Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
a3603a6e AG |
2 | #ifndef __DRBD_PROTOCOL_H |
3 | #define __DRBD_PROTOCOL_H | |
4 | ||
5 | enum drbd_packet { | |
6 | /* receiver (data socket) */ | |
7 | P_DATA = 0x00, | |
8 | P_DATA_REPLY = 0x01, /* Response to P_DATA_REQUEST */ | |
9 | P_RS_DATA_REPLY = 0x02, /* Response to P_RS_DATA_REQUEST */ | |
10 | P_BARRIER = 0x03, | |
11 | P_BITMAP = 0x04, | |
12 | P_BECOME_SYNC_TARGET = 0x05, | |
13 | P_BECOME_SYNC_SOURCE = 0x06, | |
14 | P_UNPLUG_REMOTE = 0x07, /* Used at various times to hint the peer */ | |
15 | P_DATA_REQUEST = 0x08, /* Used to ask for a data block */ | |
16 | P_RS_DATA_REQUEST = 0x09, /* Used to ask for a data block for resync */ | |
17 | P_SYNC_PARAM = 0x0a, | |
18 | P_PROTOCOL = 0x0b, | |
19 | P_UUIDS = 0x0c, | |
20 | P_SIZES = 0x0d, | |
21 | P_STATE = 0x0e, | |
22 | P_SYNC_UUID = 0x0f, | |
23 | P_AUTH_CHALLENGE = 0x10, | |
24 | P_AUTH_RESPONSE = 0x11, | |
25 | P_STATE_CHG_REQ = 0x12, | |
26 | ||
668700b4 | 27 | /* (meta socket) */ |
a3603a6e AG |
28 | P_PING = 0x13, |
29 | P_PING_ACK = 0x14, | |
30 | P_RECV_ACK = 0x15, /* Used in protocol B */ | |
31 | P_WRITE_ACK = 0x16, /* Used in protocol C */ | |
32 | P_RS_WRITE_ACK = 0x17, /* Is a P_WRITE_ACK, additionally call set_in_sync(). */ | |
33 | P_SUPERSEDED = 0x18, /* Used in proto C, two-primaries conflict detection */ | |
34 | P_NEG_ACK = 0x19, /* Sent if local disk is unusable */ | |
35 | P_NEG_DREPLY = 0x1a, /* Local disk is broken... */ | |
36 | P_NEG_RS_DREPLY = 0x1b, /* Local disk is broken... */ | |
37 | P_BARRIER_ACK = 0x1c, | |
38 | P_STATE_CHG_REPLY = 0x1d, | |
39 | ||
40 | /* "new" commands, no longer fitting into the ordering scheme above */ | |
41 | ||
42 | P_OV_REQUEST = 0x1e, /* data socket */ | |
43 | P_OV_REPLY = 0x1f, | |
44 | P_OV_RESULT = 0x20, /* meta socket */ | |
45 | P_CSUM_RS_REQUEST = 0x21, /* data socket */ | |
46 | P_RS_IS_IN_SYNC = 0x22, /* meta socket */ | |
47 | P_SYNC_PARAM89 = 0x23, /* data socket, protocol version 89 replacement for P_SYNC_PARAM */ | |
48 | P_COMPRESSED_BITMAP = 0x24, /* compressed or otherwise encoded bitmap transfer */ | |
49 | /* P_CKPT_FENCE_REQ = 0x25, * currently reserved for protocol D */ | |
50 | /* P_CKPT_DISABLE_REQ = 0x26, * currently reserved for protocol D */ | |
51 | P_DELAY_PROBE = 0x27, /* is used on BOTH sockets */ | |
52 | P_OUT_OF_SYNC = 0x28, /* Mark as out of sync (Outrunning), data socket */ | |
53 | P_RS_CANCEL = 0x29, /* meta: Used to cancel RS_DATA_REQUEST packet by SyncSource */ | |
54 | P_CONN_ST_CHG_REQ = 0x2a, /* data sock: Connection wide state request */ | |
55 | P_CONN_ST_CHG_REPLY = 0x2b, /* meta sock: Connection side state req reply */ | |
56 | P_RETRY_WRITE = 0x2c, /* Protocol C: retry conflicting write request */ | |
57 | P_PROTOCOL_UPDATE = 0x2d, /* data sock: is used in established connections */ | |
a0fb3c47 LE |
58 | /* 0x2e to 0x30 reserved, used in drbd 9 */ |
59 | ||
60 | /* REQ_DISCARD. We used "discard" in different contexts before, | |
61 | * which is why I chose TRIM here, to disambiguate. */ | |
62 | P_TRIM = 0x31, | |
a3603a6e | 63 | |
700ca8c0 PR |
64 | /* Only use these two if both support FF_THIN_RESYNC */ |
65 | P_RS_THIN_REQ = 0x32, /* Request a block for resync or reply P_RS_DEALLOCATED */ | |
66 | P_RS_DEALLOCATED = 0x33, /* Contains only zeros on sync source node */ | |
67 | ||
9104d31a LE |
68 | /* REQ_WRITE_SAME. |
69 | * On a receiving side without REQ_WRITE_SAME, | |
70 | * we may fall back to an opencoded loop instead. */ | |
71 | P_WSAME = 0x34, | |
72 | ||
a3603a6e AG |
73 | P_MAY_IGNORE = 0x100, /* Flag to test if (cmd > P_MAY_IGNORE) ... */ |
74 | P_MAX_OPT_CMD = 0x101, | |
75 | ||
76 | /* special command ids for handshake */ | |
77 | ||
78 | P_INITIAL_META = 0xfff1, /* First Packet on the MetaSock */ | |
79 | P_INITIAL_DATA = 0xfff2, /* First Packet on the Socket */ | |
80 | ||
81 | P_CONNECTION_FEATURES = 0xfffe /* FIXED for the next century! */ | |
82 | }; | |
83 | ||
84 | #ifndef __packed | |
85 | #define __packed __attribute__((packed)) | |
86 | #endif | |
87 | ||
88 | /* This is the layout for a packet on the wire. | |
89 | * The byteorder is the network byte order. | |
90 | * (except block_id and barrier fields. | |
91 | * these are pointers to local structs | |
92 | * and have no relevance for the partner, | |
93 | * which just echoes them as received.) | |
94 | * | |
95 | * NOTE that the payload starts at a long aligned offset, | |
96 | * regardless of 32 or 64 bit arch! | |
97 | */ | |
98 | struct p_header80 { | |
99 | u32 magic; | |
100 | u16 command; | |
101 | u16 length; /* bytes of data after this header */ | |
102 | } __packed; | |
103 | ||
104 | /* Header for big packets, Used for data packets exceeding 64kB */ | |
105 | struct p_header95 { | |
106 | u16 magic; /* use DRBD_MAGIC_BIG here */ | |
107 | u16 command; | |
108 | u32 length; | |
109 | } __packed; | |
110 | ||
111 | struct p_header100 { | |
112 | u32 magic; | |
113 | u16 volume; | |
114 | u16 command; | |
115 | u32 length; | |
116 | u32 pad; | |
117 | } __packed; | |
118 | ||
9104d31a LE |
119 | /* These defines must not be changed without changing the protocol version. |
120 | * New defines may only be introduced together with protocol version bump or | |
121 | * new protocol feature flags. | |
122 | */ | |
123 | #define DP_HARDBARRIER 1 /* no longer used */ | |
a3603a6e AG |
124 | #define DP_RW_SYNC 2 /* equals REQ_SYNC */ |
125 | #define DP_MAY_SET_IN_SYNC 4 | |
126 | #define DP_UNPLUG 8 /* not used anymore */ | |
127 | #define DP_FUA 16 /* equals REQ_FUA */ | |
28a8f0d3 | 128 | #define DP_FLUSH 32 /* equals REQ_PREFLUSH */ |
a3603a6e AG |
129 | #define DP_DISCARD 64 /* equals REQ_DISCARD */ |
130 | #define DP_SEND_RECEIVE_ACK 128 /* This is a proto B write request */ | |
131 | #define DP_SEND_WRITE_ACK 256 /* This is a proto C write request */ | |
9104d31a | 132 | #define DP_WSAME 512 /* equiv. REQ_WRITE_SAME */ |
a3603a6e AG |
133 | |
134 | struct p_data { | |
135 | u64 sector; /* 64 bits sector number */ | |
136 | u64 block_id; /* to identify the request in protocol B&C */ | |
137 | u32 seq_num; | |
138 | u32 dp_flags; | |
139 | } __packed; | |
140 | ||
a0fb3c47 LE |
141 | struct p_trim { |
142 | struct p_data p_data; | |
143 | u32 size; /* == bio->bi_size */ | |
144 | } __packed; | |
145 | ||
9104d31a LE |
146 | struct p_wsame { |
147 | struct p_data p_data; | |
148 | u32 size; /* == bio->bi_size */ | |
149 | } __packed; | |
150 | ||
a3603a6e AG |
151 | /* |
152 | * commands which share a struct: | |
153 | * p_block_ack: | |
154 | * P_RECV_ACK (proto B), P_WRITE_ACK (proto C), | |
155 | * P_SUPERSEDED (proto C, two-primaries conflict detection) | |
156 | * p_block_req: | |
157 | * P_DATA_REQUEST, P_RS_DATA_REQUEST | |
158 | */ | |
159 | struct p_block_ack { | |
160 | u64 sector; | |
161 | u64 block_id; | |
162 | u32 blksize; | |
163 | u32 seq_num; | |
164 | } __packed; | |
165 | ||
166 | struct p_block_req { | |
167 | u64 sector; | |
168 | u64 block_id; | |
169 | u32 blksize; | |
170 | u32 pad; /* to multiple of 8 Byte */ | |
171 | } __packed; | |
172 | ||
173 | /* | |
174 | * commands with their own struct for additional fields: | |
175 | * P_CONNECTION_FEATURES | |
176 | * P_BARRIER | |
177 | * P_BARRIER_ACK | |
178 | * P_SYNC_PARAM | |
179 | * ReportParams | |
180 | */ | |
181 | ||
9104d31a LE |
182 | /* supports TRIM/DISCARD on the "wire" protocol */ |
183 | #define DRBD_FF_TRIM 1 | |
184 | ||
185 | /* Detect all-zeros during resync, and rather TRIM/UNMAP/DISCARD those blocks | |
186 | * instead of fully allocate a supposedly thin volume on initial resync */ | |
187 | #define DRBD_FF_THIN_RESYNC 2 | |
188 | ||
189 | /* supports REQ_WRITE_SAME on the "wire" protocol. | |
190 | * Note: this flag is overloaded, | |
191 | * its presence also | |
192 | * - indicates support for 128 MiB "batch bios", | |
193 | * max discard size of 128 MiB | |
194 | * instead of 4M before that. | |
195 | * - indicates that we exchange additional settings in p_sizes | |
196 | * drbd_send_sizes()/receive_sizes() | |
197 | */ | |
198 | #define DRBD_FF_WSAME 4 | |
20c68fde | 199 | |
a3603a6e AG |
200 | struct p_connection_features { |
201 | u32 protocol_min; | |
202 | u32 feature_flags; | |
203 | u32 protocol_max; | |
204 | ||
205 | /* should be more than enough for future enhancements | |
206 | * for now, feature_flags and the reserved array shall be zero. | |
207 | */ | |
208 | ||
209 | u32 _pad; | |
210 | u64 reserved[7]; | |
211 | } __packed; | |
212 | ||
213 | struct p_barrier { | |
214 | u32 barrier; /* barrier number _handle_ only */ | |
215 | u32 pad; /* to multiple of 8 Byte */ | |
216 | } __packed; | |
217 | ||
218 | struct p_barrier_ack { | |
219 | u32 barrier; | |
220 | u32 set_size; | |
221 | } __packed; | |
222 | ||
223 | struct p_rs_param { | |
224 | u32 resync_rate; | |
225 | ||
226 | /* Since protocol version 88 and higher. */ | |
227 | char verify_alg[0]; | |
228 | } __packed; | |
229 | ||
230 | struct p_rs_param_89 { | |
231 | u32 resync_rate; | |
232 | /* protocol version 89: */ | |
233 | char verify_alg[SHARED_SECRET_MAX]; | |
234 | char csums_alg[SHARED_SECRET_MAX]; | |
235 | } __packed; | |
236 | ||
237 | struct p_rs_param_95 { | |
238 | u32 resync_rate; | |
239 | char verify_alg[SHARED_SECRET_MAX]; | |
240 | char csums_alg[SHARED_SECRET_MAX]; | |
241 | u32 c_plan_ahead; | |
242 | u32 c_delay_target; | |
243 | u32 c_fill_target; | |
244 | u32 c_max_rate; | |
245 | } __packed; | |
246 | ||
247 | enum drbd_conn_flags { | |
248 | CF_DISCARD_MY_DATA = 1, | |
249 | CF_DRY_RUN = 2, | |
250 | }; | |
251 | ||
252 | struct p_protocol { | |
253 | u32 protocol; | |
254 | u32 after_sb_0p; | |
255 | u32 after_sb_1p; | |
256 | u32 after_sb_2p; | |
257 | u32 conn_flags; | |
258 | u32 two_primaries; | |
259 | ||
260 | /* Since protocol version 87 and higher. */ | |
261 | char integrity_alg[0]; | |
262 | ||
263 | } __packed; | |
264 | ||
265 | struct p_uuids { | |
266 | u64 uuid[UI_EXTENDED_SIZE]; | |
267 | } __packed; | |
268 | ||
269 | struct p_rs_uuid { | |
270 | u64 uuid; | |
271 | } __packed; | |
272 | ||
9104d31a LE |
273 | /* optional queue_limits if (agreed_features & DRBD_FF_WSAME) |
274 | * see also struct queue_limits, as of late 2015 */ | |
275 | struct o_qlim { | |
276 | /* we don't need it yet, but we may as well communicate it now */ | |
277 | u32 physical_block_size; | |
278 | ||
279 | /* so the original in struct queue_limits is unsigned short, | |
280 | * but I'd have to put in padding anyways. */ | |
281 | u32 logical_block_size; | |
282 | ||
283 | /* One incoming bio becomes one DRBD request, | |
284 | * which may be translated to several bio on the receiving side. | |
285 | * We don't need to communicate chunk/boundary/segment ... limits. | |
286 | */ | |
287 | ||
288 | /* various IO hints may be useful with "diskless client" setups */ | |
289 | u32 alignment_offset; | |
290 | u32 io_min; | |
291 | u32 io_opt; | |
292 | ||
293 | /* We may need to communicate integrity stuff at some point, | |
294 | * but let's not get ahead of ourselves. */ | |
295 | ||
296 | /* Backend discard capabilities. | |
297 | * Receiving side uses "blkdev_issue_discard()", no need to communicate | |
298 | * more specifics. If the backend cannot do discards, the DRBD peer | |
299 | * may fall back to blkdev_issue_zeroout(). | |
300 | */ | |
301 | u8 discard_enabled; | |
302 | u8 discard_zeroes_data; | |
303 | u8 write_same_capable; | |
304 | u8 _pad; | |
305 | } __packed; | |
306 | ||
a3603a6e AG |
307 | struct p_sizes { |
308 | u64 d_size; /* size of disk */ | |
309 | u64 u_size; /* user requested size */ | |
310 | u64 c_size; /* current exported size */ | |
311 | u32 max_bio_size; /* Maximal size of a BIO */ | |
312 | u16 queue_order_type; /* not yet implemented in DRBD*/ | |
313 | u16 dds_flags; /* use enum dds_flags here. */ | |
9104d31a LE |
314 | |
315 | /* optional queue_limits if (agreed_features & DRBD_FF_WSAME) */ | |
316 | struct o_qlim qlim[0]; | |
a3603a6e AG |
317 | } __packed; |
318 | ||
319 | struct p_state { | |
320 | u32 state; | |
321 | } __packed; | |
322 | ||
323 | struct p_req_state { | |
324 | u32 mask; | |
325 | u32 val; | |
326 | } __packed; | |
327 | ||
328 | struct p_req_state_reply { | |
329 | u32 retcode; | |
330 | } __packed; | |
331 | ||
332 | struct p_drbd06_param { | |
333 | u64 size; | |
334 | u32 state; | |
335 | u32 blksize; | |
336 | u32 protocol; | |
337 | u32 version; | |
338 | u32 gen_cnt[5]; | |
339 | u32 bit_map_gen[5]; | |
340 | } __packed; | |
341 | ||
342 | struct p_block_desc { | |
343 | u64 sector; | |
344 | u32 blksize; | |
345 | u32 pad; /* to multiple of 8 Byte */ | |
346 | } __packed; | |
347 | ||
348 | /* Valid values for the encoding field. | |
349 | * Bump proto version when changing this. */ | |
350 | enum drbd_bitmap_code { | |
351 | /* RLE_VLI_Bytes = 0, | |
352 | * and other bit variants had been defined during | |
353 | * algorithm evaluation. */ | |
354 | RLE_VLI_Bits = 2, | |
355 | }; | |
356 | ||
357 | struct p_compressed_bm { | |
358 | /* (encoding & 0x0f): actual encoding, see enum drbd_bitmap_code | |
359 | * (encoding & 0x80): polarity (set/unset) of first runlength | |
360 | * ((encoding >> 4) & 0x07): pad_bits, number of trailing zero bits | |
361 | * used to pad up to head.length bytes | |
362 | */ | |
363 | u8 encoding; | |
364 | ||
365 | u8 code[0]; | |
366 | } __packed; | |
367 | ||
368 | struct p_delay_probe93 { | |
369 | u32 seq_num; /* sequence number to match the two probe packets */ | |
370 | u32 offset; /* usecs the probe got sent after the reference time point */ | |
371 | } __packed; | |
372 | ||
373 | /* | |
374 | * Bitmap packets need to fit within a single page on the sender and receiver, | |
375 | * so we are limited to 4 KiB (and not to PAGE_SIZE, which can be bigger). | |
376 | */ | |
377 | #define DRBD_SOCKET_BUFFER_SIZE 4096 | |
378 | ||
379 | #endif /* __DRBD_PROTOCOL_H */ |