Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_int.h | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | drbd is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
14 | ||
15 | drbd is distributed in the hope that it will be useful, | |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | GNU General Public License for more details. | |
19 | ||
20 | You should have received a copy of the GNU General Public License | |
21 | along with drbd; see the file COPYING. If not, write to | |
22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | ||
24 | */ | |
25 | ||
26 | #ifndef _DRBD_INT_H | |
27 | #define _DRBD_INT_H | |
28 | ||
29 | #include <linux/compiler.h> | |
30 | #include <linux/types.h> | |
31 | #include <linux/version.h> | |
32 | #include <linux/list.h> | |
33 | #include <linux/sched.h> | |
34 | #include <linux/bitops.h> | |
35 | #include <linux/slab.h> | |
36 | #include <linux/crypto.h> | |
132cc538 | 37 | #include <linux/ratelimit.h> |
b411b363 PR |
38 | #include <linux/tcp.h> |
39 | #include <linux/mutex.h> | |
40 | #include <linux/major.h> | |
41 | #include <linux/blkdev.h> | |
42 | #include <linux/genhd.h> | |
43 | #include <net/tcp.h> | |
44 | #include <linux/lru_cache.h> | |
45 | ||
46 | #ifdef __CHECKER__ | |
47 | # define __protected_by(x) __attribute__((require_context(x,1,999,"rdwr"))) | |
48 | # define __protected_read_by(x) __attribute__((require_context(x,1,999,"read"))) | |
49 | # define __protected_write_by(x) __attribute__((require_context(x,1,999,"write"))) | |
50 | # define __must_hold(x) __attribute__((context(x,1,1), require_context(x,1,999,"call"))) | |
51 | #else | |
52 | # define __protected_by(x) | |
53 | # define __protected_read_by(x) | |
54 | # define __protected_write_by(x) | |
55 | # define __must_hold(x) | |
56 | #endif | |
57 | ||
58 | #define __no_warn(lock, stmt) do { __acquire(lock); stmt; __release(lock); } while (0) | |
59 | ||
60 | /* module parameter, defined in drbd_main.c */ | |
61 | extern unsigned int minor_count; | |
62 | extern int disable_sendpage; | |
63 | extern int allow_oos; | |
64 | extern unsigned int cn_idx; | |
65 | ||
66 | #ifdef CONFIG_DRBD_FAULT_INJECTION | |
67 | extern int enable_faults; | |
68 | extern int fault_rate; | |
69 | extern int fault_devs; | |
70 | #endif | |
71 | ||
72 | extern char usermode_helper[]; | |
73 | ||
74 | ||
b411b363 PR |
75 | /* I don't remember why XCPU ... |
76 | * This is used to wake the asender, | |
77 | * and to interrupt sending the sending task | |
78 | * on disconnect. | |
79 | */ | |
80 | #define DRBD_SIG SIGXCPU | |
81 | ||
82 | /* This is used to stop/restart our threads. | |
83 | * Cannot use SIGTERM nor SIGKILL, since these | |
84 | * are sent out by init on runlevel changes | |
85 | * I choose SIGHUP for now. | |
86 | */ | |
87 | #define DRBD_SIGKILL SIGHUP | |
88 | ||
89 | /* All EEs on the free list should have ID_VACANT (== 0) | |
90 | * freshly allocated EEs get !ID_VACANT (== 1) | |
3ad2f3fb | 91 | * so if it says "cannot dereference null pointer at address 0x00000001", |
b411b363 PR |
92 | * it is most likely one of these :( */ |
93 | ||
94 | #define ID_IN_SYNC (4711ULL) | |
95 | #define ID_OUT_OF_SYNC (4712ULL) | |
96 | ||
97 | #define ID_SYNCER (-1ULL) | |
98 | #define ID_VACANT 0 | |
99 | #define is_syncer_block_id(id) ((id) == ID_SYNCER) | |
4a23f264 | 100 | #define UUID_NEW_BM_OFFSET ((u64)0x0001000000000000ULL) |
b411b363 PR |
101 | |
102 | struct drbd_conf; | |
103 | ||
104 | ||
105 | /* to shorten dev_warn(DEV, "msg"); and relatives statements */ | |
106 | #define DEV (disk_to_dev(mdev->vdisk)) | |
107 | ||
108 | #define D_ASSERT(exp) if (!(exp)) \ | |
109 | dev_err(DEV, "ASSERT( " #exp " ) in %s:%d\n", __FILE__, __LINE__) | |
110 | ||
3beec1d4 LE |
111 | #define ERR_IF(exp) if (({ \ |
112 | int _b = (exp) != 0; \ | |
113 | if (_b) dev_err(DEV, "ASSERT FAILED: %s: (%s) in %s:%d\n", \ | |
114 | __func__, #exp, __FILE__, __LINE__); \ | |
115 | _b; \ | |
b411b363 PR |
116 | })) |
117 | ||
118 | /* Defines to control fault insertion */ | |
119 | enum { | |
120 | DRBD_FAULT_MD_WR = 0, /* meta data write */ | |
121 | DRBD_FAULT_MD_RD = 1, /* read */ | |
122 | DRBD_FAULT_RS_WR = 2, /* resync */ | |
123 | DRBD_FAULT_RS_RD = 3, | |
124 | DRBD_FAULT_DT_WR = 4, /* data */ | |
125 | DRBD_FAULT_DT_RD = 5, | |
126 | DRBD_FAULT_DT_RA = 6, /* data read ahead */ | |
127 | DRBD_FAULT_BM_ALLOC = 7, /* bitmap allocation */ | |
128 | DRBD_FAULT_AL_EE = 8, /* alloc ee */ | |
6b4388ac | 129 | DRBD_FAULT_RECEIVE = 9, /* Changes some bytes upon receiving a [rs]data block */ |
b411b363 PR |
130 | |
131 | DRBD_FAULT_MAX, | |
132 | }; | |
133 | ||
b411b363 PR |
134 | extern unsigned int |
135 | _drbd_insert_fault(struct drbd_conf *mdev, unsigned int type); | |
0cf9d27e | 136 | |
b411b363 PR |
137 | static inline int |
138 | drbd_insert_fault(struct drbd_conf *mdev, unsigned int type) { | |
0cf9d27e | 139 | #ifdef CONFIG_DRBD_FAULT_INJECTION |
b411b363 PR |
140 | return fault_rate && |
141 | (enable_faults & (1<<type)) && | |
142 | _drbd_insert_fault(mdev, type); | |
b411b363 | 143 | #else |
0cf9d27e | 144 | return 0; |
b411b363 | 145 | #endif |
0cf9d27e | 146 | } |
b411b363 PR |
147 | |
148 | /* integer division, round _UP_ to the next integer */ | |
149 | #define div_ceil(A, B) ((A)/(B) + ((A)%(B) ? 1 : 0)) | |
150 | /* usual integer division */ | |
151 | #define div_floor(A, B) ((A)/(B)) | |
152 | ||
153 | /* drbd_meta-data.c (still in drbd_main.c) */ | |
154 | /* 4th incarnation of the disk layout. */ | |
155 | #define DRBD_MD_MAGIC (DRBD_MAGIC+4) | |
156 | ||
157 | extern struct drbd_conf **minor_table; | |
158 | extern struct ratelimit_state drbd_ratelimit_state; | |
159 | ||
160 | /* on the wire */ | |
161 | enum drbd_packets { | |
162 | /* receiver (data socket) */ | |
163 | P_DATA = 0x00, | |
164 | P_DATA_REPLY = 0x01, /* Response to P_DATA_REQUEST */ | |
165 | P_RS_DATA_REPLY = 0x02, /* Response to P_RS_DATA_REQUEST */ | |
166 | P_BARRIER = 0x03, | |
167 | P_BITMAP = 0x04, | |
168 | P_BECOME_SYNC_TARGET = 0x05, | |
169 | P_BECOME_SYNC_SOURCE = 0x06, | |
170 | P_UNPLUG_REMOTE = 0x07, /* Used at various times to hint the peer */ | |
171 | P_DATA_REQUEST = 0x08, /* Used to ask for a data block */ | |
172 | P_RS_DATA_REQUEST = 0x09, /* Used to ask for a data block for resync */ | |
173 | P_SYNC_PARAM = 0x0a, | |
174 | P_PROTOCOL = 0x0b, | |
175 | P_UUIDS = 0x0c, | |
176 | P_SIZES = 0x0d, | |
177 | P_STATE = 0x0e, | |
178 | P_SYNC_UUID = 0x0f, | |
179 | P_AUTH_CHALLENGE = 0x10, | |
180 | P_AUTH_RESPONSE = 0x11, | |
181 | P_STATE_CHG_REQ = 0x12, | |
182 | ||
183 | /* asender (meta socket */ | |
184 | P_PING = 0x13, | |
185 | P_PING_ACK = 0x14, | |
186 | P_RECV_ACK = 0x15, /* Used in protocol B */ | |
187 | P_WRITE_ACK = 0x16, /* Used in protocol C */ | |
188 | P_RS_WRITE_ACK = 0x17, /* Is a P_WRITE_ACK, additionally call set_in_sync(). */ | |
189 | P_DISCARD_ACK = 0x18, /* Used in proto C, two-primaries conflict detection */ | |
190 | P_NEG_ACK = 0x19, /* Sent if local disk is unusable */ | |
191 | P_NEG_DREPLY = 0x1a, /* Local disk is broken... */ | |
192 | P_NEG_RS_DREPLY = 0x1b, /* Local disk is broken... */ | |
193 | P_BARRIER_ACK = 0x1c, | |
194 | P_STATE_CHG_REPLY = 0x1d, | |
195 | ||
196 | /* "new" commands, no longer fitting into the ordering scheme above */ | |
197 | ||
198 | P_OV_REQUEST = 0x1e, /* data socket */ | |
199 | P_OV_REPLY = 0x1f, | |
200 | P_OV_RESULT = 0x20, /* meta socket */ | |
201 | P_CSUM_RS_REQUEST = 0x21, /* data socket */ | |
202 | P_RS_IS_IN_SYNC = 0x22, /* meta socket */ | |
203 | P_SYNC_PARAM89 = 0x23, /* data socket, protocol version 89 replacement for P_SYNC_PARAM */ | |
204 | P_COMPRESSED_BITMAP = 0x24, /* compressed or otherwise encoded bitmap transfer */ | |
0ced55a3 PR |
205 | /* P_CKPT_FENCE_REQ = 0x25, * currently reserved for protocol D */ |
206 | /* P_CKPT_DISABLE_REQ = 0x26, * currently reserved for protocol D */ | |
207 | P_DELAY_PROBE = 0x27, /* is used on BOTH sockets */ | |
73a01a18 | 208 | P_OUT_OF_SYNC = 0x28, /* Mark as out of sync (Outrunning), data socket */ |
d612d309 | 209 | P_RS_CANCEL = 0x29, /* meta: Used to cancel RS_DATA_REQUEST packet by SyncSource */ |
b411b363 | 210 | |
d612d309 | 211 | P_MAX_CMD = 0x2A, |
b411b363 PR |
212 | P_MAY_IGNORE = 0x100, /* Flag to test if (cmd > P_MAY_IGNORE) ... */ |
213 | P_MAX_OPT_CMD = 0x101, | |
214 | ||
215 | /* special command ids for handshake */ | |
216 | ||
217 | P_HAND_SHAKE_M = 0xfff1, /* First Packet on the MetaSock */ | |
218 | P_HAND_SHAKE_S = 0xfff2, /* First Packet on the Socket */ | |
219 | ||
220 | P_HAND_SHAKE = 0xfffe /* FIXED for the next century! */ | |
221 | }; | |
222 | ||
223 | static inline const char *cmdname(enum drbd_packets cmd) | |
224 | { | |
225 | /* THINK may need to become several global tables | |
226 | * when we want to support more than | |
227 | * one PRO_VERSION */ | |
228 | static const char *cmdnames[] = { | |
229 | [P_DATA] = "Data", | |
230 | [P_DATA_REPLY] = "DataReply", | |
231 | [P_RS_DATA_REPLY] = "RSDataReply", | |
232 | [P_BARRIER] = "Barrier", | |
233 | [P_BITMAP] = "ReportBitMap", | |
234 | [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget", | |
235 | [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource", | |
236 | [P_UNPLUG_REMOTE] = "UnplugRemote", | |
237 | [P_DATA_REQUEST] = "DataRequest", | |
238 | [P_RS_DATA_REQUEST] = "RSDataRequest", | |
239 | [P_SYNC_PARAM] = "SyncParam", | |
240 | [P_SYNC_PARAM89] = "SyncParam89", | |
241 | [P_PROTOCOL] = "ReportProtocol", | |
242 | [P_UUIDS] = "ReportUUIDs", | |
243 | [P_SIZES] = "ReportSizes", | |
244 | [P_STATE] = "ReportState", | |
245 | [P_SYNC_UUID] = "ReportSyncUUID", | |
246 | [P_AUTH_CHALLENGE] = "AuthChallenge", | |
247 | [P_AUTH_RESPONSE] = "AuthResponse", | |
248 | [P_PING] = "Ping", | |
249 | [P_PING_ACK] = "PingAck", | |
250 | [P_RECV_ACK] = "RecvAck", | |
251 | [P_WRITE_ACK] = "WriteAck", | |
252 | [P_RS_WRITE_ACK] = "RSWriteAck", | |
253 | [P_DISCARD_ACK] = "DiscardAck", | |
254 | [P_NEG_ACK] = "NegAck", | |
255 | [P_NEG_DREPLY] = "NegDReply", | |
256 | [P_NEG_RS_DREPLY] = "NegRSDReply", | |
257 | [P_BARRIER_ACK] = "BarrierAck", | |
258 | [P_STATE_CHG_REQ] = "StateChgRequest", | |
259 | [P_STATE_CHG_REPLY] = "StateChgReply", | |
260 | [P_OV_REQUEST] = "OVRequest", | |
261 | [P_OV_REPLY] = "OVReply", | |
262 | [P_OV_RESULT] = "OVResult", | |
c42b6cf4 LE |
263 | [P_CSUM_RS_REQUEST] = "CsumRSRequest", |
264 | [P_RS_IS_IN_SYNC] = "CsumRSIsInSync", | |
265 | [P_COMPRESSED_BITMAP] = "CBitmap", | |
a8cdfd8d | 266 | [P_DELAY_PROBE] = "DelayProbe", |
73a01a18 | 267 | [P_OUT_OF_SYNC] = "OutOfSync", |
b411b363 PR |
268 | [P_MAX_CMD] = NULL, |
269 | }; | |
270 | ||
271 | if (cmd == P_HAND_SHAKE_M) | |
272 | return "HandShakeM"; | |
273 | if (cmd == P_HAND_SHAKE_S) | |
274 | return "HandShakeS"; | |
275 | if (cmd == P_HAND_SHAKE) | |
276 | return "HandShake"; | |
277 | if (cmd >= P_MAX_CMD) | |
278 | return "Unknown"; | |
279 | return cmdnames[cmd]; | |
280 | } | |
281 | ||
282 | /* for sending/receiving the bitmap, | |
283 | * possibly in some encoding scheme */ | |
284 | struct bm_xfer_ctx { | |
285 | /* "const" | |
286 | * stores total bits and long words | |
287 | * of the bitmap, so we don't need to | |
288 | * call the accessor functions over and again. */ | |
289 | unsigned long bm_bits; | |
290 | unsigned long bm_words; | |
291 | /* during xfer, current position within the bitmap */ | |
292 | unsigned long bit_offset; | |
293 | unsigned long word_offset; | |
294 | ||
295 | /* statistics; index: (h->command == P_BITMAP) */ | |
296 | unsigned packets[2]; | |
297 | unsigned bytes[2]; | |
298 | }; | |
299 | ||
300 | extern void INFO_bm_xfer_stats(struct drbd_conf *mdev, | |
301 | const char *direction, struct bm_xfer_ctx *c); | |
302 | ||
303 | static inline void bm_xfer_ctx_bit_to_word_offset(struct bm_xfer_ctx *c) | |
304 | { | |
305 | /* word_offset counts "native long words" (32 or 64 bit), | |
306 | * aligned at 64 bit. | |
307 | * Encoded packet may end at an unaligned bit offset. | |
308 | * In case a fallback clear text packet is transmitted in | |
309 | * between, we adjust this offset back to the last 64bit | |
310 | * aligned "native long word", which makes coding and decoding | |
311 | * the plain text bitmap much more convenient. */ | |
312 | #if BITS_PER_LONG == 64 | |
313 | c->word_offset = c->bit_offset >> 6; | |
314 | #elif BITS_PER_LONG == 32 | |
315 | c->word_offset = c->bit_offset >> 5; | |
316 | c->word_offset &= ~(1UL); | |
317 | #else | |
318 | # error "unsupported BITS_PER_LONG" | |
319 | #endif | |
320 | } | |
321 | ||
322 | #ifndef __packed | |
323 | #define __packed __attribute__((packed)) | |
324 | #endif | |
325 | ||
326 | /* This is the layout for a packet on the wire. | |
327 | * The byteorder is the network byte order. | |
328 | * (except block_id and barrier fields. | |
329 | * these are pointers to local structs | |
330 | * and have no relevance for the partner, | |
331 | * which just echoes them as received.) | |
332 | * | |
333 | * NOTE that the payload starts at a long aligned offset, | |
334 | * regardless of 32 or 64 bit arch! | |
335 | */ | |
0b70a13d | 336 | struct p_header80 { |
b411b363 PR |
337 | u32 magic; |
338 | u16 command; | |
339 | u16 length; /* bytes of data after this header */ | |
340 | u8 payload[0]; | |
341 | } __packed; | |
0b70a13d PR |
342 | |
343 | /* Header for big packets, Used for data packets exceeding 64kB */ | |
344 | struct p_header95 { | |
345 | u16 magic; /* use DRBD_MAGIC_BIG here */ | |
346 | u16 command; | |
00b42537 | 347 | u32 length; /* Use only 24 bits of that. Ignore the highest 8 bit. */ |
0b70a13d PR |
348 | u8 payload[0]; |
349 | } __packed; | |
350 | ||
351 | union p_header { | |
352 | struct p_header80 h80; | |
353 | struct p_header95 h95; | |
354 | }; | |
b411b363 PR |
355 | |
356 | /* | |
357 | * short commands, packets without payload, plain p_header: | |
358 | * P_PING | |
359 | * P_PING_ACK | |
360 | * P_BECOME_SYNC_TARGET | |
361 | * P_BECOME_SYNC_SOURCE | |
362 | * P_UNPLUG_REMOTE | |
363 | */ | |
364 | ||
365 | /* | |
366 | * commands with out-of-struct payload: | |
367 | * P_BITMAP (no additional fields) | |
368 | * P_DATA, P_DATA_REPLY (see p_data) | |
369 | * P_COMPRESSED_BITMAP (see receive_compressed_bitmap) | |
370 | */ | |
371 | ||
372 | /* these defines must not be changed without changing the protocol version */ | |
76d2e7ec PR |
373 | #define DP_HARDBARRIER 1 /* depricated */ |
374 | #define DP_RW_SYNC 2 /* equals REQ_SYNC */ | |
b411b363 | 375 | #define DP_MAY_SET_IN_SYNC 4 |
721a9602 | 376 | #define DP_UNPLUG 8 /* not used anymore */ |
76d2e7ec PR |
377 | #define DP_FUA 16 /* equals REQ_FUA */ |
378 | #define DP_FLUSH 32 /* equals REQ_FLUSH */ | |
379 | #define DP_DISCARD 64 /* equals REQ_DISCARD */ | |
b411b363 PR |
380 | |
381 | struct p_data { | |
0b70a13d | 382 | union p_header head; |
b411b363 PR |
383 | u64 sector; /* 64 bits sector number */ |
384 | u64 block_id; /* to identify the request in protocol B&C */ | |
385 | u32 seq_num; | |
386 | u32 dp_flags; | |
387 | } __packed; | |
388 | ||
389 | /* | |
390 | * commands which share a struct: | |
391 | * p_block_ack: | |
392 | * P_RECV_ACK (proto B), P_WRITE_ACK (proto C), | |
393 | * P_DISCARD_ACK (proto C, two-primaries conflict detection) | |
394 | * p_block_req: | |
395 | * P_DATA_REQUEST, P_RS_DATA_REQUEST | |
396 | */ | |
397 | struct p_block_ack { | |
0b70a13d | 398 | struct p_header80 head; |
b411b363 PR |
399 | u64 sector; |
400 | u64 block_id; | |
401 | u32 blksize; | |
402 | u32 seq_num; | |
403 | } __packed; | |
404 | ||
405 | ||
406 | struct p_block_req { | |
0b70a13d | 407 | struct p_header80 head; |
b411b363 PR |
408 | u64 sector; |
409 | u64 block_id; | |
410 | u32 blksize; | |
411 | u32 pad; /* to multiple of 8 Byte */ | |
412 | } __packed; | |
413 | ||
414 | /* | |
415 | * commands with their own struct for additional fields: | |
416 | * P_HAND_SHAKE | |
417 | * P_BARRIER | |
418 | * P_BARRIER_ACK | |
419 | * P_SYNC_PARAM | |
420 | * ReportParams | |
421 | */ | |
422 | ||
423 | struct p_handshake { | |
0b70a13d | 424 | struct p_header80 head; /* 8 bytes */ |
b411b363 PR |
425 | u32 protocol_min; |
426 | u32 feature_flags; | |
427 | u32 protocol_max; | |
428 | ||
429 | /* should be more than enough for future enhancements | |
430 | * for now, feature_flags and the reserverd array shall be zero. | |
431 | */ | |
432 | ||
433 | u32 _pad; | |
434 | u64 reserverd[7]; | |
435 | } __packed; | |
436 | /* 80 bytes, FIXED for the next century */ | |
437 | ||
438 | struct p_barrier { | |
0b70a13d | 439 | struct p_header80 head; |
b411b363 PR |
440 | u32 barrier; /* barrier number _handle_ only */ |
441 | u32 pad; /* to multiple of 8 Byte */ | |
442 | } __packed; | |
443 | ||
444 | struct p_barrier_ack { | |
0b70a13d | 445 | struct p_header80 head; |
b411b363 PR |
446 | u32 barrier; |
447 | u32 set_size; | |
448 | } __packed; | |
449 | ||
450 | struct p_rs_param { | |
0b70a13d | 451 | struct p_header80 head; |
b411b363 PR |
452 | u32 rate; |
453 | ||
454 | /* Since protocol version 88 and higher. */ | |
455 | char verify_alg[0]; | |
456 | } __packed; | |
457 | ||
458 | struct p_rs_param_89 { | |
0b70a13d | 459 | struct p_header80 head; |
b411b363 PR |
460 | u32 rate; |
461 | /* protocol version 89: */ | |
462 | char verify_alg[SHARED_SECRET_MAX]; | |
463 | char csums_alg[SHARED_SECRET_MAX]; | |
464 | } __packed; | |
465 | ||
8e26f9cc | 466 | struct p_rs_param_95 { |
0b70a13d | 467 | struct p_header80 head; |
8e26f9cc PR |
468 | u32 rate; |
469 | char verify_alg[SHARED_SECRET_MAX]; | |
470 | char csums_alg[SHARED_SECRET_MAX]; | |
471 | u32 c_plan_ahead; | |
472 | u32 c_delay_target; | |
473 | u32 c_fill_target; | |
474 | u32 c_max_rate; | |
475 | } __packed; | |
476 | ||
cf14c2e9 PR |
477 | enum drbd_conn_flags { |
478 | CF_WANT_LOSE = 1, | |
479 | CF_DRY_RUN = 2, | |
480 | }; | |
481 | ||
b411b363 | 482 | struct p_protocol { |
0b70a13d | 483 | struct p_header80 head; |
b411b363 PR |
484 | u32 protocol; |
485 | u32 after_sb_0p; | |
486 | u32 after_sb_1p; | |
487 | u32 after_sb_2p; | |
cf14c2e9 | 488 | u32 conn_flags; |
b411b363 PR |
489 | u32 two_primaries; |
490 | ||
491 | /* Since protocol version 87 and higher. */ | |
492 | char integrity_alg[0]; | |
493 | ||
494 | } __packed; | |
495 | ||
496 | struct p_uuids { | |
0b70a13d | 497 | struct p_header80 head; |
b411b363 PR |
498 | u64 uuid[UI_EXTENDED_SIZE]; |
499 | } __packed; | |
500 | ||
501 | struct p_rs_uuid { | |
0b70a13d | 502 | struct p_header80 head; |
b411b363 PR |
503 | u64 uuid; |
504 | } __packed; | |
505 | ||
506 | struct p_sizes { | |
0b70a13d | 507 | struct p_header80 head; |
b411b363 PR |
508 | u64 d_size; /* size of disk */ |
509 | u64 u_size; /* user requested size */ | |
510 | u64 c_size; /* current exported size */ | |
1816a2b4 | 511 | u32 max_bio_size; /* Maximal size of a BIO */ |
e89b591c PR |
512 | u16 queue_order_type; /* not yet implemented in DRBD*/ |
513 | u16 dds_flags; /* use enum dds_flags here. */ | |
b411b363 PR |
514 | } __packed; |
515 | ||
516 | struct p_state { | |
0b70a13d | 517 | struct p_header80 head; |
b411b363 PR |
518 | u32 state; |
519 | } __packed; | |
520 | ||
521 | struct p_req_state { | |
0b70a13d | 522 | struct p_header80 head; |
b411b363 PR |
523 | u32 mask; |
524 | u32 val; | |
525 | } __packed; | |
526 | ||
527 | struct p_req_state_reply { | |
0b70a13d | 528 | struct p_header80 head; |
b411b363 PR |
529 | u32 retcode; |
530 | } __packed; | |
531 | ||
532 | struct p_drbd06_param { | |
533 | u64 size; | |
534 | u32 state; | |
535 | u32 blksize; | |
536 | u32 protocol; | |
537 | u32 version; | |
538 | u32 gen_cnt[5]; | |
539 | u32 bit_map_gen[5]; | |
540 | } __packed; | |
541 | ||
542 | struct p_discard { | |
0b70a13d | 543 | struct p_header80 head; |
b411b363 PR |
544 | u64 block_id; |
545 | u32 seq_num; | |
546 | u32 pad; | |
547 | } __packed; | |
548 | ||
73a01a18 PR |
549 | struct p_block_desc { |
550 | struct p_header80 head; | |
551 | u64 sector; | |
552 | u32 blksize; | |
553 | u32 pad; /* to multiple of 8 Byte */ | |
554 | } __packed; | |
555 | ||
b411b363 PR |
556 | /* Valid values for the encoding field. |
557 | * Bump proto version when changing this. */ | |
558 | enum drbd_bitmap_code { | |
559 | /* RLE_VLI_Bytes = 0, | |
560 | * and other bit variants had been defined during | |
561 | * algorithm evaluation. */ | |
562 | RLE_VLI_Bits = 2, | |
563 | }; | |
564 | ||
565 | struct p_compressed_bm { | |
0b70a13d | 566 | struct p_header80 head; |
b411b363 PR |
567 | /* (encoding & 0x0f): actual encoding, see enum drbd_bitmap_code |
568 | * (encoding & 0x80): polarity (set/unset) of first runlength | |
569 | * ((encoding >> 4) & 0x07): pad_bits, number of trailing zero bits | |
570 | * used to pad up to head.length bytes | |
571 | */ | |
572 | u8 encoding; | |
573 | ||
574 | u8 code[0]; | |
575 | } __packed; | |
576 | ||
0b70a13d PR |
577 | struct p_delay_probe93 { |
578 | struct p_header80 head; | |
579 | u32 seq_num; /* sequence number to match the two probe packets */ | |
580 | u32 offset; /* usecs the probe got sent after the reference time point */ | |
0ced55a3 PR |
581 | } __packed; |
582 | ||
b411b363 PR |
583 | /* DCBP: Drbd Compressed Bitmap Packet ... */ |
584 | static inline enum drbd_bitmap_code | |
585 | DCBP_get_code(struct p_compressed_bm *p) | |
586 | { | |
587 | return (enum drbd_bitmap_code)(p->encoding & 0x0f); | |
588 | } | |
589 | ||
590 | static inline void | |
591 | DCBP_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code) | |
592 | { | |
593 | BUG_ON(code & ~0xf); | |
594 | p->encoding = (p->encoding & ~0xf) | code; | |
595 | } | |
596 | ||
597 | static inline int | |
598 | DCBP_get_start(struct p_compressed_bm *p) | |
599 | { | |
600 | return (p->encoding & 0x80) != 0; | |
601 | } | |
602 | ||
603 | static inline void | |
604 | DCBP_set_start(struct p_compressed_bm *p, int set) | |
605 | { | |
606 | p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0); | |
607 | } | |
608 | ||
609 | static inline int | |
610 | DCBP_get_pad_bits(struct p_compressed_bm *p) | |
611 | { | |
612 | return (p->encoding >> 4) & 0x7; | |
613 | } | |
614 | ||
615 | static inline void | |
616 | DCBP_set_pad_bits(struct p_compressed_bm *p, int n) | |
617 | { | |
618 | BUG_ON(n & ~0x7); | |
619 | p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4); | |
620 | } | |
621 | ||
622 | /* one bitmap packet, including the p_header, | |
623 | * should fit within one _architecture independend_ page. | |
624 | * so we need to use the fixed size 4KiB page size | |
625 | * most architechtures have used for a long time. | |
626 | */ | |
0b70a13d | 627 | #define BM_PACKET_PAYLOAD_BYTES (4096 - sizeof(struct p_header80)) |
b411b363 PR |
628 | #define BM_PACKET_WORDS (BM_PACKET_PAYLOAD_BYTES/sizeof(long)) |
629 | #define BM_PACKET_VLI_BYTES_MAX (4096 - sizeof(struct p_compressed_bm)) | |
630 | #if (PAGE_SIZE < 4096) | |
631 | /* drbd_send_bitmap / receive_bitmap would break horribly */ | |
632 | #error "PAGE_SIZE too small" | |
633 | #endif | |
634 | ||
635 | union p_polymorph { | |
02918be2 | 636 | union p_header header; |
b411b363 PR |
637 | struct p_handshake handshake; |
638 | struct p_data data; | |
639 | struct p_block_ack block_ack; | |
640 | struct p_barrier barrier; | |
641 | struct p_barrier_ack barrier_ack; | |
642 | struct p_rs_param_89 rs_param_89; | |
8e26f9cc | 643 | struct p_rs_param_95 rs_param_95; |
b411b363 PR |
644 | struct p_protocol protocol; |
645 | struct p_sizes sizes; | |
646 | struct p_uuids uuids; | |
647 | struct p_state state; | |
648 | struct p_req_state req_state; | |
649 | struct p_req_state_reply req_state_reply; | |
650 | struct p_block_req block_req; | |
02918be2 PR |
651 | struct p_delay_probe93 delay_probe93; |
652 | struct p_rs_uuid rs_uuid; | |
73a01a18 | 653 | struct p_block_desc block_desc; |
b411b363 PR |
654 | } __packed; |
655 | ||
656 | /**********************************************************************/ | |
657 | enum drbd_thread_state { | |
658 | None, | |
659 | Running, | |
660 | Exiting, | |
661 | Restarting | |
662 | }; | |
663 | ||
664 | struct drbd_thread { | |
665 | spinlock_t t_lock; | |
666 | struct task_struct *task; | |
667 | struct completion stop; | |
668 | enum drbd_thread_state t_state; | |
669 | int (*function) (struct drbd_thread *); | |
670 | struct drbd_conf *mdev; | |
671 | int reset_cpu_mask; | |
672 | }; | |
673 | ||
674 | static inline enum drbd_thread_state get_t_state(struct drbd_thread *thi) | |
675 | { | |
676 | /* THINK testing the t_state seems to be uncritical in all cases | |
677 | * (but thread_{start,stop}), so we can read it *without* the lock. | |
678 | * --lge */ | |
679 | ||
680 | smp_rmb(); | |
681 | return thi->t_state; | |
682 | } | |
683 | ||
b411b363 PR |
684 | struct drbd_work; |
685 | typedef int (*drbd_work_cb)(struct drbd_conf *, struct drbd_work *, int cancel); | |
686 | struct drbd_work { | |
687 | struct list_head list; | |
688 | drbd_work_cb cb; | |
689 | }; | |
690 | ||
691 | struct drbd_tl_epoch; | |
692 | struct drbd_request { | |
693 | struct drbd_work w; | |
694 | struct drbd_conf *mdev; | |
695 | ||
696 | /* if local IO is not allowed, will be NULL. | |
697 | * if local IO _is_ allowed, holds the locally submitted bio clone, | |
698 | * or, after local IO completion, the ERR_PTR(error). | |
699 | * see drbd_endio_pri(). */ | |
700 | struct bio *private_bio; | |
701 | ||
702 | struct hlist_node colision; | |
703 | sector_t sector; | |
704 | unsigned int size; | |
705 | unsigned int epoch; /* barrier_nr */ | |
706 | ||
707 | /* barrier_nr: used to check on "completion" whether this req was in | |
708 | * the current epoch, and we therefore have to close it, | |
709 | * starting a new epoch... | |
710 | */ | |
711 | ||
b411b363 PR |
712 | struct list_head tl_requests; /* ring list in the transfer log */ |
713 | struct bio *master_bio; /* master bio pointer */ | |
714 | unsigned long rq_state; /* see comments above _req_mod() */ | |
715 | int seq_num; | |
716 | unsigned long start_time; | |
717 | }; | |
718 | ||
719 | struct drbd_tl_epoch { | |
720 | struct drbd_work w; | |
721 | struct list_head requests; /* requests before */ | |
722 | struct drbd_tl_epoch *next; /* pointer to the next barrier */ | |
723 | unsigned int br_number; /* the barriers identifier. */ | |
7e602c0a | 724 | int n_writes; /* number of requests attached before this barrier */ |
b411b363 PR |
725 | }; |
726 | ||
727 | struct drbd_request; | |
728 | ||
729 | /* These Tl_epoch_entries may be in one of 6 lists: | |
730 | active_ee .. data packet being written | |
731 | sync_ee .. syncer block being written | |
732 | done_ee .. block written, need to send P_WRITE_ACK | |
733 | read_ee .. [RS]P_DATA_REQUEST being read | |
734 | */ | |
735 | ||
736 | struct drbd_epoch { | |
737 | struct list_head list; | |
738 | unsigned int barrier_nr; | |
739 | atomic_t epoch_size; /* increased on every request added. */ | |
740 | atomic_t active; /* increased on every req. added, and dec on every finished. */ | |
741 | unsigned long flags; | |
742 | }; | |
743 | ||
744 | /* drbd_epoch flag bits */ | |
745 | enum { | |
b411b363 | 746 | DE_HAVE_BARRIER_NUMBER, |
b411b363 PR |
747 | }; |
748 | ||
749 | enum epoch_event { | |
750 | EV_PUT, | |
751 | EV_GOT_BARRIER_NR, | |
b411b363 | 752 | EV_BECAME_LAST, |
b411b363 PR |
753 | EV_CLEANUP = 32, /* used as flag */ |
754 | }; | |
755 | ||
b411b363 PR |
756 | struct drbd_wq_barrier { |
757 | struct drbd_work w; | |
758 | struct completion done; | |
759 | }; | |
760 | ||
761 | struct digest_info { | |
762 | int digest_size; | |
763 | void *digest; | |
764 | }; | |
765 | ||
45bb912b LE |
766 | struct drbd_epoch_entry { |
767 | struct drbd_work w; | |
768 | struct hlist_node colision; | |
85719573 | 769 | struct drbd_epoch *epoch; /* for writes */ |
45bb912b LE |
770 | struct drbd_conf *mdev; |
771 | struct page *pages; | |
772 | atomic_t pending_bios; | |
773 | unsigned int size; | |
774 | /* see comments on ee flag bits below */ | |
775 | unsigned long flags; | |
776 | sector_t sector; | |
85719573 PR |
777 | union { |
778 | u64 block_id; | |
779 | struct digest_info *digest; | |
780 | }; | |
45bb912b LE |
781 | }; |
782 | ||
783 | /* ee flag bits. | |
784 | * While corresponding bios are in flight, the only modification will be | |
785 | * set_bit WAS_ERROR, which has to be atomic. | |
786 | * If no bios are in flight yet, or all have been completed, | |
787 | * non-atomic modification to ee->flags is ok. | |
788 | */ | |
b411b363 PR |
789 | enum { |
790 | __EE_CALL_AL_COMPLETE_IO, | |
b411b363 | 791 | __EE_MAY_SET_IN_SYNC, |
45bb912b | 792 | |
45bb912b LE |
793 | /* In case a barrier failed, |
794 | * we need to resubmit without the barrier flag. */ | |
795 | __EE_RESUBMITTED, | |
796 | ||
797 | /* we may have several bios per epoch entry. | |
798 | * if any of those fail, we set this flag atomically | |
799 | * from the endio callback */ | |
800 | __EE_WAS_ERROR, | |
c36c3ced LE |
801 | |
802 | /* This ee has a pointer to a digest instead of a block id */ | |
803 | __EE_HAS_DIGEST, | |
b411b363 PR |
804 | }; |
805 | #define EE_CALL_AL_COMPLETE_IO (1<<__EE_CALL_AL_COMPLETE_IO) | |
b411b363 | 806 | #define EE_MAY_SET_IN_SYNC (1<<__EE_MAY_SET_IN_SYNC) |
45bb912b LE |
807 | #define EE_RESUBMITTED (1<<__EE_RESUBMITTED) |
808 | #define EE_WAS_ERROR (1<<__EE_WAS_ERROR) | |
c36c3ced | 809 | #define EE_HAS_DIGEST (1<<__EE_HAS_DIGEST) |
b411b363 PR |
810 | |
811 | /* global flag bits */ | |
812 | enum { | |
813 | CREATE_BARRIER, /* next P_DATA is preceeded by a P_BARRIER */ | |
814 | SIGNAL_ASENDER, /* whether asender wants to be interrupted */ | |
815 | SEND_PING, /* whether asender should send a ping asap */ | |
816 | ||
b411b363 PR |
817 | UNPLUG_QUEUED, /* only relevant with kernel 2.4 */ |
818 | UNPLUG_REMOTE, /* sending a "UnplugRemote" could help */ | |
819 | MD_DIRTY, /* current uuids and flags not yet on disk */ | |
820 | DISCARD_CONCURRENT, /* Set on one node, cleared on the peer! */ | |
821 | USE_DEGR_WFC_T, /* degr-wfc-timeout instead of wfc-timeout. */ | |
822 | CLUSTER_ST_CHANGE, /* Cluster wide state change going on... */ | |
823 | CL_ST_CHG_SUCCESS, | |
824 | CL_ST_CHG_FAIL, | |
825 | CRASHED_PRIMARY, /* This node was a crashed primary. | |
826 | * Gets cleared when the state.conn | |
827 | * goes into C_CONNECTED state. */ | |
19f843aa | 828 | NO_BARRIER_SUPP, /* underlying block device doesn't implement barriers */ |
b411b363 PR |
829 | CONSIDER_RESYNC, |
830 | ||
a8a4e51e | 831 | MD_NO_FUA, /* Users wants us to not use FUA/FLUSH on meta data dev */ |
b411b363 PR |
832 | SUSPEND_IO, /* suspend application io */ |
833 | BITMAP_IO, /* suspend application io; | |
834 | once no more io in flight, start bitmap io */ | |
835 | BITMAP_IO_QUEUED, /* Started bitmap IO */ | |
82f59cc6 LE |
836 | GO_DISKLESS, /* Disk is being detached, on io-error or admin request. */ |
837 | WAS_IO_ERROR, /* Local disk failed returned IO error */ | |
b411b363 PR |
838 | RESYNC_AFTER_NEG, /* Resync after online grow after the attach&negotiate finished. */ |
839 | NET_CONGESTED, /* The data socket is congested */ | |
840 | ||
841 | CONFIG_PENDING, /* serialization of (re)configuration requests. | |
842 | * if set, also prevents the device from dying */ | |
843 | DEVICE_DYING, /* device became unconfigured, | |
844 | * but worker thread is still handling the cleanup. | |
845 | * reconfiguring (nl_disk_conf, nl_net_conf) is dissalowed, | |
846 | * while this is set. */ | |
847 | RESIZE_PENDING, /* Size change detected locally, waiting for the response from | |
848 | * the peer, if it changed there as well. */ | |
cf14c2e9 | 849 | CONN_DRY_RUN, /* Expect disconnect after resync handshake. */ |
309d1608 | 850 | GOT_PING_ACK, /* set when we receive a ping_ack packet, misc wait gets woken */ |
43a5182c | 851 | NEW_CUR_UUID, /* Create new current UUID when thawing IO */ |
0778286a | 852 | AL_SUSPENDED, /* Activity logging is currently suspended. */ |
370a43e7 | 853 | AHEAD_TO_SYNC_SOURCE, /* Ahead -> SyncSource queued */ |
b411b363 PR |
854 | }; |
855 | ||
856 | struct drbd_bitmap; /* opaque for drbd_conf */ | |
857 | ||
858 | /* TODO sort members for performance | |
859 | * MAYBE group them further */ | |
860 | ||
861 | /* THINK maybe we actually want to use the default "event/%s" worker threads | |
862 | * or similar in linux 2.6, which uses per cpu data and threads. | |
b411b363 PR |
863 | */ |
864 | struct drbd_work_queue { | |
865 | struct list_head q; | |
866 | struct semaphore s; /* producers up it, worker down()s it */ | |
867 | spinlock_t q_lock; /* to protect the list. */ | |
868 | }; | |
869 | ||
870 | struct drbd_socket { | |
871 | struct drbd_work_queue work; | |
872 | struct mutex mutex; | |
873 | struct socket *socket; | |
874 | /* this way we get our | |
875 | * send/receive buffers off the stack */ | |
876 | union p_polymorph sbuf; | |
877 | union p_polymorph rbuf; | |
878 | }; | |
879 | ||
880 | struct drbd_md { | |
881 | u64 md_offset; /* sector offset to 'super' block */ | |
882 | ||
883 | u64 la_size_sect; /* last agreed size, unit sectors */ | |
884 | u64 uuid[UI_SIZE]; | |
885 | u64 device_uuid; | |
886 | u32 flags; | |
887 | u32 md_size_sect; | |
888 | ||
889 | s32 al_offset; /* signed relative sector offset to al area */ | |
890 | s32 bm_offset; /* signed relative sector offset to bitmap */ | |
891 | ||
892 | /* u32 al_nr_extents; important for restoring the AL | |
893 | * is stored into sync_conf.al_extents, which in turn | |
894 | * gets applied to act_log->nr_elements | |
895 | */ | |
896 | }; | |
897 | ||
898 | /* for sync_conf and other types... */ | |
899 | #define NL_PACKET(name, number, fields) struct name { fields }; | |
900 | #define NL_INTEGER(pn,pr,member) int member; | |
901 | #define NL_INT64(pn,pr,member) __u64 member; | |
902 | #define NL_BIT(pn,pr,member) unsigned member:1; | |
903 | #define NL_STRING(pn,pr,member,len) unsigned char member[len]; int member ## _len; | |
904 | #include "linux/drbd_nl.h" | |
905 | ||
906 | struct drbd_backing_dev { | |
907 | struct block_device *backing_bdev; | |
908 | struct block_device *md_bdev; | |
b411b363 PR |
909 | struct drbd_md md; |
910 | struct disk_conf dc; /* The user provided config... */ | |
911 | sector_t known_size; /* last known size of that backing device */ | |
912 | }; | |
913 | ||
914 | struct drbd_md_io { | |
915 | struct drbd_conf *mdev; | |
916 | struct completion event; | |
917 | int error; | |
918 | }; | |
919 | ||
920 | struct bm_io_work { | |
921 | struct drbd_work w; | |
922 | char *why; | |
923 | int (*io_fn)(struct drbd_conf *mdev); | |
924 | void (*done)(struct drbd_conf *mdev, int rv); | |
925 | }; | |
926 | ||
927 | enum write_ordering_e { | |
928 | WO_none, | |
929 | WO_drain_io, | |
930 | WO_bdev_flush, | |
b411b363 PR |
931 | }; |
932 | ||
778f271d PR |
933 | struct fifo_buffer { |
934 | int *values; | |
935 | unsigned int head_index; | |
936 | unsigned int size; | |
937 | }; | |
938 | ||
b411b363 PR |
939 | struct drbd_conf { |
940 | /* things that are stored as / read from meta data on disk */ | |
941 | unsigned long flags; | |
942 | ||
943 | /* configured by drbdsetup */ | |
944 | struct net_conf *net_conf; /* protected by get_net_conf() and put_net_conf() */ | |
945 | struct syncer_conf sync_conf; | |
946 | struct drbd_backing_dev *ldev __protected_by(local); | |
947 | ||
948 | sector_t p_size; /* partner's disk size */ | |
949 | struct request_queue *rq_queue; | |
950 | struct block_device *this_bdev; | |
951 | struct gendisk *vdisk; | |
952 | ||
953 | struct drbd_socket data; /* data/barrier/cstate/parameter packets */ | |
954 | struct drbd_socket meta; /* ping/ack (metadata) packets */ | |
955 | int agreed_pro_version; /* actually used protocol version */ | |
956 | unsigned long last_received; /* in jiffies, either socket */ | |
957 | unsigned int ko_count; | |
958 | struct drbd_work resync_work, | |
959 | unplug_work, | |
e9e6f3ec | 960 | go_diskless, |
c4752ef1 PR |
961 | md_sync_work, |
962 | start_resync_work; | |
b411b363 PR |
963 | struct timer_list resync_timer; |
964 | struct timer_list md_sync_timer; | |
370a43e7 | 965 | struct timer_list start_resync_timer; |
ee15b038 LE |
966 | #ifdef DRBD_DEBUG_MD_SYNC |
967 | struct { | |
968 | unsigned int line; | |
969 | const char* func; | |
970 | } last_md_mark_dirty; | |
971 | #endif | |
b411b363 PR |
972 | |
973 | /* Used after attach while negotiating new disk state. */ | |
974 | union drbd_state new_state_tmp; | |
975 | ||
976 | union drbd_state state; | |
977 | wait_queue_head_t misc_wait; | |
978 | wait_queue_head_t state_wait; /* upon each state change. */ | |
84dfb9f5 | 979 | wait_queue_head_t net_cnt_wait; |
b411b363 PR |
980 | unsigned int send_cnt; |
981 | unsigned int recv_cnt; | |
982 | unsigned int read_cnt; | |
983 | unsigned int writ_cnt; | |
984 | unsigned int al_writ_cnt; | |
985 | unsigned int bm_writ_cnt; | |
986 | atomic_t ap_bio_cnt; /* Requests we need to complete */ | |
987 | atomic_t ap_pending_cnt; /* AP data packets on the wire, ack expected */ | |
988 | atomic_t rs_pending_cnt; /* RS request/data packets on the wire */ | |
989 | atomic_t unacked_cnt; /* Need to send replys for */ | |
990 | atomic_t local_cnt; /* Waiting for local completion */ | |
991 | atomic_t net_cnt; /* Users of net_conf */ | |
992 | spinlock_t req_lock; | |
993 | struct drbd_tl_epoch *unused_spare_tle; /* for pre-allocation */ | |
994 | struct drbd_tl_epoch *newest_tle; | |
995 | struct drbd_tl_epoch *oldest_tle; | |
996 | struct list_head out_of_sequence_requests; | |
997 | struct hlist_head *tl_hash; | |
998 | unsigned int tl_hash_s; | |
999 | ||
4b0715f0 | 1000 | /* blocks to resync in this run [unit BM_BLOCK_SIZE] */ |
b411b363 | 1001 | unsigned long rs_total; |
4b0715f0 | 1002 | /* number of resync blocks that failed in this run */ |
b411b363 PR |
1003 | unsigned long rs_failed; |
1004 | /* Syncer's start time [unit jiffies] */ | |
1005 | unsigned long rs_start; | |
1006 | /* cumulated time in PausedSyncX state [unit jiffies] */ | |
1007 | unsigned long rs_paused; | |
1d7734a0 LE |
1008 | /* skipped because csum was equal [unit BM_BLOCK_SIZE] */ |
1009 | unsigned long rs_same_csum; | |
1010 | #define DRBD_SYNC_MARKS 8 | |
1011 | #define DRBD_SYNC_MARK_STEP (3*HZ) | |
b411b363 | 1012 | /* block not up-to-date at mark [unit BM_BLOCK_SIZE] */ |
1d7734a0 | 1013 | unsigned long rs_mark_left[DRBD_SYNC_MARKS]; |
b411b363 | 1014 | /* marks's time [unit jiffies] */ |
1d7734a0 LE |
1015 | unsigned long rs_mark_time[DRBD_SYNC_MARKS]; |
1016 | /* current index into rs_mark_{left,time} */ | |
1017 | int rs_last_mark; | |
b411b363 PR |
1018 | |
1019 | /* where does the admin want us to start? (sector) */ | |
1020 | sector_t ov_start_sector; | |
1021 | /* where are we now? (sector) */ | |
1022 | sector_t ov_position; | |
1023 | /* Start sector of out of sync range (to merge printk reporting). */ | |
1024 | sector_t ov_last_oos_start; | |
1025 | /* size of out-of-sync range in sectors. */ | |
1026 | sector_t ov_last_oos_size; | |
1027 | unsigned long ov_left; /* in bits */ | |
1028 | struct crypto_hash *csums_tfm; | |
1029 | struct crypto_hash *verify_tfm; | |
1030 | ||
1031 | struct drbd_thread receiver; | |
1032 | struct drbd_thread worker; | |
1033 | struct drbd_thread asender; | |
1034 | struct drbd_bitmap *bitmap; | |
1035 | unsigned long bm_resync_fo; /* bit offset for drbd_bm_find_next */ | |
1036 | ||
1037 | /* Used to track operations of resync... */ | |
1038 | struct lru_cache *resync; | |
1039 | /* Number of locked elements in resync LRU */ | |
1040 | unsigned int resync_locked; | |
1041 | /* resync extent number waiting for application requests */ | |
1042 | unsigned int resync_wenr; | |
1043 | ||
1044 | int open_cnt; | |
1045 | u64 *p_uuid; | |
1046 | struct drbd_epoch *current_epoch; | |
1047 | spinlock_t epoch_lock; | |
1048 | unsigned int epochs; | |
1049 | enum write_ordering_e write_ordering; | |
85719573 PR |
1050 | struct list_head active_ee; /* IO in progress (P_DATA gets written to disk) */ |
1051 | struct list_head sync_ee; /* IO in progress (P_RS_DATA_REPLY gets written to disk) */ | |
b411b363 | 1052 | struct list_head done_ee; /* send ack */ |
85719573 | 1053 | struct list_head read_ee; /* IO in progress (any read) */ |
b411b363 PR |
1054 | struct list_head net_ee; /* zero-copy network send in progress */ |
1055 | struct hlist_head *ee_hash; /* is proteced by req_lock! */ | |
1056 | unsigned int ee_hash_s; | |
1057 | ||
1058 | /* this one is protected by ee_lock, single thread */ | |
1059 | struct drbd_epoch_entry *last_write_w_barrier; | |
1060 | ||
1061 | int next_barrier_nr; | |
1062 | struct hlist_head *app_reads_hash; /* is proteced by req_lock */ | |
1063 | struct list_head resync_reads; | |
435f0740 LE |
1064 | atomic_t pp_in_use; /* allocated from page pool */ |
1065 | atomic_t pp_in_use_by_net; /* sendpage()d, still referenced by tcp */ | |
b411b363 PR |
1066 | wait_queue_head_t ee_wait; |
1067 | struct page *md_io_page; /* one page buffer for md_io */ | |
1068 | struct page *md_io_tmpp; /* for logical_block_size != 512 */ | |
1069 | struct mutex md_io_mutex; /* protects the md_io_buffer */ | |
1070 | spinlock_t al_lock; | |
1071 | wait_queue_head_t al_wait; | |
1072 | struct lru_cache *act_log; /* activity log */ | |
1073 | unsigned int al_tr_number; | |
1074 | int al_tr_cycle; | |
1075 | int al_tr_pos; /* position of the next transaction in the journal */ | |
1076 | struct crypto_hash *cram_hmac_tfm; | |
1077 | struct crypto_hash *integrity_w_tfm; /* to be used by the worker thread */ | |
1078 | struct crypto_hash *integrity_r_tfm; /* to be used by the receiver thread */ | |
1079 | void *int_dig_out; | |
1080 | void *int_dig_in; | |
1081 | void *int_dig_vv; | |
1082 | wait_queue_head_t seq_wait; | |
1083 | atomic_t packet_seq; | |
1084 | unsigned int peer_seq; | |
1085 | spinlock_t peer_seq_lock; | |
1086 | unsigned int minor; | |
1087 | unsigned long comm_bm_set; /* communicated number of set bits. */ | |
1088 | cpumask_var_t cpu_mask; | |
1089 | struct bm_io_work bm_io_work; | |
1090 | u64 ed_uuid; /* UUID of the exposed data */ | |
1091 | struct mutex state_mutex; | |
1092 | char congestion_reason; /* Why we where congested... */ | |
1d7734a0 LE |
1093 | atomic_t rs_sect_in; /* for incoming resync data rate, SyncTarget */ |
1094 | atomic_t rs_sect_ev; /* for submitted resync data rate, both */ | |
1095 | int rs_last_sect_ev; /* counter to compare with */ | |
1096 | int rs_last_events; /* counter of read or write "events" (unit sectors) | |
1097 | * on the lower level device when we last looked. */ | |
1098 | int c_sync_rate; /* current resync rate after syncer throttle magic */ | |
778f271d PR |
1099 | struct fifo_buffer rs_plan_s; /* correction values of resync planer */ |
1100 | int rs_in_flight; /* resync sectors in flight (to proxy, in proxy and from proxy) */ | |
1101 | int rs_planed; /* resync sectors already planed */ | |
759fbdfb | 1102 | atomic_t ap_in_flight; /* App sectors in flight (waiting for ack) */ |
b411b363 PR |
1103 | }; |
1104 | ||
1105 | static inline struct drbd_conf *minor_to_mdev(unsigned int minor) | |
1106 | { | |
1107 | struct drbd_conf *mdev; | |
1108 | ||
1109 | mdev = minor < minor_count ? minor_table[minor] : NULL; | |
1110 | ||
1111 | return mdev; | |
1112 | } | |
1113 | ||
1114 | static inline unsigned int mdev_to_minor(struct drbd_conf *mdev) | |
1115 | { | |
1116 | return mdev->minor; | |
1117 | } | |
1118 | ||
1119 | /* returns 1 if it was successfull, | |
1120 | * returns 0 if there was no data socket. | |
1121 | * so wherever you are going to use the data.socket, e.g. do | |
1122 | * if (!drbd_get_data_sock(mdev)) | |
1123 | * return 0; | |
1124 | * CODE(); | |
1125 | * drbd_put_data_sock(mdev); | |
1126 | */ | |
1127 | static inline int drbd_get_data_sock(struct drbd_conf *mdev) | |
1128 | { | |
1129 | mutex_lock(&mdev->data.mutex); | |
1130 | /* drbd_disconnect() could have called drbd_free_sock() | |
1131 | * while we were waiting in down()... */ | |
1132 | if (unlikely(mdev->data.socket == NULL)) { | |
1133 | mutex_unlock(&mdev->data.mutex); | |
1134 | return 0; | |
1135 | } | |
1136 | return 1; | |
1137 | } | |
1138 | ||
1139 | static inline void drbd_put_data_sock(struct drbd_conf *mdev) | |
1140 | { | |
1141 | mutex_unlock(&mdev->data.mutex); | |
1142 | } | |
1143 | ||
1144 | /* | |
1145 | * function declarations | |
1146 | *************************/ | |
1147 | ||
1148 | /* drbd_main.c */ | |
1149 | ||
1150 | enum chg_state_flags { | |
1151 | CS_HARD = 1, | |
1152 | CS_VERBOSE = 2, | |
1153 | CS_WAIT_COMPLETE = 4, | |
1154 | CS_SERIALIZE = 8, | |
1155 | CS_ORDERED = CS_WAIT_COMPLETE + CS_SERIALIZE, | |
1156 | }; | |
1157 | ||
e89b591c PR |
1158 | enum dds_flags { |
1159 | DDSF_FORCED = 1, | |
1160 | DDSF_NO_RESYNC = 2, /* Do not run a resync for the new space */ | |
1161 | }; | |
1162 | ||
b411b363 | 1163 | extern void drbd_init_set_defaults(struct drbd_conf *mdev); |
bf885f8a AG |
1164 | extern enum drbd_state_rv drbd_change_state(struct drbd_conf *mdev, |
1165 | enum chg_state_flags f, | |
1166 | union drbd_state mask, | |
1167 | union drbd_state val); | |
b411b363 PR |
1168 | extern void drbd_force_state(struct drbd_conf *, union drbd_state, |
1169 | union drbd_state); | |
bf885f8a AG |
1170 | extern enum drbd_state_rv _drbd_request_state(struct drbd_conf *, |
1171 | union drbd_state, | |
1172 | union drbd_state, | |
1173 | enum chg_state_flags); | |
1174 | extern enum drbd_state_rv __drbd_set_state(struct drbd_conf *, union drbd_state, | |
1175 | enum chg_state_flags, | |
1176 | struct completion *done); | |
b411b363 PR |
1177 | extern void print_st_err(struct drbd_conf *, union drbd_state, |
1178 | union drbd_state, int); | |
1179 | extern int drbd_thread_start(struct drbd_thread *thi); | |
1180 | extern void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait); | |
1181 | #ifdef CONFIG_SMP | |
1182 | extern void drbd_thread_current_set_cpu(struct drbd_conf *mdev); | |
1183 | extern void drbd_calc_cpu_mask(struct drbd_conf *mdev); | |
1184 | #else | |
1185 | #define drbd_thread_current_set_cpu(A) ({}) | |
1186 | #define drbd_calc_cpu_mask(A) ({}) | |
1187 | #endif | |
1188 | extern void drbd_free_resources(struct drbd_conf *mdev); | |
1189 | extern void tl_release(struct drbd_conf *mdev, unsigned int barrier_nr, | |
1190 | unsigned int set_size); | |
1191 | extern void tl_clear(struct drbd_conf *mdev); | |
11b58e73 PR |
1192 | enum drbd_req_event; |
1193 | extern void tl_restart(struct drbd_conf *mdev, enum drbd_req_event what); | |
b411b363 PR |
1194 | extern void _tl_add_barrier(struct drbd_conf *, struct drbd_tl_epoch *); |
1195 | extern void drbd_free_sock(struct drbd_conf *mdev); | |
1196 | extern int drbd_send(struct drbd_conf *mdev, struct socket *sock, | |
1197 | void *buf, size_t size, unsigned msg_flags); | |
1198 | extern int drbd_send_protocol(struct drbd_conf *mdev); | |
1199 | extern int drbd_send_uuids(struct drbd_conf *mdev); | |
1200 | extern int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev); | |
5a22db89 | 1201 | extern int drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev); |
e89b591c | 1202 | extern int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags); |
b411b363 PR |
1203 | extern int _drbd_send_state(struct drbd_conf *mdev); |
1204 | extern int drbd_send_state(struct drbd_conf *mdev); | |
1205 | extern int _drbd_send_cmd(struct drbd_conf *mdev, struct socket *sock, | |
0b70a13d | 1206 | enum drbd_packets cmd, struct p_header80 *h, |
b411b363 PR |
1207 | size_t size, unsigned msg_flags); |
1208 | #define USE_DATA_SOCKET 1 | |
1209 | #define USE_META_SOCKET 0 | |
1210 | extern int drbd_send_cmd(struct drbd_conf *mdev, int use_data_socket, | |
0b70a13d | 1211 | enum drbd_packets cmd, struct p_header80 *h, |
b411b363 PR |
1212 | size_t size); |
1213 | extern int drbd_send_cmd2(struct drbd_conf *mdev, enum drbd_packets cmd, | |
1214 | char *data, size_t size); | |
1215 | extern int drbd_send_sync_param(struct drbd_conf *mdev, struct syncer_conf *sc); | |
1216 | extern int drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr, | |
1217 | u32 set_size); | |
1218 | extern int drbd_send_ack(struct drbd_conf *mdev, enum drbd_packets cmd, | |
1219 | struct drbd_epoch_entry *e); | |
1220 | extern int drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packets cmd, | |
1221 | struct p_block_req *rp); | |
1222 | extern int drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packets cmd, | |
2b2bf214 | 1223 | struct p_data *dp, int data_size); |
b411b363 PR |
1224 | extern int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packets cmd, |
1225 | sector_t sector, int blksize, u64 block_id); | |
73a01a18 | 1226 | extern int drbd_send_oos(struct drbd_conf *mdev, struct drbd_request *req); |
b411b363 PR |
1227 | extern int drbd_send_block(struct drbd_conf *mdev, enum drbd_packets cmd, |
1228 | struct drbd_epoch_entry *e); | |
1229 | extern int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req); | |
b411b363 PR |
1230 | extern int drbd_send_drequest(struct drbd_conf *mdev, int cmd, |
1231 | sector_t sector, int size, u64 block_id); | |
1232 | extern int drbd_send_drequest_csum(struct drbd_conf *mdev, | |
1233 | sector_t sector,int size, | |
1234 | void *digest, int digest_size, | |
1235 | enum drbd_packets cmd); | |
1236 | extern int drbd_send_ov_request(struct drbd_conf *mdev,sector_t sector,int size); | |
1237 | ||
1238 | extern int drbd_send_bitmap(struct drbd_conf *mdev); | |
1239 | extern int _drbd_send_bitmap(struct drbd_conf *mdev); | |
bf885f8a | 1240 | extern int drbd_send_sr_reply(struct drbd_conf *mdev, enum drbd_state_rv retcode); |
b411b363 PR |
1241 | extern void drbd_free_bc(struct drbd_backing_dev *ldev); |
1242 | extern void drbd_mdev_cleanup(struct drbd_conf *mdev); | |
62b0da3a | 1243 | void drbd_print_uuids(struct drbd_conf *mdev, const char *text); |
b411b363 PR |
1244 | |
1245 | /* drbd_meta-data.c (still in drbd_main.c) */ | |
1246 | extern void drbd_md_sync(struct drbd_conf *mdev); | |
1247 | extern int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev); | |
b411b363 PR |
1248 | extern void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local); |
1249 | extern void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local); | |
1250 | extern void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local); | |
1251 | extern void _drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local); | |
1252 | extern void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local); | |
1253 | extern void drbd_md_set_flag(struct drbd_conf *mdev, int flags) __must_hold(local); | |
1254 | extern void drbd_md_clear_flag(struct drbd_conf *mdev, int flags)__must_hold(local); | |
1255 | extern int drbd_md_test_flag(struct drbd_backing_dev *, int); | |
ee15b038 | 1256 | #ifndef DRBD_DEBUG_MD_SYNC |
b411b363 | 1257 | extern void drbd_md_mark_dirty(struct drbd_conf *mdev); |
ee15b038 LE |
1258 | #else |
1259 | #define drbd_md_mark_dirty(m) drbd_md_mark_dirty_(m, __LINE__ , __func__ ) | |
1260 | extern void drbd_md_mark_dirty_(struct drbd_conf *mdev, | |
1261 | unsigned int line, const char *func); | |
1262 | #endif | |
b411b363 PR |
1263 | extern void drbd_queue_bitmap_io(struct drbd_conf *mdev, |
1264 | int (*io_fn)(struct drbd_conf *), | |
1265 | void (*done)(struct drbd_conf *, int), | |
1266 | char *why); | |
1267 | extern int drbd_bmio_set_n_write(struct drbd_conf *mdev); | |
1268 | extern int drbd_bmio_clear_n_write(struct drbd_conf *mdev); | |
1269 | extern int drbd_bitmap_io(struct drbd_conf *mdev, int (*io_fn)(struct drbd_conf *), char *why); | |
e9e6f3ec | 1270 | extern void drbd_go_diskless(struct drbd_conf *mdev); |
82f59cc6 | 1271 | extern void drbd_ldev_destroy(struct drbd_conf *mdev); |
b411b363 PR |
1272 | |
1273 | ||
1274 | /* Meta data layout | |
1275 | We reserve a 128MB Block (4k aligned) | |
1276 | * either at the end of the backing device | |
3ad2f3fb | 1277 | * or on a separate meta data device. */ |
b411b363 PR |
1278 | |
1279 | #define MD_RESERVED_SECT (128LU << 11) /* 128 MB, unit sectors */ | |
1280 | /* The following numbers are sectors */ | |
1281 | #define MD_AL_OFFSET 8 /* 8 Sectors after start of meta area */ | |
1282 | #define MD_AL_MAX_SIZE 64 /* = 32 kb LOG ~ 3776 extents ~ 14 GB Storage */ | |
1283 | /* Allows up to about 3.8TB */ | |
1284 | #define MD_BM_OFFSET (MD_AL_OFFSET + MD_AL_MAX_SIZE) | |
1285 | ||
1286 | /* Since the smalles IO unit is usually 512 byte */ | |
1287 | #define MD_SECTOR_SHIFT 9 | |
1288 | #define MD_SECTOR_SIZE (1<<MD_SECTOR_SHIFT) | |
1289 | ||
1290 | /* activity log */ | |
1291 | #define AL_EXTENTS_PT ((MD_SECTOR_SIZE-12)/8-1) /* 61 ; Extents per 512B sector */ | |
1292 | #define AL_EXTENT_SHIFT 22 /* One extent represents 4M Storage */ | |
1293 | #define AL_EXTENT_SIZE (1<<AL_EXTENT_SHIFT) | |
1294 | ||
1295 | #if BITS_PER_LONG == 32 | |
1296 | #define LN2_BPL 5 | |
1297 | #define cpu_to_lel(A) cpu_to_le32(A) | |
1298 | #define lel_to_cpu(A) le32_to_cpu(A) | |
1299 | #elif BITS_PER_LONG == 64 | |
1300 | #define LN2_BPL 6 | |
1301 | #define cpu_to_lel(A) cpu_to_le64(A) | |
1302 | #define lel_to_cpu(A) le64_to_cpu(A) | |
1303 | #else | |
1304 | #error "LN2 of BITS_PER_LONG unknown!" | |
1305 | #endif | |
1306 | ||
1307 | /* resync bitmap */ | |
1308 | /* 16MB sized 'bitmap extent' to track syncer usage */ | |
1309 | struct bm_extent { | |
1310 | int rs_left; /* number of bits set (out of sync) in this extent. */ | |
1311 | int rs_failed; /* number of failed resync requests in this extent. */ | |
1312 | unsigned long flags; | |
1313 | struct lc_element lce; | |
1314 | }; | |
1315 | ||
1316 | #define BME_NO_WRITES 0 /* bm_extent.flags: no more requests on this one! */ | |
1317 | #define BME_LOCKED 1 /* bm_extent.flags: syncer active on this one. */ | |
e3555d85 | 1318 | #define BME_PRIORITY 2 /* finish resync IO on this extent ASAP! App IO waiting! */ |
b411b363 PR |
1319 | |
1320 | /* drbd_bitmap.c */ | |
1321 | /* | |
1322 | * We need to store one bit for a block. | |
1323 | * Example: 1GB disk @ 4096 byte blocks ==> we need 32 KB bitmap. | |
1324 | * Bit 0 ==> local node thinks this block is binary identical on both nodes | |
1325 | * Bit 1 ==> local node thinks this block needs to be synced. | |
1326 | */ | |
1327 | ||
8e26f9cc PR |
1328 | #define SLEEP_TIME (HZ/10) |
1329 | ||
b411b363 PR |
1330 | #define BM_BLOCK_SHIFT 12 /* 4k per bit */ |
1331 | #define BM_BLOCK_SIZE (1<<BM_BLOCK_SHIFT) | |
1332 | /* (9+3) : 512 bytes @ 8 bits; representing 16M storage | |
1333 | * per sector of on disk bitmap */ | |
1334 | #define BM_EXT_SHIFT (BM_BLOCK_SHIFT + MD_SECTOR_SHIFT + 3) /* = 24 */ | |
1335 | #define BM_EXT_SIZE (1<<BM_EXT_SHIFT) | |
1336 | ||
1337 | #if (BM_EXT_SHIFT != 24) || (BM_BLOCK_SHIFT != 12) | |
1338 | #error "HAVE YOU FIXED drbdmeta AS WELL??" | |
1339 | #endif | |
1340 | ||
1341 | /* thus many _storage_ sectors are described by one bit */ | |
1342 | #define BM_SECT_TO_BIT(x) ((x)>>(BM_BLOCK_SHIFT-9)) | |
1343 | #define BM_BIT_TO_SECT(x) ((sector_t)(x)<<(BM_BLOCK_SHIFT-9)) | |
1344 | #define BM_SECT_PER_BIT BM_BIT_TO_SECT(1) | |
1345 | ||
1346 | /* bit to represented kilo byte conversion */ | |
1347 | #define Bit2KB(bits) ((bits)<<(BM_BLOCK_SHIFT-10)) | |
1348 | ||
1349 | /* in which _bitmap_ extent (resp. sector) the bit for a certain | |
1350 | * _storage_ sector is located in */ | |
1351 | #define BM_SECT_TO_EXT(x) ((x)>>(BM_EXT_SHIFT-9)) | |
1352 | ||
1353 | /* how much _storage_ sectors we have per bitmap sector */ | |
1354 | #define BM_EXT_TO_SECT(x) ((sector_t)(x) << (BM_EXT_SHIFT-9)) | |
1355 | #define BM_SECT_PER_EXT BM_EXT_TO_SECT(1) | |
1356 | ||
1357 | /* in one sector of the bitmap, we have this many activity_log extents. */ | |
1358 | #define AL_EXT_PER_BM_SECT (1 << (BM_EXT_SHIFT - AL_EXTENT_SHIFT)) | |
1359 | #define BM_WORDS_PER_AL_EXT (1 << (AL_EXTENT_SHIFT-BM_BLOCK_SHIFT-LN2_BPL)) | |
1360 | ||
1361 | #define BM_BLOCKS_PER_BM_EXT_B (BM_EXT_SHIFT - BM_BLOCK_SHIFT) | |
1362 | #define BM_BLOCKS_PER_BM_EXT_MASK ((1<<BM_BLOCKS_PER_BM_EXT_B) - 1) | |
1363 | ||
1364 | /* the extent in "PER_EXTENT" below is an activity log extent | |
1365 | * we need that many (long words/bytes) to store the bitmap | |
1366 | * of one AL_EXTENT_SIZE chunk of storage. | |
1367 | * we can store the bitmap for that many AL_EXTENTS within | |
1368 | * one sector of the _on_disk_ bitmap: | |
1369 | * bit 0 bit 37 bit 38 bit (512*8)-1 | |
1370 | * ...|........|........|.. // ..|........| | |
1371 | * sect. 0 `296 `304 ^(512*8*8)-1 | |
1372 | * | |
1373 | #define BM_WORDS_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / BITS_PER_LONG ) | |
1374 | #define BM_BYTES_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / 8 ) // 128 | |
1375 | #define BM_EXT_PER_SECT ( 512 / BM_BYTES_PER_EXTENT ) // 4 | |
1376 | */ | |
1377 | ||
1378 | #define DRBD_MAX_SECTORS_32 (0xffffffffLU) | |
1379 | #define DRBD_MAX_SECTORS_BM \ | |
1380 | ((MD_RESERVED_SECT - MD_BM_OFFSET) * (1LL<<(BM_EXT_SHIFT-9))) | |
1381 | #if DRBD_MAX_SECTORS_BM < DRBD_MAX_SECTORS_32 | |
1382 | #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_BM | |
1383 | #define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_BM | |
36bfc7e2 | 1384 | #elif !defined(CONFIG_LBDAF) && BITS_PER_LONG == 32 |
b411b363 PR |
1385 | #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_32 |
1386 | #define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_32 | |
1387 | #else | |
1388 | #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_BM | |
1389 | /* 16 TB in units of sectors */ | |
1390 | #if BITS_PER_LONG == 32 | |
1391 | /* adjust by one page worth of bitmap, | |
1392 | * so we won't wrap around in drbd_bm_find_next_bit. | |
1393 | * you should use 64bit OS for that much storage, anyways. */ | |
1394 | #define DRBD_MAX_SECTORS_FLEX BM_BIT_TO_SECT(0xffff7fff) | |
1395 | #else | |
4b0715f0 LE |
1396 | /* we allow up to 1 PiB now on 64bit architecture with "flexible" meta data */ |
1397 | #define DRBD_MAX_SECTORS_FLEX (1UL << 51) | |
1398 | /* corresponds to (1UL << 38) bits right now. */ | |
b411b363 PR |
1399 | #endif |
1400 | #endif | |
1401 | ||
1402 | /* Sector shift value for the "hash" functions of tl_hash and ee_hash tables. | |
d5373389 | 1403 | * With a value of 8 all IO in one 128K block make it to the same slot of the |
b411b363 | 1404 | * hash table. */ |
d5373389 | 1405 | #define HT_SHIFT 8 |
1816a2b4 | 1406 | #define DRBD_MAX_BIO_SIZE (1U<<(9+HT_SHIFT)) |
b411b363 | 1407 | |
d5373389 PR |
1408 | #define DRBD_MAX_SIZE_H80_PACKET (1 << 15) /* The old header only allows packets up to 32Kib data */ |
1409 | ||
b411b363 PR |
1410 | /* Number of elements in the app_reads_hash */ |
1411 | #define APP_R_HSIZE 15 | |
1412 | ||
1413 | extern int drbd_bm_init(struct drbd_conf *mdev); | |
02d9a94b | 1414 | extern int drbd_bm_resize(struct drbd_conf *mdev, sector_t sectors, int set_new_bits); |
b411b363 PR |
1415 | extern void drbd_bm_cleanup(struct drbd_conf *mdev); |
1416 | extern void drbd_bm_set_all(struct drbd_conf *mdev); | |
1417 | extern void drbd_bm_clear_all(struct drbd_conf *mdev); | |
4b0715f0 | 1418 | /* set/clear/test only a few bits at a time */ |
b411b363 PR |
1419 | extern int drbd_bm_set_bits( |
1420 | struct drbd_conf *mdev, unsigned long s, unsigned long e); | |
1421 | extern int drbd_bm_clear_bits( | |
1422 | struct drbd_conf *mdev, unsigned long s, unsigned long e); | |
4b0715f0 LE |
1423 | extern int drbd_bm_count_bits( |
1424 | struct drbd_conf *mdev, const unsigned long s, const unsigned long e); | |
1425 | /* bm_set_bits variant for use while holding drbd_bm_lock, | |
1426 | * may process the whole bitmap in one go */ | |
b411b363 PR |
1427 | extern void _drbd_bm_set_bits(struct drbd_conf *mdev, |
1428 | const unsigned long s, const unsigned long e); | |
1429 | extern int drbd_bm_test_bit(struct drbd_conf *mdev, unsigned long bitnr); | |
1430 | extern int drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr); | |
19f843aa | 1431 | extern int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(local); |
b411b363 PR |
1432 | extern int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local); |
1433 | extern int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local); | |
1434 | extern unsigned long drbd_bm_ALe_set_all(struct drbd_conf *mdev, | |
1435 | unsigned long al_enr); | |
1436 | extern size_t drbd_bm_words(struct drbd_conf *mdev); | |
1437 | extern unsigned long drbd_bm_bits(struct drbd_conf *mdev); | |
1438 | extern sector_t drbd_bm_capacity(struct drbd_conf *mdev); | |
4b0715f0 LE |
1439 | |
1440 | #define DRBD_END_OF_BITMAP (~(unsigned long)0) | |
b411b363 PR |
1441 | extern unsigned long drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo); |
1442 | /* bm_find_next variants for use while you hold drbd_bm_lock() */ | |
1443 | extern unsigned long _drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo); | |
1444 | extern unsigned long _drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo); | |
0778286a | 1445 | extern unsigned long _drbd_bm_total_weight(struct drbd_conf *mdev); |
b411b363 PR |
1446 | extern unsigned long drbd_bm_total_weight(struct drbd_conf *mdev); |
1447 | extern int drbd_bm_rs_done(struct drbd_conf *mdev); | |
1448 | /* for receive_bitmap */ | |
1449 | extern void drbd_bm_merge_lel(struct drbd_conf *mdev, size_t offset, | |
1450 | size_t number, unsigned long *buffer); | |
19f843aa | 1451 | /* for _drbd_send_bitmap */ |
b411b363 PR |
1452 | extern void drbd_bm_get_lel(struct drbd_conf *mdev, size_t offset, |
1453 | size_t number, unsigned long *buffer); | |
1454 | ||
1455 | extern void drbd_bm_lock(struct drbd_conf *mdev, char *why); | |
1456 | extern void drbd_bm_unlock(struct drbd_conf *mdev); | |
b411b363 PR |
1457 | /* drbd_main.c */ |
1458 | ||
1459 | extern struct kmem_cache *drbd_request_cache; | |
1460 | extern struct kmem_cache *drbd_ee_cache; /* epoch entries */ | |
1461 | extern struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */ | |
1462 | extern struct kmem_cache *drbd_al_ext_cache; /* activity log extents */ | |
1463 | extern mempool_t *drbd_request_mempool; | |
1464 | extern mempool_t *drbd_ee_mempool; | |
1465 | ||
1466 | extern struct page *drbd_pp_pool; /* drbd's page pool */ | |
1467 | extern spinlock_t drbd_pp_lock; | |
1468 | extern int drbd_pp_vacant; | |
1469 | extern wait_queue_head_t drbd_pp_wait; | |
1470 | ||
1471 | extern rwlock_t global_state_lock; | |
1472 | ||
1473 | extern struct drbd_conf *drbd_new_device(unsigned int minor); | |
1474 | extern void drbd_free_mdev(struct drbd_conf *mdev); | |
1475 | ||
1476 | extern int proc_details; | |
1477 | ||
1478 | /* drbd_req */ | |
2f58dcfc | 1479 | extern int drbd_make_request(struct request_queue *q, struct bio *bio); |
b411b363 PR |
1480 | extern int drbd_read_remote(struct drbd_conf *mdev, struct drbd_request *req); |
1481 | extern int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct bio_vec *bvec); | |
1482 | extern int is_valid_ar_handle(struct drbd_request *, sector_t); | |
1483 | ||
1484 | ||
1485 | /* drbd_nl.c */ | |
1486 | extern void drbd_suspend_io(struct drbd_conf *mdev); | |
1487 | extern void drbd_resume_io(struct drbd_conf *mdev); | |
1488 | extern char *ppsize(char *buf, unsigned long long size); | |
a393db6f | 1489 | extern sector_t drbd_new_dev_size(struct drbd_conf *, struct drbd_backing_dev *, int); |
b411b363 | 1490 | enum determine_dev_size { dev_size_error = -1, unchanged = 0, shrunk = 1, grew = 2 }; |
d845030f | 1491 | extern enum determine_dev_size drbd_determin_dev_size(struct drbd_conf *, enum dds_flags) __must_hold(local); |
b411b363 PR |
1492 | extern void resync_after_online_grow(struct drbd_conf *); |
1493 | extern void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int) __must_hold(local); | |
bf885f8a AG |
1494 | extern enum drbd_state_rv drbd_set_role(struct drbd_conf *mdev, |
1495 | enum drbd_role new_role, | |
1496 | int force); | |
87f7be4c PR |
1497 | extern enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev); |
1498 | extern void drbd_try_outdate_peer_async(struct drbd_conf *mdev); | |
b411b363 PR |
1499 | extern int drbd_khelper(struct drbd_conf *mdev, char *cmd); |
1500 | ||
1501 | /* drbd_worker.c */ | |
1502 | extern int drbd_worker(struct drbd_thread *thi); | |
1503 | extern int drbd_alter_sa(struct drbd_conf *mdev, int na); | |
1504 | extern void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side); | |
1505 | extern void resume_next_sg(struct drbd_conf *mdev); | |
1506 | extern void suspend_other_sg(struct drbd_conf *mdev); | |
1507 | extern int drbd_resync_finished(struct drbd_conf *mdev); | |
1508 | /* maybe rather drbd_main.c ? */ | |
1509 | extern int drbd_md_sync_page_io(struct drbd_conf *mdev, | |
1510 | struct drbd_backing_dev *bdev, sector_t sector, int rw); | |
1511 | extern void drbd_ov_oos_found(struct drbd_conf*, sector_t, int); | |
9bd28d3c | 1512 | extern void drbd_rs_controller_reset(struct drbd_conf *mdev); |
b411b363 PR |
1513 | |
1514 | static inline void ov_oos_print(struct drbd_conf *mdev) | |
1515 | { | |
1516 | if (mdev->ov_last_oos_size) { | |
1517 | dev_err(DEV, "Out of sync: start=%llu, size=%lu (sectors)\n", | |
1518 | (unsigned long long)mdev->ov_last_oos_start, | |
1519 | (unsigned long)mdev->ov_last_oos_size); | |
1520 | } | |
1521 | mdev->ov_last_oos_size=0; | |
1522 | } | |
1523 | ||
1524 | ||
45bb912b LE |
1525 | extern void drbd_csum_bio(struct drbd_conf *, struct crypto_hash *, struct bio *, void *); |
1526 | extern void drbd_csum_ee(struct drbd_conf *, struct crypto_hash *, struct drbd_epoch_entry *, void *); | |
b411b363 PR |
1527 | /* worker callbacks */ |
1528 | extern int w_req_cancel_conflict(struct drbd_conf *, struct drbd_work *, int); | |
1529 | extern int w_read_retry_remote(struct drbd_conf *, struct drbd_work *, int); | |
1530 | extern int w_e_end_data_req(struct drbd_conf *, struct drbd_work *, int); | |
1531 | extern int w_e_end_rsdata_req(struct drbd_conf *, struct drbd_work *, int); | |
1532 | extern int w_e_end_csum_rs_req(struct drbd_conf *, struct drbd_work *, int); | |
1533 | extern int w_e_end_ov_reply(struct drbd_conf *, struct drbd_work *, int); | |
1534 | extern int w_e_end_ov_req(struct drbd_conf *, struct drbd_work *, int); | |
1535 | extern int w_ov_finished(struct drbd_conf *, struct drbd_work *, int); | |
794abb75 | 1536 | extern int w_resync_timer(struct drbd_conf *, struct drbd_work *, int); |
b411b363 | 1537 | extern int w_resume_next_sg(struct drbd_conf *, struct drbd_work *, int); |
b411b363 | 1538 | extern int w_send_write_hint(struct drbd_conf *, struct drbd_work *, int); |
b411b363 PR |
1539 | extern int w_send_dblock(struct drbd_conf *, struct drbd_work *, int); |
1540 | extern int w_send_barrier(struct drbd_conf *, struct drbd_work *, int); | |
1541 | extern int w_send_read_req(struct drbd_conf *, struct drbd_work *, int); | |
1542 | extern int w_prev_work_done(struct drbd_conf *, struct drbd_work *, int); | |
1543 | extern int w_e_reissue(struct drbd_conf *, struct drbd_work *, int); | |
265be2d0 | 1544 | extern int w_restart_disk_io(struct drbd_conf *, struct drbd_work *, int); |
73a01a18 | 1545 | extern int w_send_oos(struct drbd_conf *, struct drbd_work *, int); |
c4752ef1 | 1546 | extern int w_start_resync(struct drbd_conf *, struct drbd_work *, int); |
b411b363 PR |
1547 | |
1548 | extern void resync_timer_fn(unsigned long data); | |
370a43e7 | 1549 | extern void start_resync_timer_fn(unsigned long data); |
b411b363 PR |
1550 | |
1551 | /* drbd_receiver.c */ | |
e3555d85 | 1552 | extern int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector); |
45bb912b LE |
1553 | extern int drbd_submit_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, |
1554 | const unsigned rw, const int fault_type); | |
b411b363 PR |
1555 | extern int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list); |
1556 | extern struct drbd_epoch_entry *drbd_alloc_ee(struct drbd_conf *mdev, | |
1557 | u64 id, | |
1558 | sector_t sector, | |
1559 | unsigned int data_size, | |
1560 | gfp_t gfp_mask) __must_hold(local); | |
435f0740 LE |
1561 | extern void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, |
1562 | int is_net); | |
1563 | #define drbd_free_ee(m,e) drbd_free_some_ee(m, e, 0) | |
1564 | #define drbd_free_net_ee(m,e) drbd_free_some_ee(m, e, 1) | |
b411b363 PR |
1565 | extern void drbd_wait_ee_list_empty(struct drbd_conf *mdev, |
1566 | struct list_head *head); | |
1567 | extern void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, | |
1568 | struct list_head *head); | |
1569 | extern void drbd_set_recv_tcq(struct drbd_conf *mdev, int tcq_enabled); | |
1570 | extern void _drbd_clear_done_ee(struct drbd_conf *mdev, struct list_head *to_be_freed); | |
1571 | extern void drbd_flush_workqueue(struct drbd_conf *mdev); | |
f70b3511 | 1572 | extern void drbd_free_tl_hash(struct drbd_conf *mdev); |
b411b363 PR |
1573 | |
1574 | /* yes, there is kernel_setsockopt, but only since 2.6.18. we don't need to | |
1575 | * mess with get_fs/set_fs, we know we are KERNEL_DS always. */ | |
1576 | static inline int drbd_setsockopt(struct socket *sock, int level, int optname, | |
1577 | char __user *optval, int optlen) | |
1578 | { | |
1579 | int err; | |
1580 | if (level == SOL_SOCKET) | |
1581 | err = sock_setsockopt(sock, level, optname, optval, optlen); | |
1582 | else | |
1583 | err = sock->ops->setsockopt(sock, level, optname, optval, | |
1584 | optlen); | |
1585 | return err; | |
1586 | } | |
1587 | ||
1588 | static inline void drbd_tcp_cork(struct socket *sock) | |
1589 | { | |
1590 | int __user val = 1; | |
1591 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK, | |
1592 | (char __user *)&val, sizeof(val)); | |
1593 | } | |
1594 | ||
1595 | static inline void drbd_tcp_uncork(struct socket *sock) | |
1596 | { | |
1597 | int __user val = 0; | |
1598 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK, | |
1599 | (char __user *)&val, sizeof(val)); | |
1600 | } | |
1601 | ||
1602 | static inline void drbd_tcp_nodelay(struct socket *sock) | |
1603 | { | |
1604 | int __user val = 1; | |
1605 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_NODELAY, | |
1606 | (char __user *)&val, sizeof(val)); | |
1607 | } | |
1608 | ||
1609 | static inline void drbd_tcp_quickack(struct socket *sock) | |
1610 | { | |
344fa462 | 1611 | int __user val = 2; |
b411b363 PR |
1612 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_QUICKACK, |
1613 | (char __user *)&val, sizeof(val)); | |
1614 | } | |
1615 | ||
1616 | void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo); | |
1617 | ||
1618 | /* drbd_proc.c */ | |
1619 | extern struct proc_dir_entry *drbd_proc; | |
7d4e9d09 | 1620 | extern const struct file_operations drbd_proc_fops; |
b411b363 PR |
1621 | extern const char *drbd_conn_str(enum drbd_conns s); |
1622 | extern const char *drbd_role_str(enum drbd_role s); | |
1623 | ||
1624 | /* drbd_actlog.c */ | |
1625 | extern void drbd_al_begin_io(struct drbd_conf *mdev, sector_t sector); | |
1626 | extern void drbd_al_complete_io(struct drbd_conf *mdev, sector_t sector); | |
1627 | extern void drbd_rs_complete_io(struct drbd_conf *mdev, sector_t sector); | |
1628 | extern int drbd_rs_begin_io(struct drbd_conf *mdev, sector_t sector); | |
1629 | extern int drbd_try_rs_begin_io(struct drbd_conf *mdev, sector_t sector); | |
1630 | extern void drbd_rs_cancel_all(struct drbd_conf *mdev); | |
1631 | extern int drbd_rs_del_all(struct drbd_conf *mdev); | |
1632 | extern void drbd_rs_failed_io(struct drbd_conf *mdev, | |
1633 | sector_t sector, int size); | |
1634 | extern int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *); | |
ea5442af | 1635 | extern void drbd_advance_rs_marks(struct drbd_conf *mdev, unsigned long still_to_go); |
b411b363 PR |
1636 | extern void __drbd_set_in_sync(struct drbd_conf *mdev, sector_t sector, |
1637 | int size, const char *file, const unsigned int line); | |
1638 | #define drbd_set_in_sync(mdev, sector, size) \ | |
1639 | __drbd_set_in_sync(mdev, sector, size, __FILE__, __LINE__) | |
73a01a18 | 1640 | extern int __drbd_set_out_of_sync(struct drbd_conf *mdev, sector_t sector, |
b411b363 PR |
1641 | int size, const char *file, const unsigned int line); |
1642 | #define drbd_set_out_of_sync(mdev, sector, size) \ | |
1643 | __drbd_set_out_of_sync(mdev, sector, size, __FILE__, __LINE__) | |
1644 | extern void drbd_al_apply_to_bm(struct drbd_conf *mdev); | |
b411b363 PR |
1645 | extern void drbd_al_shrink(struct drbd_conf *mdev); |
1646 | ||
1647 | ||
1648 | /* drbd_nl.c */ | |
1649 | ||
1650 | void drbd_nl_cleanup(void); | |
1651 | int __init drbd_nl_init(void); | |
1652 | void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state); | |
1653 | void drbd_bcast_sync_progress(struct drbd_conf *mdev); | |
1654 | void drbd_bcast_ee(struct drbd_conf *mdev, | |
1655 | const char *reason, const int dgs, | |
1656 | const char* seen_hash, const char* calc_hash, | |
1657 | const struct drbd_epoch_entry* e); | |
1658 | ||
1659 | ||
1660 | /** | |
1661 | * DOC: DRBD State macros | |
1662 | * | |
1663 | * These macros are used to express state changes in easily readable form. | |
1664 | * | |
1665 | * The NS macros expand to a mask and a value, that can be bit ored onto the | |
1666 | * current state as soon as the spinlock (req_lock) was taken. | |
1667 | * | |
1668 | * The _NS macros are used for state functions that get called with the | |
1669 | * spinlock. These macros expand directly to the new state value. | |
1670 | * | |
1671 | * Besides the basic forms NS() and _NS() additional _?NS[23] are defined | |
1672 | * to express state changes that affect more than one aspect of the state. | |
1673 | * | |
1674 | * E.g. NS2(conn, C_CONNECTED, peer, R_SECONDARY) | |
1675 | * Means that the network connection was established and that the peer | |
1676 | * is in secondary role. | |
1677 | */ | |
1678 | #define role_MASK R_MASK | |
1679 | #define peer_MASK R_MASK | |
1680 | #define disk_MASK D_MASK | |
1681 | #define pdsk_MASK D_MASK | |
1682 | #define conn_MASK C_MASK | |
1683 | #define susp_MASK 1 | |
1684 | #define user_isp_MASK 1 | |
1685 | #define aftr_isp_MASK 1 | |
fb22c402 PR |
1686 | #define susp_nod_MASK 1 |
1687 | #define susp_fen_MASK 1 | |
b411b363 PR |
1688 | |
1689 | #define NS(T, S) \ | |
1690 | ({ union drbd_state mask; mask.i = 0; mask.T = T##_MASK; mask; }), \ | |
1691 | ({ union drbd_state val; val.i = 0; val.T = (S); val; }) | |
1692 | #define NS2(T1, S1, T2, S2) \ | |
1693 | ({ union drbd_state mask; mask.i = 0; mask.T1 = T1##_MASK; \ | |
1694 | mask.T2 = T2##_MASK; mask; }), \ | |
1695 | ({ union drbd_state val; val.i = 0; val.T1 = (S1); \ | |
1696 | val.T2 = (S2); val; }) | |
1697 | #define NS3(T1, S1, T2, S2, T3, S3) \ | |
1698 | ({ union drbd_state mask; mask.i = 0; mask.T1 = T1##_MASK; \ | |
1699 | mask.T2 = T2##_MASK; mask.T3 = T3##_MASK; mask; }), \ | |
1700 | ({ union drbd_state val; val.i = 0; val.T1 = (S1); \ | |
1701 | val.T2 = (S2); val.T3 = (S3); val; }) | |
1702 | ||
1703 | #define _NS(D, T, S) \ | |
1704 | D, ({ union drbd_state __ns; __ns.i = D->state.i; __ns.T = (S); __ns; }) | |
1705 | #define _NS2(D, T1, S1, T2, S2) \ | |
1706 | D, ({ union drbd_state __ns; __ns.i = D->state.i; __ns.T1 = (S1); \ | |
1707 | __ns.T2 = (S2); __ns; }) | |
1708 | #define _NS3(D, T1, S1, T2, S2, T3, S3) \ | |
1709 | D, ({ union drbd_state __ns; __ns.i = D->state.i; __ns.T1 = (S1); \ | |
1710 | __ns.T2 = (S2); __ns.T3 = (S3); __ns; }) | |
1711 | ||
1712 | /* | |
1713 | * inline helper functions | |
1714 | *************************/ | |
1715 | ||
45bb912b LE |
1716 | /* see also page_chain_add and friends in drbd_receiver.c */ |
1717 | static inline struct page *page_chain_next(struct page *page) | |
1718 | { | |
1719 | return (struct page *)page_private(page); | |
1720 | } | |
1721 | #define page_chain_for_each(page) \ | |
1722 | for (; page && ({ prefetch(page_chain_next(page)); 1; }); \ | |
1723 | page = page_chain_next(page)) | |
1724 | #define page_chain_for_each_safe(page, n) \ | |
1725 | for (; page && ({ n = page_chain_next(page); 1; }); page = n) | |
1726 | ||
1727 | static inline int drbd_bio_has_active_page(struct bio *bio) | |
1728 | { | |
1729 | struct bio_vec *bvec; | |
1730 | int i; | |
1731 | ||
1732 | __bio_for_each_segment(bvec, bio, i, 0) { | |
1733 | if (page_count(bvec->bv_page) > 1) | |
1734 | return 1; | |
1735 | } | |
1736 | ||
1737 | return 0; | |
1738 | } | |
1739 | ||
1740 | static inline int drbd_ee_has_active_page(struct drbd_epoch_entry *e) | |
1741 | { | |
1742 | struct page *page = e->pages; | |
1743 | page_chain_for_each(page) { | |
1744 | if (page_count(page) > 1) | |
1745 | return 1; | |
1746 | } | |
1747 | return 0; | |
1748 | } | |
1749 | ||
1750 | ||
b411b363 PR |
1751 | static inline void drbd_state_lock(struct drbd_conf *mdev) |
1752 | { | |
1753 | wait_event(mdev->misc_wait, | |
1754 | !test_and_set_bit(CLUSTER_ST_CHANGE, &mdev->flags)); | |
1755 | } | |
1756 | ||
1757 | static inline void drbd_state_unlock(struct drbd_conf *mdev) | |
1758 | { | |
1759 | clear_bit(CLUSTER_ST_CHANGE, &mdev->flags); | |
1760 | wake_up(&mdev->misc_wait); | |
1761 | } | |
1762 | ||
bf885f8a AG |
1763 | static inline enum drbd_state_rv |
1764 | _drbd_set_state(struct drbd_conf *mdev, union drbd_state ns, | |
1765 | enum chg_state_flags flags, struct completion *done) | |
b411b363 | 1766 | { |
bf885f8a | 1767 | enum drbd_state_rv rv; |
b411b363 PR |
1768 | |
1769 | read_lock(&global_state_lock); | |
1770 | rv = __drbd_set_state(mdev, ns, flags, done); | |
1771 | read_unlock(&global_state_lock); | |
1772 | ||
1773 | return rv; | |
1774 | } | |
1775 | ||
1776 | /** | |
1777 | * drbd_request_state() - Reqest a state change | |
1778 | * @mdev: DRBD device. | |
1779 | * @mask: mask of state bits to change. | |
1780 | * @val: value of new state bits. | |
1781 | * | |
1782 | * This is the most graceful way of requesting a state change. It is verbose | |
1783 | * quite verbose in case the state change is not possible, and all those | |
1784 | * state changes are globally serialized. | |
1785 | */ | |
1786 | static inline int drbd_request_state(struct drbd_conf *mdev, | |
1787 | union drbd_state mask, | |
1788 | union drbd_state val) | |
1789 | { | |
1790 | return _drbd_request_state(mdev, mask, val, CS_VERBOSE + CS_ORDERED); | |
1791 | } | |
1792 | ||
1793 | #define __drbd_chk_io_error(m,f) __drbd_chk_io_error_(m,f, __func__) | |
1794 | static inline void __drbd_chk_io_error_(struct drbd_conf *mdev, int forcedetach, const char *where) | |
1795 | { | |
1796 | switch (mdev->ldev->dc.on_io_error) { | |
1797 | case EP_PASS_ON: | |
1798 | if (!forcedetach) { | |
7383506c | 1799 | if (__ratelimit(&drbd_ratelimit_state)) |
82f59cc6 | 1800 | dev_err(DEV, "Local IO failed in %s.\n", where); |
b411b363 PR |
1801 | break; |
1802 | } | |
1803 | /* NOTE fall through to detach case if forcedetach set */ | |
1804 | case EP_DETACH: | |
1805 | case EP_CALL_HELPER: | |
82f59cc6 | 1806 | set_bit(WAS_IO_ERROR, &mdev->flags); |
b411b363 PR |
1807 | if (mdev->state.disk > D_FAILED) { |
1808 | _drbd_set_state(_NS(mdev, disk, D_FAILED), CS_HARD, NULL); | |
82f59cc6 LE |
1809 | dev_err(DEV, |
1810 | "Local IO failed in %s. Detaching...\n", where); | |
b411b363 PR |
1811 | } |
1812 | break; | |
1813 | } | |
1814 | } | |
1815 | ||
1816 | /** | |
1817 | * drbd_chk_io_error: Handle the on_io_error setting, should be called from all io completion handlers | |
1818 | * @mdev: DRBD device. | |
1819 | * @error: Error code passed to the IO completion callback | |
1820 | * @forcedetach: Force detach. I.e. the error happened while accessing the meta data | |
1821 | * | |
1822 | * See also drbd_main.c:after_state_ch() if (os.disk > D_FAILED && ns.disk == D_FAILED) | |
1823 | */ | |
1824 | #define drbd_chk_io_error(m,e,f) drbd_chk_io_error_(m,e,f, __func__) | |
1825 | static inline void drbd_chk_io_error_(struct drbd_conf *mdev, | |
1826 | int error, int forcedetach, const char *where) | |
1827 | { | |
1828 | if (error) { | |
1829 | unsigned long flags; | |
1830 | spin_lock_irqsave(&mdev->req_lock, flags); | |
1831 | __drbd_chk_io_error_(mdev, forcedetach, where); | |
1832 | spin_unlock_irqrestore(&mdev->req_lock, flags); | |
1833 | } | |
1834 | } | |
1835 | ||
1836 | ||
1837 | /** | |
1838 | * drbd_md_first_sector() - Returns the first sector number of the meta data area | |
1839 | * @bdev: Meta data block device. | |
1840 | * | |
1841 | * BTW, for internal meta data, this happens to be the maximum capacity | |
1842 | * we could agree upon with our peer node. | |
1843 | */ | |
1844 | static inline sector_t drbd_md_first_sector(struct drbd_backing_dev *bdev) | |
1845 | { | |
1846 | switch (bdev->dc.meta_dev_idx) { | |
1847 | case DRBD_MD_INDEX_INTERNAL: | |
1848 | case DRBD_MD_INDEX_FLEX_INT: | |
1849 | return bdev->md.md_offset + bdev->md.bm_offset; | |
1850 | case DRBD_MD_INDEX_FLEX_EXT: | |
1851 | default: | |
1852 | return bdev->md.md_offset; | |
1853 | } | |
1854 | } | |
1855 | ||
1856 | /** | |
1857 | * drbd_md_last_sector() - Return the last sector number of the meta data area | |
1858 | * @bdev: Meta data block device. | |
1859 | */ | |
1860 | static inline sector_t drbd_md_last_sector(struct drbd_backing_dev *bdev) | |
1861 | { | |
1862 | switch (bdev->dc.meta_dev_idx) { | |
1863 | case DRBD_MD_INDEX_INTERNAL: | |
1864 | case DRBD_MD_INDEX_FLEX_INT: | |
1865 | return bdev->md.md_offset + MD_AL_OFFSET - 1; | |
1866 | case DRBD_MD_INDEX_FLEX_EXT: | |
1867 | default: | |
1868 | return bdev->md.md_offset + bdev->md.md_size_sect; | |
1869 | } | |
1870 | } | |
1871 | ||
1872 | /* Returns the number of 512 byte sectors of the device */ | |
1873 | static inline sector_t drbd_get_capacity(struct block_device *bdev) | |
1874 | { | |
1875 | /* return bdev ? get_capacity(bdev->bd_disk) : 0; */ | |
77304d2a | 1876 | return bdev ? i_size_read(bdev->bd_inode) >> 9 : 0; |
b411b363 PR |
1877 | } |
1878 | ||
1879 | /** | |
1880 | * drbd_get_max_capacity() - Returns the capacity we announce to out peer | |
1881 | * @bdev: Meta data block device. | |
1882 | * | |
1883 | * returns the capacity we announce to out peer. we clip ourselves at the | |
1884 | * various MAX_SECTORS, because if we don't, current implementation will | |
1885 | * oops sooner or later | |
1886 | */ | |
1887 | static inline sector_t drbd_get_max_capacity(struct drbd_backing_dev *bdev) | |
1888 | { | |
1889 | sector_t s; | |
1890 | switch (bdev->dc.meta_dev_idx) { | |
1891 | case DRBD_MD_INDEX_INTERNAL: | |
1892 | case DRBD_MD_INDEX_FLEX_INT: | |
1893 | s = drbd_get_capacity(bdev->backing_bdev) | |
1894 | ? min_t(sector_t, DRBD_MAX_SECTORS_FLEX, | |
1895 | drbd_md_first_sector(bdev)) | |
1896 | : 0; | |
1897 | break; | |
1898 | case DRBD_MD_INDEX_FLEX_EXT: | |
1899 | s = min_t(sector_t, DRBD_MAX_SECTORS_FLEX, | |
1900 | drbd_get_capacity(bdev->backing_bdev)); | |
1901 | /* clip at maximum size the meta device can support */ | |
1902 | s = min_t(sector_t, s, | |
1903 | BM_EXT_TO_SECT(bdev->md.md_size_sect | |
1904 | - bdev->md.bm_offset)); | |
1905 | break; | |
1906 | default: | |
1907 | s = min_t(sector_t, DRBD_MAX_SECTORS, | |
1908 | drbd_get_capacity(bdev->backing_bdev)); | |
1909 | } | |
1910 | return s; | |
1911 | } | |
1912 | ||
1913 | /** | |
1914 | * drbd_md_ss__() - Return the sector number of our meta data super block | |
1915 | * @mdev: DRBD device. | |
1916 | * @bdev: Meta data block device. | |
1917 | */ | |
1918 | static inline sector_t drbd_md_ss__(struct drbd_conf *mdev, | |
1919 | struct drbd_backing_dev *bdev) | |
1920 | { | |
1921 | switch (bdev->dc.meta_dev_idx) { | |
1922 | default: /* external, some index */ | |
1923 | return MD_RESERVED_SECT * bdev->dc.meta_dev_idx; | |
1924 | case DRBD_MD_INDEX_INTERNAL: | |
1925 | /* with drbd08, internal meta data is always "flexible" */ | |
1926 | case DRBD_MD_INDEX_FLEX_INT: | |
1927 | /* sizeof(struct md_on_disk_07) == 4k | |
1928 | * position: last 4k aligned block of 4k size */ | |
1929 | if (!bdev->backing_bdev) { | |
1930 | if (__ratelimit(&drbd_ratelimit_state)) { | |
1931 | dev_err(DEV, "bdev->backing_bdev==NULL\n"); | |
1932 | dump_stack(); | |
1933 | } | |
1934 | return 0; | |
1935 | } | |
1936 | return (drbd_get_capacity(bdev->backing_bdev) & ~7ULL) | |
1937 | - MD_AL_OFFSET; | |
1938 | case DRBD_MD_INDEX_FLEX_EXT: | |
1939 | return 0; | |
1940 | } | |
1941 | } | |
1942 | ||
b411b363 PR |
1943 | static inline void |
1944 | drbd_queue_work_front(struct drbd_work_queue *q, struct drbd_work *w) | |
1945 | { | |
1946 | unsigned long flags; | |
1947 | spin_lock_irqsave(&q->q_lock, flags); | |
1948 | list_add(&w->list, &q->q); | |
1949 | up(&q->s); /* within the spinlock, | |
1950 | see comment near end of drbd_worker() */ | |
1951 | spin_unlock_irqrestore(&q->q_lock, flags); | |
1952 | } | |
1953 | ||
1954 | static inline void | |
1955 | drbd_queue_work(struct drbd_work_queue *q, struct drbd_work *w) | |
1956 | { | |
1957 | unsigned long flags; | |
1958 | spin_lock_irqsave(&q->q_lock, flags); | |
1959 | list_add_tail(&w->list, &q->q); | |
1960 | up(&q->s); /* within the spinlock, | |
1961 | see comment near end of drbd_worker() */ | |
1962 | spin_unlock_irqrestore(&q->q_lock, flags); | |
1963 | } | |
1964 | ||
1965 | static inline void wake_asender(struct drbd_conf *mdev) | |
1966 | { | |
1967 | if (test_bit(SIGNAL_ASENDER, &mdev->flags)) | |
1968 | force_sig(DRBD_SIG, mdev->asender.task); | |
1969 | } | |
1970 | ||
1971 | static inline void request_ping(struct drbd_conf *mdev) | |
1972 | { | |
1973 | set_bit(SEND_PING, &mdev->flags); | |
1974 | wake_asender(mdev); | |
1975 | } | |
1976 | ||
1977 | static inline int drbd_send_short_cmd(struct drbd_conf *mdev, | |
1978 | enum drbd_packets cmd) | |
1979 | { | |
0b70a13d | 1980 | struct p_header80 h; |
b411b363 PR |
1981 | return drbd_send_cmd(mdev, USE_DATA_SOCKET, cmd, &h, sizeof(h)); |
1982 | } | |
1983 | ||
1984 | static inline int drbd_send_ping(struct drbd_conf *mdev) | |
1985 | { | |
0b70a13d | 1986 | struct p_header80 h; |
b411b363 PR |
1987 | return drbd_send_cmd(mdev, USE_META_SOCKET, P_PING, &h, sizeof(h)); |
1988 | } | |
1989 | ||
1990 | static inline int drbd_send_ping_ack(struct drbd_conf *mdev) | |
1991 | { | |
0b70a13d | 1992 | struct p_header80 h; |
b411b363 PR |
1993 | return drbd_send_cmd(mdev, USE_META_SOCKET, P_PING_ACK, &h, sizeof(h)); |
1994 | } | |
1995 | ||
1996 | static inline void drbd_thread_stop(struct drbd_thread *thi) | |
1997 | { | |
81e84650 | 1998 | _drbd_thread_stop(thi, false, true); |
b411b363 PR |
1999 | } |
2000 | ||
2001 | static inline void drbd_thread_stop_nowait(struct drbd_thread *thi) | |
2002 | { | |
81e84650 | 2003 | _drbd_thread_stop(thi, false, false); |
b411b363 PR |
2004 | } |
2005 | ||
2006 | static inline void drbd_thread_restart_nowait(struct drbd_thread *thi) | |
2007 | { | |
81e84650 | 2008 | _drbd_thread_stop(thi, true, false); |
b411b363 PR |
2009 | } |
2010 | ||
2011 | /* counts how many answer packets packets we expect from our peer, | |
2012 | * for either explicit application requests, | |
2013 | * or implicit barrier packets as necessary. | |
2014 | * increased: | |
2015 | * w_send_barrier | |
2016 | * _req_mod(req, queue_for_net_write or queue_for_net_read); | |
2017 | * it is much easier and equally valid to count what we queue for the | |
2018 | * worker, even before it actually was queued or send. | |
2019 | * (drbd_make_request_common; recovery path on read io-error) | |
2020 | * decreased: | |
2021 | * got_BarrierAck (respective tl_clear, tl_clear_barrier) | |
2022 | * _req_mod(req, data_received) | |
2023 | * [from receive_DataReply] | |
2024 | * _req_mod(req, write_acked_by_peer or recv_acked_by_peer or neg_acked) | |
2025 | * [from got_BlockAck (P_WRITE_ACK, P_RECV_ACK)] | |
2026 | * for some reason it is NOT decreased in got_NegAck, | |
2027 | * but in the resulting cleanup code from report_params. | |
2028 | * we should try to remember the reason for that... | |
2029 | * _req_mod(req, send_failed or send_canceled) | |
2030 | * _req_mod(req, connection_lost_while_pending) | |
2031 | * [from tl_clear_barrier] | |
2032 | */ | |
2033 | static inline void inc_ap_pending(struct drbd_conf *mdev) | |
2034 | { | |
2035 | atomic_inc(&mdev->ap_pending_cnt); | |
2036 | } | |
2037 | ||
2038 | #define ERR_IF_CNT_IS_NEGATIVE(which) \ | |
2039 | if (atomic_read(&mdev->which) < 0) \ | |
2040 | dev_err(DEV, "in %s:%d: " #which " = %d < 0 !\n", \ | |
2041 | __func__ , __LINE__ , \ | |
2042 | atomic_read(&mdev->which)) | |
2043 | ||
2044 | #define dec_ap_pending(mdev) do { \ | |
2045 | typecheck(struct drbd_conf *, mdev); \ | |
2046 | if (atomic_dec_and_test(&mdev->ap_pending_cnt)) \ | |
2047 | wake_up(&mdev->misc_wait); \ | |
2048 | ERR_IF_CNT_IS_NEGATIVE(ap_pending_cnt); } while (0) | |
2049 | ||
2050 | /* counts how many resync-related answers we still expect from the peer | |
2051 | * increase decrease | |
2052 | * C_SYNC_TARGET sends P_RS_DATA_REQUEST (and expects P_RS_DATA_REPLY) | |
2053 | * C_SYNC_SOURCE sends P_RS_DATA_REPLY (and expects P_WRITE_ACK whith ID_SYNCER) | |
2054 | * (or P_NEG_ACK with ID_SYNCER) | |
2055 | */ | |
2056 | static inline void inc_rs_pending(struct drbd_conf *mdev) | |
2057 | { | |
2058 | atomic_inc(&mdev->rs_pending_cnt); | |
2059 | } | |
2060 | ||
2061 | #define dec_rs_pending(mdev) do { \ | |
2062 | typecheck(struct drbd_conf *, mdev); \ | |
2063 | atomic_dec(&mdev->rs_pending_cnt); \ | |
2064 | ERR_IF_CNT_IS_NEGATIVE(rs_pending_cnt); } while (0) | |
2065 | ||
2066 | /* counts how many answers we still need to send to the peer. | |
2067 | * increased on | |
2068 | * receive_Data unless protocol A; | |
2069 | * we need to send a P_RECV_ACK (proto B) | |
2070 | * or P_WRITE_ACK (proto C) | |
2071 | * receive_RSDataReply (recv_resync_read) we need to send a P_WRITE_ACK | |
2072 | * receive_DataRequest (receive_RSDataRequest) we need to send back P_DATA | |
2073 | * receive_Barrier_* we need to send a P_BARRIER_ACK | |
2074 | */ | |
2075 | static inline void inc_unacked(struct drbd_conf *mdev) | |
2076 | { | |
2077 | atomic_inc(&mdev->unacked_cnt); | |
2078 | } | |
2079 | ||
2080 | #define dec_unacked(mdev) do { \ | |
2081 | typecheck(struct drbd_conf *, mdev); \ | |
2082 | atomic_dec(&mdev->unacked_cnt); \ | |
2083 | ERR_IF_CNT_IS_NEGATIVE(unacked_cnt); } while (0) | |
2084 | ||
2085 | #define sub_unacked(mdev, n) do { \ | |
2086 | typecheck(struct drbd_conf *, mdev); \ | |
2087 | atomic_sub(n, &mdev->unacked_cnt); \ | |
2088 | ERR_IF_CNT_IS_NEGATIVE(unacked_cnt); } while (0) | |
2089 | ||
2090 | ||
2091 | static inline void put_net_conf(struct drbd_conf *mdev) | |
2092 | { | |
2093 | if (atomic_dec_and_test(&mdev->net_cnt)) | |
84dfb9f5 | 2094 | wake_up(&mdev->net_cnt_wait); |
b411b363 PR |
2095 | } |
2096 | ||
2097 | /** | |
2098 | * get_net_conf() - Increase ref count on mdev->net_conf; Returns 0 if nothing there | |
2099 | * @mdev: DRBD device. | |
2100 | * | |
2101 | * You have to call put_net_conf() when finished working with mdev->net_conf. | |
2102 | */ | |
2103 | static inline int get_net_conf(struct drbd_conf *mdev) | |
2104 | { | |
2105 | int have_net_conf; | |
2106 | ||
2107 | atomic_inc(&mdev->net_cnt); | |
2108 | have_net_conf = mdev->state.conn >= C_UNCONNECTED; | |
2109 | if (!have_net_conf) | |
2110 | put_net_conf(mdev); | |
2111 | return have_net_conf; | |
2112 | } | |
2113 | ||
2114 | /** | |
2115 | * get_ldev() - Increase the ref count on mdev->ldev. Returns 0 if there is no ldev | |
2116 | * @M: DRBD device. | |
2117 | * | |
2118 | * You have to call put_ldev() when finished working with mdev->ldev. | |
2119 | */ | |
2120 | #define get_ldev(M) __cond_lock(local, _get_ldev_if_state(M,D_INCONSISTENT)) | |
2121 | #define get_ldev_if_state(M,MINS) __cond_lock(local, _get_ldev_if_state(M,MINS)) | |
2122 | ||
2123 | static inline void put_ldev(struct drbd_conf *mdev) | |
2124 | { | |
1d7734a0 | 2125 | int i = atomic_dec_return(&mdev->local_cnt); |
b411b363 | 2126 | __release(local); |
1d7734a0 | 2127 | D_ASSERT(i >= 0); |
e9e6f3ec | 2128 | if (i == 0) { |
82f59cc6 LE |
2129 | if (mdev->state.disk == D_DISKLESS) |
2130 | /* even internal references gone, safe to destroy */ | |
2131 | drbd_ldev_destroy(mdev); | |
e9e6f3ec | 2132 | if (mdev->state.disk == D_FAILED) |
82f59cc6 | 2133 | /* all application IO references gone. */ |
e9e6f3ec | 2134 | drbd_go_diskless(mdev); |
b411b363 | 2135 | wake_up(&mdev->misc_wait); |
e9e6f3ec | 2136 | } |
b411b363 PR |
2137 | } |
2138 | ||
2139 | #ifndef __CHECKER__ | |
2140 | static inline int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins) | |
2141 | { | |
2142 | int io_allowed; | |
2143 | ||
82f59cc6 LE |
2144 | /* never get a reference while D_DISKLESS */ |
2145 | if (mdev->state.disk == D_DISKLESS) | |
2146 | return 0; | |
2147 | ||
b411b363 PR |
2148 | atomic_inc(&mdev->local_cnt); |
2149 | io_allowed = (mdev->state.disk >= mins); | |
2150 | if (!io_allowed) | |
2151 | put_ldev(mdev); | |
2152 | return io_allowed; | |
2153 | } | |
2154 | #else | |
2155 | extern int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins); | |
2156 | #endif | |
2157 | ||
2158 | /* you must have an "get_ldev" reference */ | |
2159 | static inline void drbd_get_syncer_progress(struct drbd_conf *mdev, | |
2160 | unsigned long *bits_left, unsigned int *per_mil_done) | |
2161 | { | |
4b0715f0 LE |
2162 | /* this is to break it at compile time when we change that, in case we |
2163 | * want to support more than (1<<32) bits on a 32bit arch. */ | |
b411b363 PR |
2164 | typecheck(unsigned long, mdev->rs_total); |
2165 | ||
2166 | /* note: both rs_total and rs_left are in bits, i.e. in | |
2167 | * units of BM_BLOCK_SIZE. | |
2168 | * for the percentage, we don't care. */ | |
2169 | ||
439d5953 LE |
2170 | if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T) |
2171 | *bits_left = mdev->ov_left; | |
2172 | else | |
2173 | *bits_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; | |
b411b363 PR |
2174 | /* >> 10 to prevent overflow, |
2175 | * +1 to prevent division by zero */ | |
2176 | if (*bits_left > mdev->rs_total) { | |
2177 | /* doh. maybe a logic bug somewhere. | |
2178 | * may also be just a race condition | |
2179 | * between this and a disconnect during sync. | |
2180 | * for now, just prevent in-kernel buffer overflow. | |
2181 | */ | |
2182 | smp_rmb(); | |
2183 | dev_warn(DEV, "cs:%s rs_left=%lu > rs_total=%lu (rs_failed %lu)\n", | |
2184 | drbd_conn_str(mdev->state.conn), | |
2185 | *bits_left, mdev->rs_total, mdev->rs_failed); | |
2186 | *per_mil_done = 0; | |
2187 | } else { | |
4b0715f0 LE |
2188 | /* Make sure the division happens in long context. |
2189 | * We allow up to one petabyte storage right now, | |
2190 | * at a granularity of 4k per bit that is 2**38 bits. | |
2191 | * After shift right and multiplication by 1000, | |
2192 | * this should still fit easily into a 32bit long, | |
2193 | * so we don't need a 64bit division on 32bit arch. | |
2194 | * Note: currently we don't support such large bitmaps on 32bit | |
2195 | * arch anyways, but no harm done to be prepared for it here. | |
2196 | */ | |
2197 | unsigned int shift = mdev->rs_total >= (1ULL << 32) ? 16 : 10; | |
2198 | unsigned long left = *bits_left >> shift; | |
2199 | unsigned long total = 1UL + (mdev->rs_total >> shift); | |
2200 | unsigned long tmp = 1000UL - left * 1000UL/total; | |
b411b363 PR |
2201 | *per_mil_done = tmp; |
2202 | } | |
2203 | } | |
2204 | ||
2205 | ||
2206 | /* this throttles on-the-fly application requests | |
2207 | * according to max_buffers settings; | |
2208 | * maybe re-implement using semaphores? */ | |
2209 | static inline int drbd_get_max_buffers(struct drbd_conf *mdev) | |
2210 | { | |
2211 | int mxb = 1000000; /* arbitrary limit on open requests */ | |
2212 | if (get_net_conf(mdev)) { | |
2213 | mxb = mdev->net_conf->max_buffers; | |
2214 | put_net_conf(mdev); | |
2215 | } | |
2216 | return mxb; | |
2217 | } | |
2218 | ||
3719094e | 2219 | static inline int drbd_state_is_stable(struct drbd_conf *mdev) |
b411b363 | 2220 | { |
3719094e | 2221 | union drbd_state s = mdev->state; |
b411b363 PR |
2222 | |
2223 | /* DO NOT add a default clause, we want the compiler to warn us | |
2224 | * for any newly introduced state we may have forgotten to add here */ | |
2225 | ||
2226 | switch ((enum drbd_conns)s.conn) { | |
2227 | /* new io only accepted when there is no connection, ... */ | |
2228 | case C_STANDALONE: | |
2229 | case C_WF_CONNECTION: | |
2230 | /* ... or there is a well established connection. */ | |
2231 | case C_CONNECTED: | |
2232 | case C_SYNC_SOURCE: | |
2233 | case C_SYNC_TARGET: | |
2234 | case C_VERIFY_S: | |
2235 | case C_VERIFY_T: | |
2236 | case C_PAUSED_SYNC_S: | |
2237 | case C_PAUSED_SYNC_T: | |
67531718 PR |
2238 | case C_AHEAD: |
2239 | case C_BEHIND: | |
3719094e | 2240 | /* transitional states, IO allowed */ |
b411b363 PR |
2241 | case C_DISCONNECTING: |
2242 | case C_UNCONNECTED: | |
2243 | case C_TIMEOUT: | |
2244 | case C_BROKEN_PIPE: | |
2245 | case C_NETWORK_FAILURE: | |
2246 | case C_PROTOCOL_ERROR: | |
2247 | case C_TEAR_DOWN: | |
2248 | case C_WF_REPORT_PARAMS: | |
2249 | case C_STARTING_SYNC_S: | |
2250 | case C_STARTING_SYNC_T: | |
3719094e PR |
2251 | break; |
2252 | ||
2253 | /* Allow IO in BM exchange states with new protocols */ | |
b411b363 | 2254 | case C_WF_BITMAP_S: |
3719094e PR |
2255 | if (mdev->agreed_pro_version < 96) |
2256 | return 0; | |
2257 | break; | |
2258 | ||
2259 | /* no new io accepted in these states */ | |
b411b363 PR |
2260 | case C_WF_BITMAP_T: |
2261 | case C_WF_SYNC_UUID: | |
2262 | case C_MASK: | |
2263 | /* not "stable" */ | |
2264 | return 0; | |
2265 | } | |
2266 | ||
2267 | switch ((enum drbd_disk_state)s.disk) { | |
2268 | case D_DISKLESS: | |
2269 | case D_INCONSISTENT: | |
2270 | case D_OUTDATED: | |
2271 | case D_CONSISTENT: | |
2272 | case D_UP_TO_DATE: | |
2273 | /* disk state is stable as well. */ | |
2274 | break; | |
2275 | ||
2276 | /* no new io accepted during tansitional states */ | |
2277 | case D_ATTACHING: | |
2278 | case D_FAILED: | |
2279 | case D_NEGOTIATING: | |
2280 | case D_UNKNOWN: | |
2281 | case D_MASK: | |
2282 | /* not "stable" */ | |
2283 | return 0; | |
2284 | } | |
2285 | ||
2286 | return 1; | |
2287 | } | |
2288 | ||
fb22c402 PR |
2289 | static inline int is_susp(union drbd_state s) |
2290 | { | |
2291 | return s.susp || s.susp_nod || s.susp_fen; | |
2292 | } | |
2293 | ||
1b881ef7 | 2294 | static inline bool may_inc_ap_bio(struct drbd_conf *mdev) |
b411b363 PR |
2295 | { |
2296 | int mxb = drbd_get_max_buffers(mdev); | |
2297 | ||
fb22c402 | 2298 | if (is_susp(mdev->state)) |
1b881ef7 | 2299 | return false; |
b411b363 | 2300 | if (test_bit(SUSPEND_IO, &mdev->flags)) |
1b881ef7 | 2301 | return false; |
b411b363 PR |
2302 | |
2303 | /* to avoid potential deadlock or bitmap corruption, | |
2304 | * in various places, we only allow new application io | |
2305 | * to start during "stable" states. */ | |
2306 | ||
2307 | /* no new io accepted when attaching or detaching the disk */ | |
3719094e | 2308 | if (!drbd_state_is_stable(mdev)) |
1b881ef7 | 2309 | return false; |
b411b363 PR |
2310 | |
2311 | /* since some older kernels don't have atomic_add_unless, | |
2312 | * and we are within the spinlock anyways, we have this workaround. */ | |
2313 | if (atomic_read(&mdev->ap_bio_cnt) > mxb) | |
1b881ef7 | 2314 | return false; |
b411b363 | 2315 | if (test_bit(BITMAP_IO, &mdev->flags)) |
1b881ef7 AG |
2316 | return false; |
2317 | return true; | |
b411b363 PR |
2318 | } |
2319 | ||
1b881ef7 | 2320 | static inline bool inc_ap_bio_cond(struct drbd_conf *mdev, int count) |
b411b363 | 2321 | { |
1b881ef7 | 2322 | bool rv = false; |
8869d683 PR |
2323 | |
2324 | spin_lock_irq(&mdev->req_lock); | |
1b881ef7 | 2325 | rv = may_inc_ap_bio(mdev); |
8869d683 PR |
2326 | if (rv) |
2327 | atomic_add(count, &mdev->ap_bio_cnt); | |
2328 | spin_unlock_irq(&mdev->req_lock); | |
2329 | ||
2330 | return rv; | |
2331 | } | |
b411b363 | 2332 | |
8869d683 PR |
2333 | static inline void inc_ap_bio(struct drbd_conf *mdev, int count) |
2334 | { | |
b411b363 PR |
2335 | /* we wait here |
2336 | * as long as the device is suspended | |
2337 | * until the bitmap is no longer on the fly during connection | |
2338 | * handshake as long as we would exeed the max_buffer limit. | |
2339 | * | |
2340 | * to avoid races with the reconnect code, | |
2341 | * we need to atomic_inc within the spinlock. */ | |
2342 | ||
1b881ef7 | 2343 | wait_event(mdev->misc_wait, inc_ap_bio_cond(mdev, count)); |
b411b363 PR |
2344 | } |
2345 | ||
2346 | static inline void dec_ap_bio(struct drbd_conf *mdev) | |
2347 | { | |
2348 | int mxb = drbd_get_max_buffers(mdev); | |
2349 | int ap_bio = atomic_dec_return(&mdev->ap_bio_cnt); | |
2350 | ||
2351 | D_ASSERT(ap_bio >= 0); | |
2352 | /* this currently does wake_up for every dec_ap_bio! | |
2353 | * maybe rather introduce some type of hysteresis? | |
2354 | * e.g. (ap_bio == mxb/2 || ap_bio == 0) ? */ | |
2355 | if (ap_bio < mxb) | |
2356 | wake_up(&mdev->misc_wait); | |
2357 | if (ap_bio == 0 && test_bit(BITMAP_IO, &mdev->flags)) { | |
2358 | if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags)) | |
2359 | drbd_queue_work(&mdev->data.work, &mdev->bm_io_work.w); | |
2360 | } | |
2361 | } | |
2362 | ||
62b0da3a | 2363 | static inline int drbd_set_ed_uuid(struct drbd_conf *mdev, u64 val) |
b411b363 | 2364 | { |
62b0da3a | 2365 | int changed = mdev->ed_uuid != val; |
b411b363 | 2366 | mdev->ed_uuid = val; |
62b0da3a | 2367 | return changed; |
b411b363 PR |
2368 | } |
2369 | ||
2370 | static inline int seq_cmp(u32 a, u32 b) | |
2371 | { | |
2372 | /* we assume wrap around at 32bit. | |
2373 | * for wrap around at 24bit (old atomic_t), | |
2374 | * we'd have to | |
2375 | * a <<= 8; b <<= 8; | |
2376 | */ | |
2377 | return (s32)(a) - (s32)(b); | |
2378 | } | |
2379 | #define seq_lt(a, b) (seq_cmp((a), (b)) < 0) | |
2380 | #define seq_gt(a, b) (seq_cmp((a), (b)) > 0) | |
2381 | #define seq_ge(a, b) (seq_cmp((a), (b)) >= 0) | |
2382 | #define seq_le(a, b) (seq_cmp((a), (b)) <= 0) | |
2383 | /* CAUTION: please no side effects in arguments! */ | |
2384 | #define seq_max(a, b) ((u32)(seq_gt((a), (b)) ? (a) : (b))) | |
2385 | ||
2386 | static inline void update_peer_seq(struct drbd_conf *mdev, unsigned int new_seq) | |
2387 | { | |
2388 | unsigned int m; | |
2389 | spin_lock(&mdev->peer_seq_lock); | |
2390 | m = seq_max(mdev->peer_seq, new_seq); | |
2391 | mdev->peer_seq = m; | |
2392 | spin_unlock(&mdev->peer_seq_lock); | |
2393 | if (m == new_seq) | |
2394 | wake_up(&mdev->seq_wait); | |
2395 | } | |
2396 | ||
2397 | static inline void drbd_update_congested(struct drbd_conf *mdev) | |
2398 | { | |
2399 | struct sock *sk = mdev->data.socket->sk; | |
2400 | if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5) | |
2401 | set_bit(NET_CONGESTED, &mdev->flags); | |
2402 | } | |
2403 | ||
2404 | static inline int drbd_queue_order_type(struct drbd_conf *mdev) | |
2405 | { | |
2406 | /* sorry, we currently have no working implementation | |
2407 | * of distributed TCQ stuff */ | |
2408 | #ifndef QUEUE_ORDERED_NONE | |
2409 | #define QUEUE_ORDERED_NONE 0 | |
2410 | #endif | |
2411 | return QUEUE_ORDERED_NONE; | |
2412 | } | |
2413 | ||
b411b363 PR |
2414 | static inline void drbd_md_flush(struct drbd_conf *mdev) |
2415 | { | |
2416 | int r; | |
2417 | ||
a8a4e51e | 2418 | if (test_bit(MD_NO_FUA, &mdev->flags)) |
b411b363 PR |
2419 | return; |
2420 | ||
dd3932ed | 2421 | r = blkdev_issue_flush(mdev->ldev->md_bdev, GFP_KERNEL, NULL); |
b411b363 | 2422 | if (r) { |
a8a4e51e | 2423 | set_bit(MD_NO_FUA, &mdev->flags); |
b411b363 PR |
2424 | dev_err(DEV, "meta data flush failed with status %d, disabling md-flushes\n", r); |
2425 | } | |
2426 | } | |
2427 | ||
2428 | #endif |