blob: 2bf3a6ef3684396071f16ff044db1d5e537ee71b [file] [log] [blame]
Philipp Reisnerb411b362009-09-25 16:07:19 -07001/*
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>
Randy Dunlap132cc532009-10-07 19:26:00 +020037#include <linux/ratelimit.h>
Philipp Reisnerb411b362009-09-25 16:07:19 -070038#include <linux/tcp.h>
39#include <linux/mutex.h>
40#include <linux/major.h>
41#include <linux/blkdev.h>
42#include <linux/genhd.h>
43#include <net/tcp.h>
44#include <linux/lru_cache.h>
45
46#ifdef __CHECKER__
47# define __protected_by(x) __attribute__((require_context(x,1,999,"rdwr")))
48# define __protected_read_by(x) __attribute__((require_context(x,1,999,"read")))
49# define __protected_write_by(x) __attribute__((require_context(x,1,999,"write")))
50# define __must_hold(x) __attribute__((context(x,1,1), require_context(x,1,999,"call")))
51#else
52# define __protected_by(x)
53# define __protected_read_by(x)
54# define __protected_write_by(x)
55# define __must_hold(x)
56#endif
57
58#define __no_warn(lock, stmt) do { __acquire(lock); stmt; __release(lock); } while (0)
59
60/* module parameter, defined in drbd_main.c */
61extern unsigned int minor_count;
62extern int disable_sendpage;
63extern int allow_oos;
64extern unsigned int cn_idx;
65
66#ifdef CONFIG_DRBD_FAULT_INJECTION
67extern int enable_faults;
68extern int fault_rate;
69extern int fault_devs;
70#endif
71
72extern char usermode_helper[];
73
74
75#ifndef TRUE
76#define TRUE 1
77#endif
78#ifndef FALSE
79#define FALSE 0
80#endif
81
82/* I don't remember why XCPU ...
83 * This is used to wake the asender,
84 * and to interrupt sending the sending task
85 * on disconnect.
86 */
87#define DRBD_SIG SIGXCPU
88
89/* This is used to stop/restart our threads.
90 * Cannot use SIGTERM nor SIGKILL, since these
91 * are sent out by init on runlevel changes
92 * I choose SIGHUP for now.
93 */
94#define DRBD_SIGKILL SIGHUP
95
96/* All EEs on the free list should have ID_VACANT (== 0)
97 * freshly allocated EEs get !ID_VACANT (== 1)
98 * so if it says "cannot dereference null pointer at adress 0x00000001",
99 * it is most likely one of these :( */
100
101#define ID_IN_SYNC (4711ULL)
102#define ID_OUT_OF_SYNC (4712ULL)
103
104#define ID_SYNCER (-1ULL)
105#define ID_VACANT 0
106#define is_syncer_block_id(id) ((id) == ID_SYNCER)
107
108struct drbd_conf;
109
110
111/* to shorten dev_warn(DEV, "msg"); and relatives statements */
112#define DEV (disk_to_dev(mdev->vdisk))
113
114#define D_ASSERT(exp) if (!(exp)) \
115 dev_err(DEV, "ASSERT( " #exp " ) in %s:%d\n", __FILE__, __LINE__)
116
117#define ERR_IF(exp) if (({ \
118 int _b = (exp) != 0; \
119 if (_b) dev_err(DEV, "%s: (%s) in %s:%d\n", \
120 __func__, #exp, __FILE__, __LINE__); \
121 _b; \
122 }))
123
124/* Defines to control fault insertion */
125enum {
126 DRBD_FAULT_MD_WR = 0, /* meta data write */
127 DRBD_FAULT_MD_RD = 1, /* read */
128 DRBD_FAULT_RS_WR = 2, /* resync */
129 DRBD_FAULT_RS_RD = 3,
130 DRBD_FAULT_DT_WR = 4, /* data */
131 DRBD_FAULT_DT_RD = 5,
132 DRBD_FAULT_DT_RA = 6, /* data read ahead */
133 DRBD_FAULT_BM_ALLOC = 7, /* bitmap allocation */
134 DRBD_FAULT_AL_EE = 8, /* alloc ee */
135
136 DRBD_FAULT_MAX,
137};
138
Philipp Reisnerb411b362009-09-25 16:07:19 -0700139#ifdef CONFIG_DRBD_FAULT_INJECTION
140extern unsigned int
141_drbd_insert_fault(struct drbd_conf *mdev, unsigned int type);
142static inline int
143drbd_insert_fault(struct drbd_conf *mdev, unsigned int type) {
144 return fault_rate &&
145 (enable_faults & (1<<type)) &&
146 _drbd_insert_fault(mdev, type);
147}
148#define FAULT_ACTIVE(_m, _t) (drbd_insert_fault((_m), (_t)))
149
150#else
151#define FAULT_ACTIVE(_m, _t) (0)
152#endif
153
154/* integer division, round _UP_ to the next integer */
155#define div_ceil(A, B) ((A)/(B) + ((A)%(B) ? 1 : 0))
156/* usual integer division */
157#define div_floor(A, B) ((A)/(B))
158
159/* drbd_meta-data.c (still in drbd_main.c) */
160/* 4th incarnation of the disk layout. */
161#define DRBD_MD_MAGIC (DRBD_MAGIC+4)
162
163extern struct drbd_conf **minor_table;
164extern struct ratelimit_state drbd_ratelimit_state;
165
166/* on the wire */
167enum drbd_packets {
168 /* receiver (data socket) */
169 P_DATA = 0x00,
170 P_DATA_REPLY = 0x01, /* Response to P_DATA_REQUEST */
171 P_RS_DATA_REPLY = 0x02, /* Response to P_RS_DATA_REQUEST */
172 P_BARRIER = 0x03,
173 P_BITMAP = 0x04,
174 P_BECOME_SYNC_TARGET = 0x05,
175 P_BECOME_SYNC_SOURCE = 0x06,
176 P_UNPLUG_REMOTE = 0x07, /* Used at various times to hint the peer */
177 P_DATA_REQUEST = 0x08, /* Used to ask for a data block */
178 P_RS_DATA_REQUEST = 0x09, /* Used to ask for a data block for resync */
179 P_SYNC_PARAM = 0x0a,
180 P_PROTOCOL = 0x0b,
181 P_UUIDS = 0x0c,
182 P_SIZES = 0x0d,
183 P_STATE = 0x0e,
184 P_SYNC_UUID = 0x0f,
185 P_AUTH_CHALLENGE = 0x10,
186 P_AUTH_RESPONSE = 0x11,
187 P_STATE_CHG_REQ = 0x12,
188
189 /* asender (meta socket */
190 P_PING = 0x13,
191 P_PING_ACK = 0x14,
192 P_RECV_ACK = 0x15, /* Used in protocol B */
193 P_WRITE_ACK = 0x16, /* Used in protocol C */
194 P_RS_WRITE_ACK = 0x17, /* Is a P_WRITE_ACK, additionally call set_in_sync(). */
195 P_DISCARD_ACK = 0x18, /* Used in proto C, two-primaries conflict detection */
196 P_NEG_ACK = 0x19, /* Sent if local disk is unusable */
197 P_NEG_DREPLY = 0x1a, /* Local disk is broken... */
198 P_NEG_RS_DREPLY = 0x1b, /* Local disk is broken... */
199 P_BARRIER_ACK = 0x1c,
200 P_STATE_CHG_REPLY = 0x1d,
201
202 /* "new" commands, no longer fitting into the ordering scheme above */
203
204 P_OV_REQUEST = 0x1e, /* data socket */
205 P_OV_REPLY = 0x1f,
206 P_OV_RESULT = 0x20, /* meta socket */
207 P_CSUM_RS_REQUEST = 0x21, /* data socket */
208 P_RS_IS_IN_SYNC = 0x22, /* meta socket */
209 P_SYNC_PARAM89 = 0x23, /* data socket, protocol version 89 replacement for P_SYNC_PARAM */
210 P_COMPRESSED_BITMAP = 0x24, /* compressed or otherwise encoded bitmap transfer */
211
212 P_MAX_CMD = 0x25,
213 P_MAY_IGNORE = 0x100, /* Flag to test if (cmd > P_MAY_IGNORE) ... */
214 P_MAX_OPT_CMD = 0x101,
215
216 /* special command ids for handshake */
217
218 P_HAND_SHAKE_M = 0xfff1, /* First Packet on the MetaSock */
219 P_HAND_SHAKE_S = 0xfff2, /* First Packet on the Socket */
220
221 P_HAND_SHAKE = 0xfffe /* FIXED for the next century! */
222};
223
224static inline const char *cmdname(enum drbd_packets cmd)
225{
226 /* THINK may need to become several global tables
227 * when we want to support more than
228 * one PRO_VERSION */
229 static const char *cmdnames[] = {
230 [P_DATA] = "Data",
231 [P_DATA_REPLY] = "DataReply",
232 [P_RS_DATA_REPLY] = "RSDataReply",
233 [P_BARRIER] = "Barrier",
234 [P_BITMAP] = "ReportBitMap",
235 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
236 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
237 [P_UNPLUG_REMOTE] = "UnplugRemote",
238 [P_DATA_REQUEST] = "DataRequest",
239 [P_RS_DATA_REQUEST] = "RSDataRequest",
240 [P_SYNC_PARAM] = "SyncParam",
241 [P_SYNC_PARAM89] = "SyncParam89",
242 [P_PROTOCOL] = "ReportProtocol",
243 [P_UUIDS] = "ReportUUIDs",
244 [P_SIZES] = "ReportSizes",
245 [P_STATE] = "ReportState",
246 [P_SYNC_UUID] = "ReportSyncUUID",
247 [P_AUTH_CHALLENGE] = "AuthChallenge",
248 [P_AUTH_RESPONSE] = "AuthResponse",
249 [P_PING] = "Ping",
250 [P_PING_ACK] = "PingAck",
251 [P_RECV_ACK] = "RecvAck",
252 [P_WRITE_ACK] = "WriteAck",
253 [P_RS_WRITE_ACK] = "RSWriteAck",
254 [P_DISCARD_ACK] = "DiscardAck",
255 [P_NEG_ACK] = "NegAck",
256 [P_NEG_DREPLY] = "NegDReply",
257 [P_NEG_RS_DREPLY] = "NegRSDReply",
258 [P_BARRIER_ACK] = "BarrierAck",
259 [P_STATE_CHG_REQ] = "StateChgRequest",
260 [P_STATE_CHG_REPLY] = "StateChgReply",
261 [P_OV_REQUEST] = "OVRequest",
262 [P_OV_REPLY] = "OVReply",
263 [P_OV_RESULT] = "OVResult",
264 [P_MAX_CMD] = NULL,
265 };
266
267 if (cmd == P_HAND_SHAKE_M)
268 return "HandShakeM";
269 if (cmd == P_HAND_SHAKE_S)
270 return "HandShakeS";
271 if (cmd == P_HAND_SHAKE)
272 return "HandShake";
273 if (cmd >= P_MAX_CMD)
274 return "Unknown";
275 return cmdnames[cmd];
276}
277
278/* for sending/receiving the bitmap,
279 * possibly in some encoding scheme */
280struct bm_xfer_ctx {
281 /* "const"
282 * stores total bits and long words
283 * of the bitmap, so we don't need to
284 * call the accessor functions over and again. */
285 unsigned long bm_bits;
286 unsigned long bm_words;
287 /* during xfer, current position within the bitmap */
288 unsigned long bit_offset;
289 unsigned long word_offset;
290
291 /* statistics; index: (h->command == P_BITMAP) */
292 unsigned packets[2];
293 unsigned bytes[2];
294};
295
296extern void INFO_bm_xfer_stats(struct drbd_conf *mdev,
297 const char *direction, struct bm_xfer_ctx *c);
298
299static inline void bm_xfer_ctx_bit_to_word_offset(struct bm_xfer_ctx *c)
300{
301 /* word_offset counts "native long words" (32 or 64 bit),
302 * aligned at 64 bit.
303 * Encoded packet may end at an unaligned bit offset.
304 * In case a fallback clear text packet is transmitted in
305 * between, we adjust this offset back to the last 64bit
306 * aligned "native long word", which makes coding and decoding
307 * the plain text bitmap much more convenient. */
308#if BITS_PER_LONG == 64
309 c->word_offset = c->bit_offset >> 6;
310#elif BITS_PER_LONG == 32
311 c->word_offset = c->bit_offset >> 5;
312 c->word_offset &= ~(1UL);
313#else
314# error "unsupported BITS_PER_LONG"
315#endif
316}
317
318#ifndef __packed
319#define __packed __attribute__((packed))
320#endif
321
322/* This is the layout for a packet on the wire.
323 * The byteorder is the network byte order.
324 * (except block_id and barrier fields.
325 * these are pointers to local structs
326 * and have no relevance for the partner,
327 * which just echoes them as received.)
328 *
329 * NOTE that the payload starts at a long aligned offset,
330 * regardless of 32 or 64 bit arch!
331 */
332struct p_header {
333 u32 magic;
334 u16 command;
335 u16 length; /* bytes of data after this header */
336 u8 payload[0];
337} __packed;
338/* 8 bytes. packet FIXED for the next century! */
339
340/*
341 * short commands, packets without payload, plain p_header:
342 * P_PING
343 * P_PING_ACK
344 * P_BECOME_SYNC_TARGET
345 * P_BECOME_SYNC_SOURCE
346 * P_UNPLUG_REMOTE
347 */
348
349/*
350 * commands with out-of-struct payload:
351 * P_BITMAP (no additional fields)
352 * P_DATA, P_DATA_REPLY (see p_data)
353 * P_COMPRESSED_BITMAP (see receive_compressed_bitmap)
354 */
355
356/* these defines must not be changed without changing the protocol version */
357#define DP_HARDBARRIER 1
358#define DP_RW_SYNC 2
359#define DP_MAY_SET_IN_SYNC 4
360
361struct p_data {
362 struct p_header head;
363 u64 sector; /* 64 bits sector number */
364 u64 block_id; /* to identify the request in protocol B&C */
365 u32 seq_num;
366 u32 dp_flags;
367} __packed;
368
369/*
370 * commands which share a struct:
371 * p_block_ack:
372 * P_RECV_ACK (proto B), P_WRITE_ACK (proto C),
373 * P_DISCARD_ACK (proto C, two-primaries conflict detection)
374 * p_block_req:
375 * P_DATA_REQUEST, P_RS_DATA_REQUEST
376 */
377struct p_block_ack {
378 struct p_header head;
379 u64 sector;
380 u64 block_id;
381 u32 blksize;
382 u32 seq_num;
383} __packed;
384
385
386struct p_block_req {
387 struct p_header head;
388 u64 sector;
389 u64 block_id;
390 u32 blksize;
391 u32 pad; /* to multiple of 8 Byte */
392} __packed;
393
394/*
395 * commands with their own struct for additional fields:
396 * P_HAND_SHAKE
397 * P_BARRIER
398 * P_BARRIER_ACK
399 * P_SYNC_PARAM
400 * ReportParams
401 */
402
403struct p_handshake {
404 struct p_header head; /* 8 bytes */
405 u32 protocol_min;
406 u32 feature_flags;
407 u32 protocol_max;
408
409 /* should be more than enough for future enhancements
410 * for now, feature_flags and the reserverd array shall be zero.
411 */
412
413 u32 _pad;
414 u64 reserverd[7];
415} __packed;
416/* 80 bytes, FIXED for the next century */
417
418struct p_barrier {
419 struct p_header head;
420 u32 barrier; /* barrier number _handle_ only */
421 u32 pad; /* to multiple of 8 Byte */
422} __packed;
423
424struct p_barrier_ack {
425 struct p_header head;
426 u32 barrier;
427 u32 set_size;
428} __packed;
429
430struct p_rs_param {
431 struct p_header head;
432 u32 rate;
433
434 /* Since protocol version 88 and higher. */
435 char verify_alg[0];
436} __packed;
437
438struct p_rs_param_89 {
439 struct p_header head;
440 u32 rate;
441 /* protocol version 89: */
442 char verify_alg[SHARED_SECRET_MAX];
443 char csums_alg[SHARED_SECRET_MAX];
444} __packed;
445
446struct p_protocol {
447 struct p_header head;
448 u32 protocol;
449 u32 after_sb_0p;
450 u32 after_sb_1p;
451 u32 after_sb_2p;
452 u32 want_lose;
453 u32 two_primaries;
454
455 /* Since protocol version 87 and higher. */
456 char integrity_alg[0];
457
458} __packed;
459
460struct p_uuids {
461 struct p_header head;
462 u64 uuid[UI_EXTENDED_SIZE];
463} __packed;
464
465struct p_rs_uuid {
466 struct p_header head;
467 u64 uuid;
468} __packed;
469
470struct p_sizes {
471 struct p_header head;
472 u64 d_size; /* size of disk */
473 u64 u_size; /* user requested size */
474 u64 c_size; /* current exported size */
475 u32 max_segment_size; /* Maximal size of a BIO */
476 u32 queue_order_type;
477} __packed;
478
479struct p_state {
480 struct p_header head;
481 u32 state;
482} __packed;
483
484struct p_req_state {
485 struct p_header head;
486 u32 mask;
487 u32 val;
488} __packed;
489
490struct p_req_state_reply {
491 struct p_header head;
492 u32 retcode;
493} __packed;
494
495struct p_drbd06_param {
496 u64 size;
497 u32 state;
498 u32 blksize;
499 u32 protocol;
500 u32 version;
501 u32 gen_cnt[5];
502 u32 bit_map_gen[5];
503} __packed;
504
505struct p_discard {
506 struct p_header head;
507 u64 block_id;
508 u32 seq_num;
509 u32 pad;
510} __packed;
511
512/* Valid values for the encoding field.
513 * Bump proto version when changing this. */
514enum drbd_bitmap_code {
515 /* RLE_VLI_Bytes = 0,
516 * and other bit variants had been defined during
517 * algorithm evaluation. */
518 RLE_VLI_Bits = 2,
519};
520
521struct p_compressed_bm {
522 struct p_header head;
523 /* (encoding & 0x0f): actual encoding, see enum drbd_bitmap_code
524 * (encoding & 0x80): polarity (set/unset) of first runlength
525 * ((encoding >> 4) & 0x07): pad_bits, number of trailing zero bits
526 * used to pad up to head.length bytes
527 */
528 u8 encoding;
529
530 u8 code[0];
531} __packed;
532
533/* DCBP: Drbd Compressed Bitmap Packet ... */
534static inline enum drbd_bitmap_code
535DCBP_get_code(struct p_compressed_bm *p)
536{
537 return (enum drbd_bitmap_code)(p->encoding & 0x0f);
538}
539
540static inline void
541DCBP_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
542{
543 BUG_ON(code & ~0xf);
544 p->encoding = (p->encoding & ~0xf) | code;
545}
546
547static inline int
548DCBP_get_start(struct p_compressed_bm *p)
549{
550 return (p->encoding & 0x80) != 0;
551}
552
553static inline void
554DCBP_set_start(struct p_compressed_bm *p, int set)
555{
556 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
557}
558
559static inline int
560DCBP_get_pad_bits(struct p_compressed_bm *p)
561{
562 return (p->encoding >> 4) & 0x7;
563}
564
565static inline void
566DCBP_set_pad_bits(struct p_compressed_bm *p, int n)
567{
568 BUG_ON(n & ~0x7);
569 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
570}
571
572/* one bitmap packet, including the p_header,
573 * should fit within one _architecture independend_ page.
574 * so we need to use the fixed size 4KiB page size
575 * most architechtures have used for a long time.
576 */
577#define BM_PACKET_PAYLOAD_BYTES (4096 - sizeof(struct p_header))
578#define BM_PACKET_WORDS (BM_PACKET_PAYLOAD_BYTES/sizeof(long))
579#define BM_PACKET_VLI_BYTES_MAX (4096 - sizeof(struct p_compressed_bm))
580#if (PAGE_SIZE < 4096)
581/* drbd_send_bitmap / receive_bitmap would break horribly */
582#error "PAGE_SIZE too small"
583#endif
584
585union p_polymorph {
586 struct p_header header;
587 struct p_handshake handshake;
588 struct p_data data;
589 struct p_block_ack block_ack;
590 struct p_barrier barrier;
591 struct p_barrier_ack barrier_ack;
592 struct p_rs_param_89 rs_param_89;
593 struct p_protocol protocol;
594 struct p_sizes sizes;
595 struct p_uuids uuids;
596 struct p_state state;
597 struct p_req_state req_state;
598 struct p_req_state_reply req_state_reply;
599 struct p_block_req block_req;
600} __packed;
601
602/**********************************************************************/
603enum drbd_thread_state {
604 None,
605 Running,
606 Exiting,
607 Restarting
608};
609
610struct drbd_thread {
611 spinlock_t t_lock;
612 struct task_struct *task;
613 struct completion stop;
614 enum drbd_thread_state t_state;
615 int (*function) (struct drbd_thread *);
616 struct drbd_conf *mdev;
617 int reset_cpu_mask;
618};
619
620static inline enum drbd_thread_state get_t_state(struct drbd_thread *thi)
621{
622 /* THINK testing the t_state seems to be uncritical in all cases
623 * (but thread_{start,stop}), so we can read it *without* the lock.
624 * --lge */
625
626 smp_rmb();
627 return thi->t_state;
628}
629
630
631/*
632 * Having this as the first member of a struct provides sort of "inheritance".
633 * "derived" structs can be "drbd_queue_work()"ed.
634 * The callback should know and cast back to the descendant struct.
635 * drbd_request and drbd_epoch_entry are descendants of drbd_work.
636 */
637struct drbd_work;
638typedef int (*drbd_work_cb)(struct drbd_conf *, struct drbd_work *, int cancel);
639struct drbd_work {
640 struct list_head list;
641 drbd_work_cb cb;
642};
643
644struct drbd_tl_epoch;
645struct drbd_request {
646 struct drbd_work w;
647 struct drbd_conf *mdev;
648
649 /* if local IO is not allowed, will be NULL.
650 * if local IO _is_ allowed, holds the locally submitted bio clone,
651 * or, after local IO completion, the ERR_PTR(error).
652 * see drbd_endio_pri(). */
653 struct bio *private_bio;
654
655 struct hlist_node colision;
656 sector_t sector;
657 unsigned int size;
658 unsigned int epoch; /* barrier_nr */
659
660 /* barrier_nr: used to check on "completion" whether this req was in
661 * the current epoch, and we therefore have to close it,
662 * starting a new epoch...
663 */
664
665 /* up to here, the struct layout is identical to drbd_epoch_entry;
666 * we might be able to use that to our advantage... */
667
668 struct list_head tl_requests; /* ring list in the transfer log */
669 struct bio *master_bio; /* master bio pointer */
670 unsigned long rq_state; /* see comments above _req_mod() */
671 int seq_num;
672 unsigned long start_time;
673};
674
675struct drbd_tl_epoch {
676 struct drbd_work w;
677 struct list_head requests; /* requests before */
678 struct drbd_tl_epoch *next; /* pointer to the next barrier */
679 unsigned int br_number; /* the barriers identifier. */
680 int n_req; /* number of requests attached before this barrier */
681};
682
683struct drbd_request;
684
685/* These Tl_epoch_entries may be in one of 6 lists:
686 active_ee .. data packet being written
687 sync_ee .. syncer block being written
688 done_ee .. block written, need to send P_WRITE_ACK
689 read_ee .. [RS]P_DATA_REQUEST being read
690*/
691
692struct drbd_epoch {
693 struct list_head list;
694 unsigned int barrier_nr;
695 atomic_t epoch_size; /* increased on every request added. */
696 atomic_t active; /* increased on every req. added, and dec on every finished. */
697 unsigned long flags;
698};
699
700/* drbd_epoch flag bits */
701enum {
702 DE_BARRIER_IN_NEXT_EPOCH_ISSUED,
703 DE_BARRIER_IN_NEXT_EPOCH_DONE,
704 DE_CONTAINS_A_BARRIER,
705 DE_HAVE_BARRIER_NUMBER,
706 DE_IS_FINISHING,
707};
708
709enum epoch_event {
710 EV_PUT,
711 EV_GOT_BARRIER_NR,
712 EV_BARRIER_DONE,
713 EV_BECAME_LAST,
Philipp Reisnerb411b362009-09-25 16:07:19 -0700714 EV_CLEANUP = 32, /* used as flag */
715};
716
717struct drbd_epoch_entry {
718 struct drbd_work w;
719 struct drbd_conf *mdev;
720 struct bio *private_bio;
721 struct hlist_node colision;
722 sector_t sector;
723 unsigned int size;
724 struct drbd_epoch *epoch;
725
726 /* up to here, the struct layout is identical to drbd_request;
727 * we might be able to use that to our advantage... */
728
729 unsigned int flags;
730 u64 block_id;
731};
732
733struct drbd_wq_barrier {
734 struct drbd_work w;
735 struct completion done;
736};
737
738struct digest_info {
739 int digest_size;
740 void *digest;
741};
742
743/* ee flag bits */
744enum {
745 __EE_CALL_AL_COMPLETE_IO,
746 __EE_CONFLICT_PENDING,
747 __EE_MAY_SET_IN_SYNC,
748 __EE_IS_BARRIER,
749};
750#define EE_CALL_AL_COMPLETE_IO (1<<__EE_CALL_AL_COMPLETE_IO)
751#define EE_CONFLICT_PENDING (1<<__EE_CONFLICT_PENDING)
752#define EE_MAY_SET_IN_SYNC (1<<__EE_MAY_SET_IN_SYNC)
753#define EE_IS_BARRIER (1<<__EE_IS_BARRIER)
754
755/* global flag bits */
756enum {
757 CREATE_BARRIER, /* next P_DATA is preceeded by a P_BARRIER */
758 SIGNAL_ASENDER, /* whether asender wants to be interrupted */
759 SEND_PING, /* whether asender should send a ping asap */
760
761 STOP_SYNC_TIMER, /* tell timer to cancel itself */
762 UNPLUG_QUEUED, /* only relevant with kernel 2.4 */
763 UNPLUG_REMOTE, /* sending a "UnplugRemote" could help */
764 MD_DIRTY, /* current uuids and flags not yet on disk */
765 DISCARD_CONCURRENT, /* Set on one node, cleared on the peer! */
766 USE_DEGR_WFC_T, /* degr-wfc-timeout instead of wfc-timeout. */
767 CLUSTER_ST_CHANGE, /* Cluster wide state change going on... */
768 CL_ST_CHG_SUCCESS,
769 CL_ST_CHG_FAIL,
770 CRASHED_PRIMARY, /* This node was a crashed primary.
771 * Gets cleared when the state.conn
772 * goes into C_CONNECTED state. */
773 WRITE_BM_AFTER_RESYNC, /* A kmalloc() during resync failed */
774 NO_BARRIER_SUPP, /* underlying block device doesn't implement barriers */
775 CONSIDER_RESYNC,
776
777 MD_NO_BARRIER, /* meta data device does not support barriers,
778 so don't even try */
779 SUSPEND_IO, /* suspend application io */
780 BITMAP_IO, /* suspend application io;
781 once no more io in flight, start bitmap io */
782 BITMAP_IO_QUEUED, /* Started bitmap IO */
783 RESYNC_AFTER_NEG, /* Resync after online grow after the attach&negotiate finished. */
784 NET_CONGESTED, /* The data socket is congested */
785
786 CONFIG_PENDING, /* serialization of (re)configuration requests.
787 * if set, also prevents the device from dying */
788 DEVICE_DYING, /* device became unconfigured,
789 * but worker thread is still handling the cleanup.
790 * reconfiguring (nl_disk_conf, nl_net_conf) is dissalowed,
791 * while this is set. */
792 RESIZE_PENDING, /* Size change detected locally, waiting for the response from
793 * the peer, if it changed there as well. */
794};
795
796struct drbd_bitmap; /* opaque for drbd_conf */
797
798/* TODO sort members for performance
799 * MAYBE group them further */
800
801/* THINK maybe we actually want to use the default "event/%s" worker threads
802 * or similar in linux 2.6, which uses per cpu data and threads.
803 *
804 * To be general, this might need a spin_lock member.
805 * For now, please use the mdev->req_lock to protect list_head,
806 * see drbd_queue_work below.
807 */
808struct drbd_work_queue {
809 struct list_head q;
810 struct semaphore s; /* producers up it, worker down()s it */
811 spinlock_t q_lock; /* to protect the list. */
812};
813
814struct drbd_socket {
815 struct drbd_work_queue work;
816 struct mutex mutex;
817 struct socket *socket;
818 /* this way we get our
819 * send/receive buffers off the stack */
820 union p_polymorph sbuf;
821 union p_polymorph rbuf;
822};
823
824struct drbd_md {
825 u64 md_offset; /* sector offset to 'super' block */
826
827 u64 la_size_sect; /* last agreed size, unit sectors */
828 u64 uuid[UI_SIZE];
829 u64 device_uuid;
830 u32 flags;
831 u32 md_size_sect;
832
833 s32 al_offset; /* signed relative sector offset to al area */
834 s32 bm_offset; /* signed relative sector offset to bitmap */
835
836 /* u32 al_nr_extents; important for restoring the AL
837 * is stored into sync_conf.al_extents, which in turn
838 * gets applied to act_log->nr_elements
839 */
840};
841
842/* for sync_conf and other types... */
843#define NL_PACKET(name, number, fields) struct name { fields };
844#define NL_INTEGER(pn,pr,member) int member;
845#define NL_INT64(pn,pr,member) __u64 member;
846#define NL_BIT(pn,pr,member) unsigned member:1;
847#define NL_STRING(pn,pr,member,len) unsigned char member[len]; int member ## _len;
848#include "linux/drbd_nl.h"
849
850struct drbd_backing_dev {
851 struct block_device *backing_bdev;
852 struct block_device *md_bdev;
853 struct file *lo_file;
854 struct file *md_file;
855 struct drbd_md md;
856 struct disk_conf dc; /* The user provided config... */
857 sector_t known_size; /* last known size of that backing device */
858};
859
860struct drbd_md_io {
861 struct drbd_conf *mdev;
862 struct completion event;
863 int error;
864};
865
866struct bm_io_work {
867 struct drbd_work w;
868 char *why;
869 int (*io_fn)(struct drbd_conf *mdev);
870 void (*done)(struct drbd_conf *mdev, int rv);
871};
872
873enum write_ordering_e {
874 WO_none,
875 WO_drain_io,
876 WO_bdev_flush,
877 WO_bio_barrier
878};
879
880struct drbd_conf {
881 /* things that are stored as / read from meta data on disk */
882 unsigned long flags;
883
884 /* configured by drbdsetup */
885 struct net_conf *net_conf; /* protected by get_net_conf() and put_net_conf() */
886 struct syncer_conf sync_conf;
887 struct drbd_backing_dev *ldev __protected_by(local);
888
889 sector_t p_size; /* partner's disk size */
890 struct request_queue *rq_queue;
891 struct block_device *this_bdev;
892 struct gendisk *vdisk;
893
894 struct drbd_socket data; /* data/barrier/cstate/parameter packets */
895 struct drbd_socket meta; /* ping/ack (metadata) packets */
896 int agreed_pro_version; /* actually used protocol version */
897 unsigned long last_received; /* in jiffies, either socket */
898 unsigned int ko_count;
899 struct drbd_work resync_work,
900 unplug_work,
901 md_sync_work;
902 struct timer_list resync_timer;
903 struct timer_list md_sync_timer;
904
905 /* Used after attach while negotiating new disk state. */
906 union drbd_state new_state_tmp;
907
908 union drbd_state state;
909 wait_queue_head_t misc_wait;
910 wait_queue_head_t state_wait; /* upon each state change. */
911 unsigned int send_cnt;
912 unsigned int recv_cnt;
913 unsigned int read_cnt;
914 unsigned int writ_cnt;
915 unsigned int al_writ_cnt;
916 unsigned int bm_writ_cnt;
917 atomic_t ap_bio_cnt; /* Requests we need to complete */
918 atomic_t ap_pending_cnt; /* AP data packets on the wire, ack expected */
919 atomic_t rs_pending_cnt; /* RS request/data packets on the wire */
920 atomic_t unacked_cnt; /* Need to send replys for */
921 atomic_t local_cnt; /* Waiting for local completion */
922 atomic_t net_cnt; /* Users of net_conf */
923 spinlock_t req_lock;
924 struct drbd_tl_epoch *unused_spare_tle; /* for pre-allocation */
925 struct drbd_tl_epoch *newest_tle;
926 struct drbd_tl_epoch *oldest_tle;
927 struct list_head out_of_sequence_requests;
928 struct hlist_head *tl_hash;
929 unsigned int tl_hash_s;
930
931 /* blocks to sync in this run [unit BM_BLOCK_SIZE] */
932 unsigned long rs_total;
933 /* number of sync IOs that failed in this run */
934 unsigned long rs_failed;
935 /* Syncer's start time [unit jiffies] */
936 unsigned long rs_start;
937 /* cumulated time in PausedSyncX state [unit jiffies] */
938 unsigned long rs_paused;
939 /* block not up-to-date at mark [unit BM_BLOCK_SIZE] */
940 unsigned long rs_mark_left;
941 /* marks's time [unit jiffies] */
942 unsigned long rs_mark_time;
943 /* skipped because csum was equeal [unit BM_BLOCK_SIZE] */
944 unsigned long rs_same_csum;
945
946 /* where does the admin want us to start? (sector) */
947 sector_t ov_start_sector;
948 /* where are we now? (sector) */
949 sector_t ov_position;
950 /* Start sector of out of sync range (to merge printk reporting). */
951 sector_t ov_last_oos_start;
952 /* size of out-of-sync range in sectors. */
953 sector_t ov_last_oos_size;
954 unsigned long ov_left; /* in bits */
955 struct crypto_hash *csums_tfm;
956 struct crypto_hash *verify_tfm;
957
958 struct drbd_thread receiver;
959 struct drbd_thread worker;
960 struct drbd_thread asender;
961 struct drbd_bitmap *bitmap;
962 unsigned long bm_resync_fo; /* bit offset for drbd_bm_find_next */
963
964 /* Used to track operations of resync... */
965 struct lru_cache *resync;
966 /* Number of locked elements in resync LRU */
967 unsigned int resync_locked;
968 /* resync extent number waiting for application requests */
969 unsigned int resync_wenr;
970
971 int open_cnt;
972 u64 *p_uuid;
973 struct drbd_epoch *current_epoch;
974 spinlock_t epoch_lock;
975 unsigned int epochs;
976 enum write_ordering_e write_ordering;
977 struct list_head active_ee; /* IO in progress */
978 struct list_head sync_ee; /* IO in progress */
979 struct list_head done_ee; /* send ack */
980 struct list_head read_ee; /* IO in progress */
981 struct list_head net_ee; /* zero-copy network send in progress */
982 struct hlist_head *ee_hash; /* is proteced by req_lock! */
983 unsigned int ee_hash_s;
984
985 /* this one is protected by ee_lock, single thread */
986 struct drbd_epoch_entry *last_write_w_barrier;
987
988 int next_barrier_nr;
989 struct hlist_head *app_reads_hash; /* is proteced by req_lock */
990 struct list_head resync_reads;
991 atomic_t pp_in_use;
992 wait_queue_head_t ee_wait;
993 struct page *md_io_page; /* one page buffer for md_io */
994 struct page *md_io_tmpp; /* for logical_block_size != 512 */
995 struct mutex md_io_mutex; /* protects the md_io_buffer */
996 spinlock_t al_lock;
997 wait_queue_head_t al_wait;
998 struct lru_cache *act_log; /* activity log */
999 unsigned int al_tr_number;
1000 int al_tr_cycle;
1001 int al_tr_pos; /* position of the next transaction in the journal */
1002 struct crypto_hash *cram_hmac_tfm;
1003 struct crypto_hash *integrity_w_tfm; /* to be used by the worker thread */
1004 struct crypto_hash *integrity_r_tfm; /* to be used by the receiver thread */
1005 void *int_dig_out;
1006 void *int_dig_in;
1007 void *int_dig_vv;
1008 wait_queue_head_t seq_wait;
1009 atomic_t packet_seq;
1010 unsigned int peer_seq;
1011 spinlock_t peer_seq_lock;
1012 unsigned int minor;
1013 unsigned long comm_bm_set; /* communicated number of set bits. */
1014 cpumask_var_t cpu_mask;
1015 struct bm_io_work bm_io_work;
1016 u64 ed_uuid; /* UUID of the exposed data */
1017 struct mutex state_mutex;
1018 char congestion_reason; /* Why we where congested... */
1019};
1020
1021static inline struct drbd_conf *minor_to_mdev(unsigned int minor)
1022{
1023 struct drbd_conf *mdev;
1024
1025 mdev = minor < minor_count ? minor_table[minor] : NULL;
1026
1027 return mdev;
1028}
1029
1030static inline unsigned int mdev_to_minor(struct drbd_conf *mdev)
1031{
1032 return mdev->minor;
1033}
1034
1035/* returns 1 if it was successfull,
1036 * returns 0 if there was no data socket.
1037 * so wherever you are going to use the data.socket, e.g. do
1038 * if (!drbd_get_data_sock(mdev))
1039 * return 0;
1040 * CODE();
1041 * drbd_put_data_sock(mdev);
1042 */
1043static inline int drbd_get_data_sock(struct drbd_conf *mdev)
1044{
1045 mutex_lock(&mdev->data.mutex);
1046 /* drbd_disconnect() could have called drbd_free_sock()
1047 * while we were waiting in down()... */
1048 if (unlikely(mdev->data.socket == NULL)) {
1049 mutex_unlock(&mdev->data.mutex);
1050 return 0;
1051 }
1052 return 1;
1053}
1054
1055static inline void drbd_put_data_sock(struct drbd_conf *mdev)
1056{
1057 mutex_unlock(&mdev->data.mutex);
1058}
1059
1060/*
1061 * function declarations
1062 *************************/
1063
1064/* drbd_main.c */
1065
1066enum chg_state_flags {
1067 CS_HARD = 1,
1068 CS_VERBOSE = 2,
1069 CS_WAIT_COMPLETE = 4,
1070 CS_SERIALIZE = 8,
1071 CS_ORDERED = CS_WAIT_COMPLETE + CS_SERIALIZE,
1072};
1073
1074extern void drbd_init_set_defaults(struct drbd_conf *mdev);
1075extern int drbd_change_state(struct drbd_conf *mdev, enum chg_state_flags f,
1076 union drbd_state mask, union drbd_state val);
1077extern void drbd_force_state(struct drbd_conf *, union drbd_state,
1078 union drbd_state);
1079extern int _drbd_request_state(struct drbd_conf *, union drbd_state,
1080 union drbd_state, enum chg_state_flags);
1081extern int __drbd_set_state(struct drbd_conf *, union drbd_state,
1082 enum chg_state_flags, struct completion *done);
1083extern void print_st_err(struct drbd_conf *, union drbd_state,
1084 union drbd_state, int);
1085extern int drbd_thread_start(struct drbd_thread *thi);
1086extern void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait);
1087#ifdef CONFIG_SMP
1088extern void drbd_thread_current_set_cpu(struct drbd_conf *mdev);
1089extern void drbd_calc_cpu_mask(struct drbd_conf *mdev);
1090#else
1091#define drbd_thread_current_set_cpu(A) ({})
1092#define drbd_calc_cpu_mask(A) ({})
1093#endif
1094extern void drbd_free_resources(struct drbd_conf *mdev);
1095extern void tl_release(struct drbd_conf *mdev, unsigned int barrier_nr,
1096 unsigned int set_size);
1097extern void tl_clear(struct drbd_conf *mdev);
1098extern void _tl_add_barrier(struct drbd_conf *, struct drbd_tl_epoch *);
1099extern void drbd_free_sock(struct drbd_conf *mdev);
1100extern int drbd_send(struct drbd_conf *mdev, struct socket *sock,
1101 void *buf, size_t size, unsigned msg_flags);
1102extern int drbd_send_protocol(struct drbd_conf *mdev);
1103extern int drbd_send_uuids(struct drbd_conf *mdev);
1104extern int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev);
1105extern int drbd_send_sync_uuid(struct drbd_conf *mdev, u64 val);
1106extern int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply);
1107extern int _drbd_send_state(struct drbd_conf *mdev);
1108extern int drbd_send_state(struct drbd_conf *mdev);
1109extern int _drbd_send_cmd(struct drbd_conf *mdev, struct socket *sock,
1110 enum drbd_packets cmd, struct p_header *h,
1111 size_t size, unsigned msg_flags);
1112#define USE_DATA_SOCKET 1
1113#define USE_META_SOCKET 0
1114extern int drbd_send_cmd(struct drbd_conf *mdev, int use_data_socket,
1115 enum drbd_packets cmd, struct p_header *h,
1116 size_t size);
1117extern int drbd_send_cmd2(struct drbd_conf *mdev, enum drbd_packets cmd,
1118 char *data, size_t size);
1119extern int drbd_send_sync_param(struct drbd_conf *mdev, struct syncer_conf *sc);
1120extern int drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr,
1121 u32 set_size);
1122extern int drbd_send_ack(struct drbd_conf *mdev, enum drbd_packets cmd,
1123 struct drbd_epoch_entry *e);
1124extern int drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packets cmd,
1125 struct p_block_req *rp);
1126extern int drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packets cmd,
1127 struct p_data *dp);
1128extern int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packets cmd,
1129 sector_t sector, int blksize, u64 block_id);
1130extern int drbd_send_block(struct drbd_conf *mdev, enum drbd_packets cmd,
1131 struct drbd_epoch_entry *e);
1132extern int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req);
1133extern int _drbd_send_barrier(struct drbd_conf *mdev,
1134 struct drbd_tl_epoch *barrier);
1135extern int drbd_send_drequest(struct drbd_conf *mdev, int cmd,
1136 sector_t sector, int size, u64 block_id);
1137extern int drbd_send_drequest_csum(struct drbd_conf *mdev,
1138 sector_t sector,int size,
1139 void *digest, int digest_size,
1140 enum drbd_packets cmd);
1141extern int drbd_send_ov_request(struct drbd_conf *mdev,sector_t sector,int size);
1142
1143extern int drbd_send_bitmap(struct drbd_conf *mdev);
1144extern int _drbd_send_bitmap(struct drbd_conf *mdev);
1145extern int drbd_send_sr_reply(struct drbd_conf *mdev, int retcode);
1146extern void drbd_free_bc(struct drbd_backing_dev *ldev);
1147extern void drbd_mdev_cleanup(struct drbd_conf *mdev);
1148
1149/* drbd_meta-data.c (still in drbd_main.c) */
1150extern void drbd_md_sync(struct drbd_conf *mdev);
1151extern int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev);
1152/* maybe define them below as inline? */
1153extern void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local);
1154extern void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local);
1155extern void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local);
1156extern void _drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local);
1157extern void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local);
1158extern void drbd_md_set_flag(struct drbd_conf *mdev, int flags) __must_hold(local);
1159extern void drbd_md_clear_flag(struct drbd_conf *mdev, int flags)__must_hold(local);
1160extern int drbd_md_test_flag(struct drbd_backing_dev *, int);
1161extern void drbd_md_mark_dirty(struct drbd_conf *mdev);
1162extern void drbd_queue_bitmap_io(struct drbd_conf *mdev,
1163 int (*io_fn)(struct drbd_conf *),
1164 void (*done)(struct drbd_conf *, int),
1165 char *why);
1166extern int drbd_bmio_set_n_write(struct drbd_conf *mdev);
1167extern int drbd_bmio_clear_n_write(struct drbd_conf *mdev);
1168extern int drbd_bitmap_io(struct drbd_conf *mdev, int (*io_fn)(struct drbd_conf *), char *why);
1169
1170
1171/* Meta data layout
1172 We reserve a 128MB Block (4k aligned)
1173 * either at the end of the backing device
1174 * or on a seperate meta data device. */
1175
1176#define MD_RESERVED_SECT (128LU << 11) /* 128 MB, unit sectors */
1177/* The following numbers are sectors */
1178#define MD_AL_OFFSET 8 /* 8 Sectors after start of meta area */
1179#define MD_AL_MAX_SIZE 64 /* = 32 kb LOG ~ 3776 extents ~ 14 GB Storage */
1180/* Allows up to about 3.8TB */
1181#define MD_BM_OFFSET (MD_AL_OFFSET + MD_AL_MAX_SIZE)
1182
1183/* Since the smalles IO unit is usually 512 byte */
1184#define MD_SECTOR_SHIFT 9
1185#define MD_SECTOR_SIZE (1<<MD_SECTOR_SHIFT)
1186
1187/* activity log */
1188#define AL_EXTENTS_PT ((MD_SECTOR_SIZE-12)/8-1) /* 61 ; Extents per 512B sector */
1189#define AL_EXTENT_SHIFT 22 /* One extent represents 4M Storage */
1190#define AL_EXTENT_SIZE (1<<AL_EXTENT_SHIFT)
1191
1192#if BITS_PER_LONG == 32
1193#define LN2_BPL 5
1194#define cpu_to_lel(A) cpu_to_le32(A)
1195#define lel_to_cpu(A) le32_to_cpu(A)
1196#elif BITS_PER_LONG == 64
1197#define LN2_BPL 6
1198#define cpu_to_lel(A) cpu_to_le64(A)
1199#define lel_to_cpu(A) le64_to_cpu(A)
1200#else
1201#error "LN2 of BITS_PER_LONG unknown!"
1202#endif
1203
1204/* resync bitmap */
1205/* 16MB sized 'bitmap extent' to track syncer usage */
1206struct bm_extent {
1207 int rs_left; /* number of bits set (out of sync) in this extent. */
1208 int rs_failed; /* number of failed resync requests in this extent. */
1209 unsigned long flags;
1210 struct lc_element lce;
1211};
1212
1213#define BME_NO_WRITES 0 /* bm_extent.flags: no more requests on this one! */
1214#define BME_LOCKED 1 /* bm_extent.flags: syncer active on this one. */
1215
1216/* drbd_bitmap.c */
1217/*
1218 * We need to store one bit for a block.
1219 * Example: 1GB disk @ 4096 byte blocks ==> we need 32 KB bitmap.
1220 * Bit 0 ==> local node thinks this block is binary identical on both nodes
1221 * Bit 1 ==> local node thinks this block needs to be synced.
1222 */
1223
1224#define BM_BLOCK_SHIFT 12 /* 4k per bit */
1225#define BM_BLOCK_SIZE (1<<BM_BLOCK_SHIFT)
1226/* (9+3) : 512 bytes @ 8 bits; representing 16M storage
1227 * per sector of on disk bitmap */
1228#define BM_EXT_SHIFT (BM_BLOCK_SHIFT + MD_SECTOR_SHIFT + 3) /* = 24 */
1229#define BM_EXT_SIZE (1<<BM_EXT_SHIFT)
1230
1231#if (BM_EXT_SHIFT != 24) || (BM_BLOCK_SHIFT != 12)
1232#error "HAVE YOU FIXED drbdmeta AS WELL??"
1233#endif
1234
1235/* thus many _storage_ sectors are described by one bit */
1236#define BM_SECT_TO_BIT(x) ((x)>>(BM_BLOCK_SHIFT-9))
1237#define BM_BIT_TO_SECT(x) ((sector_t)(x)<<(BM_BLOCK_SHIFT-9))
1238#define BM_SECT_PER_BIT BM_BIT_TO_SECT(1)
1239
1240/* bit to represented kilo byte conversion */
1241#define Bit2KB(bits) ((bits)<<(BM_BLOCK_SHIFT-10))
1242
1243/* in which _bitmap_ extent (resp. sector) the bit for a certain
1244 * _storage_ sector is located in */
1245#define BM_SECT_TO_EXT(x) ((x)>>(BM_EXT_SHIFT-9))
1246
1247/* how much _storage_ sectors we have per bitmap sector */
1248#define BM_EXT_TO_SECT(x) ((sector_t)(x) << (BM_EXT_SHIFT-9))
1249#define BM_SECT_PER_EXT BM_EXT_TO_SECT(1)
1250
1251/* in one sector of the bitmap, we have this many activity_log extents. */
1252#define AL_EXT_PER_BM_SECT (1 << (BM_EXT_SHIFT - AL_EXTENT_SHIFT))
1253#define BM_WORDS_PER_AL_EXT (1 << (AL_EXTENT_SHIFT-BM_BLOCK_SHIFT-LN2_BPL))
1254
1255#define BM_BLOCKS_PER_BM_EXT_B (BM_EXT_SHIFT - BM_BLOCK_SHIFT)
1256#define BM_BLOCKS_PER_BM_EXT_MASK ((1<<BM_BLOCKS_PER_BM_EXT_B) - 1)
1257
1258/* the extent in "PER_EXTENT" below is an activity log extent
1259 * we need that many (long words/bytes) to store the bitmap
1260 * of one AL_EXTENT_SIZE chunk of storage.
1261 * we can store the bitmap for that many AL_EXTENTS within
1262 * one sector of the _on_disk_ bitmap:
1263 * bit 0 bit 37 bit 38 bit (512*8)-1
1264 * ...|........|........|.. // ..|........|
1265 * sect. 0 `296 `304 ^(512*8*8)-1
1266 *
1267#define BM_WORDS_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / BITS_PER_LONG )
1268#define BM_BYTES_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / 8 ) // 128
1269#define BM_EXT_PER_SECT ( 512 / BM_BYTES_PER_EXTENT ) // 4
1270 */
1271
1272#define DRBD_MAX_SECTORS_32 (0xffffffffLU)
1273#define DRBD_MAX_SECTORS_BM \
1274 ((MD_RESERVED_SECT - MD_BM_OFFSET) * (1LL<<(BM_EXT_SHIFT-9)))
1275#if DRBD_MAX_SECTORS_BM < DRBD_MAX_SECTORS_32
1276#define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_BM
1277#define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_BM
Lars Ellenberg36bfc7e2010-01-05 19:33:54 +01001278#elif !defined(CONFIG_LBDAF) && BITS_PER_LONG == 32
Philipp Reisnerb411b362009-09-25 16:07:19 -07001279#define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_32
1280#define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_32
1281#else
1282#define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_BM
1283/* 16 TB in units of sectors */
1284#if BITS_PER_LONG == 32
1285/* adjust by one page worth of bitmap,
1286 * so we won't wrap around in drbd_bm_find_next_bit.
1287 * you should use 64bit OS for that much storage, anyways. */
1288#define DRBD_MAX_SECTORS_FLEX BM_BIT_TO_SECT(0xffff7fff)
1289#else
1290#define DRBD_MAX_SECTORS_FLEX BM_BIT_TO_SECT(0x1LU << 32)
1291#endif
1292#endif
1293
1294/* Sector shift value for the "hash" functions of tl_hash and ee_hash tables.
1295 * With a value of 6 all IO in one 32K block make it to the same slot of the
1296 * hash table. */
1297#define HT_SHIFT 6
1298#define DRBD_MAX_SEGMENT_SIZE (1U<<(9+HT_SHIFT))
1299
1300/* Number of elements in the app_reads_hash */
1301#define APP_R_HSIZE 15
1302
1303extern int drbd_bm_init(struct drbd_conf *mdev);
1304extern int drbd_bm_resize(struct drbd_conf *mdev, sector_t sectors);
1305extern void drbd_bm_cleanup(struct drbd_conf *mdev);
1306extern void drbd_bm_set_all(struct drbd_conf *mdev);
1307extern void drbd_bm_clear_all(struct drbd_conf *mdev);
1308extern int drbd_bm_set_bits(
1309 struct drbd_conf *mdev, unsigned long s, unsigned long e);
1310extern int drbd_bm_clear_bits(
1311 struct drbd_conf *mdev, unsigned long s, unsigned long e);
1312/* bm_set_bits variant for use while holding drbd_bm_lock */
1313extern void _drbd_bm_set_bits(struct drbd_conf *mdev,
1314 const unsigned long s, const unsigned long e);
1315extern int drbd_bm_test_bit(struct drbd_conf *mdev, unsigned long bitnr);
1316extern int drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr);
1317extern int drbd_bm_write_sect(struct drbd_conf *mdev, unsigned long enr) __must_hold(local);
1318extern int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local);
1319extern int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local);
1320extern unsigned long drbd_bm_ALe_set_all(struct drbd_conf *mdev,
1321 unsigned long al_enr);
1322extern size_t drbd_bm_words(struct drbd_conf *mdev);
1323extern unsigned long drbd_bm_bits(struct drbd_conf *mdev);
1324extern sector_t drbd_bm_capacity(struct drbd_conf *mdev);
1325extern unsigned long drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo);
1326/* bm_find_next variants for use while you hold drbd_bm_lock() */
1327extern unsigned long _drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo);
1328extern unsigned long _drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo);
1329extern unsigned long drbd_bm_total_weight(struct drbd_conf *mdev);
1330extern int drbd_bm_rs_done(struct drbd_conf *mdev);
1331/* for receive_bitmap */
1332extern void drbd_bm_merge_lel(struct drbd_conf *mdev, size_t offset,
1333 size_t number, unsigned long *buffer);
1334/* for _drbd_send_bitmap and drbd_bm_write_sect */
1335extern void drbd_bm_get_lel(struct drbd_conf *mdev, size_t offset,
1336 size_t number, unsigned long *buffer);
1337
1338extern void drbd_bm_lock(struct drbd_conf *mdev, char *why);
1339extern void drbd_bm_unlock(struct drbd_conf *mdev);
1340
1341extern int drbd_bm_count_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e);
1342/* drbd_main.c */
1343
1344extern struct kmem_cache *drbd_request_cache;
1345extern struct kmem_cache *drbd_ee_cache; /* epoch entries */
1346extern struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
1347extern struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
1348extern mempool_t *drbd_request_mempool;
1349extern mempool_t *drbd_ee_mempool;
1350
1351extern struct page *drbd_pp_pool; /* drbd's page pool */
1352extern spinlock_t drbd_pp_lock;
1353extern int drbd_pp_vacant;
1354extern wait_queue_head_t drbd_pp_wait;
1355
1356extern rwlock_t global_state_lock;
1357
1358extern struct drbd_conf *drbd_new_device(unsigned int minor);
1359extern void drbd_free_mdev(struct drbd_conf *mdev);
1360
1361extern int proc_details;
1362
1363/* drbd_req */
1364extern int drbd_make_request_26(struct request_queue *q, struct bio *bio);
1365extern int drbd_read_remote(struct drbd_conf *mdev, struct drbd_request *req);
1366extern int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct bio_vec *bvec);
1367extern int is_valid_ar_handle(struct drbd_request *, sector_t);
1368
1369
1370/* drbd_nl.c */
1371extern void drbd_suspend_io(struct drbd_conf *mdev);
1372extern void drbd_resume_io(struct drbd_conf *mdev);
1373extern char *ppsize(char *buf, unsigned long long size);
Philipp Reisnera393db62009-12-22 13:35:52 +01001374extern sector_t drbd_new_dev_size(struct drbd_conf *, struct drbd_backing_dev *, int);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001375enum determine_dev_size { dev_size_error = -1, unchanged = 0, shrunk = 1, grew = 2 };
Philipp Reisnera393db62009-12-22 13:35:52 +01001376extern enum determine_dev_size drbd_determin_dev_size(struct drbd_conf *, int force) __must_hold(local);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001377extern void resync_after_online_grow(struct drbd_conf *);
1378extern void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int) __must_hold(local);
1379extern int drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role,
1380 int force);
1381enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev);
1382extern int drbd_khelper(struct drbd_conf *mdev, char *cmd);
1383
1384/* drbd_worker.c */
1385extern int drbd_worker(struct drbd_thread *thi);
1386extern int drbd_alter_sa(struct drbd_conf *mdev, int na);
1387extern void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side);
1388extern void resume_next_sg(struct drbd_conf *mdev);
1389extern void suspend_other_sg(struct drbd_conf *mdev);
1390extern int drbd_resync_finished(struct drbd_conf *mdev);
1391/* maybe rather drbd_main.c ? */
1392extern int drbd_md_sync_page_io(struct drbd_conf *mdev,
1393 struct drbd_backing_dev *bdev, sector_t sector, int rw);
1394extern void drbd_ov_oos_found(struct drbd_conf*, sector_t, int);
1395
1396static inline void ov_oos_print(struct drbd_conf *mdev)
1397{
1398 if (mdev->ov_last_oos_size) {
1399 dev_err(DEV, "Out of sync: start=%llu, size=%lu (sectors)\n",
1400 (unsigned long long)mdev->ov_last_oos_start,
1401 (unsigned long)mdev->ov_last_oos_size);
1402 }
1403 mdev->ov_last_oos_size=0;
1404}
1405
1406
1407extern void drbd_csum(struct drbd_conf *, struct crypto_hash *, struct bio *, void *);
1408/* worker callbacks */
1409extern int w_req_cancel_conflict(struct drbd_conf *, struct drbd_work *, int);
1410extern int w_read_retry_remote(struct drbd_conf *, struct drbd_work *, int);
1411extern int w_e_end_data_req(struct drbd_conf *, struct drbd_work *, int);
1412extern int w_e_end_rsdata_req(struct drbd_conf *, struct drbd_work *, int);
1413extern int w_e_end_csum_rs_req(struct drbd_conf *, struct drbd_work *, int);
1414extern int w_e_end_ov_reply(struct drbd_conf *, struct drbd_work *, int);
1415extern int w_e_end_ov_req(struct drbd_conf *, struct drbd_work *, int);
1416extern int w_ov_finished(struct drbd_conf *, struct drbd_work *, int);
1417extern int w_resync_inactive(struct drbd_conf *, struct drbd_work *, int);
1418extern int w_resume_next_sg(struct drbd_conf *, struct drbd_work *, int);
1419extern int w_io_error(struct drbd_conf *, struct drbd_work *, int);
1420extern int w_send_write_hint(struct drbd_conf *, struct drbd_work *, int);
1421extern int w_make_resync_request(struct drbd_conf *, struct drbd_work *, int);
1422extern int w_send_dblock(struct drbd_conf *, struct drbd_work *, int);
1423extern int w_send_barrier(struct drbd_conf *, struct drbd_work *, int);
1424extern int w_send_read_req(struct drbd_conf *, struct drbd_work *, int);
1425extern int w_prev_work_done(struct drbd_conf *, struct drbd_work *, int);
1426extern int w_e_reissue(struct drbd_conf *, struct drbd_work *, int);
1427
1428extern void resync_timer_fn(unsigned long data);
1429
1430/* drbd_receiver.c */
1431extern int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list);
1432extern struct drbd_epoch_entry *drbd_alloc_ee(struct drbd_conf *mdev,
1433 u64 id,
1434 sector_t sector,
1435 unsigned int data_size,
1436 gfp_t gfp_mask) __must_hold(local);
1437extern void drbd_free_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e);
1438extern void drbd_wait_ee_list_empty(struct drbd_conf *mdev,
1439 struct list_head *head);
1440extern void _drbd_wait_ee_list_empty(struct drbd_conf *mdev,
1441 struct list_head *head);
1442extern void drbd_set_recv_tcq(struct drbd_conf *mdev, int tcq_enabled);
1443extern void _drbd_clear_done_ee(struct drbd_conf *mdev, struct list_head *to_be_freed);
1444extern void drbd_flush_workqueue(struct drbd_conf *mdev);
1445
1446/* yes, there is kernel_setsockopt, but only since 2.6.18. we don't need to
1447 * mess with get_fs/set_fs, we know we are KERNEL_DS always. */
1448static inline int drbd_setsockopt(struct socket *sock, int level, int optname,
1449 char __user *optval, int optlen)
1450{
1451 int err;
1452 if (level == SOL_SOCKET)
1453 err = sock_setsockopt(sock, level, optname, optval, optlen);
1454 else
1455 err = sock->ops->setsockopt(sock, level, optname, optval,
1456 optlen);
1457 return err;
1458}
1459
1460static inline void drbd_tcp_cork(struct socket *sock)
1461{
1462 int __user val = 1;
1463 (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK,
1464 (char __user *)&val, sizeof(val));
1465}
1466
1467static inline void drbd_tcp_uncork(struct socket *sock)
1468{
1469 int __user val = 0;
1470 (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK,
1471 (char __user *)&val, sizeof(val));
1472}
1473
1474static inline void drbd_tcp_nodelay(struct socket *sock)
1475{
1476 int __user val = 1;
1477 (void) drbd_setsockopt(sock, SOL_TCP, TCP_NODELAY,
1478 (char __user *)&val, sizeof(val));
1479}
1480
1481static inline void drbd_tcp_quickack(struct socket *sock)
1482{
1483 int __user val = 1;
1484 (void) drbd_setsockopt(sock, SOL_TCP, TCP_QUICKACK,
1485 (char __user *)&val, sizeof(val));
1486}
1487
1488void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo);
1489
1490/* drbd_proc.c */
1491extern struct proc_dir_entry *drbd_proc;
Emese Revfy7d4e9d02009-12-14 00:59:30 +01001492extern const struct file_operations drbd_proc_fops;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001493extern const char *drbd_conn_str(enum drbd_conns s);
1494extern const char *drbd_role_str(enum drbd_role s);
1495
1496/* drbd_actlog.c */
1497extern void drbd_al_begin_io(struct drbd_conf *mdev, sector_t sector);
1498extern void drbd_al_complete_io(struct drbd_conf *mdev, sector_t sector);
1499extern void drbd_rs_complete_io(struct drbd_conf *mdev, sector_t sector);
1500extern int drbd_rs_begin_io(struct drbd_conf *mdev, sector_t sector);
1501extern int drbd_try_rs_begin_io(struct drbd_conf *mdev, sector_t sector);
1502extern void drbd_rs_cancel_all(struct drbd_conf *mdev);
1503extern int drbd_rs_del_all(struct drbd_conf *mdev);
1504extern void drbd_rs_failed_io(struct drbd_conf *mdev,
1505 sector_t sector, int size);
1506extern int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *);
1507extern void __drbd_set_in_sync(struct drbd_conf *mdev, sector_t sector,
1508 int size, const char *file, const unsigned int line);
1509#define drbd_set_in_sync(mdev, sector, size) \
1510 __drbd_set_in_sync(mdev, sector, size, __FILE__, __LINE__)
1511extern void __drbd_set_out_of_sync(struct drbd_conf *mdev, sector_t sector,
1512 int size, const char *file, const unsigned int line);
1513#define drbd_set_out_of_sync(mdev, sector, size) \
1514 __drbd_set_out_of_sync(mdev, sector, size, __FILE__, __LINE__)
1515extern void drbd_al_apply_to_bm(struct drbd_conf *mdev);
1516extern void drbd_al_to_on_disk_bm(struct drbd_conf *mdev);
1517extern void drbd_al_shrink(struct drbd_conf *mdev);
1518
1519
1520/* drbd_nl.c */
1521
1522void drbd_nl_cleanup(void);
1523int __init drbd_nl_init(void);
1524void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state);
1525void drbd_bcast_sync_progress(struct drbd_conf *mdev);
1526void drbd_bcast_ee(struct drbd_conf *mdev,
1527 const char *reason, const int dgs,
1528 const char* seen_hash, const char* calc_hash,
1529 const struct drbd_epoch_entry* e);
1530
1531
1532/**
1533 * DOC: DRBD State macros
1534 *
1535 * These macros are used to express state changes in easily readable form.
1536 *
1537 * The NS macros expand to a mask and a value, that can be bit ored onto the
1538 * current state as soon as the spinlock (req_lock) was taken.
1539 *
1540 * The _NS macros are used for state functions that get called with the
1541 * spinlock. These macros expand directly to the new state value.
1542 *
1543 * Besides the basic forms NS() and _NS() additional _?NS[23] are defined
1544 * to express state changes that affect more than one aspect of the state.
1545 *
1546 * E.g. NS2(conn, C_CONNECTED, peer, R_SECONDARY)
1547 * Means that the network connection was established and that the peer
1548 * is in secondary role.
1549 */
1550#define role_MASK R_MASK
1551#define peer_MASK R_MASK
1552#define disk_MASK D_MASK
1553#define pdsk_MASK D_MASK
1554#define conn_MASK C_MASK
1555#define susp_MASK 1
1556#define user_isp_MASK 1
1557#define aftr_isp_MASK 1
1558
1559#define NS(T, S) \
1560 ({ union drbd_state mask; mask.i = 0; mask.T = T##_MASK; mask; }), \
1561 ({ union drbd_state val; val.i = 0; val.T = (S); val; })
1562#define NS2(T1, S1, T2, S2) \
1563 ({ union drbd_state mask; mask.i = 0; mask.T1 = T1##_MASK; \
1564 mask.T2 = T2##_MASK; mask; }), \
1565 ({ union drbd_state val; val.i = 0; val.T1 = (S1); \
1566 val.T2 = (S2); val; })
1567#define NS3(T1, S1, T2, S2, T3, S3) \
1568 ({ union drbd_state mask; mask.i = 0; mask.T1 = T1##_MASK; \
1569 mask.T2 = T2##_MASK; mask.T3 = T3##_MASK; mask; }), \
1570 ({ union drbd_state val; val.i = 0; val.T1 = (S1); \
1571 val.T2 = (S2); val.T3 = (S3); val; })
1572
1573#define _NS(D, T, S) \
1574 D, ({ union drbd_state __ns; __ns.i = D->state.i; __ns.T = (S); __ns; })
1575#define _NS2(D, T1, S1, T2, S2) \
1576 D, ({ union drbd_state __ns; __ns.i = D->state.i; __ns.T1 = (S1); \
1577 __ns.T2 = (S2); __ns; })
1578#define _NS3(D, T1, S1, T2, S2, T3, S3) \
1579 D, ({ union drbd_state __ns; __ns.i = D->state.i; __ns.T1 = (S1); \
1580 __ns.T2 = (S2); __ns.T3 = (S3); __ns; })
1581
1582/*
1583 * inline helper functions
1584 *************************/
1585
1586static inline void drbd_state_lock(struct drbd_conf *mdev)
1587{
1588 wait_event(mdev->misc_wait,
1589 !test_and_set_bit(CLUSTER_ST_CHANGE, &mdev->flags));
1590}
1591
1592static inline void drbd_state_unlock(struct drbd_conf *mdev)
1593{
1594 clear_bit(CLUSTER_ST_CHANGE, &mdev->flags);
1595 wake_up(&mdev->misc_wait);
1596}
1597
1598static inline int _drbd_set_state(struct drbd_conf *mdev,
1599 union drbd_state ns, enum chg_state_flags flags,
1600 struct completion *done)
1601{
1602 int rv;
1603
1604 read_lock(&global_state_lock);
1605 rv = __drbd_set_state(mdev, ns, flags, done);
1606 read_unlock(&global_state_lock);
1607
1608 return rv;
1609}
1610
1611/**
1612 * drbd_request_state() - Reqest a state change
1613 * @mdev: DRBD device.
1614 * @mask: mask of state bits to change.
1615 * @val: value of new state bits.
1616 *
1617 * This is the most graceful way of requesting a state change. It is verbose
1618 * quite verbose in case the state change is not possible, and all those
1619 * state changes are globally serialized.
1620 */
1621static inline int drbd_request_state(struct drbd_conf *mdev,
1622 union drbd_state mask,
1623 union drbd_state val)
1624{
1625 return _drbd_request_state(mdev, mask, val, CS_VERBOSE + CS_ORDERED);
1626}
1627
1628#define __drbd_chk_io_error(m,f) __drbd_chk_io_error_(m,f, __func__)
1629static inline void __drbd_chk_io_error_(struct drbd_conf *mdev, int forcedetach, const char *where)
1630{
1631 switch (mdev->ldev->dc.on_io_error) {
1632 case EP_PASS_ON:
1633 if (!forcedetach) {
1634 if (printk_ratelimit())
1635 dev_err(DEV, "Local IO failed in %s."
1636 "Passing error on...\n", where);
1637 break;
1638 }
1639 /* NOTE fall through to detach case if forcedetach set */
1640 case EP_DETACH:
1641 case EP_CALL_HELPER:
1642 if (mdev->state.disk > D_FAILED) {
1643 _drbd_set_state(_NS(mdev, disk, D_FAILED), CS_HARD, NULL);
1644 dev_err(DEV, "Local IO failed in %s."
1645 "Detaching...\n", where);
1646 }
1647 break;
1648 }
1649}
1650
1651/**
1652 * drbd_chk_io_error: Handle the on_io_error setting, should be called from all io completion handlers
1653 * @mdev: DRBD device.
1654 * @error: Error code passed to the IO completion callback
1655 * @forcedetach: Force detach. I.e. the error happened while accessing the meta data
1656 *
1657 * See also drbd_main.c:after_state_ch() if (os.disk > D_FAILED && ns.disk == D_FAILED)
1658 */
1659#define drbd_chk_io_error(m,e,f) drbd_chk_io_error_(m,e,f, __func__)
1660static inline void drbd_chk_io_error_(struct drbd_conf *mdev,
1661 int error, int forcedetach, const char *where)
1662{
1663 if (error) {
1664 unsigned long flags;
1665 spin_lock_irqsave(&mdev->req_lock, flags);
1666 __drbd_chk_io_error_(mdev, forcedetach, where);
1667 spin_unlock_irqrestore(&mdev->req_lock, flags);
1668 }
1669}
1670
1671
1672/**
1673 * drbd_md_first_sector() - Returns the first sector number of the meta data area
1674 * @bdev: Meta data block device.
1675 *
1676 * BTW, for internal meta data, this happens to be the maximum capacity
1677 * we could agree upon with our peer node.
1678 */
1679static inline sector_t drbd_md_first_sector(struct drbd_backing_dev *bdev)
1680{
1681 switch (bdev->dc.meta_dev_idx) {
1682 case DRBD_MD_INDEX_INTERNAL:
1683 case DRBD_MD_INDEX_FLEX_INT:
1684 return bdev->md.md_offset + bdev->md.bm_offset;
1685 case DRBD_MD_INDEX_FLEX_EXT:
1686 default:
1687 return bdev->md.md_offset;
1688 }
1689}
1690
1691/**
1692 * drbd_md_last_sector() - Return the last sector number of the meta data area
1693 * @bdev: Meta data block device.
1694 */
1695static inline sector_t drbd_md_last_sector(struct drbd_backing_dev *bdev)
1696{
1697 switch (bdev->dc.meta_dev_idx) {
1698 case DRBD_MD_INDEX_INTERNAL:
1699 case DRBD_MD_INDEX_FLEX_INT:
1700 return bdev->md.md_offset + MD_AL_OFFSET - 1;
1701 case DRBD_MD_INDEX_FLEX_EXT:
1702 default:
1703 return bdev->md.md_offset + bdev->md.md_size_sect;
1704 }
1705}
1706
1707/* Returns the number of 512 byte sectors of the device */
1708static inline sector_t drbd_get_capacity(struct block_device *bdev)
1709{
1710 /* return bdev ? get_capacity(bdev->bd_disk) : 0; */
1711 return bdev ? bdev->bd_inode->i_size >> 9 : 0;
1712}
1713
1714/**
1715 * drbd_get_max_capacity() - Returns the capacity we announce to out peer
1716 * @bdev: Meta data block device.
1717 *
1718 * returns the capacity we announce to out peer. we clip ourselves at the
1719 * various MAX_SECTORS, because if we don't, current implementation will
1720 * oops sooner or later
1721 */
1722static inline sector_t drbd_get_max_capacity(struct drbd_backing_dev *bdev)
1723{
1724 sector_t s;
1725 switch (bdev->dc.meta_dev_idx) {
1726 case DRBD_MD_INDEX_INTERNAL:
1727 case DRBD_MD_INDEX_FLEX_INT:
1728 s = drbd_get_capacity(bdev->backing_bdev)
1729 ? min_t(sector_t, DRBD_MAX_SECTORS_FLEX,
1730 drbd_md_first_sector(bdev))
1731 : 0;
1732 break;
1733 case DRBD_MD_INDEX_FLEX_EXT:
1734 s = min_t(sector_t, DRBD_MAX_SECTORS_FLEX,
1735 drbd_get_capacity(bdev->backing_bdev));
1736 /* clip at maximum size the meta device can support */
1737 s = min_t(sector_t, s,
1738 BM_EXT_TO_SECT(bdev->md.md_size_sect
1739 - bdev->md.bm_offset));
1740 break;
1741 default:
1742 s = min_t(sector_t, DRBD_MAX_SECTORS,
1743 drbd_get_capacity(bdev->backing_bdev));
1744 }
1745 return s;
1746}
1747
1748/**
1749 * drbd_md_ss__() - Return the sector number of our meta data super block
1750 * @mdev: DRBD device.
1751 * @bdev: Meta data block device.
1752 */
1753static inline sector_t drbd_md_ss__(struct drbd_conf *mdev,
1754 struct drbd_backing_dev *bdev)
1755{
1756 switch (bdev->dc.meta_dev_idx) {
1757 default: /* external, some index */
1758 return MD_RESERVED_SECT * bdev->dc.meta_dev_idx;
1759 case DRBD_MD_INDEX_INTERNAL:
1760 /* with drbd08, internal meta data is always "flexible" */
1761 case DRBD_MD_INDEX_FLEX_INT:
1762 /* sizeof(struct md_on_disk_07) == 4k
1763 * position: last 4k aligned block of 4k size */
1764 if (!bdev->backing_bdev) {
1765 if (__ratelimit(&drbd_ratelimit_state)) {
1766 dev_err(DEV, "bdev->backing_bdev==NULL\n");
1767 dump_stack();
1768 }
1769 return 0;
1770 }
1771 return (drbd_get_capacity(bdev->backing_bdev) & ~7ULL)
1772 - MD_AL_OFFSET;
1773 case DRBD_MD_INDEX_FLEX_EXT:
1774 return 0;
1775 }
1776}
1777
1778static inline void
1779_drbd_queue_work(struct drbd_work_queue *q, struct drbd_work *w)
1780{
1781 list_add_tail(&w->list, &q->q);
1782 up(&q->s);
1783}
1784
1785static inline void
1786drbd_queue_work_front(struct drbd_work_queue *q, struct drbd_work *w)
1787{
1788 unsigned long flags;
1789 spin_lock_irqsave(&q->q_lock, flags);
1790 list_add(&w->list, &q->q);
1791 up(&q->s); /* within the spinlock,
1792 see comment near end of drbd_worker() */
1793 spin_unlock_irqrestore(&q->q_lock, flags);
1794}
1795
1796static inline void
1797drbd_queue_work(struct drbd_work_queue *q, struct drbd_work *w)
1798{
1799 unsigned long flags;
1800 spin_lock_irqsave(&q->q_lock, flags);
1801 list_add_tail(&w->list, &q->q);
1802 up(&q->s); /* within the spinlock,
1803 see comment near end of drbd_worker() */
1804 spin_unlock_irqrestore(&q->q_lock, flags);
1805}
1806
1807static inline void wake_asender(struct drbd_conf *mdev)
1808{
1809 if (test_bit(SIGNAL_ASENDER, &mdev->flags))
1810 force_sig(DRBD_SIG, mdev->asender.task);
1811}
1812
1813static inline void request_ping(struct drbd_conf *mdev)
1814{
1815 set_bit(SEND_PING, &mdev->flags);
1816 wake_asender(mdev);
1817}
1818
1819static inline int drbd_send_short_cmd(struct drbd_conf *mdev,
1820 enum drbd_packets cmd)
1821{
1822 struct p_header h;
1823 return drbd_send_cmd(mdev, USE_DATA_SOCKET, cmd, &h, sizeof(h));
1824}
1825
1826static inline int drbd_send_ping(struct drbd_conf *mdev)
1827{
1828 struct p_header h;
1829 return drbd_send_cmd(mdev, USE_META_SOCKET, P_PING, &h, sizeof(h));
1830}
1831
1832static inline int drbd_send_ping_ack(struct drbd_conf *mdev)
1833{
1834 struct p_header h;
1835 return drbd_send_cmd(mdev, USE_META_SOCKET, P_PING_ACK, &h, sizeof(h));
1836}
1837
1838static inline void drbd_thread_stop(struct drbd_thread *thi)
1839{
1840 _drbd_thread_stop(thi, FALSE, TRUE);
1841}
1842
1843static inline void drbd_thread_stop_nowait(struct drbd_thread *thi)
1844{
1845 _drbd_thread_stop(thi, FALSE, FALSE);
1846}
1847
1848static inline void drbd_thread_restart_nowait(struct drbd_thread *thi)
1849{
1850 _drbd_thread_stop(thi, TRUE, FALSE);
1851}
1852
1853/* counts how many answer packets packets we expect from our peer,
1854 * for either explicit application requests,
1855 * or implicit barrier packets as necessary.
1856 * increased:
1857 * w_send_barrier
1858 * _req_mod(req, queue_for_net_write or queue_for_net_read);
1859 * it is much easier and equally valid to count what we queue for the
1860 * worker, even before it actually was queued or send.
1861 * (drbd_make_request_common; recovery path on read io-error)
1862 * decreased:
1863 * got_BarrierAck (respective tl_clear, tl_clear_barrier)
1864 * _req_mod(req, data_received)
1865 * [from receive_DataReply]
1866 * _req_mod(req, write_acked_by_peer or recv_acked_by_peer or neg_acked)
1867 * [from got_BlockAck (P_WRITE_ACK, P_RECV_ACK)]
1868 * for some reason it is NOT decreased in got_NegAck,
1869 * but in the resulting cleanup code from report_params.
1870 * we should try to remember the reason for that...
1871 * _req_mod(req, send_failed or send_canceled)
1872 * _req_mod(req, connection_lost_while_pending)
1873 * [from tl_clear_barrier]
1874 */
1875static inline void inc_ap_pending(struct drbd_conf *mdev)
1876{
1877 atomic_inc(&mdev->ap_pending_cnt);
1878}
1879
1880#define ERR_IF_CNT_IS_NEGATIVE(which) \
1881 if (atomic_read(&mdev->which) < 0) \
1882 dev_err(DEV, "in %s:%d: " #which " = %d < 0 !\n", \
1883 __func__ , __LINE__ , \
1884 atomic_read(&mdev->which))
1885
1886#define dec_ap_pending(mdev) do { \
1887 typecheck(struct drbd_conf *, mdev); \
1888 if (atomic_dec_and_test(&mdev->ap_pending_cnt)) \
1889 wake_up(&mdev->misc_wait); \
1890 ERR_IF_CNT_IS_NEGATIVE(ap_pending_cnt); } while (0)
1891
1892/* counts how many resync-related answers we still expect from the peer
1893 * increase decrease
1894 * C_SYNC_TARGET sends P_RS_DATA_REQUEST (and expects P_RS_DATA_REPLY)
1895 * C_SYNC_SOURCE sends P_RS_DATA_REPLY (and expects P_WRITE_ACK whith ID_SYNCER)
1896 * (or P_NEG_ACK with ID_SYNCER)
1897 */
1898static inline void inc_rs_pending(struct drbd_conf *mdev)
1899{
1900 atomic_inc(&mdev->rs_pending_cnt);
1901}
1902
1903#define dec_rs_pending(mdev) do { \
1904 typecheck(struct drbd_conf *, mdev); \
1905 atomic_dec(&mdev->rs_pending_cnt); \
1906 ERR_IF_CNT_IS_NEGATIVE(rs_pending_cnt); } while (0)
1907
1908/* counts how many answers we still need to send to the peer.
1909 * increased on
1910 * receive_Data unless protocol A;
1911 * we need to send a P_RECV_ACK (proto B)
1912 * or P_WRITE_ACK (proto C)
1913 * receive_RSDataReply (recv_resync_read) we need to send a P_WRITE_ACK
1914 * receive_DataRequest (receive_RSDataRequest) we need to send back P_DATA
1915 * receive_Barrier_* we need to send a P_BARRIER_ACK
1916 */
1917static inline void inc_unacked(struct drbd_conf *mdev)
1918{
1919 atomic_inc(&mdev->unacked_cnt);
1920}
1921
1922#define dec_unacked(mdev) do { \
1923 typecheck(struct drbd_conf *, mdev); \
1924 atomic_dec(&mdev->unacked_cnt); \
1925 ERR_IF_CNT_IS_NEGATIVE(unacked_cnt); } while (0)
1926
1927#define sub_unacked(mdev, n) do { \
1928 typecheck(struct drbd_conf *, mdev); \
1929 atomic_sub(n, &mdev->unacked_cnt); \
1930 ERR_IF_CNT_IS_NEGATIVE(unacked_cnt); } while (0)
1931
1932
1933static inline void put_net_conf(struct drbd_conf *mdev)
1934{
1935 if (atomic_dec_and_test(&mdev->net_cnt))
1936 wake_up(&mdev->misc_wait);
1937}
1938
1939/**
1940 * get_net_conf() - Increase ref count on mdev->net_conf; Returns 0 if nothing there
1941 * @mdev: DRBD device.
1942 *
1943 * You have to call put_net_conf() when finished working with mdev->net_conf.
1944 */
1945static inline int get_net_conf(struct drbd_conf *mdev)
1946{
1947 int have_net_conf;
1948
1949 atomic_inc(&mdev->net_cnt);
1950 have_net_conf = mdev->state.conn >= C_UNCONNECTED;
1951 if (!have_net_conf)
1952 put_net_conf(mdev);
1953 return have_net_conf;
1954}
1955
1956/**
1957 * get_ldev() - Increase the ref count on mdev->ldev. Returns 0 if there is no ldev
1958 * @M: DRBD device.
1959 *
1960 * You have to call put_ldev() when finished working with mdev->ldev.
1961 */
1962#define get_ldev(M) __cond_lock(local, _get_ldev_if_state(M,D_INCONSISTENT))
1963#define get_ldev_if_state(M,MINS) __cond_lock(local, _get_ldev_if_state(M,MINS))
1964
1965static inline void put_ldev(struct drbd_conf *mdev)
1966{
1967 __release(local);
1968 if (atomic_dec_and_test(&mdev->local_cnt))
1969 wake_up(&mdev->misc_wait);
1970 D_ASSERT(atomic_read(&mdev->local_cnt) >= 0);
1971}
1972
1973#ifndef __CHECKER__
1974static inline int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins)
1975{
1976 int io_allowed;
1977
1978 atomic_inc(&mdev->local_cnt);
1979 io_allowed = (mdev->state.disk >= mins);
1980 if (!io_allowed)
1981 put_ldev(mdev);
1982 return io_allowed;
1983}
1984#else
1985extern int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins);
1986#endif
1987
1988/* you must have an "get_ldev" reference */
1989static inline void drbd_get_syncer_progress(struct drbd_conf *mdev,
1990 unsigned long *bits_left, unsigned int *per_mil_done)
1991{
1992 /*
1993 * this is to break it at compile time when we change that
1994 * (we may feel 4TB maximum storage per drbd is not enough)
1995 */
1996 typecheck(unsigned long, mdev->rs_total);
1997
1998 /* note: both rs_total and rs_left are in bits, i.e. in
1999 * units of BM_BLOCK_SIZE.
2000 * for the percentage, we don't care. */
2001
2002 *bits_left = drbd_bm_total_weight(mdev) - mdev->rs_failed;
2003 /* >> 10 to prevent overflow,
2004 * +1 to prevent division by zero */
2005 if (*bits_left > mdev->rs_total) {
2006 /* doh. maybe a logic bug somewhere.
2007 * may also be just a race condition
2008 * between this and a disconnect during sync.
2009 * for now, just prevent in-kernel buffer overflow.
2010 */
2011 smp_rmb();
2012 dev_warn(DEV, "cs:%s rs_left=%lu > rs_total=%lu (rs_failed %lu)\n",
2013 drbd_conn_str(mdev->state.conn),
2014 *bits_left, mdev->rs_total, mdev->rs_failed);
2015 *per_mil_done = 0;
2016 } else {
2017 /* make sure the calculation happens in long context */
2018 unsigned long tmp = 1000UL -
2019 (*bits_left >> 10)*1000UL
2020 / ((mdev->rs_total >> 10) + 1UL);
2021 *per_mil_done = tmp;
2022 }
2023}
2024
2025
2026/* this throttles on-the-fly application requests
2027 * according to max_buffers settings;
2028 * maybe re-implement using semaphores? */
2029static inline int drbd_get_max_buffers(struct drbd_conf *mdev)
2030{
2031 int mxb = 1000000; /* arbitrary limit on open requests */
2032 if (get_net_conf(mdev)) {
2033 mxb = mdev->net_conf->max_buffers;
2034 put_net_conf(mdev);
2035 }
2036 return mxb;
2037}
2038
2039static inline int drbd_state_is_stable(union drbd_state s)
2040{
2041
2042 /* DO NOT add a default clause, we want the compiler to warn us
2043 * for any newly introduced state we may have forgotten to add here */
2044
2045 switch ((enum drbd_conns)s.conn) {
2046 /* new io only accepted when there is no connection, ... */
2047 case C_STANDALONE:
2048 case C_WF_CONNECTION:
2049 /* ... or there is a well established connection. */
2050 case C_CONNECTED:
2051 case C_SYNC_SOURCE:
2052 case C_SYNC_TARGET:
2053 case C_VERIFY_S:
2054 case C_VERIFY_T:
2055 case C_PAUSED_SYNC_S:
2056 case C_PAUSED_SYNC_T:
2057 /* maybe stable, look at the disk state */
2058 break;
2059
2060 /* no new io accepted during tansitional states
2061 * like handshake or teardown */
2062 case C_DISCONNECTING:
2063 case C_UNCONNECTED:
2064 case C_TIMEOUT:
2065 case C_BROKEN_PIPE:
2066 case C_NETWORK_FAILURE:
2067 case C_PROTOCOL_ERROR:
2068 case C_TEAR_DOWN:
2069 case C_WF_REPORT_PARAMS:
2070 case C_STARTING_SYNC_S:
2071 case C_STARTING_SYNC_T:
2072 case C_WF_BITMAP_S:
2073 case C_WF_BITMAP_T:
2074 case C_WF_SYNC_UUID:
2075 case C_MASK:
2076 /* not "stable" */
2077 return 0;
2078 }
2079
2080 switch ((enum drbd_disk_state)s.disk) {
2081 case D_DISKLESS:
2082 case D_INCONSISTENT:
2083 case D_OUTDATED:
2084 case D_CONSISTENT:
2085 case D_UP_TO_DATE:
2086 /* disk state is stable as well. */
2087 break;
2088
2089 /* no new io accepted during tansitional states */
2090 case D_ATTACHING:
2091 case D_FAILED:
2092 case D_NEGOTIATING:
2093 case D_UNKNOWN:
2094 case D_MASK:
2095 /* not "stable" */
2096 return 0;
2097 }
2098
2099 return 1;
2100}
2101
2102static inline int __inc_ap_bio_cond(struct drbd_conf *mdev)
2103{
2104 int mxb = drbd_get_max_buffers(mdev);
2105
2106 if (mdev->state.susp)
2107 return 0;
2108 if (test_bit(SUSPEND_IO, &mdev->flags))
2109 return 0;
2110
2111 /* to avoid potential deadlock or bitmap corruption,
2112 * in various places, we only allow new application io
2113 * to start during "stable" states. */
2114
2115 /* no new io accepted when attaching or detaching the disk */
2116 if (!drbd_state_is_stable(mdev->state))
2117 return 0;
2118
2119 /* since some older kernels don't have atomic_add_unless,
2120 * and we are within the spinlock anyways, we have this workaround. */
2121 if (atomic_read(&mdev->ap_bio_cnt) > mxb)
2122 return 0;
2123 if (test_bit(BITMAP_IO, &mdev->flags))
2124 return 0;
2125 return 1;
2126}
2127
2128/* I'd like to use wait_event_lock_irq,
2129 * but I'm not sure when it got introduced,
2130 * and not sure when it has 3 or 4 arguments */
2131static inline void inc_ap_bio(struct drbd_conf *mdev, int one_or_two)
2132{
2133 /* compare with after_state_ch,
2134 * os.conn != C_WF_BITMAP_S && ns.conn == C_WF_BITMAP_S */
2135 DEFINE_WAIT(wait);
2136
2137 /* we wait here
2138 * as long as the device is suspended
2139 * until the bitmap is no longer on the fly during connection
2140 * handshake as long as we would exeed the max_buffer limit.
2141 *
2142 * to avoid races with the reconnect code,
2143 * we need to atomic_inc within the spinlock. */
2144
2145 spin_lock_irq(&mdev->req_lock);
2146 while (!__inc_ap_bio_cond(mdev)) {
2147 prepare_to_wait(&mdev->misc_wait, &wait, TASK_UNINTERRUPTIBLE);
2148 spin_unlock_irq(&mdev->req_lock);
2149 schedule();
2150 finish_wait(&mdev->misc_wait, &wait);
2151 spin_lock_irq(&mdev->req_lock);
2152 }
2153 atomic_add(one_or_two, &mdev->ap_bio_cnt);
2154 spin_unlock_irq(&mdev->req_lock);
2155}
2156
2157static inline void dec_ap_bio(struct drbd_conf *mdev)
2158{
2159 int mxb = drbd_get_max_buffers(mdev);
2160 int ap_bio = atomic_dec_return(&mdev->ap_bio_cnt);
2161
2162 D_ASSERT(ap_bio >= 0);
2163 /* this currently does wake_up for every dec_ap_bio!
2164 * maybe rather introduce some type of hysteresis?
2165 * e.g. (ap_bio == mxb/2 || ap_bio == 0) ? */
2166 if (ap_bio < mxb)
2167 wake_up(&mdev->misc_wait);
2168 if (ap_bio == 0 && test_bit(BITMAP_IO, &mdev->flags)) {
2169 if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags))
2170 drbd_queue_work(&mdev->data.work, &mdev->bm_io_work.w);
2171 }
2172}
2173
2174static inline void drbd_set_ed_uuid(struct drbd_conf *mdev, u64 val)
2175{
2176 mdev->ed_uuid = val;
2177}
2178
2179static inline int seq_cmp(u32 a, u32 b)
2180{
2181 /* we assume wrap around at 32bit.
2182 * for wrap around at 24bit (old atomic_t),
2183 * we'd have to
2184 * a <<= 8; b <<= 8;
2185 */
2186 return (s32)(a) - (s32)(b);
2187}
2188#define seq_lt(a, b) (seq_cmp((a), (b)) < 0)
2189#define seq_gt(a, b) (seq_cmp((a), (b)) > 0)
2190#define seq_ge(a, b) (seq_cmp((a), (b)) >= 0)
2191#define seq_le(a, b) (seq_cmp((a), (b)) <= 0)
2192/* CAUTION: please no side effects in arguments! */
2193#define seq_max(a, b) ((u32)(seq_gt((a), (b)) ? (a) : (b)))
2194
2195static inline void update_peer_seq(struct drbd_conf *mdev, unsigned int new_seq)
2196{
2197 unsigned int m;
2198 spin_lock(&mdev->peer_seq_lock);
2199 m = seq_max(mdev->peer_seq, new_seq);
2200 mdev->peer_seq = m;
2201 spin_unlock(&mdev->peer_seq_lock);
2202 if (m == new_seq)
2203 wake_up(&mdev->seq_wait);
2204}
2205
2206static inline void drbd_update_congested(struct drbd_conf *mdev)
2207{
2208 struct sock *sk = mdev->data.socket->sk;
2209 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
2210 set_bit(NET_CONGESTED, &mdev->flags);
2211}
2212
2213static inline int drbd_queue_order_type(struct drbd_conf *mdev)
2214{
2215 /* sorry, we currently have no working implementation
2216 * of distributed TCQ stuff */
2217#ifndef QUEUE_ORDERED_NONE
2218#define QUEUE_ORDERED_NONE 0
2219#endif
2220 return QUEUE_ORDERED_NONE;
2221}
2222
2223static inline void drbd_blk_run_queue(struct request_queue *q)
2224{
2225 if (q && q->unplug_fn)
2226 q->unplug_fn(q);
2227}
2228
2229static inline void drbd_kick_lo(struct drbd_conf *mdev)
2230{
2231 if (get_ldev(mdev)) {
2232 drbd_blk_run_queue(bdev_get_queue(mdev->ldev->backing_bdev));
2233 put_ldev(mdev);
2234 }
2235}
2236
2237static inline void drbd_md_flush(struct drbd_conf *mdev)
2238{
2239 int r;
2240
2241 if (test_bit(MD_NO_BARRIER, &mdev->flags))
2242 return;
2243
2244 r = blkdev_issue_flush(mdev->ldev->md_bdev, NULL);
2245 if (r) {
2246 set_bit(MD_NO_BARRIER, &mdev->flags);
2247 dev_err(DEV, "meta data flush failed with status %d, disabling md-flushes\n", r);
2248 }
2249}
2250
2251#endif