1 /* SPDX-License-Identifier: MIT */
2 #define _POSIX_C_SOURCE 200112L
3
4 #include "lib.h"
5 #include "syscall.h"
6 #include "liburing.h"
7 #include "int_flags.h"
8 #include "liburing/compat.h"
9 #include "liburing/io_uring.h"
10 #include "liburing/sanitize.h"
11
do_register(struct io_uring * ring,unsigned int opcode,const void * arg,unsigned int nr_args)12 static inline int do_register(struct io_uring *ring, unsigned int opcode,
13 const void *arg, unsigned int nr_args)
14 {
15 int fd;
16
17 liburing_sanitize_address(arg);
18
19 if (ring->int_flags & INT_FLAG_REG_REG_RING) {
20 opcode |= IORING_REGISTER_USE_REGISTERED_RING;
21 fd = ring->enter_ring_fd;
22 } else {
23 fd = ring->ring_fd;
24 }
25
26 return __sys_io_uring_register(fd, opcode, arg, nr_args);
27 }
28
io_uring_register_buffers_update_tag(struct io_uring * ring,unsigned off,const struct iovec * iovecs,const __u64 * tags,unsigned nr)29 int io_uring_register_buffers_update_tag(struct io_uring *ring, unsigned off,
30 const struct iovec *iovecs,
31 const __u64 *tags,
32 unsigned nr)
33 {
34 liburing_sanitize_iovecs(iovecs, nr);
35
36 struct io_uring_rsrc_update2 up = {
37 .offset = off,
38 .data = (unsigned long)iovecs,
39 .tags = (unsigned long)tags,
40 .nr = nr,
41 };
42
43 return do_register(ring, IORING_REGISTER_BUFFERS_UPDATE, &up, sizeof(up));
44 }
45
io_uring_register_buffers_tags(struct io_uring * ring,const struct iovec * iovecs,const __u64 * tags,unsigned nr)46 int io_uring_register_buffers_tags(struct io_uring *ring,
47 const struct iovec *iovecs,
48 const __u64 *tags,
49 unsigned nr)
50 {
51 liburing_sanitize_iovecs(iovecs, nr);
52
53 struct io_uring_rsrc_register reg = {
54 .nr = nr,
55 .data = (unsigned long)iovecs,
56 .tags = (unsigned long)tags,
57 };
58
59 return do_register(ring, IORING_REGISTER_BUFFERS2, ®, sizeof(reg));
60 }
61
io_uring_register_buffers_sparse(struct io_uring * ring,unsigned nr)62 int io_uring_register_buffers_sparse(struct io_uring *ring, unsigned nr)
63 {
64 struct io_uring_rsrc_register reg = {
65 .flags = IORING_RSRC_REGISTER_SPARSE,
66 .nr = nr,
67 };
68
69 return do_register(ring, IORING_REGISTER_BUFFERS2, ®, sizeof(reg));
70 }
71
io_uring_register_buffers(struct io_uring * ring,const struct iovec * iovecs,unsigned nr_iovecs)72 int io_uring_register_buffers(struct io_uring *ring, const struct iovec *iovecs,
73 unsigned nr_iovecs)
74 {
75 liburing_sanitize_iovecs(iovecs, nr_iovecs);
76
77 return do_register(ring, IORING_REGISTER_BUFFERS, iovecs, nr_iovecs);
78 }
79
io_uring_unregister_buffers(struct io_uring * ring)80 int io_uring_unregister_buffers(struct io_uring *ring)
81 {
82 return do_register(ring, IORING_UNREGISTER_BUFFERS, NULL, 0);
83 }
84
io_uring_register_files_update_tag(struct io_uring * ring,unsigned off,const int * files,const __u64 * tags,unsigned nr_files)85 int io_uring_register_files_update_tag(struct io_uring *ring, unsigned off,
86 const int *files, const __u64 *tags,
87 unsigned nr_files)
88 {
89 liburing_sanitize_address(files);
90 liburing_sanitize_address(tags);
91
92 struct io_uring_rsrc_update2 up = {
93 .offset = off,
94 .data = (unsigned long)files,
95 .tags = (unsigned long)tags,
96 .nr = nr_files,
97 };
98
99 return do_register(ring, IORING_REGISTER_FILES_UPDATE2, &up, sizeof(up));
100 }
101
102 /*
103 * Register an update for an existing file set. The updates will start at
104 * 'off' in the original array, and 'nr_files' is the number of files we'll
105 * update.
106 *
107 * Returns number of files updated on success, -ERROR on failure.
108 */
io_uring_register_files_update(struct io_uring * ring,unsigned off,const int * files,unsigned nr_files)109 int io_uring_register_files_update(struct io_uring *ring, unsigned off,
110 const int *files, unsigned nr_files)
111 {
112 liburing_sanitize_address(files);
113
114 struct io_uring_files_update up = {
115 .offset = off,
116 .fds = (unsigned long) files,
117 };
118
119 return do_register(ring, IORING_REGISTER_FILES_UPDATE, &up, nr_files);
120 }
121
increase_rlimit_nofile(unsigned nr)122 static int increase_rlimit_nofile(unsigned nr)
123 {
124 int ret;
125 struct rlimit rlim;
126
127 ret = __sys_getrlimit(RLIMIT_NOFILE, &rlim);
128 if (ret < 0)
129 return ret;
130
131 if (rlim.rlim_cur < nr) {
132 rlim.rlim_cur += nr;
133 __sys_setrlimit(RLIMIT_NOFILE, &rlim);
134 }
135
136 return 0;
137 }
138
io_uring_register_files_sparse(struct io_uring * ring,unsigned nr)139 int io_uring_register_files_sparse(struct io_uring *ring, unsigned nr)
140 {
141 struct io_uring_rsrc_register reg = {
142 .flags = IORING_RSRC_REGISTER_SPARSE,
143 .nr = nr,
144 };
145 int ret, did_increase = 0;
146
147 do {
148 ret = do_register(ring, IORING_REGISTER_FILES2, ®, sizeof(reg));
149 if (ret >= 0)
150 break;
151 if (ret == -EMFILE && !did_increase) {
152 did_increase = 1;
153 increase_rlimit_nofile(nr);
154 continue;
155 }
156 break;
157 } while (1);
158
159 return ret;
160 }
161
io_uring_register_files_tags(struct io_uring * ring,const int * files,const __u64 * tags,unsigned nr)162 int io_uring_register_files_tags(struct io_uring *ring, const int *files,
163 const __u64 *tags, unsigned nr)
164 {
165 liburing_sanitize_address(files);
166 liburing_sanitize_address(tags);
167
168 struct io_uring_rsrc_register reg = {
169 .nr = nr,
170 .data = (unsigned long)files,
171 .tags = (unsigned long)tags,
172 };
173 int ret, did_increase = 0;
174
175 do {
176 ret = do_register(ring, IORING_REGISTER_FILES2, ®, sizeof(reg));
177 if (ret >= 0)
178 break;
179 if (ret == -EMFILE && !did_increase) {
180 did_increase = 1;
181 increase_rlimit_nofile(nr);
182 continue;
183 }
184 break;
185 } while (1);
186
187 return ret;
188 }
189
io_uring_register_files(struct io_uring * ring,const int * files,unsigned nr_files)190 int io_uring_register_files(struct io_uring *ring, const int *files,
191 unsigned nr_files)
192 {
193 int ret, did_increase = 0;
194
195 liburing_sanitize_address(files);
196
197 do {
198 ret = do_register(ring, IORING_REGISTER_FILES, files, nr_files);
199 if (ret >= 0)
200 break;
201 if (ret == -EMFILE && !did_increase) {
202 did_increase = 1;
203 increase_rlimit_nofile(nr_files);
204 continue;
205 }
206 break;
207 } while (1);
208
209 return ret;
210 }
211
io_uring_unregister_files(struct io_uring * ring)212 int io_uring_unregister_files(struct io_uring *ring)
213 {
214 return do_register(ring, IORING_UNREGISTER_FILES, NULL, 0);
215 }
216
io_uring_register_eventfd(struct io_uring * ring,int event_fd)217 int io_uring_register_eventfd(struct io_uring *ring, int event_fd)
218 {
219 return do_register(ring, IORING_REGISTER_EVENTFD, &event_fd, 1);
220 }
221
io_uring_unregister_eventfd(struct io_uring * ring)222 int io_uring_unregister_eventfd(struct io_uring *ring)
223 {
224 return do_register(ring, IORING_UNREGISTER_EVENTFD, NULL, 0);
225 }
226
io_uring_register_eventfd_async(struct io_uring * ring,int event_fd)227 int io_uring_register_eventfd_async(struct io_uring *ring, int event_fd)
228 {
229 return do_register(ring, IORING_REGISTER_EVENTFD_ASYNC, &event_fd, 1);
230 }
231
io_uring_register_probe(struct io_uring * ring,struct io_uring_probe * p,unsigned int nr_ops)232 int io_uring_register_probe(struct io_uring *ring, struct io_uring_probe *p,
233 unsigned int nr_ops)
234 {
235 return do_register(ring, IORING_REGISTER_PROBE, p, nr_ops);
236 }
237
io_uring_register_personality(struct io_uring * ring)238 int io_uring_register_personality(struct io_uring *ring)
239 {
240 return do_register(ring, IORING_REGISTER_PERSONALITY, NULL, 0);
241 }
242
io_uring_unregister_personality(struct io_uring * ring,int id)243 int io_uring_unregister_personality(struct io_uring *ring, int id)
244 {
245 return do_register(ring, IORING_UNREGISTER_PERSONALITY, NULL, id);
246 }
247
io_uring_register_restrictions(struct io_uring * ring,struct io_uring_restriction * res,unsigned int nr_res)248 int io_uring_register_restrictions(struct io_uring *ring,
249 struct io_uring_restriction *res,
250 unsigned int nr_res)
251 {
252 return do_register(ring, IORING_REGISTER_RESTRICTIONS, res, nr_res);
253 }
254
io_uring_enable_rings(struct io_uring * ring)255 int io_uring_enable_rings(struct io_uring *ring)
256 {
257 return do_register(ring, IORING_REGISTER_ENABLE_RINGS, NULL, 0);
258 }
259
io_uring_register_iowq_aff(struct io_uring * ring,size_t cpusz,const cpu_set_t * mask)260 int io_uring_register_iowq_aff(struct io_uring *ring, size_t cpusz,
261 const cpu_set_t *mask)
262 {
263 if (cpusz >= (1U << 31))
264 return -EINVAL;
265
266 return do_register(ring, IORING_REGISTER_IOWQ_AFF, mask, (int) cpusz);
267 }
268
io_uring_unregister_iowq_aff(struct io_uring * ring)269 int io_uring_unregister_iowq_aff(struct io_uring *ring)
270 {
271 return do_register(ring, IORING_UNREGISTER_IOWQ_AFF, NULL, 0);
272 }
273
io_uring_register_iowq_max_workers(struct io_uring * ring,unsigned int * val)274 int io_uring_register_iowq_max_workers(struct io_uring *ring, unsigned int *val)
275 {
276 return do_register(ring, IORING_REGISTER_IOWQ_MAX_WORKERS, val, 2);
277 }
278
io_uring_register_ring_fd(struct io_uring * ring)279 int io_uring_register_ring_fd(struct io_uring *ring)
280 {
281 struct io_uring_rsrc_update up = {
282 .data = ring->ring_fd,
283 .offset = -1U,
284 };
285 int ret;
286
287 if (ring->int_flags & INT_FLAG_REG_RING)
288 return -EEXIST;
289
290 ret = do_register(ring, IORING_REGISTER_RING_FDS, &up, 1);
291 if (ret == 1) {
292 ring->enter_ring_fd = up.offset;
293 ring->int_flags |= INT_FLAG_REG_RING;
294 if (ring->features & IORING_FEAT_REG_REG_RING) {
295 ring->int_flags |= INT_FLAG_REG_REG_RING;
296 }
297 }
298 return ret;
299 }
300
301
io_uring_unregister_ring_fd(struct io_uring * ring)302 int io_uring_unregister_ring_fd(struct io_uring *ring)
303 {
304 struct io_uring_rsrc_update up = {
305 .offset = ring->enter_ring_fd,
306 };
307 int ret;
308
309 if (!(ring->int_flags & INT_FLAG_REG_RING))
310 return -EINVAL;
311
312 ret = do_register(ring, IORING_UNREGISTER_RING_FDS, &up, 1);
313 if (ret == 1) {
314 ring->enter_ring_fd = ring->ring_fd;
315 ring->int_flags &= ~(INT_FLAG_REG_RING | INT_FLAG_REG_REG_RING);
316 }
317 return ret;
318 }
319
io_uring_close_ring_fd(struct io_uring * ring)320 int io_uring_close_ring_fd(struct io_uring *ring)
321 {
322 if (!(ring->features & IORING_FEAT_REG_REG_RING))
323 return -EOPNOTSUPP;
324 if (!(ring->int_flags & INT_FLAG_REG_RING))
325 return -EINVAL;
326 if (ring->ring_fd == -1)
327 return -EBADF;
328
329 __sys_close(ring->ring_fd);
330 ring->ring_fd = -1;
331 return 1;
332 }
333
io_uring_register_buf_ring(struct io_uring * ring,struct io_uring_buf_reg * reg,unsigned int __maybe_unused flags)334 int io_uring_register_buf_ring(struct io_uring *ring,
335 struct io_uring_buf_reg *reg,
336 unsigned int __maybe_unused flags)
337 {
338 reg->flags |= flags;
339 return do_register(ring, IORING_REGISTER_PBUF_RING, reg, 1);
340 }
341
io_uring_unregister_buf_ring(struct io_uring * ring,int bgid)342 int io_uring_unregister_buf_ring(struct io_uring *ring, int bgid)
343 {
344 struct io_uring_buf_reg reg = { .bgid = bgid };
345
346 return do_register(ring, IORING_UNREGISTER_PBUF_RING, ®, 1);
347 }
348
io_uring_buf_ring_head(struct io_uring * ring,int buf_group,uint16_t * head)349 int io_uring_buf_ring_head(struct io_uring *ring, int buf_group, uint16_t *head)
350 {
351 liburing_sanitize_address(head);
352
353 struct io_uring_buf_status buf_status = {
354 .buf_group = buf_group,
355 };
356 int ret;
357
358 ret = do_register(ring, IORING_REGISTER_PBUF_STATUS, &buf_status, 1);
359 if (ret)
360 return ret;
361 *head = buf_status.head;
362 return 0;
363 }
364
io_uring_register_sync_cancel(struct io_uring * ring,struct io_uring_sync_cancel_reg * reg)365 int io_uring_register_sync_cancel(struct io_uring *ring,
366 struct io_uring_sync_cancel_reg *reg)
367 {
368 return do_register(ring, IORING_REGISTER_SYNC_CANCEL, reg, 1);
369 }
370
io_uring_register_file_alloc_range(struct io_uring * ring,unsigned off,unsigned len)371 int io_uring_register_file_alloc_range(struct io_uring *ring,
372 unsigned off, unsigned len)
373 {
374 struct io_uring_file_index_range range = {
375 .off = off,
376 .len = len
377 };
378
379 return do_register(ring, IORING_REGISTER_FILE_ALLOC_RANGE, &range, 0);
380 }
381
io_uring_register_napi(struct io_uring * ring,struct io_uring_napi * napi)382 int io_uring_register_napi(struct io_uring *ring, struct io_uring_napi *napi)
383 {
384 return do_register(ring, IORING_REGISTER_NAPI, napi, 1);
385 }
386
io_uring_unregister_napi(struct io_uring * ring,struct io_uring_napi * napi)387 int io_uring_unregister_napi(struct io_uring *ring, struct io_uring_napi *napi)
388 {
389 return do_register(ring, IORING_UNREGISTER_NAPI, napi, 1);
390 }
391
io_uring_register_clock(struct io_uring * ring,struct io_uring_clock_register * arg)392 int io_uring_register_clock(struct io_uring *ring,
393 struct io_uring_clock_register *arg)
394 {
395 return do_register(ring, IORING_REGISTER_CLOCK, arg, 0);
396 }
397
io_uring_clone_buffers(struct io_uring * dst,struct io_uring * src)398 int io_uring_clone_buffers(struct io_uring *dst, struct io_uring *src)
399 {
400 struct io_uring_clone_buffers buf = { .src_fd = src->ring_fd, };
401
402 if (src->int_flags & INT_FLAG_REG_REG_RING) {
403 buf.src_fd = src->enter_ring_fd;
404 buf.flags = IORING_REGISTER_SRC_REGISTERED;
405 } else {
406 buf.src_fd = src->ring_fd;
407 }
408
409 return do_register(dst, IORING_REGISTER_CLONE_BUFFERS, &buf, 1);
410 }
411