); | |
87 | req->flags &= ~REQ_F_ASYNC_DATA; | |
88 | __io_futex_complete(req, ts); | |
89 | } | |
90 | ||
91 | static bool io_futexv_claim(struct io_futex *iof) | |
92 | { | |
93 | if (test_bit(0, &iof->futexv_owned) || | |
94 | test_and_set_bit_lock(0, &iof->futexv_owned)) | |
194bb58c | 95 | return false; |
8f350194 JA |
96 | return true; |
97 | } | |
98 | ||
99 | static bool __io_futex_cancel(struct io_ring_ctx *ctx, struct io_kiocb *req) | |
100 | { | |
101 | /* futex wake already done or in progress */ | |
102 | if (req->opcode == IORING_OP_FUTEX_WAIT) { | |
103 | struct io_futex_data *ifd = req->async_data; | |
104 | ||
105 | if (!futex_unqueue(&ifd->q)) | |
106 | return false; | |
107 | req->io_task_work.func = io_futex_complete; | |
108 | } else { | |
109 | struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex); | |
110 | ||
111 | if (!io_futexv_claim(iof)) | |
112 | return false; | |
113 | req->io_task_work.func = io_futexv_complete; | |
114 | } | |
194bb58c JA |
115 | |
116 | hlist_del_init(&req->hash_node); | |
117 | io_req_set_res(req, -ECANCELED, 0); | |
194bb58c JA |
118 | io_req_task_work_add(req); |
119 | return true; | |
120 | } | |
121 | ||
122 | int io_futex_cancel(struct io_ring_ctx *ctx, struct io_cancel_data *cd, | |
123 | unsigned int issue_flags) | |
124 | { | |
125 | struct hlist_node *tmp; | |
126 | struct io_kiocb *req; | |
127 | int nr = 0; | |
128 | ||
129 | if (cd->flags & (IORING_ASYNC_CANCEL_FD|IORING_ASYNC_CANCEL_FD_FIXED)) | |
130 | return -ENOENT; | |
131 | ||
132 | io_ring_submit_lock(ctx, issue_flags); | |
133 | hlist_for_each_entry_safe(req, tmp, &ctx->futex_list, hash_node) { | |
134 | if (req->cqe.user_data != cd->data && | |
135 | !(cd->flags & IORING_ASYNC_CANCEL_ANY)) | |
136 | continue; | |
137 | if (__io_futex_cancel(ctx, req)) | |
138 | nr++; | |
139 | if (!(cd->flags & IORING_ASYNC_CANCEL_ALL)) | |
140 | break; | |
141 | } | |
142 | io_ring_submit_unlock(ctx, issue_flags); | |
143 | ||
144 | if (nr) | |
145 | return nr; | |
146 | ||
147 | return -ENOENT; | |
148 | } | |
149 | ||
150 | bool io_futex_remove_all(struct io_ring_ctx *ctx, struct task_struct *task, | |
151 | bool cancel_all) | |
152 | { | |
153 | struct hlist_node *tmp; | |
154 | struct io_kiocb *req; | |
155 | bool found = false; | |
156 | ||
157 | lockdep_assert_held(&ctx->uring_lock); | |
158 | ||
159 | hlist_for_each_entry_safe(req, tmp, &ctx->futex_list, hash_node) { | |
160 | if (!io_match_task_safe(req, task, cancel_all)) | |
161 | continue; | |
162 | __io_futex_cancel(ctx, req); | |
163 | found = true; | |
164 | } | |
165 | ||
166 | return found; | |
167 | } | |
168 | ||
169 | int io_futex_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
170 | { | |
171 | struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex); | |
172 | u32 flags; | |
173 | ||
174 | if (unlikely(sqe->len || sqe->futex_flags || sqe->buf_index || | |
175 | sqe->file_index)) | |
176 | return -EINVAL; | |
177 | ||
178 | iof->uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
179 | iof->futex_val = READ_ONCE(sqe->addr2); | |
180 | iof->futex_mask = READ_ONCE(sqe->addr3); | |
181 | flags = READ_ONCE(sqe->fd); | |
182 | ||
183 | if (flags & ~FUTEX2_VALID_MASK) | |
184 | return -EINVAL; | |
185 | ||
186 | iof->futex_flags = futex2_to_flags(flags); | |
187 | if (!futex_flags_valid(iof->futex_flags)) | |
188 | return -EINVAL; | |
189 | ||
190 | if (!futex_validate_input(iof->futex_flags, iof->futex_val) || | |
191 | !futex_validate_input(iof->futex_flags, iof->futex_mask)) | |
192 | return -EINVAL; | |
193 | ||
194 | return 0; | |
195 | } | |
196 | ||
8f350194 JA |
197 | static void io_futex_wakev_fn(struct wake_q_head *wake_q, struct futex_q *q) |
198 | { | |
199 | struct io_kiocb *req = q->wake_data; | |
200 | struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex); | |
201 | ||
202 | if (!io_futexv_claim(iof)) | |
203 | return; | |
204 | if (unlikely(!__futex_wake_mark(q))) | |
205 | return; | |
206 | ||
207 | io_req_set_res(req, 0, 0); | |
208 | req->io_task_work.func = io_futexv_complete; | |
209 | io_req_task_work_add(req); | |
210 | } | |
211 | ||
212 | int io_futexv_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) | |
213 | { | |
214 | struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex); | |
215 | struct futex_vector *futexv; | |
216 | int ret; | |
217 | ||
218 | /* No flags or mask supported for waitv */ | |
219 | if (unlikely(sqe->fd || sqe->buf_index || sqe->file_index || | |
220 | sqe->addr2 || sqe->futex_flags || sqe->addr3)) | |
221 | return -EINVAL; | |
222 | ||
223 | iof->uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr)); | |
224 | iof->futex_nr = READ_ONCE(sqe->len); | |
225 | if (!iof->futex_nr || iof->futex_nr > FUTEX_WAITV_MAX) | |
226 | return -EINVAL; | |
227 | ||
228 | futexv = kcalloc(iof->futex_nr, sizeof(*futexv), GFP_KERNEL); | |
229 | if (!futexv) | |
230 | return -ENOMEM; | |
231 | ||
232 | ret = futex_parse_waitv(futexv, iof->uwaitv, iof->futex_nr, | |
233 | io_futex_wakev_fn, req); | |
234 | if (ret) { | |
235 | kfree(futexv); | |
236 | return ret; | |
237 | } | |
238 | ||
239 | iof->futexv_owned = 0; | |
240 | iof->futexv_unqueued = 0; | |
241 | req->flags |= REQ_F_ASYNC_DATA; | |
242 | req->async_data = futexv; | |
243 | return 0; | |
244 | } | |
245 | ||
194bb58c JA |
246 | static void io_futex_wake_fn(struct wake_q_head *wake_q, struct futex_q *q) |
247 | { | |
248 | struct io_futex_data *ifd = container_of(q, struct io_futex_data, q); | |
249 | struct io_kiocb *req = ifd->req; | |
250 | ||
251 | if (unlikely(!__futex_wake_mark(q))) | |
252 | return; | |
253 | ||
254 | io_req_set_res(req, 0, 0); | |
255 | req->io_task_work.func = io_futex_complete; | |
256 | io_req_task_work_add(req); | |
257 | } | |
258 | ||
259 | static struct io_futex_data *io_alloc_ifd(struct io_ring_ctx *ctx) | |
260 | { | |
261 | struct io_cache_entry *entry; | |
262 | ||
263 | entry = io_alloc_cache_get(&ctx->futex_cache); | |
264 | if (entry) | |
265 | return container_of(entry, struct io_futex_data, cache); | |
266 | ||
267 | return kmalloc(sizeof(struct io_futex_data), GFP_NOWAIT); | |
268 | } | |
269 | ||
8f350194 JA |
270 | int io_futexv_wait(struct io_kiocb *req, unsigned int issue_flags) |
271 | { | |
272 | struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex); | |
273 | struct futex_vector *futexv = req->async_data; | |
274 | struct io_ring_ctx *ctx = req->ctx; | |
275 | int ret, woken = -1; | |
276 | ||
277 | io_ring_submit_lock(ctx, issue_flags); | |
278 | ||
279 | ret = futex_wait_multiple_setup(futexv, iof->futex_nr, &woken); | |
280 | ||
281 | /* | |
282 | * Error case, ret is < 0. Mark the request as failed. | |
283 | */ | |
284 | if (unlikely(ret < 0)) { | |
285 | io_ring_submit_unlock(ctx, issue_flags); | |
286 | req_set_fail(req); | |
287 | io_req_set_res(req, ret, 0); | |
288 | kfree(futexv); | |
289 | req->async_data = NULL; | |
290 | req->flags &= ~REQ_F_ASYNC_DATA; | |
291 | return IOU_OK; | |
292 | } | |
293 | ||
294 | /* | |
295 | * 0 return means that we successfully setup the waiters, and that | |
296 | * nobody triggered a wakeup while we were doing so. If the wakeup | |
297 | * happened post setup, the task_work will be run post this issue and | |
298 | * under the submission lock. 1 means We got woken while setting up, | |
299 | * let that side do the completion. Note that | |
300 | * futex_wait_multiple_setup() will have unqueued all the futexes in | |
301 | * this case. Mark us as having done that already, since this is | |
302 | * different from normal wakeup. | |
303 | */ | |
304 | if (!ret) { | |
305 | /* | |
306 | * If futex_wait_multiple_setup() returns 0 for a | |
307 | * successful setup, then the task state will not be | |
308 | * runnable. This is fine for the sync syscall, as | |
309 | * it'll be blocking unless we already got one of the | |
310 | * futexes woken, but it obviously won't work for an | |
311 | * async invocation. Mark us runnable again. | |
312 | */ | |
313 | __set_current_state(TASK_RUNNING); | |
314 | hlist_add_head(&req->hash_node, &ctx->futex_list); | |
315 | } else { | |
316 | iof->futexv_unqueued = 1; | |
317 | if (woken != -1) | |
318 | io_req_set_res(req, woken, 0); | |
319 | } | |
320 | ||
321 | io_ring_submit_unlock(ctx, issue_flags); | |
322 | return IOU_ISSUE_SKIP_COMPLETE; | |
323 | } | |
324 | ||
194bb58c JA |
325 | int io_futex_wait(struct io_kiocb *req, unsigned int issue_flags) |
326 | { | |
327 | struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex); | |
328 | struct io_ring_ctx *ctx = req->ctx; | |
329 | struct io_futex_data *ifd = NULL; | |
330 | struct futex_hash_bucket *hb; | |
331 | int ret; | |
332 | ||
333 | if (!iof->futex_mask) { | |
334 | ret = -EINVAL; | |
335 | goto done; | |
336 | } | |
337 | ||
338 | io_ring_submit_lock(ctx, issue_flags); | |
339 | ifd = io_alloc_ifd(ctx); | |
340 | if (!ifd) { | |
341 | ret = -ENOMEM; | |
342 | goto done_unlock; | |
343 | } | |
344 | ||
345 | req->async_data = ifd; | |
346 | ifd->q = futex_q_init; | |
347 | ifd->q.bitset = iof->futex_mask; | |
348 | ifd->q.wake = io_futex_wake_fn; | |
349 | ifd->req = req; | |
350 | ||
351 | ret = futex_wait_setup(iof->uaddr, iof->futex_val, iof->futex_flags, | |
352 | &ifd->q, &hb); | |
353 | if (!ret) { | |
354 | hlist_add_head(&req->hash_node, &ctx->futex_list); | |
355 | io_ring_submit_unlock(ctx, issue_flags); | |
356 | ||
357 | futex_queue(&ifd->q, hb); | |
358 | return IOU_ISSUE_SKIP_COMPLETE; | |
359 | } | |
360 | ||
361 | done_unlock: | |
362 | io_ring_submit_unlock(ctx, issue_flags); | |
363 | done: | |
364 | if (ret < 0) | |
365 | req_set_fail(req); | |
366 | io_req_set_res(req, ret, 0); | |
367 | kfree(ifd); | |
368 | return IOU_OK; | |
369 | } | |
370 | ||
371 | int io_futex_wake(struct io_kiocb *req, unsigned int issue_flags) | |
372 | { | |
373 | struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex); | |
374 | int ret; | |
375 | ||
376 | /* | |
377 | * Strict flags - ensure that waking 0 futexes yields a 0 result. | |
378 | * See commit 43adf8449510 ("futex: FLAGS_STRICT") for details. | |
379 | */ | |
380 | ret = futex_wake(iof->uaddr, FLAGS_STRICT | iof->futex_flags, | |
381 | iof->futex_val, iof->futex_mask); | |
382 | if (ret < 0) | |
383 | req_set_fail(req); | |
384 | io_req_set_res(req, ret, 0); | |
385 | return IOU_OK; | |
386 | } |