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