1 | /* SPDX-License-Identifier: GPL-2.0-only */ |
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 | |
27 | /* (meta socket) */ |
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 */ |
58 | /* 0x2e to 0x30 reserved, used in drbd 9 */ |
59 | |
60 | /* REQ_OP_DISCARD. We used "discard" in different contexts before, |
61 | * which is why I chose TRIM here, to disambiguate. */ |
62 | P_TRIM = 0x31, |
63 | |
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 | |
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 | |
73 | /* 0x35 already claimed in DRBD 9 */ |
74 | P_ZEROES = 0x36, /* data sock: zero-out, WRITE_ZEROES */ |
75 | |
76 | /* 0x40 .. 0x48 already claimed in DRBD 9 */ |
77 | |
78 | P_MAY_IGNORE = 0x100, /* Flag to test if (cmd > P_MAY_IGNORE) ... */ |
79 | P_MAX_OPT_CMD = 0x101, |
80 | |
81 | /* special command ids for handshake */ |
82 | |
83 | P_INITIAL_META = 0xfff1, /* First Packet on the MetaSock */ |
84 | P_INITIAL_DATA = 0xfff2, /* First Packet on the Socket */ |
85 | |
86 | P_CONNECTION_FEATURES = 0xfffe /* FIXED for the next century! */ |
87 | }; |
88 | |
89 | #ifndef __packed |
90 | #define __packed __attribute__((packed)) |
91 | #endif |
92 | |
93 | /* This is the layout for a packet on the wire. |
94 | * The byteorder is the network byte order. |
95 | * (except block_id and barrier fields. |
96 | * these are pointers to local structs |
97 | * and have no relevance for the partner, |
98 | * which just echoes them as received.) |
99 | * |
100 | * NOTE that the payload starts at a long aligned offset, |
101 | * regardless of 32 or 64 bit arch! |
102 | */ |
103 | struct { |
104 | u32 ; |
105 | u16 command; |
106 | u16 ; /* bytes of data after this header */ |
107 | } __packed; |
108 | |
109 | /* Header for big packets, Used for data packets exceeding 64kB */ |
110 | struct { |
111 | u16 ; /* use DRBD_MAGIC_BIG here */ |
112 | u16 command; |
113 | u32 ; |
114 | } __packed; |
115 | |
116 | struct { |
117 | u32 ; |
118 | u16 ; |
119 | u16 command; |
120 | u32 ; |
121 | u32 ; |
122 | } __packed; |
123 | |
124 | /* These defines must not be changed without changing the protocol version. |
125 | * New defines may only be introduced together with protocol version bump or |
126 | * new protocol feature flags. |
127 | */ |
128 | #define DP_HARDBARRIER 1 /* no longer used */ |
129 | #define DP_RW_SYNC 2 /* equals REQ_SYNC */ |
130 | #define DP_MAY_SET_IN_SYNC 4 |
131 | #define DP_UNPLUG 8 /* not used anymore */ |
132 | #define DP_FUA 16 /* equals REQ_FUA */ |
133 | #define DP_FLUSH 32 /* equals REQ_PREFLUSH */ |
134 | #define DP_DISCARD 64 /* equals REQ_OP_DISCARD */ |
135 | #define DP_SEND_RECEIVE_ACK 128 /* This is a proto B write request */ |
136 | #define DP_SEND_WRITE_ACK 256 /* This is a proto C write request */ |
137 | #define DP_WSAME 512 /* equiv. REQ_WRITE_SAME */ |
138 | #define DP_ZEROES 1024 /* equiv. REQ_OP_WRITE_ZEROES */ |
139 | |
140 | /* possible combinations: |
141 | * REQ_OP_WRITE_ZEROES: DP_DISCARD | DP_ZEROES |
142 | * REQ_OP_WRITE_ZEROES + REQ_NOUNMAP: DP_ZEROES |
143 | */ |
144 | |
145 | struct p_data { |
146 | u64 sector; /* 64 bits sector number */ |
147 | u64 block_id; /* to identify the request in protocol B&C */ |
148 | u32 seq_num; |
149 | u32 dp_flags; |
150 | } __packed; |
151 | |
152 | struct p_trim { |
153 | struct p_data p_data; |
154 | u32 size; /* == bio->bi_size */ |
155 | } __packed; |
156 | |
157 | struct p_wsame { |
158 | struct p_data p_data; |
159 | u32 size; /* == bio->bi_size */ |
160 | } __packed; |
161 | |
162 | /* |
163 | * commands which share a struct: |
164 | * p_block_ack: |
165 | * P_RECV_ACK (proto B), P_WRITE_ACK (proto C), |
166 | * P_SUPERSEDED (proto C, two-primaries conflict detection) |
167 | * p_block_req: |
168 | * P_DATA_REQUEST, P_RS_DATA_REQUEST |
169 | */ |
170 | struct p_block_ack { |
171 | u64 sector; |
172 | u64 block_id; |
173 | u32 blksize; |
174 | u32 seq_num; |
175 | } __packed; |
176 | |
177 | struct p_block_req { |
178 | u64 sector; |
179 | u64 block_id; |
180 | u32 blksize; |
181 | u32 pad; /* to multiple of 8 Byte */ |
182 | } __packed; |
183 | |
184 | /* |
185 | * commands with their own struct for additional fields: |
186 | * P_CONNECTION_FEATURES |
187 | * P_BARRIER |
188 | * P_BARRIER_ACK |
189 | * P_SYNC_PARAM |
190 | * ReportParams |
191 | */ |
192 | |
193 | /* supports TRIM/DISCARD on the "wire" protocol */ |
194 | #define DRBD_FF_TRIM 1 |
195 | |
196 | /* Detect all-zeros during resync, and rather TRIM/UNMAP/DISCARD those blocks |
197 | * instead of fully allocate a supposedly thin volume on initial resync */ |
198 | #define DRBD_FF_THIN_RESYNC 2 |
199 | |
200 | /* supports REQ_WRITE_SAME on the "wire" protocol. |
201 | * Note: this flag is overloaded, |
202 | * its presence also |
203 | * - indicates support for 128 MiB "batch bios", |
204 | * max discard size of 128 MiB |
205 | * instead of 4M before that. |
206 | * - indicates that we exchange additional settings in p_sizes |
207 | * drbd_send_sizes()/receive_sizes() |
208 | */ |
209 | #define DRBD_FF_WSAME 4 |
210 | |
211 | /* supports REQ_OP_WRITE_ZEROES on the "wire" protocol. |
212 | * |
213 | * We used to map that to "discard" on the sending side, and if we cannot |
214 | * guarantee that discard zeroes data, the receiving side would map discard |
215 | * back to zero-out. |
216 | * |
217 | * With the introduction of REQ_OP_WRITE_ZEROES, |
218 | * we started to use that for both WRITE_ZEROES and DISCARDS, |
219 | * hoping that WRITE_ZEROES would "do what we want", |
220 | * UNMAP if possible, zero-out the rest. |
221 | * |
222 | * The example scenario is some LVM "thin" backend. |
223 | * |
224 | * While an un-allocated block on dm-thin reads as zeroes, on a dm-thin |
225 | * with "skip_block_zeroing=true", after a partial block write allocated |
226 | * that block, that same block may well map "undefined old garbage" from |
227 | * the backends on LBAs that have not yet been written to. |
228 | * |
229 | * If we cannot distinguish between zero-out and discard on the receiving |
230 | * side, to avoid "undefined old garbage" to pop up randomly at later times |
231 | * on supposedly zero-initialized blocks, we'd need to map all discards to |
232 | * zero-out on the receiving side. But that would potentially do a full |
233 | * alloc on thinly provisioned backends, even when the expectation was to |
234 | * unmap/trim/discard/de-allocate. |
235 | * |
236 | * We need to distinguish on the protocol level, whether we need to guarantee |
237 | * zeroes (and thus use zero-out, potentially doing the mentioned full-alloc), |
238 | * or if we want to put the emphasis on discard, and only do a "best effort |
239 | * zeroing" (by "discarding" blocks aligned to discard-granularity, and zeroing |
240 | * only potential unaligned head and tail clippings), to at least *try* to |
241 | * avoid "false positives" in an online-verify later, hoping that someone |
242 | * set skip_block_zeroing=false. |
243 | */ |
244 | #define DRBD_FF_WZEROES 8 |
245 | |
246 | |
247 | struct p_connection_features { |
248 | u32 protocol_min; |
249 | u32 feature_flags; |
250 | u32 protocol_max; |
251 | |
252 | /* should be more than enough for future enhancements |
253 | * for now, feature_flags and the reserved array shall be zero. |
254 | */ |
255 | |
256 | u32 _pad; |
257 | u64 reserved[7]; |
258 | } __packed; |
259 | |
260 | struct p_barrier { |
261 | u32 barrier; /* barrier number _handle_ only */ |
262 | u32 pad; /* to multiple of 8 Byte */ |
263 | } __packed; |
264 | |
265 | struct p_barrier_ack { |
266 | u32 barrier; |
267 | u32 set_size; |
268 | } __packed; |
269 | |
270 | struct p_rs_param { |
271 | u32 resync_rate; |
272 | |
273 | /* Since protocol version 88 and higher. */ |
274 | char verify_alg[]; |
275 | } __packed; |
276 | |
277 | struct p_rs_param_89 { |
278 | u32 resync_rate; |
279 | /* protocol version 89: */ |
280 | char verify_alg[SHARED_SECRET_MAX]; |
281 | char csums_alg[SHARED_SECRET_MAX]; |
282 | } __packed; |
283 | |
284 | struct p_rs_param_95 { |
285 | u32 resync_rate; |
286 | struct_group(algs, |
287 | char verify_alg[SHARED_SECRET_MAX]; |
288 | char csums_alg[SHARED_SECRET_MAX]; |
289 | ); |
290 | u32 c_plan_ahead; |
291 | u32 c_delay_target; |
292 | u32 c_fill_target; |
293 | u32 c_max_rate; |
294 | } __packed; |
295 | |
296 | enum drbd_conn_flags { |
297 | CF_DISCARD_MY_DATA = 1, |
298 | CF_DRY_RUN = 2, |
299 | }; |
300 | |
301 | struct p_protocol { |
302 | u32 protocol; |
303 | u32 after_sb_0p; |
304 | u32 after_sb_1p; |
305 | u32 after_sb_2p; |
306 | u32 conn_flags; |
307 | u32 two_primaries; |
308 | |
309 | /* Since protocol version 87 and higher. */ |
310 | char integrity_alg[]; |
311 | |
312 | } __packed; |
313 | |
314 | struct p_uuids { |
315 | u64 uuid[UI_EXTENDED_SIZE]; |
316 | } __packed; |
317 | |
318 | struct p_rs_uuid { |
319 | u64 uuid; |
320 | } __packed; |
321 | |
322 | /* optional queue_limits if (agreed_features & DRBD_FF_WSAME) |
323 | * see also struct queue_limits, as of late 2015 */ |
324 | struct o_qlim { |
325 | /* we don't need it yet, but we may as well communicate it now */ |
326 | u32 physical_block_size; |
327 | |
328 | /* so the original in struct queue_limits is unsigned short, |
329 | * but I'd have to put in padding anyways. */ |
330 | u32 logical_block_size; |
331 | |
332 | /* One incoming bio becomes one DRBD request, |
333 | * which may be translated to several bio on the receiving side. |
334 | * We don't need to communicate chunk/boundary/segment ... limits. |
335 | */ |
336 | |
337 | /* various IO hints may be useful with "diskless client" setups */ |
338 | u32 alignment_offset; |
339 | u32 io_min; |
340 | u32 io_opt; |
341 | |
342 | /* We may need to communicate integrity stuff at some point, |
343 | * but let's not get ahead of ourselves. */ |
344 | |
345 | /* Backend discard capabilities. |
346 | * Receiving side uses "blkdev_issue_discard()", no need to communicate |
347 | * more specifics. If the backend cannot do discards, the DRBD peer |
348 | * may fall back to blkdev_issue_zeroout(). |
349 | */ |
350 | u8 discard_enabled; |
351 | u8 discard_zeroes_data; |
352 | u8 write_same_capable; |
353 | u8 _pad; |
354 | } __packed; |
355 | |
356 | struct p_sizes { |
357 | u64 d_size; /* size of disk */ |
358 | u64 u_size; /* user requested size */ |
359 | u64 c_size; /* current exported size */ |
360 | u32 max_bio_size; /* Maximal size of a BIO */ |
361 | u16 queue_order_type; /* not yet implemented in DRBD*/ |
362 | u16 dds_flags; /* use enum dds_flags here. */ |
363 | |
364 | /* optional queue_limits if (agreed_features & DRBD_FF_WSAME) */ |
365 | struct o_qlim qlim[]; |
366 | } __packed; |
367 | |
368 | struct p_state { |
369 | u32 state; |
370 | } __packed; |
371 | |
372 | struct p_req_state { |
373 | u32 mask; |
374 | u32 val; |
375 | } __packed; |
376 | |
377 | struct p_req_state_reply { |
378 | u32 retcode; |
379 | } __packed; |
380 | |
381 | struct p_drbd06_param { |
382 | u64 size; |
383 | u32 state; |
384 | u32 blksize; |
385 | u32 protocol; |
386 | u32 version; |
387 | u32 gen_cnt[5]; |
388 | u32 bit_map_gen[5]; |
389 | } __packed; |
390 | |
391 | struct p_block_desc { |
392 | u64 sector; |
393 | u32 blksize; |
394 | u32 pad; /* to multiple of 8 Byte */ |
395 | } __packed; |
396 | |
397 | /* Valid values for the encoding field. |
398 | * Bump proto version when changing this. */ |
399 | enum drbd_bitmap_code { |
400 | /* RLE_VLI_Bytes = 0, |
401 | * and other bit variants had been defined during |
402 | * algorithm evaluation. */ |
403 | RLE_VLI_Bits = 2, |
404 | }; |
405 | |
406 | struct p_compressed_bm { |
407 | /* (encoding & 0x0f): actual encoding, see enum drbd_bitmap_code |
408 | * (encoding & 0x80): polarity (set/unset) of first runlength |
409 | * ((encoding >> 4) & 0x07): pad_bits, number of trailing zero bits |
410 | * used to pad up to head.length bytes |
411 | */ |
412 | u8 encoding; |
413 | |
414 | u8 code[]; |
415 | } __packed; |
416 | |
417 | struct p_delay_probe93 { |
418 | u32 seq_num; /* sequence number to match the two probe packets */ |
419 | u32 offset; /* usecs the probe got sent after the reference time point */ |
420 | } __packed; |
421 | |
422 | /* |
423 | * Bitmap packets need to fit within a single page on the sender and receiver, |
424 | * so we are limited to 4 KiB (and not to PAGE_SIZE, which can be bigger). |
425 | */ |
426 | #define DRBD_SOCKET_BUFFER_SIZE 4096 |
427 | |
428 | #endif /* __DRBD_PROTOCOL_H */ |
429 | |