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> | |
062e879c | 43 | #include <linux/idr.h> |
b411b363 PR |
44 | #include <net/tcp.h> |
45 | #include <linux/lru_cache.h> | |
70c71606 | 46 | #include <linux/prefetch.h> |
b8907339 PR |
47 | #include <linux/drbd.h> |
48 | #include "drbd_state.h" | |
b411b363 PR |
49 | |
50 | #ifdef __CHECKER__ | |
51 | # define __protected_by(x) __attribute__((require_context(x,1,999,"rdwr"))) | |
52 | # define __protected_read_by(x) __attribute__((require_context(x,1,999,"read"))) | |
53 | # define __protected_write_by(x) __attribute__((require_context(x,1,999,"write"))) | |
54 | # define __must_hold(x) __attribute__((context(x,1,1), require_context(x,1,999,"call"))) | |
55 | #else | |
56 | # define __protected_by(x) | |
57 | # define __protected_read_by(x) | |
58 | # define __protected_write_by(x) | |
59 | # define __must_hold(x) | |
60 | #endif | |
61 | ||
62 | #define __no_warn(lock, stmt) do { __acquire(lock); stmt; __release(lock); } while (0) | |
63 | ||
64 | /* module parameter, defined in drbd_main.c */ | |
65 | extern unsigned int minor_count; | |
66 | extern int disable_sendpage; | |
67 | extern int allow_oos; | |
68 | extern unsigned int cn_idx; | |
69 | ||
70 | #ifdef CONFIG_DRBD_FAULT_INJECTION | |
71 | extern int enable_faults; | |
72 | extern int fault_rate; | |
73 | extern int fault_devs; | |
74 | #endif | |
75 | ||
76 | extern char usermode_helper[]; | |
77 | ||
78 | ||
b411b363 PR |
79 | /* I don't remember why XCPU ... |
80 | * This is used to wake the asender, | |
81 | * and to interrupt sending the sending task | |
82 | * on disconnect. | |
83 | */ | |
84 | #define DRBD_SIG SIGXCPU | |
85 | ||
86 | /* This is used to stop/restart our threads. | |
87 | * Cannot use SIGTERM nor SIGKILL, since these | |
88 | * are sent out by init on runlevel changes | |
89 | * I choose SIGHUP for now. | |
90 | */ | |
91 | #define DRBD_SIGKILL SIGHUP | |
92 | ||
b411b363 PR |
93 | #define ID_IN_SYNC (4711ULL) |
94 | #define ID_OUT_OF_SYNC (4712ULL) | |
b411b363 | 95 | #define ID_SYNCER (-1ULL) |
579b57ed | 96 | |
4a23f264 | 97 | #define UUID_NEW_BM_OFFSET ((u64)0x0001000000000000ULL) |
b411b363 PR |
98 | |
99 | struct drbd_conf; | |
2111438b | 100 | struct drbd_tconn; |
b411b363 PR |
101 | |
102 | ||
103 | /* to shorten dev_warn(DEV, "msg"); and relatives statements */ | |
104 | #define DEV (disk_to_dev(mdev->vdisk)) | |
105 | ||
60ae4966 PR |
106 | #define conn_printk(LEVEL, TCONN, FMT, ARGS...) \ |
107 | printk(LEVEL "d-con %s: " FMT, TCONN->name , ## ARGS) | |
108 | #define conn_alert(TCONN, FMT, ARGS...) conn_printk(KERN_ALERT, TCONN, FMT, ## ARGS) | |
109 | #define conn_crit(TCONN, FMT, ARGS...) conn_printk(KERN_CRIT, TCONN, FMT, ## ARGS) | |
110 | #define conn_err(TCONN, FMT, ARGS...) conn_printk(KERN_ERR, TCONN, FMT, ## ARGS) | |
111 | #define conn_warn(TCONN, FMT, ARGS...) conn_printk(KERN_WARNING, TCONN, FMT, ## ARGS) | |
112 | #define conn_notice(TCONN, FMT, ARGS...) conn_printk(KERN_NOTICE, TCONN, FMT, ## ARGS) | |
113 | #define conn_info(TCONN, FMT, ARGS...) conn_printk(KERN_INFO, TCONN, FMT, ## ARGS) | |
114 | #define conn_dbg(TCONN, FMT, ARGS...) conn_printk(KERN_DEBUG, TCONN, FMT, ## ARGS) | |
115 | ||
b411b363 PR |
116 | #define D_ASSERT(exp) if (!(exp)) \ |
117 | dev_err(DEV, "ASSERT( " #exp " ) in %s:%d\n", __FILE__, __LINE__) | |
118 | ||
841ce241 AG |
119 | /** |
120 | * expect - Make an assertion | |
121 | * | |
122 | * Unlike the assert macro, this macro returns a boolean result. | |
123 | */ | |
124 | #define expect(exp) ({ \ | |
125 | bool _bool = (exp); \ | |
126 | if (!_bool) \ | |
127 | dev_err(DEV, "ASSERTION %s FAILED in %s\n", \ | |
128 | #exp, __func__); \ | |
129 | _bool; \ | |
130 | }) | |
b411b363 PR |
131 | |
132 | /* Defines to control fault insertion */ | |
133 | enum { | |
134 | DRBD_FAULT_MD_WR = 0, /* meta data write */ | |
135 | DRBD_FAULT_MD_RD = 1, /* read */ | |
136 | DRBD_FAULT_RS_WR = 2, /* resync */ | |
137 | DRBD_FAULT_RS_RD = 3, | |
138 | DRBD_FAULT_DT_WR = 4, /* data */ | |
139 | DRBD_FAULT_DT_RD = 5, | |
140 | DRBD_FAULT_DT_RA = 6, /* data read ahead */ | |
141 | DRBD_FAULT_BM_ALLOC = 7, /* bitmap allocation */ | |
142 | DRBD_FAULT_AL_EE = 8, /* alloc ee */ | |
6b4388ac | 143 | DRBD_FAULT_RECEIVE = 9, /* Changes some bytes upon receiving a [rs]data block */ |
b411b363 PR |
144 | |
145 | DRBD_FAULT_MAX, | |
146 | }; | |
147 | ||
b411b363 PR |
148 | extern unsigned int |
149 | _drbd_insert_fault(struct drbd_conf *mdev, unsigned int type); | |
0cf9d27e | 150 | |
b411b363 PR |
151 | static inline int |
152 | drbd_insert_fault(struct drbd_conf *mdev, unsigned int type) { | |
0cf9d27e | 153 | #ifdef CONFIG_DRBD_FAULT_INJECTION |
b411b363 PR |
154 | return fault_rate && |
155 | (enable_faults & (1<<type)) && | |
156 | _drbd_insert_fault(mdev, type); | |
b411b363 | 157 | #else |
0cf9d27e | 158 | return 0; |
b411b363 | 159 | #endif |
0cf9d27e | 160 | } |
b411b363 PR |
161 | |
162 | /* integer division, round _UP_ to the next integer */ | |
163 | #define div_ceil(A, B) ((A)/(B) + ((A)%(B) ? 1 : 0)) | |
164 | /* usual integer division */ | |
165 | #define div_floor(A, B) ((A)/(B)) | |
166 | ||
167 | /* drbd_meta-data.c (still in drbd_main.c) */ | |
168 | /* 4th incarnation of the disk layout. */ | |
169 | #define DRBD_MD_MAGIC (DRBD_MAGIC+4) | |
170 | ||
171 | extern struct drbd_conf **minor_table; | |
172 | extern struct ratelimit_state drbd_ratelimit_state; | |
173 | ||
174 | /* on the wire */ | |
d8763023 | 175 | enum drbd_packet { |
b411b363 PR |
176 | /* receiver (data socket) */ |
177 | P_DATA = 0x00, | |
178 | P_DATA_REPLY = 0x01, /* Response to P_DATA_REQUEST */ | |
179 | P_RS_DATA_REPLY = 0x02, /* Response to P_RS_DATA_REQUEST */ | |
180 | P_BARRIER = 0x03, | |
181 | P_BITMAP = 0x04, | |
182 | P_BECOME_SYNC_TARGET = 0x05, | |
183 | P_BECOME_SYNC_SOURCE = 0x06, | |
184 | P_UNPLUG_REMOTE = 0x07, /* Used at various times to hint the peer */ | |
185 | P_DATA_REQUEST = 0x08, /* Used to ask for a data block */ | |
186 | P_RS_DATA_REQUEST = 0x09, /* Used to ask for a data block for resync */ | |
187 | P_SYNC_PARAM = 0x0a, | |
188 | P_PROTOCOL = 0x0b, | |
189 | P_UUIDS = 0x0c, | |
190 | P_SIZES = 0x0d, | |
191 | P_STATE = 0x0e, | |
192 | P_SYNC_UUID = 0x0f, | |
193 | P_AUTH_CHALLENGE = 0x10, | |
194 | P_AUTH_RESPONSE = 0x11, | |
195 | P_STATE_CHG_REQ = 0x12, | |
196 | ||
197 | /* asender (meta socket */ | |
198 | P_PING = 0x13, | |
199 | P_PING_ACK = 0x14, | |
200 | P_RECV_ACK = 0x15, /* Used in protocol B */ | |
201 | P_WRITE_ACK = 0x16, /* Used in protocol C */ | |
202 | P_RS_WRITE_ACK = 0x17, /* Is a P_WRITE_ACK, additionally call set_in_sync(). */ | |
203 | P_DISCARD_ACK = 0x18, /* Used in proto C, two-primaries conflict detection */ | |
204 | P_NEG_ACK = 0x19, /* Sent if local disk is unusable */ | |
205 | P_NEG_DREPLY = 0x1a, /* Local disk is broken... */ | |
206 | P_NEG_RS_DREPLY = 0x1b, /* Local disk is broken... */ | |
207 | P_BARRIER_ACK = 0x1c, | |
208 | P_STATE_CHG_REPLY = 0x1d, | |
209 | ||
210 | /* "new" commands, no longer fitting into the ordering scheme above */ | |
211 | ||
212 | P_OV_REQUEST = 0x1e, /* data socket */ | |
213 | P_OV_REPLY = 0x1f, | |
214 | P_OV_RESULT = 0x20, /* meta socket */ | |
215 | P_CSUM_RS_REQUEST = 0x21, /* data socket */ | |
216 | P_RS_IS_IN_SYNC = 0x22, /* meta socket */ | |
217 | P_SYNC_PARAM89 = 0x23, /* data socket, protocol version 89 replacement for P_SYNC_PARAM */ | |
218 | P_COMPRESSED_BITMAP = 0x24, /* compressed or otherwise encoded bitmap transfer */ | |
0ced55a3 PR |
219 | /* P_CKPT_FENCE_REQ = 0x25, * currently reserved for protocol D */ |
220 | /* P_CKPT_DISABLE_REQ = 0x26, * currently reserved for protocol D */ | |
221 | P_DELAY_PROBE = 0x27, /* is used on BOTH sockets */ | |
73a01a18 | 222 | P_OUT_OF_SYNC = 0x28, /* Mark as out of sync (Outrunning), data socket */ |
d612d309 | 223 | P_RS_CANCEL = 0x29, /* meta: Used to cancel RS_DATA_REQUEST packet by SyncSource */ |
cf29c9d8 PR |
224 | P_CONN_ST_CHG_REQ = 0x2a, /* data sock: Connection wide state request */ |
225 | P_CONN_ST_CHG_REPLY = 0x2b, /* meta sock: Connection side state req reply */ | |
b411b363 | 226 | |
cf29c9d8 | 227 | P_MAX_CMD = 0x2c, |
b411b363 PR |
228 | P_MAY_IGNORE = 0x100, /* Flag to test if (cmd > P_MAY_IGNORE) ... */ |
229 | P_MAX_OPT_CMD = 0x101, | |
230 | ||
231 | /* special command ids for handshake */ | |
232 | ||
233 | P_HAND_SHAKE_M = 0xfff1, /* First Packet on the MetaSock */ | |
234 | P_HAND_SHAKE_S = 0xfff2, /* First Packet on the Socket */ | |
235 | ||
236 | P_HAND_SHAKE = 0xfffe /* FIXED for the next century! */ | |
237 | }; | |
238 | ||
d8763023 | 239 | extern const char *cmdname(enum drbd_packet cmd); |
b411b363 PR |
240 | |
241 | /* for sending/receiving the bitmap, | |
242 | * possibly in some encoding scheme */ | |
243 | struct bm_xfer_ctx { | |
244 | /* "const" | |
245 | * stores total bits and long words | |
246 | * of the bitmap, so we don't need to | |
247 | * call the accessor functions over and again. */ | |
248 | unsigned long bm_bits; | |
249 | unsigned long bm_words; | |
250 | /* during xfer, current position within the bitmap */ | |
251 | unsigned long bit_offset; | |
252 | unsigned long word_offset; | |
253 | ||
254 | /* statistics; index: (h->command == P_BITMAP) */ | |
255 | unsigned packets[2]; | |
256 | unsigned bytes[2]; | |
257 | }; | |
258 | ||
259 | extern void INFO_bm_xfer_stats(struct drbd_conf *mdev, | |
260 | const char *direction, struct bm_xfer_ctx *c); | |
261 | ||
262 | static inline void bm_xfer_ctx_bit_to_word_offset(struct bm_xfer_ctx *c) | |
263 | { | |
264 | /* word_offset counts "native long words" (32 or 64 bit), | |
265 | * aligned at 64 bit. | |
266 | * Encoded packet may end at an unaligned bit offset. | |
267 | * In case a fallback clear text packet is transmitted in | |
268 | * between, we adjust this offset back to the last 64bit | |
269 | * aligned "native long word", which makes coding and decoding | |
270 | * the plain text bitmap much more convenient. */ | |
271 | #if BITS_PER_LONG == 64 | |
272 | c->word_offset = c->bit_offset >> 6; | |
273 | #elif BITS_PER_LONG == 32 | |
274 | c->word_offset = c->bit_offset >> 5; | |
275 | c->word_offset &= ~(1UL); | |
276 | #else | |
277 | # error "unsupported BITS_PER_LONG" | |
278 | #endif | |
279 | } | |
280 | ||
281 | #ifndef __packed | |
282 | #define __packed __attribute__((packed)) | |
283 | #endif | |
284 | ||
285 | /* This is the layout for a packet on the wire. | |
286 | * The byteorder is the network byte order. | |
287 | * (except block_id and barrier fields. | |
288 | * these are pointers to local structs | |
289 | * and have no relevance for the partner, | |
290 | * which just echoes them as received.) | |
291 | * | |
292 | * NOTE that the payload starts at a long aligned offset, | |
293 | * regardless of 32 or 64 bit arch! | |
294 | */ | |
0b70a13d | 295 | struct p_header80 { |
b411b363 PR |
296 | u32 magic; |
297 | u16 command; | |
298 | u16 length; /* bytes of data after this header */ | |
b411b363 | 299 | } __packed; |
0b70a13d PR |
300 | |
301 | /* Header for big packets, Used for data packets exceeding 64kB */ | |
302 | struct p_header95 { | |
303 | u16 magic; /* use DRBD_MAGIC_BIG here */ | |
304 | u16 command; | |
00b42537 | 305 | u32 length; /* Use only 24 bits of that. Ignore the highest 8 bit. */ |
0b70a13d PR |
306 | } __packed; |
307 | ||
c012949a PR |
308 | struct p_header { |
309 | union { | |
310 | struct p_header80 h80; | |
311 | struct p_header95 h95; | |
312 | }; | |
313 | u8 payload[0]; | |
0b70a13d | 314 | }; |
b411b363 PR |
315 | |
316 | /* | |
317 | * short commands, packets without payload, plain p_header: | |
318 | * P_PING | |
319 | * P_PING_ACK | |
320 | * P_BECOME_SYNC_TARGET | |
321 | * P_BECOME_SYNC_SOURCE | |
322 | * P_UNPLUG_REMOTE | |
323 | */ | |
324 | ||
325 | /* | |
326 | * commands with out-of-struct payload: | |
327 | * P_BITMAP (no additional fields) | |
328 | * P_DATA, P_DATA_REPLY (see p_data) | |
329 | * P_COMPRESSED_BITMAP (see receive_compressed_bitmap) | |
330 | */ | |
331 | ||
332 | /* these defines must not be changed without changing the protocol version */ | |
76d2e7ec PR |
333 | #define DP_HARDBARRIER 1 /* depricated */ |
334 | #define DP_RW_SYNC 2 /* equals REQ_SYNC */ | |
b411b363 | 335 | #define DP_MAY_SET_IN_SYNC 4 |
721a9602 | 336 | #define DP_UNPLUG 8 /* not used anymore */ |
76d2e7ec PR |
337 | #define DP_FUA 16 /* equals REQ_FUA */ |
338 | #define DP_FLUSH 32 /* equals REQ_FLUSH */ | |
339 | #define DP_DISCARD 64 /* equals REQ_DISCARD */ | |
b411b363 PR |
340 | |
341 | struct p_data { | |
c012949a | 342 | struct p_header head; |
b411b363 PR |
343 | u64 sector; /* 64 bits sector number */ |
344 | u64 block_id; /* to identify the request in protocol B&C */ | |
345 | u32 seq_num; | |
346 | u32 dp_flags; | |
347 | } __packed; | |
348 | ||
349 | /* | |
350 | * commands which share a struct: | |
351 | * p_block_ack: | |
352 | * P_RECV_ACK (proto B), P_WRITE_ACK (proto C), | |
353 | * P_DISCARD_ACK (proto C, two-primaries conflict detection) | |
354 | * p_block_req: | |
355 | * P_DATA_REQUEST, P_RS_DATA_REQUEST | |
356 | */ | |
357 | struct p_block_ack { | |
c012949a | 358 | struct p_header head; |
b411b363 PR |
359 | u64 sector; |
360 | u64 block_id; | |
361 | u32 blksize; | |
362 | u32 seq_num; | |
363 | } __packed; | |
364 | ||
365 | ||
366 | struct p_block_req { | |
c012949a | 367 | struct p_header head; |
b411b363 PR |
368 | u64 sector; |
369 | u64 block_id; | |
370 | u32 blksize; | |
371 | u32 pad; /* to multiple of 8 Byte */ | |
372 | } __packed; | |
373 | ||
374 | /* | |
375 | * commands with their own struct for additional fields: | |
376 | * P_HAND_SHAKE | |
377 | * P_BARRIER | |
378 | * P_BARRIER_ACK | |
379 | * P_SYNC_PARAM | |
380 | * ReportParams | |
381 | */ | |
382 | ||
383 | struct p_handshake { | |
8a22cccc | 384 | struct p_header head; /* Note: vnr will be ignored */ |
b411b363 PR |
385 | u32 protocol_min; |
386 | u32 feature_flags; | |
387 | u32 protocol_max; | |
388 | ||
389 | /* should be more than enough for future enhancements | |
390 | * for now, feature_flags and the reserverd array shall be zero. | |
391 | */ | |
392 | ||
393 | u32 _pad; | |
394 | u64 reserverd[7]; | |
395 | } __packed; | |
396 | /* 80 bytes, FIXED for the next century */ | |
397 | ||
398 | struct p_barrier { | |
c012949a | 399 | struct p_header head; |
b411b363 PR |
400 | u32 barrier; /* barrier number _handle_ only */ |
401 | u32 pad; /* to multiple of 8 Byte */ | |
402 | } __packed; | |
403 | ||
404 | struct p_barrier_ack { | |
c012949a | 405 | struct p_header head; |
b411b363 PR |
406 | u32 barrier; |
407 | u32 set_size; | |
408 | } __packed; | |
409 | ||
410 | struct p_rs_param { | |
c012949a | 411 | struct p_header head; |
b411b363 PR |
412 | u32 rate; |
413 | ||
414 | /* Since protocol version 88 and higher. */ | |
415 | char verify_alg[0]; | |
416 | } __packed; | |
417 | ||
418 | struct p_rs_param_89 { | |
c012949a | 419 | struct p_header head; |
b411b363 PR |
420 | u32 rate; |
421 | /* protocol version 89: */ | |
422 | char verify_alg[SHARED_SECRET_MAX]; | |
423 | char csums_alg[SHARED_SECRET_MAX]; | |
424 | } __packed; | |
425 | ||
8e26f9cc | 426 | struct p_rs_param_95 { |
c012949a | 427 | struct p_header head; |
8e26f9cc PR |
428 | u32 rate; |
429 | char verify_alg[SHARED_SECRET_MAX]; | |
430 | char csums_alg[SHARED_SECRET_MAX]; | |
431 | u32 c_plan_ahead; | |
432 | u32 c_delay_target; | |
433 | u32 c_fill_target; | |
434 | u32 c_max_rate; | |
435 | } __packed; | |
436 | ||
cf14c2e9 PR |
437 | enum drbd_conn_flags { |
438 | CF_WANT_LOSE = 1, | |
439 | CF_DRY_RUN = 2, | |
440 | }; | |
441 | ||
b411b363 | 442 | struct p_protocol { |
c012949a | 443 | struct p_header head; |
b411b363 PR |
444 | u32 protocol; |
445 | u32 after_sb_0p; | |
446 | u32 after_sb_1p; | |
447 | u32 after_sb_2p; | |
cf14c2e9 | 448 | u32 conn_flags; |
b411b363 PR |
449 | u32 two_primaries; |
450 | ||
451 | /* Since protocol version 87 and higher. */ | |
452 | char integrity_alg[0]; | |
453 | ||
454 | } __packed; | |
455 | ||
456 | struct p_uuids { | |
c012949a | 457 | struct p_header head; |
b411b363 PR |
458 | u64 uuid[UI_EXTENDED_SIZE]; |
459 | } __packed; | |
460 | ||
461 | struct p_rs_uuid { | |
c012949a | 462 | struct p_header head; |
b411b363 PR |
463 | u64 uuid; |
464 | } __packed; | |
465 | ||
466 | struct p_sizes { | |
c012949a | 467 | struct p_header head; |
b411b363 PR |
468 | u64 d_size; /* size of disk */ |
469 | u64 u_size; /* user requested size */ | |
470 | u64 c_size; /* current exported size */ | |
1816a2b4 | 471 | u32 max_bio_size; /* Maximal size of a BIO */ |
e89b591c PR |
472 | u16 queue_order_type; /* not yet implemented in DRBD*/ |
473 | u16 dds_flags; /* use enum dds_flags here. */ | |
b411b363 PR |
474 | } __packed; |
475 | ||
476 | struct p_state { | |
c012949a | 477 | struct p_header head; |
b411b363 PR |
478 | u32 state; |
479 | } __packed; | |
480 | ||
481 | struct p_req_state { | |
c012949a | 482 | struct p_header head; |
b411b363 PR |
483 | u32 mask; |
484 | u32 val; | |
485 | } __packed; | |
486 | ||
487 | struct p_req_state_reply { | |
c012949a | 488 | struct p_header head; |
b411b363 PR |
489 | u32 retcode; |
490 | } __packed; | |
491 | ||
492 | struct p_drbd06_param { | |
493 | u64 size; | |
494 | u32 state; | |
495 | u32 blksize; | |
496 | u32 protocol; | |
497 | u32 version; | |
498 | u32 gen_cnt[5]; | |
499 | u32 bit_map_gen[5]; | |
500 | } __packed; | |
501 | ||
502 | struct p_discard { | |
c012949a | 503 | struct p_header head; |
b411b363 PR |
504 | u64 block_id; |
505 | u32 seq_num; | |
506 | u32 pad; | |
507 | } __packed; | |
508 | ||
73a01a18 | 509 | struct p_block_desc { |
c012949a | 510 | struct p_header head; |
73a01a18 PR |
511 | u64 sector; |
512 | u32 blksize; | |
513 | u32 pad; /* to multiple of 8 Byte */ | |
514 | } __packed; | |
515 | ||
b411b363 PR |
516 | /* Valid values for the encoding field. |
517 | * Bump proto version when changing this. */ | |
518 | enum drbd_bitmap_code { | |
519 | /* RLE_VLI_Bytes = 0, | |
520 | * and other bit variants had been defined during | |
521 | * algorithm evaluation. */ | |
522 | RLE_VLI_Bits = 2, | |
523 | }; | |
524 | ||
525 | struct p_compressed_bm { | |
c012949a | 526 | struct p_header head; |
b411b363 PR |
527 | /* (encoding & 0x0f): actual encoding, see enum drbd_bitmap_code |
528 | * (encoding & 0x80): polarity (set/unset) of first runlength | |
529 | * ((encoding >> 4) & 0x07): pad_bits, number of trailing zero bits | |
530 | * used to pad up to head.length bytes | |
531 | */ | |
532 | u8 encoding; | |
533 | ||
534 | u8 code[0]; | |
535 | } __packed; | |
536 | ||
0b70a13d | 537 | struct p_delay_probe93 { |
c012949a | 538 | struct p_header head; |
0b70a13d PR |
539 | u32 seq_num; /* sequence number to match the two probe packets */ |
540 | u32 offset; /* usecs the probe got sent after the reference time point */ | |
0ced55a3 PR |
541 | } __packed; |
542 | ||
b411b363 PR |
543 | /* DCBP: Drbd Compressed Bitmap Packet ... */ |
544 | static inline enum drbd_bitmap_code | |
545 | DCBP_get_code(struct p_compressed_bm *p) | |
546 | { | |
547 | return (enum drbd_bitmap_code)(p->encoding & 0x0f); | |
548 | } | |
549 | ||
550 | static inline void | |
551 | DCBP_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code) | |
552 | { | |
553 | BUG_ON(code & ~0xf); | |
554 | p->encoding = (p->encoding & ~0xf) | code; | |
555 | } | |
556 | ||
557 | static inline int | |
558 | DCBP_get_start(struct p_compressed_bm *p) | |
559 | { | |
560 | return (p->encoding & 0x80) != 0; | |
561 | } | |
562 | ||
563 | static inline void | |
564 | DCBP_set_start(struct p_compressed_bm *p, int set) | |
565 | { | |
566 | p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0); | |
567 | } | |
568 | ||
569 | static inline int | |
570 | DCBP_get_pad_bits(struct p_compressed_bm *p) | |
571 | { | |
572 | return (p->encoding >> 4) & 0x7; | |
573 | } | |
574 | ||
575 | static inline void | |
576 | DCBP_set_pad_bits(struct p_compressed_bm *p, int n) | |
577 | { | |
578 | BUG_ON(n & ~0x7); | |
579 | p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4); | |
580 | } | |
581 | ||
582 | /* one bitmap packet, including the p_header, | |
583 | * should fit within one _architecture independend_ page. | |
584 | * so we need to use the fixed size 4KiB page size | |
25985edc | 585 | * most architectures have used for a long time. |
b411b363 | 586 | */ |
c012949a | 587 | #define BM_PACKET_PAYLOAD_BYTES (4096 - sizeof(struct p_header)) |
b411b363 PR |
588 | #define BM_PACKET_WORDS (BM_PACKET_PAYLOAD_BYTES/sizeof(long)) |
589 | #define BM_PACKET_VLI_BYTES_MAX (4096 - sizeof(struct p_compressed_bm)) | |
590 | #if (PAGE_SIZE < 4096) | |
591 | /* drbd_send_bitmap / receive_bitmap would break horribly */ | |
592 | #error "PAGE_SIZE too small" | |
593 | #endif | |
594 | ||
595 | union p_polymorph { | |
c012949a | 596 | struct p_header header; |
b411b363 PR |
597 | struct p_handshake handshake; |
598 | struct p_data data; | |
599 | struct p_block_ack block_ack; | |
600 | struct p_barrier barrier; | |
601 | struct p_barrier_ack barrier_ack; | |
602 | struct p_rs_param_89 rs_param_89; | |
8e26f9cc | 603 | struct p_rs_param_95 rs_param_95; |
b411b363 PR |
604 | struct p_protocol protocol; |
605 | struct p_sizes sizes; | |
606 | struct p_uuids uuids; | |
607 | struct p_state state; | |
608 | struct p_req_state req_state; | |
609 | struct p_req_state_reply req_state_reply; | |
610 | struct p_block_req block_req; | |
02918be2 PR |
611 | struct p_delay_probe93 delay_probe93; |
612 | struct p_rs_uuid rs_uuid; | |
73a01a18 | 613 | struct p_block_desc block_desc; |
b411b363 PR |
614 | } __packed; |
615 | ||
616 | /**********************************************************************/ | |
617 | enum drbd_thread_state { | |
e77a0a5c AG |
618 | NONE, |
619 | RUNNING, | |
620 | EXITING, | |
621 | RESTARTING | |
b411b363 PR |
622 | }; |
623 | ||
624 | struct drbd_thread { | |
625 | spinlock_t t_lock; | |
626 | struct task_struct *task; | |
627 | struct completion stop; | |
628 | enum drbd_thread_state t_state; | |
629 | int (*function) (struct drbd_thread *); | |
392c8801 | 630 | struct drbd_tconn *tconn; |
b411b363 | 631 | int reset_cpu_mask; |
bed879ae | 632 | char name[9]; |
b411b363 PR |
633 | }; |
634 | ||
635 | static inline enum drbd_thread_state get_t_state(struct drbd_thread *thi) | |
636 | { | |
637 | /* THINK testing the t_state seems to be uncritical in all cases | |
638 | * (but thread_{start,stop}), so we can read it *without* the lock. | |
639 | * --lge */ | |
640 | ||
641 | smp_rmb(); | |
642 | return thi->t_state; | |
643 | } | |
644 | ||
b411b363 | 645 | struct drbd_work; |
00d56944 | 646 | typedef int (*drbd_work_cb)(struct drbd_work *, int cancel); |
b411b363 PR |
647 | struct drbd_work { |
648 | struct list_head list; | |
649 | drbd_work_cb cb; | |
00d56944 PR |
650 | union { |
651 | struct drbd_conf *mdev; | |
652 | struct drbd_tconn *tconn; | |
653 | }; | |
b411b363 PR |
654 | }; |
655 | ||
ace652ac AG |
656 | #include "drbd_interval.h" |
657 | ||
b411b363 PR |
658 | struct drbd_request { |
659 | struct drbd_work w; | |
b411b363 PR |
660 | |
661 | /* if local IO is not allowed, will be NULL. | |
662 | * if local IO _is_ allowed, holds the locally submitted bio clone, | |
663 | * or, after local IO completion, the ERR_PTR(error). | |
fcefa62e | 664 | * see drbd_request_endio(). */ |
b411b363 PR |
665 | struct bio *private_bio; |
666 | ||
ace652ac | 667 | struct drbd_interval i; |
b411b363 PR |
668 | unsigned int epoch; /* barrier_nr */ |
669 | ||
670 | /* barrier_nr: used to check on "completion" whether this req was in | |
671 | * the current epoch, and we therefore have to close it, | |
672 | * starting a new epoch... | |
673 | */ | |
674 | ||
b411b363 PR |
675 | struct list_head tl_requests; /* ring list in the transfer log */ |
676 | struct bio *master_bio; /* master bio pointer */ | |
677 | unsigned long rq_state; /* see comments above _req_mod() */ | |
678 | int seq_num; | |
679 | unsigned long start_time; | |
680 | }; | |
681 | ||
682 | struct drbd_tl_epoch { | |
683 | struct drbd_work w; | |
684 | struct list_head requests; /* requests before */ | |
685 | struct drbd_tl_epoch *next; /* pointer to the next barrier */ | |
686 | unsigned int br_number; /* the barriers identifier. */ | |
7e602c0a | 687 | int n_writes; /* number of requests attached before this barrier */ |
b411b363 PR |
688 | }; |
689 | ||
b411b363 PR |
690 | struct drbd_epoch { |
691 | struct list_head list; | |
692 | unsigned int barrier_nr; | |
693 | atomic_t epoch_size; /* increased on every request added. */ | |
694 | atomic_t active; /* increased on every req. added, and dec on every finished. */ | |
695 | unsigned long flags; | |
696 | }; | |
697 | ||
698 | /* drbd_epoch flag bits */ | |
699 | enum { | |
b411b363 | 700 | DE_HAVE_BARRIER_NUMBER, |
b411b363 PR |
701 | }; |
702 | ||
703 | enum epoch_event { | |
704 | EV_PUT, | |
705 | EV_GOT_BARRIER_NR, | |
b411b363 | 706 | EV_BECAME_LAST, |
b411b363 PR |
707 | EV_CLEANUP = 32, /* used as flag */ |
708 | }; | |
709 | ||
b411b363 PR |
710 | struct drbd_wq_barrier { |
711 | struct drbd_work w; | |
712 | struct completion done; | |
713 | }; | |
714 | ||
715 | struct digest_info { | |
716 | int digest_size; | |
717 | void *digest; | |
718 | }; | |
719 | ||
f6ffca9f | 720 | struct drbd_peer_request { |
45bb912b | 721 | struct drbd_work w; |
85719573 | 722 | struct drbd_epoch *epoch; /* for writes */ |
45bb912b LE |
723 | struct page *pages; |
724 | atomic_t pending_bios; | |
010f6e67 | 725 | struct drbd_interval i; |
45bb912b LE |
726 | /* see comments on ee flag bits below */ |
727 | unsigned long flags; | |
85719573 PR |
728 | union { |
729 | u64 block_id; | |
730 | struct digest_info *digest; | |
731 | }; | |
45bb912b LE |
732 | }; |
733 | ||
734 | /* ee flag bits. | |
735 | * While corresponding bios are in flight, the only modification will be | |
736 | * set_bit WAS_ERROR, which has to be atomic. | |
737 | * If no bios are in flight yet, or all have been completed, | |
738 | * non-atomic modification to ee->flags is ok. | |
739 | */ | |
b411b363 PR |
740 | enum { |
741 | __EE_CALL_AL_COMPLETE_IO, | |
b411b363 | 742 | __EE_MAY_SET_IN_SYNC, |
45bb912b | 743 | |
45bb912b LE |
744 | /* In case a barrier failed, |
745 | * we need to resubmit without the barrier flag. */ | |
746 | __EE_RESUBMITTED, | |
747 | ||
6c852bec | 748 | /* we may have several bios per peer request. |
45bb912b LE |
749 | * if any of those fail, we set this flag atomically |
750 | * from the endio callback */ | |
751 | __EE_WAS_ERROR, | |
c36c3ced LE |
752 | |
753 | /* This ee has a pointer to a digest instead of a block id */ | |
754 | __EE_HAS_DIGEST, | |
b411b363 PR |
755 | }; |
756 | #define EE_CALL_AL_COMPLETE_IO (1<<__EE_CALL_AL_COMPLETE_IO) | |
b411b363 | 757 | #define EE_MAY_SET_IN_SYNC (1<<__EE_MAY_SET_IN_SYNC) |
45bb912b LE |
758 | #define EE_RESUBMITTED (1<<__EE_RESUBMITTED) |
759 | #define EE_WAS_ERROR (1<<__EE_WAS_ERROR) | |
c36c3ced | 760 | #define EE_HAS_DIGEST (1<<__EE_HAS_DIGEST) |
b411b363 | 761 | |
01a311a5 | 762 | /* flag bits per mdev */ |
b411b363 | 763 | enum { |
25985edc | 764 | CREATE_BARRIER, /* next P_DATA is preceded by a P_BARRIER */ |
b411b363 PR |
765 | UNPLUG_QUEUED, /* only relevant with kernel 2.4 */ |
766 | UNPLUG_REMOTE, /* sending a "UnplugRemote" could help */ | |
767 | MD_DIRTY, /* current uuids and flags not yet on disk */ | |
b411b363 | 768 | USE_DEGR_WFC_T, /* degr-wfc-timeout instead of wfc-timeout. */ |
b411b363 PR |
769 | CL_ST_CHG_SUCCESS, |
770 | CL_ST_CHG_FAIL, | |
771 | CRASHED_PRIMARY, /* This node was a crashed primary. | |
772 | * Gets cleared when the state.conn | |
773 | * goes into C_CONNECTED state. */ | |
19f843aa | 774 | NO_BARRIER_SUPP, /* underlying block device doesn't implement barriers */ |
b411b363 PR |
775 | CONSIDER_RESYNC, |
776 | ||
a8a4e51e | 777 | MD_NO_FUA, /* Users wants us to not use FUA/FLUSH on meta data dev */ |
b411b363 PR |
778 | SUSPEND_IO, /* suspend application io */ |
779 | BITMAP_IO, /* suspend application io; | |
780 | once no more io in flight, start bitmap io */ | |
781 | BITMAP_IO_QUEUED, /* Started bitmap IO */ | |
82f59cc6 LE |
782 | GO_DISKLESS, /* Disk is being detached, on io-error or admin request. */ |
783 | WAS_IO_ERROR, /* Local disk failed returned IO error */ | |
b411b363 | 784 | RESYNC_AFTER_NEG, /* Resync after online grow after the attach&negotiate finished. */ |
b411b363 PR |
785 | CONFIG_PENDING, /* serialization of (re)configuration requests. |
786 | * if set, also prevents the device from dying */ | |
787 | DEVICE_DYING, /* device became unconfigured, | |
788 | * but worker thread is still handling the cleanup. | |
789 | * reconfiguring (nl_disk_conf, nl_net_conf) is dissalowed, | |
790 | * while this is set. */ | |
791 | RESIZE_PENDING, /* Size change detected locally, waiting for the response from | |
792 | * the peer, if it changed there as well. */ | |
cf14c2e9 | 793 | CONN_DRY_RUN, /* Expect disconnect after resync handshake. */ |
43a5182c | 794 | NEW_CUR_UUID, /* Create new current UUID when thawing IO */ |
0778286a | 795 | AL_SUSPENDED, /* Activity logging is currently suspended. */ |
370a43e7 | 796 | AHEAD_TO_SYNC_SOURCE, /* Ahead -> SyncSource queued */ |
e64a3294 | 797 | B_RS_H_DONE, /* Before resync handler done (already executed) */ |
b411b363 PR |
798 | }; |
799 | ||
800 | struct drbd_bitmap; /* opaque for drbd_conf */ | |
801 | ||
20ceb2b2 LE |
802 | /* definition of bits in bm_flags to be used in drbd_bm_lock |
803 | * and drbd_bitmap_io and friends. */ | |
804 | enum bm_flag { | |
805 | /* do we need to kfree, or vfree bm_pages? */ | |
806 | BM_P_VMALLOCED = 0x10000, /* internal use only, will be masked out */ | |
807 | ||
808 | /* currently locked for bulk operation */ | |
809 | BM_LOCKED_MASK = 0x7, | |
810 | ||
811 | /* in detail, that is: */ | |
812 | BM_DONT_CLEAR = 0x1, | |
813 | BM_DONT_SET = 0x2, | |
814 | BM_DONT_TEST = 0x4, | |
815 | ||
816 | /* (test bit, count bit) allowed (common case) */ | |
817 | BM_LOCKED_TEST_ALLOWED = 0x3, | |
818 | ||
819 | /* testing bits, as well as setting new bits allowed, but clearing bits | |
820 | * would be unexpected. Used during bitmap receive. Setting new bits | |
821 | * requires sending of "out-of-sync" information, though. */ | |
822 | BM_LOCKED_SET_ALLOWED = 0x1, | |
823 | ||
824 | /* clear is not expected while bitmap is locked for bulk operation */ | |
825 | }; | |
826 | ||
827 | ||
b411b363 PR |
828 | /* TODO sort members for performance |
829 | * MAYBE group them further */ | |
830 | ||
831 | /* THINK maybe we actually want to use the default "event/%s" worker threads | |
832 | * or similar in linux 2.6, which uses per cpu data and threads. | |
b411b363 PR |
833 | */ |
834 | struct drbd_work_queue { | |
835 | struct list_head q; | |
836 | struct semaphore s; /* producers up it, worker down()s it */ | |
837 | spinlock_t q_lock; /* to protect the list. */ | |
838 | }; | |
839 | ||
840 | struct drbd_socket { | |
841 | struct drbd_work_queue work; | |
842 | struct mutex mutex; | |
843 | struct socket *socket; | |
844 | /* this way we get our | |
845 | * send/receive buffers off the stack */ | |
846 | union p_polymorph sbuf; | |
847 | union p_polymorph rbuf; | |
848 | }; | |
849 | ||
850 | struct drbd_md { | |
851 | u64 md_offset; /* sector offset to 'super' block */ | |
852 | ||
853 | u64 la_size_sect; /* last agreed size, unit sectors */ | |
854 | u64 uuid[UI_SIZE]; | |
855 | u64 device_uuid; | |
856 | u32 flags; | |
857 | u32 md_size_sect; | |
858 | ||
859 | s32 al_offset; /* signed relative sector offset to al area */ | |
860 | s32 bm_offset; /* signed relative sector offset to bitmap */ | |
861 | ||
862 | /* u32 al_nr_extents; important for restoring the AL | |
863 | * is stored into sync_conf.al_extents, which in turn | |
864 | * gets applied to act_log->nr_elements | |
865 | */ | |
866 | }; | |
867 | ||
868 | /* for sync_conf and other types... */ | |
869 | #define NL_PACKET(name, number, fields) struct name { fields }; | |
870 | #define NL_INTEGER(pn,pr,member) int member; | |
871 | #define NL_INT64(pn,pr,member) __u64 member; | |
872 | #define NL_BIT(pn,pr,member) unsigned member:1; | |
873 | #define NL_STRING(pn,pr,member,len) unsigned char member[len]; int member ## _len; | |
874 | #include "linux/drbd_nl.h" | |
875 | ||
876 | struct drbd_backing_dev { | |
877 | struct block_device *backing_bdev; | |
878 | struct block_device *md_bdev; | |
b411b363 PR |
879 | struct drbd_md md; |
880 | struct disk_conf dc; /* The user provided config... */ | |
881 | sector_t known_size; /* last known size of that backing device */ | |
882 | }; | |
883 | ||
884 | struct drbd_md_io { | |
885 | struct drbd_conf *mdev; | |
886 | struct completion event; | |
887 | int error; | |
888 | }; | |
889 | ||
890 | struct bm_io_work { | |
891 | struct drbd_work w; | |
892 | char *why; | |
20ceb2b2 | 893 | enum bm_flag flags; |
b411b363 PR |
894 | int (*io_fn)(struct drbd_conf *mdev); |
895 | void (*done)(struct drbd_conf *mdev, int rv); | |
896 | }; | |
897 | ||
898 | enum write_ordering_e { | |
899 | WO_none, | |
900 | WO_drain_io, | |
901 | WO_bdev_flush, | |
b411b363 PR |
902 | }; |
903 | ||
778f271d PR |
904 | struct fifo_buffer { |
905 | int *values; | |
906 | unsigned int head_index; | |
907 | unsigned int size; | |
908 | }; | |
909 | ||
01a311a5 PR |
910 | /* flag bits per tconn */ |
911 | enum { | |
912 | NET_CONGESTED, /* The data socket is congested */ | |
25703f83 | 913 | DISCARD_CONCURRENT, /* Set on one node, cleared on the peer! */ |
e43ef195 | 914 | SEND_PING, /* whether asender should send a ping asap */ |
808e37b8 | 915 | SIGNAL_ASENDER, /* whether asender wants to be interrupted */ |
2a67d8b9 | 916 | GOT_PING_ACK, /* set when we receive a ping_ack packet, ping_wait gets woken */ |
fc3b10a4 PR |
917 | CONN_WD_ST_CHG_OKAY, |
918 | CONN_WD_ST_CHG_FAIL, | |
01a311a5 PR |
919 | }; |
920 | ||
2111438b PR |
921 | struct drbd_tconn { /* is a resource from the config file */ |
922 | char *name; /* Resource name */ | |
923 | struct list_head all_tconn; /* List of all drbd_tconn, prot by global_state_lock */ | |
8410da8f PR |
924 | struct idr volumes; /* <tconn, vnr> to mdev mapping */ |
925 | enum drbd_conns cstate; /* Only C_STANDALONE to C_WF_REPORT_PARAMS */ | |
926 | struct mutex cstate_mutex; /* Protects graceful disconnects */ | |
2111438b | 927 | |
062e879c | 928 | unsigned long flags; |
2111438b | 929 | struct net_conf *net_conf; /* protected by get_net_conf() and put_net_conf() */ |
b2fb6dbe PR |
930 | atomic_t net_cnt; /* Users of net_conf */ |
931 | wait_queue_head_t net_cnt_wait; | |
2a67d8b9 | 932 | wait_queue_head_t ping_wait; /* Woken upon reception of a ping, and a state change */ |
e42325a5 PR |
933 | |
934 | struct drbd_socket data; /* data/barrier/cstate/parameter packets */ | |
935 | struct drbd_socket meta; /* ping/ack (metadata) packets */ | |
31890f4a PR |
936 | int agreed_pro_version; /* actually used protocol version */ |
937 | unsigned long last_received; /* in jiffies, either socket */ | |
938 | unsigned int ko_count; | |
e6b3ea83 | 939 | |
87eeee41 PR |
940 | spinlock_t req_lock; |
941 | struct drbd_tl_epoch *unused_spare_tle; /* for pre-allocation */ | |
942 | struct drbd_tl_epoch *newest_tle; | |
943 | struct drbd_tl_epoch *oldest_tle; | |
944 | struct list_head out_of_sequence_requests; | |
945 | ||
a0638456 PR |
946 | struct crypto_hash *cram_hmac_tfm; |
947 | struct crypto_hash *integrity_w_tfm; /* to be used by the worker thread */ | |
948 | struct crypto_hash *integrity_r_tfm; /* to be used by the receiver thread */ | |
949 | void *int_dig_out; | |
950 | void *int_dig_in; | |
951 | void *int_dig_vv; | |
952 | ||
e6b3ea83 PR |
953 | struct drbd_thread receiver; |
954 | struct drbd_thread worker; | |
955 | struct drbd_thread asender; | |
80822284 | 956 | cpumask_var_t cpu_mask; |
2111438b PR |
957 | }; |
958 | ||
b411b363 | 959 | struct drbd_conf { |
2111438b PR |
960 | struct drbd_tconn *tconn; |
961 | int vnr; /* volume number within the connection */ | |
962 | ||
b411b363 PR |
963 | /* things that are stored as / read from meta data on disk */ |
964 | unsigned long flags; | |
965 | ||
966 | /* configured by drbdsetup */ | |
b411b363 PR |
967 | struct syncer_conf sync_conf; |
968 | struct drbd_backing_dev *ldev __protected_by(local); | |
969 | ||
970 | sector_t p_size; /* partner's disk size */ | |
971 | struct request_queue *rq_queue; | |
972 | struct block_device *this_bdev; | |
973 | struct gendisk *vdisk; | |
974 | ||
b411b363 PR |
975 | struct drbd_work resync_work, |
976 | unplug_work, | |
e9e6f3ec | 977 | go_diskless, |
c4752ef1 PR |
978 | md_sync_work, |
979 | start_resync_work; | |
b411b363 PR |
980 | struct timer_list resync_timer; |
981 | struct timer_list md_sync_timer; | |
370a43e7 | 982 | struct timer_list start_resync_timer; |
7fde2be9 | 983 | struct timer_list request_timer; |
ee15b038 LE |
984 | #ifdef DRBD_DEBUG_MD_SYNC |
985 | struct { | |
986 | unsigned int line; | |
987 | const char* func; | |
988 | } last_md_mark_dirty; | |
989 | #endif | |
b411b363 PR |
990 | |
991 | /* Used after attach while negotiating new disk state. */ | |
992 | union drbd_state new_state_tmp; | |
993 | ||
994 | union drbd_state state; | |
995 | wait_queue_head_t misc_wait; | |
996 | wait_queue_head_t state_wait; /* upon each state change. */ | |
997 | unsigned int send_cnt; | |
998 | unsigned int recv_cnt; | |
999 | unsigned int read_cnt; | |
1000 | unsigned int writ_cnt; | |
1001 | unsigned int al_writ_cnt; | |
1002 | unsigned int bm_writ_cnt; | |
1003 | atomic_t ap_bio_cnt; /* Requests we need to complete */ | |
1004 | atomic_t ap_pending_cnt; /* AP data packets on the wire, ack expected */ | |
1005 | atomic_t rs_pending_cnt; /* RS request/data packets on the wire */ | |
1006 | atomic_t unacked_cnt; /* Need to send replys for */ | |
1007 | atomic_t local_cnt; /* Waiting for local completion */ | |
b2fb6dbe | 1008 | |
dac1389c AG |
1009 | /* Interval tree of pending local requests */ |
1010 | struct rb_root read_requests; | |
de696716 AG |
1011 | struct rb_root write_requests; |
1012 | ||
4b0715f0 | 1013 | /* blocks to resync in this run [unit BM_BLOCK_SIZE] */ |
b411b363 | 1014 | unsigned long rs_total; |
4b0715f0 | 1015 | /* number of resync blocks that failed in this run */ |
b411b363 PR |
1016 | unsigned long rs_failed; |
1017 | /* Syncer's start time [unit jiffies] */ | |
1018 | unsigned long rs_start; | |
1019 | /* cumulated time in PausedSyncX state [unit jiffies] */ | |
1020 | unsigned long rs_paused; | |
1d7734a0 LE |
1021 | /* skipped because csum was equal [unit BM_BLOCK_SIZE] */ |
1022 | unsigned long rs_same_csum; | |
1023 | #define DRBD_SYNC_MARKS 8 | |
1024 | #define DRBD_SYNC_MARK_STEP (3*HZ) | |
b411b363 | 1025 | /* block not up-to-date at mark [unit BM_BLOCK_SIZE] */ |
1d7734a0 | 1026 | unsigned long rs_mark_left[DRBD_SYNC_MARKS]; |
b411b363 | 1027 | /* marks's time [unit jiffies] */ |
1d7734a0 LE |
1028 | unsigned long rs_mark_time[DRBD_SYNC_MARKS]; |
1029 | /* current index into rs_mark_{left,time} */ | |
1030 | int rs_last_mark; | |
b411b363 PR |
1031 | |
1032 | /* where does the admin want us to start? (sector) */ | |
1033 | sector_t ov_start_sector; | |
1034 | /* where are we now? (sector) */ | |
1035 | sector_t ov_position; | |
1036 | /* Start sector of out of sync range (to merge printk reporting). */ | |
1037 | sector_t ov_last_oos_start; | |
1038 | /* size of out-of-sync range in sectors. */ | |
1039 | sector_t ov_last_oos_size; | |
1040 | unsigned long ov_left; /* in bits */ | |
1041 | struct crypto_hash *csums_tfm; | |
1042 | struct crypto_hash *verify_tfm; | |
1043 | ||
b411b363 PR |
1044 | struct drbd_bitmap *bitmap; |
1045 | unsigned long bm_resync_fo; /* bit offset for drbd_bm_find_next */ | |
1046 | ||
1047 | /* Used to track operations of resync... */ | |
1048 | struct lru_cache *resync; | |
1049 | /* Number of locked elements in resync LRU */ | |
1050 | unsigned int resync_locked; | |
1051 | /* resync extent number waiting for application requests */ | |
1052 | unsigned int resync_wenr; | |
1053 | ||
1054 | int open_cnt; | |
1055 | u64 *p_uuid; | |
1056 | struct drbd_epoch *current_epoch; | |
1057 | spinlock_t epoch_lock; | |
1058 | unsigned int epochs; | |
1059 | enum write_ordering_e write_ordering; | |
85719573 PR |
1060 | struct list_head active_ee; /* IO in progress (P_DATA gets written to disk) */ |
1061 | struct list_head sync_ee; /* IO in progress (P_RS_DATA_REPLY gets written to disk) */ | |
18b75d75 AG |
1062 | struct list_head done_ee; /* need to send P_WRITE_ACK */ |
1063 | struct list_head read_ee; /* [RS]P_DATA_REQUEST being read */ | |
b411b363 | 1064 | struct list_head net_ee; /* zero-copy network send in progress */ |
b411b363 | 1065 | |
b411b363 | 1066 | int next_barrier_nr; |
b411b363 | 1067 | struct list_head resync_reads; |
435f0740 LE |
1068 | atomic_t pp_in_use; /* allocated from page pool */ |
1069 | atomic_t pp_in_use_by_net; /* sendpage()d, still referenced by tcp */ | |
b411b363 PR |
1070 | wait_queue_head_t ee_wait; |
1071 | struct page *md_io_page; /* one page buffer for md_io */ | |
b411b363 PR |
1072 | struct mutex md_io_mutex; /* protects the md_io_buffer */ |
1073 | spinlock_t al_lock; | |
1074 | wait_queue_head_t al_wait; | |
1075 | struct lru_cache *act_log; /* activity log */ | |
1076 | unsigned int al_tr_number; | |
1077 | int al_tr_cycle; | |
1078 | int al_tr_pos; /* position of the next transaction in the journal */ | |
b411b363 PR |
1079 | wait_queue_head_t seq_wait; |
1080 | atomic_t packet_seq; | |
1081 | unsigned int peer_seq; | |
1082 | spinlock_t peer_seq_lock; | |
1083 | unsigned int minor; | |
1084 | unsigned long comm_bm_set; /* communicated number of set bits. */ | |
b411b363 PR |
1085 | struct bm_io_work bm_io_work; |
1086 | u64 ed_uuid; /* UUID of the exposed data */ | |
8410da8f PR |
1087 | struct mutex own_state_mutex; |
1088 | struct mutex *state_mutex; /* either own_state_mutex or mdev->tconn->cstate_mutex */ | |
b411b363 | 1089 | char congestion_reason; /* Why we where congested... */ |
1d7734a0 LE |
1090 | atomic_t rs_sect_in; /* for incoming resync data rate, SyncTarget */ |
1091 | atomic_t rs_sect_ev; /* for submitted resync data rate, both */ | |
1092 | int rs_last_sect_ev; /* counter to compare with */ | |
1093 | int rs_last_events; /* counter of read or write "events" (unit sectors) | |
1094 | * on the lower level device when we last looked. */ | |
1095 | int c_sync_rate; /* current resync rate after syncer throttle magic */ | |
778f271d PR |
1096 | struct fifo_buffer rs_plan_s; /* correction values of resync planer */ |
1097 | int rs_in_flight; /* resync sectors in flight (to proxy, in proxy and from proxy) */ | |
25985edc | 1098 | int rs_planed; /* resync sectors already planned */ |
759fbdfb | 1099 | atomic_t ap_in_flight; /* App sectors in flight (waiting for ack) */ |
99432fcc PR |
1100 | int peer_max_bio_size; |
1101 | int local_max_bio_size; | |
b411b363 PR |
1102 | }; |
1103 | ||
1104 | static inline struct drbd_conf *minor_to_mdev(unsigned int minor) | |
1105 | { | |
1106 | struct drbd_conf *mdev; | |
1107 | ||
1108 | mdev = minor < minor_count ? minor_table[minor] : NULL; | |
1109 | ||
1110 | return mdev; | |
1111 | } | |
1112 | ||
1113 | static inline unsigned int mdev_to_minor(struct drbd_conf *mdev) | |
1114 | { | |
1115 | return mdev->minor; | |
1116 | } | |
1117 | ||
eefc2f7d PR |
1118 | static inline struct drbd_conf *vnr_to_mdev(struct drbd_tconn *tconn, int vnr) |
1119 | { | |
1120 | return (struct drbd_conf *)idr_find(&tconn->volumes, vnr); | |
1121 | } | |
1122 | ||
25985edc | 1123 | /* returns 1 if it was successful, |
b411b363 PR |
1124 | * returns 0 if there was no data socket. |
1125 | * so wherever you are going to use the data.socket, e.g. do | |
61120870 | 1126 | * if (!drbd_get_data_sock(mdev->tconn)) |
b411b363 PR |
1127 | * return 0; |
1128 | * CODE(); | |
61120870 | 1129 | * drbd_get_data_sock(mdev->tconn); |
b411b363 | 1130 | */ |
61120870 | 1131 | static inline int drbd_get_data_sock(struct drbd_tconn *tconn) |
b411b363 | 1132 | { |
61120870 | 1133 | mutex_lock(&tconn->data.mutex); |
b411b363 PR |
1134 | /* drbd_disconnect() could have called drbd_free_sock() |
1135 | * while we were waiting in down()... */ | |
61120870 PR |
1136 | if (unlikely(tconn->data.socket == NULL)) { |
1137 | mutex_unlock(&tconn->data.mutex); | |
b411b363 PR |
1138 | return 0; |
1139 | } | |
1140 | return 1; | |
1141 | } | |
1142 | ||
61120870 | 1143 | static inline void drbd_put_data_sock(struct drbd_tconn *tconn) |
b411b363 | 1144 | { |
61120870 | 1145 | mutex_unlock(&tconn->data.mutex); |
b411b363 PR |
1146 | } |
1147 | ||
1148 | /* | |
1149 | * function declarations | |
1150 | *************************/ | |
1151 | ||
1152 | /* drbd_main.c */ | |
1153 | ||
e89b591c PR |
1154 | enum dds_flags { |
1155 | DDSF_FORCED = 1, | |
1156 | DDSF_NO_RESYNC = 2, /* Do not run a resync for the new space */ | |
1157 | }; | |
1158 | ||
b411b363 | 1159 | extern void drbd_init_set_defaults(struct drbd_conf *mdev); |
b411b363 PR |
1160 | extern int drbd_thread_start(struct drbd_thread *thi); |
1161 | extern void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait); | |
392c8801 | 1162 | extern char *drbd_task_to_thread_name(struct drbd_tconn *tconn, struct task_struct *task); |
b411b363 | 1163 | #ifdef CONFIG_SMP |
80822284 PR |
1164 | extern void drbd_thread_current_set_cpu(struct drbd_thread *thi); |
1165 | extern void drbd_calc_cpu_mask(struct drbd_tconn *tconn); | |
b411b363 | 1166 | #else |
80822284 | 1167 | #define drbd_thread_current_set_cpu(A) ({}) |
b411b363 PR |
1168 | #define drbd_calc_cpu_mask(A) ({}) |
1169 | #endif | |
1170 | extern void drbd_free_resources(struct drbd_conf *mdev); | |
1171 | extern void tl_release(struct drbd_conf *mdev, unsigned int barrier_nr, | |
1172 | unsigned int set_size); | |
1173 | extern void tl_clear(struct drbd_conf *mdev); | |
1174 | extern void _tl_add_barrier(struct drbd_conf *, struct drbd_tl_epoch *); | |
360cc740 | 1175 | extern void drbd_free_sock(struct drbd_tconn *tconn); |
bedbd2a5 PR |
1176 | extern int drbd_send(struct drbd_tconn *tconn, struct socket *sock, |
1177 | void *buf, size_t size, unsigned msg_flags); | |
dc8228d1 | 1178 | extern int drbd_send_protocol(struct drbd_tconn *tconn); |
b411b363 PR |
1179 | extern int drbd_send_uuids(struct drbd_conf *mdev); |
1180 | extern int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev); | |
5a22db89 | 1181 | extern int drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev); |
e89b591c | 1182 | extern int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags); |
cf29c9d8 PR |
1183 | extern int _conn_send_state_req(struct drbd_tconn *, int vnr, enum drbd_packet cmd, |
1184 | union drbd_state, union drbd_state); | |
b411b363 PR |
1185 | extern int _drbd_send_state(struct drbd_conf *mdev); |
1186 | extern int drbd_send_state(struct drbd_conf *mdev); | |
d38e787e PR |
1187 | extern int _conn_send_cmd(struct drbd_tconn *tconn, int vnr, struct socket *sock, |
1188 | enum drbd_packet cmd, struct p_header *h, size_t size, | |
1189 | unsigned msg_flags); | |
2a67d8b9 PR |
1190 | extern int conn_send_cmd(struct drbd_tconn *tconn, int vnr, int use_data_socket, |
1191 | enum drbd_packet cmd, struct p_header *h, size_t size); | |
61120870 PR |
1192 | extern int conn_send_cmd2(struct drbd_tconn *tconn, enum drbd_packet cmd, |
1193 | char *data, size_t size); | |
b411b363 PR |
1194 | #define USE_DATA_SOCKET 1 |
1195 | #define USE_META_SOCKET 0 | |
b411b363 PR |
1196 | extern int drbd_send_sync_param(struct drbd_conf *mdev, struct syncer_conf *sc); |
1197 | extern int drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr, | |
1198 | u32 set_size); | |
f6ffca9f AG |
1199 | extern int drbd_send_ack(struct drbd_conf *, enum drbd_packet, |
1200 | struct drbd_peer_request *); | |
d8763023 AG |
1201 | extern int drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packet cmd, |
1202 | struct p_block_req *rp); | |
1203 | extern int drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packet cmd, | |
1204 | struct p_data *dp, int data_size); | |
1205 | extern int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packet cmd, | |
b411b363 | 1206 | sector_t sector, int blksize, u64 block_id); |
73a01a18 | 1207 | extern int drbd_send_oos(struct drbd_conf *mdev, struct drbd_request *req); |
f6ffca9f AG |
1208 | extern int drbd_send_block(struct drbd_conf *, enum drbd_packet, |
1209 | struct drbd_peer_request *); | |
b411b363 | 1210 | extern int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req); |
b411b363 PR |
1211 | extern int drbd_send_drequest(struct drbd_conf *mdev, int cmd, |
1212 | sector_t sector, int size, u64 block_id); | |
d8763023 AG |
1213 | extern int drbd_send_drequest_csum(struct drbd_conf *mdev, sector_t sector, |
1214 | int size, void *digest, int digest_size, | |
1215 | enum drbd_packet cmd); | |
b411b363 PR |
1216 | extern int drbd_send_ov_request(struct drbd_conf *mdev,sector_t sector,int size); |
1217 | ||
1218 | extern int drbd_send_bitmap(struct drbd_conf *mdev); | |
1219 | extern int _drbd_send_bitmap(struct drbd_conf *mdev); | |
bf885f8a | 1220 | extern int drbd_send_sr_reply(struct drbd_conf *mdev, enum drbd_state_rv retcode); |
047cd4a6 | 1221 | extern int conn_send_sr_reply(struct drbd_tconn *tconn, enum drbd_state_rv retcode); |
b411b363 PR |
1222 | extern void drbd_free_bc(struct drbd_backing_dev *ldev); |
1223 | extern void drbd_mdev_cleanup(struct drbd_conf *mdev); | |
62b0da3a | 1224 | void drbd_print_uuids(struct drbd_conf *mdev, const char *text); |
b411b363 | 1225 | |
b411b363 PR |
1226 | extern void drbd_md_sync(struct drbd_conf *mdev); |
1227 | extern int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev); | |
b411b363 PR |
1228 | extern void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local); |
1229 | extern void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local); | |
1230 | extern void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local); | |
1231 | extern void _drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local); | |
1232 | extern void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local); | |
1233 | extern void drbd_md_set_flag(struct drbd_conf *mdev, int flags) __must_hold(local); | |
1234 | extern void drbd_md_clear_flag(struct drbd_conf *mdev, int flags)__must_hold(local); | |
1235 | extern int drbd_md_test_flag(struct drbd_backing_dev *, int); | |
ee15b038 | 1236 | #ifndef DRBD_DEBUG_MD_SYNC |
b411b363 | 1237 | extern void drbd_md_mark_dirty(struct drbd_conf *mdev); |
ee15b038 LE |
1238 | #else |
1239 | #define drbd_md_mark_dirty(m) drbd_md_mark_dirty_(m, __LINE__ , __func__ ) | |
1240 | extern void drbd_md_mark_dirty_(struct drbd_conf *mdev, | |
1241 | unsigned int line, const char *func); | |
1242 | #endif | |
b411b363 PR |
1243 | extern void drbd_queue_bitmap_io(struct drbd_conf *mdev, |
1244 | int (*io_fn)(struct drbd_conf *), | |
1245 | void (*done)(struct drbd_conf *, int), | |
20ceb2b2 LE |
1246 | char *why, enum bm_flag flags); |
1247 | extern int drbd_bitmap_io(struct drbd_conf *mdev, | |
1248 | int (*io_fn)(struct drbd_conf *), | |
1249 | char *why, enum bm_flag flags); | |
b411b363 PR |
1250 | extern int drbd_bmio_set_n_write(struct drbd_conf *mdev); |
1251 | extern int drbd_bmio_clear_n_write(struct drbd_conf *mdev); | |
e9e6f3ec | 1252 | extern void drbd_go_diskless(struct drbd_conf *mdev); |
82f59cc6 | 1253 | extern void drbd_ldev_destroy(struct drbd_conf *mdev); |
b411b363 PR |
1254 | |
1255 | ||
1256 | /* Meta data layout | |
1257 | We reserve a 128MB Block (4k aligned) | |
1258 | * either at the end of the backing device | |
3ad2f3fb | 1259 | * or on a separate meta data device. */ |
b411b363 | 1260 | |
b411b363 | 1261 | /* The following numbers are sectors */ |
7ad651b5 LE |
1262 | /* Allows up to about 3.8TB, so if you want more, |
1263 | * you need to use the "flexible" meta data format. */ | |
1264 | #define MD_RESERVED_SECT (128LU << 11) /* 128 MB, unit sectors */ | |
1265 | #define MD_AL_OFFSET 8 /* 8 Sectors after start of meta area */ | |
1266 | #define MD_AL_SECTORS 64 /* = 32 kB on disk activity log ring buffer */ | |
1267 | #define MD_BM_OFFSET (MD_AL_OFFSET + MD_AL_SECTORS) | |
1268 | ||
1269 | /* we do all meta data IO in 4k blocks */ | |
1270 | #define MD_BLOCK_SHIFT 12 | |
1271 | #define MD_BLOCK_SIZE (1<<MD_BLOCK_SHIFT) | |
1272 | ||
1273 | /* One activity log extent represents 4M of storage */ | |
1274 | #define AL_EXTENT_SHIFT 22 | |
b411b363 PR |
1275 | #define AL_EXTENT_SIZE (1<<AL_EXTENT_SHIFT) |
1276 | ||
7ad651b5 LE |
1277 | /* We could make these currently hardcoded constants configurable |
1278 | * variables at create-md time (or even re-configurable at runtime?). | |
1279 | * Which will require some more changes to the DRBD "super block" | |
1280 | * and attach code. | |
1281 | * | |
1282 | * updates per transaction: | |
1283 | * This many changes to the active set can be logged with one transaction. | |
1284 | * This number is arbitrary. | |
1285 | * context per transaction: | |
1286 | * This many context extent numbers are logged with each transaction. | |
1287 | * This number is resulting from the transaction block size (4k), the layout | |
1288 | * of the transaction header, and the number of updates per transaction. | |
1289 | * See drbd_actlog.c:struct al_transaction_on_disk | |
1290 | * */ | |
1291 | #define AL_UPDATES_PER_TRANSACTION 64 // arbitrary | |
1292 | #define AL_CONTEXT_PER_TRANSACTION 919 // (4096 - 36 - 6*64)/4 | |
1293 | ||
b411b363 PR |
1294 | #if BITS_PER_LONG == 32 |
1295 | #define LN2_BPL 5 | |
1296 | #define cpu_to_lel(A) cpu_to_le32(A) | |
1297 | #define lel_to_cpu(A) le32_to_cpu(A) | |
1298 | #elif BITS_PER_LONG == 64 | |
1299 | #define LN2_BPL 6 | |
1300 | #define cpu_to_lel(A) cpu_to_le64(A) | |
1301 | #define lel_to_cpu(A) le64_to_cpu(A) | |
1302 | #else | |
1303 | #error "LN2 of BITS_PER_LONG unknown!" | |
1304 | #endif | |
1305 | ||
1306 | /* resync bitmap */ | |
1307 | /* 16MB sized 'bitmap extent' to track syncer usage */ | |
1308 | struct bm_extent { | |
1309 | int rs_left; /* number of bits set (out of sync) in this extent. */ | |
1310 | int rs_failed; /* number of failed resync requests in this extent. */ | |
1311 | unsigned long flags; | |
1312 | struct lc_element lce; | |
1313 | }; | |
1314 | ||
1315 | #define BME_NO_WRITES 0 /* bm_extent.flags: no more requests on this one! */ | |
1316 | #define BME_LOCKED 1 /* bm_extent.flags: syncer active on this one. */ | |
e3555d85 | 1317 | #define BME_PRIORITY 2 /* finish resync IO on this extent ASAP! App IO waiting! */ |
b411b363 PR |
1318 | |
1319 | /* drbd_bitmap.c */ | |
1320 | /* | |
1321 | * We need to store one bit for a block. | |
1322 | * Example: 1GB disk @ 4096 byte blocks ==> we need 32 KB bitmap. | |
1323 | * Bit 0 ==> local node thinks this block is binary identical on both nodes | |
1324 | * Bit 1 ==> local node thinks this block needs to be synced. | |
1325 | */ | |
1326 | ||
8e26f9cc PR |
1327 | #define SLEEP_TIME (HZ/10) |
1328 | ||
45dfffeb LE |
1329 | /* We do bitmap IO in units of 4k blocks. |
1330 | * We also still have a hardcoded 4k per bit relation. */ | |
1331 | #define BM_BLOCK_SHIFT 12 /* 4k per bit */ | |
b411b363 | 1332 | #define BM_BLOCK_SIZE (1<<BM_BLOCK_SHIFT) |
45dfffeb LE |
1333 | /* mostly arbitrarily set the represented size of one bitmap extent, |
1334 | * aka resync extent, to 16 MiB (which is also 512 Byte worth of bitmap | |
1335 | * at 4k per bit resolution) */ | |
1336 | #define BM_EXT_SHIFT 24 /* 16 MiB per resync extent */ | |
b411b363 PR |
1337 | #define BM_EXT_SIZE (1<<BM_EXT_SHIFT) |
1338 | ||
1339 | #if (BM_EXT_SHIFT != 24) || (BM_BLOCK_SHIFT != 12) | |
1340 | #error "HAVE YOU FIXED drbdmeta AS WELL??" | |
1341 | #endif | |
1342 | ||
1343 | /* thus many _storage_ sectors are described by one bit */ | |
1344 | #define BM_SECT_TO_BIT(x) ((x)>>(BM_BLOCK_SHIFT-9)) | |
1345 | #define BM_BIT_TO_SECT(x) ((sector_t)(x)<<(BM_BLOCK_SHIFT-9)) | |
1346 | #define BM_SECT_PER_BIT BM_BIT_TO_SECT(1) | |
1347 | ||
1348 | /* bit to represented kilo byte conversion */ | |
1349 | #define Bit2KB(bits) ((bits)<<(BM_BLOCK_SHIFT-10)) | |
1350 | ||
1351 | /* in which _bitmap_ extent (resp. sector) the bit for a certain | |
1352 | * _storage_ sector is located in */ | |
1353 | #define BM_SECT_TO_EXT(x) ((x)>>(BM_EXT_SHIFT-9)) | |
1354 | ||
1355 | /* how much _storage_ sectors we have per bitmap sector */ | |
1356 | #define BM_EXT_TO_SECT(x) ((sector_t)(x) << (BM_EXT_SHIFT-9)) | |
1357 | #define BM_SECT_PER_EXT BM_EXT_TO_SECT(1) | |
1358 | ||
1359 | /* in one sector of the bitmap, we have this many activity_log extents. */ | |
1360 | #define AL_EXT_PER_BM_SECT (1 << (BM_EXT_SHIFT - AL_EXTENT_SHIFT)) | |
1361 | #define BM_WORDS_PER_AL_EXT (1 << (AL_EXTENT_SHIFT-BM_BLOCK_SHIFT-LN2_BPL)) | |
1362 | ||
1363 | #define BM_BLOCKS_PER_BM_EXT_B (BM_EXT_SHIFT - BM_BLOCK_SHIFT) | |
1364 | #define BM_BLOCKS_PER_BM_EXT_MASK ((1<<BM_BLOCKS_PER_BM_EXT_B) - 1) | |
1365 | ||
1366 | /* the extent in "PER_EXTENT" below is an activity log extent | |
1367 | * we need that many (long words/bytes) to store the bitmap | |
1368 | * of one AL_EXTENT_SIZE chunk of storage. | |
1369 | * we can store the bitmap for that many AL_EXTENTS within | |
1370 | * one sector of the _on_disk_ bitmap: | |
1371 | * bit 0 bit 37 bit 38 bit (512*8)-1 | |
1372 | * ...|........|........|.. // ..|........| | |
1373 | * sect. 0 `296 `304 ^(512*8*8)-1 | |
1374 | * | |
1375 | #define BM_WORDS_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / BITS_PER_LONG ) | |
1376 | #define BM_BYTES_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / 8 ) // 128 | |
1377 | #define BM_EXT_PER_SECT ( 512 / BM_BYTES_PER_EXTENT ) // 4 | |
1378 | */ | |
1379 | ||
1380 | #define DRBD_MAX_SECTORS_32 (0xffffffffLU) | |
1381 | #define DRBD_MAX_SECTORS_BM \ | |
1382 | ((MD_RESERVED_SECT - MD_BM_OFFSET) * (1LL<<(BM_EXT_SHIFT-9))) | |
1383 | #if DRBD_MAX_SECTORS_BM < DRBD_MAX_SECTORS_32 | |
1384 | #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_BM | |
1385 | #define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_BM | |
36bfc7e2 | 1386 | #elif !defined(CONFIG_LBDAF) && BITS_PER_LONG == 32 |
b411b363 PR |
1387 | #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_32 |
1388 | #define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_32 | |
1389 | #else | |
1390 | #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_BM | |
1391 | /* 16 TB in units of sectors */ | |
1392 | #if BITS_PER_LONG == 32 | |
1393 | /* adjust by one page worth of bitmap, | |
1394 | * so we won't wrap around in drbd_bm_find_next_bit. | |
1395 | * you should use 64bit OS for that much storage, anyways. */ | |
1396 | #define DRBD_MAX_SECTORS_FLEX BM_BIT_TO_SECT(0xffff7fff) | |
1397 | #else | |
4b0715f0 LE |
1398 | /* we allow up to 1 PiB now on 64bit architecture with "flexible" meta data */ |
1399 | #define DRBD_MAX_SECTORS_FLEX (1UL << 51) | |
1400 | /* corresponds to (1UL << 38) bits right now. */ | |
b411b363 PR |
1401 | #endif |
1402 | #endif | |
1403 | ||
d5373389 | 1404 | #define HT_SHIFT 8 |
1816a2b4 | 1405 | #define DRBD_MAX_BIO_SIZE (1U<<(9+HT_SHIFT)) |
99432fcc | 1406 | #define DRBD_MAX_BIO_SIZE_SAFE (1 << 12) /* Works always = 4k */ |
b411b363 | 1407 | |
d5373389 | 1408 | #define DRBD_MAX_SIZE_H80_PACKET (1 << 15) /* The old header only allows packets up to 32Kib data */ |
b411b363 PR |
1409 | |
1410 | extern int drbd_bm_init(struct drbd_conf *mdev); | |
02d9a94b | 1411 | extern int drbd_bm_resize(struct drbd_conf *mdev, sector_t sectors, int set_new_bits); |
b411b363 PR |
1412 | extern void drbd_bm_cleanup(struct drbd_conf *mdev); |
1413 | extern void drbd_bm_set_all(struct drbd_conf *mdev); | |
1414 | extern void drbd_bm_clear_all(struct drbd_conf *mdev); | |
4b0715f0 | 1415 | /* set/clear/test only a few bits at a time */ |
b411b363 PR |
1416 | extern int drbd_bm_set_bits( |
1417 | struct drbd_conf *mdev, unsigned long s, unsigned long e); | |
1418 | extern int drbd_bm_clear_bits( | |
1419 | struct drbd_conf *mdev, unsigned long s, unsigned long e); | |
4b0715f0 LE |
1420 | extern int drbd_bm_count_bits( |
1421 | struct drbd_conf *mdev, const unsigned long s, const unsigned long e); | |
1422 | /* bm_set_bits variant for use while holding drbd_bm_lock, | |
1423 | * may process the whole bitmap in one go */ | |
b411b363 PR |
1424 | extern void _drbd_bm_set_bits(struct drbd_conf *mdev, |
1425 | const unsigned long s, const unsigned long e); | |
1426 | extern int drbd_bm_test_bit(struct drbd_conf *mdev, unsigned long bitnr); | |
1427 | extern int drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr); | |
19f843aa | 1428 | extern int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(local); |
b411b363 | 1429 | extern int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local); |
45dfffeb | 1430 | extern void drbd_bm_mark_for_writeout(struct drbd_conf *mdev, int page_nr); |
b411b363 | 1431 | extern int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local); |
45dfffeb | 1432 | extern int drbd_bm_write_hinted(struct drbd_conf *mdev) __must_hold(local); |
b411b363 PR |
1433 | extern unsigned long drbd_bm_ALe_set_all(struct drbd_conf *mdev, |
1434 | unsigned long al_enr); | |
1435 | extern size_t drbd_bm_words(struct drbd_conf *mdev); | |
1436 | extern unsigned long drbd_bm_bits(struct drbd_conf *mdev); | |
1437 | extern sector_t drbd_bm_capacity(struct drbd_conf *mdev); | |
4b0715f0 LE |
1438 | |
1439 | #define DRBD_END_OF_BITMAP (~(unsigned long)0) | |
b411b363 PR |
1440 | extern unsigned long drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo); |
1441 | /* bm_find_next variants for use while you hold drbd_bm_lock() */ | |
1442 | extern unsigned long _drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo); | |
1443 | extern unsigned long _drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo); | |
0778286a | 1444 | extern unsigned long _drbd_bm_total_weight(struct drbd_conf *mdev); |
b411b363 PR |
1445 | extern unsigned long drbd_bm_total_weight(struct drbd_conf *mdev); |
1446 | extern int drbd_bm_rs_done(struct drbd_conf *mdev); | |
1447 | /* for receive_bitmap */ | |
1448 | extern void drbd_bm_merge_lel(struct drbd_conf *mdev, size_t offset, | |
1449 | size_t number, unsigned long *buffer); | |
19f843aa | 1450 | /* for _drbd_send_bitmap */ |
b411b363 PR |
1451 | extern void drbd_bm_get_lel(struct drbd_conf *mdev, size_t offset, |
1452 | size_t number, unsigned long *buffer); | |
1453 | ||
20ceb2b2 | 1454 | extern void drbd_bm_lock(struct drbd_conf *mdev, char *why, enum bm_flag flags); |
b411b363 | 1455 | extern void drbd_bm_unlock(struct drbd_conf *mdev); |
b411b363 PR |
1456 | /* drbd_main.c */ |
1457 | ||
1458 | extern struct kmem_cache *drbd_request_cache; | |
6c852bec | 1459 | extern struct kmem_cache *drbd_ee_cache; /* peer requests */ |
b411b363 PR |
1460 | extern struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */ |
1461 | extern struct kmem_cache *drbd_al_ext_cache; /* activity log extents */ | |
1462 | extern mempool_t *drbd_request_mempool; | |
1463 | extern mempool_t *drbd_ee_mempool; | |
1464 | ||
1465 | extern struct page *drbd_pp_pool; /* drbd's page pool */ | |
1466 | extern spinlock_t drbd_pp_lock; | |
1467 | extern int drbd_pp_vacant; | |
1468 | extern wait_queue_head_t drbd_pp_wait; | |
1469 | ||
1470 | extern rwlock_t global_state_lock; | |
1471 | ||
1472 | extern struct drbd_conf *drbd_new_device(unsigned int minor); | |
1473 | extern void drbd_free_mdev(struct drbd_conf *mdev); | |
1474 | ||
2111438b PR |
1475 | struct drbd_tconn *drbd_new_tconn(char *name); |
1476 | extern void drbd_free_tconn(struct drbd_tconn *tconn); | |
1477 | ||
b411b363 PR |
1478 | extern int proc_details; |
1479 | ||
1480 | /* drbd_req */ | |
2f58dcfc | 1481 | extern int drbd_make_request(struct request_queue *q, struct bio *bio); |
b411b363 PR |
1482 | extern int drbd_read_remote(struct drbd_conf *mdev, struct drbd_request *req); |
1483 | extern int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct bio_vec *bvec); | |
1484 | extern int is_valid_ar_handle(struct drbd_request *, sector_t); | |
1485 | ||
1486 | ||
1487 | /* drbd_nl.c */ | |
1488 | extern void drbd_suspend_io(struct drbd_conf *mdev); | |
1489 | extern void drbd_resume_io(struct drbd_conf *mdev); | |
1490 | extern char *ppsize(char *buf, unsigned long long size); | |
a393db6f | 1491 | extern sector_t drbd_new_dev_size(struct drbd_conf *, struct drbd_backing_dev *, int); |
b411b363 | 1492 | enum determine_dev_size { dev_size_error = -1, unchanged = 0, shrunk = 1, grew = 2 }; |
24c4830c | 1493 | extern enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *, enum dds_flags) __must_hold(local); |
b411b363 | 1494 | extern void resync_after_online_grow(struct drbd_conf *); |
99432fcc | 1495 | extern void drbd_reconsider_max_bio_size(struct drbd_conf *mdev); |
bf885f8a AG |
1496 | extern enum drbd_state_rv drbd_set_role(struct drbd_conf *mdev, |
1497 | enum drbd_role new_role, | |
1498 | int force); | |
87f7be4c PR |
1499 | extern enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev); |
1500 | extern void drbd_try_outdate_peer_async(struct drbd_conf *mdev); | |
b411b363 PR |
1501 | extern int drbd_khelper(struct drbd_conf *mdev, char *cmd); |
1502 | ||
1503 | /* drbd_worker.c */ | |
1504 | extern int drbd_worker(struct drbd_thread *thi); | |
1505 | extern int drbd_alter_sa(struct drbd_conf *mdev, int na); | |
1506 | extern void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side); | |
1507 | extern void resume_next_sg(struct drbd_conf *mdev); | |
1508 | extern void suspend_other_sg(struct drbd_conf *mdev); | |
1509 | extern int drbd_resync_finished(struct drbd_conf *mdev); | |
1510 | /* maybe rather drbd_main.c ? */ | |
1511 | extern int drbd_md_sync_page_io(struct drbd_conf *mdev, | |
1512 | struct drbd_backing_dev *bdev, sector_t sector, int rw); | |
1513 | extern void drbd_ov_oos_found(struct drbd_conf*, sector_t, int); | |
9bd28d3c | 1514 | extern void drbd_rs_controller_reset(struct drbd_conf *mdev); |
b411b363 PR |
1515 | |
1516 | static inline void ov_oos_print(struct drbd_conf *mdev) | |
1517 | { | |
1518 | if (mdev->ov_last_oos_size) { | |
1519 | dev_err(DEV, "Out of sync: start=%llu, size=%lu (sectors)\n", | |
1520 | (unsigned long long)mdev->ov_last_oos_start, | |
1521 | (unsigned long)mdev->ov_last_oos_size); | |
1522 | } | |
1523 | mdev->ov_last_oos_size=0; | |
1524 | } | |
1525 | ||
1526 | ||
45bb912b | 1527 | extern void drbd_csum_bio(struct drbd_conf *, struct crypto_hash *, struct bio *, void *); |
f6ffca9f AG |
1528 | extern void drbd_csum_ee(struct drbd_conf *, struct crypto_hash *, |
1529 | struct drbd_peer_request *, void *); | |
b411b363 | 1530 | /* worker callbacks */ |
00d56944 PR |
1531 | extern int w_req_cancel_conflict(struct drbd_work *, int); |
1532 | extern int w_read_retry_remote(struct drbd_work *, int); | |
1533 | extern int w_e_end_data_req(struct drbd_work *, int); | |
1534 | extern int w_e_end_rsdata_req(struct drbd_work *, int); | |
1535 | extern int w_e_end_csum_rs_req(struct drbd_work *, int); | |
1536 | extern int w_e_end_ov_reply(struct drbd_work *, int); | |
1537 | extern int w_e_end_ov_req(struct drbd_work *, int); | |
1538 | extern int w_ov_finished(struct drbd_work *, int); | |
1539 | extern int w_resync_timer(struct drbd_work *, int); | |
1540 | extern int w_resume_next_sg(struct drbd_work *, int); | |
1541 | extern int w_send_write_hint(struct drbd_work *, int); | |
1542 | extern int w_send_dblock(struct drbd_work *, int); | |
1543 | extern int w_send_barrier(struct drbd_work *, int); | |
1544 | extern int w_send_read_req(struct drbd_work *, int); | |
1545 | extern int w_prev_work_done(struct drbd_work *, int); | |
1546 | extern int w_e_reissue(struct drbd_work *, int); | |
1547 | extern int w_restart_disk_io(struct drbd_work *, int); | |
1548 | extern int w_send_oos(struct drbd_work *, int); | |
1549 | extern int w_start_resync(struct drbd_work *, int); | |
b411b363 PR |
1550 | |
1551 | extern void resync_timer_fn(unsigned long data); | |
370a43e7 | 1552 | extern void start_resync_timer_fn(unsigned long data); |
b411b363 PR |
1553 | |
1554 | /* drbd_receiver.c */ | |
e3555d85 | 1555 | extern int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector); |
fbe29dec AG |
1556 | extern int drbd_submit_peer_request(struct drbd_conf *, |
1557 | struct drbd_peer_request *, const unsigned, | |
1558 | const int); | |
b411b363 | 1559 | extern int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list); |
f6ffca9f AG |
1560 | extern struct drbd_peer_request *drbd_alloc_ee(struct drbd_conf *, |
1561 | u64, sector_t, unsigned int, | |
1562 | gfp_t) __must_hold(local); | |
1563 | extern void drbd_free_some_ee(struct drbd_conf *, struct drbd_peer_request *, | |
1564 | int); | |
435f0740 LE |
1565 | #define drbd_free_ee(m,e) drbd_free_some_ee(m, e, 0) |
1566 | #define drbd_free_net_ee(m,e) drbd_free_some_ee(m, e, 1) | |
b411b363 PR |
1567 | extern void drbd_wait_ee_list_empty(struct drbd_conf *mdev, |
1568 | struct list_head *head); | |
1569 | extern void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, | |
1570 | struct list_head *head); | |
1571 | extern void drbd_set_recv_tcq(struct drbd_conf *mdev, int tcq_enabled); | |
1572 | extern void _drbd_clear_done_ee(struct drbd_conf *mdev, struct list_head *to_be_freed); | |
a21e9298 | 1573 | extern void drbd_flush_workqueue(struct drbd_conf *mdev); |
b411b363 PR |
1574 | |
1575 | /* yes, there is kernel_setsockopt, but only since 2.6.18. we don't need to | |
1576 | * mess with get_fs/set_fs, we know we are KERNEL_DS always. */ | |
1577 | static inline int drbd_setsockopt(struct socket *sock, int level, int optname, | |
1578 | char __user *optval, int optlen) | |
1579 | { | |
1580 | int err; | |
1581 | if (level == SOL_SOCKET) | |
1582 | err = sock_setsockopt(sock, level, optname, optval, optlen); | |
1583 | else | |
1584 | err = sock->ops->setsockopt(sock, level, optname, optval, | |
1585 | optlen); | |
1586 | return err; | |
1587 | } | |
1588 | ||
1589 | static inline void drbd_tcp_cork(struct socket *sock) | |
1590 | { | |
1591 | int __user val = 1; | |
1592 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK, | |
1593 | (char __user *)&val, sizeof(val)); | |
1594 | } | |
1595 | ||
1596 | static inline void drbd_tcp_uncork(struct socket *sock) | |
1597 | { | |
1598 | int __user val = 0; | |
1599 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK, | |
1600 | (char __user *)&val, sizeof(val)); | |
1601 | } | |
1602 | ||
1603 | static inline void drbd_tcp_nodelay(struct socket *sock) | |
1604 | { | |
1605 | int __user val = 1; | |
1606 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_NODELAY, | |
1607 | (char __user *)&val, sizeof(val)); | |
1608 | } | |
1609 | ||
1610 | static inline void drbd_tcp_quickack(struct socket *sock) | |
1611 | { | |
344fa462 | 1612 | int __user val = 2; |
b411b363 PR |
1613 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_QUICKACK, |
1614 | (char __user *)&val, sizeof(val)); | |
1615 | } | |
1616 | ||
1617 | void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo); | |
1618 | ||
1619 | /* drbd_proc.c */ | |
1620 | extern struct proc_dir_entry *drbd_proc; | |
7d4e9d09 | 1621 | extern const struct file_operations drbd_proc_fops; |
b411b363 PR |
1622 | extern const char *drbd_conn_str(enum drbd_conns s); |
1623 | extern const char *drbd_role_str(enum drbd_role s); | |
1624 | ||
1625 | /* drbd_actlog.c */ | |
1626 | extern void drbd_al_begin_io(struct drbd_conf *mdev, sector_t sector); | |
1627 | extern void drbd_al_complete_io(struct drbd_conf *mdev, sector_t sector); | |
1628 | extern void drbd_rs_complete_io(struct drbd_conf *mdev, sector_t sector); | |
1629 | extern int drbd_rs_begin_io(struct drbd_conf *mdev, sector_t sector); | |
1630 | extern int drbd_try_rs_begin_io(struct drbd_conf *mdev, sector_t sector); | |
1631 | extern void drbd_rs_cancel_all(struct drbd_conf *mdev); | |
1632 | extern int drbd_rs_del_all(struct drbd_conf *mdev); | |
1633 | extern void drbd_rs_failed_io(struct drbd_conf *mdev, | |
1634 | sector_t sector, int size); | |
1635 | extern int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *); | |
ea5442af | 1636 | extern void drbd_advance_rs_marks(struct drbd_conf *mdev, unsigned long still_to_go); |
b411b363 PR |
1637 | extern void __drbd_set_in_sync(struct drbd_conf *mdev, sector_t sector, |
1638 | int size, const char *file, const unsigned int line); | |
1639 | #define drbd_set_in_sync(mdev, sector, size) \ | |
1640 | __drbd_set_in_sync(mdev, sector, size, __FILE__, __LINE__) | |
73a01a18 | 1641 | extern int __drbd_set_out_of_sync(struct drbd_conf *mdev, sector_t sector, |
b411b363 PR |
1642 | int size, const char *file, const unsigned int line); |
1643 | #define drbd_set_out_of_sync(mdev, sector, size) \ | |
1644 | __drbd_set_out_of_sync(mdev, sector, size, __FILE__, __LINE__) | |
1645 | extern void drbd_al_apply_to_bm(struct drbd_conf *mdev); | |
b411b363 PR |
1646 | extern void drbd_al_shrink(struct drbd_conf *mdev); |
1647 | ||
1648 | ||
1649 | /* drbd_nl.c */ | |
1650 | ||
1651 | void drbd_nl_cleanup(void); | |
1652 | int __init drbd_nl_init(void); | |
1653 | void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state); | |
1654 | void drbd_bcast_sync_progress(struct drbd_conf *mdev); | |
f6ffca9f AG |
1655 | void drbd_bcast_ee(struct drbd_conf *, const char *, const int, const char *, |
1656 | const char *, const struct drbd_peer_request *); | |
b411b363 PR |
1657 | |
1658 | ||
b411b363 PR |
1659 | /* |
1660 | * inline helper functions | |
1661 | *************************/ | |
1662 | ||
45bb912b LE |
1663 | /* see also page_chain_add and friends in drbd_receiver.c */ |
1664 | static inline struct page *page_chain_next(struct page *page) | |
1665 | { | |
1666 | return (struct page *)page_private(page); | |
1667 | } | |
1668 | #define page_chain_for_each(page) \ | |
1669 | for (; page && ({ prefetch(page_chain_next(page)); 1; }); \ | |
1670 | page = page_chain_next(page)) | |
1671 | #define page_chain_for_each_safe(page, n) \ | |
1672 | for (; page && ({ n = page_chain_next(page); 1; }); page = n) | |
1673 | ||
1674 | static inline int drbd_bio_has_active_page(struct bio *bio) | |
1675 | { | |
1676 | struct bio_vec *bvec; | |
1677 | int i; | |
1678 | ||
1679 | __bio_for_each_segment(bvec, bio, i, 0) { | |
1680 | if (page_count(bvec->bv_page) > 1) | |
1681 | return 1; | |
1682 | } | |
1683 | ||
1684 | return 0; | |
1685 | } | |
1686 | ||
db830c46 | 1687 | static inline int drbd_ee_has_active_page(struct drbd_peer_request *peer_req) |
45bb912b | 1688 | { |
db830c46 | 1689 | struct page *page = peer_req->pages; |
45bb912b LE |
1690 | page_chain_for_each(page) { |
1691 | if (page_count(page) > 1) | |
1692 | return 1; | |
1693 | } | |
1694 | return 0; | |
1695 | } | |
1696 | ||
bf885f8a AG |
1697 | static inline enum drbd_state_rv |
1698 | _drbd_set_state(struct drbd_conf *mdev, union drbd_state ns, | |
1699 | enum chg_state_flags flags, struct completion *done) | |
b411b363 | 1700 | { |
bf885f8a | 1701 | enum drbd_state_rv rv; |
b411b363 PR |
1702 | |
1703 | read_lock(&global_state_lock); | |
1704 | rv = __drbd_set_state(mdev, ns, flags, done); | |
1705 | read_unlock(&global_state_lock); | |
1706 | ||
1707 | return rv; | |
1708 | } | |
1709 | ||
b411b363 PR |
1710 | #define __drbd_chk_io_error(m,f) __drbd_chk_io_error_(m,f, __func__) |
1711 | static inline void __drbd_chk_io_error_(struct drbd_conf *mdev, int forcedetach, const char *where) | |
1712 | { | |
1713 | switch (mdev->ldev->dc.on_io_error) { | |
1714 | case EP_PASS_ON: | |
1715 | if (!forcedetach) { | |
7383506c | 1716 | if (__ratelimit(&drbd_ratelimit_state)) |
82f59cc6 | 1717 | dev_err(DEV, "Local IO failed in %s.\n", where); |
d2e17807 PR |
1718 | if (mdev->state.disk > D_INCONSISTENT) |
1719 | _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_HARD, NULL); | |
b411b363 PR |
1720 | break; |
1721 | } | |
1722 | /* NOTE fall through to detach case if forcedetach set */ | |
1723 | case EP_DETACH: | |
1724 | case EP_CALL_HELPER: | |
82f59cc6 | 1725 | set_bit(WAS_IO_ERROR, &mdev->flags); |
b411b363 PR |
1726 | if (mdev->state.disk > D_FAILED) { |
1727 | _drbd_set_state(_NS(mdev, disk, D_FAILED), CS_HARD, NULL); | |
82f59cc6 LE |
1728 | dev_err(DEV, |
1729 | "Local IO failed in %s. Detaching...\n", where); | |
b411b363 PR |
1730 | } |
1731 | break; | |
1732 | } | |
1733 | } | |
1734 | ||
1735 | /** | |
1736 | * drbd_chk_io_error: Handle the on_io_error setting, should be called from all io completion handlers | |
1737 | * @mdev: DRBD device. | |
1738 | * @error: Error code passed to the IO completion callback | |
1739 | * @forcedetach: Force detach. I.e. the error happened while accessing the meta data | |
1740 | * | |
1741 | * See also drbd_main.c:after_state_ch() if (os.disk > D_FAILED && ns.disk == D_FAILED) | |
1742 | */ | |
1743 | #define drbd_chk_io_error(m,e,f) drbd_chk_io_error_(m,e,f, __func__) | |
1744 | static inline void drbd_chk_io_error_(struct drbd_conf *mdev, | |
1745 | int error, int forcedetach, const char *where) | |
1746 | { | |
1747 | if (error) { | |
1748 | unsigned long flags; | |
87eeee41 | 1749 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
b411b363 | 1750 | __drbd_chk_io_error_(mdev, forcedetach, where); |
87eeee41 | 1751 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
b411b363 PR |
1752 | } |
1753 | } | |
1754 | ||
1755 | ||
1756 | /** | |
1757 | * drbd_md_first_sector() - Returns the first sector number of the meta data area | |
1758 | * @bdev: Meta data block device. | |
1759 | * | |
1760 | * BTW, for internal meta data, this happens to be the maximum capacity | |
1761 | * we could agree upon with our peer node. | |
1762 | */ | |
1763 | static inline sector_t drbd_md_first_sector(struct drbd_backing_dev *bdev) | |
1764 | { | |
1765 | switch (bdev->dc.meta_dev_idx) { | |
1766 | case DRBD_MD_INDEX_INTERNAL: | |
1767 | case DRBD_MD_INDEX_FLEX_INT: | |
1768 | return bdev->md.md_offset + bdev->md.bm_offset; | |
1769 | case DRBD_MD_INDEX_FLEX_EXT: | |
1770 | default: | |
1771 | return bdev->md.md_offset; | |
1772 | } | |
1773 | } | |
1774 | ||
1775 | /** | |
1776 | * drbd_md_last_sector() - Return the last sector number of the meta data area | |
1777 | * @bdev: Meta data block device. | |
1778 | */ | |
1779 | static inline sector_t drbd_md_last_sector(struct drbd_backing_dev *bdev) | |
1780 | { | |
1781 | switch (bdev->dc.meta_dev_idx) { | |
1782 | case DRBD_MD_INDEX_INTERNAL: | |
1783 | case DRBD_MD_INDEX_FLEX_INT: | |
1784 | return bdev->md.md_offset + MD_AL_OFFSET - 1; | |
1785 | case DRBD_MD_INDEX_FLEX_EXT: | |
1786 | default: | |
1787 | return bdev->md.md_offset + bdev->md.md_size_sect; | |
1788 | } | |
1789 | } | |
1790 | ||
1791 | /* Returns the number of 512 byte sectors of the device */ | |
1792 | static inline sector_t drbd_get_capacity(struct block_device *bdev) | |
1793 | { | |
1794 | /* return bdev ? get_capacity(bdev->bd_disk) : 0; */ | |
77304d2a | 1795 | return bdev ? i_size_read(bdev->bd_inode) >> 9 : 0; |
b411b363 PR |
1796 | } |
1797 | ||
1798 | /** | |
1799 | * drbd_get_max_capacity() - Returns the capacity we announce to out peer | |
1800 | * @bdev: Meta data block device. | |
1801 | * | |
1802 | * returns the capacity we announce to out peer. we clip ourselves at the | |
1803 | * various MAX_SECTORS, because if we don't, current implementation will | |
1804 | * oops sooner or later | |
1805 | */ | |
1806 | static inline sector_t drbd_get_max_capacity(struct drbd_backing_dev *bdev) | |
1807 | { | |
1808 | sector_t s; | |
1809 | switch (bdev->dc.meta_dev_idx) { | |
1810 | case DRBD_MD_INDEX_INTERNAL: | |
1811 | case DRBD_MD_INDEX_FLEX_INT: | |
1812 | s = drbd_get_capacity(bdev->backing_bdev) | |
1813 | ? min_t(sector_t, DRBD_MAX_SECTORS_FLEX, | |
1814 | drbd_md_first_sector(bdev)) | |
1815 | : 0; | |
1816 | break; | |
1817 | case DRBD_MD_INDEX_FLEX_EXT: | |
1818 | s = min_t(sector_t, DRBD_MAX_SECTORS_FLEX, | |
1819 | drbd_get_capacity(bdev->backing_bdev)); | |
1820 | /* clip at maximum size the meta device can support */ | |
1821 | s = min_t(sector_t, s, | |
1822 | BM_EXT_TO_SECT(bdev->md.md_size_sect | |
1823 | - bdev->md.bm_offset)); | |
1824 | break; | |
1825 | default: | |
1826 | s = min_t(sector_t, DRBD_MAX_SECTORS, | |
1827 | drbd_get_capacity(bdev->backing_bdev)); | |
1828 | } | |
1829 | return s; | |
1830 | } | |
1831 | ||
1832 | /** | |
1833 | * drbd_md_ss__() - Return the sector number of our meta data super block | |
1834 | * @mdev: DRBD device. | |
1835 | * @bdev: Meta data block device. | |
1836 | */ | |
1837 | static inline sector_t drbd_md_ss__(struct drbd_conf *mdev, | |
1838 | struct drbd_backing_dev *bdev) | |
1839 | { | |
1840 | switch (bdev->dc.meta_dev_idx) { | |
1841 | default: /* external, some index */ | |
1842 | return MD_RESERVED_SECT * bdev->dc.meta_dev_idx; | |
1843 | case DRBD_MD_INDEX_INTERNAL: | |
1844 | /* with drbd08, internal meta data is always "flexible" */ | |
1845 | case DRBD_MD_INDEX_FLEX_INT: | |
1846 | /* sizeof(struct md_on_disk_07) == 4k | |
1847 | * position: last 4k aligned block of 4k size */ | |
1848 | if (!bdev->backing_bdev) { | |
1849 | if (__ratelimit(&drbd_ratelimit_state)) { | |
1850 | dev_err(DEV, "bdev->backing_bdev==NULL\n"); | |
1851 | dump_stack(); | |
1852 | } | |
1853 | return 0; | |
1854 | } | |
1855 | return (drbd_get_capacity(bdev->backing_bdev) & ~7ULL) | |
1856 | - MD_AL_OFFSET; | |
1857 | case DRBD_MD_INDEX_FLEX_EXT: | |
1858 | return 0; | |
1859 | } | |
1860 | } | |
1861 | ||
b411b363 PR |
1862 | static inline void |
1863 | drbd_queue_work_front(struct drbd_work_queue *q, struct drbd_work *w) | |
1864 | { | |
1865 | unsigned long flags; | |
1866 | spin_lock_irqsave(&q->q_lock, flags); | |
1867 | list_add(&w->list, &q->q); | |
1868 | up(&q->s); /* within the spinlock, | |
1869 | see comment near end of drbd_worker() */ | |
1870 | spin_unlock_irqrestore(&q->q_lock, flags); | |
1871 | } | |
1872 | ||
1873 | static inline void | |
1874 | drbd_queue_work(struct drbd_work_queue *q, struct drbd_work *w) | |
1875 | { | |
1876 | unsigned long flags; | |
1877 | spin_lock_irqsave(&q->q_lock, flags); | |
1878 | list_add_tail(&w->list, &q->q); | |
1879 | up(&q->s); /* within the spinlock, | |
1880 | see comment near end of drbd_worker() */ | |
1881 | spin_unlock_irqrestore(&q->q_lock, flags); | |
1882 | } | |
1883 | ||
0625ac19 | 1884 | static inline void wake_asender(struct drbd_tconn *tconn) |
b411b363 | 1885 | { |
0625ac19 PR |
1886 | if (test_bit(SIGNAL_ASENDER, &tconn->flags)) |
1887 | force_sig(DRBD_SIG, tconn->asender.task); | |
b411b363 PR |
1888 | } |
1889 | ||
0625ac19 | 1890 | static inline void request_ping(struct drbd_tconn *tconn) |
b411b363 | 1891 | { |
0625ac19 PR |
1892 | set_bit(SEND_PING, &tconn->flags); |
1893 | wake_asender(tconn); | |
b411b363 PR |
1894 | } |
1895 | ||
d38e787e PR |
1896 | static inline int _drbd_send_cmd(struct drbd_conf *mdev, struct socket *sock, |
1897 | enum drbd_packet cmd, struct p_header *h, size_t size, | |
1898 | unsigned msg_flags) | |
1899 | { | |
1900 | return _conn_send_cmd(mdev->tconn, mdev->vnr, sock, cmd, h, size, msg_flags); | |
1901 | } | |
1902 | ||
2a67d8b9 PR |
1903 | static inline int drbd_send_cmd(struct drbd_conf *mdev, int use_data_socket, |
1904 | enum drbd_packet cmd, struct p_header *h, size_t size) | |
1905 | { | |
1906 | return conn_send_cmd(mdev->tconn, mdev->vnr, use_data_socket, cmd, h, size); | |
1907 | } | |
1908 | ||
b411b363 | 1909 | static inline int drbd_send_short_cmd(struct drbd_conf *mdev, |
d8763023 | 1910 | enum drbd_packet cmd) |
b411b363 | 1911 | { |
c012949a | 1912 | struct p_header h; |
b411b363 PR |
1913 | return drbd_send_cmd(mdev, USE_DATA_SOCKET, cmd, &h, sizeof(h)); |
1914 | } | |
1915 | ||
2a67d8b9 | 1916 | static inline int drbd_send_ping(struct drbd_tconn *tconn) |
b411b363 | 1917 | { |
c012949a | 1918 | struct p_header h; |
2a67d8b9 | 1919 | return conn_send_cmd(tconn, 0, USE_META_SOCKET, P_PING, &h, sizeof(h)); |
b411b363 PR |
1920 | } |
1921 | ||
2a67d8b9 | 1922 | static inline int drbd_send_ping_ack(struct drbd_tconn *tconn) |
b411b363 | 1923 | { |
c012949a | 1924 | struct p_header h; |
2a67d8b9 | 1925 | return conn_send_cmd(tconn, 0, USE_META_SOCKET, P_PING_ACK, &h, sizeof(h)); |
b411b363 PR |
1926 | } |
1927 | ||
cf29c9d8 PR |
1928 | static inline int drbd_send_state_req(struct drbd_conf *mdev, |
1929 | union drbd_state mask, union drbd_state val) | |
1930 | { | |
1931 | return _conn_send_state_req(mdev->tconn, mdev->vnr, P_STATE_CHG_REQ, mask, val); | |
1932 | } | |
1933 | ||
1934 | static inline int conn_send_state_req(struct drbd_tconn *tconn, | |
1935 | union drbd_state mask, union drbd_state val) | |
1936 | { | |
1937 | enum drbd_packet cmd = tconn->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ; | |
1938 | return _conn_send_state_req(tconn, 0, cmd, mask, val); | |
1939 | } | |
1940 | ||
b411b363 PR |
1941 | static inline void drbd_thread_stop(struct drbd_thread *thi) |
1942 | { | |
81e84650 | 1943 | _drbd_thread_stop(thi, false, true); |
b411b363 PR |
1944 | } |
1945 | ||
1946 | static inline void drbd_thread_stop_nowait(struct drbd_thread *thi) | |
1947 | { | |
81e84650 | 1948 | _drbd_thread_stop(thi, false, false); |
b411b363 PR |
1949 | } |
1950 | ||
1951 | static inline void drbd_thread_restart_nowait(struct drbd_thread *thi) | |
1952 | { | |
81e84650 | 1953 | _drbd_thread_stop(thi, true, false); |
b411b363 PR |
1954 | } |
1955 | ||
1956 | /* counts how many answer packets packets we expect from our peer, | |
1957 | * for either explicit application requests, | |
1958 | * or implicit barrier packets as necessary. | |
1959 | * increased: | |
1960 | * w_send_barrier | |
8554df1c | 1961 | * _req_mod(req, QUEUE_FOR_NET_WRITE or QUEUE_FOR_NET_READ); |
b411b363 PR |
1962 | * it is much easier and equally valid to count what we queue for the |
1963 | * worker, even before it actually was queued or send. | |
1964 | * (drbd_make_request_common; recovery path on read io-error) | |
1965 | * decreased: | |
1966 | * got_BarrierAck (respective tl_clear, tl_clear_barrier) | |
8554df1c | 1967 | * _req_mod(req, DATA_RECEIVED) |
b411b363 | 1968 | * [from receive_DataReply] |
8554df1c | 1969 | * _req_mod(req, WRITE_ACKED_BY_PEER or RECV_ACKED_BY_PEER or NEG_ACKED) |
b411b363 PR |
1970 | * [from got_BlockAck (P_WRITE_ACK, P_RECV_ACK)] |
1971 | * for some reason it is NOT decreased in got_NegAck, | |
1972 | * but in the resulting cleanup code from report_params. | |
1973 | * we should try to remember the reason for that... | |
8554df1c AG |
1974 | * _req_mod(req, SEND_FAILED or SEND_CANCELED) |
1975 | * _req_mod(req, CONNECTION_LOST_WHILE_PENDING) | |
b411b363 PR |
1976 | * [from tl_clear_barrier] |
1977 | */ | |
1978 | static inline void inc_ap_pending(struct drbd_conf *mdev) | |
1979 | { | |
1980 | atomic_inc(&mdev->ap_pending_cnt); | |
1981 | } | |
1982 | ||
1983 | #define ERR_IF_CNT_IS_NEGATIVE(which) \ | |
1984 | if (atomic_read(&mdev->which) < 0) \ | |
1985 | dev_err(DEV, "in %s:%d: " #which " = %d < 0 !\n", \ | |
1986 | __func__ , __LINE__ , \ | |
1987 | atomic_read(&mdev->which)) | |
1988 | ||
1989 | #define dec_ap_pending(mdev) do { \ | |
1990 | typecheck(struct drbd_conf *, mdev); \ | |
1991 | if (atomic_dec_and_test(&mdev->ap_pending_cnt)) \ | |
1992 | wake_up(&mdev->misc_wait); \ | |
1993 | ERR_IF_CNT_IS_NEGATIVE(ap_pending_cnt); } while (0) | |
1994 | ||
1995 | /* counts how many resync-related answers we still expect from the peer | |
1996 | * increase decrease | |
1997 | * C_SYNC_TARGET sends P_RS_DATA_REQUEST (and expects P_RS_DATA_REPLY) | |
25985edc | 1998 | * C_SYNC_SOURCE sends P_RS_DATA_REPLY (and expects P_WRITE_ACK with ID_SYNCER) |
b411b363 PR |
1999 | * (or P_NEG_ACK with ID_SYNCER) |
2000 | */ | |
2001 | static inline void inc_rs_pending(struct drbd_conf *mdev) | |
2002 | { | |
2003 | atomic_inc(&mdev->rs_pending_cnt); | |
2004 | } | |
2005 | ||
2006 | #define dec_rs_pending(mdev) do { \ | |
2007 | typecheck(struct drbd_conf *, mdev); \ | |
2008 | atomic_dec(&mdev->rs_pending_cnt); \ | |
2009 | ERR_IF_CNT_IS_NEGATIVE(rs_pending_cnt); } while (0) | |
2010 | ||
2011 | /* counts how many answers we still need to send to the peer. | |
2012 | * increased on | |
2013 | * receive_Data unless protocol A; | |
2014 | * we need to send a P_RECV_ACK (proto B) | |
2015 | * or P_WRITE_ACK (proto C) | |
2016 | * receive_RSDataReply (recv_resync_read) we need to send a P_WRITE_ACK | |
2017 | * receive_DataRequest (receive_RSDataRequest) we need to send back P_DATA | |
2018 | * receive_Barrier_* we need to send a P_BARRIER_ACK | |
2019 | */ | |
2020 | static inline void inc_unacked(struct drbd_conf *mdev) | |
2021 | { | |
2022 | atomic_inc(&mdev->unacked_cnt); | |
2023 | } | |
2024 | ||
2025 | #define dec_unacked(mdev) do { \ | |
2026 | typecheck(struct drbd_conf *, mdev); \ | |
2027 | atomic_dec(&mdev->unacked_cnt); \ | |
2028 | ERR_IF_CNT_IS_NEGATIVE(unacked_cnt); } while (0) | |
2029 | ||
2030 | #define sub_unacked(mdev, n) do { \ | |
2031 | typecheck(struct drbd_conf *, mdev); \ | |
2032 | atomic_sub(n, &mdev->unacked_cnt); \ | |
2033 | ERR_IF_CNT_IS_NEGATIVE(unacked_cnt); } while (0) | |
2034 | ||
2035 | ||
b2fb6dbe | 2036 | static inline void put_net_conf(struct drbd_tconn *tconn) |
b411b363 | 2037 | { |
b2fb6dbe PR |
2038 | if (atomic_dec_and_test(&tconn->net_cnt)) |
2039 | wake_up(&tconn->net_cnt_wait); | |
b411b363 PR |
2040 | } |
2041 | ||
2042 | /** | |
89e58e75 | 2043 | * get_net_conf() - Increase ref count on mdev->tconn->net_conf; Returns 0 if nothing there |
b411b363 PR |
2044 | * @mdev: DRBD device. |
2045 | * | |
89e58e75 | 2046 | * You have to call put_net_conf() when finished working with mdev->tconn->net_conf. |
b411b363 | 2047 | */ |
b2fb6dbe | 2048 | static inline int get_net_conf(struct drbd_tconn *tconn) |
b411b363 PR |
2049 | { |
2050 | int have_net_conf; | |
2051 | ||
b2fb6dbe | 2052 | atomic_inc(&tconn->net_cnt); |
bbeb641c | 2053 | have_net_conf = tconn->cstate >= C_UNCONNECTED; |
b411b363 | 2054 | if (!have_net_conf) |
b2fb6dbe | 2055 | put_net_conf(tconn); |
b411b363 PR |
2056 | return have_net_conf; |
2057 | } | |
2058 | ||
2059 | /** | |
2060 | * get_ldev() - Increase the ref count on mdev->ldev. Returns 0 if there is no ldev | |
2061 | * @M: DRBD device. | |
2062 | * | |
2063 | * You have to call put_ldev() when finished working with mdev->ldev. | |
2064 | */ | |
2065 | #define get_ldev(M) __cond_lock(local, _get_ldev_if_state(M,D_INCONSISTENT)) | |
2066 | #define get_ldev_if_state(M,MINS) __cond_lock(local, _get_ldev_if_state(M,MINS)) | |
2067 | ||
2068 | static inline void put_ldev(struct drbd_conf *mdev) | |
2069 | { | |
1d7734a0 | 2070 | int i = atomic_dec_return(&mdev->local_cnt); |
9a0d9d03 LE |
2071 | |
2072 | /* This may be called from some endio handler, | |
2073 | * so we must not sleep here. */ | |
2074 | ||
b411b363 | 2075 | __release(local); |
1d7734a0 | 2076 | D_ASSERT(i >= 0); |
e9e6f3ec | 2077 | if (i == 0) { |
82f59cc6 LE |
2078 | if (mdev->state.disk == D_DISKLESS) |
2079 | /* even internal references gone, safe to destroy */ | |
2080 | drbd_ldev_destroy(mdev); | |
e9e6f3ec | 2081 | if (mdev->state.disk == D_FAILED) |
82f59cc6 | 2082 | /* all application IO references gone. */ |
e9e6f3ec | 2083 | drbd_go_diskless(mdev); |
b411b363 | 2084 | wake_up(&mdev->misc_wait); |
e9e6f3ec | 2085 | } |
b411b363 PR |
2086 | } |
2087 | ||
2088 | #ifndef __CHECKER__ | |
2089 | static inline int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins) | |
2090 | { | |
2091 | int io_allowed; | |
2092 | ||
82f59cc6 LE |
2093 | /* never get a reference while D_DISKLESS */ |
2094 | if (mdev->state.disk == D_DISKLESS) | |
2095 | return 0; | |
2096 | ||
b411b363 PR |
2097 | atomic_inc(&mdev->local_cnt); |
2098 | io_allowed = (mdev->state.disk >= mins); | |
2099 | if (!io_allowed) | |
2100 | put_ldev(mdev); | |
2101 | return io_allowed; | |
2102 | } | |
2103 | #else | |
2104 | extern int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins); | |
2105 | #endif | |
2106 | ||
2107 | /* you must have an "get_ldev" reference */ | |
2108 | static inline void drbd_get_syncer_progress(struct drbd_conf *mdev, | |
2109 | unsigned long *bits_left, unsigned int *per_mil_done) | |
2110 | { | |
4b0715f0 LE |
2111 | /* this is to break it at compile time when we change that, in case we |
2112 | * want to support more than (1<<32) bits on a 32bit arch. */ | |
b411b363 PR |
2113 | typecheck(unsigned long, mdev->rs_total); |
2114 | ||
2115 | /* note: both rs_total and rs_left are in bits, i.e. in | |
2116 | * units of BM_BLOCK_SIZE. | |
2117 | * for the percentage, we don't care. */ | |
2118 | ||
439d5953 LE |
2119 | if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T) |
2120 | *bits_left = mdev->ov_left; | |
2121 | else | |
2122 | *bits_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; | |
b411b363 PR |
2123 | /* >> 10 to prevent overflow, |
2124 | * +1 to prevent division by zero */ | |
2125 | if (*bits_left > mdev->rs_total) { | |
2126 | /* doh. maybe a logic bug somewhere. | |
2127 | * may also be just a race condition | |
2128 | * between this and a disconnect during sync. | |
2129 | * for now, just prevent in-kernel buffer overflow. | |
2130 | */ | |
2131 | smp_rmb(); | |
2132 | dev_warn(DEV, "cs:%s rs_left=%lu > rs_total=%lu (rs_failed %lu)\n", | |
2133 | drbd_conn_str(mdev->state.conn), | |
2134 | *bits_left, mdev->rs_total, mdev->rs_failed); | |
2135 | *per_mil_done = 0; | |
2136 | } else { | |
4b0715f0 LE |
2137 | /* Make sure the division happens in long context. |
2138 | * We allow up to one petabyte storage right now, | |
2139 | * at a granularity of 4k per bit that is 2**38 bits. | |
2140 | * After shift right and multiplication by 1000, | |
2141 | * this should still fit easily into a 32bit long, | |
2142 | * so we don't need a 64bit division on 32bit arch. | |
2143 | * Note: currently we don't support such large bitmaps on 32bit | |
2144 | * arch anyways, but no harm done to be prepared for it here. | |
2145 | */ | |
2146 | unsigned int shift = mdev->rs_total >= (1ULL << 32) ? 16 : 10; | |
2147 | unsigned long left = *bits_left >> shift; | |
2148 | unsigned long total = 1UL + (mdev->rs_total >> shift); | |
2149 | unsigned long tmp = 1000UL - left * 1000UL/total; | |
b411b363 PR |
2150 | *per_mil_done = tmp; |
2151 | } | |
2152 | } | |
2153 | ||
2154 | ||
2155 | /* this throttles on-the-fly application requests | |
2156 | * according to max_buffers settings; | |
2157 | * maybe re-implement using semaphores? */ | |
2158 | static inline int drbd_get_max_buffers(struct drbd_conf *mdev) | |
2159 | { | |
2160 | int mxb = 1000000; /* arbitrary limit on open requests */ | |
b2fb6dbe | 2161 | if (get_net_conf(mdev->tconn)) { |
89e58e75 | 2162 | mxb = mdev->tconn->net_conf->max_buffers; |
b2fb6dbe | 2163 | put_net_conf(mdev->tconn); |
b411b363 PR |
2164 | } |
2165 | return mxb; | |
2166 | } | |
2167 | ||
3719094e | 2168 | static inline int drbd_state_is_stable(struct drbd_conf *mdev) |
b411b363 | 2169 | { |
3719094e | 2170 | union drbd_state s = mdev->state; |
b411b363 PR |
2171 | |
2172 | /* DO NOT add a default clause, we want the compiler to warn us | |
2173 | * for any newly introduced state we may have forgotten to add here */ | |
2174 | ||
2175 | switch ((enum drbd_conns)s.conn) { | |
2176 | /* new io only accepted when there is no connection, ... */ | |
2177 | case C_STANDALONE: | |
2178 | case C_WF_CONNECTION: | |
2179 | /* ... or there is a well established connection. */ | |
2180 | case C_CONNECTED: | |
2181 | case C_SYNC_SOURCE: | |
2182 | case C_SYNC_TARGET: | |
2183 | case C_VERIFY_S: | |
2184 | case C_VERIFY_T: | |
2185 | case C_PAUSED_SYNC_S: | |
2186 | case C_PAUSED_SYNC_T: | |
67531718 PR |
2187 | case C_AHEAD: |
2188 | case C_BEHIND: | |
3719094e | 2189 | /* transitional states, IO allowed */ |
b411b363 PR |
2190 | case C_DISCONNECTING: |
2191 | case C_UNCONNECTED: | |
2192 | case C_TIMEOUT: | |
2193 | case C_BROKEN_PIPE: | |
2194 | case C_NETWORK_FAILURE: | |
2195 | case C_PROTOCOL_ERROR: | |
2196 | case C_TEAR_DOWN: | |
2197 | case C_WF_REPORT_PARAMS: | |
2198 | case C_STARTING_SYNC_S: | |
2199 | case C_STARTING_SYNC_T: | |
3719094e PR |
2200 | break; |
2201 | ||
2202 | /* Allow IO in BM exchange states with new protocols */ | |
b411b363 | 2203 | case C_WF_BITMAP_S: |
31890f4a | 2204 | if (mdev->tconn->agreed_pro_version < 96) |
3719094e PR |
2205 | return 0; |
2206 | break; | |
2207 | ||
2208 | /* no new io accepted in these states */ | |
b411b363 PR |
2209 | case C_WF_BITMAP_T: |
2210 | case C_WF_SYNC_UUID: | |
2211 | case C_MASK: | |
2212 | /* not "stable" */ | |
2213 | return 0; | |
2214 | } | |
2215 | ||
2216 | switch ((enum drbd_disk_state)s.disk) { | |
2217 | case D_DISKLESS: | |
2218 | case D_INCONSISTENT: | |
2219 | case D_OUTDATED: | |
2220 | case D_CONSISTENT: | |
2221 | case D_UP_TO_DATE: | |
2222 | /* disk state is stable as well. */ | |
2223 | break; | |
2224 | ||
2225 | /* no new io accepted during tansitional states */ | |
2226 | case D_ATTACHING: | |
2227 | case D_FAILED: | |
2228 | case D_NEGOTIATING: | |
2229 | case D_UNKNOWN: | |
2230 | case D_MASK: | |
2231 | /* not "stable" */ | |
2232 | return 0; | |
2233 | } | |
2234 | ||
2235 | return 1; | |
2236 | } | |
2237 | ||
fb22c402 PR |
2238 | static inline int is_susp(union drbd_state s) |
2239 | { | |
2240 | return s.susp || s.susp_nod || s.susp_fen; | |
2241 | } | |
2242 | ||
1b881ef7 | 2243 | static inline bool may_inc_ap_bio(struct drbd_conf *mdev) |
b411b363 PR |
2244 | { |
2245 | int mxb = drbd_get_max_buffers(mdev); | |
2246 | ||
fb22c402 | 2247 | if (is_susp(mdev->state)) |
1b881ef7 | 2248 | return false; |
b411b363 | 2249 | if (test_bit(SUSPEND_IO, &mdev->flags)) |
1b881ef7 | 2250 | return false; |
b411b363 PR |
2251 | |
2252 | /* to avoid potential deadlock or bitmap corruption, | |
2253 | * in various places, we only allow new application io | |
2254 | * to start during "stable" states. */ | |
2255 | ||
2256 | /* no new io accepted when attaching or detaching the disk */ | |
3719094e | 2257 | if (!drbd_state_is_stable(mdev)) |
1b881ef7 | 2258 | return false; |
b411b363 PR |
2259 | |
2260 | /* since some older kernels don't have atomic_add_unless, | |
2261 | * and we are within the spinlock anyways, we have this workaround. */ | |
2262 | if (atomic_read(&mdev->ap_bio_cnt) > mxb) | |
1b881ef7 | 2263 | return false; |
b411b363 | 2264 | if (test_bit(BITMAP_IO, &mdev->flags)) |
1b881ef7 AG |
2265 | return false; |
2266 | return true; | |
b411b363 PR |
2267 | } |
2268 | ||
1b881ef7 | 2269 | static inline bool inc_ap_bio_cond(struct drbd_conf *mdev, int count) |
b411b363 | 2270 | { |
1b881ef7 | 2271 | bool rv = false; |
8869d683 | 2272 | |
87eeee41 | 2273 | spin_lock_irq(&mdev->tconn->req_lock); |
1b881ef7 | 2274 | rv = may_inc_ap_bio(mdev); |
8869d683 PR |
2275 | if (rv) |
2276 | atomic_add(count, &mdev->ap_bio_cnt); | |
87eeee41 | 2277 | spin_unlock_irq(&mdev->tconn->req_lock); |
8869d683 PR |
2278 | |
2279 | return rv; | |
2280 | } | |
b411b363 | 2281 | |
8869d683 PR |
2282 | static inline void inc_ap_bio(struct drbd_conf *mdev, int count) |
2283 | { | |
b411b363 PR |
2284 | /* we wait here |
2285 | * as long as the device is suspended | |
2286 | * until the bitmap is no longer on the fly during connection | |
2287 | * handshake as long as we would exeed the max_buffer limit. | |
2288 | * | |
2289 | * to avoid races with the reconnect code, | |
2290 | * we need to atomic_inc within the spinlock. */ | |
2291 | ||
1b881ef7 | 2292 | wait_event(mdev->misc_wait, inc_ap_bio_cond(mdev, count)); |
b411b363 PR |
2293 | } |
2294 | ||
2295 | static inline void dec_ap_bio(struct drbd_conf *mdev) | |
2296 | { | |
2297 | int mxb = drbd_get_max_buffers(mdev); | |
2298 | int ap_bio = atomic_dec_return(&mdev->ap_bio_cnt); | |
2299 | ||
2300 | D_ASSERT(ap_bio >= 0); | |
2301 | /* this currently does wake_up for every dec_ap_bio! | |
2302 | * maybe rather introduce some type of hysteresis? | |
2303 | * e.g. (ap_bio == mxb/2 || ap_bio == 0) ? */ | |
2304 | if (ap_bio < mxb) | |
2305 | wake_up(&mdev->misc_wait); | |
2306 | if (ap_bio == 0 && test_bit(BITMAP_IO, &mdev->flags)) { | |
2307 | if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags)) | |
e42325a5 | 2308 | drbd_queue_work(&mdev->tconn->data.work, &mdev->bm_io_work.w); |
b411b363 PR |
2309 | } |
2310 | } | |
2311 | ||
62b0da3a | 2312 | static inline int drbd_set_ed_uuid(struct drbd_conf *mdev, u64 val) |
b411b363 | 2313 | { |
62b0da3a | 2314 | int changed = mdev->ed_uuid != val; |
b411b363 | 2315 | mdev->ed_uuid = val; |
62b0da3a | 2316 | return changed; |
b411b363 PR |
2317 | } |
2318 | ||
b411b363 PR |
2319 | static inline int drbd_queue_order_type(struct drbd_conf *mdev) |
2320 | { | |
2321 | /* sorry, we currently have no working implementation | |
2322 | * of distributed TCQ stuff */ | |
2323 | #ifndef QUEUE_ORDERED_NONE | |
2324 | #define QUEUE_ORDERED_NONE 0 | |
2325 | #endif | |
2326 | return QUEUE_ORDERED_NONE; | |
2327 | } | |
2328 | ||
b411b363 PR |
2329 | static inline void drbd_md_flush(struct drbd_conf *mdev) |
2330 | { | |
2331 | int r; | |
2332 | ||
a8a4e51e | 2333 | if (test_bit(MD_NO_FUA, &mdev->flags)) |
b411b363 PR |
2334 | return; |
2335 | ||
dd3932ed | 2336 | r = blkdev_issue_flush(mdev->ldev->md_bdev, GFP_KERNEL, NULL); |
b411b363 | 2337 | if (r) { |
a8a4e51e | 2338 | set_bit(MD_NO_FUA, &mdev->flags); |
b411b363 PR |
2339 | dev_err(DEV, "meta data flush failed with status %d, disabling md-flushes\n", r); |
2340 | } | |
2341 | } | |
2342 | ||
2343 | #endif |