Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_int.h | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | drbd is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
14 | ||
15 | drbd is distributed in the hope that it will be useful, | |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | GNU General Public License for more details. | |
19 | ||
20 | You should have received a copy of the GNU General Public License | |
21 | along with drbd; see the file COPYING. If not, write to | |
22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | ||
24 | */ | |
25 | ||
26 | #ifndef _DRBD_INT_H | |
27 | #define _DRBD_INT_H | |
28 | ||
29 | #include <linux/compiler.h> | |
30 | #include <linux/types.h> | |
31 | #include <linux/version.h> | |
32 | #include <linux/list.h> | |
33 | #include <linux/sched.h> | |
34 | #include <linux/bitops.h> | |
35 | #include <linux/slab.h> | |
36 | #include <linux/crypto.h> | |
132cc538 | 37 | #include <linux/ratelimit.h> |
b411b363 PR |
38 | #include <linux/tcp.h> |
39 | #include <linux/mutex.h> | |
40 | #include <linux/major.h> | |
41 | #include <linux/blkdev.h> | |
42 | #include <linux/genhd.h> | |
062e879c | 43 | #include <linux/idr.h> |
b411b363 PR |
44 | #include <net/tcp.h> |
45 | #include <linux/lru_cache.h> | |
70c71606 | 46 | #include <linux/prefetch.h> |
3b98c0c2 | 47 | #include <linux/drbd_genl_api.h> |
b8907339 PR |
48 | #include <linux/drbd.h> |
49 | #include "drbd_state.h" | |
b411b363 PR |
50 | |
51 | #ifdef __CHECKER__ | |
52 | # define __protected_by(x) __attribute__((require_context(x,1,999,"rdwr"))) | |
53 | # define __protected_read_by(x) __attribute__((require_context(x,1,999,"read"))) | |
54 | # define __protected_write_by(x) __attribute__((require_context(x,1,999,"write"))) | |
55 | # define __must_hold(x) __attribute__((context(x,1,1), require_context(x,1,999,"call"))) | |
56 | #else | |
57 | # define __protected_by(x) | |
58 | # define __protected_read_by(x) | |
59 | # define __protected_write_by(x) | |
60 | # define __must_hold(x) | |
61 | #endif | |
62 | ||
63 | #define __no_warn(lock, stmt) do { __acquire(lock); stmt; __release(lock); } while (0) | |
64 | ||
65 | /* module parameter, defined in drbd_main.c */ | |
66 | extern unsigned int minor_count; | |
67 | extern int disable_sendpage; | |
68 | extern int allow_oos; | |
b411b363 PR |
69 | |
70 | #ifdef CONFIG_DRBD_FAULT_INJECTION | |
71 | extern int enable_faults; | |
72 | extern int fault_rate; | |
73 | extern int fault_devs; | |
74 | #endif | |
75 | ||
76 | extern char usermode_helper[]; | |
77 | ||
78 | ||
b411b363 PR |
79 | /* I don't remember why XCPU ... |
80 | * This is used to wake the asender, | |
81 | * and to interrupt sending the sending task | |
82 | * on disconnect. | |
83 | */ | |
84 | #define DRBD_SIG SIGXCPU | |
85 | ||
86 | /* This is used to stop/restart our threads. | |
87 | * Cannot use SIGTERM nor SIGKILL, since these | |
88 | * are sent out by init on runlevel changes | |
89 | * I choose SIGHUP for now. | |
90 | */ | |
91 | #define DRBD_SIGKILL SIGHUP | |
92 | ||
b411b363 PR |
93 | #define ID_IN_SYNC (4711ULL) |
94 | #define ID_OUT_OF_SYNC (4712ULL) | |
b411b363 | 95 | #define ID_SYNCER (-1ULL) |
579b57ed | 96 | |
4a23f264 | 97 | #define UUID_NEW_BM_OFFSET ((u64)0x0001000000000000ULL) |
b411b363 PR |
98 | |
99 | struct drbd_conf; | |
2111438b | 100 | struct drbd_tconn; |
b411b363 PR |
101 | |
102 | ||
103 | /* to shorten dev_warn(DEV, "msg"); and relatives statements */ | |
104 | #define DEV (disk_to_dev(mdev->vdisk)) | |
105 | ||
60ae4966 PR |
106 | #define conn_printk(LEVEL, TCONN, FMT, ARGS...) \ |
107 | printk(LEVEL "d-con %s: " FMT, TCONN->name , ## ARGS) | |
108 | #define conn_alert(TCONN, FMT, ARGS...) conn_printk(KERN_ALERT, TCONN, FMT, ## ARGS) | |
109 | #define conn_crit(TCONN, FMT, ARGS...) conn_printk(KERN_CRIT, TCONN, FMT, ## ARGS) | |
110 | #define conn_err(TCONN, FMT, ARGS...) conn_printk(KERN_ERR, TCONN, FMT, ## ARGS) | |
111 | #define conn_warn(TCONN, FMT, ARGS...) conn_printk(KERN_WARNING, TCONN, FMT, ## ARGS) | |
112 | #define conn_notice(TCONN, FMT, ARGS...) conn_printk(KERN_NOTICE, TCONN, FMT, ## ARGS) | |
113 | #define conn_info(TCONN, FMT, ARGS...) conn_printk(KERN_INFO, TCONN, FMT, ## ARGS) | |
114 | #define conn_dbg(TCONN, FMT, ARGS...) conn_printk(KERN_DEBUG, TCONN, FMT, ## ARGS) | |
115 | ||
b411b363 PR |
116 | #define D_ASSERT(exp) if (!(exp)) \ |
117 | dev_err(DEV, "ASSERT( " #exp " ) in %s:%d\n", __FILE__, __LINE__) | |
118 | ||
841ce241 AG |
119 | /** |
120 | * expect - Make an assertion | |
121 | * | |
122 | * Unlike the assert macro, this macro returns a boolean result. | |
123 | */ | |
124 | #define expect(exp) ({ \ | |
125 | bool _bool = (exp); \ | |
126 | if (!_bool) \ | |
127 | dev_err(DEV, "ASSERTION %s FAILED in %s\n", \ | |
128 | #exp, __func__); \ | |
129 | _bool; \ | |
130 | }) | |
b411b363 PR |
131 | |
132 | /* Defines to control fault insertion */ | |
133 | enum { | |
134 | DRBD_FAULT_MD_WR = 0, /* meta data write */ | |
135 | DRBD_FAULT_MD_RD = 1, /* read */ | |
136 | DRBD_FAULT_RS_WR = 2, /* resync */ | |
137 | DRBD_FAULT_RS_RD = 3, | |
138 | DRBD_FAULT_DT_WR = 4, /* data */ | |
139 | DRBD_FAULT_DT_RD = 5, | |
140 | DRBD_FAULT_DT_RA = 6, /* data read ahead */ | |
141 | DRBD_FAULT_BM_ALLOC = 7, /* bitmap allocation */ | |
142 | DRBD_FAULT_AL_EE = 8, /* alloc ee */ | |
6b4388ac | 143 | DRBD_FAULT_RECEIVE = 9, /* Changes some bytes upon receiving a [rs]data block */ |
b411b363 PR |
144 | |
145 | DRBD_FAULT_MAX, | |
146 | }; | |
147 | ||
b411b363 PR |
148 | extern unsigned int |
149 | _drbd_insert_fault(struct drbd_conf *mdev, unsigned int type); | |
0cf9d27e | 150 | |
b411b363 PR |
151 | static inline int |
152 | drbd_insert_fault(struct drbd_conf *mdev, unsigned int type) { | |
0cf9d27e | 153 | #ifdef CONFIG_DRBD_FAULT_INJECTION |
b411b363 PR |
154 | return fault_rate && |
155 | (enable_faults & (1<<type)) && | |
156 | _drbd_insert_fault(mdev, type); | |
b411b363 | 157 | #else |
0cf9d27e | 158 | return 0; |
b411b363 | 159 | #endif |
0cf9d27e | 160 | } |
b411b363 PR |
161 | |
162 | /* integer division, round _UP_ to the next integer */ | |
163 | #define div_ceil(A, B) ((A)/(B) + ((A)%(B) ? 1 : 0)) | |
164 | /* usual integer division */ | |
165 | #define div_floor(A, B) ((A)/(B)) | |
166 | ||
b411b363 | 167 | extern struct ratelimit_state drbd_ratelimit_state; |
c141ebda PR |
168 | extern struct idr minors; /* RCU, updates: genl_lock() */ |
169 | extern struct list_head drbd_tconns; /* RCU, updates: genl_lock() */ | |
b411b363 PR |
170 | |
171 | /* on the wire */ | |
d8763023 | 172 | enum drbd_packet { |
b411b363 PR |
173 | /* receiver (data socket) */ |
174 | P_DATA = 0x00, | |
175 | P_DATA_REPLY = 0x01, /* Response to P_DATA_REQUEST */ | |
176 | P_RS_DATA_REPLY = 0x02, /* Response to P_RS_DATA_REQUEST */ | |
177 | P_BARRIER = 0x03, | |
178 | P_BITMAP = 0x04, | |
179 | P_BECOME_SYNC_TARGET = 0x05, | |
180 | P_BECOME_SYNC_SOURCE = 0x06, | |
181 | P_UNPLUG_REMOTE = 0x07, /* Used at various times to hint the peer */ | |
182 | P_DATA_REQUEST = 0x08, /* Used to ask for a data block */ | |
183 | P_RS_DATA_REQUEST = 0x09, /* Used to ask for a data block for resync */ | |
184 | P_SYNC_PARAM = 0x0a, | |
185 | P_PROTOCOL = 0x0b, | |
186 | P_UUIDS = 0x0c, | |
187 | P_SIZES = 0x0d, | |
188 | P_STATE = 0x0e, | |
189 | P_SYNC_UUID = 0x0f, | |
190 | P_AUTH_CHALLENGE = 0x10, | |
191 | P_AUTH_RESPONSE = 0x11, | |
192 | P_STATE_CHG_REQ = 0x12, | |
193 | ||
194 | /* asender (meta socket */ | |
195 | P_PING = 0x13, | |
196 | P_PING_ACK = 0x14, | |
197 | P_RECV_ACK = 0x15, /* Used in protocol B */ | |
198 | P_WRITE_ACK = 0x16, /* Used in protocol C */ | |
199 | P_RS_WRITE_ACK = 0x17, /* Is a P_WRITE_ACK, additionally call set_in_sync(). */ | |
7be8da07 | 200 | P_DISCARD_WRITE = 0x18, /* Used in proto C, two-primaries conflict detection */ |
b411b363 PR |
201 | P_NEG_ACK = 0x19, /* Sent if local disk is unusable */ |
202 | P_NEG_DREPLY = 0x1a, /* Local disk is broken... */ | |
203 | P_NEG_RS_DREPLY = 0x1b, /* Local disk is broken... */ | |
204 | P_BARRIER_ACK = 0x1c, | |
205 | P_STATE_CHG_REPLY = 0x1d, | |
206 | ||
207 | /* "new" commands, no longer fitting into the ordering scheme above */ | |
208 | ||
209 | P_OV_REQUEST = 0x1e, /* data socket */ | |
210 | P_OV_REPLY = 0x1f, | |
211 | P_OV_RESULT = 0x20, /* meta socket */ | |
212 | P_CSUM_RS_REQUEST = 0x21, /* data socket */ | |
213 | P_RS_IS_IN_SYNC = 0x22, /* meta socket */ | |
214 | P_SYNC_PARAM89 = 0x23, /* data socket, protocol version 89 replacement for P_SYNC_PARAM */ | |
215 | P_COMPRESSED_BITMAP = 0x24, /* compressed or otherwise encoded bitmap transfer */ | |
0ced55a3 PR |
216 | /* P_CKPT_FENCE_REQ = 0x25, * currently reserved for protocol D */ |
217 | /* P_CKPT_DISABLE_REQ = 0x26, * currently reserved for protocol D */ | |
218 | P_DELAY_PROBE = 0x27, /* is used on BOTH sockets */ | |
73a01a18 | 219 | P_OUT_OF_SYNC = 0x28, /* Mark as out of sync (Outrunning), data socket */ |
d612d309 | 220 | P_RS_CANCEL = 0x29, /* meta: Used to cancel RS_DATA_REQUEST packet by SyncSource */ |
cf29c9d8 PR |
221 | P_CONN_ST_CHG_REQ = 0x2a, /* data sock: Connection wide state request */ |
222 | P_CONN_ST_CHG_REPLY = 0x2b, /* meta sock: Connection side state req reply */ | |
7be8da07 | 223 | P_RETRY_WRITE = 0x2c, /* Protocol C: retry conflicting write request */ |
036b17ea | 224 | P_PROTOCOL_UPDATE = 0x2d, /* data sock: is used in established connections */ |
b411b363 | 225 | |
b411b363 PR |
226 | P_MAY_IGNORE = 0x100, /* Flag to test if (cmd > P_MAY_IGNORE) ... */ |
227 | P_MAX_OPT_CMD = 0x101, | |
228 | ||
229 | /* special command ids for handshake */ | |
230 | ||
e5d6f33a AG |
231 | P_INITIAL_META = 0xfff1, /* First Packet on the MetaSock */ |
232 | P_INITIAL_DATA = 0xfff2, /* First Packet on the Socket */ | |
b411b363 | 233 | |
6038178e | 234 | P_CONNECTION_FEATURES = 0xfffe /* FIXED for the next century! */ |
b411b363 PR |
235 | }; |
236 | ||
d8763023 | 237 | extern const char *cmdname(enum drbd_packet cmd); |
b411b363 PR |
238 | |
239 | /* for sending/receiving the bitmap, | |
240 | * possibly in some encoding scheme */ | |
241 | struct bm_xfer_ctx { | |
242 | /* "const" | |
243 | * stores total bits and long words | |
244 | * of the bitmap, so we don't need to | |
245 | * call the accessor functions over and again. */ | |
246 | unsigned long bm_bits; | |
247 | unsigned long bm_words; | |
248 | /* during xfer, current position within the bitmap */ | |
249 | unsigned long bit_offset; | |
250 | unsigned long word_offset; | |
251 | ||
252 | /* statistics; index: (h->command == P_BITMAP) */ | |
253 | unsigned packets[2]; | |
254 | unsigned bytes[2]; | |
255 | }; | |
256 | ||
257 | extern void INFO_bm_xfer_stats(struct drbd_conf *mdev, | |
258 | const char *direction, struct bm_xfer_ctx *c); | |
259 | ||
260 | static inline void bm_xfer_ctx_bit_to_word_offset(struct bm_xfer_ctx *c) | |
261 | { | |
262 | /* word_offset counts "native long words" (32 or 64 bit), | |
263 | * aligned at 64 bit. | |
264 | * Encoded packet may end at an unaligned bit offset. | |
265 | * In case a fallback clear text packet is transmitted in | |
266 | * between, we adjust this offset back to the last 64bit | |
267 | * aligned "native long word", which makes coding and decoding | |
268 | * the plain text bitmap much more convenient. */ | |
269 | #if BITS_PER_LONG == 64 | |
270 | c->word_offset = c->bit_offset >> 6; | |
271 | #elif BITS_PER_LONG == 32 | |
272 | c->word_offset = c->bit_offset >> 5; | |
273 | c->word_offset &= ~(1UL); | |
274 | #else | |
275 | # error "unsupported BITS_PER_LONG" | |
276 | #endif | |
277 | } | |
278 | ||
279 | #ifndef __packed | |
280 | #define __packed __attribute__((packed)) | |
281 | #endif | |
282 | ||
283 | /* This is the layout for a packet on the wire. | |
284 | * The byteorder is the network byte order. | |
285 | * (except block_id and barrier fields. | |
286 | * these are pointers to local structs | |
287 | * and have no relevance for the partner, | |
288 | * which just echoes them as received.) | |
289 | * | |
290 | * NOTE that the payload starts at a long aligned offset, | |
291 | * regardless of 32 or 64 bit arch! | |
292 | */ | |
0b70a13d | 293 | struct p_header80 { |
b411b363 PR |
294 | u32 magic; |
295 | u16 command; | |
296 | u16 length; /* bytes of data after this header */ | |
b411b363 | 297 | } __packed; |
0b70a13d PR |
298 | |
299 | /* Header for big packets, Used for data packets exceeding 64kB */ | |
300 | struct p_header95 { | |
301 | u16 magic; /* use DRBD_MAGIC_BIG here */ | |
302 | u16 command; | |
b55d84ba | 303 | u32 length; |
0b70a13d PR |
304 | } __packed; |
305 | ||
0c8e36d9 AG |
306 | struct p_header100 { |
307 | u32 magic; | |
308 | u16 volume; | |
309 | u16 command; | |
310 | u32 length; | |
311 | u32 pad; | |
312 | } __packed; | |
313 | ||
52b061a4 AG |
314 | extern unsigned int drbd_header_size(struct drbd_tconn *tconn); |
315 | ||
b411b363 | 316 | /* these defines must not be changed without changing the protocol version */ |
76d2e7ec PR |
317 | #define DP_HARDBARRIER 1 /* depricated */ |
318 | #define DP_RW_SYNC 2 /* equals REQ_SYNC */ | |
b411b363 | 319 | #define DP_MAY_SET_IN_SYNC 4 |
721a9602 | 320 | #define DP_UNPLUG 8 /* not used anymore */ |
76d2e7ec PR |
321 | #define DP_FUA 16 /* equals REQ_FUA */ |
322 | #define DP_FLUSH 32 /* equals REQ_FLUSH */ | |
323 | #define DP_DISCARD 64 /* equals REQ_DISCARD */ | |
303d1448 PR |
324 | #define DP_SEND_RECEIVE_ACK 128 /* This is a proto B write request */ |
325 | #define DP_SEND_WRITE_ACK 256 /* This is a proto C write request */ | |
b411b363 PR |
326 | |
327 | struct p_data { | |
b411b363 PR |
328 | u64 sector; /* 64 bits sector number */ |
329 | u64 block_id; /* to identify the request in protocol B&C */ | |
330 | u32 seq_num; | |
331 | u32 dp_flags; | |
332 | } __packed; | |
333 | ||
334 | /* | |
335 | * commands which share a struct: | |
336 | * p_block_ack: | |
337 | * P_RECV_ACK (proto B), P_WRITE_ACK (proto C), | |
7be8da07 | 338 | * P_DISCARD_WRITE (proto C, two-primaries conflict detection) |
b411b363 PR |
339 | * p_block_req: |
340 | * P_DATA_REQUEST, P_RS_DATA_REQUEST | |
341 | */ | |
342 | struct p_block_ack { | |
b411b363 PR |
343 | u64 sector; |
344 | u64 block_id; | |
345 | u32 blksize; | |
346 | u32 seq_num; | |
347 | } __packed; | |
348 | ||
b411b363 | 349 | struct p_block_req { |
b411b363 PR |
350 | u64 sector; |
351 | u64 block_id; | |
352 | u32 blksize; | |
353 | u32 pad; /* to multiple of 8 Byte */ | |
354 | } __packed; | |
355 | ||
356 | /* | |
357 | * commands with their own struct for additional fields: | |
6038178e | 358 | * P_CONNECTION_FEATURES |
b411b363 PR |
359 | * P_BARRIER |
360 | * P_BARRIER_ACK | |
361 | * P_SYNC_PARAM | |
362 | * ReportParams | |
363 | */ | |
364 | ||
6038178e | 365 | struct p_connection_features { |
b411b363 PR |
366 | u32 protocol_min; |
367 | u32 feature_flags; | |
368 | u32 protocol_max; | |
369 | ||
370 | /* should be more than enough for future enhancements | |
d942ae44 | 371 | * for now, feature_flags and the reserved array shall be zero. |
b411b363 PR |
372 | */ |
373 | ||
374 | u32 _pad; | |
d942ae44 | 375 | u64 reserved[7]; |
b411b363 | 376 | } __packed; |
b411b363 PR |
377 | |
378 | struct p_barrier { | |
b411b363 PR |
379 | u32 barrier; /* barrier number _handle_ only */ |
380 | u32 pad; /* to multiple of 8 Byte */ | |
381 | } __packed; | |
382 | ||
383 | struct p_barrier_ack { | |
b411b363 PR |
384 | u32 barrier; |
385 | u32 set_size; | |
386 | } __packed; | |
387 | ||
388 | struct p_rs_param { | |
6394b935 | 389 | u32 resync_rate; |
b411b363 PR |
390 | |
391 | /* Since protocol version 88 and higher. */ | |
392 | char verify_alg[0]; | |
393 | } __packed; | |
394 | ||
395 | struct p_rs_param_89 { | |
6394b935 | 396 | u32 resync_rate; |
b411b363 PR |
397 | /* protocol version 89: */ |
398 | char verify_alg[SHARED_SECRET_MAX]; | |
399 | char csums_alg[SHARED_SECRET_MAX]; | |
400 | } __packed; | |
401 | ||
8e26f9cc | 402 | struct p_rs_param_95 { |
6394b935 | 403 | u32 resync_rate; |
8e26f9cc PR |
404 | char verify_alg[SHARED_SECRET_MAX]; |
405 | char csums_alg[SHARED_SECRET_MAX]; | |
406 | u32 c_plan_ahead; | |
407 | u32 c_delay_target; | |
408 | u32 c_fill_target; | |
409 | u32 c_max_rate; | |
410 | } __packed; | |
411 | ||
cf14c2e9 | 412 | enum drbd_conn_flags { |
6139f60d | 413 | CF_DISCARD_MY_DATA = 1, |
cf14c2e9 PR |
414 | CF_DRY_RUN = 2, |
415 | }; | |
416 | ||
b411b363 | 417 | struct p_protocol { |
b411b363 PR |
418 | u32 protocol; |
419 | u32 after_sb_0p; | |
420 | u32 after_sb_1p; | |
421 | u32 after_sb_2p; | |
cf14c2e9 | 422 | u32 conn_flags; |
b411b363 PR |
423 | u32 two_primaries; |
424 | ||
425 | /* Since protocol version 87 and higher. */ | |
426 | char integrity_alg[0]; | |
427 | ||
428 | } __packed; | |
429 | ||
430 | struct p_uuids { | |
b411b363 PR |
431 | u64 uuid[UI_EXTENDED_SIZE]; |
432 | } __packed; | |
433 | ||
434 | struct p_rs_uuid { | |
b411b363 PR |
435 | u64 uuid; |
436 | } __packed; | |
437 | ||
438 | struct p_sizes { | |
b411b363 PR |
439 | u64 d_size; /* size of disk */ |
440 | u64 u_size; /* user requested size */ | |
441 | u64 c_size; /* current exported size */ | |
1816a2b4 | 442 | u32 max_bio_size; /* Maximal size of a BIO */ |
e89b591c PR |
443 | u16 queue_order_type; /* not yet implemented in DRBD*/ |
444 | u16 dds_flags; /* use enum dds_flags here. */ | |
b411b363 PR |
445 | } __packed; |
446 | ||
447 | struct p_state { | |
b411b363 PR |
448 | u32 state; |
449 | } __packed; | |
450 | ||
451 | struct p_req_state { | |
b411b363 PR |
452 | u32 mask; |
453 | u32 val; | |
454 | } __packed; | |
455 | ||
456 | struct p_req_state_reply { | |
b411b363 PR |
457 | u32 retcode; |
458 | } __packed; | |
459 | ||
460 | struct p_drbd06_param { | |
461 | u64 size; | |
462 | u32 state; | |
463 | u32 blksize; | |
464 | u32 protocol; | |
465 | u32 version; | |
466 | u32 gen_cnt[5]; | |
467 | u32 bit_map_gen[5]; | |
468 | } __packed; | |
469 | ||
470 | struct p_discard { | |
b411b363 PR |
471 | u64 block_id; |
472 | u32 seq_num; | |
473 | u32 pad; | |
474 | } __packed; | |
475 | ||
73a01a18 | 476 | struct p_block_desc { |
73a01a18 PR |
477 | u64 sector; |
478 | u32 blksize; | |
479 | u32 pad; /* to multiple of 8 Byte */ | |
480 | } __packed; | |
481 | ||
b411b363 PR |
482 | /* Valid values for the encoding field. |
483 | * Bump proto version when changing this. */ | |
484 | enum drbd_bitmap_code { | |
485 | /* RLE_VLI_Bytes = 0, | |
486 | * and other bit variants had been defined during | |
487 | * algorithm evaluation. */ | |
488 | RLE_VLI_Bits = 2, | |
489 | }; | |
490 | ||
491 | struct p_compressed_bm { | |
b411b363 PR |
492 | /* (encoding & 0x0f): actual encoding, see enum drbd_bitmap_code |
493 | * (encoding & 0x80): polarity (set/unset) of first runlength | |
494 | * ((encoding >> 4) & 0x07): pad_bits, number of trailing zero bits | |
495 | * used to pad up to head.length bytes | |
496 | */ | |
497 | u8 encoding; | |
498 | ||
499 | u8 code[0]; | |
500 | } __packed; | |
501 | ||
0b70a13d | 502 | struct p_delay_probe93 { |
0b70a13d PR |
503 | u32 seq_num; /* sequence number to match the two probe packets */ |
504 | u32 offset; /* usecs the probe got sent after the reference time point */ | |
0ced55a3 PR |
505 | } __packed; |
506 | ||
50d0b1ad AG |
507 | /* |
508 | * Bitmap packets need to fit within a single page on the sender and receiver, | |
509 | * so we are limited to 4 KiB (and not to PAGE_SIZE, which can be bigger). | |
b411b363 | 510 | */ |
e6ef8a5c AG |
511 | #define DRBD_SOCKET_BUFFER_SIZE 4096 |
512 | ||
b411b363 PR |
513 | /**********************************************************************/ |
514 | enum drbd_thread_state { | |
e77a0a5c AG |
515 | NONE, |
516 | RUNNING, | |
517 | EXITING, | |
518 | RESTARTING | |
b411b363 PR |
519 | }; |
520 | ||
521 | struct drbd_thread { | |
522 | spinlock_t t_lock; | |
523 | struct task_struct *task; | |
524 | struct completion stop; | |
525 | enum drbd_thread_state t_state; | |
526 | int (*function) (struct drbd_thread *); | |
392c8801 | 527 | struct drbd_tconn *tconn; |
b411b363 | 528 | int reset_cpu_mask; |
bed879ae | 529 | char name[9]; |
b411b363 PR |
530 | }; |
531 | ||
532 | static inline enum drbd_thread_state get_t_state(struct drbd_thread *thi) | |
533 | { | |
534 | /* THINK testing the t_state seems to be uncritical in all cases | |
535 | * (but thread_{start,stop}), so we can read it *without* the lock. | |
536 | * --lge */ | |
537 | ||
538 | smp_rmb(); | |
539 | return thi->t_state; | |
540 | } | |
541 | ||
b411b363 PR |
542 | struct drbd_work { |
543 | struct list_head list; | |
309a8348 | 544 | int (*cb)(struct drbd_work *, int cancel); |
00d56944 PR |
545 | union { |
546 | struct drbd_conf *mdev; | |
547 | struct drbd_tconn *tconn; | |
548 | }; | |
b411b363 PR |
549 | }; |
550 | ||
ace652ac AG |
551 | #include "drbd_interval.h" |
552 | ||
7be8da07 AG |
553 | extern int drbd_wait_misc(struct drbd_conf *, struct drbd_interval *); |
554 | ||
b411b363 PR |
555 | struct drbd_request { |
556 | struct drbd_work w; | |
b411b363 PR |
557 | |
558 | /* if local IO is not allowed, will be NULL. | |
559 | * if local IO _is_ allowed, holds the locally submitted bio clone, | |
560 | * or, after local IO completion, the ERR_PTR(error). | |
fcefa62e | 561 | * see drbd_request_endio(). */ |
b411b363 PR |
562 | struct bio *private_bio; |
563 | ||
ace652ac | 564 | struct drbd_interval i; |
b411b363 PR |
565 | unsigned int epoch; /* barrier_nr */ |
566 | ||
567 | /* barrier_nr: used to check on "completion" whether this req was in | |
568 | * the current epoch, and we therefore have to close it, | |
569 | * starting a new epoch... | |
570 | */ | |
571 | ||
b411b363 PR |
572 | struct list_head tl_requests; /* ring list in the transfer log */ |
573 | struct bio *master_bio; /* master bio pointer */ | |
574 | unsigned long rq_state; /* see comments above _req_mod() */ | |
575 | int seq_num; | |
576 | unsigned long start_time; | |
577 | }; | |
578 | ||
579 | struct drbd_tl_epoch { | |
580 | struct drbd_work w; | |
581 | struct list_head requests; /* requests before */ | |
582 | struct drbd_tl_epoch *next; /* pointer to the next barrier */ | |
583 | unsigned int br_number; /* the barriers identifier. */ | |
7e602c0a | 584 | int n_writes; /* number of requests attached before this barrier */ |
b411b363 PR |
585 | }; |
586 | ||
b411b363 PR |
587 | struct drbd_epoch { |
588 | struct list_head list; | |
589 | unsigned int barrier_nr; | |
590 | atomic_t epoch_size; /* increased on every request added. */ | |
591 | atomic_t active; /* increased on every req. added, and dec on every finished. */ | |
592 | unsigned long flags; | |
593 | }; | |
594 | ||
595 | /* drbd_epoch flag bits */ | |
596 | enum { | |
b411b363 | 597 | DE_HAVE_BARRIER_NUMBER, |
b411b363 PR |
598 | }; |
599 | ||
600 | enum epoch_event { | |
601 | EV_PUT, | |
602 | EV_GOT_BARRIER_NR, | |
b411b363 | 603 | EV_BECAME_LAST, |
b411b363 PR |
604 | EV_CLEANUP = 32, /* used as flag */ |
605 | }; | |
606 | ||
b411b363 PR |
607 | struct drbd_wq_barrier { |
608 | struct drbd_work w; | |
609 | struct completion done; | |
610 | }; | |
611 | ||
612 | struct digest_info { | |
613 | int digest_size; | |
614 | void *digest; | |
615 | }; | |
616 | ||
f6ffca9f | 617 | struct drbd_peer_request { |
45bb912b | 618 | struct drbd_work w; |
85719573 | 619 | struct drbd_epoch *epoch; /* for writes */ |
45bb912b LE |
620 | struct page *pages; |
621 | atomic_t pending_bios; | |
010f6e67 | 622 | struct drbd_interval i; |
45bb912b LE |
623 | /* see comments on ee flag bits below */ |
624 | unsigned long flags; | |
85719573 PR |
625 | union { |
626 | u64 block_id; | |
627 | struct digest_info *digest; | |
628 | }; | |
45bb912b LE |
629 | }; |
630 | ||
631 | /* ee flag bits. | |
632 | * While corresponding bios are in flight, the only modification will be | |
633 | * set_bit WAS_ERROR, which has to be atomic. | |
634 | * If no bios are in flight yet, or all have been completed, | |
635 | * non-atomic modification to ee->flags is ok. | |
636 | */ | |
b411b363 PR |
637 | enum { |
638 | __EE_CALL_AL_COMPLETE_IO, | |
b411b363 | 639 | __EE_MAY_SET_IN_SYNC, |
45bb912b | 640 | |
45bb912b LE |
641 | /* In case a barrier failed, |
642 | * we need to resubmit without the barrier flag. */ | |
643 | __EE_RESUBMITTED, | |
644 | ||
6c852bec | 645 | /* we may have several bios per peer request. |
45bb912b LE |
646 | * if any of those fail, we set this flag atomically |
647 | * from the endio callback */ | |
648 | __EE_WAS_ERROR, | |
c36c3ced LE |
649 | |
650 | /* This ee has a pointer to a digest instead of a block id */ | |
651 | __EE_HAS_DIGEST, | |
7be8da07 AG |
652 | |
653 | /* Conflicting local requests need to be restarted after this request */ | |
654 | __EE_RESTART_REQUESTS, | |
303d1448 PR |
655 | |
656 | /* The peer wants a write ACK for this (wire proto C) */ | |
657 | __EE_SEND_WRITE_ACK, | |
302bdeae PR |
658 | |
659 | /* Is set when net_conf had two_primaries set while creating this peer_req */ | |
660 | __EE_IN_INTERVAL_TREE, | |
b411b363 PR |
661 | }; |
662 | #define EE_CALL_AL_COMPLETE_IO (1<<__EE_CALL_AL_COMPLETE_IO) | |
b411b363 | 663 | #define EE_MAY_SET_IN_SYNC (1<<__EE_MAY_SET_IN_SYNC) |
45bb912b LE |
664 | #define EE_RESUBMITTED (1<<__EE_RESUBMITTED) |
665 | #define EE_WAS_ERROR (1<<__EE_WAS_ERROR) | |
c36c3ced | 666 | #define EE_HAS_DIGEST (1<<__EE_HAS_DIGEST) |
7be8da07 | 667 | #define EE_RESTART_REQUESTS (1<<__EE_RESTART_REQUESTS) |
303d1448 | 668 | #define EE_SEND_WRITE_ACK (1<<__EE_SEND_WRITE_ACK) |
302bdeae | 669 | #define EE_IN_INTERVAL_TREE (1<<__EE_IN_INTERVAL_TREE) |
b411b363 | 670 | |
01a311a5 | 671 | /* flag bits per mdev */ |
b411b363 | 672 | enum { |
b411b363 PR |
673 | UNPLUG_QUEUED, /* only relevant with kernel 2.4 */ |
674 | UNPLUG_REMOTE, /* sending a "UnplugRemote" could help */ | |
675 | MD_DIRTY, /* current uuids and flags not yet on disk */ | |
b411b363 | 676 | USE_DEGR_WFC_T, /* degr-wfc-timeout instead of wfc-timeout. */ |
b411b363 PR |
677 | CL_ST_CHG_SUCCESS, |
678 | CL_ST_CHG_FAIL, | |
679 | CRASHED_PRIMARY, /* This node was a crashed primary. | |
680 | * Gets cleared when the state.conn | |
681 | * goes into C_CONNECTED state. */ | |
19f843aa | 682 | NO_BARRIER_SUPP, /* underlying block device doesn't implement barriers */ |
b411b363 PR |
683 | CONSIDER_RESYNC, |
684 | ||
a8a4e51e | 685 | MD_NO_FUA, /* Users wants us to not use FUA/FLUSH on meta data dev */ |
b411b363 PR |
686 | SUSPEND_IO, /* suspend application io */ |
687 | BITMAP_IO, /* suspend application io; | |
688 | once no more io in flight, start bitmap io */ | |
689 | BITMAP_IO_QUEUED, /* Started bitmap IO */ | |
82f59cc6 LE |
690 | GO_DISKLESS, /* Disk is being detached, on io-error or admin request. */ |
691 | WAS_IO_ERROR, /* Local disk failed returned IO error */ | |
b411b363 | 692 | RESYNC_AFTER_NEG, /* Resync after online grow after the attach&negotiate finished. */ |
b411b363 PR |
693 | RESIZE_PENDING, /* Size change detected locally, waiting for the response from |
694 | * the peer, if it changed there as well. */ | |
43a5182c | 695 | NEW_CUR_UUID, /* Create new current UUID when thawing IO */ |
0778286a | 696 | AL_SUSPENDED, /* Activity logging is currently suspended. */ |
370a43e7 | 697 | AHEAD_TO_SYNC_SOURCE, /* Ahead -> SyncSource queued */ |
e64a3294 | 698 | B_RS_H_DONE, /* Before resync handler done (already executed) */ |
08b165ba | 699 | DISCARD_MY_DATA, /* discard_my_data flag per volume */ |
b411b363 PR |
700 | }; |
701 | ||
702 | struct drbd_bitmap; /* opaque for drbd_conf */ | |
703 | ||
20ceb2b2 LE |
704 | /* definition of bits in bm_flags to be used in drbd_bm_lock |
705 | * and drbd_bitmap_io and friends. */ | |
706 | enum bm_flag { | |
707 | /* do we need to kfree, or vfree bm_pages? */ | |
708 | BM_P_VMALLOCED = 0x10000, /* internal use only, will be masked out */ | |
709 | ||
710 | /* currently locked for bulk operation */ | |
711 | BM_LOCKED_MASK = 0x7, | |
712 | ||
713 | /* in detail, that is: */ | |
714 | BM_DONT_CLEAR = 0x1, | |
715 | BM_DONT_SET = 0x2, | |
716 | BM_DONT_TEST = 0x4, | |
717 | ||
718 | /* (test bit, count bit) allowed (common case) */ | |
719 | BM_LOCKED_TEST_ALLOWED = 0x3, | |
720 | ||
721 | /* testing bits, as well as setting new bits allowed, but clearing bits | |
722 | * would be unexpected. Used during bitmap receive. Setting new bits | |
723 | * requires sending of "out-of-sync" information, though. */ | |
724 | BM_LOCKED_SET_ALLOWED = 0x1, | |
725 | ||
726 | /* clear is not expected while bitmap is locked for bulk operation */ | |
727 | }; | |
728 | ||
729 | ||
b411b363 PR |
730 | /* TODO sort members for performance |
731 | * MAYBE group them further */ | |
732 | ||
733 | /* THINK maybe we actually want to use the default "event/%s" worker threads | |
734 | * or similar in linux 2.6, which uses per cpu data and threads. | |
b411b363 PR |
735 | */ |
736 | struct drbd_work_queue { | |
737 | struct list_head q; | |
738 | struct semaphore s; /* producers up it, worker down()s it */ | |
739 | spinlock_t q_lock; /* to protect the list. */ | |
740 | }; | |
741 | ||
742 | struct drbd_socket { | |
743 | struct drbd_work_queue work; | |
744 | struct mutex mutex; | |
745 | struct socket *socket; | |
746 | /* this way we get our | |
747 | * send/receive buffers off the stack */ | |
5a87d920 | 748 | void *sbuf; |
e6ef8a5c | 749 | void *rbuf; |
b411b363 PR |
750 | }; |
751 | ||
752 | struct drbd_md { | |
753 | u64 md_offset; /* sector offset to 'super' block */ | |
754 | ||
755 | u64 la_size_sect; /* last agreed size, unit sectors */ | |
756 | u64 uuid[UI_SIZE]; | |
757 | u64 device_uuid; | |
758 | u32 flags; | |
759 | u32 md_size_sect; | |
760 | ||
761 | s32 al_offset; /* signed relative sector offset to al area */ | |
762 | s32 bm_offset; /* signed relative sector offset to bitmap */ | |
763 | ||
764 | /* u32 al_nr_extents; important for restoring the AL | |
f399002e | 765 | * is stored into ldev->dc.al_extents, which in turn |
b411b363 PR |
766 | * gets applied to act_log->nr_elements |
767 | */ | |
768 | }; | |
769 | ||
b411b363 PR |
770 | struct drbd_backing_dev { |
771 | struct block_device *backing_bdev; | |
772 | struct block_device *md_bdev; | |
b411b363 | 773 | struct drbd_md md; |
daeda1cc | 774 | struct disk_conf *disk_conf; /* RCU, for updates: mdev->tconn->conf_update */ |
b411b363 PR |
775 | sector_t known_size; /* last known size of that backing device */ |
776 | }; | |
777 | ||
778 | struct drbd_md_io { | |
cdfda633 | 779 | unsigned int done; |
b411b363 PR |
780 | int error; |
781 | }; | |
782 | ||
783 | struct bm_io_work { | |
784 | struct drbd_work w; | |
785 | char *why; | |
20ceb2b2 | 786 | enum bm_flag flags; |
b411b363 PR |
787 | int (*io_fn)(struct drbd_conf *mdev); |
788 | void (*done)(struct drbd_conf *mdev, int rv); | |
789 | }; | |
790 | ||
791 | enum write_ordering_e { | |
792 | WO_none, | |
793 | WO_drain_io, | |
794 | WO_bdev_flush, | |
b411b363 PR |
795 | }; |
796 | ||
778f271d | 797 | struct fifo_buffer { |
778f271d PR |
798 | unsigned int head_index; |
799 | unsigned int size; | |
9958c857 PR |
800 | int total; /* sum of all values */ |
801 | int values[0]; | |
778f271d | 802 | }; |
9958c857 | 803 | extern struct fifo_buffer *fifo_alloc(int fifo_size); |
778f271d | 804 | |
01a311a5 PR |
805 | /* flag bits per tconn */ |
806 | enum { | |
807 | NET_CONGESTED, /* The data socket is congested */ | |
25703f83 | 808 | DISCARD_CONCURRENT, /* Set on one node, cleared on the peer! */ |
e43ef195 | 809 | SEND_PING, /* whether asender should send a ping asap */ |
808e37b8 | 810 | SIGNAL_ASENDER, /* whether asender wants to be interrupted */ |
2a67d8b9 | 811 | GOT_PING_ACK, /* set when we receive a ping_ack packet, ping_wait gets woken */ |
fc3b10a4 PR |
812 | CONN_WD_ST_CHG_OKAY, |
813 | CONN_WD_ST_CHG_FAIL, | |
8169e41b | 814 | CONN_DRY_RUN, /* Expect disconnect after resync handshake. */ |
6936fcb4 | 815 | CREATE_BARRIER, /* next P_DATA is preceded by a P_BARRIER */ |
01a311a5 PR |
816 | }; |
817 | ||
2111438b PR |
818 | struct drbd_tconn { /* is a resource from the config file */ |
819 | char *name; /* Resource name */ | |
543cc10b | 820 | struct list_head all_tconn; /* linked on global drbd_tconns */ |
9dc9fbb3 | 821 | struct kref kref; |
8410da8f PR |
822 | struct idr volumes; /* <tconn, vnr> to mdev mapping */ |
823 | enum drbd_conns cstate; /* Only C_STANDALONE to C_WF_REPORT_PARAMS */ | |
8e0af25f PR |
824 | unsigned susp:1; /* IO suspended by user */ |
825 | unsigned susp_nod:1; /* IO suspended because no data */ | |
826 | unsigned susp_fen:1; /* IO suspended because fence peer handler runs */ | |
8410da8f | 827 | struct mutex cstate_mutex; /* Protects graceful disconnects */ |
2111438b | 828 | |
062e879c | 829 | unsigned long flags; |
44ed167d | 830 | struct net_conf *net_conf; /* content protected by rcu */ |
a0095508 | 831 | struct mutex conf_update; /* mutex for ready-copy-update of net_conf and disk_conf */ |
91fd4dad | 832 | wait_queue_head_t ping_wait; /* Woken upon reception of a ping, and a state change */ |
f399002e | 833 | struct res_opts res_opts; |
e42325a5 | 834 | |
089c075d AG |
835 | struct sockaddr_storage my_addr; |
836 | int my_addr_len; | |
837 | struct sockaddr_storage peer_addr; | |
838 | int peer_addr_len; | |
839 | ||
e42325a5 PR |
840 | struct drbd_socket data; /* data/barrier/cstate/parameter packets */ |
841 | struct drbd_socket meta; /* ping/ack (metadata) packets */ | |
31890f4a PR |
842 | int agreed_pro_version; /* actually used protocol version */ |
843 | unsigned long last_received; /* in jiffies, either socket */ | |
844 | unsigned int ko_count; | |
e6b3ea83 | 845 | |
87eeee41 PR |
846 | spinlock_t req_lock; |
847 | struct drbd_tl_epoch *unused_spare_tle; /* for pre-allocation */ | |
848 | struct drbd_tl_epoch *newest_tle; | |
849 | struct drbd_tl_epoch *oldest_tle; | |
850 | struct list_head out_of_sequence_requests; | |
cdfda633 | 851 | struct list_head barrier_acked_requests; |
87eeee41 | 852 | |
a0638456 | 853 | struct crypto_hash *cram_hmac_tfm; |
46e1ce41 | 854 | struct crypto_hash *integrity_tfm; /* checksums we compute, updates protected by tconn->data->mutex */ |
036b17ea | 855 | struct crypto_hash *peer_integrity_tfm; /* checksums we verify, only accessed from receiver thread */ |
f399002e LE |
856 | struct crypto_hash *csums_tfm; |
857 | struct crypto_hash *verify_tfm; | |
a0638456 PR |
858 | void *int_dig_in; |
859 | void *int_dig_vv; | |
860 | ||
e6b3ea83 PR |
861 | struct drbd_thread receiver; |
862 | struct drbd_thread worker; | |
863 | struct drbd_thread asender; | |
80822284 | 864 | cpumask_var_t cpu_mask; |
2111438b PR |
865 | }; |
866 | ||
b411b363 | 867 | struct drbd_conf { |
2111438b PR |
868 | struct drbd_tconn *tconn; |
869 | int vnr; /* volume number within the connection */ | |
81fa2e67 | 870 | struct kref kref; |
2111438b | 871 | |
b411b363 PR |
872 | /* things that are stored as / read from meta data on disk */ |
873 | unsigned long flags; | |
874 | ||
875 | /* configured by drbdsetup */ | |
b411b363 PR |
876 | struct drbd_backing_dev *ldev __protected_by(local); |
877 | ||
878 | sector_t p_size; /* partner's disk size */ | |
879 | struct request_queue *rq_queue; | |
880 | struct block_device *this_bdev; | |
881 | struct gendisk *vdisk; | |
882 | ||
b411b363 PR |
883 | struct drbd_work resync_work, |
884 | unplug_work, | |
e9e6f3ec | 885 | go_diskless, |
c4752ef1 PR |
886 | md_sync_work, |
887 | start_resync_work; | |
b411b363 PR |
888 | struct timer_list resync_timer; |
889 | struct timer_list md_sync_timer; | |
370a43e7 | 890 | struct timer_list start_resync_timer; |
7fde2be9 | 891 | struct timer_list request_timer; |
ee15b038 LE |
892 | #ifdef DRBD_DEBUG_MD_SYNC |
893 | struct { | |
894 | unsigned int line; | |
895 | const char* func; | |
896 | } last_md_mark_dirty; | |
897 | #endif | |
b411b363 PR |
898 | |
899 | /* Used after attach while negotiating new disk state. */ | |
900 | union drbd_state new_state_tmp; | |
901 | ||
da9fbc27 | 902 | union drbd_dev_state state; |
b411b363 PR |
903 | wait_queue_head_t misc_wait; |
904 | wait_queue_head_t state_wait; /* upon each state change. */ | |
905 | unsigned int send_cnt; | |
906 | unsigned int recv_cnt; | |
907 | unsigned int read_cnt; | |
908 | unsigned int writ_cnt; | |
909 | unsigned int al_writ_cnt; | |
910 | unsigned int bm_writ_cnt; | |
911 | atomic_t ap_bio_cnt; /* Requests we need to complete */ | |
912 | atomic_t ap_pending_cnt; /* AP data packets on the wire, ack expected */ | |
913 | atomic_t rs_pending_cnt; /* RS request/data packets on the wire */ | |
d942ae44 | 914 | atomic_t unacked_cnt; /* Need to send replies for */ |
b411b363 | 915 | atomic_t local_cnt; /* Waiting for local completion */ |
b2fb6dbe | 916 | |
dac1389c AG |
917 | /* Interval tree of pending local requests */ |
918 | struct rb_root read_requests; | |
de696716 AG |
919 | struct rb_root write_requests; |
920 | ||
4b0715f0 | 921 | /* blocks to resync in this run [unit BM_BLOCK_SIZE] */ |
b411b363 | 922 | unsigned long rs_total; |
4b0715f0 | 923 | /* number of resync blocks that failed in this run */ |
b411b363 PR |
924 | unsigned long rs_failed; |
925 | /* Syncer's start time [unit jiffies] */ | |
926 | unsigned long rs_start; | |
927 | /* cumulated time in PausedSyncX state [unit jiffies] */ | |
928 | unsigned long rs_paused; | |
1d7734a0 LE |
929 | /* skipped because csum was equal [unit BM_BLOCK_SIZE] */ |
930 | unsigned long rs_same_csum; | |
931 | #define DRBD_SYNC_MARKS 8 | |
932 | #define DRBD_SYNC_MARK_STEP (3*HZ) | |
b411b363 | 933 | /* block not up-to-date at mark [unit BM_BLOCK_SIZE] */ |
1d7734a0 | 934 | unsigned long rs_mark_left[DRBD_SYNC_MARKS]; |
b411b363 | 935 | /* marks's time [unit jiffies] */ |
1d7734a0 LE |
936 | unsigned long rs_mark_time[DRBD_SYNC_MARKS]; |
937 | /* current index into rs_mark_{left,time} */ | |
938 | int rs_last_mark; | |
b411b363 PR |
939 | |
940 | /* where does the admin want us to start? (sector) */ | |
941 | sector_t ov_start_sector; | |
942 | /* where are we now? (sector) */ | |
943 | sector_t ov_position; | |
944 | /* Start sector of out of sync range (to merge printk reporting). */ | |
945 | sector_t ov_last_oos_start; | |
946 | /* size of out-of-sync range in sectors. */ | |
947 | sector_t ov_last_oos_size; | |
948 | unsigned long ov_left; /* in bits */ | |
b411b363 | 949 | |
b411b363 PR |
950 | struct drbd_bitmap *bitmap; |
951 | unsigned long bm_resync_fo; /* bit offset for drbd_bm_find_next */ | |
952 | ||
953 | /* Used to track operations of resync... */ | |
954 | struct lru_cache *resync; | |
955 | /* Number of locked elements in resync LRU */ | |
956 | unsigned int resync_locked; | |
957 | /* resync extent number waiting for application requests */ | |
958 | unsigned int resync_wenr; | |
959 | ||
960 | int open_cnt; | |
961 | u64 *p_uuid; | |
962 | struct drbd_epoch *current_epoch; | |
963 | spinlock_t epoch_lock; | |
964 | unsigned int epochs; | |
965 | enum write_ordering_e write_ordering; | |
85719573 PR |
966 | struct list_head active_ee; /* IO in progress (P_DATA gets written to disk) */ |
967 | struct list_head sync_ee; /* IO in progress (P_RS_DATA_REPLY gets written to disk) */ | |
18b75d75 AG |
968 | struct list_head done_ee; /* need to send P_WRITE_ACK */ |
969 | struct list_head read_ee; /* [RS]P_DATA_REQUEST being read */ | |
b411b363 | 970 | struct list_head net_ee; /* zero-copy network send in progress */ |
b411b363 | 971 | |
b411b363 | 972 | int next_barrier_nr; |
b411b363 | 973 | struct list_head resync_reads; |
435f0740 LE |
974 | atomic_t pp_in_use; /* allocated from page pool */ |
975 | atomic_t pp_in_use_by_net; /* sendpage()d, still referenced by tcp */ | |
b411b363 PR |
976 | wait_queue_head_t ee_wait; |
977 | struct page *md_io_page; /* one page buffer for md_io */ | |
cdfda633 PR |
978 | struct drbd_md_io md_io; |
979 | atomic_t md_io_in_use; /* protects the md_io, md_io_page and md_io_tmpp */ | |
b411b363 PR |
980 | spinlock_t al_lock; |
981 | wait_queue_head_t al_wait; | |
982 | struct lru_cache *act_log; /* activity log */ | |
983 | unsigned int al_tr_number; | |
984 | int al_tr_cycle; | |
985 | int al_tr_pos; /* position of the next transaction in the journal */ | |
b411b363 PR |
986 | wait_queue_head_t seq_wait; |
987 | atomic_t packet_seq; | |
988 | unsigned int peer_seq; | |
989 | spinlock_t peer_seq_lock; | |
990 | unsigned int minor; | |
991 | unsigned long comm_bm_set; /* communicated number of set bits. */ | |
b411b363 PR |
992 | struct bm_io_work bm_io_work; |
993 | u64 ed_uuid; /* UUID of the exposed data */ | |
8410da8f PR |
994 | struct mutex own_state_mutex; |
995 | struct mutex *state_mutex; /* either own_state_mutex or mdev->tconn->cstate_mutex */ | |
b411b363 | 996 | char congestion_reason; /* Why we where congested... */ |
1d7734a0 LE |
997 | atomic_t rs_sect_in; /* for incoming resync data rate, SyncTarget */ |
998 | atomic_t rs_sect_ev; /* for submitted resync data rate, both */ | |
999 | int rs_last_sect_ev; /* counter to compare with */ | |
1000 | int rs_last_events; /* counter of read or write "events" (unit sectors) | |
1001 | * on the lower level device when we last looked. */ | |
1002 | int c_sync_rate; /* current resync rate after syncer throttle magic */ | |
813472ce | 1003 | struct fifo_buffer *rs_plan_s; /* correction values of resync planer (RCU, tconn->conn_update) */ |
778f271d | 1004 | int rs_in_flight; /* resync sectors in flight (to proxy, in proxy and from proxy) */ |
759fbdfb | 1005 | atomic_t ap_in_flight; /* App sectors in flight (waiting for ack) */ |
99432fcc PR |
1006 | int peer_max_bio_size; |
1007 | int local_max_bio_size; | |
b411b363 PR |
1008 | }; |
1009 | ||
1010 | static inline struct drbd_conf *minor_to_mdev(unsigned int minor) | |
1011 | { | |
81a5d60e | 1012 | return (struct drbd_conf *)idr_find(&minors, minor); |
b411b363 PR |
1013 | } |
1014 | ||
1015 | static inline unsigned int mdev_to_minor(struct drbd_conf *mdev) | |
1016 | { | |
1017 | return mdev->minor; | |
1018 | } | |
1019 | ||
eefc2f7d PR |
1020 | static inline struct drbd_conf *vnr_to_mdev(struct drbd_tconn *tconn, int vnr) |
1021 | { | |
1022 | return (struct drbd_conf *)idr_find(&tconn->volumes, vnr); | |
1023 | } | |
1024 | ||
b411b363 PR |
1025 | /* |
1026 | * function declarations | |
1027 | *************************/ | |
1028 | ||
1029 | /* drbd_main.c */ | |
1030 | ||
e89b591c PR |
1031 | enum dds_flags { |
1032 | DDSF_FORCED = 1, | |
1033 | DDSF_NO_RESYNC = 2, /* Do not run a resync for the new space */ | |
1034 | }; | |
1035 | ||
b411b363 | 1036 | extern void drbd_init_set_defaults(struct drbd_conf *mdev); |
b411b363 PR |
1037 | extern int drbd_thread_start(struct drbd_thread *thi); |
1038 | extern void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait); | |
392c8801 | 1039 | extern char *drbd_task_to_thread_name(struct drbd_tconn *tconn, struct task_struct *task); |
b411b363 | 1040 | #ifdef CONFIG_SMP |
80822284 PR |
1041 | extern void drbd_thread_current_set_cpu(struct drbd_thread *thi); |
1042 | extern void drbd_calc_cpu_mask(struct drbd_tconn *tconn); | |
b411b363 | 1043 | #else |
80822284 | 1044 | #define drbd_thread_current_set_cpu(A) ({}) |
b411b363 PR |
1045 | #define drbd_calc_cpu_mask(A) ({}) |
1046 | #endif | |
2f5cdd0b | 1047 | extern void tl_release(struct drbd_tconn *, unsigned int barrier_nr, |
b411b363 | 1048 | unsigned int set_size); |
2f5cdd0b PR |
1049 | extern void tl_clear(struct drbd_tconn *); |
1050 | extern void _tl_add_barrier(struct drbd_tconn *, struct drbd_tl_epoch *); | |
360cc740 | 1051 | extern void drbd_free_sock(struct drbd_tconn *tconn); |
bedbd2a5 PR |
1052 | extern int drbd_send(struct drbd_tconn *tconn, struct socket *sock, |
1053 | void *buf, size_t size, unsigned msg_flags); | |
fb708e40 AG |
1054 | extern int drbd_send_all(struct drbd_tconn *, struct socket *, void *, size_t, |
1055 | unsigned); | |
1056 | ||
d659f2aa | 1057 | extern int __drbd_send_protocol(struct drbd_tconn *tconn, enum drbd_packet cmd); |
dc8228d1 | 1058 | extern int drbd_send_protocol(struct drbd_tconn *tconn); |
b411b363 PR |
1059 | extern int drbd_send_uuids(struct drbd_conf *mdev); |
1060 | extern int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev); | |
9c1b7f72 | 1061 | extern void drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev); |
e89b591c | 1062 | extern int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags); |
cf29c9d8 PR |
1063 | extern int _conn_send_state_req(struct drbd_tconn *, int vnr, enum drbd_packet cmd, |
1064 | union drbd_state, union drbd_state); | |
43de7c85 PR |
1065 | extern int drbd_send_state(struct drbd_conf *mdev, union drbd_state s); |
1066 | extern int drbd_send_current_state(struct drbd_conf *mdev); | |
f399002e | 1067 | extern int drbd_send_sync_param(struct drbd_conf *mdev); |
d4e67d7c AG |
1068 | extern void drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr, |
1069 | u32 set_size); | |
f6ffca9f AG |
1070 | extern int drbd_send_ack(struct drbd_conf *, enum drbd_packet, |
1071 | struct drbd_peer_request *); | |
a9a9994d AG |
1072 | extern void drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packet cmd, |
1073 | struct p_block_req *rp); | |
1074 | extern void drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packet cmd, | |
1075 | struct p_data *dp, int data_size); | |
d8763023 | 1076 | extern int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packet cmd, |
b411b363 | 1077 | sector_t sector, int blksize, u64 block_id); |
8f7bed77 | 1078 | extern int drbd_send_out_of_sync(struct drbd_conf *, struct drbd_request *); |
f6ffca9f AG |
1079 | extern int drbd_send_block(struct drbd_conf *, enum drbd_packet, |
1080 | struct drbd_peer_request *); | |
b411b363 | 1081 | extern int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req); |
b411b363 PR |
1082 | extern int drbd_send_drequest(struct drbd_conf *mdev, int cmd, |
1083 | sector_t sector, int size, u64 block_id); | |
d8763023 AG |
1084 | extern int drbd_send_drequest_csum(struct drbd_conf *mdev, sector_t sector, |
1085 | int size, void *digest, int digest_size, | |
1086 | enum drbd_packet cmd); | |
b411b363 PR |
1087 | extern int drbd_send_ov_request(struct drbd_conf *mdev,sector_t sector,int size); |
1088 | ||
1089 | extern int drbd_send_bitmap(struct drbd_conf *mdev); | |
2f4e7abe | 1090 | extern void drbd_send_sr_reply(struct drbd_conf *mdev, enum drbd_state_rv retcode); |
9f5bdc33 | 1091 | extern void conn_send_sr_reply(struct drbd_tconn *tconn, enum drbd_state_rv retcode); |
b411b363 PR |
1092 | extern void drbd_free_bc(struct drbd_backing_dev *ldev); |
1093 | extern void drbd_mdev_cleanup(struct drbd_conf *mdev); | |
62b0da3a | 1094 | void drbd_print_uuids(struct drbd_conf *mdev, const char *text); |
b411b363 | 1095 | |
b411b363 PR |
1096 | extern void drbd_md_sync(struct drbd_conf *mdev); |
1097 | extern int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev); | |
b411b363 PR |
1098 | extern void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local); |
1099 | extern void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local); | |
1100 | extern void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local); | |
1101 | extern void _drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local); | |
1102 | extern void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local); | |
1103 | extern void drbd_md_set_flag(struct drbd_conf *mdev, int flags) __must_hold(local); | |
1104 | extern void drbd_md_clear_flag(struct drbd_conf *mdev, int flags)__must_hold(local); | |
1105 | extern int drbd_md_test_flag(struct drbd_backing_dev *, int); | |
ee15b038 | 1106 | #ifndef DRBD_DEBUG_MD_SYNC |
b411b363 | 1107 | extern void drbd_md_mark_dirty(struct drbd_conf *mdev); |
ee15b038 LE |
1108 | #else |
1109 | #define drbd_md_mark_dirty(m) drbd_md_mark_dirty_(m, __LINE__ , __func__ ) | |
1110 | extern void drbd_md_mark_dirty_(struct drbd_conf *mdev, | |
1111 | unsigned int line, const char *func); | |
1112 | #endif | |
b411b363 PR |
1113 | extern void drbd_queue_bitmap_io(struct drbd_conf *mdev, |
1114 | int (*io_fn)(struct drbd_conf *), | |
1115 | void (*done)(struct drbd_conf *, int), | |
20ceb2b2 LE |
1116 | char *why, enum bm_flag flags); |
1117 | extern int drbd_bitmap_io(struct drbd_conf *mdev, | |
1118 | int (*io_fn)(struct drbd_conf *), | |
1119 | char *why, enum bm_flag flags); | |
b411b363 PR |
1120 | extern int drbd_bmio_set_n_write(struct drbd_conf *mdev); |
1121 | extern int drbd_bmio_clear_n_write(struct drbd_conf *mdev); | |
e9e6f3ec | 1122 | extern void drbd_go_diskless(struct drbd_conf *mdev); |
82f59cc6 | 1123 | extern void drbd_ldev_destroy(struct drbd_conf *mdev); |
b411b363 | 1124 | |
b411b363 PR |
1125 | /* Meta data layout |
1126 | We reserve a 128MB Block (4k aligned) | |
1127 | * either at the end of the backing device | |
3ad2f3fb | 1128 | * or on a separate meta data device. */ |
b411b363 | 1129 | |
b411b363 | 1130 | /* The following numbers are sectors */ |
7ad651b5 LE |
1131 | /* Allows up to about 3.8TB, so if you want more, |
1132 | * you need to use the "flexible" meta data format. */ | |
1133 | #define MD_RESERVED_SECT (128LU << 11) /* 128 MB, unit sectors */ | |
1134 | #define MD_AL_OFFSET 8 /* 8 Sectors after start of meta area */ | |
1135 | #define MD_AL_SECTORS 64 /* = 32 kB on disk activity log ring buffer */ | |
1136 | #define MD_BM_OFFSET (MD_AL_OFFSET + MD_AL_SECTORS) | |
1137 | ||
1138 | /* we do all meta data IO in 4k blocks */ | |
1139 | #define MD_BLOCK_SHIFT 12 | |
1140 | #define MD_BLOCK_SIZE (1<<MD_BLOCK_SHIFT) | |
1141 | ||
1142 | /* One activity log extent represents 4M of storage */ | |
1143 | #define AL_EXTENT_SHIFT 22 | |
b411b363 PR |
1144 | #define AL_EXTENT_SIZE (1<<AL_EXTENT_SHIFT) |
1145 | ||
7ad651b5 LE |
1146 | /* We could make these currently hardcoded constants configurable |
1147 | * variables at create-md time (or even re-configurable at runtime?). | |
1148 | * Which will require some more changes to the DRBD "super block" | |
1149 | * and attach code. | |
1150 | * | |
1151 | * updates per transaction: | |
1152 | * This many changes to the active set can be logged with one transaction. | |
1153 | * This number is arbitrary. | |
1154 | * context per transaction: | |
1155 | * This many context extent numbers are logged with each transaction. | |
1156 | * This number is resulting from the transaction block size (4k), the layout | |
1157 | * of the transaction header, and the number of updates per transaction. | |
1158 | * See drbd_actlog.c:struct al_transaction_on_disk | |
1159 | * */ | |
1160 | #define AL_UPDATES_PER_TRANSACTION 64 // arbitrary | |
1161 | #define AL_CONTEXT_PER_TRANSACTION 919 // (4096 - 36 - 6*64)/4 | |
1162 | ||
b411b363 PR |
1163 | #if BITS_PER_LONG == 32 |
1164 | #define LN2_BPL 5 | |
1165 | #define cpu_to_lel(A) cpu_to_le32(A) | |
1166 | #define lel_to_cpu(A) le32_to_cpu(A) | |
1167 | #elif BITS_PER_LONG == 64 | |
1168 | #define LN2_BPL 6 | |
1169 | #define cpu_to_lel(A) cpu_to_le64(A) | |
1170 | #define lel_to_cpu(A) le64_to_cpu(A) | |
1171 | #else | |
1172 | #error "LN2 of BITS_PER_LONG unknown!" | |
1173 | #endif | |
1174 | ||
1175 | /* resync bitmap */ | |
1176 | /* 16MB sized 'bitmap extent' to track syncer usage */ | |
1177 | struct bm_extent { | |
1178 | int rs_left; /* number of bits set (out of sync) in this extent. */ | |
1179 | int rs_failed; /* number of failed resync requests in this extent. */ | |
1180 | unsigned long flags; | |
1181 | struct lc_element lce; | |
1182 | }; | |
1183 | ||
1184 | #define BME_NO_WRITES 0 /* bm_extent.flags: no more requests on this one! */ | |
1185 | #define BME_LOCKED 1 /* bm_extent.flags: syncer active on this one. */ | |
e3555d85 | 1186 | #define BME_PRIORITY 2 /* finish resync IO on this extent ASAP! App IO waiting! */ |
b411b363 PR |
1187 | |
1188 | /* drbd_bitmap.c */ | |
1189 | /* | |
1190 | * We need to store one bit for a block. | |
1191 | * Example: 1GB disk @ 4096 byte blocks ==> we need 32 KB bitmap. | |
1192 | * Bit 0 ==> local node thinks this block is binary identical on both nodes | |
1193 | * Bit 1 ==> local node thinks this block needs to be synced. | |
1194 | */ | |
1195 | ||
8e26f9cc PR |
1196 | #define SLEEP_TIME (HZ/10) |
1197 | ||
45dfffeb LE |
1198 | /* We do bitmap IO in units of 4k blocks. |
1199 | * We also still have a hardcoded 4k per bit relation. */ | |
1200 | #define BM_BLOCK_SHIFT 12 /* 4k per bit */ | |
b411b363 | 1201 | #define BM_BLOCK_SIZE (1<<BM_BLOCK_SHIFT) |
45dfffeb LE |
1202 | /* mostly arbitrarily set the represented size of one bitmap extent, |
1203 | * aka resync extent, to 16 MiB (which is also 512 Byte worth of bitmap | |
1204 | * at 4k per bit resolution) */ | |
1205 | #define BM_EXT_SHIFT 24 /* 16 MiB per resync extent */ | |
b411b363 PR |
1206 | #define BM_EXT_SIZE (1<<BM_EXT_SHIFT) |
1207 | ||
1208 | #if (BM_EXT_SHIFT != 24) || (BM_BLOCK_SHIFT != 12) | |
1209 | #error "HAVE YOU FIXED drbdmeta AS WELL??" | |
1210 | #endif | |
1211 | ||
1212 | /* thus many _storage_ sectors are described by one bit */ | |
1213 | #define BM_SECT_TO_BIT(x) ((x)>>(BM_BLOCK_SHIFT-9)) | |
1214 | #define BM_BIT_TO_SECT(x) ((sector_t)(x)<<(BM_BLOCK_SHIFT-9)) | |
1215 | #define BM_SECT_PER_BIT BM_BIT_TO_SECT(1) | |
1216 | ||
1217 | /* bit to represented kilo byte conversion */ | |
1218 | #define Bit2KB(bits) ((bits)<<(BM_BLOCK_SHIFT-10)) | |
1219 | ||
1220 | /* in which _bitmap_ extent (resp. sector) the bit for a certain | |
1221 | * _storage_ sector is located in */ | |
1222 | #define BM_SECT_TO_EXT(x) ((x)>>(BM_EXT_SHIFT-9)) | |
1223 | ||
1224 | /* how much _storage_ sectors we have per bitmap sector */ | |
1225 | #define BM_EXT_TO_SECT(x) ((sector_t)(x) << (BM_EXT_SHIFT-9)) | |
1226 | #define BM_SECT_PER_EXT BM_EXT_TO_SECT(1) | |
1227 | ||
1228 | /* in one sector of the bitmap, we have this many activity_log extents. */ | |
1229 | #define AL_EXT_PER_BM_SECT (1 << (BM_EXT_SHIFT - AL_EXTENT_SHIFT)) | |
1230 | #define BM_WORDS_PER_AL_EXT (1 << (AL_EXTENT_SHIFT-BM_BLOCK_SHIFT-LN2_BPL)) | |
1231 | ||
1232 | #define BM_BLOCKS_PER_BM_EXT_B (BM_EXT_SHIFT - BM_BLOCK_SHIFT) | |
1233 | #define BM_BLOCKS_PER_BM_EXT_MASK ((1<<BM_BLOCKS_PER_BM_EXT_B) - 1) | |
1234 | ||
1235 | /* the extent in "PER_EXTENT" below is an activity log extent | |
1236 | * we need that many (long words/bytes) to store the bitmap | |
1237 | * of one AL_EXTENT_SIZE chunk of storage. | |
1238 | * we can store the bitmap for that many AL_EXTENTS within | |
1239 | * one sector of the _on_disk_ bitmap: | |
1240 | * bit 0 bit 37 bit 38 bit (512*8)-1 | |
1241 | * ...|........|........|.. // ..|........| | |
1242 | * sect. 0 `296 `304 ^(512*8*8)-1 | |
1243 | * | |
1244 | #define BM_WORDS_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / BITS_PER_LONG ) | |
1245 | #define BM_BYTES_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / 8 ) // 128 | |
1246 | #define BM_EXT_PER_SECT ( 512 / BM_BYTES_PER_EXTENT ) // 4 | |
1247 | */ | |
1248 | ||
1249 | #define DRBD_MAX_SECTORS_32 (0xffffffffLU) | |
1250 | #define DRBD_MAX_SECTORS_BM \ | |
1251 | ((MD_RESERVED_SECT - MD_BM_OFFSET) * (1LL<<(BM_EXT_SHIFT-9))) | |
1252 | #if DRBD_MAX_SECTORS_BM < DRBD_MAX_SECTORS_32 | |
1253 | #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_BM | |
1254 | #define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_BM | |
36bfc7e2 | 1255 | #elif !defined(CONFIG_LBDAF) && BITS_PER_LONG == 32 |
b411b363 PR |
1256 | #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_32 |
1257 | #define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_32 | |
1258 | #else | |
1259 | #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_BM | |
1260 | /* 16 TB in units of sectors */ | |
1261 | #if BITS_PER_LONG == 32 | |
1262 | /* adjust by one page worth of bitmap, | |
1263 | * so we won't wrap around in drbd_bm_find_next_bit. | |
1264 | * you should use 64bit OS for that much storage, anyways. */ | |
1265 | #define DRBD_MAX_SECTORS_FLEX BM_BIT_TO_SECT(0xffff7fff) | |
1266 | #else | |
4b0715f0 LE |
1267 | /* we allow up to 1 PiB now on 64bit architecture with "flexible" meta data */ |
1268 | #define DRBD_MAX_SECTORS_FLEX (1UL << 51) | |
1269 | /* corresponds to (1UL << 38) bits right now. */ | |
b411b363 PR |
1270 | #endif |
1271 | #endif | |
1272 | ||
23361cf3 LE |
1273 | /* BIO_MAX_SIZE is 256 * PAGE_CACHE_SIZE, |
1274 | * so for typical PAGE_CACHE_SIZE of 4k, that is (1<<20) Byte. | |
1275 | * Since we may live in a mixed-platform cluster, | |
1276 | * we limit us to a platform agnostic constant here for now. | |
1277 | * A followup commit may allow even bigger BIO sizes, | |
1278 | * once we thought that through. */ | |
1279 | #define DRBD_MAX_BIO_SIZE (1 << 20) | |
1280 | #if DRBD_MAX_BIO_SIZE > BIO_MAX_SIZE | |
1281 | #error Architecture not supported: DRBD_MAX_BIO_SIZE > BIO_MAX_SIZE | |
1282 | #endif | |
99432fcc | 1283 | #define DRBD_MAX_BIO_SIZE_SAFE (1 << 12) /* Works always = 4k */ |
b411b363 | 1284 | |
2ffca4f3 PR |
1285 | #define DRBD_MAX_SIZE_H80_PACKET (1 << 15) /* Header 80 only allows packets up to 32KiB data */ |
1286 | #define DRBD_MAX_BIO_SIZE_P95 (1 << 17) /* Protocol 95 to 99 allows bios up to 128KiB */ | |
b411b363 PR |
1287 | |
1288 | extern int drbd_bm_init(struct drbd_conf *mdev); | |
02d9a94b | 1289 | extern int drbd_bm_resize(struct drbd_conf *mdev, sector_t sectors, int set_new_bits); |
b411b363 PR |
1290 | extern void drbd_bm_cleanup(struct drbd_conf *mdev); |
1291 | extern void drbd_bm_set_all(struct drbd_conf *mdev); | |
1292 | extern void drbd_bm_clear_all(struct drbd_conf *mdev); | |
4b0715f0 | 1293 | /* set/clear/test only a few bits at a time */ |
b411b363 PR |
1294 | extern int drbd_bm_set_bits( |
1295 | struct drbd_conf *mdev, unsigned long s, unsigned long e); | |
1296 | extern int drbd_bm_clear_bits( | |
1297 | struct drbd_conf *mdev, unsigned long s, unsigned long e); | |
4b0715f0 LE |
1298 | extern int drbd_bm_count_bits( |
1299 | struct drbd_conf *mdev, const unsigned long s, const unsigned long e); | |
1300 | /* bm_set_bits variant for use while holding drbd_bm_lock, | |
1301 | * may process the whole bitmap in one go */ | |
b411b363 PR |
1302 | extern void _drbd_bm_set_bits(struct drbd_conf *mdev, |
1303 | const unsigned long s, const unsigned long e); | |
1304 | extern int drbd_bm_test_bit(struct drbd_conf *mdev, unsigned long bitnr); | |
1305 | extern int drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr); | |
19f843aa | 1306 | extern int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(local); |
b411b363 | 1307 | extern int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local); |
45dfffeb | 1308 | extern void drbd_bm_mark_for_writeout(struct drbd_conf *mdev, int page_nr); |
b411b363 | 1309 | extern int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local); |
45dfffeb | 1310 | extern int drbd_bm_write_hinted(struct drbd_conf *mdev) __must_hold(local); |
b411b363 PR |
1311 | extern unsigned long drbd_bm_ALe_set_all(struct drbd_conf *mdev, |
1312 | unsigned long al_enr); | |
1313 | extern size_t drbd_bm_words(struct drbd_conf *mdev); | |
1314 | extern unsigned long drbd_bm_bits(struct drbd_conf *mdev); | |
1315 | extern sector_t drbd_bm_capacity(struct drbd_conf *mdev); | |
4b0715f0 LE |
1316 | |
1317 | #define DRBD_END_OF_BITMAP (~(unsigned long)0) | |
b411b363 PR |
1318 | extern unsigned long drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo); |
1319 | /* bm_find_next variants for use while you hold drbd_bm_lock() */ | |
1320 | extern unsigned long _drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo); | |
1321 | extern unsigned long _drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo); | |
0778286a | 1322 | extern unsigned long _drbd_bm_total_weight(struct drbd_conf *mdev); |
b411b363 PR |
1323 | extern unsigned long drbd_bm_total_weight(struct drbd_conf *mdev); |
1324 | extern int drbd_bm_rs_done(struct drbd_conf *mdev); | |
1325 | /* for receive_bitmap */ | |
1326 | extern void drbd_bm_merge_lel(struct drbd_conf *mdev, size_t offset, | |
1327 | size_t number, unsigned long *buffer); | |
19f843aa | 1328 | /* for _drbd_send_bitmap */ |
b411b363 PR |
1329 | extern void drbd_bm_get_lel(struct drbd_conf *mdev, size_t offset, |
1330 | size_t number, unsigned long *buffer); | |
1331 | ||
20ceb2b2 | 1332 | extern void drbd_bm_lock(struct drbd_conf *mdev, char *why, enum bm_flag flags); |
b411b363 | 1333 | extern void drbd_bm_unlock(struct drbd_conf *mdev); |
b411b363 PR |
1334 | /* drbd_main.c */ |
1335 | ||
1336 | extern struct kmem_cache *drbd_request_cache; | |
6c852bec | 1337 | extern struct kmem_cache *drbd_ee_cache; /* peer requests */ |
b411b363 PR |
1338 | extern struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */ |
1339 | extern struct kmem_cache *drbd_al_ext_cache; /* activity log extents */ | |
1340 | extern mempool_t *drbd_request_mempool; | |
1341 | extern mempool_t *drbd_ee_mempool; | |
1342 | ||
35abf594 LE |
1343 | /* drbd's page pool, used to buffer data received from the peer, |
1344 | * or data requested by the peer. | |
1345 | * | |
1346 | * This does not have an emergency reserve. | |
1347 | * | |
1348 | * When allocating from this pool, it first takes pages from the pool. | |
1349 | * Only if the pool is depleted will try to allocate from the system. | |
1350 | * | |
1351 | * The assumption is that pages taken from this pool will be processed, | |
1352 | * and given back, "quickly", and then can be recycled, so we can avoid | |
1353 | * frequent calls to alloc_page(), and still will be able to make progress even | |
1354 | * under memory pressure. | |
1355 | */ | |
1356 | extern struct page *drbd_pp_pool; | |
b411b363 PR |
1357 | extern spinlock_t drbd_pp_lock; |
1358 | extern int drbd_pp_vacant; | |
1359 | extern wait_queue_head_t drbd_pp_wait; | |
1360 | ||
35abf594 LE |
1361 | /* We also need a standard (emergency-reserve backed) page pool |
1362 | * for meta data IO (activity log, bitmap). | |
1363 | * We can keep it global, as long as it is used as "N pages at a time". | |
1364 | * 128 should be plenty, currently we probably can get away with as few as 1. | |
1365 | */ | |
1366 | #define DRBD_MIN_POOL_PAGES 128 | |
1367 | extern mempool_t *drbd_md_io_page_pool; | |
1368 | ||
da4a75d2 LE |
1369 | /* We also need to make sure we get a bio |
1370 | * when we need it for housekeeping purposes */ | |
1371 | extern struct bio_set *drbd_md_io_bio_set; | |
1372 | /* to allocate from that set */ | |
1373 | extern struct bio *bio_alloc_drbd(gfp_t gfp_mask); | |
1374 | ||
b411b363 PR |
1375 | extern rwlock_t global_state_lock; |
1376 | ||
80883197 | 1377 | extern int conn_lowest_minor(struct drbd_tconn *tconn); |
774b3055 | 1378 | enum drbd_ret_code conn_new_minor(struct drbd_tconn *tconn, unsigned int minor, int vnr); |
81fa2e67 | 1379 | extern void drbd_minor_destroy(struct kref *kref); |
b411b363 | 1380 | |
afbbfa88 AG |
1381 | extern int set_resource_options(struct drbd_tconn *tconn, struct res_opts *res_opts); |
1382 | extern struct drbd_tconn *conn_create(const char *name, struct res_opts *res_opts); | |
9dc9fbb3 | 1383 | extern void conn_destroy(struct kref *kref); |
0ace9dfa | 1384 | struct drbd_tconn *conn_get_by_name(const char *name); |
089c075d AG |
1385 | extern struct drbd_tconn *conn_get_by_addrs(void *my_addr, int my_addr_len, |
1386 | void *peer_addr, int peer_addr_len); | |
91fd4dad | 1387 | extern void conn_free_crypto(struct drbd_tconn *tconn); |
2111438b | 1388 | |
b411b363 PR |
1389 | extern int proc_details; |
1390 | ||
1391 | /* drbd_req */ | |
7be8da07 | 1392 | extern int __drbd_make_request(struct drbd_conf *, struct bio *, unsigned long); |
2f58dcfc | 1393 | extern int drbd_make_request(struct request_queue *q, struct bio *bio); |
b411b363 PR |
1394 | extern int drbd_read_remote(struct drbd_conf *mdev, struct drbd_request *req); |
1395 | extern int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct bio_vec *bvec); | |
1396 | extern int is_valid_ar_handle(struct drbd_request *, sector_t); | |
1397 | ||
1398 | ||
1399 | /* drbd_nl.c */ | |
8432b314 | 1400 | extern int drbd_msg_put_info(const char *info); |
b411b363 PR |
1401 | extern void drbd_suspend_io(struct drbd_conf *mdev); |
1402 | extern void drbd_resume_io(struct drbd_conf *mdev); | |
1403 | extern char *ppsize(char *buf, unsigned long long size); | |
ef5e44a6 | 1404 | extern sector_t drbd_new_dev_size(struct drbd_conf *, struct drbd_backing_dev *, sector_t, int); |
b411b363 | 1405 | enum determine_dev_size { dev_size_error = -1, unchanged = 0, shrunk = 1, grew = 2 }; |
24c4830c | 1406 | extern enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *, enum dds_flags) __must_hold(local); |
b411b363 | 1407 | extern void resync_after_online_grow(struct drbd_conf *); |
99432fcc | 1408 | extern void drbd_reconsider_max_bio_size(struct drbd_conf *mdev); |
bf885f8a AG |
1409 | extern enum drbd_state_rv drbd_set_role(struct drbd_conf *mdev, |
1410 | enum drbd_role new_role, | |
1411 | int force); | |
cb703454 PR |
1412 | extern bool conn_try_outdate_peer(struct drbd_tconn *tconn); |
1413 | extern void conn_try_outdate_peer_async(struct drbd_tconn *tconn); | |
b411b363 PR |
1414 | extern int drbd_khelper(struct drbd_conf *mdev, char *cmd); |
1415 | ||
1416 | /* drbd_worker.c */ | |
1417 | extern int drbd_worker(struct drbd_thread *thi); | |
95f8efd0 AG |
1418 | enum drbd_ret_code drbd_resync_after_valid(struct drbd_conf *mdev, int o_minor); |
1419 | void drbd_resync_after_changed(struct drbd_conf *mdev); | |
b411b363 PR |
1420 | extern void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side); |
1421 | extern void resume_next_sg(struct drbd_conf *mdev); | |
1422 | extern void suspend_other_sg(struct drbd_conf *mdev); | |
1423 | extern int drbd_resync_finished(struct drbd_conf *mdev); | |
1424 | /* maybe rather drbd_main.c ? */ | |
cdfda633 PR |
1425 | extern void *drbd_md_get_buffer(struct drbd_conf *mdev); |
1426 | extern void drbd_md_put_buffer(struct drbd_conf *mdev); | |
b411b363 PR |
1427 | extern int drbd_md_sync_page_io(struct drbd_conf *mdev, |
1428 | struct drbd_backing_dev *bdev, sector_t sector, int rw); | |
8f7bed77 | 1429 | extern void drbd_ov_out_of_sync_found(struct drbd_conf *, sector_t, int); |
cdfda633 | 1430 | extern void wait_until_done_or_disk_failure(struct drbd_conf *mdev, unsigned int *done); |
9bd28d3c | 1431 | extern void drbd_rs_controller_reset(struct drbd_conf *mdev); |
b411b363 | 1432 | |
8f7bed77 | 1433 | static inline void ov_out_of_sync_print(struct drbd_conf *mdev) |
b411b363 PR |
1434 | { |
1435 | if (mdev->ov_last_oos_size) { | |
1436 | dev_err(DEV, "Out of sync: start=%llu, size=%lu (sectors)\n", | |
1437 | (unsigned long long)mdev->ov_last_oos_start, | |
1438 | (unsigned long)mdev->ov_last_oos_size); | |
1439 | } | |
1440 | mdev->ov_last_oos_size=0; | |
1441 | } | |
1442 | ||
1443 | ||
45bb912b | 1444 | extern void drbd_csum_bio(struct drbd_conf *, struct crypto_hash *, struct bio *, void *); |
f6ffca9f AG |
1445 | extern void drbd_csum_ee(struct drbd_conf *, struct crypto_hash *, |
1446 | struct drbd_peer_request *, void *); | |
b411b363 | 1447 | /* worker callbacks */ |
99920dc5 AG |
1448 | extern int w_read_retry_remote(struct drbd_work *, int); |
1449 | extern int w_e_end_data_req(struct drbd_work *, int); | |
1450 | extern int w_e_end_rsdata_req(struct drbd_work *, int); | |
1451 | extern int w_e_end_csum_rs_req(struct drbd_work *, int); | |
1452 | extern int w_e_end_ov_reply(struct drbd_work *, int); | |
1453 | extern int w_e_end_ov_req(struct drbd_work *, int); | |
1454 | extern int w_ov_finished(struct drbd_work *, int); | |
1455 | extern int w_resync_timer(struct drbd_work *, int); | |
1456 | extern int w_send_write_hint(struct drbd_work *, int); | |
1457 | extern int w_make_resync_request(struct drbd_work *, int); | |
1458 | extern int w_send_dblock(struct drbd_work *, int); | |
1459 | extern int w_send_barrier(struct drbd_work *, int); | |
1460 | extern int w_send_read_req(struct drbd_work *, int); | |
1461 | extern int w_prev_work_done(struct drbd_work *, int); | |
1462 | extern int w_e_reissue(struct drbd_work *, int); | |
1463 | extern int w_restart_disk_io(struct drbd_work *, int); | |
8f7bed77 | 1464 | extern int w_send_out_of_sync(struct drbd_work *, int); |
99920dc5 | 1465 | extern int w_start_resync(struct drbd_work *, int); |
b411b363 PR |
1466 | |
1467 | extern void resync_timer_fn(unsigned long data); | |
370a43e7 | 1468 | extern void start_resync_timer_fn(unsigned long data); |
b411b363 PR |
1469 | |
1470 | /* drbd_receiver.c */ | |
e3555d85 | 1471 | extern int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector); |
fbe29dec AG |
1472 | extern int drbd_submit_peer_request(struct drbd_conf *, |
1473 | struct drbd_peer_request *, const unsigned, | |
1474 | const int); | |
7721f567 | 1475 | extern int drbd_free_peer_reqs(struct drbd_conf *, struct list_head *); |
0db55363 AG |
1476 | extern struct drbd_peer_request *drbd_alloc_peer_req(struct drbd_conf *, u64, |
1477 | sector_t, unsigned int, | |
1478 | gfp_t) __must_hold(local); | |
3967deb1 AG |
1479 | extern void __drbd_free_peer_req(struct drbd_conf *, struct drbd_peer_request *, |
1480 | int); | |
1481 | #define drbd_free_peer_req(m,e) __drbd_free_peer_req(m, e, 0) | |
1482 | #define drbd_free_net_peer_req(m,e) __drbd_free_peer_req(m, e, 1) | |
c37c8ecf | 1483 | extern struct page *drbd_alloc_pages(struct drbd_conf *, unsigned int, bool); |
b411b363 PR |
1484 | extern void drbd_set_recv_tcq(struct drbd_conf *mdev, int tcq_enabled); |
1485 | extern void _drbd_clear_done_ee(struct drbd_conf *mdev, struct list_head *to_be_freed); | |
0e29d163 | 1486 | extern void conn_flush_workqueue(struct drbd_tconn *tconn); |
c141ebda | 1487 | extern int drbd_connected(struct drbd_conf *mdev); |
0e29d163 PR |
1488 | static inline void drbd_flush_workqueue(struct drbd_conf *mdev) |
1489 | { | |
1490 | conn_flush_workqueue(mdev->tconn); | |
1491 | } | |
b411b363 | 1492 | |
ed439848 LE |
1493 | /* Yes, there is kernel_setsockopt, but only since 2.6.18. |
1494 | * So we have our own copy of it here. */ | |
b411b363 | 1495 | static inline int drbd_setsockopt(struct socket *sock, int level, int optname, |
ed439848 | 1496 | char *optval, int optlen) |
b411b363 | 1497 | { |
ed439848 LE |
1498 | mm_segment_t oldfs = get_fs(); |
1499 | char __user *uoptval; | |
b411b363 | 1500 | int err; |
ed439848 LE |
1501 | |
1502 | uoptval = (char __user __force *)optval; | |
1503 | ||
1504 | set_fs(KERNEL_DS); | |
b411b363 | 1505 | if (level == SOL_SOCKET) |
ed439848 | 1506 | err = sock_setsockopt(sock, level, optname, uoptval, optlen); |
b411b363 | 1507 | else |
ed439848 | 1508 | err = sock->ops->setsockopt(sock, level, optname, uoptval, |
b411b363 | 1509 | optlen); |
ed439848 | 1510 | set_fs(oldfs); |
b411b363 PR |
1511 | return err; |
1512 | } | |
1513 | ||
1514 | static inline void drbd_tcp_cork(struct socket *sock) | |
1515 | { | |
ed439848 | 1516 | int val = 1; |
b411b363 | 1517 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK, |
ed439848 | 1518 | (char*)&val, sizeof(val)); |
b411b363 PR |
1519 | } |
1520 | ||
1521 | static inline void drbd_tcp_uncork(struct socket *sock) | |
1522 | { | |
ed439848 | 1523 | int val = 0; |
b411b363 | 1524 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK, |
ed439848 | 1525 | (char*)&val, sizeof(val)); |
b411b363 PR |
1526 | } |
1527 | ||
1528 | static inline void drbd_tcp_nodelay(struct socket *sock) | |
1529 | { | |
ed439848 | 1530 | int val = 1; |
b411b363 | 1531 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_NODELAY, |
ed439848 | 1532 | (char*)&val, sizeof(val)); |
b411b363 PR |
1533 | } |
1534 | ||
1535 | static inline void drbd_tcp_quickack(struct socket *sock) | |
1536 | { | |
ed439848 | 1537 | int val = 2; |
b411b363 | 1538 | (void) drbd_setsockopt(sock, SOL_TCP, TCP_QUICKACK, |
ed439848 | 1539 | (char*)&val, sizeof(val)); |
b411b363 PR |
1540 | } |
1541 | ||
1542 | void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo); | |
1543 | ||
1544 | /* drbd_proc.c */ | |
1545 | extern struct proc_dir_entry *drbd_proc; | |
7d4e9d09 | 1546 | extern const struct file_operations drbd_proc_fops; |
b411b363 PR |
1547 | extern const char *drbd_conn_str(enum drbd_conns s); |
1548 | extern const char *drbd_role_str(enum drbd_role s); | |
1549 | ||
1550 | /* drbd_actlog.c */ | |
181286ad LE |
1551 | extern void drbd_al_begin_io(struct drbd_conf *mdev, struct drbd_interval *i); |
1552 | extern void drbd_al_complete_io(struct drbd_conf *mdev, struct drbd_interval *i); | |
b411b363 PR |
1553 | extern void drbd_rs_complete_io(struct drbd_conf *mdev, sector_t sector); |
1554 | extern int drbd_rs_begin_io(struct drbd_conf *mdev, sector_t sector); | |
1555 | extern int drbd_try_rs_begin_io(struct drbd_conf *mdev, sector_t sector); | |
1556 | extern void drbd_rs_cancel_all(struct drbd_conf *mdev); | |
1557 | extern int drbd_rs_del_all(struct drbd_conf *mdev); | |
1558 | extern void drbd_rs_failed_io(struct drbd_conf *mdev, | |
1559 | sector_t sector, int size); | |
ea5442af | 1560 | extern void drbd_advance_rs_marks(struct drbd_conf *mdev, unsigned long still_to_go); |
b411b363 PR |
1561 | extern void __drbd_set_in_sync(struct drbd_conf *mdev, sector_t sector, |
1562 | int size, const char *file, const unsigned int line); | |
1563 | #define drbd_set_in_sync(mdev, sector, size) \ | |
1564 | __drbd_set_in_sync(mdev, sector, size, __FILE__, __LINE__) | |
73a01a18 | 1565 | extern int __drbd_set_out_of_sync(struct drbd_conf *mdev, sector_t sector, |
b411b363 PR |
1566 | int size, const char *file, const unsigned int line); |
1567 | #define drbd_set_out_of_sync(mdev, sector, size) \ | |
1568 | __drbd_set_out_of_sync(mdev, sector, size, __FILE__, __LINE__) | |
b411b363 PR |
1569 | extern void drbd_al_shrink(struct drbd_conf *mdev); |
1570 | ||
b411b363 | 1571 | /* drbd_nl.c */ |
3b98c0c2 LE |
1572 | /* state info broadcast */ |
1573 | struct sib_info { | |
1574 | enum drbd_state_info_bcast_reason sib_reason; | |
1575 | union { | |
1576 | struct { | |
1577 | char *helper_name; | |
1578 | unsigned helper_exit_code; | |
1579 | }; | |
1580 | struct { | |
1581 | union drbd_state os; | |
1582 | union drbd_state ns; | |
1583 | }; | |
1584 | }; | |
1585 | }; | |
1586 | void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib); | |
b411b363 | 1587 | |
b411b363 PR |
1588 | /* |
1589 | * inline helper functions | |
1590 | *************************/ | |
1591 | ||
45bb912b LE |
1592 | /* see also page_chain_add and friends in drbd_receiver.c */ |
1593 | static inline struct page *page_chain_next(struct page *page) | |
1594 | { | |
1595 | return (struct page *)page_private(page); | |
1596 | } | |
1597 | #define page_chain_for_each(page) \ | |
1598 | for (; page && ({ prefetch(page_chain_next(page)); 1; }); \ | |
1599 | page = page_chain_next(page)) | |
1600 | #define page_chain_for_each_safe(page, n) \ | |
1601 | for (; page && ({ n = page_chain_next(page); 1; }); page = n) | |
1602 | ||
1603 | static inline int drbd_bio_has_active_page(struct bio *bio) | |
1604 | { | |
1605 | struct bio_vec *bvec; | |
1606 | int i; | |
1607 | ||
1608 | __bio_for_each_segment(bvec, bio, i, 0) { | |
1609 | if (page_count(bvec->bv_page) > 1) | |
1610 | return 1; | |
1611 | } | |
1612 | ||
1613 | return 0; | |
1614 | } | |
1615 | ||
045417f7 | 1616 | static inline int drbd_peer_req_has_active_page(struct drbd_peer_request *peer_req) |
45bb912b | 1617 | { |
db830c46 | 1618 | struct page *page = peer_req->pages; |
45bb912b LE |
1619 | page_chain_for_each(page) { |
1620 | if (page_count(page) > 1) | |
1621 | return 1; | |
1622 | } | |
1623 | return 0; | |
1624 | } | |
1625 | ||
bf885f8a AG |
1626 | static inline enum drbd_state_rv |
1627 | _drbd_set_state(struct drbd_conf *mdev, union drbd_state ns, | |
1628 | enum chg_state_flags flags, struct completion *done) | |
b411b363 | 1629 | { |
bf885f8a | 1630 | enum drbd_state_rv rv; |
b411b363 PR |
1631 | |
1632 | read_lock(&global_state_lock); | |
1633 | rv = __drbd_set_state(mdev, ns, flags, done); | |
1634 | read_unlock(&global_state_lock); | |
1635 | ||
1636 | return rv; | |
1637 | } | |
1638 | ||
78bae59b PR |
1639 | static inline union drbd_state drbd_read_state(struct drbd_conf *mdev) |
1640 | { | |
1641 | union drbd_state rv; | |
1642 | ||
da9fbc27 | 1643 | rv.i = mdev->state.i; |
8e0af25f PR |
1644 | rv.susp = mdev->tconn->susp; |
1645 | rv.susp_nod = mdev->tconn->susp_nod; | |
1646 | rv.susp_fen = mdev->tconn->susp_fen; | |
78bae59b PR |
1647 | |
1648 | return rv; | |
1649 | } | |
1650 | ||
b411b363 PR |
1651 | #define __drbd_chk_io_error(m,f) __drbd_chk_io_error_(m,f, __func__) |
1652 | static inline void __drbd_chk_io_error_(struct drbd_conf *mdev, int forcedetach, const char *where) | |
1653 | { | |
daeda1cc PR |
1654 | enum drbd_io_error_p ep; |
1655 | ||
1656 | rcu_read_lock(); | |
1657 | ep = rcu_dereference(mdev->ldev->disk_conf)->on_io_error; | |
1658 | rcu_read_unlock(); | |
1659 | switch (ep) { | |
1660 | case EP_PASS_ON: /* FIXME would this be better named "Ignore"? */ | |
b411b363 | 1661 | if (!forcedetach) { |
7383506c | 1662 | if (__ratelimit(&drbd_ratelimit_state)) |
82f59cc6 | 1663 | dev_err(DEV, "Local IO failed in %s.\n", where); |
d2e17807 PR |
1664 | if (mdev->state.disk > D_INCONSISTENT) |
1665 | _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_HARD, NULL); | |
b411b363 PR |
1666 | break; |
1667 | } | |
1668 | /* NOTE fall through to detach case if forcedetach set */ | |
1669 | case EP_DETACH: | |
1670 | case EP_CALL_HELPER: | |
82f59cc6 | 1671 | set_bit(WAS_IO_ERROR, &mdev->flags); |
b411b363 PR |
1672 | if (mdev->state.disk > D_FAILED) { |
1673 | _drbd_set_state(_NS(mdev, disk, D_FAILED), CS_HARD, NULL); | |
82f59cc6 LE |
1674 | dev_err(DEV, |
1675 | "Local IO failed in %s. Detaching...\n", where); | |
b411b363 PR |
1676 | } |
1677 | break; | |
1678 | } | |
1679 | } | |
1680 | ||
1681 | /** | |
1682 | * drbd_chk_io_error: Handle the on_io_error setting, should be called from all io completion handlers | |
1683 | * @mdev: DRBD device. | |
1684 | * @error: Error code passed to the IO completion callback | |
1685 | * @forcedetach: Force detach. I.e. the error happened while accessing the meta data | |
1686 | * | |
1687 | * See also drbd_main.c:after_state_ch() if (os.disk > D_FAILED && ns.disk == D_FAILED) | |
1688 | */ | |
1689 | #define drbd_chk_io_error(m,e,f) drbd_chk_io_error_(m,e,f, __func__) | |
1690 | static inline void drbd_chk_io_error_(struct drbd_conf *mdev, | |
1691 | int error, int forcedetach, const char *where) | |
1692 | { | |
1693 | if (error) { | |
1694 | unsigned long flags; | |
87eeee41 | 1695 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
b411b363 | 1696 | __drbd_chk_io_error_(mdev, forcedetach, where); |
87eeee41 | 1697 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
b411b363 PR |
1698 | } |
1699 | } | |
1700 | ||
1701 | ||
1702 | /** | |
1703 | * drbd_md_first_sector() - Returns the first sector number of the meta data area | |
1704 | * @bdev: Meta data block device. | |
1705 | * | |
1706 | * BTW, for internal meta data, this happens to be the maximum capacity | |
1707 | * we could agree upon with our peer node. | |
1708 | */ | |
daeda1cc | 1709 | static inline sector_t _drbd_md_first_sector(int meta_dev_idx, struct drbd_backing_dev *bdev) |
b411b363 | 1710 | { |
daeda1cc | 1711 | switch (meta_dev_idx) { |
b411b363 PR |
1712 | case DRBD_MD_INDEX_INTERNAL: |
1713 | case DRBD_MD_INDEX_FLEX_INT: | |
1714 | return bdev->md.md_offset + bdev->md.bm_offset; | |
1715 | case DRBD_MD_INDEX_FLEX_EXT: | |
1716 | default: | |
1717 | return bdev->md.md_offset; | |
1718 | } | |
1719 | } | |
1720 | ||
daeda1cc PR |
1721 | static inline sector_t drbd_md_first_sector(struct drbd_backing_dev *bdev) |
1722 | { | |
1723 | int meta_dev_idx; | |
1724 | ||
1725 | rcu_read_lock(); | |
1726 | meta_dev_idx = rcu_dereference(bdev->disk_conf)->meta_dev_idx; | |
1727 | rcu_read_unlock(); | |
1728 | ||
1729 | return _drbd_md_first_sector(meta_dev_idx, bdev); | |
1730 | } | |
1731 | ||
b411b363 PR |
1732 | /** |
1733 | * drbd_md_last_sector() - Return the last sector number of the meta data area | |
1734 | * @bdev: Meta data block device. | |
1735 | */ | |
1736 | static inline sector_t drbd_md_last_sector(struct drbd_backing_dev *bdev) | |
1737 | { | |
daeda1cc PR |
1738 | int meta_dev_idx; |
1739 | ||
1740 | rcu_read_lock(); | |
1741 | meta_dev_idx = rcu_dereference(bdev->disk_conf)->meta_dev_idx; | |
1742 | rcu_read_unlock(); | |
1743 | ||
1744 | switch (meta_dev_idx) { | |
b411b363 PR |
1745 | case DRBD_MD_INDEX_INTERNAL: |
1746 | case DRBD_MD_INDEX_FLEX_INT: | |
1747 | return bdev->md.md_offset + MD_AL_OFFSET - 1; | |
1748 | case DRBD_MD_INDEX_FLEX_EXT: | |
1749 | default: | |
1750 | return bdev->md.md_offset + bdev->md.md_size_sect; | |
1751 | } | |
1752 | } | |
1753 | ||
1754 | /* Returns the number of 512 byte sectors of the device */ | |
1755 | static inline sector_t drbd_get_capacity(struct block_device *bdev) | |
1756 | { | |
1757 | /* return bdev ? get_capacity(bdev->bd_disk) : 0; */ | |
77304d2a | 1758 | return bdev ? i_size_read(bdev->bd_inode) >> 9 : 0; |
b411b363 PR |
1759 | } |
1760 | ||
1761 | /** | |
1762 | * drbd_get_max_capacity() - Returns the capacity we announce to out peer | |
1763 | * @bdev: Meta data block device. | |
1764 | * | |
1765 | * returns the capacity we announce to out peer. we clip ourselves at the | |
1766 | * various MAX_SECTORS, because if we don't, current implementation will | |
1767 | * oops sooner or later | |
1768 | */ | |
1769 | static inline sector_t drbd_get_max_capacity(struct drbd_backing_dev *bdev) | |
1770 | { | |
1771 | sector_t s; | |
daeda1cc PR |
1772 | int meta_dev_idx; |
1773 | ||
1774 | rcu_read_lock(); | |
1775 | meta_dev_idx = rcu_dereference(bdev->disk_conf)->meta_dev_idx; | |
1776 | rcu_read_unlock(); | |
1777 | ||
1778 | switch (meta_dev_idx) { | |
b411b363 PR |
1779 | case DRBD_MD_INDEX_INTERNAL: |
1780 | case DRBD_MD_INDEX_FLEX_INT: | |
1781 | s = drbd_get_capacity(bdev->backing_bdev) | |
1782 | ? min_t(sector_t, DRBD_MAX_SECTORS_FLEX, | |
daeda1cc | 1783 | _drbd_md_first_sector(meta_dev_idx, bdev)) |
b411b363 PR |
1784 | : 0; |
1785 | break; | |
1786 | case DRBD_MD_INDEX_FLEX_EXT: | |
1787 | s = min_t(sector_t, DRBD_MAX_SECTORS_FLEX, | |
1788 | drbd_get_capacity(bdev->backing_bdev)); | |
1789 | /* clip at maximum size the meta device can support */ | |
1790 | s = min_t(sector_t, s, | |
1791 | BM_EXT_TO_SECT(bdev->md.md_size_sect | |
1792 | - bdev->md.bm_offset)); | |
1793 | break; | |
1794 | default: | |
1795 | s = min_t(sector_t, DRBD_MAX_SECTORS, | |
1796 | drbd_get_capacity(bdev->backing_bdev)); | |
1797 | } | |
1798 | return s; | |
1799 | } | |
1800 | ||
1801 | /** | |
1802 | * drbd_md_ss__() - Return the sector number of our meta data super block | |
1803 | * @mdev: DRBD device. | |
1804 | * @bdev: Meta data block device. | |
1805 | */ | |
1806 | static inline sector_t drbd_md_ss__(struct drbd_conf *mdev, | |
1807 | struct drbd_backing_dev *bdev) | |
1808 | { | |
daeda1cc PR |
1809 | int meta_dev_idx; |
1810 | ||
1811 | rcu_read_lock(); | |
1812 | meta_dev_idx = rcu_dereference(bdev->disk_conf)->meta_dev_idx; | |
1813 | rcu_read_unlock(); | |
1814 | ||
1815 | switch (meta_dev_idx) { | |
b411b363 | 1816 | default: /* external, some index */ |
daeda1cc | 1817 | return MD_RESERVED_SECT * meta_dev_idx; |
b411b363 PR |
1818 | case DRBD_MD_INDEX_INTERNAL: |
1819 | /* with drbd08, internal meta data is always "flexible" */ | |
1820 | case DRBD_MD_INDEX_FLEX_INT: | |
1821 | /* sizeof(struct md_on_disk_07) == 4k | |
1822 | * position: last 4k aligned block of 4k size */ | |
1823 | if (!bdev->backing_bdev) { | |
1824 | if (__ratelimit(&drbd_ratelimit_state)) { | |
1825 | dev_err(DEV, "bdev->backing_bdev==NULL\n"); | |
1826 | dump_stack(); | |
1827 | } | |
1828 | return 0; | |
1829 | } | |
1830 | return (drbd_get_capacity(bdev->backing_bdev) & ~7ULL) | |
1831 | - MD_AL_OFFSET; | |
1832 | case DRBD_MD_INDEX_FLEX_EXT: | |
1833 | return 0; | |
1834 | } | |
1835 | } | |
1836 | ||
b411b363 PR |
1837 | static inline void |
1838 | drbd_queue_work_front(struct drbd_work_queue *q, struct drbd_work *w) | |
1839 | { | |
1840 | unsigned long flags; | |
1841 | spin_lock_irqsave(&q->q_lock, flags); | |
1842 | list_add(&w->list, &q->q); | |
1843 | up(&q->s); /* within the spinlock, | |
1844 | see comment near end of drbd_worker() */ | |
1845 | spin_unlock_irqrestore(&q->q_lock, flags); | |
1846 | } | |
1847 | ||
1848 | static inline void | |
1849 | drbd_queue_work(struct drbd_work_queue *q, struct drbd_work *w) | |
1850 | { | |
1851 | unsigned long flags; | |
1852 | spin_lock_irqsave(&q->q_lock, flags); | |
1853 | list_add_tail(&w->list, &q->q); | |
1854 | up(&q->s); /* within the spinlock, | |
1855 | see comment near end of drbd_worker() */ | |
1856 | spin_unlock_irqrestore(&q->q_lock, flags); | |
1857 | } | |
1858 | ||
0625ac19 | 1859 | static inline void wake_asender(struct drbd_tconn *tconn) |
b411b363 | 1860 | { |
0625ac19 PR |
1861 | if (test_bit(SIGNAL_ASENDER, &tconn->flags)) |
1862 | force_sig(DRBD_SIG, tconn->asender.task); | |
b411b363 PR |
1863 | } |
1864 | ||
0625ac19 | 1865 | static inline void request_ping(struct drbd_tconn *tconn) |
b411b363 | 1866 | { |
0625ac19 PR |
1867 | set_bit(SEND_PING, &tconn->flags); |
1868 | wake_asender(tconn); | |
b411b363 PR |
1869 | } |
1870 | ||
dba58587 AG |
1871 | extern void *conn_prepare_command(struct drbd_tconn *, struct drbd_socket *); |
1872 | extern void *drbd_prepare_command(struct drbd_conf *, struct drbd_socket *); | |
1873 | extern int conn_send_command(struct drbd_tconn *, struct drbd_socket *, | |
1874 | enum drbd_packet, unsigned int, void *, | |
1875 | unsigned int); | |
1876 | extern int drbd_send_command(struct drbd_conf *, struct drbd_socket *, | |
1877 | enum drbd_packet, unsigned int, void *, | |
1878 | unsigned int); | |
1879 | ||
e307f352 AG |
1880 | extern int drbd_send_ping(struct drbd_tconn *tconn); |
1881 | extern int drbd_send_ping_ack(struct drbd_tconn *tconn); | |
9f5bdc33 AG |
1882 | extern int drbd_send_state_req(struct drbd_conf *, union drbd_state, union drbd_state); |
1883 | extern int conn_send_state_req(struct drbd_tconn *, union drbd_state, union drbd_state); | |
cf29c9d8 | 1884 | |
b411b363 PR |
1885 | static inline void drbd_thread_stop(struct drbd_thread *thi) |
1886 | { | |
81e84650 | 1887 | _drbd_thread_stop(thi, false, true); |
b411b363 PR |
1888 | } |
1889 | ||
1890 | static inline void drbd_thread_stop_nowait(struct drbd_thread *thi) | |
1891 | { | |
81e84650 | 1892 | _drbd_thread_stop(thi, false, false); |
b411b363 PR |
1893 | } |
1894 | ||
1895 | static inline void drbd_thread_restart_nowait(struct drbd_thread *thi) | |
1896 | { | |
81e84650 | 1897 | _drbd_thread_stop(thi, true, false); |
b411b363 PR |
1898 | } |
1899 | ||
1900 | /* counts how many answer packets packets we expect from our peer, | |
1901 | * for either explicit application requests, | |
1902 | * or implicit barrier packets as necessary. | |
1903 | * increased: | |
1904 | * w_send_barrier | |
8554df1c | 1905 | * _req_mod(req, QUEUE_FOR_NET_WRITE or QUEUE_FOR_NET_READ); |
b411b363 PR |
1906 | * it is much easier and equally valid to count what we queue for the |
1907 | * worker, even before it actually was queued or send. | |
1908 | * (drbd_make_request_common; recovery path on read io-error) | |
1909 | * decreased: | |
1910 | * got_BarrierAck (respective tl_clear, tl_clear_barrier) | |
8554df1c | 1911 | * _req_mod(req, DATA_RECEIVED) |
b411b363 | 1912 | * [from receive_DataReply] |
8554df1c | 1913 | * _req_mod(req, WRITE_ACKED_BY_PEER or RECV_ACKED_BY_PEER or NEG_ACKED) |
b411b363 PR |
1914 | * [from got_BlockAck (P_WRITE_ACK, P_RECV_ACK)] |
1915 | * for some reason it is NOT decreased in got_NegAck, | |
1916 | * but in the resulting cleanup code from report_params. | |
1917 | * we should try to remember the reason for that... | |
8554df1c AG |
1918 | * _req_mod(req, SEND_FAILED or SEND_CANCELED) |
1919 | * _req_mod(req, CONNECTION_LOST_WHILE_PENDING) | |
b411b363 PR |
1920 | * [from tl_clear_barrier] |
1921 | */ | |
1922 | static inline void inc_ap_pending(struct drbd_conf *mdev) | |
1923 | { | |
1924 | atomic_inc(&mdev->ap_pending_cnt); | |
1925 | } | |
1926 | ||
49559d87 PR |
1927 | #define ERR_IF_CNT_IS_NEGATIVE(which, func, line) \ |
1928 | if (atomic_read(&mdev->which) < 0) \ | |
b411b363 | 1929 | dev_err(DEV, "in %s:%d: " #which " = %d < 0 !\n", \ |
49559d87 PR |
1930 | func, line, \ |
1931 | atomic_read(&mdev->which)) | |
b411b363 | 1932 | |
49559d87 PR |
1933 | #define dec_ap_pending(mdev) _dec_ap_pending(mdev, __FUNCTION__, __LINE__) |
1934 | static inline void _dec_ap_pending(struct drbd_conf *mdev, const char *func, int line) | |
1935 | { | |
1936 | if (atomic_dec_and_test(&mdev->ap_pending_cnt)) | |
1937 | wake_up(&mdev->misc_wait); | |
1938 | ERR_IF_CNT_IS_NEGATIVE(ap_pending_cnt, func, line); | |
1939 | } | |
b411b363 PR |
1940 | |
1941 | /* counts how many resync-related answers we still expect from the peer | |
1942 | * increase decrease | |
1943 | * C_SYNC_TARGET sends P_RS_DATA_REQUEST (and expects P_RS_DATA_REPLY) | |
25985edc | 1944 | * C_SYNC_SOURCE sends P_RS_DATA_REPLY (and expects P_WRITE_ACK with ID_SYNCER) |
b411b363 PR |
1945 | * (or P_NEG_ACK with ID_SYNCER) |
1946 | */ | |
1947 | static inline void inc_rs_pending(struct drbd_conf *mdev) | |
1948 | { | |
1949 | atomic_inc(&mdev->rs_pending_cnt); | |
1950 | } | |
1951 | ||
49559d87 PR |
1952 | #define dec_rs_pending(mdev) _dec_rs_pending(mdev, __FUNCTION__, __LINE__) |
1953 | static inline void _dec_rs_pending(struct drbd_conf *mdev, const char *func, int line) | |
1954 | { | |
1955 | atomic_dec(&mdev->rs_pending_cnt); | |
1956 | ERR_IF_CNT_IS_NEGATIVE(rs_pending_cnt, func, line); | |
1957 | } | |
b411b363 PR |
1958 | |
1959 | /* counts how many answers we still need to send to the peer. | |
1960 | * increased on | |
1961 | * receive_Data unless protocol A; | |
1962 | * we need to send a P_RECV_ACK (proto B) | |
1963 | * or P_WRITE_ACK (proto C) | |
1964 | * receive_RSDataReply (recv_resync_read) we need to send a P_WRITE_ACK | |
1965 | * receive_DataRequest (receive_RSDataRequest) we need to send back P_DATA | |
1966 | * receive_Barrier_* we need to send a P_BARRIER_ACK | |
1967 | */ | |
1968 | static inline void inc_unacked(struct drbd_conf *mdev) | |
1969 | { | |
1970 | atomic_inc(&mdev->unacked_cnt); | |
1971 | } | |
1972 | ||
49559d87 PR |
1973 | #define dec_unacked(mdev) _dec_unacked(mdev, __FUNCTION__, __LINE__) |
1974 | static inline void _dec_unacked(struct drbd_conf *mdev, const char *func, int line) | |
1975 | { | |
1976 | atomic_dec(&mdev->unacked_cnt); | |
1977 | ERR_IF_CNT_IS_NEGATIVE(unacked_cnt, func, line); | |
1978 | } | |
b411b363 | 1979 | |
49559d87 PR |
1980 | #define sub_unacked(mdev, n) _sub_unacked(mdev, n, __FUNCTION__, __LINE__) |
1981 | static inline void _sub_unacked(struct drbd_conf *mdev, int n, const char *func, int line) | |
1982 | { | |
1983 | atomic_sub(n, &mdev->unacked_cnt); | |
1984 | ERR_IF_CNT_IS_NEGATIVE(unacked_cnt, func, line); | |
1985 | } | |
b411b363 | 1986 | |
b411b363 PR |
1987 | /** |
1988 | * get_ldev() - Increase the ref count on mdev->ldev. Returns 0 if there is no ldev | |
1989 | * @M: DRBD device. | |
1990 | * | |
1991 | * You have to call put_ldev() when finished working with mdev->ldev. | |
1992 | */ | |
1993 | #define get_ldev(M) __cond_lock(local, _get_ldev_if_state(M,D_INCONSISTENT)) | |
1994 | #define get_ldev_if_state(M,MINS) __cond_lock(local, _get_ldev_if_state(M,MINS)) | |
1995 | ||
1996 | static inline void put_ldev(struct drbd_conf *mdev) | |
1997 | { | |
1d7734a0 | 1998 | int i = atomic_dec_return(&mdev->local_cnt); |
9a0d9d03 LE |
1999 | |
2000 | /* This may be called from some endio handler, | |
2001 | * so we must not sleep here. */ | |
2002 | ||
b411b363 | 2003 | __release(local); |
1d7734a0 | 2004 | D_ASSERT(i >= 0); |
e9e6f3ec | 2005 | if (i == 0) { |
82f59cc6 LE |
2006 | if (mdev->state.disk == D_DISKLESS) |
2007 | /* even internal references gone, safe to destroy */ | |
2008 | drbd_ldev_destroy(mdev); | |
e9e6f3ec | 2009 | if (mdev->state.disk == D_FAILED) |
82f59cc6 | 2010 | /* all application IO references gone. */ |
e9e6f3ec | 2011 | drbd_go_diskless(mdev); |
b411b363 | 2012 | wake_up(&mdev->misc_wait); |
e9e6f3ec | 2013 | } |
b411b363 PR |
2014 | } |
2015 | ||
2016 | #ifndef __CHECKER__ | |
2017 | static inline int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins) | |
2018 | { | |
2019 | int io_allowed; | |
2020 | ||
82f59cc6 LE |
2021 | /* never get a reference while D_DISKLESS */ |
2022 | if (mdev->state.disk == D_DISKLESS) | |
2023 | return 0; | |
2024 | ||
b411b363 PR |
2025 | atomic_inc(&mdev->local_cnt); |
2026 | io_allowed = (mdev->state.disk >= mins); | |
2027 | if (!io_allowed) | |
2028 | put_ldev(mdev); | |
2029 | return io_allowed; | |
2030 | } | |
2031 | #else | |
2032 | extern int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins); | |
2033 | #endif | |
2034 | ||
2035 | /* you must have an "get_ldev" reference */ | |
2036 | static inline void drbd_get_syncer_progress(struct drbd_conf *mdev, | |
2037 | unsigned long *bits_left, unsigned int *per_mil_done) | |
2038 | { | |
4b0715f0 LE |
2039 | /* this is to break it at compile time when we change that, in case we |
2040 | * want to support more than (1<<32) bits on a 32bit arch. */ | |
b411b363 PR |
2041 | typecheck(unsigned long, mdev->rs_total); |
2042 | ||
2043 | /* note: both rs_total and rs_left are in bits, i.e. in | |
2044 | * units of BM_BLOCK_SIZE. | |
2045 | * for the percentage, we don't care. */ | |
2046 | ||
439d5953 LE |
2047 | if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T) |
2048 | *bits_left = mdev->ov_left; | |
2049 | else | |
2050 | *bits_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; | |
b411b363 PR |
2051 | /* >> 10 to prevent overflow, |
2052 | * +1 to prevent division by zero */ | |
2053 | if (*bits_left > mdev->rs_total) { | |
2054 | /* doh. maybe a logic bug somewhere. | |
2055 | * may also be just a race condition | |
2056 | * between this and a disconnect during sync. | |
2057 | * for now, just prevent in-kernel buffer overflow. | |
2058 | */ | |
2059 | smp_rmb(); | |
2060 | dev_warn(DEV, "cs:%s rs_left=%lu > rs_total=%lu (rs_failed %lu)\n", | |
2061 | drbd_conn_str(mdev->state.conn), | |
2062 | *bits_left, mdev->rs_total, mdev->rs_failed); | |
2063 | *per_mil_done = 0; | |
2064 | } else { | |
4b0715f0 LE |
2065 | /* Make sure the division happens in long context. |
2066 | * We allow up to one petabyte storage right now, | |
2067 | * at a granularity of 4k per bit that is 2**38 bits. | |
2068 | * After shift right and multiplication by 1000, | |
2069 | * this should still fit easily into a 32bit long, | |
2070 | * so we don't need a 64bit division on 32bit arch. | |
2071 | * Note: currently we don't support such large bitmaps on 32bit | |
2072 | * arch anyways, but no harm done to be prepared for it here. | |
2073 | */ | |
3b7cd457 | 2074 | unsigned int shift = mdev->rs_total > UINT_MAX ? 16 : 10; |
4b0715f0 LE |
2075 | unsigned long left = *bits_left >> shift; |
2076 | unsigned long total = 1UL + (mdev->rs_total >> shift); | |
2077 | unsigned long tmp = 1000UL - left * 1000UL/total; | |
b411b363 PR |
2078 | *per_mil_done = tmp; |
2079 | } | |
2080 | } | |
2081 | ||
2082 | ||
2083 | /* this throttles on-the-fly application requests | |
2084 | * according to max_buffers settings; | |
2085 | * maybe re-implement using semaphores? */ | |
2086 | static inline int drbd_get_max_buffers(struct drbd_conf *mdev) | |
2087 | { | |
44ed167d PR |
2088 | struct net_conf *nc; |
2089 | int mxb; | |
2090 | ||
2091 | rcu_read_lock(); | |
2092 | nc = rcu_dereference(mdev->tconn->net_conf); | |
2093 | mxb = nc ? nc->max_buffers : 1000000; /* arbitrary limit on open requests */ | |
2094 | rcu_read_unlock(); | |
2095 | ||
b411b363 PR |
2096 | return mxb; |
2097 | } | |
2098 | ||
3719094e | 2099 | static inline int drbd_state_is_stable(struct drbd_conf *mdev) |
b411b363 | 2100 | { |
da9fbc27 | 2101 | union drbd_dev_state s = mdev->state; |
b411b363 PR |
2102 | |
2103 | /* DO NOT add a default clause, we want the compiler to warn us | |
2104 | * for any newly introduced state we may have forgotten to add here */ | |
2105 | ||
2106 | switch ((enum drbd_conns)s.conn) { | |
2107 | /* new io only accepted when there is no connection, ... */ | |
2108 | case C_STANDALONE: | |
2109 | case C_WF_CONNECTION: | |
2110 | /* ... or there is a well established connection. */ | |
2111 | case C_CONNECTED: | |
2112 | case C_SYNC_SOURCE: | |
2113 | case C_SYNC_TARGET: | |
2114 | case C_VERIFY_S: | |
2115 | case C_VERIFY_T: | |
2116 | case C_PAUSED_SYNC_S: | |
2117 | case C_PAUSED_SYNC_T: | |
67531718 PR |
2118 | case C_AHEAD: |
2119 | case C_BEHIND: | |
3719094e | 2120 | /* transitional states, IO allowed */ |
b411b363 PR |
2121 | case C_DISCONNECTING: |
2122 | case C_UNCONNECTED: | |
2123 | case C_TIMEOUT: | |
2124 | case C_BROKEN_PIPE: | |
2125 | case C_NETWORK_FAILURE: | |
2126 | case C_PROTOCOL_ERROR: | |
2127 | case C_TEAR_DOWN: | |
2128 | case C_WF_REPORT_PARAMS: | |
2129 | case C_STARTING_SYNC_S: | |
2130 | case C_STARTING_SYNC_T: | |
3719094e PR |
2131 | break; |
2132 | ||
2133 | /* Allow IO in BM exchange states with new protocols */ | |
b411b363 | 2134 | case C_WF_BITMAP_S: |
31890f4a | 2135 | if (mdev->tconn->agreed_pro_version < 96) |
3719094e PR |
2136 | return 0; |
2137 | break; | |
2138 | ||
2139 | /* no new io accepted in these states */ | |
b411b363 PR |
2140 | case C_WF_BITMAP_T: |
2141 | case C_WF_SYNC_UUID: | |
2142 | case C_MASK: | |
2143 | /* not "stable" */ | |
2144 | return 0; | |
2145 | } | |
2146 | ||
2147 | switch ((enum drbd_disk_state)s.disk) { | |
2148 | case D_DISKLESS: | |
2149 | case D_INCONSISTENT: | |
2150 | case D_OUTDATED: | |
2151 | case D_CONSISTENT: | |
2152 | case D_UP_TO_DATE: | |
cdfda633 | 2153 | case D_FAILED: |
b411b363 PR |
2154 | /* disk state is stable as well. */ |
2155 | break; | |
2156 | ||
d942ae44 | 2157 | /* no new io accepted during transitional states */ |
b411b363 | 2158 | case D_ATTACHING: |
b411b363 PR |
2159 | case D_NEGOTIATING: |
2160 | case D_UNKNOWN: | |
2161 | case D_MASK: | |
2162 | /* not "stable" */ | |
2163 | return 0; | |
2164 | } | |
2165 | ||
2166 | return 1; | |
2167 | } | |
2168 | ||
2aebfabb | 2169 | static inline int drbd_suspended(struct drbd_conf *mdev) |
fb22c402 | 2170 | { |
8e0af25f PR |
2171 | struct drbd_tconn *tconn = mdev->tconn; |
2172 | ||
2173 | return tconn->susp || tconn->susp_fen || tconn->susp_nod; | |
fb22c402 PR |
2174 | } |
2175 | ||
1b881ef7 | 2176 | static inline bool may_inc_ap_bio(struct drbd_conf *mdev) |
b411b363 PR |
2177 | { |
2178 | int mxb = drbd_get_max_buffers(mdev); | |
2179 | ||
2aebfabb | 2180 | if (drbd_suspended(mdev)) |
1b881ef7 | 2181 | return false; |
b411b363 | 2182 | if (test_bit(SUSPEND_IO, &mdev->flags)) |
1b881ef7 | 2183 | return false; |
b411b363 PR |
2184 | |
2185 | /* to avoid potential deadlock or bitmap corruption, | |
2186 | * in various places, we only allow new application io | |
2187 | * to start during "stable" states. */ | |
2188 | ||
2189 | /* no new io accepted when attaching or detaching the disk */ | |
3719094e | 2190 | if (!drbd_state_is_stable(mdev)) |
1b881ef7 | 2191 | return false; |
b411b363 PR |
2192 | |
2193 | /* since some older kernels don't have atomic_add_unless, | |
2194 | * and we are within the spinlock anyways, we have this workaround. */ | |
2195 | if (atomic_read(&mdev->ap_bio_cnt) > mxb) | |
1b881ef7 | 2196 | return false; |
b411b363 | 2197 | if (test_bit(BITMAP_IO, &mdev->flags)) |
1b881ef7 AG |
2198 | return false; |
2199 | return true; | |
b411b363 PR |
2200 | } |
2201 | ||
23361cf3 | 2202 | static inline bool inc_ap_bio_cond(struct drbd_conf *mdev) |
b411b363 | 2203 | { |
1b881ef7 | 2204 | bool rv = false; |
8869d683 | 2205 | |
87eeee41 | 2206 | spin_lock_irq(&mdev->tconn->req_lock); |
1b881ef7 | 2207 | rv = may_inc_ap_bio(mdev); |
8869d683 | 2208 | if (rv) |
23361cf3 | 2209 | atomic_inc(&mdev->ap_bio_cnt); |
87eeee41 | 2210 | spin_unlock_irq(&mdev->tconn->req_lock); |
8869d683 PR |
2211 | |
2212 | return rv; | |
2213 | } | |
b411b363 | 2214 | |
23361cf3 | 2215 | static inline void inc_ap_bio(struct drbd_conf *mdev) |
8869d683 | 2216 | { |
b411b363 PR |
2217 | /* we wait here |
2218 | * as long as the device is suspended | |
2219 | * until the bitmap is no longer on the fly during connection | |
d942ae44 | 2220 | * handshake as long as we would exceed the max_buffer limit. |
b411b363 PR |
2221 | * |
2222 | * to avoid races with the reconnect code, | |
2223 | * we need to atomic_inc within the spinlock. */ | |
2224 | ||
23361cf3 | 2225 | wait_event(mdev->misc_wait, inc_ap_bio_cond(mdev)); |
b411b363 PR |
2226 | } |
2227 | ||
2228 | static inline void dec_ap_bio(struct drbd_conf *mdev) | |
2229 | { | |
2230 | int mxb = drbd_get_max_buffers(mdev); | |
2231 | int ap_bio = atomic_dec_return(&mdev->ap_bio_cnt); | |
2232 | ||
2233 | D_ASSERT(ap_bio >= 0); | |
2234 | /* this currently does wake_up for every dec_ap_bio! | |
2235 | * maybe rather introduce some type of hysteresis? | |
2236 | * e.g. (ap_bio == mxb/2 || ap_bio == 0) ? */ | |
2237 | if (ap_bio < mxb) | |
2238 | wake_up(&mdev->misc_wait); | |
2239 | if (ap_bio == 0 && test_bit(BITMAP_IO, &mdev->flags)) { | |
2240 | if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags)) | |
e42325a5 | 2241 | drbd_queue_work(&mdev->tconn->data.work, &mdev->bm_io_work.w); |
b411b363 PR |
2242 | } |
2243 | } | |
2244 | ||
62b0da3a | 2245 | static inline int drbd_set_ed_uuid(struct drbd_conf *mdev, u64 val) |
b411b363 | 2246 | { |
62b0da3a | 2247 | int changed = mdev->ed_uuid != val; |
b411b363 | 2248 | mdev->ed_uuid = val; |
62b0da3a | 2249 | return changed; |
b411b363 PR |
2250 | } |
2251 | ||
b411b363 PR |
2252 | static inline int drbd_queue_order_type(struct drbd_conf *mdev) |
2253 | { | |
2254 | /* sorry, we currently have no working implementation | |
2255 | * of distributed TCQ stuff */ | |
2256 | #ifndef QUEUE_ORDERED_NONE | |
2257 | #define QUEUE_ORDERED_NONE 0 | |
2258 | #endif | |
2259 | return QUEUE_ORDERED_NONE; | |
2260 | } | |
2261 | ||
b411b363 PR |
2262 | static inline void drbd_md_flush(struct drbd_conf *mdev) |
2263 | { | |
2264 | int r; | |
2265 | ||
a8a4e51e | 2266 | if (test_bit(MD_NO_FUA, &mdev->flags)) |
b411b363 PR |
2267 | return; |
2268 | ||
dd3932ed | 2269 | r = blkdev_issue_flush(mdev->ldev->md_bdev, GFP_KERNEL, NULL); |
b411b363 | 2270 | if (r) { |
a8a4e51e | 2271 | set_bit(MD_NO_FUA, &mdev->flags); |
b411b363 PR |
2272 | dev_err(DEV, "meta data flush failed with status %d, disabling md-flushes\n", r); |
2273 | } | |
2274 | } | |
2275 | ||
2276 | #endif |