Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_int.h | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | drbd is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
14 | ||
15 | drbd is distributed in the hope that it will be useful, | |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | GNU General Public License for more details. | |
19 | ||
20 | You should have received a copy of the GNU General Public License | |
21 | along with drbd; see the file COPYING. If not, write to | |
22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | ||
24 | */ | |
25 | ||
26 | #ifndef _DRBD_INT_H | |
27 | #define _DRBD_INT_H | |
28 | ||
29 | #include <linux/compiler.h> | |
30 | #include <linux/types.h> | |
31 | #include <linux/version.h> | |
32 | #include <linux/list.h> | |
33 | #include <linux/sched.h> | |
34 | #include <linux/bitops.h> | |
35 | #include <linux/slab.h> | |
36 | #include <linux/crypto.h> | |
132cc538 | 37 | #include <linux/ratelimit.h> |
b411b363 PR |
38 | #include <linux/tcp.h> |
39 | #include <linux/mutex.h> | |
40 | #include <linux/major.h> | |
41 | #include <linux/blkdev.h> | |
42 | #include <linux/genhd.h> | |
43 | #include <net/tcp.h> | |
44 | #include <linux/lru_cache.h> | |
70c71606 | 45 | #include <linux/prefetch.h> |
b8907339 PR |
46 | #include <linux/drbd.h> |
47 | #include "drbd_state.h" | |
b411b363 PR |
48 | |
49 | #ifdef __CHECKER__ | |
50 | # define __protected_by(x) __attribute__((require_context(x,1,999,"rdwr"))) | |
51 | # define __protected_read_by(x) __attribute__((require_context(x,1,999,"read"))) | |
52 | # define __protected_write_by(x) __attribute__((require_context(x,1,999,"write"))) | |
53 | # define __must_hold(x) __attribute__((context(x,1,1), require_context(x,1,999,"call"))) | |
54 | #else | |
55 | # define __protected_by(x) | |
56 | # define __protected_read_by(x) | |
57 | # define __protected_write_by(x) | |
58 | # define __must_hold(x) | |
59 | #endif | |
60 | ||
61 | #define __no_warn(lock, stmt) do { __acquire(lock); stmt; __release(lock); } while (0) | |
62 | ||
63 | /* module parameter, defined in drbd_main.c */ | |
64 | extern unsigned int minor_count; | |
65 | extern int disable_sendpage; | |
66 | extern int allow_oos; | |
67 | extern unsigned int cn_idx; | |
68 | ||
69 | #ifdef CONFIG_DRBD_FAULT_INJECTION | |
70 | extern int enable_faults; | |
71 | extern int fault_rate; | |
72 | extern int fault_devs; | |
73 | #endif | |
74 | ||
75 | extern char usermode_helper[]; | |
76 | ||
77 | ||
b411b363 PR |
78 | /* I don't remember why XCPU ... |
79 | * This is used to wake the asender, | |
80 | * and to interrupt sending the sending task | |
81 | * on disconnect. | |
82 | */ | |
83 | #define DRBD_SIG SIGXCPU | |
84 | ||
85 | /* This is used to stop/restart our threads. | |
86 | * Cannot use SIGTERM nor SIGKILL, since these | |
87 | * are sent out by init on runlevel changes | |
88 | * I choose SIGHUP for now. | |
89 | */ | |
90 | #define DRBD_SIGKILL SIGHUP | |
91 | ||
b411b363 PR |
92 | #define ID_IN_SYNC (4711ULL) |
93 | #define ID_OUT_OF_SYNC (4712ULL) | |
b411b363 | 94 | #define ID_SYNCER (-1ULL) |
579b57ed | 95 | |
4a23f264 | 96 | #define UUID_NEW_BM_OFFSET ((u64)0x0001000000000000ULL) |
b411b363 PR |
97 | |
98 | struct drbd_conf; | |
2111438b | 99 | struct drbd_tconn; |
b411b363 PR |
100 | |
101 | ||
102 | /* to shorten dev_warn(DEV, "msg"); and relatives statements */ | |
103 | #define DEV (disk_to_dev(mdev->vdisk)) | |
104 | ||
60ae4966 PR |
105 | #define conn_printk(LEVEL, TCONN, FMT, ARGS...) \ |
106 | printk(LEVEL "d-con %s: " FMT, TCONN->name , ## ARGS) | |
107 | #define conn_alert(TCONN, FMT, ARGS...) conn_printk(KERN_ALERT, TCONN, FMT, ## ARGS) | |
108 | #define conn_crit(TCONN, FMT, ARGS...) conn_printk(KERN_CRIT, TCONN, FMT, ## ARGS) | |
109 | #define conn_err(TCONN, FMT, ARGS...) conn_printk(KERN_ERR, TCONN, FMT, ## ARGS) | |
110 | #define conn_warn(TCONN, FMT, ARGS...) conn_printk(KERN_WARNING, TCONN, FMT, ## ARGS) | |
111 | #define conn_notice(TCONN, FMT, ARGS...) conn_printk(KERN_NOTICE, TCONN, FMT, ## ARGS) | |
112 | #define conn_info(TCONN, FMT, ARGS...) conn_printk(KERN_INFO, TCONN, FMT, ## ARGS) | |
113 | #define conn_dbg(TCONN, FMT, ARGS...) conn_printk(KERN_DEBUG, TCONN, FMT, ## ARGS) | |
114 | ||
b411b363 PR |
115 | #define D_ASSERT(exp) if (!(exp)) \ |
116 | dev_err(DEV, "ASSERT( " #exp " ) in %s:%d\n", __FILE__, __LINE__) | |
117 | ||
841ce241 AG |
118 | /** |
119 | * expect - Make an assertion | |
120 | * | |
121 | * Unlike the assert macro, this macro returns a boolean result. | |
122 | */ | |
123 | #define expect(exp) ({ \ | |
124 | bool _bool = (exp); \ | |
125 | if (!_bool) \ | |
126 | dev_err(DEV, "ASSERTION %s FAILED in %s\n", \ | |
127 | #exp, __func__); \ | |
128 | _bool; \ | |
129 | }) | |
b411b363 PR |
130 | |
131 | /* Defines to control fault insertion */ | |
132 | enum { | |
133 | DRBD_FAULT_MD_WR = 0, /* meta data write */ | |
134 | DRBD_FAULT_MD_RD = 1, /* read */ | |
135 | DRBD_FAULT_RS_WR = 2, /* resync */ | |
136 | DRBD_FAULT_RS_RD = 3, | |
137 | DRBD_FAULT_DT_WR = 4, /* data */ | |
138 | DRBD_FAULT_DT_RD = 5, | |
139 | DRBD_FAULT_DT_RA = 6, /* data read ahead */ | |
140 | DRBD_FAULT_BM_ALLOC = 7, /* bitmap allocation */ | |
141 | DRBD_FAULT_AL_EE = 8, /* alloc ee */ | |
6b4388ac | 142 | DRBD_FAULT_RECEIVE = 9, /* Changes some bytes upon receiving a [rs]data block */ |
b411b363 PR |
143 | |
144 | DRBD_FAULT_MAX, | |
145 | }; | |
146 | ||
b411b363 PR |
147 | extern unsigned int |
148 | _drbd_insert_fault(struct drbd_conf *mdev, unsigned int type); | |
0cf9d27e | 149 | |
b411b363 PR |
150 | static inline int |
151 | drbd_insert_fault(struct drbd_conf *mdev, unsigned int type) { | |
0cf9d27e | 152 | #ifdef CONFIG_DRBD_FAULT_INJECTION |
b411b363 PR |
153 | return fault_rate && |
154 | (enable_faults & (1<<type)) && | |
155 | _drbd_insert_fault(mdev, type); | |
b411b363 | 156 | #else |
0cf9d27e | 157 | return 0; |
b411b363 | 158 | #endif |
0cf9d27e | 159 | } |
b411b363 PR |
160 | |
161 | /* integer division, round _UP_ to the next integer */ | |
162 | #define div_ceil(A, B) ((A)/(B) + ((A)%(B) ? 1 : 0)) | |
163 | /* usual integer division */ | |
164 | #define div_floor(A, B) ((A)/(B)) | |
165 | ||
166 | /* drbd_meta-data.c (still in drbd_main.c) */ | |
167 | /* 4th incarnation of the disk layout. */ | |
168 | #define DRBD_MD_MAGIC (DRBD_MAGIC+4) | |
169 | ||
170 | extern struct drbd_conf **minor_table; | |
171 | extern struct ratelimit_state drbd_ratelimit_state; | |
172 | ||
173 | /* on the wire */ | |
d8763023 | 174 | enum drbd_packet { |
b411b363 PR |
175 | /* receiver (data socket) */ |
176 | P_DATA = 0x00, | |
177 | P_DATA_REPLY = 0x01, /* Response to P_DATA_REQUEST */ | |
178 | P_RS_DATA_REPLY = 0x02, /* Response to P_RS_DATA_REQUEST */ | |
179 | P_BARRIER = 0x03, | |
180 | P_BITMAP = 0x04, | |
181 | P_BECOME_SYNC_TARGET = 0x05, | |
182 | P_BECOME_SYNC_SOURCE = 0x06, | |
183 | P_UNPLUG_REMOTE = 0x07, /* Used at various times to hint the peer */ | |
184 | P_DATA_REQUEST = 0x08, /* Used to ask for a data block */ | |
185 | P_RS_DATA_REQUEST = 0x09, /* Used to ask for a data block for resync */ | |
186 | P_SYNC_PARAM = 0x0a, | |
187 | P_PROTOCOL = 0x0b, | |
188 | P_UUIDS = 0x0c, | |
189 | P_SIZES = 0x0d, | |
190 | P_STATE = 0x0e, | |
191 | P_SYNC_UUID = 0x0f, | |
192 | P_AUTH_CHALLENGE = 0x10, | |
193 | P_AUTH_RESPONSE = 0x11, | |
194 | P_STATE_CHG_REQ = 0x12, | |
195 | ||
196 | /* asender (meta socket */ | |
197 | P_PING = 0x13, | |
198 | P_PING_ACK = 0x14, | |
199 | P_RECV_ACK = 0x15, /* Used in protocol B */ | |
200 | P_WRITE_ACK = 0x16, /* Used in protocol C */ | |
201 | P_RS_WRITE_ACK = 0x17, /* Is a P_WRITE_ACK, additionally call set_in_sync(). */ | |
202 | P_DISCARD_ACK = 0x18, /* Used in proto C, two-primaries conflict detection */ | |
203 | P_NEG_ACK = 0x19, /* Sent if local disk is unusable */ | |
204 | P_NEG_DREPLY = 0x1a, /* Local disk is broken... */ | |
205 | P_NEG_RS_DREPLY = 0x1b, /* Local disk is broken... */ | |
206 | P_BARRIER_ACK = 0x1c, | |
207 | P_STATE_CHG_REPLY = 0x1d, | |
208 | ||
209 | /* "new" commands, no longer fitting into the ordering scheme above */ | |
210 | ||
211 | P_OV_REQUEST = 0x1e, /* data socket */ | |
212 | P_OV_REPLY = 0x1f, | |
213 | P_OV_RESULT = 0x20, /* meta socket */ | |
214 | P_CSUM_RS_REQUEST = 0x21, /* data socket */ | |
215 | P_RS_IS_IN_SYNC = 0x22, /* meta socket */ | |
216 | P_SYNC_PARAM89 = 0x23, /* data socket, protocol version 89 replacement for P_SYNC_PARAM */ | |
217 | P_COMPRESSED_BITMAP = 0x24, /* compressed or otherwise encoded bitmap transfer */ | |
0ced55a3 PR |
218 | /* P_CKPT_FENCE_REQ = 0x25, * currently reserved for protocol D */ |
219 | /* P_CKPT_DISABLE_REQ = 0x26, * currently reserved for protocol D */ | |
220 | P_DELAY_PROBE = 0x27, /* is used on BOTH sockets */ | |
73a01a18 | 221 | P_OUT_OF_SYNC = 0x28, /* Mark as out of sync (Outrunning), data socket */ |
d612d309 | 222 | P_RS_CANCEL = 0x29, /* meta: Used to cancel RS_DATA_REQUEST packet by SyncSource */ |
b411b363 | 223 | |
d612d309 | 224 | P_MAX_CMD = 0x2A, |
b411b363 PR |
225 | P_MAY_IGNORE = 0x100, /* Flag to test if (cmd > P_MAY_IGNORE) ... */ |
226 | P_MAX_OPT_CMD = 0x101, | |
227 | ||
228 | /* special command ids for handshake */ | |
229 | ||
230 | P_HAND_SHAKE_M = 0xfff1, /* First Packet on the MetaSock */ | |
231 | P_HAND_SHAKE_S = 0xfff2, /* First Packet on the Socket */ | |
232 | ||
233 | P_HAND_SHAKE = 0xfffe /* FIXED for the next century! */ | |
234 | }; | |
235 | ||
d8763023 | 236 | extern const char *cmdname(enum drbd_packet cmd); |
b411b363 PR |
237 | |
238 | /* for sending/receiving the bitmap, | |
239 | * possibly in some encoding scheme */ | |
240 | struct bm_xfer_ctx { | |
241 | /* "const" | |
242 | * stores total bits and long words | |
243 | * of the bitmap, so we don't need to | |
244 | * call the accessor functions over and again. */ | |
245 | unsigned long bm_bits; | |
246 | unsigned long bm_words; | |
247 | /* during xfer, current position within the bitmap */ | |
248 | unsigned long bit_offset; | |
249 | unsigned long word_offset; | |
250 | ||
251 | /* statistics; index: (h->command == P_BITMAP) */ | |
252 | unsigned packets[2]; | |
253 | unsigned bytes[2]; | |
254 | }; | |
255 | ||
256 | extern void INFO_bm_xfer_stats(struct drbd_conf *mdev, | |
257 | const char *direction, struct bm_xfer_ctx *c); | |
258 | ||
259 | static inline void bm_xfer_ctx_bit_to_word_offset(struct bm_xfer_ctx *c) | |
260 | { | |
261 | /* word_offset counts "native long words" (32 or 64 bit), | |
262 | * aligned at 64 bit. | |
263 | * Encoded packet may end at an unaligned bit offset. | |
264 | * In case a fallback clear text packet is transmitted in | |
265 | * between, we adjust this offset back to the last 64bit | |
266 | * aligned "native long word", which makes coding and decoding | |
267 | * the plain text bitmap much more convenient. */ | |
268 | #if BITS_PER_LONG == 64 | |
269 | c->word_offset = c->bit_offset >> 6; | |
270 | #elif BITS_PER_LONG == 32 | |
271 | c->word_offset = c->bit_offset >> 5; | |
272 | c->word_offset &= ~(1UL); | |
273 | #else | |
274 | # error "unsupported BITS_PER_LONG" | |
275 | #endif | |
276 | } | |
277 | ||
278 | #ifndef __packed | |
279 | #define __packed __attribute__((packed)) | |
280 | #endif | |
281 | ||
282 | /* This is the layout for a packet on the wire. | |
283 | * The byteorder is the network byte order. | |
284 | * (except block_id and barrier fields. | |
285 | * these are pointers to local structs | |
286 | * and have no relevance for the partner, | |
287 | * which just echoes them as received.) | |
288 | * | |
289 | * NOTE that the payload starts at a long aligned offset, | |
290 | * regardless of 32 or 64 bit arch! | |
291 | */ | |
0b70a13d | 292 | struct p_header80 { |
b411b363 PR |
293 | u32 magic; |
294 | u16 command; | |
295 | u16 length; /* bytes of data after this header */ | |
b411b363 | 296 | } __packed; |
0b70a13d PR |
297 | |
298 | /* Header for big packets, Used for data packets exceeding 64kB */ | |
299 | struct p_header95 { | |
300 | u16 magic; /* use DRBD_MAGIC_BIG here */ | |
301 | u16 command; | |
00b42537 | 302 | u32 length; /* Use only 24 bits of that. Ignore the highest 8 bit. */ |
0b70a13d PR |
303 | } __packed; |
304 | ||
c012949a PR |
305 | struct p_header { |
306 | union { | |
307 | struct p_header80 h80; | |
308 | struct p_header95 h95; | |
309 | }; | |
310 | u8 payload[0]; | |
0b70a13d | 311 | }; |
b411b363 PR |
312 | |
313 | /* | |
314 | * short commands, packets without payload, plain p_header: | |
315 | * P_PING | |
316 | * P_PING_ACK | |
317 | * P_BECOME_SYNC_TARGET | |
318 | * P_BECOME_SYNC_SOURCE | |
319 | * P_UNPLUG_REMOTE | |
320 | */ | |
321 | ||
322 | /* | |
323 | * commands with out-of-struct payload: | |
324 | * P_BITMAP (no additional fields) | |
325 | * P_DATA, P_DATA_REPLY (see p_data) | |
326 | * P_COMPRESSED_BITMAP (see receive_compressed_bitmap) | |
327 | */ | |
328 | ||
329 | /* these defines must not be changed without changing the protocol version */ | |
76d2e7ec PR |
330 | #define DP_HARDBARRIER 1 /* depricated */ |
331 | #define DP_RW_SYNC 2 /* equals REQ_SYNC */ | |
b411b363 | 332 | #define DP_MAY_SET_IN_SYNC 4 |
721a9602 | 333 | #define DP_UNPLUG 8 /* not used anymore */ |
76d2e7ec PR |
334 | #define DP_FUA 16 /* equals REQ_FUA */ |
335 | #define DP_FLUSH 32 /* equals REQ_FLUSH */ | |
336 | #define DP_DISCARD 64 /* equals REQ_DISCARD */ | |
b411b363 PR |
337 | |
338 | struct p_data { | |
c012949a | 339 | struct p_header head; |
b411b363 PR |
340 | u64 sector; /* 64 bits sector number */ |
341 | u64 block_id; /* to identify the request in protocol B&C */ | |
342 | u32 seq_num; | |
343 | u32 dp_flags; | |
344 | } __packed; | |
345 | ||
346 | /* | |
347 | * commands which share a struct: | |
348 | * p_block_ack: | |
349 | * P_RECV_ACK (proto B), P_WRITE_ACK (proto C), | |
350 | * P_DISCARD_ACK (proto C, two-primaries conflict detection) | |
351 | * p_block_req: | |
352 | * P_DATA_REQUEST, P_RS_DATA_REQUEST | |
353 | */ | |
354 | struct p_block_ack { | |
c012949a | 355 | struct p_header head; |
b411b363 PR |
356 | u64 sector; |
357 | u64 block_id; | |
358 | u32 blksize; | |
359 | u32 seq_num; | |
360 | } __packed; | |
361 | ||
362 | ||
363 | struct p_block_req { | |
c012949a | 364 | struct p_header head; |
b411b363 PR |
365 | u64 sector; |
366 | u64 block_id; | |
367 | u32 blksize; | |
368 | u32 pad; /* to multiple of 8 Byte */ | |
369 | } __packed; | |
370 | ||
371 | /* | |
372 | * commands with their own struct for additional fields: | |
373 | * P_HAND_SHAKE | |
374 | * P_BARRIER | |
375 | * P_BARRIER_ACK | |
376 | * P_SYNC_PARAM | |
377 | * ReportParams | |
378 | */ | |
379 | ||
380 | struct p_handshake { | |
8a22cccc | 381 | struct p_header head; /* Note: vnr will be ignored */ |
b411b363 PR |
382 | u32 protocol_min; |
383 | u32 feature_flags; | |
384 | u32 protocol_max; | |
385 | ||
386 | /* should be more than enough for future enhancements | |
387 | * for now, feature_flags and the reserverd array shall be zero. | |
388 | */ | |
389 | ||
390 | u32 _pad; | |
391 | u64 reserverd[7]; | |
392 | } __packed; | |
393 | /* 80 bytes, FIXED for the next century */ | |
394 | ||
395 | struct p_barrier { | |
c012949a | 396 | struct p_header head; |
b411b363 PR |
397 | u32 barrier; /* barrier number _handle_ only */ |
398 | u32 pad; /* to multiple of 8 Byte */ | |
399 | } __packed; | |
400 | ||
401 | struct p_barrier_ack { | |
c012949a | 402 | struct p_header head; |
b411b363 PR |
403 | u32 barrier; |
404 | u32 set_size; | |
405 | } __packed; | |
406 | ||
407 | struct p_rs_param { | |
c012949a | 408 | struct p_header head; |
b411b363 PR |
409 | u32 rate; |
410 | ||
411 | /* Since protocol version 88 and higher. */ | |
412 | char verify_alg[0]; | |
413 | } __packed; | |
414 | ||
415 | struct p_rs_param_89 { | |
c012949a | 416 | struct p_header head; |
b411b363 PR |
417 | u32 rate; |
418 | /* protocol version 89: */ | |
419 | char verify_alg[SHARED_SECRET_MAX]; | |
420 | char csums_alg[SHARED_SECRET_MAX]; | |
421 | } __packed; | |
422 | ||
8e26f9cc | 423 | struct p_rs_param_95 { |
c012949a | 424 | struct p_header head; |
8e26f9cc PR |
425 | u32 rate; |
426 | char verify_alg[SHARED_SECRET_MAX]; | |
427 | char csums_alg[SHARED_SECRET_MAX]; | |
428 | u32 c_plan_ahead; | |
429 | u32 c_delay_target; | |
430 | u32 c_fill_target; | |
431 | u32 c_max_rate; | |
432 | } __packed; | |
433 | ||
cf14c2e9 PR |
434 | enum drbd_conn_flags { |
435 | CF_WANT_LOSE = 1, | |
436 | CF_DRY_RUN = 2, | |
437 | }; | |
438 | ||
b411b363 | 439 | struct p_protocol { |
c012949a | 440 | struct p_header head; |
b411b363 PR |
441 | u32 protocol; |
442 | u32 after_sb_0p; | |
443 | u32 after_sb_1p; | |
444 | u32 after_sb_2p; | |
cf14c2e9 | 445 | u32 conn_flags; |
b411b363 PR |
446 | u32 two_primaries; |
447 | ||
448 | /* Since protocol version 87 and higher. */ | |
449 | char integrity_alg[0]; | |
450 | ||
451 | } __packed; | |
452 | ||
453 | struct p_uuids { | |
c012949a | 454 | struct p_header head; |
b411b363 PR |
455 | u64 uuid[UI_EXTENDED_SIZE]; |
456 | } __packed; | |
457 | ||
458 | struct p_rs_uuid { | |
c012949a | 459 | struct p_header head; |
b411b363 PR |
460 | u64 uuid; |
461 | } __packed; | |
462 | ||
463 | struct p_sizes { | |
c012949a | 464 | struct p_header head; |
b411b363 PR |
465 | u64 d_size; /* size of disk */ |
466 | u64 u_size; /* user requested size */ | |
467 | u64 c_size; /* current exported size */ | |
1816a2b4 | 468 | u32 max_bio_size; /* Maximal size of a BIO */ |
e89b591c PR |
469 | u16 queue_order_type; /* not yet implemented in DRBD*/ |
470 | u16 dds_flags; /* use enum dds_flags here. */ | |
b411b363 PR |
471 | } __packed; |
472 | ||
473 | struct p_state { | |
c012949a | 474 | struct p_header head; |
b411b363 PR |
475 | u32 state; |
476 | } __packed; | |
477 | ||
478 | struct p_req_state { | |
c012949a | 479 | struct p_header head; |
b411b363 PR |
480 | u32 mask; |
481 | u32 val; | |
482 | } __packed; | |
483 | ||
484 | struct p_req_state_reply { | |
c012949a | 485 | struct p_header head; |
b411b363 PR |
486 | u32 retcode; |
487 | } __packed; | |
488 | ||
489 | struct p_drbd06_param { | |
490 | u64 size; | |
491 | u32 state; | |
492 | u32 blksize; | |
493 | u32 protocol; | |
494 | u32 version; | |
495 | u32 gen_cnt[5]; | |
496 | u32 bit_map_gen[5]; | |
497 | } __packed; | |
498 | ||
499 | struct p_discard { | |
c012949a | 500 | struct p_header head; |
b411b363 PR |
501 | u64 block_id; |
502 | u32 seq_num; | |
503 | u32 pad; | |
504 | } __packed; | |
505 | ||
73a01a18 | 506 | struct p_block_desc { |
c012949a | 507 | struct p_header head; |
73a01a18 PR |
508 | u64 sector; |
509 | u32 blksize; | |
510 | u32 pad; /* to multiple of 8 Byte */ | |
511 | } __packed; | |
512 | ||
b411b363 PR |
513 | /* Valid values for the encoding field. |
514 | * Bump proto version when changing this. */ | |
515 | enum drbd_bitmap_code { | |
516 | /* RLE_VLI_Bytes = 0, | |
517 | * and other bit variants had been defined during | |
518 | * algorithm evaluation. */ | |
519 | RLE_VLI_Bits = 2, | |
520 | }; | |
521 | ||
522 | struct p_compressed_bm { | |
c012949a | 523 | struct p_header head; |
b411b363 PR |
524 | /* (encoding & 0x0f): actual encoding, see enum drbd_bitmap_code |
525 | * (encoding & 0x80): polarity (set/unset) of first runlength | |
526 | * ((encoding >> 4) & 0x07): pad_bits, number of trailing zero bits | |
527 | * used to pad up to head.length bytes | |
528 | */ | |
529 | u8 encoding; | |
530 | ||
531 | u8 code[0]; | |
532 | } __packed; | |
533 | ||
0b70a13d | 534 | struct p_delay_probe93 { |
c012949a | 535 | struct p_header head; |
0b70a13d PR |
536 | u32 seq_num; /* sequence number to match the two probe packets */ |
537 | u32 offset; /* usecs the probe got sent after the reference time point */ | |
0ced55a3 PR |
538 | } __packed; |
539 | ||
b411b363 PR |
540 | /* DCBP: Drbd Compressed Bitmap Packet ... */ |
541 | static inline enum drbd_bitmap_code | |
542 | DCBP_get_code(struct p_compressed_bm *p) | |
543 | { | |
544 | return (enum drbd_bitmap_code)(p->encoding & 0x0f); | |
545 | } | |
546 | ||
547 | static inline void | |
548 | DCBP_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code) | |
549 | { | |
550 | BUG_ON(code & ~0xf); | |
551 | p->encoding = (p->encoding & ~0xf) | code; | |
552 | } | |
553 | ||
554 | static inline int | |
555 | DCBP_get_start(struct p_compressed_bm *p) | |
556 | { | |
557 | return (p->encoding & 0x80) != 0; | |
558 | } | |
559 | ||
560 | static inline void | |
561 | DCBP_set_start(struct p_compressed_bm *p, int set) | |
562 | { | |
563 | p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0); | |
564 | } | |
565 | ||
566 | static inline int | |
567 | DCBP_get_pad_bits(struct p_compressed_bm *p) | |
568 | { | |
569 | return (p->encoding >> 4) & 0x7; | |
570 | } | |
571 | ||
572 | static inline void | |
573 | DCBP_set_pad_bits(struct p_compressed_bm *p, int n) | |
574 | { | |
575 | BUG_ON(n & ~0x7); | |
576 | p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4); | |
577 | } | |
578 | ||
579 | /* one bitmap packet, including the p_header, | |
580 | * should fit within one _architecture independend_ page. | |
581 | * so we need to use the fixed size 4KiB page size | |
25985edc | 582 | * most architectures have used for a long time. |
b411b363 | 583 | */ |
c012949a | 584 | #define BM_PACKET_PAYLOAD_BYTES (4096 - sizeof(struct p_header)) |
b411b363 PR |
585 | #define BM_PACKET_WORDS (BM_PACKET_PAYLOAD_BYTES/sizeof(long)) |
586 | #define BM_PACKET_VLI_BYTES_MAX (4096 - sizeof(struct p_compressed_bm)) | |
587 | #if (PAGE_SIZE < 4096) | |
588 | /* drbd_send_bitmap / receive_bitmap would break horribly */ | |
589 | #error "PAGE_SIZE too small" | |
590 | #endif | |
591 | ||
592 | union p_polymorph { | |
c012949a | 593 | struct p_header header; |
b411b363 PR |
594 | struct p_handshake handshake; |
595 | struct p_data data; | |
596 | struct p_block_ack block_ack; | |
597 | struct p_barrier barrier; | |
598 | struct p_barrier_ack barrier_ack; | |
599 | struct p_rs_param_89 rs_param_89; | |
8e26f9cc | 600 | struct p_rs_param_95 rs_param_95; |
b411b363 PR |
601 | struct p_protocol protocol; |
602 | struct p_sizes sizes; | |
603 | struct p_uuids uuids; | |
604 | struct p_state state; | |
605 | struct p_req_state req_state; | |
606 | struct p_req_state_reply req_state_reply; | |
607 | struct p_block_req block_req; | |
02918be2 PR |
608 | struct p_delay_probe93 delay_probe93; |
609 | struct p_rs_uuid rs_uuid; | |
73a01a18 | 610 | struct p_block_desc block_desc; |
b411b363 PR |
611 | } __packed; |
612 | ||
613 | /**********************************************************************/ | |
614 | enum drbd_thread_state { | |
e77a0a5c AG |
615 | NONE, |
616 | RUNNING, | |
617 | EXITING, | |
618 | RESTARTING | |
b411b363 PR |
619 | }; |
620 | ||
621 | struct drbd_thread { | |
622 | spinlock_t t_lock; | |
623 | struct task_struct *task; | |
624 | struct completion stop; | |
625 | enum drbd_thread_state t_state; | |
626 | int (*function) (struct drbd_thread *); | |
627 | struct drbd_conf *mdev; | |
628 | int reset_cpu_mask; | |
bed879ae | 629 | char name[9]; |
b411b363 PR |
630 | }; |
631 | ||
632 | static inline enum drbd_thread_state get_t_state(struct drbd_thread *thi) | |
633 | { | |
634 | /* THINK testing the t_state seems to be uncritical in all cases | |
635 | * (but thread_{start,stop}), so we can read it *without* the lock. | |
636 | * --lge */ | |
637 | ||
638 | smp_rmb(); | |
639 | return thi->t_state; | |
640 | } | |
641 | ||
b411b363 PR |
642 | struct drbd_work; |
643 | typedef int (*drbd_work_cb)(struct drbd_conf *, struct drbd_work *, int cancel); | |
644 | struct drbd_work { | |
645 | struct list_head list; | |
646 | drbd_work_cb cb; | |
647 | }; | |
648 | ||
ace652ac AG |
649 | #include "drbd_interval.h" |
650 | ||
b411b363 PR |
651 | struct drbd_request { |
652 | struct drbd_work w; | |
653 | struct drbd_conf *mdev; | |
654 | ||
655 | /* if local IO is not allowed, will be NULL. | |
656 | * if local IO _is_ allowed, holds the locally submitted bio clone, | |
657 | * or, after local IO completion, the ERR_PTR(error). | |
658 | * see drbd_endio_pri(). */ | |
659 | struct bio *private_bio; | |
660 | ||
ace652ac | 661 | struct drbd_interval i; |
b411b363 PR |
662 | unsigned int epoch; /* barrier_nr */ |
663 | ||
664 | /* barrier_nr: used to check on "completion" whether this req was in | |
665 | * the current epoch, and we therefore have to close it, | |
666 | * starting a new epoch... | |
667 | */ | |
668 | ||
b411b363 PR |
669 | struct list_head tl_requests; /* ring list in the transfer log */ |
670 | struct bio *master_bio; /* master bio pointer */ | |
671 | unsigned long rq_state; /* see comments above _req_mod() */ | |
672 | int seq_num; | |
673 | unsigned long start_time; | |
674 | }; | |
675 | ||
676 | struct drbd_tl_epoch { | |
677 | struct drbd_work w; | |
678 | struct list_head requests; /* requests before */ | |
679 | struct drbd_tl_epoch *next; /* pointer to the next barrier */ | |
680 | unsigned int br_number; /* the barriers identifier. */ | |
7e602c0a | 681 | int n_writes; /* number of requests attached before this barrier */ |
b411b363 PR |
682 | }; |
683 | ||
b411b363 PR |
684 | struct drbd_epoch { |
685 | struct list_head list; | |
686 | unsigned int barrier_nr; | |
687 | atomic_t epoch_size; /* increased on every request added. */ | |
688 | atomic_t active; /* increased on every req. added, and dec on every finished. */ | |
689 | unsigned long flags; | |
690 | }; | |
691 | ||
692 | /* drbd_epoch flag bits */ | |
693 | enum { | |
b411b363 | 694 | DE_HAVE_BARRIER_NUMBER, |
b411b363 PR |
695 | }; |
696 | ||
697 | enum epoch_event { | |
698 | EV_PUT, | |
699 | EV_GOT_BARRIER_NR, | |
b411b363 | 700 | EV_BECAME_LAST, |
b411b363 PR |
701 | EV_CLEANUP = 32, /* used as flag */ |
702 | }; | |
703 | ||
b411b363 PR |
704 | struct drbd_wq_barrier { |
705 | struct drbd_work w; | |
706 | struct completion done; | |
707 | }; | |
708 | ||
709 | struct digest_info { | |
710 | int digest_size; | |
711 | void *digest; | |
712 | }; | |
713 | ||
f6ffca9f | 714 | struct drbd_peer_request { |
45bb912b | 715 | struct drbd_work w; |
85719573 | 716 | struct drbd_epoch *epoch; /* for writes */ |
45bb912b LE |
717 | struct drbd_conf *mdev; |
718 | struct page *pages; | |
719 | atomic_t pending_bios; | |
010f6e67 | 720 | struct drbd_interval i; |
45bb912b LE |
721 | /* see comments on ee flag bits below */ |
722 | unsigned long flags; | |
85719573 PR |
723 | union { |
724 | u64 block_id; | |
725 | struct digest_info *digest; | |
726 | }; | |
45bb912b LE |
727 | }; |
728 | ||
729 | /* ee flag bits. | |
730 | * While corresponding bios are in flight, the only modification will be | |
731 | * set_bit WAS_ERROR, which has to be atomic. | |
732 | * If no bios are in flight yet, or all have been completed, | |
733 | * non-atomic modification to ee->flags is ok. | |
734 | */ | |
b411b363 PR |
735 | enum { |
736 | __EE_CALL_AL_COMPLETE_IO, | |
b411b363 | 737 | __EE_MAY_SET_IN_SYNC, |
45bb912b | 738 | |
45bb912b LE |
739 | /* In case a barrier failed, |
740 | * we need to resubmit without the barrier flag. */ | |
741 | __EE_RESUBMITTED, | |
742 | ||
6c852bec | 743 | /* we may have several bios per peer request. |
45bb912b LE |
744 | * if any of those fail, we set this flag atomically |
745 | * from the endio callback */ | |
746 | __EE_WAS_ERROR, | |
c36c3ced LE |
747 | |
748 | /* This ee has a pointer to a digest instead of a block id */ | |
749 | __EE_HAS_DIGEST, | |
b411b363 PR |
750 | }; |
751 | #define EE_CALL_AL_COMPLETE_IO (1<<__EE_CALL_AL_COMPLETE_IO) | |
b411b363 | 752 | #define EE_MAY_SET_IN_SYNC (1<<__EE_MAY_SET_IN_SYNC) |
45bb912b LE |
753 | #define EE_RESUBMITTED (1<<__EE_RESUBMITTED) |
754 | #define EE_WAS_ERROR (1<<__EE_WAS_ERROR) | |
c36c3ced | 755 | #define EE_HAS_DIGEST (1<<__EE_HAS_DIGEST) |
b411b363 | 756 | |
01a311a5 | 757 | /* flag bits per mdev */ |
b411b363 | 758 | enum { |
25985edc | 759 | CREATE_BARRIER, /* next P_DATA is preceded by a P_BARRIER */ |
b411b363 PR |
760 | UNPLUG_QUEUED, /* only relevant with kernel 2.4 */ |
761 | UNPLUG_REMOTE, /* sending a "UnplugRemote" could help */ | |
762 | MD_DIRTY, /* current uuids and flags not yet on disk */ | |
b411b363 PR |
763 | USE_DEGR_WFC_T, /* degr-wfc-timeout instead of wfc-timeout. */ |
764 | CLUSTER_ST_CHANGE, /* Cluster wide state change going on... */ | |
765 | CL_ST_CHG_SUCCESS, | |
766 | CL_ST_CHG_FAIL, | |
767 | CRASHED_PRIMARY, /* This node was a crashed primary. | |
768 | * Gets cleared when the state.conn | |
769 | * goes into C_CONNECTED state. */ | |
19f843aa | 770 | NO_BARRIER_SUPP, /* underlying block device doesn't implement barriers */ |
b411b363 PR |
771 | CONSIDER_RESYNC, |
772 | ||
a8a4e51e | 773 | MD_NO_FUA, /* Users wants us to not use FUA/FLUSH on meta data dev */ |
b411b363 PR |
774 | SUSPEND_IO, /* suspend application io */ |
775 | BITMAP_IO, /* suspend application io; | |
776 | once no more io in flight, start bitmap io */ | |
777 | BITMAP_IO_QUEUED, /* Started bitmap IO */ | |
82f59cc6 LE |
778 | GO_DISKLESS, /* Disk is being detached, on io-error or admin request. */ |
779 | WAS_IO_ERROR, /* Local disk failed returned IO error */ | |
b411b363 | 780 | RESYNC_AFTER_NEG, /* Resync after online grow after the attach&negotiate finished. */ |
b411b363 PR |
781 | CONFIG_PENDING, /* serialization of (re)configuration requests. |
782 | * if set, also prevents the device from dying */ | |
783 | DEVICE_DYING, /* device became unconfigured, | |
784 | * but worker thread is still handling the cleanup. | |
785 | * reconfiguring (nl_disk_conf, nl_net_conf) is dissalowed, | |
786 | * while this is set. */ | |
787 | RESIZE_PENDING, /* Size change detected locally, waiting for the response from | |
788 | * the peer, if it changed there as well. */ | |
cf14c2e9 | 789 | CONN_DRY_RUN, /* Expect disconnect after resync handshake. */ |
309d1608 | 790 | GOT_PING_ACK, /* set when we receive a ping_ack packet, misc wait gets woken */ |
43a5182c | 791 | NEW_CUR_UUID, /* Create new current UUID when thawing IO */ |
0778286a | 792 | AL_SUSPENDED, /* Activity logging is currently suspended. */ |
370a43e7 | 793 | AHEAD_TO_SYNC_SOURCE, /* Ahead -> SyncSource queued */ |
e64a3294 | 794 | B_RS_H_DONE, /* Before resync handler done (already executed) */ |
b411b363 PR |
795 | }; |
796 | ||
797 | struct drbd_bitmap; /* opaque for drbd_conf */ | |
798 | ||
20ceb2b2 LE |
799 | /* definition of bits in bm_flags to be used in drbd_bm_lock |
800 | * and drbd_bitmap_io and friends. */ | |
801 | enum bm_flag { | |
802 | /* do we need to kfree, or vfree bm_pages? */ | |
803 | BM_P_VMALLOCED = 0x10000, /* internal use only, will be masked out */ | |
804 | ||
805 | /* currently locked for bulk operation */ | |
806 | BM_LOCKED_MASK = 0x7, | |
807 | ||
808 | /* in detail, that is: */ | |
809 | BM_DONT_CLEAR = 0x1, | |
810 | BM_DONT_SET = 0x2, | |
811 | BM_DONT_TEST = 0x4, | |
812 | ||
813 | /* (test bit, count bit) allowed (common case) */ | |
814 | BM_LOCKED_TEST_ALLOWED = 0x3, | |
815 | ||
816 | /* testing bits, as well as setting new bits allowed, but clearing bits | |
817 | * would be unexpected. Used during bitmap receive. Setting new bits | |
818 | * requires sending of "out-of-sync" information, though. */ | |
819 | BM_LOCKED_SET_ALLOWED = 0x1, | |
820 | ||
821 | /* clear is not expected while bitmap is locked for bulk operation */ | |
822 | }; | |
823 | ||
824 | ||
b411b363 PR |
825 | /* TODO sort members for performance |
826 | * MAYBE group them further */ | |
827 | ||
828 | /* THINK maybe we actually want to use the default "event/%s" worker threads | |
829 | * or similar in linux 2.6, which uses per cpu data and threads. | |
b411b363 PR |
830 | */ |
831 | struct drbd_work_queue { | |
832 | struct list_head q; | |
833 | struct semaphore s; /* producers up it, worker down()s it */ | |
834 | spinlock_t q_lock; /* to protect the list. */ | |
835 | }; | |
836 | ||
837 | struct drbd_socket { | |
838 | struct drbd_work_queue work; | |
839 | struct mutex mutex; | |
840 | struct socket *socket; | |
841 | /* this way we get our | |
842 | * send/receive buffers off the stack */ | |
843 | union p_polymorph sbuf; | |
844 | union p_polymorph rbuf; | |
845 | }; | |
846 | ||
847 | struct drbd_md { | |
848 | u64 md_offset; /* sector offset to 'super' block */ | |
849 | ||
850 | u64 la_size_sect; /* last agreed size, unit sectors */ | |
851 | u64 uuid[UI_SIZE]; | |
852 | u64 device_uuid; | |
853 | u32 flags; | |
854 | u32 md_size_sect; | |
855 | ||
856 | s32 al_offset; /* signed relative sector offset to al area */ | |
857 | s32 bm_offset; /* signed relative sector offset to bitmap */ | |
858 | ||
859 | /* u32 al_nr_extents; important for restoring the AL | |
860 | * is stored into sync_conf.al_extents, which in turn | |
861 | * gets applied to act_log->nr_elements | |
862 | */ | |
863 | }; | |
864 | ||
865 | /* for sync_conf and other types... */ | |
866 | #define NL_PACKET(name, number, fields) struct name { fields }; | |
867 | #define NL_INTEGER(pn,pr,member) int member; | |
868 | #define NL_INT64(pn,pr,member) __u64 member; | |
869 | #define NL_BIT(pn,pr,member) unsigned member:1; | |
870 | #define NL_STRING(pn,pr,member,len) unsigned char member[len]; int member ## _len; | |
871 | #include "linux/drbd_nl.h" | |
872 | ||
873 | struct drbd_backing_dev { | |
874 | struct block_device *backing_bdev; | |
875 | struct block_device *md_bdev; | |
b411b363 PR |
876 | struct drbd_md md; |
877 | struct disk_conf dc; /* The user provided config... */ | |
878 | sector_t known_size; /* last known size of that backing device */ | |
879 | }; | |
880 | ||
881 | struct drbd_md_io { | |
882 | struct drbd_conf *mdev; | |
883 | struct completion event; | |
884 | int error; | |
885 | }; | |
886 | ||
887 | struct bm_io_work { | |
888 | struct drbd_work w; | |
889 | char *why; | |
20ceb2b2 | 890 | enum bm_flag flags; |
b411b363 PR |
891 | int (*io_fn)(struct drbd_conf *mdev); |
892 | void (*done)(struct drbd_conf *mdev, int rv); | |
893 | }; | |
894 | ||
895 | enum write_ordering_e { | |
896 | WO_none, | |
897 | WO_drain_io, | |
898 | WO_bdev_flush, | |
b411b363 PR |
899 | }; |
900 | ||
778f271d PR |
901 | struct fifo_buffer { |
902 | int *values; | |
903 | unsigned int head_index; | |
904 | unsigned int size; | |
905 | }; | |
906 | ||
01a311a5 PR |
907 | /* flag bits per tconn */ |
908 | enum { | |
909 | NET_CONGESTED, /* The data socket is congested */ | |
25703f83 | 910 | DISCARD_CONCURRENT, /* Set on one node, cleared on the peer! */ |
e43ef195 | 911 | SEND_PING, /* whether asender should send a ping asap */ |
808e37b8 | 912 | SIGNAL_ASENDER, /* whether asender wants to be interrupted */ |
01a311a5 PR |
913 | }; |
914 | ||
2111438b PR |
915 | struct drbd_tconn { /* is a resource from the config file */ |
916 | char *name; /* Resource name */ | |
917 | struct list_head all_tconn; /* List of all drbd_tconn, prot by global_state_lock */ | |
918 | struct drbd_conf *volume0; /* TODO: Remove me again */ | |
01a311a5 | 919 | unsigned long flags; |
2111438b PR |
920 | |
921 | struct net_conf *net_conf; /* protected by get_net_conf() and put_net_conf() */ | |
b2fb6dbe PR |
922 | atomic_t net_cnt; /* Users of net_conf */ |
923 | wait_queue_head_t net_cnt_wait; | |
e42325a5 PR |
924 | |
925 | struct drbd_socket data; /* data/barrier/cstate/parameter packets */ | |
926 | struct drbd_socket meta; /* ping/ack (metadata) packets */ | |
31890f4a PR |
927 | int agreed_pro_version; /* actually used protocol version */ |
928 | unsigned long last_received; /* in jiffies, either socket */ | |
929 | unsigned int ko_count; | |
e6b3ea83 | 930 | |
87eeee41 PR |
931 | spinlock_t req_lock; |
932 | struct drbd_tl_epoch *unused_spare_tle; /* for pre-allocation */ | |
933 | struct drbd_tl_epoch *newest_tle; | |
934 | struct drbd_tl_epoch *oldest_tle; | |
935 | struct list_head out_of_sequence_requests; | |
936 | ||
a0638456 PR |
937 | struct crypto_hash *cram_hmac_tfm; |
938 | struct crypto_hash *integrity_w_tfm; /* to be used by the worker thread */ | |
939 | struct crypto_hash *integrity_r_tfm; /* to be used by the receiver thread */ | |
940 | void *int_dig_out; | |
941 | void *int_dig_in; | |
942 | void *int_dig_vv; | |
943 | ||
e6b3ea83 PR |
944 | struct drbd_thread receiver; |
945 | struct drbd_thread worker; | |
946 | struct drbd_thread asender; | |
2111438b PR |
947 | }; |
948 | ||
b411b363 | 949 | struct drbd_conf { |
2111438b PR |
950 | struct drbd_tconn *tconn; |
951 | int vnr; /* volume number within the connection */ | |
952 | ||
b411b363 PR |
953 | /* things that are stored as / read from meta data on disk */ |
954 | unsigned long flags; | |
955 | ||
956 | /* configured by drbdsetup */ | |
b411b363 PR |
957 | struct syncer_conf sync_conf; |
958 | struct drbd_backing_dev *ldev __protected_by(local); | |
959 | ||
960 | sector_t p_size; /* partner's disk size */ | |
961 | struct request_queue *rq_queue; | |
962 | struct block_device *this_bdev; | |
963 | struct gendisk *vdisk; | |
964 | ||
b411b363 PR |
965 | struct drbd_work resync_work, |
966 | unplug_work, | |
e9e6f3ec | 967 | go_diskless, |
c4752ef1 PR |
968 | md_sync_work, |
969 | start_resync_work; | |
b411b363 PR |
970 | struct timer_list resync_timer; |
971 | struct timer_list md_sync_timer; | |
370a43e7 | 972 | struct timer_list start_resync_timer; |
7fde2be9 | 973 | struct timer_list request_timer; |
ee15b038 LE |
974 | #ifdef DRBD_DEBUG_MD_SYNC |
975 | struct { | |
976 | unsigned int line; | |
977 | const char* func; | |
978 | } last_md_mark_dirty; | |
979 | #endif | |
b411b363 PR |
980 | |
981 | /* Used after attach while negotiating new disk state. */ | |
982 | union drbd_state new_state_tmp; | |
983 | ||
984 | union drbd_state state; | |
985 | wait_queue_head_t misc_wait; | |
986 | wait_queue_head_t state_wait; /* upon each state change. */ | |
987 | unsigned int send_cnt; | |
988 | unsigned int recv_cnt; | |
989 | unsigned int read_cnt; | |
990 | unsigned int writ_cnt; | |
991 | unsigned int al_writ_cnt; | |
992 | unsigned int bm_writ_cnt; | |
993 | atomic_t ap_bio_cnt; /* Requests we need to complete */ | |
994 | atomic_t ap_pending_cnt; /* AP data packets on the wire, ack expected */ | |
995 | atomic_t rs_pending_cnt; /* RS request/data packets on the wire */ | |
996 | atomic_t unacked_cnt; /* Need to send replys for */ | |
997 | atomic_t local_cnt; /* Waiting for local completion */ | |
b2fb6dbe | 998 | |
dac1389c AG |
999 | /* Interval tree of pending local requests */ |
1000 | struct rb_root read_requests; | |
de696716 AG |
1001 | struct rb_root write_requests; |
1002 | ||
4b0715f0 | 1003 | /* blocks to resync in this run [unit BM_BLOCK_SIZE] */ |
b411b363 | 1004 | unsigned long rs_total; |
4b0715f0 | 1005 | /* number of resync blocks that failed in this run */ |
b411b363 PR |
1006 | unsigned long rs_failed; |
1007 | /* Syncer's start time [unit jiffies] */ | |
1008 | unsigned long rs_start; | |
1009 | /* cumulated time in PausedSyncX state [unit jiffies] */ | |
1010 | unsigned long rs_paused; | |
1d7734a0 LE |
1011 | /* skipped because csum was equal [unit BM_BLOCK_SIZE] */ |
1012 | unsigned long rs_same_csum; | |
1013 | #define DRBD_SYNC_MARKS 8 | |
1014 | #define DRBD_SYNC_MARK_STEP (3*HZ) | |
b411b363 | 1015 | /* block not up-to-date at mark [unit BM_BLOCK_SIZE] */ |
1d7734a0 | 1016 | unsigned long rs_mark_left[DRBD_SYNC_MARKS]; |
b411b363 | 1017 | /* marks's time [unit jiffies] */ |
1d7734a0 LE |
1018 | unsigned long rs_mark_time[DRBD_SYNC_MARKS]; |
1019 | /* current index into rs_mark_{left,time} */ | |
1020 | int rs_last_mark; | |
b411b363 PR |
1021 | |
1022 | /* where does the admin want us to start? (sector) */ | |
1023 | sector_t ov_start_sector; | |
1024 | /* where are we now? (sector) */ | |
1025 | sector_t ov_position; | |
1026 | /* Start sector of out of sync range (to merge printk reporting). */ | |
1027 | sector_t ov_last_oos_start; | |
1028 | /* size of out-of-sync range in sectors. */ | |
1029 | sector_t ov_last_oos_size; | |
1030 | unsigned long ov_left; /* in bits */ | |
1031 | struct crypto_hash *csums_tfm; | |
1032 | struct crypto_hash *verify_tfm; | |
1033 | ||
b411b363 PR |
1034 | struct drbd_bitmap *bitmap; |
1035 | unsigned long bm_resync_fo; /* bit offset for drbd_bm_find_next */ | |
1036 | ||
1037 | /* Used to track operations of resync... */ | |
1038 | struct lru_cache *resync; | |
1039 | /* Number of locked elements in resync LRU */ | |
1040 | unsigned int resync_locked; | |
1041 | /* resync extent number waiting for application requests */ | |
1042 | unsigned int resync_wenr; | |
1043 | ||
1044 | int open_cnt; | |
1045 | u64 *p_uuid; | |
1046 | struct drbd_epoch *current_epoch; | |
1047 | spinlock_t epoch_lock; | |
1048 | unsigned int epochs; | |
1049 | enum write_ordering_e write_ordering; | |
85719573 PR |
1050 | struct list_head active_ee; /* IO in progress (P_DATA gets written to disk) */ |
1051 | struct list_head sync_ee; /* IO in progress (P_RS_DATA_REPLY gets written to disk) */ | |
18b75d75 AG |
1052 | struct list_head done_ee; /* need to send P_WRITE_ACK */ |
1053 | struct list_head read_ee; /* [RS]P_DATA_REQUEST being read */ | |
b411b363 | 1054 | struct list_head net_ee; /* zero-copy network send in progress */ |
b411b363 | 1055 | |
b411b363 | 1056 | int next_barrier_nr; |
b411b363 | 1057 | struct list_head resync_reads; |
435f0740 LE |
1058 | atomic_t pp_in_use; /* allocated from page pool */ |
1059 | atomic_t pp_in_use_by_net; /* sendpage()d, still referenced by tcp */ | |
b411b363 PR |
1060 | wait_queue_head_t ee_wait; |
1061 | struct page *md_io_page; /* one page buffer for md_io */ | |
1062 | struct page *md_io_tmpp; /* for logical_block_size != 512 */ | |
1063 | struct mutex md_io_mutex; /* protects the md_io_buffer */ | |
1064 | spinlock_t al_lock; | |
1065 | wait_queue_head_t al_wait; | |
1066 | struct lru_cache *act_log; /* activity log */ | |
1067 | unsigned int al_tr_number; | |
1068 | int al_tr_cycle; | |
1069 | int al_tr_pos; /* position of the next transaction in the journal */ | |
b411b363 PR |
1070 | wait_queue_head_t seq_wait; |
1071 | atomic_t packet_seq; | |
1072 | unsigned int peer_seq; | |
1073 | spinlock_t peer_seq_lock; | |
1074 | unsigned int minor; | |
1075 | unsigned long comm_bm_set; /* communicated number of set bits. */ | |
1076 | cpumask_var_t cpu_mask; | |
1077 | struct bm_io_work bm_io_work; | |
1078 | u64 ed_uuid; /* UUID of the exposed data */ | |
1079 | struct mutex state_mutex; | |
1080 | char congestion_reason; /* Why we where congested... */ | |
1d7734a0 LE |
1081 | atomic_t rs_sect_in; /* for incoming resync data rate, SyncTarget */ |
1082 | atomic_t rs_sect_ev; /* for submitted resync data rate, both */ | |
1083 | int rs_last_sect_ev; /* counter to compare with */ | |
1084 | int rs_last_events; /* counter of read or write "events" (unit sectors) | |
1085 | * on the lower level device when we last looked. */ | |
1086 | int c_sync_rate; /* current resync rate after syncer throttle magic */ | |
778f271d PR |
1087 | struct fifo_buffer rs_plan_s; /* correction values of resync planer */ |
1088 | int rs_in_flight; /* resync sectors in flight (to proxy, in proxy and from proxy) */ | |
25985edc | 1089 | int rs_planed; /* resync sectors already planned */ |
759fbdfb | 1090 | atomic_t ap_in_flight; /* App sectors in flight (waiting for ack) */ |
99432fcc PR |
1091 | int peer_max_bio_size; |
1092 | int local_max_bio_size; | |
b411b363 PR |
1093 | }; |
1094 | ||
1095 | static inline struct drbd_conf *minor_to_mdev(unsigned int minor) | |
1096 | { | |
1097 | struct drbd_conf *mdev; | |
1098 | ||
1099 | mdev = minor < minor_count ? minor_table[minor] : NULL; | |
1100 | ||
1101 | return mdev; | |
1102 | } | |
1103 | ||
1104 | static inline unsigned int mdev_to_minor(struct drbd_conf *mdev) | |
1105 | { | |
1106 | return mdev->minor; | |
1107 | } | |
1108 | ||
25985edc | 1109 | /* returns 1 if it was successful, |
b411b363 PR |
1110 | * returns 0 if there was no data socket. |
1111 | * so wherever you are going to use the data.socket, e.g. do | |
61120870 | 1112 | * if (!drbd_get_data_sock(mdev->tconn)) |
b411b363 PR |
1113 | * return 0; |
1114 | * CODE(); | |
61120870 | 1115 | * drbd_get_data_sock(mdev->tconn); |
b411b363 | 1116 | */ |
61120870 | 1117 | static inline int drbd_get_data_sock(struct drbd_tconn *tconn) |
b411b363 | 1118 | { |
61120870 | 1119 | mutex_lock(&tconn->data.mutex); |
b411b363 PR |
1120 | /* drbd_disconnect() could have called drbd_free_sock() |
1121 | * while we were waiting in down()... */ | |
61120870 PR |
1122 | if (unlikely(tconn->data.socket == NULL)) { |
1123 | mutex_unlock(&tconn->data.mutex); | |
b411b363 PR |
1124 | return 0; |
1125 | } | |
1126 | return 1; | |
1127 | } | |
1128 | ||
61120870 | 1129 | static inline void drbd_put_data_sock(struct drbd_tconn *tconn) |
b411b363 | 1130 | { |
61120870 | 1131 | mutex_unlock(&tconn->data.mutex); |
b411b363 PR |
1132 | } |
1133 | ||
1134 | /* | |
1135 | * function declarations | |
1136 | *************************/ | |
1137 | ||
1138 | /* drbd_main.c */ | |
1139 | ||
e89b591c PR |
1140 | enum dds_flags { |
1141 | DDSF_FORCED = 1, | |
1142 | DDSF_NO_RESYNC = 2, /* Do not run a resync for the new space */ | |
1143 | }; | |
1144 | ||
b411b363 | 1145 | extern void drbd_init_set_defaults(struct drbd_conf *mdev); |
b411b363 PR |
1146 | extern int drbd_thread_start(struct drbd_thread *thi); |
1147 | extern void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait); | |
bed879ae | 1148 | extern char *drbd_task_to_thread_name(struct drbd_conf *mdev, struct task_struct *task); |
b411b363 | 1149 | #ifdef CONFIG_SMP |
bc31fe33 | 1150 | extern void drbd_thread_current_set_cpu(struct drbd_conf *mdev, struct drbd_thread *thi); |
b411b363 PR |
1151 | extern void drbd_calc_cpu_mask(struct drbd_conf *mdev); |
1152 | #else | |
bc31fe33 | 1153 | #define drbd_thread_current_set_cpu(A, B) ({}) |
b411b363 PR |
1154 | #define drbd_calc_cpu_mask(A) ({}) |
1155 | #endif | |
1156 | extern void drbd_free_resources(struct drbd_conf *mdev); | |
1157 | extern void tl_release(struct drbd_conf *mdev, unsigned int barrier_nr, | |
1158 | unsigned int set_size); | |
1159 | extern void tl_clear(struct drbd_conf *mdev); | |
1160 | extern void _tl_add_barrier(struct drbd_conf *, struct drbd_tl_epoch *); | |
1161 | extern void drbd_free_sock(struct drbd_conf *mdev); | |
bedbd2a5 PR |
1162 | extern int drbd_send(struct drbd_tconn *tconn, struct socket *sock, |
1163 | void *buf, size_t size, unsigned msg_flags); | |
b411b363 PR |
1164 | extern int drbd_send_protocol(struct drbd_conf *mdev); |
1165 | extern int drbd_send_uuids(struct drbd_conf *mdev); | |
1166 | extern int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev); | |
5a22db89 | 1167 | extern int drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev); |
e89b591c | 1168 | extern int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags); |
b411b363 PR |
1169 | extern int _drbd_send_state(struct drbd_conf *mdev); |
1170 | extern int drbd_send_state(struct drbd_conf *mdev); | |
d38e787e PR |
1171 | extern int _conn_send_cmd(struct drbd_tconn *tconn, int vnr, struct socket *sock, |
1172 | enum drbd_packet cmd, struct p_header *h, size_t size, | |
1173 | unsigned msg_flags); | |
61120870 PR |
1174 | extern int conn_send_cmd2(struct drbd_tconn *tconn, enum drbd_packet cmd, |
1175 | char *data, size_t size); | |
b411b363 PR |
1176 | #define USE_DATA_SOCKET 1 |
1177 | #define USE_META_SOCKET 0 | |
1178 | extern int drbd_send_cmd(struct drbd_conf *mdev, int use_data_socket, | |
d8763023 | 1179 | enum drbd_packet cmd, struct p_header *h, size_t size); |
b411b363 PR |
1180 | extern int drbd_send_sync_param(struct drbd_conf *mdev, struct syncer_conf *sc); |
1181 | extern int drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr, | |
1182 | u32 set_size); | |
f6ffca9f AG |
1183 | extern int drbd_send_ack(struct drbd_conf *, enum drbd_packet, |
1184 | struct drbd_peer_request *); | |
d8763023 AG |
1185 | extern int drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packet cmd, |
1186 | struct p_block_req *rp); | |
1187 | extern int drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packet cmd, | |
1188 | struct p_data *dp, int data_size); | |
1189 | extern int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packet cmd, | |
b411b363 | 1190 | sector_t sector, int blksize, u64 block_id); |
73a01a18 | 1191 | extern int drbd_send_oos(struct drbd_conf *mdev, struct drbd_request *req); |
f6ffca9f AG |
1192 | extern int drbd_send_block(struct drbd_conf *, enum drbd_packet, |
1193 | struct drbd_peer_request *); | |
b411b363 | 1194 | extern int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req); |
b411b363 PR |
1195 | extern int drbd_send_drequest(struct drbd_conf *mdev, int cmd, |
1196 | sector_t sector, int size, u64 block_id); | |
d8763023 AG |
1197 | extern int drbd_send_drequest_csum(struct drbd_conf *mdev, sector_t sector, |
1198 | int size, void *digest, int digest_size, | |
1199 | enum drbd_packet cmd); | |
b411b363 PR |
1200 | extern int drbd_send_ov_request(struct drbd_conf *mdev,sector_t sector,int size); |
1201 | ||
1202 | extern int drbd_send_bitmap(struct drbd_conf *mdev); | |
1203 | extern int _drbd_send_bitmap(struct drbd_conf *mdev); | |
bf885f8a | 1204 | extern int drbd_send_sr_reply(struct drbd_conf *mdev, enum drbd_state_rv retcode); |
b411b363 PR |
1205 | extern void drbd_free_bc(struct drbd_backing_dev *ldev); |
1206 | extern void drbd_mdev_cleanup(struct drbd_conf *mdev); | |
62b0da3a | 1207 | void drbd_print_uuids(struct drbd_conf *mdev, const char *text); |
b411b363 | 1208 | |
b411b363 PR |
1209 | extern void drbd_md_sync(struct drbd_conf *mdev); |
1210 | extern int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev); | |
b411b363 PR |
1211 | extern void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local); |
1212 | extern void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local); | |
1213 | extern void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local); | |
1214 | extern void _drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local); | |
1215 | extern void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local); | |
1216 | extern void drbd_md_set_flag(struct drbd_conf *mdev, int flags) __must_hold(local); | |
1217 | extern void drbd_md_clear_flag(struct drbd_conf *mdev, int flags)__must_hold(local); | |
1218 | extern int drbd_md_test_flag(struct drbd_backing_dev *, int); | |
ee15b038 | 1219 | #ifndef DRBD_DEBUG_MD_SYNC |
b411b363 | 1220 | extern void drbd_md_mark_dirty(struct drbd_conf *mdev); |
ee15b038 LE |
1221 | #else |
1222 | #define drbd_md_mark_dirty(m) drbd_md_mark_dirty_(m, __LINE__ , __func__ ) | |
1223 | extern void drbd_md_mark_dirty_(struct drbd_conf *mdev, | |
1224 | unsigned int line, const char *func); | |
1225 | #endif | |
b411b363 PR |
1226 | extern void drbd_queue_bitmap_io(struct drbd_conf *mdev, |
1227 | int (*io_fn)(struct drbd_conf *), | |
1228 | void (*done)(struct drbd_conf *, int), | |
20ceb2b2 LE |
1229 | char *why, enum bm_flag flags); |
1230 | extern int drbd_bitmap_io(struct drbd_conf *mdev, | |
1231 | int (*io_fn)(struct drbd_conf *), | |
1232 | char *why, enum bm_flag flags); | |
b411b363 PR |
1233 | extern int drbd_bmio_set_n_write(struct drbd_conf *mdev); |
1234 | extern int drbd_bmio_clear_n_write(struct drbd_conf *mdev); | |
e9e6f3ec | 1235 | extern void drbd_go_diskless(struct drbd_conf *mdev); |
82f59cc6 | 1236 | extern void drbd_ldev_destroy(struct drbd_conf *mdev); |
b411b363 PR |
1237 | |
1238 | ||
1239 | /* Meta data layout | |
1240 | We reserve a 128MB Block (4k aligned) | |
1241 | * either at the end of the backing device | |
3ad2f3fb | 1242 | * or on a separate meta data device. */ |
b411b363 PR |
1243 | |
1244 | #define MD_RESERVED_SECT (128LU << 11) /* 128 MB, unit sectors */ | |
1245 | /* The following numbers are sectors */ | |
1246 | #define MD_AL_OFFSET 8 /* 8 Sectors after start of meta area */ | |
1247 | #define MD_AL_MAX_SIZE 64 /* = 32 kb LOG ~ 3776 extents ~ 14 GB Storage */ | |
1248 | /* Allows up to about 3.8TB */ | |
1249 | #define MD_BM_OFFSET (MD_AL_OFFSET + MD_AL_MAX_SIZE) | |
1250 | ||
1251 | /* Since the smalles IO unit is usually 512 byte */ | |
1252 | #define MD_SECTOR_SHIFT 9 | |
1253 | #define MD_SECTOR_SIZE (1<<MD_SECTOR_SHIFT) | |
1254 | ||
1255 | /* activity log */ | |
1256 | #define AL_EXTENTS_PT ((MD_SECTOR_SIZE-12)/8-1) /* 61 ; Extents per 512B sector */ | |
1257 | #define AL_EXTENT_SHIFT 22 /* One extent represents 4M Storage */ | |
1258 | #define AL_EXTENT_SIZE (1<<AL_EXTENT_SHIFT) | |
1259 | ||
1260 | #if BITS_PER_LONG == 32 | |
1261 | #define LN2_BPL 5 | |
1262 | #define cpu_to_lel(A) cpu_to_le32(A) | |
1263 | #define lel_to_cpu(A) le32_to_cpu(A) | |
1264 | #elif BITS_PER_LONG == 64 | |
1265 | #define LN2_BPL 6 | |
1266 | #define cpu_to_lel(A) cpu_to_le64(A) | |
1267 | #define lel_to_cpu(A) le64_to_cpu(A) | |
1268 | #else | |
1269 | #error "LN2 of BITS_PER_LONG unknown!" | |
1270 | #endif | |
1271 | ||
1272 | /* resync bitmap */ | |
1273 | /* 16MB sized 'bitmap extent' to track syncer usage */ | |
1274 | struct bm_extent { | |
1275 | int rs_left; /* number of bits set (out of sync) in this extent. */ | |
1276 | int rs_failed; /* number of failed resync requests in this extent. */ | |
1277 | unsigned long flags; | |
1278 | struct lc_element lce; | |
1279 | }; | |
1280 | ||
1281 | #define BME_NO_WRITES 0 /* bm_extent.flags: no more requests on this one! */ | |
1282 | #define BME_LOCKED 1 /* bm_extent.flags: syncer active on this one. */ | |
e3555d85 | 1283 | #define BME_PRIORITY 2 /* finish resync IO on this extent ASAP! App IO waiting! */ |
b411b363 PR |
1284 | |
1285 | /* drbd_bitmap.c */ | |
1286 | /* | |
1287 | * We need to store one bit for a block. | |
1288 | * Example: 1GB disk @ 4096 byte blocks ==> we need 32 KB bitmap. | |
1289 | * Bit 0 ==> local node thinks this block is binary identical on both nodes | |
1290 | * Bit 1 ==> local node thinks this block needs to be synced. | |
1291 | */ | |
1292 | ||
8e26f9cc PR |
1293 | #define SLEEP_TIME (HZ/10) |
1294 | ||
b411b363 PR |
1295 | #define BM_BLOCK_SHIFT 12 /* 4k per bit */ |
1296 | #define BM_BLOCK_SIZE (1<<BM_BLOCK_SHIFT) | |
1297 | /* (9+3) : 512 bytes @ 8 bits; representing 16M storage | |
1298 | * per sector of on disk bitmap */ | |
1299 | #define BM_EXT_SHIFT (BM_BLOCK_SHIFT + MD_SECTOR_SHIFT + 3) /* = 24 */ | |
1300 | #define BM_EXT_SIZE (1<<BM_EXT_SHIFT) | |
1301 | ||
1302 | #if (BM_EXT_SHIFT != 24) || (BM_BLOCK_SHIFT != 12) | |
1303 | #error "HAVE YOU FIXED drbdmeta AS WELL??" | |
1304 | #endif | |
1305 | ||
1306 | /* thus many _storage_ sectors are described by one bit */ | |
1307 | #define BM_SECT_TO_BIT(x) ((x)>>(BM_BLOCK_SHIFT-9)) | |
1308 | #define BM_BIT_TO_SECT(x) ((sector_t)(x)<<(BM_BLOCK_SHIFT-9)) | |
1309 | #define BM_SECT_PER_BIT BM_BIT_TO_SECT(1) | |
1310 | ||
1311 | /* bit to represented kilo byte conversion */ | |
1312 | #define Bit2KB(bits) ((bits)<<(BM_BLOCK_SHIFT-10)) | |
1313 | ||
1314 | /* in which _bitmap_ extent (resp. sector) the bit for a certain | |
1315 | * _storage_ sector is located in */ | |
1316 | #define BM_SECT_TO_EXT(x) ((x)>>(BM_EXT_SHIFT-9)) | |
1317 | ||
1318 | /* how much _storage_ sectors we have per bitmap sector */ | |
1319 | #define BM_EXT_TO_SECT(x) ((sector_t)(x) << (BM_EXT_SHIFT-9)) | |
1320 | #define BM_SECT_PER_EXT BM_EXT_TO_SECT(1) | |
1321 | ||
1322 | /* in one sector of the bitmap, we have this many activity_log extents. */ | |
1323 | #define AL_EXT_PER_BM_SECT (1 << (BM_EXT_SHIFT - AL_EXTENT_SHIFT)) | |
1324 | #define BM_WORDS_PER_AL_EXT (1 << (AL_EXTENT_SHIFT-BM_BLOCK_SHIFT-LN2_BPL)) | |
1325 | ||
1326 | #define BM_BLOCKS_PER_BM_EXT_B (BM_EXT_SHIFT - BM_BLOCK_SHIFT) | |
1327 | #define BM_BLOCKS_PER_BM_EXT_MASK ((1<<BM_BLOCKS_PER_BM_EXT_B) - 1) | |
1328 | ||
1329 | /* the extent in "PER_EXTENT" below is an activity log extent | |
1330 | * we need that many (long words/bytes) to store the bitmap | |
1331 | * of one AL_EXTENT_SIZE chunk of storage. | |
1332 | * we can store the bitmap for that many AL_EXTENTS within | |
1333 | * one sector of the _on_disk_ bitmap: | |
1334 | * bit 0 bit 37 bit 38 bit (512*8)-1 | |
1335 | * ...|........|........|.. // ..|........| | |
1336 | * sect. 0 `296 `304 ^(512*8*8)-1 | |
1337 | * | |
1338 | #define BM_WORDS_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / BITS_PER_LONG ) | |
1339 | #define BM_BYTES_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / 8 ) // 128 | |
1340 | #define BM_EXT_PER_SECT ( 512 / BM_BYTES_PER_EXTENT ) // 4 | |
1341 | */ | |
1342 | ||
1343 | #define DRBD_MAX_SECTORS_32 (0xffffffffLU) | |
1344 | #define DRBD_MAX_SECTORS_BM \ | |
1345 | ((MD_RESERVED_SECT - MD_BM_OFFSET) * (1LL<<(BM_EXT_SHIFT-9))) | |
1346 | #if DRBD_MAX_SECTORS_BM < DRBD_MAX_SECTORS_32 | |
1347 | #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_BM | |
1348 | #define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_BM | |
36bfc7e2 | 1349 | #elif !defined(CONFIG_LBDAF) && BITS_PER_LONG == 32 |
b411b363 PR |
1350 | #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_32 |
1351 | #define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_32 | |
1352 | #else | |
1353 | #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_BM | |
1354 | /* 16 TB in units of sectors */ | |
1355 | #if BITS_PER_LONG == 32 | |
1356 | /* adjust by one page worth of bitmap, | |
1357 | * so we won't wrap around in drbd_bm_find_next_bit. | |
1358 | * you should use 64bit OS for that much storage, anyways. */ | |
1359 | #define DRBD_MAX_SECTORS_FLEX BM_BIT_TO_SECT(0xffff7fff) | |
1360 | #else | |
4b0715f0 LE |
1361 | /* we allow up to 1 PiB now on 64bit architecture with "flexible" meta data */ |
1362 | #define DRBD_MAX_SECTORS_FLEX (1UL << 51) | |
1363 | /* corresponds to (1UL << 38) bits right now. */ | |
b411b363 PR |
1364 | #endif |
1365 | #endif | |
1366 | ||
d5373389 | 1367 | #define HT_SHIFT 8 |
1816a2b4 | 1368 | #define DRBD_MAX_BIO_SIZE (1U<<(9+HT_SHIFT)) |
99432fcc | 1369 | #define DRBD_MAX_BIO_SIZE_SAFE (1 << 12) /* Works always = 4k */ |
b411b363 | 1370 | |
d5373389 | 1371 | #define DRBD_MAX_SIZE_H80_PACKET (1 << 15) /* The old header only allows packets up to 32Kib data */ |
b411b363 PR |
1372 | |
1373 | extern int drbd_bm_init(struct drbd_conf *mdev); | |
02d9a94b | 1374 | extern int drbd_bm_resize(struct drbd_conf *mdev, sector_t sectors, int set_new_bits); |
b411b363 PR |
1375 | extern void drbd_bm_cleanup(struct drbd_conf *mdev); |
1376 | extern void drbd_bm_set_all(struct drbd_conf *mdev); | |
1377 | extern void drbd_bm_clear_all(struct drbd_conf *mdev); | |
4b0715f0 | 1378 | /* set/clear/test only a few bits at a time */ |
b411b363 PR |
1379 | extern int drbd_bm_set_bits( |
1380 | struct drbd_conf *mdev, unsigned long s, unsigned long e); | |
1381 | extern int drbd_bm_clear_bits( | |
1382 | struct drbd_conf *mdev, unsigned long s, unsigned long e); | |
4b0715f0 LE |
1383 | extern int drbd_bm_count_bits( |
1384 | struct drbd_conf *mdev, const unsigned long s, const unsigned long e); | |
1385 | /* bm_set_bits variant for use while holding drbd_bm_lock, | |
1386 | * may process the whole bitmap in one go */ | |
b411b363 PR |
1387 | extern void _drbd_bm_set_bits(struct drbd_conf *mdev, |
1388 | const unsigned long s, const unsigned long e); | |
1389 | extern int drbd_bm_test_bit(struct drbd_conf *mdev, unsigned long bitnr); | |
1390 | extern int drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr); | |
19f843aa | 1391 | extern int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(local); |
b411b363 PR |
1392 | extern int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local); |
1393 | extern int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local); | |
1394 | extern unsigned long drbd_bm_ALe_set_all(struct drbd_conf *mdev, | |
1395 | unsigned long al_enr); | |
1396 | extern size_t drbd_bm_words(struct drbd_conf *mdev); | |
1397 | extern unsigned long drbd_bm_bits(struct drbd_conf *mdev); | |
1398 | extern sector_t drbd_bm_capacity(struct drbd_conf *mdev); | |
4b0715f0 LE |
1399 | |
1400 | #define DRBD_END_OF_BITMAP (~(unsigned long)0) | |
b411b363 PR |
1401 | extern unsigned long drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo); |
1402 | /* bm_find_next variants for use while you hold drbd_bm_lock() */ | |
1403 | extern unsigned long _drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo); | |
1404 | extern unsigned long _drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo); | |
0778286a | 1405 | extern unsigned long _drbd_bm_total_weight(struct drbd_conf *mdev); |
b411b363 PR |
1406 | extern unsigned long drbd_bm_total_weight(struct drbd_conf *mdev); |
1407 | extern int drbd_bm_rs_done(struct drbd_conf *mdev); | |
1408 | /* for receive_bitmap */ | |
1409 | extern void drbd_bm_merge_lel(struct drbd_conf *mdev, size_t offset, | |
1410 | size_t number, unsigned long *buffer); | |
19f843aa | 1411 | /* for _drbd_send_bitmap */ |
b411b363 PR |
1412 | extern void drbd_bm_get_lel(struct drbd_conf *mdev, size_t offset, |
1413 | size_t number, unsigned long *buffer); | |
1414 | ||
20ceb2b2 | 1415 | extern void drbd_bm_lock(struct drbd_conf *mdev, char *why, enum bm_flag flags); |
b411b363 | 1416 | extern void drbd_bm_unlock(struct drbd_conf *mdev); |
b411b363 PR |
1417 | /* drbd_main.c */ |
1418 | ||
1419 | extern struct kmem_cache *drbd_request_cache; | |
6c852bec | 1420 | extern struct kmem_cache *drbd_ee_cache; /* peer requests */ |
b411b363 PR |
1421 | extern struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */ |
1422 | extern struct kmem_cache *drbd_al_ext_cache; /* activity log extents */ | |
1423 | extern mempool_t *drbd_request_mempool; | |
1424 | extern mempool_t *drbd_ee_mempool; | |
1425 | ||
1426 | extern struct page *drbd_pp_pool; /* drbd's page pool */ | |
1427 | extern spinlock_t drbd_pp_lock; | |
1428 | extern int drbd_pp_vacant; | |
1429 | extern wait_queue_head_t drbd_pp_wait; | |
1430 | ||
1431 | extern rwlock_t global_state_lock; | |
1432 | ||
1433 | extern struct drbd_conf *drbd_new_device(unsigned int minor); | |
1434 | extern void drbd_free_mdev(struct drbd_conf *mdev); | |
1435 | ||
2111438b PR |
1436 | struct drbd_tconn *drbd_new_tconn(char *name); |
1437 | extern void drbd_free_tconn(struct drbd_tconn *tconn); | |
1438 | ||
b411b363 PR |
1439 | extern int proc_details; |
1440 | ||
1441 | /* drbd_req */ | |
2f58dcfc | 1442 | extern int drbd_make_request(struct request_queue *q, struct bio *bio); |
b411b363 PR |
1443 | extern int drbd_read_remote(struct drbd_conf *mdev, struct drbd_request *req); |
1444 | extern int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct bio_vec *bvec); | |
1445 | extern int is_valid_ar_handle(struct drbd_request *, sector_t); | |
1446 | ||
1447 | ||
1448 | /* drbd_nl.c */ | |
1449 | extern void drbd_suspend_io(struct drbd_conf *mdev); | |
1450 | extern void drbd_resume_io(struct drbd_conf *mdev); | |
1451 | extern char *ppsize(char *buf, unsigned long long size); | |
a393db6f | 1452 | extern sector_t drbd_new_dev_size(struct drbd_conf *, struct drbd_backing_dev *, int); |
b411b363 | 1453 | enum determine_dev_size { dev_size_error = -1, unchanged = 0, shrunk = 1, grew = 2 }; |
24c4830c | 1454 | extern enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *, enum dds_flags) __must_hold(local); |
b411b363 | 1455 | extern void resync_after_online_grow(struct drbd_conf *); |
99432fcc | 1456 | extern void drbd_reconsider_max_bio_size(struct drbd_conf *mdev); |
bf885f8a AG |
1457 | extern enum drbd_state_rv drbd_set_role(struct drbd_conf *mdev, |
1458 | enum drbd_role new_role, | |
1459 | int force); | |
87f7be4c PR |
1460 | extern enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev); |
1461 | extern void drbd_try_outdate_peer_async(struct drbd_conf *mdev); | |
b411b363 PR |
1462 | extern int drbd_khelper(struct drbd_conf *mdev, char *cmd); |
1463 | ||
1464 | /* drbd_worker.c */ | |
1465 | extern int drbd_worker(struct drbd_thread *thi); | |
1466 | extern int drbd_alter_sa(struct drbd_conf *mdev, int na); | |
1467 | extern void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side); | |
1468 | extern void resume_next_sg(struct drbd_conf *mdev); | |
1469 | extern void suspend_other_sg(struct drbd_conf *mdev); | |
1470 | extern int drbd_resync_finished(struct drbd_conf *mdev); | |
1471 | /* maybe rather drbd_main.c ? */ | |
1472 | extern int drbd_md_sync_page_io(struct drbd_conf *mdev, | |
1473 | struct drbd_backing_dev *bdev, sector_t sector, int rw); | |
1474 | extern void drbd_ov_oos_found(struct drbd_conf*, sector_t, int); | |
9bd28d3c | 1475 | extern void drbd_rs_controller_reset(struct drbd_conf *mdev); |
b411b363 PR |
1476 | |
1477 | static inline void ov_oos_print(struct drbd_conf *mdev) | |
1478 | { | |
1479 | if (mdev->ov_last_oos_size) { | |
1480 | dev_err(DEV, "Out of sync: start=%llu, size=%lu (sectors)\n", | |
1481 | (unsigned long long)mdev->ov_last_oos_start, | |
1482 | (unsigned long)mdev->ov_last_oos_size); | |
1483 | } | |
1484 | mdev->ov_last_oos_size=0; | |
1485 | } | |
1486 | ||
1487 | ||
45bb912b | 1488 | extern void drbd_csum_bio(struct drbd_conf *, struct crypto_hash *, struct bio *, void *); |
f6ffca9f AG |
1489 | extern void drbd_csum_ee(struct drbd_conf *, struct crypto_hash *, |
1490 | struct drbd_peer_request *, void *); | |
b411b363 PR |
1491 | /* worker callbacks */ |
1492 | extern int w_req_cancel_conflict(struct drbd_conf *, struct drbd_work *, int); | |
1493 | extern int w_read_retry_remote(struct drbd_conf *, struct drbd_work *, int); | |
1494 | extern int w_e_end_data_req(struct drbd_conf *, struct drbd_work *, int); | |
1495 | extern int w_e_end_rsdata_req(struct drbd_conf *, struct drbd_work *, int); | |
1496 | extern int w_e_end_csum_rs_req(struct drbd_conf *, struct drbd_work *, int); | |
1497 | extern int w_e_end_ov_reply(struct drbd_conf *, struct drbd_work *, int); | |
1498 | extern int w_e_end_ov_req(struct drbd_conf *, struct drbd_work *, int); | |
1499 | extern int w_ov_finished(struct drbd_conf *, struct drbd_work *, int); | |
794abb75 | 1500 | extern int w_resync_timer(struct drbd_conf *, struct drbd_work *, int); |
b411b363 | 1501 | extern int w_resume_next_sg(struct drbd_conf *, struct drbd_work *, int); |
b411b363 | 1502 | extern int w_send_write_hint(struct drbd_conf *, struct drbd_work *, int); |
b411b363 PR |
1503 | extern int w_send_dblock(struct drbd_conf *, struct drbd_work *, int); |
1504 | extern int w_send_barrier(struct drbd_conf *, struct drbd_work *, int); | |
1505 | extern int w_send_read_req(struct drbd_conf *, struct drbd_work *, int); | |
1506 | extern int w_prev_work_done(struct drbd_conf *, struct drbd_work *, int); | |
1507 | extern int w_e_reissue(struct drbd_conf *, struct drbd_work *, int); | |
265be2d0 | 1508 | extern int w_restart_disk_io(struct drbd_conf *, struct drbd_work *, int); |
73a01a18 | 1509 | extern int w_send_oos(struct drbd_conf *, struct drbd_work *, int); |
c4752ef1 | 1510 | extern int w_start_resync(struct drbd_conf *, struct drbd_work *, int); |
b411b363 PR |
1511 | |
1512 | extern void resync_timer_fn(unsigned long data); | |
370a43e7 | 1513 | extern void start_resync_timer_fn(unsigned long data); |
b411b363 PR |
1514 | |
1515 | /* drbd_receiver.c */ | |
e3555d85 | 1516 | extern int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector); |
f6ffca9f AG |
1517 | extern int drbd_submit_ee(struct drbd_conf *, struct drbd_peer_request *, |
1518 | const unsigned, const int); | |
b411b363 | 1519 | extern int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list); |
f6ffca9f AG |
1520 | extern struct drbd_peer_request *drbd_alloc_ee(struct drbd_conf *, |
1521 | u64, sector_t, unsigned int, | |
1522 | gfp_t) __must_hold(local); | |
1523 | extern void drbd_free_some_ee(struct drbd_conf *, struct drbd_peer_request *, | |
1524 | int); | |
435f0740 LE |
1525 | #define drbd_free_ee(m,e) drbd_free_some_ee(m, e, 0) |
1526 | #define drbd_free_net_ee(m,e) drbd_free_some_ee(m, e, 1) | |
b411b363 PR |
1527 | extern void drbd_wait_ee_list_empty(struct drbd_conf *mdev, |
1528 | struct list_head *head); | |
1529 | extern void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, | |
1530 | struct list_head *head); | |
1531 | extern void drbd_set_recv_tcq(struct drbd_conf *mdev, int tcq_enabled); | |
1532 | extern void _drbd_clear_done_ee(struct drbd_conf *mdev, struct list_head *to_be_freed); | |
191d3cc8 | 1533 | extern void drbd_flush_workqueue(struct drbd_tconn *tconn); |
b411b363 PR |
1534 | |
1535 | /* yes, there is kernel_setsockopt, but only since 2.6.18. we don't need to | |
1536 | * mess with get_fs/set_fs, we know we are KERNEL_DS always. */ | |
1537 | static inline int drbd_setsockopt(struct socket *sock, int level, int optname, | |
1538 | char __user *optval, int optlen) | |
1539 | { | |
1540 | int err; | |
1541 | if (level == SOL_SOCKET) | |
1542 | err = sock_setsockopt(sock, level, optname, optval, optlen); | |
1543 | else | |
1544 | err = sock->ops->setsockopt(sock, level, optname, optval, | |
1545 | optlen); | |
1546 | return err; | |
1547 | } | |
1548 | ||
1549 | static inline void drbd_tcp_cork(struct socket *sock) | |
1550 | { | |
1551 | int __user val = 1; | |
1552 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK, | |
1553 | (char __user *)&val, sizeof(val)); | |
1554 | } | |
1555 | ||
1556 | static inline void drbd_tcp_uncork(struct socket *sock) | |
1557 | { | |
1558 | int __user val = 0; | |
1559 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK, | |
1560 | (char __user *)&val, sizeof(val)); | |
1561 | } | |
1562 | ||
1563 | static inline void drbd_tcp_nodelay(struct socket *sock) | |
1564 | { | |
1565 | int __user val = 1; | |
1566 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_NODELAY, | |
1567 | (char __user *)&val, sizeof(val)); | |
1568 | } | |
1569 | ||
1570 | static inline void drbd_tcp_quickack(struct socket *sock) | |
1571 | { | |
344fa462 | 1572 | int __user val = 2; |
b411b363 PR |
1573 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_QUICKACK, |
1574 | (char __user *)&val, sizeof(val)); | |
1575 | } | |
1576 | ||
1577 | void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo); | |
1578 | ||
1579 | /* drbd_proc.c */ | |
1580 | extern struct proc_dir_entry *drbd_proc; | |
7d4e9d09 | 1581 | extern const struct file_operations drbd_proc_fops; |
b411b363 PR |
1582 | extern const char *drbd_conn_str(enum drbd_conns s); |
1583 | extern const char *drbd_role_str(enum drbd_role s); | |
1584 | ||
1585 | /* drbd_actlog.c */ | |
1586 | extern void drbd_al_begin_io(struct drbd_conf *mdev, sector_t sector); | |
1587 | extern void drbd_al_complete_io(struct drbd_conf *mdev, sector_t sector); | |
1588 | extern void drbd_rs_complete_io(struct drbd_conf *mdev, sector_t sector); | |
1589 | extern int drbd_rs_begin_io(struct drbd_conf *mdev, sector_t sector); | |
1590 | extern int drbd_try_rs_begin_io(struct drbd_conf *mdev, sector_t sector); | |
1591 | extern void drbd_rs_cancel_all(struct drbd_conf *mdev); | |
1592 | extern int drbd_rs_del_all(struct drbd_conf *mdev); | |
1593 | extern void drbd_rs_failed_io(struct drbd_conf *mdev, | |
1594 | sector_t sector, int size); | |
1595 | extern int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *); | |
ea5442af | 1596 | extern void drbd_advance_rs_marks(struct drbd_conf *mdev, unsigned long still_to_go); |
b411b363 PR |
1597 | extern void __drbd_set_in_sync(struct drbd_conf *mdev, sector_t sector, |
1598 | int size, const char *file, const unsigned int line); | |
1599 | #define drbd_set_in_sync(mdev, sector, size) \ | |
1600 | __drbd_set_in_sync(mdev, sector, size, __FILE__, __LINE__) | |
73a01a18 | 1601 | extern int __drbd_set_out_of_sync(struct drbd_conf *mdev, sector_t sector, |
b411b363 PR |
1602 | int size, const char *file, const unsigned int line); |
1603 | #define drbd_set_out_of_sync(mdev, sector, size) \ | |
1604 | __drbd_set_out_of_sync(mdev, sector, size, __FILE__, __LINE__) | |
1605 | extern void drbd_al_apply_to_bm(struct drbd_conf *mdev); | |
b411b363 PR |
1606 | extern void drbd_al_shrink(struct drbd_conf *mdev); |
1607 | ||
1608 | ||
1609 | /* drbd_nl.c */ | |
1610 | ||
1611 | void drbd_nl_cleanup(void); | |
1612 | int __init drbd_nl_init(void); | |
1613 | void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state); | |
1614 | void drbd_bcast_sync_progress(struct drbd_conf *mdev); | |
f6ffca9f AG |
1615 | void drbd_bcast_ee(struct drbd_conf *, const char *, const int, const char *, |
1616 | const char *, const struct drbd_peer_request *); | |
b411b363 PR |
1617 | |
1618 | ||
b411b363 PR |
1619 | /* |
1620 | * inline helper functions | |
1621 | *************************/ | |
1622 | ||
45bb912b LE |
1623 | /* see also page_chain_add and friends in drbd_receiver.c */ |
1624 | static inline struct page *page_chain_next(struct page *page) | |
1625 | { | |
1626 | return (struct page *)page_private(page); | |
1627 | } | |
1628 | #define page_chain_for_each(page) \ | |
1629 | for (; page && ({ prefetch(page_chain_next(page)); 1; }); \ | |
1630 | page = page_chain_next(page)) | |
1631 | #define page_chain_for_each_safe(page, n) \ | |
1632 | for (; page && ({ n = page_chain_next(page); 1; }); page = n) | |
1633 | ||
1634 | static inline int drbd_bio_has_active_page(struct bio *bio) | |
1635 | { | |
1636 | struct bio_vec *bvec; | |
1637 | int i; | |
1638 | ||
1639 | __bio_for_each_segment(bvec, bio, i, 0) { | |
1640 | if (page_count(bvec->bv_page) > 1) | |
1641 | return 1; | |
1642 | } | |
1643 | ||
1644 | return 0; | |
1645 | } | |
1646 | ||
db830c46 | 1647 | static inline int drbd_ee_has_active_page(struct drbd_peer_request *peer_req) |
45bb912b | 1648 | { |
db830c46 | 1649 | struct page *page = peer_req->pages; |
45bb912b LE |
1650 | page_chain_for_each(page) { |
1651 | if (page_count(page) > 1) | |
1652 | return 1; | |
1653 | } | |
1654 | return 0; | |
1655 | } | |
1656 | ||
1657 | ||
b8907339 PR |
1658 | |
1659 | ||
1660 | ||
1661 | ||
b411b363 PR |
1662 | static inline void drbd_state_lock(struct drbd_conf *mdev) |
1663 | { | |
1664 | wait_event(mdev->misc_wait, | |
1665 | !test_and_set_bit(CLUSTER_ST_CHANGE, &mdev->flags)); | |
1666 | } | |
1667 | ||
1668 | static inline void drbd_state_unlock(struct drbd_conf *mdev) | |
1669 | { | |
1670 | clear_bit(CLUSTER_ST_CHANGE, &mdev->flags); | |
1671 | wake_up(&mdev->misc_wait); | |
1672 | } | |
1673 | ||
bf885f8a AG |
1674 | static inline enum drbd_state_rv |
1675 | _drbd_set_state(struct drbd_conf *mdev, union drbd_state ns, | |
1676 | enum chg_state_flags flags, struct completion *done) | |
b411b363 | 1677 | { |
bf885f8a | 1678 | enum drbd_state_rv rv; |
b411b363 PR |
1679 | |
1680 | read_lock(&global_state_lock); | |
1681 | rv = __drbd_set_state(mdev, ns, flags, done); | |
1682 | read_unlock(&global_state_lock); | |
1683 | ||
1684 | return rv; | |
1685 | } | |
1686 | ||
b411b363 PR |
1687 | #define __drbd_chk_io_error(m,f) __drbd_chk_io_error_(m,f, __func__) |
1688 | static inline void __drbd_chk_io_error_(struct drbd_conf *mdev, int forcedetach, const char *where) | |
1689 | { | |
1690 | switch (mdev->ldev->dc.on_io_error) { | |
1691 | case EP_PASS_ON: | |
1692 | if (!forcedetach) { | |
7383506c | 1693 | if (__ratelimit(&drbd_ratelimit_state)) |
82f59cc6 | 1694 | dev_err(DEV, "Local IO failed in %s.\n", where); |
d2e17807 PR |
1695 | if (mdev->state.disk > D_INCONSISTENT) |
1696 | _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_HARD, NULL); | |
b411b363 PR |
1697 | break; |
1698 | } | |
1699 | /* NOTE fall through to detach case if forcedetach set */ | |
1700 | case EP_DETACH: | |
1701 | case EP_CALL_HELPER: | |
82f59cc6 | 1702 | set_bit(WAS_IO_ERROR, &mdev->flags); |
b411b363 PR |
1703 | if (mdev->state.disk > D_FAILED) { |
1704 | _drbd_set_state(_NS(mdev, disk, D_FAILED), CS_HARD, NULL); | |
82f59cc6 LE |
1705 | dev_err(DEV, |
1706 | "Local IO failed in %s. Detaching...\n", where); | |
b411b363 PR |
1707 | } |
1708 | break; | |
1709 | } | |
1710 | } | |
1711 | ||
1712 | /** | |
1713 | * drbd_chk_io_error: Handle the on_io_error setting, should be called from all io completion handlers | |
1714 | * @mdev: DRBD device. | |
1715 | * @error: Error code passed to the IO completion callback | |
1716 | * @forcedetach: Force detach. I.e. the error happened while accessing the meta data | |
1717 | * | |
1718 | * See also drbd_main.c:after_state_ch() if (os.disk > D_FAILED && ns.disk == D_FAILED) | |
1719 | */ | |
1720 | #define drbd_chk_io_error(m,e,f) drbd_chk_io_error_(m,e,f, __func__) | |
1721 | static inline void drbd_chk_io_error_(struct drbd_conf *mdev, | |
1722 | int error, int forcedetach, const char *where) | |
1723 | { | |
1724 | if (error) { | |
1725 | unsigned long flags; | |
87eeee41 | 1726 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
b411b363 | 1727 | __drbd_chk_io_error_(mdev, forcedetach, where); |
87eeee41 | 1728 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
b411b363 PR |
1729 | } |
1730 | } | |
1731 | ||
1732 | ||
1733 | /** | |
1734 | * drbd_md_first_sector() - Returns the first sector number of the meta data area | |
1735 | * @bdev: Meta data block device. | |
1736 | * | |
1737 | * BTW, for internal meta data, this happens to be the maximum capacity | |
1738 | * we could agree upon with our peer node. | |
1739 | */ | |
1740 | static inline sector_t drbd_md_first_sector(struct drbd_backing_dev *bdev) | |
1741 | { | |
1742 | switch (bdev->dc.meta_dev_idx) { | |
1743 | case DRBD_MD_INDEX_INTERNAL: | |
1744 | case DRBD_MD_INDEX_FLEX_INT: | |
1745 | return bdev->md.md_offset + bdev->md.bm_offset; | |
1746 | case DRBD_MD_INDEX_FLEX_EXT: | |
1747 | default: | |
1748 | return bdev->md.md_offset; | |
1749 | } | |
1750 | } | |
1751 | ||
1752 | /** | |
1753 | * drbd_md_last_sector() - Return the last sector number of the meta data area | |
1754 | * @bdev: Meta data block device. | |
1755 | */ | |
1756 | static inline sector_t drbd_md_last_sector(struct drbd_backing_dev *bdev) | |
1757 | { | |
1758 | switch (bdev->dc.meta_dev_idx) { | |
1759 | case DRBD_MD_INDEX_INTERNAL: | |
1760 | case DRBD_MD_INDEX_FLEX_INT: | |
1761 | return bdev->md.md_offset + MD_AL_OFFSET - 1; | |
1762 | case DRBD_MD_INDEX_FLEX_EXT: | |
1763 | default: | |
1764 | return bdev->md.md_offset + bdev->md.md_size_sect; | |
1765 | } | |
1766 | } | |
1767 | ||
1768 | /* Returns the number of 512 byte sectors of the device */ | |
1769 | static inline sector_t drbd_get_capacity(struct block_device *bdev) | |
1770 | { | |
1771 | /* return bdev ? get_capacity(bdev->bd_disk) : 0; */ | |
77304d2a | 1772 | return bdev ? i_size_read(bdev->bd_inode) >> 9 : 0; |
b411b363 PR |
1773 | } |
1774 | ||
1775 | /** | |
1776 | * drbd_get_max_capacity() - Returns the capacity we announce to out peer | |
1777 | * @bdev: Meta data block device. | |
1778 | * | |
1779 | * returns the capacity we announce to out peer. we clip ourselves at the | |
1780 | * various MAX_SECTORS, because if we don't, current implementation will | |
1781 | * oops sooner or later | |
1782 | */ | |
1783 | static inline sector_t drbd_get_max_capacity(struct drbd_backing_dev *bdev) | |
1784 | { | |
1785 | sector_t s; | |
1786 | switch (bdev->dc.meta_dev_idx) { | |
1787 | case DRBD_MD_INDEX_INTERNAL: | |
1788 | case DRBD_MD_INDEX_FLEX_INT: | |
1789 | s = drbd_get_capacity(bdev->backing_bdev) | |
1790 | ? min_t(sector_t, DRBD_MAX_SECTORS_FLEX, | |
1791 | drbd_md_first_sector(bdev)) | |
1792 | : 0; | |
1793 | break; | |
1794 | case DRBD_MD_INDEX_FLEX_EXT: | |
1795 | s = min_t(sector_t, DRBD_MAX_SECTORS_FLEX, | |
1796 | drbd_get_capacity(bdev->backing_bdev)); | |
1797 | /* clip at maximum size the meta device can support */ | |
1798 | s = min_t(sector_t, s, | |
1799 | BM_EXT_TO_SECT(bdev->md.md_size_sect | |
1800 | - bdev->md.bm_offset)); | |
1801 | break; | |
1802 | default: | |
1803 | s = min_t(sector_t, DRBD_MAX_SECTORS, | |
1804 | drbd_get_capacity(bdev->backing_bdev)); | |
1805 | } | |
1806 | return s; | |
1807 | } | |
1808 | ||
1809 | /** | |
1810 | * drbd_md_ss__() - Return the sector number of our meta data super block | |
1811 | * @mdev: DRBD device. | |
1812 | * @bdev: Meta data block device. | |
1813 | */ | |
1814 | static inline sector_t drbd_md_ss__(struct drbd_conf *mdev, | |
1815 | struct drbd_backing_dev *bdev) | |
1816 | { | |
1817 | switch (bdev->dc.meta_dev_idx) { | |
1818 | default: /* external, some index */ | |
1819 | return MD_RESERVED_SECT * bdev->dc.meta_dev_idx; | |
1820 | case DRBD_MD_INDEX_INTERNAL: | |
1821 | /* with drbd08, internal meta data is always "flexible" */ | |
1822 | case DRBD_MD_INDEX_FLEX_INT: | |
1823 | /* sizeof(struct md_on_disk_07) == 4k | |
1824 | * position: last 4k aligned block of 4k size */ | |
1825 | if (!bdev->backing_bdev) { | |
1826 | if (__ratelimit(&drbd_ratelimit_state)) { | |
1827 | dev_err(DEV, "bdev->backing_bdev==NULL\n"); | |
1828 | dump_stack(); | |
1829 | } | |
1830 | return 0; | |
1831 | } | |
1832 | return (drbd_get_capacity(bdev->backing_bdev) & ~7ULL) | |
1833 | - MD_AL_OFFSET; | |
1834 | case DRBD_MD_INDEX_FLEX_EXT: | |
1835 | return 0; | |
1836 | } | |
1837 | } | |
1838 | ||
b411b363 PR |
1839 | static inline void |
1840 | drbd_queue_work_front(struct drbd_work_queue *q, struct drbd_work *w) | |
1841 | { | |
1842 | unsigned long flags; | |
1843 | spin_lock_irqsave(&q->q_lock, flags); | |
1844 | list_add(&w->list, &q->q); | |
1845 | up(&q->s); /* within the spinlock, | |
1846 | see comment near end of drbd_worker() */ | |
1847 | spin_unlock_irqrestore(&q->q_lock, flags); | |
1848 | } | |
1849 | ||
1850 | static inline void | |
1851 | drbd_queue_work(struct drbd_work_queue *q, struct drbd_work *w) | |
1852 | { | |
1853 | unsigned long flags; | |
1854 | spin_lock_irqsave(&q->q_lock, flags); | |
1855 | list_add_tail(&w->list, &q->q); | |
1856 | up(&q->s); /* within the spinlock, | |
1857 | see comment near end of drbd_worker() */ | |
1858 | spin_unlock_irqrestore(&q->q_lock, flags); | |
1859 | } | |
1860 | ||
0625ac19 | 1861 | static inline void wake_asender(struct drbd_tconn *tconn) |
b411b363 | 1862 | { |
0625ac19 PR |
1863 | if (test_bit(SIGNAL_ASENDER, &tconn->flags)) |
1864 | force_sig(DRBD_SIG, tconn->asender.task); | |
b411b363 PR |
1865 | } |
1866 | ||
0625ac19 | 1867 | static inline void request_ping(struct drbd_tconn *tconn) |
b411b363 | 1868 | { |
0625ac19 PR |
1869 | set_bit(SEND_PING, &tconn->flags); |
1870 | wake_asender(tconn); | |
b411b363 PR |
1871 | } |
1872 | ||
d38e787e PR |
1873 | static inline int _drbd_send_cmd(struct drbd_conf *mdev, struct socket *sock, |
1874 | enum drbd_packet cmd, struct p_header *h, size_t size, | |
1875 | unsigned msg_flags) | |
1876 | { | |
1877 | return _conn_send_cmd(mdev->tconn, mdev->vnr, sock, cmd, h, size, msg_flags); | |
1878 | } | |
1879 | ||
b411b363 | 1880 | static inline int drbd_send_short_cmd(struct drbd_conf *mdev, |
d8763023 | 1881 | enum drbd_packet cmd) |
b411b363 | 1882 | { |
c012949a | 1883 | struct p_header h; |
b411b363 PR |
1884 | return drbd_send_cmd(mdev, USE_DATA_SOCKET, cmd, &h, sizeof(h)); |
1885 | } | |
1886 | ||
1887 | static inline int drbd_send_ping(struct drbd_conf *mdev) | |
1888 | { | |
c012949a | 1889 | struct p_header h; |
b411b363 PR |
1890 | return drbd_send_cmd(mdev, USE_META_SOCKET, P_PING, &h, sizeof(h)); |
1891 | } | |
1892 | ||
1893 | static inline int drbd_send_ping_ack(struct drbd_conf *mdev) | |
1894 | { | |
c012949a | 1895 | struct p_header h; |
b411b363 PR |
1896 | return drbd_send_cmd(mdev, USE_META_SOCKET, P_PING_ACK, &h, sizeof(h)); |
1897 | } | |
1898 | ||
1899 | static inline void drbd_thread_stop(struct drbd_thread *thi) | |
1900 | { | |
81e84650 | 1901 | _drbd_thread_stop(thi, false, true); |
b411b363 PR |
1902 | } |
1903 | ||
1904 | static inline void drbd_thread_stop_nowait(struct drbd_thread *thi) | |
1905 | { | |
81e84650 | 1906 | _drbd_thread_stop(thi, false, false); |
b411b363 PR |
1907 | } |
1908 | ||
1909 | static inline void drbd_thread_restart_nowait(struct drbd_thread *thi) | |
1910 | { | |
81e84650 | 1911 | _drbd_thread_stop(thi, true, false); |
b411b363 PR |
1912 | } |
1913 | ||
1914 | /* counts how many answer packets packets we expect from our peer, | |
1915 | * for either explicit application requests, | |
1916 | * or implicit barrier packets as necessary. | |
1917 | * increased: | |
1918 | * w_send_barrier | |
8554df1c | 1919 | * _req_mod(req, QUEUE_FOR_NET_WRITE or QUEUE_FOR_NET_READ); |
b411b363 PR |
1920 | * it is much easier and equally valid to count what we queue for the |
1921 | * worker, even before it actually was queued or send. | |
1922 | * (drbd_make_request_common; recovery path on read io-error) | |
1923 | * decreased: | |
1924 | * got_BarrierAck (respective tl_clear, tl_clear_barrier) | |
8554df1c | 1925 | * _req_mod(req, DATA_RECEIVED) |
b411b363 | 1926 | * [from receive_DataReply] |
8554df1c | 1927 | * _req_mod(req, WRITE_ACKED_BY_PEER or RECV_ACKED_BY_PEER or NEG_ACKED) |
b411b363 PR |
1928 | * [from got_BlockAck (P_WRITE_ACK, P_RECV_ACK)] |
1929 | * for some reason it is NOT decreased in got_NegAck, | |
1930 | * but in the resulting cleanup code from report_params. | |
1931 | * we should try to remember the reason for that... | |
8554df1c AG |
1932 | * _req_mod(req, SEND_FAILED or SEND_CANCELED) |
1933 | * _req_mod(req, CONNECTION_LOST_WHILE_PENDING) | |
b411b363 PR |
1934 | * [from tl_clear_barrier] |
1935 | */ | |
1936 | static inline void inc_ap_pending(struct drbd_conf *mdev) | |
1937 | { | |
1938 | atomic_inc(&mdev->ap_pending_cnt); | |
1939 | } | |
1940 | ||
1941 | #define ERR_IF_CNT_IS_NEGATIVE(which) \ | |
1942 | if (atomic_read(&mdev->which) < 0) \ | |
1943 | dev_err(DEV, "in %s:%d: " #which " = %d < 0 !\n", \ | |
1944 | __func__ , __LINE__ , \ | |
1945 | atomic_read(&mdev->which)) | |
1946 | ||
1947 | #define dec_ap_pending(mdev) do { \ | |
1948 | typecheck(struct drbd_conf *, mdev); \ | |
1949 | if (atomic_dec_and_test(&mdev->ap_pending_cnt)) \ | |
1950 | wake_up(&mdev->misc_wait); \ | |
1951 | ERR_IF_CNT_IS_NEGATIVE(ap_pending_cnt); } while (0) | |
1952 | ||
1953 | /* counts how many resync-related answers we still expect from the peer | |
1954 | * increase decrease | |
1955 | * C_SYNC_TARGET sends P_RS_DATA_REQUEST (and expects P_RS_DATA_REPLY) | |
25985edc | 1956 | * C_SYNC_SOURCE sends P_RS_DATA_REPLY (and expects P_WRITE_ACK with ID_SYNCER) |
b411b363 PR |
1957 | * (or P_NEG_ACK with ID_SYNCER) |
1958 | */ | |
1959 | static inline void inc_rs_pending(struct drbd_conf *mdev) | |
1960 | { | |
1961 | atomic_inc(&mdev->rs_pending_cnt); | |
1962 | } | |
1963 | ||
1964 | #define dec_rs_pending(mdev) do { \ | |
1965 | typecheck(struct drbd_conf *, mdev); \ | |
1966 | atomic_dec(&mdev->rs_pending_cnt); \ | |
1967 | ERR_IF_CNT_IS_NEGATIVE(rs_pending_cnt); } while (0) | |
1968 | ||
1969 | /* counts how many answers we still need to send to the peer. | |
1970 | * increased on | |
1971 | * receive_Data unless protocol A; | |
1972 | * we need to send a P_RECV_ACK (proto B) | |
1973 | * or P_WRITE_ACK (proto C) | |
1974 | * receive_RSDataReply (recv_resync_read) we need to send a P_WRITE_ACK | |
1975 | * receive_DataRequest (receive_RSDataRequest) we need to send back P_DATA | |
1976 | * receive_Barrier_* we need to send a P_BARRIER_ACK | |
1977 | */ | |
1978 | static inline void inc_unacked(struct drbd_conf *mdev) | |
1979 | { | |
1980 | atomic_inc(&mdev->unacked_cnt); | |
1981 | } | |
1982 | ||
1983 | #define dec_unacked(mdev) do { \ | |
1984 | typecheck(struct drbd_conf *, mdev); \ | |
1985 | atomic_dec(&mdev->unacked_cnt); \ | |
1986 | ERR_IF_CNT_IS_NEGATIVE(unacked_cnt); } while (0) | |
1987 | ||
1988 | #define sub_unacked(mdev, n) do { \ | |
1989 | typecheck(struct drbd_conf *, mdev); \ | |
1990 | atomic_sub(n, &mdev->unacked_cnt); \ | |
1991 | ERR_IF_CNT_IS_NEGATIVE(unacked_cnt); } while (0) | |
1992 | ||
1993 | ||
b2fb6dbe | 1994 | static inline void put_net_conf(struct drbd_tconn *tconn) |
b411b363 | 1995 | { |
b2fb6dbe PR |
1996 | if (atomic_dec_and_test(&tconn->net_cnt)) |
1997 | wake_up(&tconn->net_cnt_wait); | |
b411b363 PR |
1998 | } |
1999 | ||
2000 | /** | |
89e58e75 | 2001 | * get_net_conf() - Increase ref count on mdev->tconn->net_conf; Returns 0 if nothing there |
b411b363 PR |
2002 | * @mdev: DRBD device. |
2003 | * | |
89e58e75 | 2004 | * You have to call put_net_conf() when finished working with mdev->tconn->net_conf. |
b411b363 | 2005 | */ |
b2fb6dbe | 2006 | static inline int get_net_conf(struct drbd_tconn *tconn) |
b411b363 PR |
2007 | { |
2008 | int have_net_conf; | |
2009 | ||
b2fb6dbe PR |
2010 | atomic_inc(&tconn->net_cnt); |
2011 | have_net_conf = tconn->volume0->state.conn >= C_UNCONNECTED; | |
b411b363 | 2012 | if (!have_net_conf) |
b2fb6dbe | 2013 | put_net_conf(tconn); |
b411b363 PR |
2014 | return have_net_conf; |
2015 | } | |
2016 | ||
2017 | /** | |
2018 | * get_ldev() - Increase the ref count on mdev->ldev. Returns 0 if there is no ldev | |
2019 | * @M: DRBD device. | |
2020 | * | |
2021 | * You have to call put_ldev() when finished working with mdev->ldev. | |
2022 | */ | |
2023 | #define get_ldev(M) __cond_lock(local, _get_ldev_if_state(M,D_INCONSISTENT)) | |
2024 | #define get_ldev_if_state(M,MINS) __cond_lock(local, _get_ldev_if_state(M,MINS)) | |
2025 | ||
2026 | static inline void put_ldev(struct drbd_conf *mdev) | |
2027 | { | |
1d7734a0 | 2028 | int i = atomic_dec_return(&mdev->local_cnt); |
9a0d9d03 LE |
2029 | |
2030 | /* This may be called from some endio handler, | |
2031 | * so we must not sleep here. */ | |
2032 | ||
b411b363 | 2033 | __release(local); |
1d7734a0 | 2034 | D_ASSERT(i >= 0); |
e9e6f3ec | 2035 | if (i == 0) { |
82f59cc6 LE |
2036 | if (mdev->state.disk == D_DISKLESS) |
2037 | /* even internal references gone, safe to destroy */ | |
2038 | drbd_ldev_destroy(mdev); | |
e9e6f3ec | 2039 | if (mdev->state.disk == D_FAILED) |
82f59cc6 | 2040 | /* all application IO references gone. */ |
e9e6f3ec | 2041 | drbd_go_diskless(mdev); |
b411b363 | 2042 | wake_up(&mdev->misc_wait); |
e9e6f3ec | 2043 | } |
b411b363 PR |
2044 | } |
2045 | ||
2046 | #ifndef __CHECKER__ | |
2047 | static inline int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins) | |
2048 | { | |
2049 | int io_allowed; | |
2050 | ||
82f59cc6 LE |
2051 | /* never get a reference while D_DISKLESS */ |
2052 | if (mdev->state.disk == D_DISKLESS) | |
2053 | return 0; | |
2054 | ||
b411b363 PR |
2055 | atomic_inc(&mdev->local_cnt); |
2056 | io_allowed = (mdev->state.disk >= mins); | |
2057 | if (!io_allowed) | |
2058 | put_ldev(mdev); | |
2059 | return io_allowed; | |
2060 | } | |
2061 | #else | |
2062 | extern int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins); | |
2063 | #endif | |
2064 | ||
2065 | /* you must have an "get_ldev" reference */ | |
2066 | static inline void drbd_get_syncer_progress(struct drbd_conf *mdev, | |
2067 | unsigned long *bits_left, unsigned int *per_mil_done) | |
2068 | { | |
4b0715f0 LE |
2069 | /* this is to break it at compile time when we change that, in case we |
2070 | * want to support more than (1<<32) bits on a 32bit arch. */ | |
b411b363 PR |
2071 | typecheck(unsigned long, mdev->rs_total); |
2072 | ||
2073 | /* note: both rs_total and rs_left are in bits, i.e. in | |
2074 | * units of BM_BLOCK_SIZE. | |
2075 | * for the percentage, we don't care. */ | |
2076 | ||
439d5953 LE |
2077 | if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T) |
2078 | *bits_left = mdev->ov_left; | |
2079 | else | |
2080 | *bits_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; | |
b411b363 PR |
2081 | /* >> 10 to prevent overflow, |
2082 | * +1 to prevent division by zero */ | |
2083 | if (*bits_left > mdev->rs_total) { | |
2084 | /* doh. maybe a logic bug somewhere. | |
2085 | * may also be just a race condition | |
2086 | * between this and a disconnect during sync. | |
2087 | * for now, just prevent in-kernel buffer overflow. | |
2088 | */ | |
2089 | smp_rmb(); | |
2090 | dev_warn(DEV, "cs:%s rs_left=%lu > rs_total=%lu (rs_failed %lu)\n", | |
2091 | drbd_conn_str(mdev->state.conn), | |
2092 | *bits_left, mdev->rs_total, mdev->rs_failed); | |
2093 | *per_mil_done = 0; | |
2094 | } else { | |
4b0715f0 LE |
2095 | /* Make sure the division happens in long context. |
2096 | * We allow up to one petabyte storage right now, | |
2097 | * at a granularity of 4k per bit that is 2**38 bits. | |
2098 | * After shift right and multiplication by 1000, | |
2099 | * this should still fit easily into a 32bit long, | |
2100 | * so we don't need a 64bit division on 32bit arch. | |
2101 | * Note: currently we don't support such large bitmaps on 32bit | |
2102 | * arch anyways, but no harm done to be prepared for it here. | |
2103 | */ | |
2104 | unsigned int shift = mdev->rs_total >= (1ULL << 32) ? 16 : 10; | |
2105 | unsigned long left = *bits_left >> shift; | |
2106 | unsigned long total = 1UL + (mdev->rs_total >> shift); | |
2107 | unsigned long tmp = 1000UL - left * 1000UL/total; | |
b411b363 PR |
2108 | *per_mil_done = tmp; |
2109 | } | |
2110 | } | |
2111 | ||
2112 | ||
2113 | /* this throttles on-the-fly application requests | |
2114 | * according to max_buffers settings; | |
2115 | * maybe re-implement using semaphores? */ | |
2116 | static inline int drbd_get_max_buffers(struct drbd_conf *mdev) | |
2117 | { | |
2118 | int mxb = 1000000; /* arbitrary limit on open requests */ | |
b2fb6dbe | 2119 | if (get_net_conf(mdev->tconn)) { |
89e58e75 | 2120 | mxb = mdev->tconn->net_conf->max_buffers; |
b2fb6dbe | 2121 | put_net_conf(mdev->tconn); |
b411b363 PR |
2122 | } |
2123 | return mxb; | |
2124 | } | |
2125 | ||
3719094e | 2126 | static inline int drbd_state_is_stable(struct drbd_conf *mdev) |
b411b363 | 2127 | { |
3719094e | 2128 | union drbd_state s = mdev->state; |
b411b363 PR |
2129 | |
2130 | /* DO NOT add a default clause, we want the compiler to warn us | |
2131 | * for any newly introduced state we may have forgotten to add here */ | |
2132 | ||
2133 | switch ((enum drbd_conns)s.conn) { | |
2134 | /* new io only accepted when there is no connection, ... */ | |
2135 | case C_STANDALONE: | |
2136 | case C_WF_CONNECTION: | |
2137 | /* ... or there is a well established connection. */ | |
2138 | case C_CONNECTED: | |
2139 | case C_SYNC_SOURCE: | |
2140 | case C_SYNC_TARGET: | |
2141 | case C_VERIFY_S: | |
2142 | case C_VERIFY_T: | |
2143 | case C_PAUSED_SYNC_S: | |
2144 | case C_PAUSED_SYNC_T: | |
67531718 PR |
2145 | case C_AHEAD: |
2146 | case C_BEHIND: | |
3719094e | 2147 | /* transitional states, IO allowed */ |
b411b363 PR |
2148 | case C_DISCONNECTING: |
2149 | case C_UNCONNECTED: | |
2150 | case C_TIMEOUT: | |
2151 | case C_BROKEN_PIPE: | |
2152 | case C_NETWORK_FAILURE: | |
2153 | case C_PROTOCOL_ERROR: | |
2154 | case C_TEAR_DOWN: | |
2155 | case C_WF_REPORT_PARAMS: | |
2156 | case C_STARTING_SYNC_S: | |
2157 | case C_STARTING_SYNC_T: | |
3719094e PR |
2158 | break; |
2159 | ||
2160 | /* Allow IO in BM exchange states with new protocols */ | |
b411b363 | 2161 | case C_WF_BITMAP_S: |
31890f4a | 2162 | if (mdev->tconn->agreed_pro_version < 96) |
3719094e PR |
2163 | return 0; |
2164 | break; | |
2165 | ||
2166 | /* no new io accepted in these states */ | |
b411b363 PR |
2167 | case C_WF_BITMAP_T: |
2168 | case C_WF_SYNC_UUID: | |
2169 | case C_MASK: | |
2170 | /* not "stable" */ | |
2171 | return 0; | |
2172 | } | |
2173 | ||
2174 | switch ((enum drbd_disk_state)s.disk) { | |
2175 | case D_DISKLESS: | |
2176 | case D_INCONSISTENT: | |
2177 | case D_OUTDATED: | |
2178 | case D_CONSISTENT: | |
2179 | case D_UP_TO_DATE: | |
2180 | /* disk state is stable as well. */ | |
2181 | break; | |
2182 | ||
2183 | /* no new io accepted during tansitional states */ | |
2184 | case D_ATTACHING: | |
2185 | case D_FAILED: | |
2186 | case D_NEGOTIATING: | |
2187 | case D_UNKNOWN: | |
2188 | case D_MASK: | |
2189 | /* not "stable" */ | |
2190 | return 0; | |
2191 | } | |
2192 | ||
2193 | return 1; | |
2194 | } | |
2195 | ||
fb22c402 PR |
2196 | static inline int is_susp(union drbd_state s) |
2197 | { | |
2198 | return s.susp || s.susp_nod || s.susp_fen; | |
2199 | } | |
2200 | ||
1b881ef7 | 2201 | static inline bool may_inc_ap_bio(struct drbd_conf *mdev) |
b411b363 PR |
2202 | { |
2203 | int mxb = drbd_get_max_buffers(mdev); | |
2204 | ||
fb22c402 | 2205 | if (is_susp(mdev->state)) |
1b881ef7 | 2206 | return false; |
b411b363 | 2207 | if (test_bit(SUSPEND_IO, &mdev->flags)) |
1b881ef7 | 2208 | return false; |
b411b363 PR |
2209 | |
2210 | /* to avoid potential deadlock or bitmap corruption, | |
2211 | * in various places, we only allow new application io | |
2212 | * to start during "stable" states. */ | |
2213 | ||
2214 | /* no new io accepted when attaching or detaching the disk */ | |
3719094e | 2215 | if (!drbd_state_is_stable(mdev)) |
1b881ef7 | 2216 | return false; |
b411b363 PR |
2217 | |
2218 | /* since some older kernels don't have atomic_add_unless, | |
2219 | * and we are within the spinlock anyways, we have this workaround. */ | |
2220 | if (atomic_read(&mdev->ap_bio_cnt) > mxb) | |
1b881ef7 | 2221 | return false; |
b411b363 | 2222 | if (test_bit(BITMAP_IO, &mdev->flags)) |
1b881ef7 AG |
2223 | return false; |
2224 | return true; | |
b411b363 PR |
2225 | } |
2226 | ||
1b881ef7 | 2227 | static inline bool inc_ap_bio_cond(struct drbd_conf *mdev, int count) |
b411b363 | 2228 | { |
1b881ef7 | 2229 | bool rv = false; |
8869d683 | 2230 | |
87eeee41 | 2231 | spin_lock_irq(&mdev->tconn->req_lock); |
1b881ef7 | 2232 | rv = may_inc_ap_bio(mdev); |
8869d683 PR |
2233 | if (rv) |
2234 | atomic_add(count, &mdev->ap_bio_cnt); | |
87eeee41 | 2235 | spin_unlock_irq(&mdev->tconn->req_lock); |
8869d683 PR |
2236 | |
2237 | return rv; | |
2238 | } | |
b411b363 | 2239 | |
8869d683 PR |
2240 | static inline void inc_ap_bio(struct drbd_conf *mdev, int count) |
2241 | { | |
b411b363 PR |
2242 | /* we wait here |
2243 | * as long as the device is suspended | |
2244 | * until the bitmap is no longer on the fly during connection | |
2245 | * handshake as long as we would exeed the max_buffer limit. | |
2246 | * | |
2247 | * to avoid races with the reconnect code, | |
2248 | * we need to atomic_inc within the spinlock. */ | |
2249 | ||
1b881ef7 | 2250 | wait_event(mdev->misc_wait, inc_ap_bio_cond(mdev, count)); |
b411b363 PR |
2251 | } |
2252 | ||
2253 | static inline void dec_ap_bio(struct drbd_conf *mdev) | |
2254 | { | |
2255 | int mxb = drbd_get_max_buffers(mdev); | |
2256 | int ap_bio = atomic_dec_return(&mdev->ap_bio_cnt); | |
2257 | ||
2258 | D_ASSERT(ap_bio >= 0); | |
2259 | /* this currently does wake_up for every dec_ap_bio! | |
2260 | * maybe rather introduce some type of hysteresis? | |
2261 | * e.g. (ap_bio == mxb/2 || ap_bio == 0) ? */ | |
2262 | if (ap_bio < mxb) | |
2263 | wake_up(&mdev->misc_wait); | |
2264 | if (ap_bio == 0 && test_bit(BITMAP_IO, &mdev->flags)) { | |
2265 | if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags)) | |
e42325a5 | 2266 | drbd_queue_work(&mdev->tconn->data.work, &mdev->bm_io_work.w); |
b411b363 PR |
2267 | } |
2268 | } | |
2269 | ||
62b0da3a | 2270 | static inline int drbd_set_ed_uuid(struct drbd_conf *mdev, u64 val) |
b411b363 | 2271 | { |
62b0da3a | 2272 | int changed = mdev->ed_uuid != val; |
b411b363 | 2273 | mdev->ed_uuid = val; |
62b0da3a | 2274 | return changed; |
b411b363 PR |
2275 | } |
2276 | ||
b411b363 PR |
2277 | static inline int drbd_queue_order_type(struct drbd_conf *mdev) |
2278 | { | |
2279 | /* sorry, we currently have no working implementation | |
2280 | * of distributed TCQ stuff */ | |
2281 | #ifndef QUEUE_ORDERED_NONE | |
2282 | #define QUEUE_ORDERED_NONE 0 | |
2283 | #endif | |
2284 | return QUEUE_ORDERED_NONE; | |
2285 | } | |
2286 | ||
b411b363 PR |
2287 | static inline void drbd_md_flush(struct drbd_conf *mdev) |
2288 | { | |
2289 | int r; | |
2290 | ||
a8a4e51e | 2291 | if (test_bit(MD_NO_FUA, &mdev->flags)) |
b411b363 PR |
2292 | return; |
2293 | ||
dd3932ed | 2294 | r = blkdev_issue_flush(mdev->ldev->md_bdev, GFP_KERNEL, NULL); |
b411b363 | 2295 | if (r) { |
a8a4e51e | 2296 | set_bit(MD_NO_FUA, &mdev->flags); |
b411b363 PR |
2297 | dev_err(DEV, "meta data flush failed with status %d, disabling md-flushes\n", r); |
2298 | } | |
2299 | } | |
2300 | ||
2301 | #endif |