1 /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
2 #ifndef USER_BLK_DRV_CMD_INC_H
3 #define USER_BLK_DRV_CMD_INC_H
4
5 #include <linux/types.h>
6
7 /* ublk server command definition */
8
9 /*
10 * Admin commands, issued by ublk server, and handled by ublk driver.
11 *
12 * Legacy command definition, don't use in new application, and don't
13 * add new such definition any more
14 */
15 #define UBLK_CMD_GET_QUEUE_AFFINITY 0x01
16 #define UBLK_CMD_GET_DEV_INFO 0x02
17 #define UBLK_CMD_ADD_DEV 0x04
18 #define UBLK_CMD_DEL_DEV 0x05
19 #define UBLK_CMD_START_DEV 0x06
20 #define UBLK_CMD_STOP_DEV 0x07
21 #define UBLK_CMD_SET_PARAMS 0x08
22 #define UBLK_CMD_GET_PARAMS 0x09
23 #define UBLK_CMD_START_USER_RECOVERY 0x10
24 #define UBLK_CMD_END_USER_RECOVERY 0x11
25 #define UBLK_CMD_GET_DEV_INFO2 0x12
26
27 /* Any new ctrl command should encode by __IO*() */
28 #define UBLK_U_CMD_GET_QUEUE_AFFINITY \
29 _IOR('u', UBLK_CMD_GET_QUEUE_AFFINITY, struct ublksrv_ctrl_cmd)
30 #define UBLK_U_CMD_GET_DEV_INFO \
31 _IOR('u', UBLK_CMD_GET_DEV_INFO, struct ublksrv_ctrl_cmd)
32 #define UBLK_U_CMD_ADD_DEV \
33 _IOWR('u', UBLK_CMD_ADD_DEV, struct ublksrv_ctrl_cmd)
34 #define UBLK_U_CMD_DEL_DEV \
35 _IOWR('u', UBLK_CMD_DEL_DEV, struct ublksrv_ctrl_cmd)
36 #define UBLK_U_CMD_START_DEV \
37 _IOWR('u', UBLK_CMD_START_DEV, struct ublksrv_ctrl_cmd)
38 #define UBLK_U_CMD_STOP_DEV \
39 _IOWR('u', UBLK_CMD_STOP_DEV, struct ublksrv_ctrl_cmd)
40 #define UBLK_U_CMD_SET_PARAMS \
41 _IOWR('u', UBLK_CMD_SET_PARAMS, struct ublksrv_ctrl_cmd)
42 #define UBLK_U_CMD_GET_PARAMS \
43 _IOR('u', UBLK_CMD_GET_PARAMS, struct ublksrv_ctrl_cmd)
44 #define UBLK_U_CMD_START_USER_RECOVERY \
45 _IOWR('u', UBLK_CMD_START_USER_RECOVERY, struct ublksrv_ctrl_cmd)
46 #define UBLK_U_CMD_END_USER_RECOVERY \
47 _IOWR('u', UBLK_CMD_END_USER_RECOVERY, struct ublksrv_ctrl_cmd)
48 #define UBLK_U_CMD_GET_DEV_INFO2 \
49 _IOR('u', UBLK_CMD_GET_DEV_INFO2, struct ublksrv_ctrl_cmd)
50 #define UBLK_U_CMD_GET_FEATURES \
51 _IOR('u', 0x13, struct ublksrv_ctrl_cmd)
52 #define UBLK_U_CMD_DEL_DEV_ASYNC \
53 _IOR('u', 0x14, struct ublksrv_ctrl_cmd)
54
55 /*
56 * 64bits are enough now, and it should be easy to extend in case of
57 * running out of feature flags
58 */
59 #define UBLK_FEATURES_LEN 8
60
61 /*
62 * IO commands, issued by ublk server, and handled by ublk driver.
63 *
64 * FETCH_REQ: issued via sqe(URING_CMD) beforehand for fetching IO request
65 * from ublk driver, should be issued only when starting device. After
66 * the associated cqe is returned, request's tag can be retrieved via
67 * cqe->userdata.
68 *
69 * COMMIT_AND_FETCH_REQ: issued via sqe(URING_CMD) after ublkserver handled
70 * this IO request, request's handling result is committed to ublk
71 * driver, meantime FETCH_REQ is piggyback, and FETCH_REQ has to be
72 * handled before completing io request.
73 *
74 * NEED_GET_DATA: only used for write requests to set io addr and copy data
75 * When NEED_GET_DATA is set, ublksrv has to issue UBLK_IO_NEED_GET_DATA
76 * command after ublk driver returns UBLK_IO_RES_NEED_GET_DATA.
77 *
78 * It is only used if ublksrv set UBLK_F_NEED_GET_DATA flag
79 * while starting a ublk device.
80 */
81
82 /*
83 * Legacy IO command definition, don't use in new application, and don't
84 * add new such definition any more
85 */
86 #define UBLK_IO_FETCH_REQ 0x20
87 #define UBLK_IO_COMMIT_AND_FETCH_REQ 0x21
88 #define UBLK_IO_NEED_GET_DATA 0x22
89
90 /* Any new IO command should encode by __IOWR() */
91 #define UBLK_U_IO_FETCH_REQ \
92 _IOWR('u', UBLK_IO_FETCH_REQ, struct ublksrv_io_cmd)
93 #define UBLK_U_IO_COMMIT_AND_FETCH_REQ \
94 _IOWR('u', UBLK_IO_COMMIT_AND_FETCH_REQ, struct ublksrv_io_cmd)
95 #define UBLK_U_IO_NEED_GET_DATA \
96 _IOWR('u', UBLK_IO_NEED_GET_DATA, struct ublksrv_io_cmd)
97
98 /* only ABORT means that no re-fetch */
99 #define UBLK_IO_RES_OK 0
100 #define UBLK_IO_RES_NEED_GET_DATA 1
101 #define UBLK_IO_RES_ABORT (-ENODEV)
102
103 #define UBLKSRV_CMD_BUF_OFFSET 0
104 #define UBLKSRV_IO_BUF_OFFSET 0x80000000
105
106 /* tag bit is 16bit, so far limit at most 4096 IOs for each queue */
107 #define UBLK_MAX_QUEUE_DEPTH 4096
108
109 /* single IO buffer max size is 32MB */
110 #define UBLK_IO_BUF_OFF 0
111 #define UBLK_IO_BUF_BITS 25
112 #define UBLK_IO_BUF_BITS_MASK ((1ULL << UBLK_IO_BUF_BITS) - 1)
113
114 /* so at most 64K IOs for each queue */
115 #define UBLK_TAG_OFF UBLK_IO_BUF_BITS
116 #define UBLK_TAG_BITS 16
117 #define UBLK_TAG_BITS_MASK ((1ULL << UBLK_TAG_BITS) - 1)
118
119 /* max 4096 queues */
120 #define UBLK_QID_OFF (UBLK_TAG_OFF + UBLK_TAG_BITS)
121 #define UBLK_QID_BITS 12
122 #define UBLK_QID_BITS_MASK ((1ULL << UBLK_QID_BITS) - 1)
123
124 #define UBLK_MAX_NR_QUEUES (1U << UBLK_QID_BITS)
125
126 #define UBLKSRV_IO_BUF_TOTAL_BITS (UBLK_QID_OFF + UBLK_QID_BITS)
127 #define UBLKSRV_IO_BUF_TOTAL_SIZE (1ULL << UBLKSRV_IO_BUF_TOTAL_BITS)
128
129 /*
130 * zero copy requires 4k block size, and can remap ublk driver's io
131 * request into ublksrv's vm space
132 */
133 #define UBLK_F_SUPPORT_ZERO_COPY (1ULL << 0)
134
135 /*
136 * Force to complete io cmd via io_uring_cmd_complete_in_task so that
137 * performance comparison is done easily with using task_work_add
138 */
139 #define UBLK_F_URING_CMD_COMP_IN_TASK (1ULL << 1)
140
141 /*
142 * User should issue io cmd again for write requests to
143 * set io buffer address and copy data from bio vectors
144 * to the userspace io buffer.
145 *
146 * In this mode, task_work is not used.
147 */
148 #define UBLK_F_NEED_GET_DATA (1UL << 2)
149
150 /*
151 * - Block devices are recoverable if ublk server exits and restarts
152 * - Outstanding I/O when ublk server exits is met with errors
153 * - I/O issued while there is no ublk server queues
154 */
155 #define UBLK_F_USER_RECOVERY (1UL << 3)
156
157 /*
158 * - Block devices are recoverable if ublk server exits and restarts
159 * - Outstanding I/O when ublk server exits is reissued
160 * - I/O issued while there is no ublk server queues
161 */
162 #define UBLK_F_USER_RECOVERY_REISSUE (1UL << 4)
163
164 /*
165 * Unprivileged user can create /dev/ublkcN and /dev/ublkbN.
166 *
167 * /dev/ublk-control needs to be available for unprivileged user, and it
168 * can be done via udev rule to make all control commands available to
169 * unprivileged user. Except for the command of UBLK_CMD_ADD_DEV, all
170 * other commands are only allowed for the owner of the specified device.
171 *
172 * When userspace sends UBLK_CMD_ADD_DEV, the device pair's owner_uid and
173 * owner_gid are stored to ublksrv_ctrl_dev_info by kernel, so far only
174 * the current user's uid/gid is stored, that said owner of the created
175 * device is always the current user.
176 *
177 * We still need udev rule to apply OWNER/GROUP with the stored owner_uid
178 * and owner_gid.
179 *
180 * Then ublk server can be run as unprivileged user, and /dev/ublkbN can
181 * be accessed and managed by its owner represented by owner_uid/owner_gid.
182 */
183 #define UBLK_F_UNPRIVILEGED_DEV (1UL << 5)
184
185 /* use ioctl encoding for uring command */
186 #define UBLK_F_CMD_IOCTL_ENCODE (1UL << 6)
187
188 /*
189 * Copy between request and user buffer by pread()/pwrite()
190 *
191 * Not available for UBLK_F_UNPRIVILEGED_DEV, otherwise userspace may
192 * deceive us by not filling request buffer, then kernel uninitialized
193 * data may be leaked.
194 */
195 #define UBLK_F_USER_COPY (1UL << 7)
196
197 /*
198 * User space sets this flag when setting up the device to request zoned storage support. Kernel may
199 * deny the request by returning an error.
200 */
201 #define UBLK_F_ZONED (1ULL << 8)
202
203 /*
204 * - Block devices are recoverable if ublk server exits and restarts
205 * - Outstanding I/O when ublk server exits is met with errors
206 * - I/O issued while there is no ublk server is met with errors
207 */
208 #define UBLK_F_USER_RECOVERY_FAIL_IO (1ULL << 9)
209
210 /* device state */
211 #define UBLK_S_DEV_DEAD 0
212 #define UBLK_S_DEV_LIVE 1
213 #define UBLK_S_DEV_QUIESCED 2
214 #define UBLK_S_DEV_FAIL_IO 3
215
216 /* shipped via sqe->cmd of io_uring command */
217 struct ublksrv_ctrl_cmd {
218 /* sent to which device, must be valid */
219 __u32 dev_id;
220
221 /* sent to which queue, must be -1 if the cmd isn't for queue */
222 __u16 queue_id;
223 /*
224 * cmd specific buffer, can be IN or OUT.
225 */
226 __u16 len;
227 __u64 addr;
228
229 /* inline data */
230 __u64 data[1];
231
232 /*
233 * Used for UBLK_F_UNPRIVILEGED_DEV and UBLK_CMD_GET_DEV_INFO2
234 * only, include null char
235 */
236 __u16 dev_path_len;
237 __u16 pad;
238 __u32 reserved;
239 };
240
241 struct ublksrv_ctrl_dev_info {
242 __u16 nr_hw_queues;
243 __u16 queue_depth;
244 __u16 state;
245 __u16 pad0;
246
247 __u32 max_io_buf_bytes;
248 __u32 dev_id;
249
250 __s32 ublksrv_pid;
251 __u32 pad1;
252
253 __u64 flags;
254
255 /* For ublksrv internal use, invisible to ublk driver */
256 __u64 ublksrv_flags;
257
258 __u32 owner_uid; /* store by kernel */
259 __u32 owner_gid; /* store by kernel */
260 __u64 reserved1;
261 __u64 reserved2;
262 };
263
264 #define UBLK_IO_OP_READ 0
265 #define UBLK_IO_OP_WRITE 1
266 #define UBLK_IO_OP_FLUSH 2
267 #define UBLK_IO_OP_DISCARD 3
268 #define UBLK_IO_OP_WRITE_SAME 4
269 #define UBLK_IO_OP_WRITE_ZEROES 5
270 #define UBLK_IO_OP_ZONE_OPEN 10
271 #define UBLK_IO_OP_ZONE_CLOSE 11
272 #define UBLK_IO_OP_ZONE_FINISH 12
273 #define UBLK_IO_OP_ZONE_APPEND 13
274 #define UBLK_IO_OP_ZONE_RESET_ALL 14
275 #define UBLK_IO_OP_ZONE_RESET 15
276 /*
277 * Construct a zone report. The report request is carried in `struct
278 * ublksrv_io_desc`. The `start_sector` field must be the first sector of a zone
279 * and shall indicate the first zone of the report. The `nr_zones` shall
280 * indicate how many zones should be reported at most. The report shall be
281 * delivered as a `struct blk_zone` array. To report fewer zones than requested,
282 * zero the last entry of the returned array.
283 *
284 * Related definitions(blk_zone, blk_zone_cond, blk_zone_type, ...) in
285 * include/uapi/linux/blkzoned.h are part of ublk UAPI.
286 */
287 #define UBLK_IO_OP_REPORT_ZONES 18
288
289 #define UBLK_IO_F_FAILFAST_DEV (1U << 8)
290 #define UBLK_IO_F_FAILFAST_TRANSPORT (1U << 9)
291 #define UBLK_IO_F_FAILFAST_DRIVER (1U << 10)
292 #define UBLK_IO_F_META (1U << 11)
293 #define UBLK_IO_F_FUA (1U << 13)
294 #define UBLK_IO_F_NOUNMAP (1U << 15)
295 #define UBLK_IO_F_SWAP (1U << 16)
296
297 /*
298 * io cmd is described by this structure, and stored in share memory, indexed
299 * by request tag.
300 *
301 * The data is stored by ublk driver, and read by ublksrv after one fetch command
302 * returns.
303 */
304 struct ublksrv_io_desc {
305 /* op: bit 0-7, flags: bit 8-31 */
306 __u32 op_flags;
307
308 union {
309 __u32 nr_sectors;
310 __u32 nr_zones; /* for UBLK_IO_OP_REPORT_ZONES */
311 };
312
313 /* start sector for this io */
314 __u64 start_sector;
315
316 /* buffer address in ublksrv daemon vm space, from ublk driver */
317 __u64 addr;
318 };
319
ublksrv_get_op(const struct ublksrv_io_desc * iod)320 static inline __u8 ublksrv_get_op(const struct ublksrv_io_desc *iod)
321 {
322 return iod->op_flags & 0xff;
323 }
324
ublksrv_get_flags(const struct ublksrv_io_desc * iod)325 static inline __u32 ublksrv_get_flags(const struct ublksrv_io_desc *iod)
326 {
327 return iod->op_flags >> 8;
328 }
329
330 /* issued to ublk driver via /dev/ublkcN */
331 struct ublksrv_io_cmd {
332 __u16 q_id;
333
334 /* for fetch/commit which result */
335 __u16 tag;
336
337 /* io result, it is valid for COMMIT* command only */
338 __s32 result;
339
340 union {
341 /*
342 * userspace buffer address in ublksrv daemon process, valid for
343 * FETCH* command only
344 *
345 * `addr` should not be used when UBLK_F_USER_COPY is enabled,
346 * because userspace handles data copy by pread()/pwrite() over
347 * /dev/ublkcN. But in case of UBLK_F_ZONED, this union is
348 * re-used to pass back the allocated LBA for
349 * UBLK_IO_OP_ZONE_APPEND which actually depends on
350 * UBLK_F_USER_COPY
351 */
352 __u64 addr;
353 __u64 zone_append_lba;
354 };
355 };
356
357 struct ublk_param_basic {
358 #define UBLK_ATTR_READ_ONLY (1 << 0)
359 #define UBLK_ATTR_ROTATIONAL (1 << 1)
360 #define UBLK_ATTR_VOLATILE_CACHE (1 << 2)
361 #define UBLK_ATTR_FUA (1 << 3)
362 __u32 attrs;
363 __u8 logical_bs_shift;
364 __u8 physical_bs_shift;
365 __u8 io_opt_shift;
366 __u8 io_min_shift;
367
368 __u32 max_sectors;
369 __u32 chunk_sectors;
370
371 __u64 dev_sectors;
372 __u64 virt_boundary_mask;
373 };
374
375 struct ublk_param_discard {
376 __u32 discard_alignment;
377
378 __u32 discard_granularity;
379 __u32 max_discard_sectors;
380
381 __u32 max_write_zeroes_sectors;
382 __u16 max_discard_segments;
383 __u16 reserved0;
384 };
385
386 /*
387 * read-only, can't set via UBLK_CMD_SET_PARAMS, disk_devt is available
388 * after device is started
389 */
390 struct ublk_param_devt {
391 __u32 char_major;
392 __u32 char_minor;
393 __u32 disk_major;
394 __u32 disk_minor;
395 };
396
397 struct ublk_param_zoned {
398 __u32 max_open_zones;
399 __u32 max_active_zones;
400 __u32 max_zone_append_sectors;
401 __u8 reserved[20];
402 };
403
404 struct ublk_params {
405 /*
406 * Total length of parameters, userspace has to set 'len' for both
407 * SET_PARAMS and GET_PARAMS command, and driver may update len
408 * if two sides use different version of 'ublk_params', same with
409 * 'types' fields.
410 */
411 __u32 len;
412 #define UBLK_PARAM_TYPE_BASIC (1 << 0)
413 #define UBLK_PARAM_TYPE_DISCARD (1 << 1)
414 #define UBLK_PARAM_TYPE_DEVT (1 << 2)
415 #define UBLK_PARAM_TYPE_ZONED (1 << 3)
416 __u32 types; /* types of parameter included */
417
418 struct ublk_param_basic basic;
419 struct ublk_param_discard discard;
420 struct ublk_param_devt devt;
421 struct ublk_param_zoned zoned;
422 };
423
424 #endif
425