Loading...
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>
37#include <linux/ratelimit.h>
38#include <linux/tcp.h>
39#include <linux/mutex.h>
40#include <linux/major.h>
41#include <linux/blkdev.h>
42#include <linux/genhd.h>
43#include <net/tcp.h>
44#include <linux/lru_cache.h>
45#include <linux/prefetch.h>
46
47#ifdef __CHECKER__
48# define __protected_by(x) __attribute__((require_context(x,1,999,"rdwr")))
49# define __protected_read_by(x) __attribute__((require_context(x,1,999,"read")))
50# define __protected_write_by(x) __attribute__((require_context(x,1,999,"write")))
51# define __must_hold(x) __attribute__((context(x,1,1), require_context(x,1,999,"call")))
52#else
53# define __protected_by(x)
54# define __protected_read_by(x)
55# define __protected_write_by(x)
56# define __must_hold(x)
57#endif
58
59#define __no_warn(lock, stmt) do { __acquire(lock); stmt; __release(lock); } while (0)
60
61/* module parameter, defined in drbd_main.c */
62extern unsigned int minor_count;
63extern int disable_sendpage;
64extern int allow_oos;
65extern unsigned int cn_idx;
66
67#ifdef CONFIG_DRBD_FAULT_INJECTION
68extern int enable_faults;
69extern int fault_rate;
70extern int fault_devs;
71#endif
72
73extern char usermode_helper[];
74
75
76/* I don't remember why XCPU ...
77 * This is used to wake the asender,
78 * and to interrupt sending the sending task
79 * on disconnect.
80 */
81#define DRBD_SIG SIGXCPU
82
83/* This is used to stop/restart our threads.
84 * Cannot use SIGTERM nor SIGKILL, since these
85 * are sent out by init on runlevel changes
86 * I choose SIGHUP for now.
87 */
88#define DRBD_SIGKILL SIGHUP
89
90/* All EEs on the free list should have ID_VACANT (== 0)
91 * freshly allocated EEs get !ID_VACANT (== 1)
92 * so if it says "cannot dereference null pointer at address 0x00000001",
93 * it is most likely one of these :( */
94
95#define ID_IN_SYNC (4711ULL)
96#define ID_OUT_OF_SYNC (4712ULL)
97
98#define ID_SYNCER (-1ULL)
99#define ID_VACANT 0
100#define is_syncer_block_id(id) ((id) == ID_SYNCER)
101#define UUID_NEW_BM_OFFSET ((u64)0x0001000000000000ULL)
102
103struct drbd_conf;
104
105
106/* to shorten dev_warn(DEV, "msg"); and relatives statements */
107#define DEV (disk_to_dev(mdev->vdisk))
108
109#define D_ASSERT(exp) if (!(exp)) \
110 dev_err(DEV, "ASSERT( " #exp " ) in %s:%d\n", __FILE__, __LINE__)
111
112#define ERR_IF(exp) if (({ \
113 int _b = (exp) != 0; \
114 if (_b) dev_err(DEV, "ASSERT FAILED: %s: (%s) in %s:%d\n", \
115 __func__, #exp, __FILE__, __LINE__); \
116 _b; \
117 }))
118
119/* Defines to control fault insertion */
120enum {
121 DRBD_FAULT_MD_WR = 0, /* meta data write */
122 DRBD_FAULT_MD_RD = 1, /* read */
123 DRBD_FAULT_RS_WR = 2, /* resync */
124 DRBD_FAULT_RS_RD = 3,
125 DRBD_FAULT_DT_WR = 4, /* data */
126 DRBD_FAULT_DT_RD = 5,
127 DRBD_FAULT_DT_RA = 6, /* data read ahead */
128 DRBD_FAULT_BM_ALLOC = 7, /* bitmap allocation */
129 DRBD_FAULT_AL_EE = 8, /* alloc ee */
130 DRBD_FAULT_RECEIVE = 9, /* Changes some bytes upon receiving a [rs]data block */
131
132 DRBD_FAULT_MAX,
133};
134
135extern unsigned int
136_drbd_insert_fault(struct drbd_conf *mdev, unsigned int type);
137
138static inline int
139drbd_insert_fault(struct drbd_conf *mdev, unsigned int type) {
140#ifdef CONFIG_DRBD_FAULT_INJECTION
141 return fault_rate &&
142 (enable_faults & (1<<type)) &&
143 _drbd_insert_fault(mdev, type);
144#else
145 return 0;
146#endif
147}
148
149/* integer division, round _UP_ to the next integer */
150#define div_ceil(A, B) ((A)/(B) + ((A)%(B) ? 1 : 0))
151/* usual integer division */
152#define div_floor(A, B) ((A)/(B))
153
154/* drbd_meta-data.c (still in drbd_main.c) */
155/* 4th incarnation of the disk layout. */
156#define DRBD_MD_MAGIC (DRBD_MAGIC+4)
157
158extern struct drbd_conf **minor_table;
159extern struct ratelimit_state drbd_ratelimit_state;
160
161/* on the wire */
162enum drbd_packets {
163 /* receiver (data socket) */
164 P_DATA = 0x00,
165 P_DATA_REPLY = 0x01, /* Response to P_DATA_REQUEST */
166 P_RS_DATA_REPLY = 0x02, /* Response to P_RS_DATA_REQUEST */
167 P_BARRIER = 0x03,
168 P_BITMAP = 0x04,
169 P_BECOME_SYNC_TARGET = 0x05,
170 P_BECOME_SYNC_SOURCE = 0x06,
171 P_UNPLUG_REMOTE = 0x07, /* Used at various times to hint the peer */
172 P_DATA_REQUEST = 0x08, /* Used to ask for a data block */
173 P_RS_DATA_REQUEST = 0x09, /* Used to ask for a data block for resync */
174 P_SYNC_PARAM = 0x0a,
175 P_PROTOCOL = 0x0b,
176 P_UUIDS = 0x0c,
177 P_SIZES = 0x0d,
178 P_STATE = 0x0e,
179 P_SYNC_UUID = 0x0f,
180 P_AUTH_CHALLENGE = 0x10,
181 P_AUTH_RESPONSE = 0x11,
182 P_STATE_CHG_REQ = 0x12,
183
184 /* asender (meta socket */
185 P_PING = 0x13,
186 P_PING_ACK = 0x14,
187 P_RECV_ACK = 0x15, /* Used in protocol B */
188 P_WRITE_ACK = 0x16, /* Used in protocol C */
189 P_RS_WRITE_ACK = 0x17, /* Is a P_WRITE_ACK, additionally call set_in_sync(). */
190 P_DISCARD_ACK = 0x18, /* Used in proto C, two-primaries conflict detection */
191 P_NEG_ACK = 0x19, /* Sent if local disk is unusable */
192 P_NEG_DREPLY = 0x1a, /* Local disk is broken... */
193 P_NEG_RS_DREPLY = 0x1b, /* Local disk is broken... */
194 P_BARRIER_ACK = 0x1c,
195 P_STATE_CHG_REPLY = 0x1d,
196
197 /* "new" commands, no longer fitting into the ordering scheme above */
198
199 P_OV_REQUEST = 0x1e, /* data socket */
200 P_OV_REPLY = 0x1f,
201 P_OV_RESULT = 0x20, /* meta socket */
202 P_CSUM_RS_REQUEST = 0x21, /* data socket */
203 P_RS_IS_IN_SYNC = 0x22, /* meta socket */
204 P_SYNC_PARAM89 = 0x23, /* data socket, protocol version 89 replacement for P_SYNC_PARAM */
205 P_COMPRESSED_BITMAP = 0x24, /* compressed or otherwise encoded bitmap transfer */
206 /* P_CKPT_FENCE_REQ = 0x25, * currently reserved for protocol D */
207 /* P_CKPT_DISABLE_REQ = 0x26, * currently reserved for protocol D */
208 P_DELAY_PROBE = 0x27, /* is used on BOTH sockets */
209 P_OUT_OF_SYNC = 0x28, /* Mark as out of sync (Outrunning), data socket */
210 P_RS_CANCEL = 0x29, /* meta: Used to cancel RS_DATA_REQUEST packet by SyncSource */
211
212 P_MAX_CMD = 0x2A,
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_CSUM_RS_REQUEST] = "CsumRSRequest",
265 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
266 [P_COMPRESSED_BITMAP] = "CBitmap",
267 [P_DELAY_PROBE] = "DelayProbe",
268 [P_OUT_OF_SYNC] = "OutOfSync",
269 [P_MAX_CMD] = NULL,
270 };
271
272 if (cmd == P_HAND_SHAKE_M)
273 return "HandShakeM";
274 if (cmd == P_HAND_SHAKE_S)
275 return "HandShakeS";
276 if (cmd == P_HAND_SHAKE)
277 return "HandShake";
278 if (cmd >= P_MAX_CMD)
279 return "Unknown";
280 return cmdnames[cmd];
281}
282
283/* for sending/receiving the bitmap,
284 * possibly in some encoding scheme */
285struct bm_xfer_ctx {
286 /* "const"
287 * stores total bits and long words
288 * of the bitmap, so we don't need to
289 * call the accessor functions over and again. */
290 unsigned long bm_bits;
291 unsigned long bm_words;
292 /* during xfer, current position within the bitmap */
293 unsigned long bit_offset;
294 unsigned long word_offset;
295
296 /* statistics; index: (h->command == P_BITMAP) */
297 unsigned packets[2];
298 unsigned bytes[2];
299};
300
301extern void INFO_bm_xfer_stats(struct drbd_conf *mdev,
302 const char *direction, struct bm_xfer_ctx *c);
303
304static inline void bm_xfer_ctx_bit_to_word_offset(struct bm_xfer_ctx *c)
305{
306 /* word_offset counts "native long words" (32 or 64 bit),
307 * aligned at 64 bit.
308 * Encoded packet may end at an unaligned bit offset.
309 * In case a fallback clear text packet is transmitted in
310 * between, we adjust this offset back to the last 64bit
311 * aligned "native long word", which makes coding and decoding
312 * the plain text bitmap much more convenient. */
313#if BITS_PER_LONG == 64
314 c->word_offset = c->bit_offset >> 6;
315#elif BITS_PER_LONG == 32
316 c->word_offset = c->bit_offset >> 5;
317 c->word_offset &= ~(1UL);
318#else
319# error "unsupported BITS_PER_LONG"
320#endif
321}
322
323#ifndef __packed
324#define __packed __attribute__((packed))
325#endif
326
327/* This is the layout for a packet on the wire.
328 * The byteorder is the network byte order.
329 * (except block_id and barrier fields.
330 * these are pointers to local structs
331 * and have no relevance for the partner,
332 * which just echoes them as received.)
333 *
334 * NOTE that the payload starts at a long aligned offset,
335 * regardless of 32 or 64 bit arch!
336 */
337struct p_header80 {
338 u32 magic;
339 u16 command;
340 u16 length; /* bytes of data after this header */
341 u8 payload[0];
342} __packed;
343
344/* Header for big packets, Used for data packets exceeding 64kB */
345struct p_header95 {
346 u16 magic; /* use DRBD_MAGIC_BIG here */
347 u16 command;
348 u32 length; /* Use only 24 bits of that. Ignore the highest 8 bit. */
349 u8 payload[0];
350} __packed;
351
352union p_header {
353 struct p_header80 h80;
354 struct p_header95 h95;
355};
356
357/*
358 * short commands, packets without payload, plain p_header:
359 * P_PING
360 * P_PING_ACK
361 * P_BECOME_SYNC_TARGET
362 * P_BECOME_SYNC_SOURCE
363 * P_UNPLUG_REMOTE
364 */
365
366/*
367 * commands with out-of-struct payload:
368 * P_BITMAP (no additional fields)
369 * P_DATA, P_DATA_REPLY (see p_data)
370 * P_COMPRESSED_BITMAP (see receive_compressed_bitmap)
371 */
372
373/* these defines must not be changed without changing the protocol version */
374#define DP_HARDBARRIER 1 /* depricated */
375#define DP_RW_SYNC 2 /* equals REQ_SYNC */
376#define DP_MAY_SET_IN_SYNC 4
377#define DP_UNPLUG 8 /* not used anymore */
378#define DP_FUA 16 /* equals REQ_FUA */
379#define DP_FLUSH 32 /* equals REQ_FLUSH */
380#define DP_DISCARD 64 /* equals REQ_DISCARD */
381
382struct p_data {
383 union p_header head;
384 u64 sector; /* 64 bits sector number */
385 u64 block_id; /* to identify the request in protocol B&C */
386 u32 seq_num;
387 u32 dp_flags;
388} __packed;
389
390/*
391 * commands which share a struct:
392 * p_block_ack:
393 * P_RECV_ACK (proto B), P_WRITE_ACK (proto C),
394 * P_DISCARD_ACK (proto C, two-primaries conflict detection)
395 * p_block_req:
396 * P_DATA_REQUEST, P_RS_DATA_REQUEST
397 */
398struct p_block_ack {
399 struct p_header80 head;
400 u64 sector;
401 u64 block_id;
402 u32 blksize;
403 u32 seq_num;
404} __packed;
405
406
407struct p_block_req {
408 struct p_header80 head;
409 u64 sector;
410 u64 block_id;
411 u32 blksize;
412 u32 pad; /* to multiple of 8 Byte */
413} __packed;
414
415/*
416 * commands with their own struct for additional fields:
417 * P_HAND_SHAKE
418 * P_BARRIER
419 * P_BARRIER_ACK
420 * P_SYNC_PARAM
421 * ReportParams
422 */
423
424struct p_handshake {
425 struct p_header80 head; /* 8 bytes */
426 u32 protocol_min;
427 u32 feature_flags;
428 u32 protocol_max;
429
430 /* should be more than enough for future enhancements
431 * for now, feature_flags and the reserverd array shall be zero.
432 */
433
434 u32 _pad;
435 u64 reserverd[7];
436} __packed;
437/* 80 bytes, FIXED for the next century */
438
439struct p_barrier {
440 struct p_header80 head;
441 u32 barrier; /* barrier number _handle_ only */
442 u32 pad; /* to multiple of 8 Byte */
443} __packed;
444
445struct p_barrier_ack {
446 struct p_header80 head;
447 u32 barrier;
448 u32 set_size;
449} __packed;
450
451struct p_rs_param {
452 struct p_header80 head;
453 u32 rate;
454
455 /* Since protocol version 88 and higher. */
456 char verify_alg[0];
457} __packed;
458
459struct p_rs_param_89 {
460 struct p_header80 head;
461 u32 rate;
462 /* protocol version 89: */
463 char verify_alg[SHARED_SECRET_MAX];
464 char csums_alg[SHARED_SECRET_MAX];
465} __packed;
466
467struct p_rs_param_95 {
468 struct p_header80 head;
469 u32 rate;
470 char verify_alg[SHARED_SECRET_MAX];
471 char csums_alg[SHARED_SECRET_MAX];
472 u32 c_plan_ahead;
473 u32 c_delay_target;
474 u32 c_fill_target;
475 u32 c_max_rate;
476} __packed;
477
478enum drbd_conn_flags {
479 CF_WANT_LOSE = 1,
480 CF_DRY_RUN = 2,
481};
482
483struct p_protocol {
484 struct p_header80 head;
485 u32 protocol;
486 u32 after_sb_0p;
487 u32 after_sb_1p;
488 u32 after_sb_2p;
489 u32 conn_flags;
490 u32 two_primaries;
491
492 /* Since protocol version 87 and higher. */
493 char integrity_alg[0];
494
495} __packed;
496
497struct p_uuids {
498 struct p_header80 head;
499 u64 uuid[UI_EXTENDED_SIZE];
500} __packed;
501
502struct p_rs_uuid {
503 struct p_header80 head;
504 u64 uuid;
505} __packed;
506
507struct p_sizes {
508 struct p_header80 head;
509 u64 d_size; /* size of disk */
510 u64 u_size; /* user requested size */
511 u64 c_size; /* current exported size */
512 u32 max_bio_size; /* Maximal size of a BIO */
513 u16 queue_order_type; /* not yet implemented in DRBD*/
514 u16 dds_flags; /* use enum dds_flags here. */
515} __packed;
516
517struct p_state {
518 struct p_header80 head;
519 u32 state;
520} __packed;
521
522struct p_req_state {
523 struct p_header80 head;
524 u32 mask;
525 u32 val;
526} __packed;
527
528struct p_req_state_reply {
529 struct p_header80 head;
530 u32 retcode;
531} __packed;
532
533struct p_drbd06_param {
534 u64 size;
535 u32 state;
536 u32 blksize;
537 u32 protocol;
538 u32 version;
539 u32 gen_cnt[5];
540 u32 bit_map_gen[5];
541} __packed;
542
543struct p_discard {
544 struct p_header80 head;
545 u64 block_id;
546 u32 seq_num;
547 u32 pad;
548} __packed;
549
550struct p_block_desc {
551 struct p_header80 head;
552 u64 sector;
553 u32 blksize;
554 u32 pad; /* to multiple of 8 Byte */
555} __packed;
556
557/* Valid values for the encoding field.
558 * Bump proto version when changing this. */
559enum drbd_bitmap_code {
560 /* RLE_VLI_Bytes = 0,
561 * and other bit variants had been defined during
562 * algorithm evaluation. */
563 RLE_VLI_Bits = 2,
564};
565
566struct p_compressed_bm {
567 struct p_header80 head;
568 /* (encoding & 0x0f): actual encoding, see enum drbd_bitmap_code
569 * (encoding & 0x80): polarity (set/unset) of first runlength
570 * ((encoding >> 4) & 0x07): pad_bits, number of trailing zero bits
571 * used to pad up to head.length bytes
572 */
573 u8 encoding;
574
575 u8 code[0];
576} __packed;
577
578struct p_delay_probe93 {
579 struct p_header80 head;
580 u32 seq_num; /* sequence number to match the two probe packets */
581 u32 offset; /* usecs the probe got sent after the reference time point */
582} __packed;
583
584/* DCBP: Drbd Compressed Bitmap Packet ... */
585static inline enum drbd_bitmap_code
586DCBP_get_code(struct p_compressed_bm *p)
587{
588 return (enum drbd_bitmap_code)(p->encoding & 0x0f);
589}
590
591static inline void
592DCBP_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
593{
594 BUG_ON(code & ~0xf);
595 p->encoding = (p->encoding & ~0xf) | code;
596}
597
598static inline int
599DCBP_get_start(struct p_compressed_bm *p)
600{
601 return (p->encoding & 0x80) != 0;
602}
603
604static inline void
605DCBP_set_start(struct p_compressed_bm *p, int set)
606{
607 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
608}
609
610static inline int
611DCBP_get_pad_bits(struct p_compressed_bm *p)
612{
613 return (p->encoding >> 4) & 0x7;
614}
615
616static inline void
617DCBP_set_pad_bits(struct p_compressed_bm *p, int n)
618{
619 BUG_ON(n & ~0x7);
620 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
621}
622
623/* one bitmap packet, including the p_header,
624 * should fit within one _architecture independend_ page.
625 * so we need to use the fixed size 4KiB page size
626 * most architectures have used for a long time.
627 */
628#define BM_PACKET_PAYLOAD_BYTES (4096 - sizeof(struct p_header80))
629#define BM_PACKET_WORDS (BM_PACKET_PAYLOAD_BYTES/sizeof(long))
630#define BM_PACKET_VLI_BYTES_MAX (4096 - sizeof(struct p_compressed_bm))
631#if (PAGE_SIZE < 4096)
632/* drbd_send_bitmap / receive_bitmap would break horribly */
633#error "PAGE_SIZE too small"
634#endif
635
636union p_polymorph {
637 union p_header header;
638 struct p_handshake handshake;
639 struct p_data data;
640 struct p_block_ack block_ack;
641 struct p_barrier barrier;
642 struct p_barrier_ack barrier_ack;
643 struct p_rs_param_89 rs_param_89;
644 struct p_rs_param_95 rs_param_95;
645 struct p_protocol protocol;
646 struct p_sizes sizes;
647 struct p_uuids uuids;
648 struct p_state state;
649 struct p_req_state req_state;
650 struct p_req_state_reply req_state_reply;
651 struct p_block_req block_req;
652 struct p_delay_probe93 delay_probe93;
653 struct p_rs_uuid rs_uuid;
654 struct p_block_desc block_desc;
655} __packed;
656
657/**********************************************************************/
658enum drbd_thread_state {
659 None,
660 Running,
661 Exiting,
662 Restarting
663};
664
665struct drbd_thread {
666 spinlock_t t_lock;
667 struct task_struct *task;
668 struct completion stop;
669 enum drbd_thread_state t_state;
670 int (*function) (struct drbd_thread *);
671 struct drbd_conf *mdev;
672 int reset_cpu_mask;
673};
674
675static inline enum drbd_thread_state get_t_state(struct drbd_thread *thi)
676{
677 /* THINK testing the t_state seems to be uncritical in all cases
678 * (but thread_{start,stop}), so we can read it *without* the lock.
679 * --lge */
680
681 smp_rmb();
682 return thi->t_state;
683}
684
685struct drbd_work;
686typedef int (*drbd_work_cb)(struct drbd_conf *, struct drbd_work *, int cancel);
687struct drbd_work {
688 struct list_head list;
689 drbd_work_cb cb;
690};
691
692struct drbd_tl_epoch;
693struct drbd_request {
694 struct drbd_work w;
695 struct drbd_conf *mdev;
696
697 /* if local IO is not allowed, will be NULL.
698 * if local IO _is_ allowed, holds the locally submitted bio clone,
699 * or, after local IO completion, the ERR_PTR(error).
700 * see drbd_endio_pri(). */
701 struct bio *private_bio;
702
703 struct hlist_node collision;
704 sector_t sector;
705 unsigned int size;
706 unsigned int epoch; /* barrier_nr */
707
708 /* barrier_nr: used to check on "completion" whether this req was in
709 * the current epoch, and we therefore have to close it,
710 * starting a new epoch...
711 */
712
713 struct list_head tl_requests; /* ring list in the transfer log */
714 struct bio *master_bio; /* master bio pointer */
715 unsigned long rq_state; /* see comments above _req_mod() */
716 int seq_num;
717 unsigned long start_time;
718};
719
720struct drbd_tl_epoch {
721 struct drbd_work w;
722 struct list_head requests; /* requests before */
723 struct drbd_tl_epoch *next; /* pointer to the next barrier */
724 unsigned int br_number; /* the barriers identifier. */
725 int n_writes; /* number of requests attached before this barrier */
726};
727
728struct drbd_request;
729
730/* These Tl_epoch_entries may be in one of 6 lists:
731 active_ee .. data packet being written
732 sync_ee .. syncer block being written
733 done_ee .. block written, need to send P_WRITE_ACK
734 read_ee .. [RS]P_DATA_REQUEST being read
735*/
736
737struct drbd_epoch {
738 struct list_head list;
739 unsigned int barrier_nr;
740 atomic_t epoch_size; /* increased on every request added. */
741 atomic_t active; /* increased on every req. added, and dec on every finished. */
742 unsigned long flags;
743};
744
745/* drbd_epoch flag bits */
746enum {
747 DE_HAVE_BARRIER_NUMBER,
748};
749
750enum epoch_event {
751 EV_PUT,
752 EV_GOT_BARRIER_NR,
753 EV_BECAME_LAST,
754 EV_CLEANUP = 32, /* used as flag */
755};
756
757struct drbd_wq_barrier {
758 struct drbd_work w;
759 struct completion done;
760};
761
762struct digest_info {
763 int digest_size;
764 void *digest;
765};
766
767struct drbd_epoch_entry {
768 struct drbd_work w;
769 struct hlist_node collision;
770 struct drbd_epoch *epoch; /* for writes */
771 struct drbd_conf *mdev;
772 struct page *pages;
773 atomic_t pending_bios;
774 unsigned int size;
775 /* see comments on ee flag bits below */
776 unsigned long flags;
777 sector_t sector;
778 union {
779 u64 block_id;
780 struct digest_info *digest;
781 };
782};
783
784/* ee flag bits.
785 * While corresponding bios are in flight, the only modification will be
786 * set_bit WAS_ERROR, which has to be atomic.
787 * If no bios are in flight yet, or all have been completed,
788 * non-atomic modification to ee->flags is ok.
789 */
790enum {
791 __EE_CALL_AL_COMPLETE_IO,
792 __EE_MAY_SET_IN_SYNC,
793
794 /* In case a barrier failed,
795 * we need to resubmit without the barrier flag. */
796 __EE_RESUBMITTED,
797
798 /* we may have several bios per epoch entry.
799 * if any of those fail, we set this flag atomically
800 * from the endio callback */
801 __EE_WAS_ERROR,
802
803 /* This ee has a pointer to a digest instead of a block id */
804 __EE_HAS_DIGEST,
805};
806#define EE_CALL_AL_COMPLETE_IO (1<<__EE_CALL_AL_COMPLETE_IO)
807#define EE_MAY_SET_IN_SYNC (1<<__EE_MAY_SET_IN_SYNC)
808#define EE_RESUBMITTED (1<<__EE_RESUBMITTED)
809#define EE_WAS_ERROR (1<<__EE_WAS_ERROR)
810#define EE_HAS_DIGEST (1<<__EE_HAS_DIGEST)
811
812/* global flag bits */
813enum {
814 CREATE_BARRIER, /* next P_DATA is preceded by a P_BARRIER */
815 SIGNAL_ASENDER, /* whether asender wants to be interrupted */
816 SEND_PING, /* whether asender should send a ping asap */
817
818 UNPLUG_QUEUED, /* only relevant with kernel 2.4 */
819 UNPLUG_REMOTE, /* sending a "UnplugRemote" could help */
820 MD_DIRTY, /* current uuids and flags not yet on disk */
821 DISCARD_CONCURRENT, /* Set on one node, cleared on the peer! */
822 USE_DEGR_WFC_T, /* degr-wfc-timeout instead of wfc-timeout. */
823 CLUSTER_ST_CHANGE, /* Cluster wide state change going on... */
824 CL_ST_CHG_SUCCESS,
825 CL_ST_CHG_FAIL,
826 CRASHED_PRIMARY, /* This node was a crashed primary.
827 * Gets cleared when the state.conn
828 * goes into C_CONNECTED state. */
829 NO_BARRIER_SUPP, /* underlying block device doesn't implement barriers */
830 CONSIDER_RESYNC,
831
832 MD_NO_FUA, /* Users wants us to not use FUA/FLUSH on meta data dev */
833 SUSPEND_IO, /* suspend application io */
834 BITMAP_IO, /* suspend application io;
835 once no more io in flight, start bitmap io */
836 BITMAP_IO_QUEUED, /* Started bitmap IO */
837 GO_DISKLESS, /* Disk is being detached, on io-error or admin request. */
838 WAS_IO_ERROR, /* Local disk failed returned IO error */
839 RESYNC_AFTER_NEG, /* Resync after online grow after the attach&negotiate finished. */
840 NET_CONGESTED, /* The data socket is congested */
841
842 CONFIG_PENDING, /* serialization of (re)configuration requests.
843 * if set, also prevents the device from dying */
844 DEVICE_DYING, /* device became unconfigured,
845 * but worker thread is still handling the cleanup.
846 * reconfiguring (nl_disk_conf, nl_net_conf) is dissalowed,
847 * while this is set. */
848 RESIZE_PENDING, /* Size change detected locally, waiting for the response from
849 * the peer, if it changed there as well. */
850 CONN_DRY_RUN, /* Expect disconnect after resync handshake. */
851 GOT_PING_ACK, /* set when we receive a ping_ack packet, misc wait gets woken */
852 NEW_CUR_UUID, /* Create new current UUID when thawing IO */
853 AL_SUSPENDED, /* Activity logging is currently suspended. */
854 AHEAD_TO_SYNC_SOURCE, /* Ahead -> SyncSource queued */
855};
856
857struct drbd_bitmap; /* opaque for drbd_conf */
858
859/* definition of bits in bm_flags to be used in drbd_bm_lock
860 * and drbd_bitmap_io and friends. */
861enum bm_flag {
862 /* do we need to kfree, or vfree bm_pages? */
863 BM_P_VMALLOCED = 0x10000, /* internal use only, will be masked out */
864
865 /* currently locked for bulk operation */
866 BM_LOCKED_MASK = 0x7,
867
868 /* in detail, that is: */
869 BM_DONT_CLEAR = 0x1,
870 BM_DONT_SET = 0x2,
871 BM_DONT_TEST = 0x4,
872
873 /* (test bit, count bit) allowed (common case) */
874 BM_LOCKED_TEST_ALLOWED = 0x3,
875
876 /* testing bits, as well as setting new bits allowed, but clearing bits
877 * would be unexpected. Used during bitmap receive. Setting new bits
878 * requires sending of "out-of-sync" information, though. */
879 BM_LOCKED_SET_ALLOWED = 0x1,
880
881 /* clear is not expected while bitmap is locked for bulk operation */
882};
883
884
885/* TODO sort members for performance
886 * MAYBE group them further */
887
888/* THINK maybe we actually want to use the default "event/%s" worker threads
889 * or similar in linux 2.6, which uses per cpu data and threads.
890 */
891struct drbd_work_queue {
892 struct list_head q;
893 struct semaphore s; /* producers up it, worker down()s it */
894 spinlock_t q_lock; /* to protect the list. */
895};
896
897struct drbd_socket {
898 struct drbd_work_queue work;
899 struct mutex mutex;
900 struct socket *socket;
901 /* this way we get our
902 * send/receive buffers off the stack */
903 union p_polymorph sbuf;
904 union p_polymorph rbuf;
905};
906
907struct drbd_md {
908 u64 md_offset; /* sector offset to 'super' block */
909
910 u64 la_size_sect; /* last agreed size, unit sectors */
911 u64 uuid[UI_SIZE];
912 u64 device_uuid;
913 u32 flags;
914 u32 md_size_sect;
915
916 s32 al_offset; /* signed relative sector offset to al area */
917 s32 bm_offset; /* signed relative sector offset to bitmap */
918
919 /* u32 al_nr_extents; important for restoring the AL
920 * is stored into sync_conf.al_extents, which in turn
921 * gets applied to act_log->nr_elements
922 */
923};
924
925/* for sync_conf and other types... */
926#define NL_PACKET(name, number, fields) struct name { fields };
927#define NL_INTEGER(pn,pr,member) int member;
928#define NL_INT64(pn,pr,member) __u64 member;
929#define NL_BIT(pn,pr,member) unsigned member:1;
930#define NL_STRING(pn,pr,member,len) unsigned char member[len]; int member ## _len;
931#include "linux/drbd_nl.h"
932
933struct drbd_backing_dev {
934 struct block_device *backing_bdev;
935 struct block_device *md_bdev;
936 struct drbd_md md;
937 struct disk_conf dc; /* The user provided config... */
938 sector_t known_size; /* last known size of that backing device */
939};
940
941struct drbd_md_io {
942 struct drbd_conf *mdev;
943 struct completion event;
944 int error;
945};
946
947struct bm_io_work {
948 struct drbd_work w;
949 char *why;
950 enum bm_flag flags;
951 int (*io_fn)(struct drbd_conf *mdev);
952 void (*done)(struct drbd_conf *mdev, int rv);
953};
954
955enum write_ordering_e {
956 WO_none,
957 WO_drain_io,
958 WO_bdev_flush,
959};
960
961struct fifo_buffer {
962 int *values;
963 unsigned int head_index;
964 unsigned int size;
965};
966
967struct drbd_conf {
968 /* things that are stored as / read from meta data on disk */
969 unsigned long flags;
970
971 /* configured by drbdsetup */
972 struct net_conf *net_conf; /* protected by get_net_conf() and put_net_conf() */
973 struct syncer_conf sync_conf;
974 struct drbd_backing_dev *ldev __protected_by(local);
975
976 sector_t p_size; /* partner's disk size */
977 struct request_queue *rq_queue;
978 struct block_device *this_bdev;
979 struct gendisk *vdisk;
980
981 struct drbd_socket data; /* data/barrier/cstate/parameter packets */
982 struct drbd_socket meta; /* ping/ack (metadata) packets */
983 int agreed_pro_version; /* actually used protocol version */
984 unsigned long last_received; /* in jiffies, either socket */
985 unsigned int ko_count;
986 struct drbd_work resync_work,
987 unplug_work,
988 go_diskless,
989 md_sync_work,
990 start_resync_work;
991 struct timer_list resync_timer;
992 struct timer_list md_sync_timer;
993 struct timer_list start_resync_timer;
994 struct timer_list request_timer;
995#ifdef DRBD_DEBUG_MD_SYNC
996 struct {
997 unsigned int line;
998 const char* func;
999 } last_md_mark_dirty;
1000#endif
1001
1002 /* Used after attach while negotiating new disk state. */
1003 union drbd_state new_state_tmp;
1004
1005 union drbd_state state;
1006 wait_queue_head_t misc_wait;
1007 wait_queue_head_t state_wait; /* upon each state change. */
1008 wait_queue_head_t net_cnt_wait;
1009 unsigned int send_cnt;
1010 unsigned int recv_cnt;
1011 unsigned int read_cnt;
1012 unsigned int writ_cnt;
1013 unsigned int al_writ_cnt;
1014 unsigned int bm_writ_cnt;
1015 atomic_t ap_bio_cnt; /* Requests we need to complete */
1016 atomic_t ap_pending_cnt; /* AP data packets on the wire, ack expected */
1017 atomic_t rs_pending_cnt; /* RS request/data packets on the wire */
1018 atomic_t unacked_cnt; /* Need to send replys for */
1019 atomic_t local_cnt; /* Waiting for local completion */
1020 atomic_t net_cnt; /* Users of net_conf */
1021 spinlock_t req_lock;
1022 struct drbd_tl_epoch *unused_spare_tle; /* for pre-allocation */
1023 struct drbd_tl_epoch *newest_tle;
1024 struct drbd_tl_epoch *oldest_tle;
1025 struct list_head out_of_sequence_requests;
1026 struct hlist_head *tl_hash;
1027 unsigned int tl_hash_s;
1028
1029 /* blocks to resync in this run [unit BM_BLOCK_SIZE] */
1030 unsigned long rs_total;
1031 /* number of resync blocks that failed in this run */
1032 unsigned long rs_failed;
1033 /* Syncer's start time [unit jiffies] */
1034 unsigned long rs_start;
1035 /* cumulated time in PausedSyncX state [unit jiffies] */
1036 unsigned long rs_paused;
1037 /* skipped because csum was equal [unit BM_BLOCK_SIZE] */
1038 unsigned long rs_same_csum;
1039#define DRBD_SYNC_MARKS 8
1040#define DRBD_SYNC_MARK_STEP (3*HZ)
1041 /* block not up-to-date at mark [unit BM_BLOCK_SIZE] */
1042 unsigned long rs_mark_left[DRBD_SYNC_MARKS];
1043 /* marks's time [unit jiffies] */
1044 unsigned long rs_mark_time[DRBD_SYNC_MARKS];
1045 /* current index into rs_mark_{left,time} */
1046 int rs_last_mark;
1047
1048 /* where does the admin want us to start? (sector) */
1049 sector_t ov_start_sector;
1050 /* where are we now? (sector) */
1051 sector_t ov_position;
1052 /* Start sector of out of sync range (to merge printk reporting). */
1053 sector_t ov_last_oos_start;
1054 /* size of out-of-sync range in sectors. */
1055 sector_t ov_last_oos_size;
1056 unsigned long ov_left; /* in bits */
1057 struct crypto_hash *csums_tfm;
1058 struct crypto_hash *verify_tfm;
1059
1060 struct drbd_thread receiver;
1061 struct drbd_thread worker;
1062 struct drbd_thread asender;
1063 struct drbd_bitmap *bitmap;
1064 unsigned long bm_resync_fo; /* bit offset for drbd_bm_find_next */
1065
1066 /* Used to track operations of resync... */
1067 struct lru_cache *resync;
1068 /* Number of locked elements in resync LRU */
1069 unsigned int resync_locked;
1070 /* resync extent number waiting for application requests */
1071 unsigned int resync_wenr;
1072
1073 int open_cnt;
1074 u64 *p_uuid;
1075 struct drbd_epoch *current_epoch;
1076 spinlock_t epoch_lock;
1077 unsigned int epochs;
1078 enum write_ordering_e write_ordering;
1079 struct list_head active_ee; /* IO in progress (P_DATA gets written to disk) */
1080 struct list_head sync_ee; /* IO in progress (P_RS_DATA_REPLY gets written to disk) */
1081 struct list_head done_ee; /* send ack */
1082 struct list_head read_ee; /* IO in progress (any read) */
1083 struct list_head net_ee; /* zero-copy network send in progress */
1084 struct hlist_head *ee_hash; /* is proteced by req_lock! */
1085 unsigned int ee_hash_s;
1086
1087 /* this one is protected by ee_lock, single thread */
1088 struct drbd_epoch_entry *last_write_w_barrier;
1089
1090 int next_barrier_nr;
1091 struct hlist_head *app_reads_hash; /* is proteced by req_lock */
1092 struct list_head resync_reads;
1093 atomic_t pp_in_use; /* allocated from page pool */
1094 atomic_t pp_in_use_by_net; /* sendpage()d, still referenced by tcp */
1095 wait_queue_head_t ee_wait;
1096 struct page *md_io_page; /* one page buffer for md_io */
1097 struct page *md_io_tmpp; /* for logical_block_size != 512 */
1098 struct mutex md_io_mutex; /* protects the md_io_buffer */
1099 spinlock_t al_lock;
1100 wait_queue_head_t al_wait;
1101 struct lru_cache *act_log; /* activity log */
1102 unsigned int al_tr_number;
1103 int al_tr_cycle;
1104 int al_tr_pos; /* position of the next transaction in the journal */
1105 struct crypto_hash *cram_hmac_tfm;
1106 struct crypto_hash *integrity_w_tfm; /* to be used by the worker thread */
1107 struct crypto_hash *integrity_r_tfm; /* to be used by the receiver thread */
1108 void *int_dig_out;
1109 void *int_dig_in;
1110 void *int_dig_vv;
1111 wait_queue_head_t seq_wait;
1112 atomic_t packet_seq;
1113 unsigned int peer_seq;
1114 spinlock_t peer_seq_lock;
1115 unsigned int minor;
1116 unsigned long comm_bm_set; /* communicated number of set bits. */
1117 cpumask_var_t cpu_mask;
1118 struct bm_io_work bm_io_work;
1119 u64 ed_uuid; /* UUID of the exposed data */
1120 struct mutex state_mutex;
1121 char congestion_reason; /* Why we where congested... */
1122 atomic_t rs_sect_in; /* for incoming resync data rate, SyncTarget */
1123 atomic_t rs_sect_ev; /* for submitted resync data rate, both */
1124 int rs_last_sect_ev; /* counter to compare with */
1125 int rs_last_events; /* counter of read or write "events" (unit sectors)
1126 * on the lower level device when we last looked. */
1127 int c_sync_rate; /* current resync rate after syncer throttle magic */
1128 struct fifo_buffer rs_plan_s; /* correction values of resync planer */
1129 int rs_in_flight; /* resync sectors in flight (to proxy, in proxy and from proxy) */
1130 int rs_planed; /* resync sectors already planned */
1131 atomic_t ap_in_flight; /* App sectors in flight (waiting for ack) */
1132 int peer_max_bio_size;
1133 int local_max_bio_size;
1134};
1135
1136static inline struct drbd_conf *minor_to_mdev(unsigned int minor)
1137{
1138 struct drbd_conf *mdev;
1139
1140 mdev = minor < minor_count ? minor_table[minor] : NULL;
1141
1142 return mdev;
1143}
1144
1145static inline unsigned int mdev_to_minor(struct drbd_conf *mdev)
1146{
1147 return mdev->minor;
1148}
1149
1150/* returns 1 if it was successful,
1151 * returns 0 if there was no data socket.
1152 * so wherever you are going to use the data.socket, e.g. do
1153 * if (!drbd_get_data_sock(mdev))
1154 * return 0;
1155 * CODE();
1156 * drbd_put_data_sock(mdev);
1157 */
1158static inline int drbd_get_data_sock(struct drbd_conf *mdev)
1159{
1160 mutex_lock(&mdev->data.mutex);
1161 /* drbd_disconnect() could have called drbd_free_sock()
1162 * while we were waiting in down()... */
1163 if (unlikely(mdev->data.socket == NULL)) {
1164 mutex_unlock(&mdev->data.mutex);
1165 return 0;
1166 }
1167 return 1;
1168}
1169
1170static inline void drbd_put_data_sock(struct drbd_conf *mdev)
1171{
1172 mutex_unlock(&mdev->data.mutex);
1173}
1174
1175/*
1176 * function declarations
1177 *************************/
1178
1179/* drbd_main.c */
1180
1181enum chg_state_flags {
1182 CS_HARD = 1,
1183 CS_VERBOSE = 2,
1184 CS_WAIT_COMPLETE = 4,
1185 CS_SERIALIZE = 8,
1186 CS_ORDERED = CS_WAIT_COMPLETE + CS_SERIALIZE,
1187};
1188
1189enum dds_flags {
1190 DDSF_FORCED = 1,
1191 DDSF_NO_RESYNC = 2, /* Do not run a resync for the new space */
1192};
1193
1194extern void drbd_init_set_defaults(struct drbd_conf *mdev);
1195extern enum drbd_state_rv drbd_change_state(struct drbd_conf *mdev,
1196 enum chg_state_flags f,
1197 union drbd_state mask,
1198 union drbd_state val);
1199extern void drbd_force_state(struct drbd_conf *, union drbd_state,
1200 union drbd_state);
1201extern enum drbd_state_rv _drbd_request_state(struct drbd_conf *,
1202 union drbd_state,
1203 union drbd_state,
1204 enum chg_state_flags);
1205extern enum drbd_state_rv __drbd_set_state(struct drbd_conf *, union drbd_state,
1206 enum chg_state_flags,
1207 struct completion *done);
1208extern void print_st_err(struct drbd_conf *, union drbd_state,
1209 union drbd_state, int);
1210extern int drbd_thread_start(struct drbd_thread *thi);
1211extern void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait);
1212#ifdef CONFIG_SMP
1213extern void drbd_thread_current_set_cpu(struct drbd_conf *mdev);
1214extern void drbd_calc_cpu_mask(struct drbd_conf *mdev);
1215#else
1216#define drbd_thread_current_set_cpu(A) ({})
1217#define drbd_calc_cpu_mask(A) ({})
1218#endif
1219extern void drbd_free_resources(struct drbd_conf *mdev);
1220extern void tl_release(struct drbd_conf *mdev, unsigned int barrier_nr,
1221 unsigned int set_size);
1222extern void tl_clear(struct drbd_conf *mdev);
1223extern void _tl_add_barrier(struct drbd_conf *, struct drbd_tl_epoch *);
1224extern void drbd_free_sock(struct drbd_conf *mdev);
1225extern int drbd_send(struct drbd_conf *mdev, struct socket *sock,
1226 void *buf, size_t size, unsigned msg_flags);
1227extern int drbd_send_protocol(struct drbd_conf *mdev);
1228extern int drbd_send_uuids(struct drbd_conf *mdev);
1229extern int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev);
1230extern int drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev);
1231extern int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags);
1232extern int _drbd_send_state(struct drbd_conf *mdev);
1233extern int drbd_send_state(struct drbd_conf *mdev);
1234extern int _drbd_send_cmd(struct drbd_conf *mdev, struct socket *sock,
1235 enum drbd_packets cmd, struct p_header80 *h,
1236 size_t size, unsigned msg_flags);
1237#define USE_DATA_SOCKET 1
1238#define USE_META_SOCKET 0
1239extern int drbd_send_cmd(struct drbd_conf *mdev, int use_data_socket,
1240 enum drbd_packets cmd, struct p_header80 *h,
1241 size_t size);
1242extern int drbd_send_cmd2(struct drbd_conf *mdev, enum drbd_packets cmd,
1243 char *data, size_t size);
1244extern int drbd_send_sync_param(struct drbd_conf *mdev, struct syncer_conf *sc);
1245extern int drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr,
1246 u32 set_size);
1247extern int drbd_send_ack(struct drbd_conf *mdev, enum drbd_packets cmd,
1248 struct drbd_epoch_entry *e);
1249extern int drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packets cmd,
1250 struct p_block_req *rp);
1251extern int drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packets cmd,
1252 struct p_data *dp, int data_size);
1253extern int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packets cmd,
1254 sector_t sector, int blksize, u64 block_id);
1255extern int drbd_send_oos(struct drbd_conf *mdev, struct drbd_request *req);
1256extern int drbd_send_block(struct drbd_conf *mdev, enum drbd_packets cmd,
1257 struct drbd_epoch_entry *e);
1258extern int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req);
1259extern int drbd_send_drequest(struct drbd_conf *mdev, int cmd,
1260 sector_t sector, int size, u64 block_id);
1261extern int drbd_send_drequest_csum(struct drbd_conf *mdev,
1262 sector_t sector,int size,
1263 void *digest, int digest_size,
1264 enum drbd_packets cmd);
1265extern int drbd_send_ov_request(struct drbd_conf *mdev,sector_t sector,int size);
1266
1267extern int drbd_send_bitmap(struct drbd_conf *mdev);
1268extern int _drbd_send_bitmap(struct drbd_conf *mdev);
1269extern int drbd_send_sr_reply(struct drbd_conf *mdev, enum drbd_state_rv retcode);
1270extern void drbd_free_bc(struct drbd_backing_dev *ldev);
1271extern void drbd_mdev_cleanup(struct drbd_conf *mdev);
1272void drbd_print_uuids(struct drbd_conf *mdev, const char *text);
1273
1274extern void drbd_md_sync(struct drbd_conf *mdev);
1275extern int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev);
1276extern void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local);
1277extern void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local);
1278extern void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local);
1279extern void _drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local);
1280extern void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local);
1281extern void drbd_md_set_flag(struct drbd_conf *mdev, int flags) __must_hold(local);
1282extern void drbd_md_clear_flag(struct drbd_conf *mdev, int flags)__must_hold(local);
1283extern int drbd_md_test_flag(struct drbd_backing_dev *, int);
1284#ifndef DRBD_DEBUG_MD_SYNC
1285extern void drbd_md_mark_dirty(struct drbd_conf *mdev);
1286#else
1287#define drbd_md_mark_dirty(m) drbd_md_mark_dirty_(m, __LINE__ , __func__ )
1288extern void drbd_md_mark_dirty_(struct drbd_conf *mdev,
1289 unsigned int line, const char *func);
1290#endif
1291extern void drbd_queue_bitmap_io(struct drbd_conf *mdev,
1292 int (*io_fn)(struct drbd_conf *),
1293 void (*done)(struct drbd_conf *, int),
1294 char *why, enum bm_flag flags);
1295extern int drbd_bitmap_io(struct drbd_conf *mdev,
1296 int (*io_fn)(struct drbd_conf *),
1297 char *why, enum bm_flag flags);
1298extern int drbd_bmio_set_n_write(struct drbd_conf *mdev);
1299extern int drbd_bmio_clear_n_write(struct drbd_conf *mdev);
1300extern void drbd_go_diskless(struct drbd_conf *mdev);
1301extern void drbd_ldev_destroy(struct drbd_conf *mdev);
1302
1303
1304/* Meta data layout
1305 We reserve a 128MB Block (4k aligned)
1306 * either at the end of the backing device
1307 * or on a separate meta data device. */
1308
1309#define MD_RESERVED_SECT (128LU << 11) /* 128 MB, unit sectors */
1310/* The following numbers are sectors */
1311#define MD_AL_OFFSET 8 /* 8 Sectors after start of meta area */
1312#define MD_AL_MAX_SIZE 64 /* = 32 kb LOG ~ 3776 extents ~ 14 GB Storage */
1313/* Allows up to about 3.8TB */
1314#define MD_BM_OFFSET (MD_AL_OFFSET + MD_AL_MAX_SIZE)
1315
1316/* Since the smalles IO unit is usually 512 byte */
1317#define MD_SECTOR_SHIFT 9
1318#define MD_SECTOR_SIZE (1<<MD_SECTOR_SHIFT)
1319
1320/* activity log */
1321#define AL_EXTENTS_PT ((MD_SECTOR_SIZE-12)/8-1) /* 61 ; Extents per 512B sector */
1322#define AL_EXTENT_SHIFT 22 /* One extent represents 4M Storage */
1323#define AL_EXTENT_SIZE (1<<AL_EXTENT_SHIFT)
1324
1325#if BITS_PER_LONG == 32
1326#define LN2_BPL 5
1327#define cpu_to_lel(A) cpu_to_le32(A)
1328#define lel_to_cpu(A) le32_to_cpu(A)
1329#elif BITS_PER_LONG == 64
1330#define LN2_BPL 6
1331#define cpu_to_lel(A) cpu_to_le64(A)
1332#define lel_to_cpu(A) le64_to_cpu(A)
1333#else
1334#error "LN2 of BITS_PER_LONG unknown!"
1335#endif
1336
1337/* resync bitmap */
1338/* 16MB sized 'bitmap extent' to track syncer usage */
1339struct bm_extent {
1340 int rs_left; /* number of bits set (out of sync) in this extent. */
1341 int rs_failed; /* number of failed resync requests in this extent. */
1342 unsigned long flags;
1343 struct lc_element lce;
1344};
1345
1346#define BME_NO_WRITES 0 /* bm_extent.flags: no more requests on this one! */
1347#define BME_LOCKED 1 /* bm_extent.flags: syncer active on this one. */
1348#define BME_PRIORITY 2 /* finish resync IO on this extent ASAP! App IO waiting! */
1349
1350/* drbd_bitmap.c */
1351/*
1352 * We need to store one bit for a block.
1353 * Example: 1GB disk @ 4096 byte blocks ==> we need 32 KB bitmap.
1354 * Bit 0 ==> local node thinks this block is binary identical on both nodes
1355 * Bit 1 ==> local node thinks this block needs to be synced.
1356 */
1357
1358#define SLEEP_TIME (HZ/10)
1359
1360#define BM_BLOCK_SHIFT 12 /* 4k per bit */
1361#define BM_BLOCK_SIZE (1<<BM_BLOCK_SHIFT)
1362/* (9+3) : 512 bytes @ 8 bits; representing 16M storage
1363 * per sector of on disk bitmap */
1364#define BM_EXT_SHIFT (BM_BLOCK_SHIFT + MD_SECTOR_SHIFT + 3) /* = 24 */
1365#define BM_EXT_SIZE (1<<BM_EXT_SHIFT)
1366
1367#if (BM_EXT_SHIFT != 24) || (BM_BLOCK_SHIFT != 12)
1368#error "HAVE YOU FIXED drbdmeta AS WELL??"
1369#endif
1370
1371/* thus many _storage_ sectors are described by one bit */
1372#define BM_SECT_TO_BIT(x) ((x)>>(BM_BLOCK_SHIFT-9))
1373#define BM_BIT_TO_SECT(x) ((sector_t)(x)<<(BM_BLOCK_SHIFT-9))
1374#define BM_SECT_PER_BIT BM_BIT_TO_SECT(1)
1375
1376/* bit to represented kilo byte conversion */
1377#define Bit2KB(bits) ((bits)<<(BM_BLOCK_SHIFT-10))
1378
1379/* in which _bitmap_ extent (resp. sector) the bit for a certain
1380 * _storage_ sector is located in */
1381#define BM_SECT_TO_EXT(x) ((x)>>(BM_EXT_SHIFT-9))
1382
1383/* how much _storage_ sectors we have per bitmap sector */
1384#define BM_EXT_TO_SECT(x) ((sector_t)(x) << (BM_EXT_SHIFT-9))
1385#define BM_SECT_PER_EXT BM_EXT_TO_SECT(1)
1386
1387/* in one sector of the bitmap, we have this many activity_log extents. */
1388#define AL_EXT_PER_BM_SECT (1 << (BM_EXT_SHIFT - AL_EXTENT_SHIFT))
1389#define BM_WORDS_PER_AL_EXT (1 << (AL_EXTENT_SHIFT-BM_BLOCK_SHIFT-LN2_BPL))
1390
1391#define BM_BLOCKS_PER_BM_EXT_B (BM_EXT_SHIFT - BM_BLOCK_SHIFT)
1392#define BM_BLOCKS_PER_BM_EXT_MASK ((1<<BM_BLOCKS_PER_BM_EXT_B) - 1)
1393
1394/* the extent in "PER_EXTENT" below is an activity log extent
1395 * we need that many (long words/bytes) to store the bitmap
1396 * of one AL_EXTENT_SIZE chunk of storage.
1397 * we can store the bitmap for that many AL_EXTENTS within
1398 * one sector of the _on_disk_ bitmap:
1399 * bit 0 bit 37 bit 38 bit (512*8)-1
1400 * ...|........|........|.. // ..|........|
1401 * sect. 0 `296 `304 ^(512*8*8)-1
1402 *
1403#define BM_WORDS_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / BITS_PER_LONG )
1404#define BM_BYTES_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / 8 ) // 128
1405#define BM_EXT_PER_SECT ( 512 / BM_BYTES_PER_EXTENT ) // 4
1406 */
1407
1408#define DRBD_MAX_SECTORS_32 (0xffffffffLU)
1409#define DRBD_MAX_SECTORS_BM \
1410 ((MD_RESERVED_SECT - MD_BM_OFFSET) * (1LL<<(BM_EXT_SHIFT-9)))
1411#if DRBD_MAX_SECTORS_BM < DRBD_MAX_SECTORS_32
1412#define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_BM
1413#define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_BM
1414#elif !defined(CONFIG_LBDAF) && BITS_PER_LONG == 32
1415#define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_32
1416#define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_32
1417#else
1418#define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_BM
1419/* 16 TB in units of sectors */
1420#if BITS_PER_LONG == 32
1421/* adjust by one page worth of bitmap,
1422 * so we won't wrap around in drbd_bm_find_next_bit.
1423 * you should use 64bit OS for that much storage, anyways. */
1424#define DRBD_MAX_SECTORS_FLEX BM_BIT_TO_SECT(0xffff7fff)
1425#else
1426/* we allow up to 1 PiB now on 64bit architecture with "flexible" meta data */
1427#define DRBD_MAX_SECTORS_FLEX (1UL << 51)
1428/* corresponds to (1UL << 38) bits right now. */
1429#endif
1430#endif
1431
1432/* Sector shift value for the "hash" functions of tl_hash and ee_hash tables.
1433 * With a value of 8 all IO in one 128K block make it to the same slot of the
1434 * hash table. */
1435#define HT_SHIFT 8
1436#define DRBD_MAX_BIO_SIZE (1U<<(9+HT_SHIFT))
1437#define DRBD_MAX_BIO_SIZE_SAFE (1 << 12) /* Works always = 4k */
1438
1439#define DRBD_MAX_SIZE_H80_PACKET (1 << 15) /* The old header only allows packets up to 32Kib data */
1440
1441/* Number of elements in the app_reads_hash */
1442#define APP_R_HSIZE 15
1443
1444extern int drbd_bm_init(struct drbd_conf *mdev);
1445extern int drbd_bm_resize(struct drbd_conf *mdev, sector_t sectors, int set_new_bits);
1446extern void drbd_bm_cleanup(struct drbd_conf *mdev);
1447extern void drbd_bm_set_all(struct drbd_conf *mdev);
1448extern void drbd_bm_clear_all(struct drbd_conf *mdev);
1449/* set/clear/test only a few bits at a time */
1450extern int drbd_bm_set_bits(
1451 struct drbd_conf *mdev, unsigned long s, unsigned long e);
1452extern int drbd_bm_clear_bits(
1453 struct drbd_conf *mdev, unsigned long s, unsigned long e);
1454extern int drbd_bm_count_bits(
1455 struct drbd_conf *mdev, const unsigned long s, const unsigned long e);
1456/* bm_set_bits variant for use while holding drbd_bm_lock,
1457 * may process the whole bitmap in one go */
1458extern void _drbd_bm_set_bits(struct drbd_conf *mdev,
1459 const unsigned long s, const unsigned long e);
1460extern int drbd_bm_test_bit(struct drbd_conf *mdev, unsigned long bitnr);
1461extern int drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr);
1462extern int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(local);
1463extern int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local);
1464extern int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local);
1465extern unsigned long drbd_bm_ALe_set_all(struct drbd_conf *mdev,
1466 unsigned long al_enr);
1467extern size_t drbd_bm_words(struct drbd_conf *mdev);
1468extern unsigned long drbd_bm_bits(struct drbd_conf *mdev);
1469extern sector_t drbd_bm_capacity(struct drbd_conf *mdev);
1470
1471#define DRBD_END_OF_BITMAP (~(unsigned long)0)
1472extern unsigned long drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo);
1473/* bm_find_next variants for use while you hold drbd_bm_lock() */
1474extern unsigned long _drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo);
1475extern unsigned long _drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo);
1476extern unsigned long _drbd_bm_total_weight(struct drbd_conf *mdev);
1477extern unsigned long drbd_bm_total_weight(struct drbd_conf *mdev);
1478extern int drbd_bm_rs_done(struct drbd_conf *mdev);
1479/* for receive_bitmap */
1480extern void drbd_bm_merge_lel(struct drbd_conf *mdev, size_t offset,
1481 size_t number, unsigned long *buffer);
1482/* for _drbd_send_bitmap */
1483extern void drbd_bm_get_lel(struct drbd_conf *mdev, size_t offset,
1484 size_t number, unsigned long *buffer);
1485
1486extern void drbd_bm_lock(struct drbd_conf *mdev, char *why, enum bm_flag flags);
1487extern void drbd_bm_unlock(struct drbd_conf *mdev);
1488/* drbd_main.c */
1489
1490extern struct kmem_cache *drbd_request_cache;
1491extern struct kmem_cache *drbd_ee_cache; /* epoch entries */
1492extern struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
1493extern struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
1494extern mempool_t *drbd_request_mempool;
1495extern mempool_t *drbd_ee_mempool;
1496
1497extern struct page *drbd_pp_pool; /* drbd's page pool */
1498extern spinlock_t drbd_pp_lock;
1499extern int drbd_pp_vacant;
1500extern wait_queue_head_t drbd_pp_wait;
1501
1502extern rwlock_t global_state_lock;
1503
1504extern struct drbd_conf *drbd_new_device(unsigned int minor);
1505extern void drbd_free_mdev(struct drbd_conf *mdev);
1506
1507extern int proc_details;
1508
1509/* drbd_req */
1510extern int drbd_make_request(struct request_queue *q, struct bio *bio);
1511extern int drbd_read_remote(struct drbd_conf *mdev, struct drbd_request *req);
1512extern int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct bio_vec *bvec);
1513extern int is_valid_ar_handle(struct drbd_request *, sector_t);
1514
1515
1516/* drbd_nl.c */
1517extern void drbd_suspend_io(struct drbd_conf *mdev);
1518extern void drbd_resume_io(struct drbd_conf *mdev);
1519extern char *ppsize(char *buf, unsigned long long size);
1520extern sector_t drbd_new_dev_size(struct drbd_conf *, struct drbd_backing_dev *, int);
1521enum determine_dev_size { dev_size_error = -1, unchanged = 0, shrunk = 1, grew = 2 };
1522extern enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *, enum dds_flags) __must_hold(local);
1523extern void resync_after_online_grow(struct drbd_conf *);
1524extern void drbd_reconsider_max_bio_size(struct drbd_conf *mdev);
1525extern enum drbd_state_rv drbd_set_role(struct drbd_conf *mdev,
1526 enum drbd_role new_role,
1527 int force);
1528extern enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev);
1529extern void drbd_try_outdate_peer_async(struct drbd_conf *mdev);
1530extern int drbd_khelper(struct drbd_conf *mdev, char *cmd);
1531
1532/* drbd_worker.c */
1533extern int drbd_worker(struct drbd_thread *thi);
1534extern int drbd_alter_sa(struct drbd_conf *mdev, int na);
1535extern void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side);
1536extern void resume_next_sg(struct drbd_conf *mdev);
1537extern void suspend_other_sg(struct drbd_conf *mdev);
1538extern int drbd_resync_finished(struct drbd_conf *mdev);
1539/* maybe rather drbd_main.c ? */
1540extern int drbd_md_sync_page_io(struct drbd_conf *mdev,
1541 struct drbd_backing_dev *bdev, sector_t sector, int rw);
1542extern void drbd_ov_oos_found(struct drbd_conf*, sector_t, int);
1543extern void drbd_rs_controller_reset(struct drbd_conf *mdev);
1544
1545static inline void ov_oos_print(struct drbd_conf *mdev)
1546{
1547 if (mdev->ov_last_oos_size) {
1548 dev_err(DEV, "Out of sync: start=%llu, size=%lu (sectors)\n",
1549 (unsigned long long)mdev->ov_last_oos_start,
1550 (unsigned long)mdev->ov_last_oos_size);
1551 }
1552 mdev->ov_last_oos_size=0;
1553}
1554
1555
1556extern void drbd_csum_bio(struct drbd_conf *, struct crypto_hash *, struct bio *, void *);
1557extern void drbd_csum_ee(struct drbd_conf *, struct crypto_hash *, struct drbd_epoch_entry *, void *);
1558/* worker callbacks */
1559extern int w_req_cancel_conflict(struct drbd_conf *, struct drbd_work *, int);
1560extern int w_read_retry_remote(struct drbd_conf *, struct drbd_work *, int);
1561extern int w_e_end_data_req(struct drbd_conf *, struct drbd_work *, int);
1562extern int w_e_end_rsdata_req(struct drbd_conf *, struct drbd_work *, int);
1563extern int w_e_end_csum_rs_req(struct drbd_conf *, struct drbd_work *, int);
1564extern int w_e_end_ov_reply(struct drbd_conf *, struct drbd_work *, int);
1565extern int w_e_end_ov_req(struct drbd_conf *, struct drbd_work *, int);
1566extern int w_ov_finished(struct drbd_conf *, struct drbd_work *, int);
1567extern int w_resync_timer(struct drbd_conf *, struct drbd_work *, int);
1568extern int w_resume_next_sg(struct drbd_conf *, struct drbd_work *, int);
1569extern int w_send_write_hint(struct drbd_conf *, struct drbd_work *, int);
1570extern int w_send_dblock(struct drbd_conf *, struct drbd_work *, int);
1571extern int w_send_barrier(struct drbd_conf *, struct drbd_work *, int);
1572extern int w_send_read_req(struct drbd_conf *, struct drbd_work *, int);
1573extern int w_prev_work_done(struct drbd_conf *, struct drbd_work *, int);
1574extern int w_e_reissue(struct drbd_conf *, struct drbd_work *, int);
1575extern int w_restart_disk_io(struct drbd_conf *, struct drbd_work *, int);
1576extern int w_send_oos(struct drbd_conf *, struct drbd_work *, int);
1577extern int w_start_resync(struct drbd_conf *, struct drbd_work *, int);
1578
1579extern void resync_timer_fn(unsigned long data);
1580extern void start_resync_timer_fn(unsigned long data);
1581
1582/* drbd_receiver.c */
1583extern int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector);
1584extern int drbd_submit_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e,
1585 const unsigned rw, const int fault_type);
1586extern int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list);
1587extern struct drbd_epoch_entry *drbd_alloc_ee(struct drbd_conf *mdev,
1588 u64 id,
1589 sector_t sector,
1590 unsigned int data_size,
1591 gfp_t gfp_mask) __must_hold(local);
1592extern void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e,
1593 int is_net);
1594#define drbd_free_ee(m,e) drbd_free_some_ee(m, e, 0)
1595#define drbd_free_net_ee(m,e) drbd_free_some_ee(m, e, 1)
1596extern void drbd_wait_ee_list_empty(struct drbd_conf *mdev,
1597 struct list_head *head);
1598extern void _drbd_wait_ee_list_empty(struct drbd_conf *mdev,
1599 struct list_head *head);
1600extern void drbd_set_recv_tcq(struct drbd_conf *mdev, int tcq_enabled);
1601extern void _drbd_clear_done_ee(struct drbd_conf *mdev, struct list_head *to_be_freed);
1602extern void drbd_flush_workqueue(struct drbd_conf *mdev);
1603extern void drbd_free_tl_hash(struct drbd_conf *mdev);
1604
1605/* yes, there is kernel_setsockopt, but only since 2.6.18. we don't need to
1606 * mess with get_fs/set_fs, we know we are KERNEL_DS always. */
1607static inline int drbd_setsockopt(struct socket *sock, int level, int optname,
1608 char __user *optval, int optlen)
1609{
1610 int err;
1611 if (level == SOL_SOCKET)
1612 err = sock_setsockopt(sock, level, optname, optval, optlen);
1613 else
1614 err = sock->ops->setsockopt(sock, level, optname, optval,
1615 optlen);
1616 return err;
1617}
1618
1619static inline void drbd_tcp_cork(struct socket *sock)
1620{
1621 int __user val = 1;
1622 (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK,
1623 (char __user *)&val, sizeof(val));
1624}
1625
1626static inline void drbd_tcp_uncork(struct socket *sock)
1627{
1628 int __user val = 0;
1629 (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK,
1630 (char __user *)&val, sizeof(val));
1631}
1632
1633static inline void drbd_tcp_nodelay(struct socket *sock)
1634{
1635 int __user val = 1;
1636 (void) drbd_setsockopt(sock, SOL_TCP, TCP_NODELAY,
1637 (char __user *)&val, sizeof(val));
1638}
1639
1640static inline void drbd_tcp_quickack(struct socket *sock)
1641{
1642 int __user val = 2;
1643 (void) drbd_setsockopt(sock, SOL_TCP, TCP_QUICKACK,
1644 (char __user *)&val, sizeof(val));
1645}
1646
1647void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo);
1648
1649/* drbd_proc.c */
1650extern struct proc_dir_entry *drbd_proc;
1651extern const struct file_operations drbd_proc_fops;
1652extern const char *drbd_conn_str(enum drbd_conns s);
1653extern const char *drbd_role_str(enum drbd_role s);
1654
1655/* drbd_actlog.c */
1656extern void drbd_al_begin_io(struct drbd_conf *mdev, sector_t sector);
1657extern void drbd_al_complete_io(struct drbd_conf *mdev, sector_t sector);
1658extern void drbd_rs_complete_io(struct drbd_conf *mdev, sector_t sector);
1659extern int drbd_rs_begin_io(struct drbd_conf *mdev, sector_t sector);
1660extern int drbd_try_rs_begin_io(struct drbd_conf *mdev, sector_t sector);
1661extern void drbd_rs_cancel_all(struct drbd_conf *mdev);
1662extern int drbd_rs_del_all(struct drbd_conf *mdev);
1663extern void drbd_rs_failed_io(struct drbd_conf *mdev,
1664 sector_t sector, int size);
1665extern int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *);
1666extern void drbd_advance_rs_marks(struct drbd_conf *mdev, unsigned long still_to_go);
1667extern void __drbd_set_in_sync(struct drbd_conf *mdev, sector_t sector,
1668 int size, const char *file, const unsigned int line);
1669#define drbd_set_in_sync(mdev, sector, size) \
1670 __drbd_set_in_sync(mdev, sector, size, __FILE__, __LINE__)
1671extern int __drbd_set_out_of_sync(struct drbd_conf *mdev, sector_t sector,
1672 int size, const char *file, const unsigned int line);
1673#define drbd_set_out_of_sync(mdev, sector, size) \
1674 __drbd_set_out_of_sync(mdev, sector, size, __FILE__, __LINE__)
1675extern void drbd_al_apply_to_bm(struct drbd_conf *mdev);
1676extern void drbd_al_shrink(struct drbd_conf *mdev);
1677
1678
1679/* drbd_nl.c */
1680
1681void drbd_nl_cleanup(void);
1682int __init drbd_nl_init(void);
1683void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state);
1684void drbd_bcast_sync_progress(struct drbd_conf *mdev);
1685void drbd_bcast_ee(struct drbd_conf *mdev,
1686 const char *reason, const int dgs,
1687 const char* seen_hash, const char* calc_hash,
1688 const struct drbd_epoch_entry* e);
1689
1690
1691/**
1692 * DOC: DRBD State macros
1693 *
1694 * These macros are used to express state changes in easily readable form.
1695 *
1696 * The NS macros expand to a mask and a value, that can be bit ored onto the
1697 * current state as soon as the spinlock (req_lock) was taken.
1698 *
1699 * The _NS macros are used for state functions that get called with the
1700 * spinlock. These macros expand directly to the new state value.
1701 *
1702 * Besides the basic forms NS() and _NS() additional _?NS[23] are defined
1703 * to express state changes that affect more than one aspect of the state.
1704 *
1705 * E.g. NS2(conn, C_CONNECTED, peer, R_SECONDARY)
1706 * Means that the network connection was established and that the peer
1707 * is in secondary role.
1708 */
1709#define role_MASK R_MASK
1710#define peer_MASK R_MASK
1711#define disk_MASK D_MASK
1712#define pdsk_MASK D_MASK
1713#define conn_MASK C_MASK
1714#define susp_MASK 1
1715#define user_isp_MASK 1
1716#define aftr_isp_MASK 1
1717#define susp_nod_MASK 1
1718#define susp_fen_MASK 1
1719
1720#define NS(T, S) \
1721 ({ union drbd_state mask; mask.i = 0; mask.T = T##_MASK; mask; }), \
1722 ({ union drbd_state val; val.i = 0; val.T = (S); val; })
1723#define NS2(T1, S1, T2, S2) \
1724 ({ union drbd_state mask; mask.i = 0; mask.T1 = T1##_MASK; \
1725 mask.T2 = T2##_MASK; mask; }), \
1726 ({ union drbd_state val; val.i = 0; val.T1 = (S1); \
1727 val.T2 = (S2); val; })
1728#define NS3(T1, S1, T2, S2, T3, S3) \
1729 ({ union drbd_state mask; mask.i = 0; mask.T1 = T1##_MASK; \
1730 mask.T2 = T2##_MASK; mask.T3 = T3##_MASK; mask; }), \
1731 ({ union drbd_state val; val.i = 0; val.T1 = (S1); \
1732 val.T2 = (S2); val.T3 = (S3); val; })
1733
1734#define _NS(D, T, S) \
1735 D, ({ union drbd_state __ns; __ns.i = D->state.i; __ns.T = (S); __ns; })
1736#define _NS2(D, T1, S1, T2, S2) \
1737 D, ({ union drbd_state __ns; __ns.i = D->state.i; __ns.T1 = (S1); \
1738 __ns.T2 = (S2); __ns; })
1739#define _NS3(D, T1, S1, T2, S2, T3, S3) \
1740 D, ({ union drbd_state __ns; __ns.i = D->state.i; __ns.T1 = (S1); \
1741 __ns.T2 = (S2); __ns.T3 = (S3); __ns; })
1742
1743/*
1744 * inline helper functions
1745 *************************/
1746
1747/* see also page_chain_add and friends in drbd_receiver.c */
1748static inline struct page *page_chain_next(struct page *page)
1749{
1750 return (struct page *)page_private(page);
1751}
1752#define page_chain_for_each(page) \
1753 for (; page && ({ prefetch(page_chain_next(page)); 1; }); \
1754 page = page_chain_next(page))
1755#define page_chain_for_each_safe(page, n) \
1756 for (; page && ({ n = page_chain_next(page); 1; }); page = n)
1757
1758static inline int drbd_bio_has_active_page(struct bio *bio)
1759{
1760 struct bio_vec *bvec;
1761 int i;
1762
1763 __bio_for_each_segment(bvec, bio, i, 0) {
1764 if (page_count(bvec->bv_page) > 1)
1765 return 1;
1766 }
1767
1768 return 0;
1769}
1770
1771static inline int drbd_ee_has_active_page(struct drbd_epoch_entry *e)
1772{
1773 struct page *page = e->pages;
1774 page_chain_for_each(page) {
1775 if (page_count(page) > 1)
1776 return 1;
1777 }
1778 return 0;
1779}
1780
1781
1782static inline void drbd_state_lock(struct drbd_conf *mdev)
1783{
1784 wait_event(mdev->misc_wait,
1785 !test_and_set_bit(CLUSTER_ST_CHANGE, &mdev->flags));
1786}
1787
1788static inline void drbd_state_unlock(struct drbd_conf *mdev)
1789{
1790 clear_bit(CLUSTER_ST_CHANGE, &mdev->flags);
1791 wake_up(&mdev->misc_wait);
1792}
1793
1794static inline enum drbd_state_rv
1795_drbd_set_state(struct drbd_conf *mdev, union drbd_state ns,
1796 enum chg_state_flags flags, struct completion *done)
1797{
1798 enum drbd_state_rv rv;
1799
1800 read_lock(&global_state_lock);
1801 rv = __drbd_set_state(mdev, ns, flags, done);
1802 read_unlock(&global_state_lock);
1803
1804 return rv;
1805}
1806
1807/**
1808 * drbd_request_state() - Reqest a state change
1809 * @mdev: DRBD device.
1810 * @mask: mask of state bits to change.
1811 * @val: value of new state bits.
1812 *
1813 * This is the most graceful way of requesting a state change. It is verbose
1814 * quite verbose in case the state change is not possible, and all those
1815 * state changes are globally serialized.
1816 */
1817static inline int drbd_request_state(struct drbd_conf *mdev,
1818 union drbd_state mask,
1819 union drbd_state val)
1820{
1821 return _drbd_request_state(mdev, mask, val, CS_VERBOSE + CS_ORDERED);
1822}
1823
1824#define __drbd_chk_io_error(m,f) __drbd_chk_io_error_(m,f, __func__)
1825static inline void __drbd_chk_io_error_(struct drbd_conf *mdev, int forcedetach, const char *where)
1826{
1827 switch (mdev->ldev->dc.on_io_error) {
1828 case EP_PASS_ON:
1829 if (!forcedetach) {
1830 if (__ratelimit(&drbd_ratelimit_state))
1831 dev_err(DEV, "Local IO failed in %s.\n", where);
1832 if (mdev->state.disk > D_INCONSISTENT)
1833 _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_HARD, NULL);
1834 break;
1835 }
1836 /* NOTE fall through to detach case if forcedetach set */
1837 case EP_DETACH:
1838 case EP_CALL_HELPER:
1839 set_bit(WAS_IO_ERROR, &mdev->flags);
1840 if (mdev->state.disk > D_FAILED) {
1841 _drbd_set_state(_NS(mdev, disk, D_FAILED), CS_HARD, NULL);
1842 dev_err(DEV,
1843 "Local IO failed in %s. Detaching...\n", where);
1844 }
1845 break;
1846 }
1847}
1848
1849/**
1850 * drbd_chk_io_error: Handle the on_io_error setting, should be called from all io completion handlers
1851 * @mdev: DRBD device.
1852 * @error: Error code passed to the IO completion callback
1853 * @forcedetach: Force detach. I.e. the error happened while accessing the meta data
1854 *
1855 * See also drbd_main.c:after_state_ch() if (os.disk > D_FAILED && ns.disk == D_FAILED)
1856 */
1857#define drbd_chk_io_error(m,e,f) drbd_chk_io_error_(m,e,f, __func__)
1858static inline void drbd_chk_io_error_(struct drbd_conf *mdev,
1859 int error, int forcedetach, const char *where)
1860{
1861 if (error) {
1862 unsigned long flags;
1863 spin_lock_irqsave(&mdev->req_lock, flags);
1864 __drbd_chk_io_error_(mdev, forcedetach, where);
1865 spin_unlock_irqrestore(&mdev->req_lock, flags);
1866 }
1867}
1868
1869
1870/**
1871 * drbd_md_first_sector() - Returns the first sector number of the meta data area
1872 * @bdev: Meta data block device.
1873 *
1874 * BTW, for internal meta data, this happens to be the maximum capacity
1875 * we could agree upon with our peer node.
1876 */
1877static inline sector_t drbd_md_first_sector(struct drbd_backing_dev *bdev)
1878{
1879 switch (bdev->dc.meta_dev_idx) {
1880 case DRBD_MD_INDEX_INTERNAL:
1881 case DRBD_MD_INDEX_FLEX_INT:
1882 return bdev->md.md_offset + bdev->md.bm_offset;
1883 case DRBD_MD_INDEX_FLEX_EXT:
1884 default:
1885 return bdev->md.md_offset;
1886 }
1887}
1888
1889/**
1890 * drbd_md_last_sector() - Return the last sector number of the meta data area
1891 * @bdev: Meta data block device.
1892 */
1893static inline sector_t drbd_md_last_sector(struct drbd_backing_dev *bdev)
1894{
1895 switch (bdev->dc.meta_dev_idx) {
1896 case DRBD_MD_INDEX_INTERNAL:
1897 case DRBD_MD_INDEX_FLEX_INT:
1898 return bdev->md.md_offset + MD_AL_OFFSET - 1;
1899 case DRBD_MD_INDEX_FLEX_EXT:
1900 default:
1901 return bdev->md.md_offset + bdev->md.md_size_sect;
1902 }
1903}
1904
1905/* Returns the number of 512 byte sectors of the device */
1906static inline sector_t drbd_get_capacity(struct block_device *bdev)
1907{
1908 /* return bdev ? get_capacity(bdev->bd_disk) : 0; */
1909 return bdev ? i_size_read(bdev->bd_inode) >> 9 : 0;
1910}
1911
1912/**
1913 * drbd_get_max_capacity() - Returns the capacity we announce to out peer
1914 * @bdev: Meta data block device.
1915 *
1916 * returns the capacity we announce to out peer. we clip ourselves at the
1917 * various MAX_SECTORS, because if we don't, current implementation will
1918 * oops sooner or later
1919 */
1920static inline sector_t drbd_get_max_capacity(struct drbd_backing_dev *bdev)
1921{
1922 sector_t s;
1923 switch (bdev->dc.meta_dev_idx) {
1924 case DRBD_MD_INDEX_INTERNAL:
1925 case DRBD_MD_INDEX_FLEX_INT:
1926 s = drbd_get_capacity(bdev->backing_bdev)
1927 ? min_t(sector_t, DRBD_MAX_SECTORS_FLEX,
1928 drbd_md_first_sector(bdev))
1929 : 0;
1930 break;
1931 case DRBD_MD_INDEX_FLEX_EXT:
1932 s = min_t(sector_t, DRBD_MAX_SECTORS_FLEX,
1933 drbd_get_capacity(bdev->backing_bdev));
1934 /* clip at maximum size the meta device can support */
1935 s = min_t(sector_t, s,
1936 BM_EXT_TO_SECT(bdev->md.md_size_sect
1937 - bdev->md.bm_offset));
1938 break;
1939 default:
1940 s = min_t(sector_t, DRBD_MAX_SECTORS,
1941 drbd_get_capacity(bdev->backing_bdev));
1942 }
1943 return s;
1944}
1945
1946/**
1947 * drbd_md_ss__() - Return the sector number of our meta data super block
1948 * @mdev: DRBD device.
1949 * @bdev: Meta data block device.
1950 */
1951static inline sector_t drbd_md_ss__(struct drbd_conf *mdev,
1952 struct drbd_backing_dev *bdev)
1953{
1954 switch (bdev->dc.meta_dev_idx) {
1955 default: /* external, some index */
1956 return MD_RESERVED_SECT * bdev->dc.meta_dev_idx;
1957 case DRBD_MD_INDEX_INTERNAL:
1958 /* with drbd08, internal meta data is always "flexible" */
1959 case DRBD_MD_INDEX_FLEX_INT:
1960 /* sizeof(struct md_on_disk_07) == 4k
1961 * position: last 4k aligned block of 4k size */
1962 if (!bdev->backing_bdev) {
1963 if (__ratelimit(&drbd_ratelimit_state)) {
1964 dev_err(DEV, "bdev->backing_bdev==NULL\n");
1965 dump_stack();
1966 }
1967 return 0;
1968 }
1969 return (drbd_get_capacity(bdev->backing_bdev) & ~7ULL)
1970 - MD_AL_OFFSET;
1971 case DRBD_MD_INDEX_FLEX_EXT:
1972 return 0;
1973 }
1974}
1975
1976static inline void
1977drbd_queue_work_front(struct drbd_work_queue *q, struct drbd_work *w)
1978{
1979 unsigned long flags;
1980 spin_lock_irqsave(&q->q_lock, flags);
1981 list_add(&w->list, &q->q);
1982 up(&q->s); /* within the spinlock,
1983 see comment near end of drbd_worker() */
1984 spin_unlock_irqrestore(&q->q_lock, flags);
1985}
1986
1987static inline void
1988drbd_queue_work(struct drbd_work_queue *q, struct drbd_work *w)
1989{
1990 unsigned long flags;
1991 spin_lock_irqsave(&q->q_lock, flags);
1992 list_add_tail(&w->list, &q->q);
1993 up(&q->s); /* within the spinlock,
1994 see comment near end of drbd_worker() */
1995 spin_unlock_irqrestore(&q->q_lock, flags);
1996}
1997
1998static inline void wake_asender(struct drbd_conf *mdev)
1999{
2000 if (test_bit(SIGNAL_ASENDER, &mdev->flags))
2001 force_sig(DRBD_SIG, mdev->asender.task);
2002}
2003
2004static inline void request_ping(struct drbd_conf *mdev)
2005{
2006 set_bit(SEND_PING, &mdev->flags);
2007 wake_asender(mdev);
2008}
2009
2010static inline int drbd_send_short_cmd(struct drbd_conf *mdev,
2011 enum drbd_packets cmd)
2012{
2013 struct p_header80 h;
2014 return drbd_send_cmd(mdev, USE_DATA_SOCKET, cmd, &h, sizeof(h));
2015}
2016
2017static inline int drbd_send_ping(struct drbd_conf *mdev)
2018{
2019 struct p_header80 h;
2020 return drbd_send_cmd(mdev, USE_META_SOCKET, P_PING, &h, sizeof(h));
2021}
2022
2023static inline int drbd_send_ping_ack(struct drbd_conf *mdev)
2024{
2025 struct p_header80 h;
2026 return drbd_send_cmd(mdev, USE_META_SOCKET, P_PING_ACK, &h, sizeof(h));
2027}
2028
2029static inline void drbd_thread_stop(struct drbd_thread *thi)
2030{
2031 _drbd_thread_stop(thi, false, true);
2032}
2033
2034static inline void drbd_thread_stop_nowait(struct drbd_thread *thi)
2035{
2036 _drbd_thread_stop(thi, false, false);
2037}
2038
2039static inline void drbd_thread_restart_nowait(struct drbd_thread *thi)
2040{
2041 _drbd_thread_stop(thi, true, false);
2042}
2043
2044/* counts how many answer packets packets we expect from our peer,
2045 * for either explicit application requests,
2046 * or implicit barrier packets as necessary.
2047 * increased:
2048 * w_send_barrier
2049 * _req_mod(req, queue_for_net_write or queue_for_net_read);
2050 * it is much easier and equally valid to count what we queue for the
2051 * worker, even before it actually was queued or send.
2052 * (drbd_make_request_common; recovery path on read io-error)
2053 * decreased:
2054 * got_BarrierAck (respective tl_clear, tl_clear_barrier)
2055 * _req_mod(req, data_received)
2056 * [from receive_DataReply]
2057 * _req_mod(req, write_acked_by_peer or recv_acked_by_peer or neg_acked)
2058 * [from got_BlockAck (P_WRITE_ACK, P_RECV_ACK)]
2059 * for some reason it is NOT decreased in got_NegAck,
2060 * but in the resulting cleanup code from report_params.
2061 * we should try to remember the reason for that...
2062 * _req_mod(req, send_failed or send_canceled)
2063 * _req_mod(req, connection_lost_while_pending)
2064 * [from tl_clear_barrier]
2065 */
2066static inline void inc_ap_pending(struct drbd_conf *mdev)
2067{
2068 atomic_inc(&mdev->ap_pending_cnt);
2069}
2070
2071#define ERR_IF_CNT_IS_NEGATIVE(which) \
2072 if (atomic_read(&mdev->which) < 0) \
2073 dev_err(DEV, "in %s:%d: " #which " = %d < 0 !\n", \
2074 __func__ , __LINE__ , \
2075 atomic_read(&mdev->which))
2076
2077#define dec_ap_pending(mdev) do { \
2078 typecheck(struct drbd_conf *, mdev); \
2079 if (atomic_dec_and_test(&mdev->ap_pending_cnt)) \
2080 wake_up(&mdev->misc_wait); \
2081 ERR_IF_CNT_IS_NEGATIVE(ap_pending_cnt); } while (0)
2082
2083/* counts how many resync-related answers we still expect from the peer
2084 * increase decrease
2085 * C_SYNC_TARGET sends P_RS_DATA_REQUEST (and expects P_RS_DATA_REPLY)
2086 * C_SYNC_SOURCE sends P_RS_DATA_REPLY (and expects P_WRITE_ACK with ID_SYNCER)
2087 * (or P_NEG_ACK with ID_SYNCER)
2088 */
2089static inline void inc_rs_pending(struct drbd_conf *mdev)
2090{
2091 atomic_inc(&mdev->rs_pending_cnt);
2092}
2093
2094#define dec_rs_pending(mdev) do { \
2095 typecheck(struct drbd_conf *, mdev); \
2096 atomic_dec(&mdev->rs_pending_cnt); \
2097 ERR_IF_CNT_IS_NEGATIVE(rs_pending_cnt); } while (0)
2098
2099/* counts how many answers we still need to send to the peer.
2100 * increased on
2101 * receive_Data unless protocol A;
2102 * we need to send a P_RECV_ACK (proto B)
2103 * or P_WRITE_ACK (proto C)
2104 * receive_RSDataReply (recv_resync_read) we need to send a P_WRITE_ACK
2105 * receive_DataRequest (receive_RSDataRequest) we need to send back P_DATA
2106 * receive_Barrier_* we need to send a P_BARRIER_ACK
2107 */
2108static inline void inc_unacked(struct drbd_conf *mdev)
2109{
2110 atomic_inc(&mdev->unacked_cnt);
2111}
2112
2113#define dec_unacked(mdev) do { \
2114 typecheck(struct drbd_conf *, mdev); \
2115 atomic_dec(&mdev->unacked_cnt); \
2116 ERR_IF_CNT_IS_NEGATIVE(unacked_cnt); } while (0)
2117
2118#define sub_unacked(mdev, n) do { \
2119 typecheck(struct drbd_conf *, mdev); \
2120 atomic_sub(n, &mdev->unacked_cnt); \
2121 ERR_IF_CNT_IS_NEGATIVE(unacked_cnt); } while (0)
2122
2123
2124static inline void put_net_conf(struct drbd_conf *mdev)
2125{
2126 if (atomic_dec_and_test(&mdev->net_cnt))
2127 wake_up(&mdev->net_cnt_wait);
2128}
2129
2130/**
2131 * get_net_conf() - Increase ref count on mdev->net_conf; Returns 0 if nothing there
2132 * @mdev: DRBD device.
2133 *
2134 * You have to call put_net_conf() when finished working with mdev->net_conf.
2135 */
2136static inline int get_net_conf(struct drbd_conf *mdev)
2137{
2138 int have_net_conf;
2139
2140 atomic_inc(&mdev->net_cnt);
2141 have_net_conf = mdev->state.conn >= C_UNCONNECTED;
2142 if (!have_net_conf)
2143 put_net_conf(mdev);
2144 return have_net_conf;
2145}
2146
2147/**
2148 * get_ldev() - Increase the ref count on mdev->ldev. Returns 0 if there is no ldev
2149 * @M: DRBD device.
2150 *
2151 * You have to call put_ldev() when finished working with mdev->ldev.
2152 */
2153#define get_ldev(M) __cond_lock(local, _get_ldev_if_state(M,D_INCONSISTENT))
2154#define get_ldev_if_state(M,MINS) __cond_lock(local, _get_ldev_if_state(M,MINS))
2155
2156static inline void put_ldev(struct drbd_conf *mdev)
2157{
2158 int i = atomic_dec_return(&mdev->local_cnt);
2159
2160 /* This may be called from some endio handler,
2161 * so we must not sleep here. */
2162
2163 __release(local);
2164 D_ASSERT(i >= 0);
2165 if (i == 0) {
2166 if (mdev->state.disk == D_DISKLESS)
2167 /* even internal references gone, safe to destroy */
2168 drbd_ldev_destroy(mdev);
2169 if (mdev->state.disk == D_FAILED)
2170 /* all application IO references gone. */
2171 drbd_go_diskless(mdev);
2172 wake_up(&mdev->misc_wait);
2173 }
2174}
2175
2176#ifndef __CHECKER__
2177static inline int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins)
2178{
2179 int io_allowed;
2180
2181 /* never get a reference while D_DISKLESS */
2182 if (mdev->state.disk == D_DISKLESS)
2183 return 0;
2184
2185 atomic_inc(&mdev->local_cnt);
2186 io_allowed = (mdev->state.disk >= mins);
2187 if (!io_allowed)
2188 put_ldev(mdev);
2189 return io_allowed;
2190}
2191#else
2192extern int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins);
2193#endif
2194
2195/* you must have an "get_ldev" reference */
2196static inline void drbd_get_syncer_progress(struct drbd_conf *mdev,
2197 unsigned long *bits_left, unsigned int *per_mil_done)
2198{
2199 /* this is to break it at compile time when we change that, in case we
2200 * want to support more than (1<<32) bits on a 32bit arch. */
2201 typecheck(unsigned long, mdev->rs_total);
2202
2203 /* note: both rs_total and rs_left are in bits, i.e. in
2204 * units of BM_BLOCK_SIZE.
2205 * for the percentage, we don't care. */
2206
2207 if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T)
2208 *bits_left = mdev->ov_left;
2209 else
2210 *bits_left = drbd_bm_total_weight(mdev) - mdev->rs_failed;
2211 /* >> 10 to prevent overflow,
2212 * +1 to prevent division by zero */
2213 if (*bits_left > mdev->rs_total) {
2214 /* doh. maybe a logic bug somewhere.
2215 * may also be just a race condition
2216 * between this and a disconnect during sync.
2217 * for now, just prevent in-kernel buffer overflow.
2218 */
2219 smp_rmb();
2220 dev_warn(DEV, "cs:%s rs_left=%lu > rs_total=%lu (rs_failed %lu)\n",
2221 drbd_conn_str(mdev->state.conn),
2222 *bits_left, mdev->rs_total, mdev->rs_failed);
2223 *per_mil_done = 0;
2224 } else {
2225 /* Make sure the division happens in long context.
2226 * We allow up to one petabyte storage right now,
2227 * at a granularity of 4k per bit that is 2**38 bits.
2228 * After shift right and multiplication by 1000,
2229 * this should still fit easily into a 32bit long,
2230 * so we don't need a 64bit division on 32bit arch.
2231 * Note: currently we don't support such large bitmaps on 32bit
2232 * arch anyways, but no harm done to be prepared for it here.
2233 */
2234 unsigned int shift = mdev->rs_total >= (1ULL << 32) ? 16 : 10;
2235 unsigned long left = *bits_left >> shift;
2236 unsigned long total = 1UL + (mdev->rs_total >> shift);
2237 unsigned long tmp = 1000UL - left * 1000UL/total;
2238 *per_mil_done = tmp;
2239 }
2240}
2241
2242
2243/* this throttles on-the-fly application requests
2244 * according to max_buffers settings;
2245 * maybe re-implement using semaphores? */
2246static inline int drbd_get_max_buffers(struct drbd_conf *mdev)
2247{
2248 int mxb = 1000000; /* arbitrary limit on open requests */
2249 if (get_net_conf(mdev)) {
2250 mxb = mdev->net_conf->max_buffers;
2251 put_net_conf(mdev);
2252 }
2253 return mxb;
2254}
2255
2256static inline int drbd_state_is_stable(struct drbd_conf *mdev)
2257{
2258 union drbd_state s = mdev->state;
2259
2260 /* DO NOT add a default clause, we want the compiler to warn us
2261 * for any newly introduced state we may have forgotten to add here */
2262
2263 switch ((enum drbd_conns)s.conn) {
2264 /* new io only accepted when there is no connection, ... */
2265 case C_STANDALONE:
2266 case C_WF_CONNECTION:
2267 /* ... or there is a well established connection. */
2268 case C_CONNECTED:
2269 case C_SYNC_SOURCE:
2270 case C_SYNC_TARGET:
2271 case C_VERIFY_S:
2272 case C_VERIFY_T:
2273 case C_PAUSED_SYNC_S:
2274 case C_PAUSED_SYNC_T:
2275 case C_AHEAD:
2276 case C_BEHIND:
2277 /* transitional states, IO allowed */
2278 case C_DISCONNECTING:
2279 case C_UNCONNECTED:
2280 case C_TIMEOUT:
2281 case C_BROKEN_PIPE:
2282 case C_NETWORK_FAILURE:
2283 case C_PROTOCOL_ERROR:
2284 case C_TEAR_DOWN:
2285 case C_WF_REPORT_PARAMS:
2286 case C_STARTING_SYNC_S:
2287 case C_STARTING_SYNC_T:
2288 break;
2289
2290 /* Allow IO in BM exchange states with new protocols */
2291 case C_WF_BITMAP_S:
2292 if (mdev->agreed_pro_version < 96)
2293 return 0;
2294 break;
2295
2296 /* no new io accepted in these states */
2297 case C_WF_BITMAP_T:
2298 case C_WF_SYNC_UUID:
2299 case C_MASK:
2300 /* not "stable" */
2301 return 0;
2302 }
2303
2304 switch ((enum drbd_disk_state)s.disk) {
2305 case D_DISKLESS:
2306 case D_INCONSISTENT:
2307 case D_OUTDATED:
2308 case D_CONSISTENT:
2309 case D_UP_TO_DATE:
2310 /* disk state is stable as well. */
2311 break;
2312
2313 /* no new io accepted during tansitional states */
2314 case D_ATTACHING:
2315 case D_FAILED:
2316 case D_NEGOTIATING:
2317 case D_UNKNOWN:
2318 case D_MASK:
2319 /* not "stable" */
2320 return 0;
2321 }
2322
2323 return 1;
2324}
2325
2326static inline int is_susp(union drbd_state s)
2327{
2328 return s.susp || s.susp_nod || s.susp_fen;
2329}
2330
2331static inline bool may_inc_ap_bio(struct drbd_conf *mdev)
2332{
2333 int mxb = drbd_get_max_buffers(mdev);
2334
2335 if (is_susp(mdev->state))
2336 return false;
2337 if (test_bit(SUSPEND_IO, &mdev->flags))
2338 return false;
2339
2340 /* to avoid potential deadlock or bitmap corruption,
2341 * in various places, we only allow new application io
2342 * to start during "stable" states. */
2343
2344 /* no new io accepted when attaching or detaching the disk */
2345 if (!drbd_state_is_stable(mdev))
2346 return false;
2347
2348 /* since some older kernels don't have atomic_add_unless,
2349 * and we are within the spinlock anyways, we have this workaround. */
2350 if (atomic_read(&mdev->ap_bio_cnt) > mxb)
2351 return false;
2352 if (test_bit(BITMAP_IO, &mdev->flags))
2353 return false;
2354 return true;
2355}
2356
2357static inline bool inc_ap_bio_cond(struct drbd_conf *mdev, int count)
2358{
2359 bool rv = false;
2360
2361 spin_lock_irq(&mdev->req_lock);
2362 rv = may_inc_ap_bio(mdev);
2363 if (rv)
2364 atomic_add(count, &mdev->ap_bio_cnt);
2365 spin_unlock_irq(&mdev->req_lock);
2366
2367 return rv;
2368}
2369
2370static inline void inc_ap_bio(struct drbd_conf *mdev, int count)
2371{
2372 /* we wait here
2373 * as long as the device is suspended
2374 * until the bitmap is no longer on the fly during connection
2375 * handshake as long as we would exeed the max_buffer limit.
2376 *
2377 * to avoid races with the reconnect code,
2378 * we need to atomic_inc within the spinlock. */
2379
2380 wait_event(mdev->misc_wait, inc_ap_bio_cond(mdev, count));
2381}
2382
2383static inline void dec_ap_bio(struct drbd_conf *mdev)
2384{
2385 int mxb = drbd_get_max_buffers(mdev);
2386 int ap_bio = atomic_dec_return(&mdev->ap_bio_cnt);
2387
2388 D_ASSERT(ap_bio >= 0);
2389 /* this currently does wake_up for every dec_ap_bio!
2390 * maybe rather introduce some type of hysteresis?
2391 * e.g. (ap_bio == mxb/2 || ap_bio == 0) ? */
2392 if (ap_bio < mxb)
2393 wake_up(&mdev->misc_wait);
2394 if (ap_bio == 0 && test_bit(BITMAP_IO, &mdev->flags)) {
2395 if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags))
2396 drbd_queue_work(&mdev->data.work, &mdev->bm_io_work.w);
2397 }
2398}
2399
2400static inline int drbd_set_ed_uuid(struct drbd_conf *mdev, u64 val)
2401{
2402 int changed = mdev->ed_uuid != val;
2403 mdev->ed_uuid = val;
2404 return changed;
2405}
2406
2407static inline int seq_cmp(u32 a, u32 b)
2408{
2409 /* we assume wrap around at 32bit.
2410 * for wrap around at 24bit (old atomic_t),
2411 * we'd have to
2412 * a <<= 8; b <<= 8;
2413 */
2414 return (s32)(a) - (s32)(b);
2415}
2416#define seq_lt(a, b) (seq_cmp((a), (b)) < 0)
2417#define seq_gt(a, b) (seq_cmp((a), (b)) > 0)
2418#define seq_ge(a, b) (seq_cmp((a), (b)) >= 0)
2419#define seq_le(a, b) (seq_cmp((a), (b)) <= 0)
2420/* CAUTION: please no side effects in arguments! */
2421#define seq_max(a, b) ((u32)(seq_gt((a), (b)) ? (a) : (b)))
2422
2423static inline void update_peer_seq(struct drbd_conf *mdev, unsigned int new_seq)
2424{
2425 unsigned int m;
2426 spin_lock(&mdev->peer_seq_lock);
2427 m = seq_max(mdev->peer_seq, new_seq);
2428 mdev->peer_seq = m;
2429 spin_unlock(&mdev->peer_seq_lock);
2430 if (m == new_seq)
2431 wake_up(&mdev->seq_wait);
2432}
2433
2434static inline void drbd_update_congested(struct drbd_conf *mdev)
2435{
2436 struct sock *sk = mdev->data.socket->sk;
2437 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
2438 set_bit(NET_CONGESTED, &mdev->flags);
2439}
2440
2441static inline int drbd_queue_order_type(struct drbd_conf *mdev)
2442{
2443 /* sorry, we currently have no working implementation
2444 * of distributed TCQ stuff */
2445#ifndef QUEUE_ORDERED_NONE
2446#define QUEUE_ORDERED_NONE 0
2447#endif
2448 return QUEUE_ORDERED_NONE;
2449}
2450
2451static inline void drbd_md_flush(struct drbd_conf *mdev)
2452{
2453 int r;
2454
2455 if (test_bit(MD_NO_FUA, &mdev->flags))
2456 return;
2457
2458 r = blkdev_issue_flush(mdev->ldev->md_bdev, GFP_KERNEL, NULL);
2459 if (r) {
2460 set_bit(MD_NO_FUA, &mdev->flags);
2461 dev_err(DEV, "meta data flush failed with status %d, disabling md-flushes\n", r);
2462 }
2463}
2464
2465#endif
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 <crypto/hash.h>
30#include <linux/compiler.h>
31#include <linux/types.h>
32#include <linux/list.h>
33#include <linux/sched/signal.h>
34#include <linux/bitops.h>
35#include <linux/slab.h>
36#include <linux/ratelimit.h>
37#include <linux/tcp.h>
38#include <linux/mutex.h>
39#include <linux/major.h>
40#include <linux/blkdev.h>
41#include <linux/backing-dev.h>
42#include <linux/genhd.h>
43#include <linux/idr.h>
44#include <linux/dynamic_debug.h>
45#include <net/tcp.h>
46#include <linux/lru_cache.h>
47#include <linux/prefetch.h>
48#include <linux/drbd_genl_api.h>
49#include <linux/drbd.h>
50#include "drbd_strings.h"
51#include "drbd_state.h"
52#include "drbd_protocol.h"
53
54#ifdef __CHECKER__
55# define __protected_by(x) __attribute__((require_context(x,1,999,"rdwr")))
56# define __protected_read_by(x) __attribute__((require_context(x,1,999,"read")))
57# define __protected_write_by(x) __attribute__((require_context(x,1,999,"write")))
58# define __must_hold(x) __attribute__((context(x,1,1), require_context(x,1,999,"call")))
59#else
60# define __protected_by(x)
61# define __protected_read_by(x)
62# define __protected_write_by(x)
63# define __must_hold(x)
64#endif
65
66/* shared module parameters, defined in drbd_main.c */
67#ifdef CONFIG_DRBD_FAULT_INJECTION
68extern int drbd_enable_faults;
69extern int drbd_fault_rate;
70#endif
71
72extern unsigned int drbd_minor_count;
73extern char drbd_usermode_helper[];
74extern int drbd_proc_details;
75
76
77/* This is used to stop/restart our threads.
78 * Cannot use SIGTERM nor SIGKILL, since these
79 * are sent out by init on runlevel changes
80 * I choose SIGHUP for now.
81 */
82#define DRBD_SIGKILL SIGHUP
83
84#define ID_IN_SYNC (4711ULL)
85#define ID_OUT_OF_SYNC (4712ULL)
86#define ID_SYNCER (-1ULL)
87
88#define UUID_NEW_BM_OFFSET ((u64)0x0001000000000000ULL)
89
90struct drbd_device;
91struct drbd_connection;
92
93#define __drbd_printk_device(level, device, fmt, args...) \
94 dev_printk(level, disk_to_dev((device)->vdisk), fmt, ## args)
95#define __drbd_printk_peer_device(level, peer_device, fmt, args...) \
96 dev_printk(level, disk_to_dev((peer_device)->device->vdisk), fmt, ## args)
97#define __drbd_printk_resource(level, resource, fmt, args...) \
98 printk(level "drbd %s: " fmt, (resource)->name, ## args)
99#define __drbd_printk_connection(level, connection, fmt, args...) \
100 printk(level "drbd %s: " fmt, (connection)->resource->name, ## args)
101
102void drbd_printk_with_wrong_object_type(void);
103
104#define __drbd_printk_if_same_type(obj, type, func, level, fmt, args...) \
105 (__builtin_types_compatible_p(typeof(obj), type) || \
106 __builtin_types_compatible_p(typeof(obj), const type)), \
107 func(level, (const type)(obj), fmt, ## args)
108
109#define drbd_printk(level, obj, fmt, args...) \
110 __builtin_choose_expr( \
111 __drbd_printk_if_same_type(obj, struct drbd_device *, \
112 __drbd_printk_device, level, fmt, ## args), \
113 __builtin_choose_expr( \
114 __drbd_printk_if_same_type(obj, struct drbd_resource *, \
115 __drbd_printk_resource, level, fmt, ## args), \
116 __builtin_choose_expr( \
117 __drbd_printk_if_same_type(obj, struct drbd_connection *, \
118 __drbd_printk_connection, level, fmt, ## args), \
119 __builtin_choose_expr( \
120 __drbd_printk_if_same_type(obj, struct drbd_peer_device *, \
121 __drbd_printk_peer_device, level, fmt, ## args), \
122 drbd_printk_with_wrong_object_type()))))
123
124#define drbd_dbg(obj, fmt, args...) \
125 drbd_printk(KERN_DEBUG, obj, fmt, ## args)
126#define drbd_alert(obj, fmt, args...) \
127 drbd_printk(KERN_ALERT, obj, fmt, ## args)
128#define drbd_err(obj, fmt, args...) \
129 drbd_printk(KERN_ERR, obj, fmt, ## args)
130#define drbd_warn(obj, fmt, args...) \
131 drbd_printk(KERN_WARNING, obj, fmt, ## args)
132#define drbd_info(obj, fmt, args...) \
133 drbd_printk(KERN_INFO, obj, fmt, ## args)
134#define drbd_emerg(obj, fmt, args...) \
135 drbd_printk(KERN_EMERG, obj, fmt, ## args)
136
137#define dynamic_drbd_dbg(device, fmt, args...) \
138 dynamic_dev_dbg(disk_to_dev(device->vdisk), fmt, ## args)
139
140#define D_ASSERT(device, exp) do { \
141 if (!(exp)) \
142 drbd_err(device, "ASSERT( " #exp " ) in %s:%d\n", __FILE__, __LINE__); \
143 } while (0)
144
145/**
146 * expect - Make an assertion
147 *
148 * Unlike the assert macro, this macro returns a boolean result.
149 */
150#define expect(exp) ({ \
151 bool _bool = (exp); \
152 if (!_bool) \
153 drbd_err(device, "ASSERTION %s FAILED in %s\n", \
154 #exp, __func__); \
155 _bool; \
156 })
157
158/* Defines to control fault insertion */
159enum {
160 DRBD_FAULT_MD_WR = 0, /* meta data write */
161 DRBD_FAULT_MD_RD = 1, /* read */
162 DRBD_FAULT_RS_WR = 2, /* resync */
163 DRBD_FAULT_RS_RD = 3,
164 DRBD_FAULT_DT_WR = 4, /* data */
165 DRBD_FAULT_DT_RD = 5,
166 DRBD_FAULT_DT_RA = 6, /* data read ahead */
167 DRBD_FAULT_BM_ALLOC = 7, /* bitmap allocation */
168 DRBD_FAULT_AL_EE = 8, /* alloc ee */
169 DRBD_FAULT_RECEIVE = 9, /* Changes some bytes upon receiving a [rs]data block */
170
171 DRBD_FAULT_MAX,
172};
173
174extern unsigned int
175_drbd_insert_fault(struct drbd_device *device, unsigned int type);
176
177static inline int
178drbd_insert_fault(struct drbd_device *device, unsigned int type) {
179#ifdef CONFIG_DRBD_FAULT_INJECTION
180 return drbd_fault_rate &&
181 (drbd_enable_faults & (1<<type)) &&
182 _drbd_insert_fault(device, type);
183#else
184 return 0;
185#endif
186}
187
188/* integer division, round _UP_ to the next integer */
189#define div_ceil(A, B) ((A)/(B) + ((A)%(B) ? 1 : 0))
190/* usual integer division */
191#define div_floor(A, B) ((A)/(B))
192
193extern struct ratelimit_state drbd_ratelimit_state;
194extern struct idr drbd_devices; /* RCU, updates: genl_lock() */
195extern struct list_head drbd_resources; /* RCU, updates: genl_lock() */
196
197extern const char *cmdname(enum drbd_packet cmd);
198
199/* for sending/receiving the bitmap,
200 * possibly in some encoding scheme */
201struct bm_xfer_ctx {
202 /* "const"
203 * stores total bits and long words
204 * of the bitmap, so we don't need to
205 * call the accessor functions over and again. */
206 unsigned long bm_bits;
207 unsigned long bm_words;
208 /* during xfer, current position within the bitmap */
209 unsigned long bit_offset;
210 unsigned long word_offset;
211
212 /* statistics; index: (h->command == P_BITMAP) */
213 unsigned packets[2];
214 unsigned bytes[2];
215};
216
217extern void INFO_bm_xfer_stats(struct drbd_device *device,
218 const char *direction, struct bm_xfer_ctx *c);
219
220static inline void bm_xfer_ctx_bit_to_word_offset(struct bm_xfer_ctx *c)
221{
222 /* word_offset counts "native long words" (32 or 64 bit),
223 * aligned at 64 bit.
224 * Encoded packet may end at an unaligned bit offset.
225 * In case a fallback clear text packet is transmitted in
226 * between, we adjust this offset back to the last 64bit
227 * aligned "native long word", which makes coding and decoding
228 * the plain text bitmap much more convenient. */
229#if BITS_PER_LONG == 64
230 c->word_offset = c->bit_offset >> 6;
231#elif BITS_PER_LONG == 32
232 c->word_offset = c->bit_offset >> 5;
233 c->word_offset &= ~(1UL);
234#else
235# error "unsupported BITS_PER_LONG"
236#endif
237}
238
239extern unsigned int drbd_header_size(struct drbd_connection *connection);
240
241/**********************************************************************/
242enum drbd_thread_state {
243 NONE,
244 RUNNING,
245 EXITING,
246 RESTARTING
247};
248
249struct drbd_thread {
250 spinlock_t t_lock;
251 struct task_struct *task;
252 struct completion stop;
253 enum drbd_thread_state t_state;
254 int (*function) (struct drbd_thread *);
255 struct drbd_resource *resource;
256 struct drbd_connection *connection;
257 int reset_cpu_mask;
258 const char *name;
259};
260
261static inline enum drbd_thread_state get_t_state(struct drbd_thread *thi)
262{
263 /* THINK testing the t_state seems to be uncritical in all cases
264 * (but thread_{start,stop}), so we can read it *without* the lock.
265 * --lge */
266
267 smp_rmb();
268 return thi->t_state;
269}
270
271struct drbd_work {
272 struct list_head list;
273 int (*cb)(struct drbd_work *, int cancel);
274};
275
276struct drbd_device_work {
277 struct drbd_work w;
278 struct drbd_device *device;
279};
280
281#include "drbd_interval.h"
282
283extern int drbd_wait_misc(struct drbd_device *, struct drbd_interval *);
284
285extern void lock_all_resources(void);
286extern void unlock_all_resources(void);
287
288struct drbd_request {
289 struct drbd_work w;
290 struct drbd_device *device;
291
292 /* if local IO is not allowed, will be NULL.
293 * if local IO _is_ allowed, holds the locally submitted bio clone,
294 * or, after local IO completion, the ERR_PTR(error).
295 * see drbd_request_endio(). */
296 struct bio *private_bio;
297
298 struct drbd_interval i;
299
300 /* epoch: used to check on "completion" whether this req was in
301 * the current epoch, and we therefore have to close it,
302 * causing a p_barrier packet to be send, starting a new epoch.
303 *
304 * This corresponds to "barrier" in struct p_barrier[_ack],
305 * and to "barrier_nr" in struct drbd_epoch (and various
306 * comments/function parameters/local variable names).
307 */
308 unsigned int epoch;
309
310 struct list_head tl_requests; /* ring list in the transfer log */
311 struct bio *master_bio; /* master bio pointer */
312
313 /* see struct drbd_device */
314 struct list_head req_pending_master_completion;
315 struct list_head req_pending_local;
316
317 /* for generic IO accounting */
318 unsigned long start_jif;
319
320 /* for DRBD internal statistics */
321
322 /* Minimal set of time stamps to determine if we wait for activity log
323 * transactions, local disk or peer. 32 bit "jiffies" are good enough,
324 * we don't expect a DRBD request to be stalled for several month.
325 */
326
327 /* before actual request processing */
328 unsigned long in_actlog_jif;
329
330 /* local disk */
331 unsigned long pre_submit_jif;
332
333 /* per connection */
334 unsigned long pre_send_jif;
335 unsigned long acked_jif;
336 unsigned long net_done_jif;
337
338 /* Possibly even more detail to track each phase:
339 * master_completion_jif
340 * how long did it take to complete the master bio
341 * (application visible latency)
342 * allocated_jif
343 * how long the master bio was blocked until we finally allocated
344 * a tracking struct
345 * in_actlog_jif
346 * how long did we wait for activity log transactions
347 *
348 * net_queued_jif
349 * when did we finally queue it for sending
350 * pre_send_jif
351 * when did we start sending it
352 * post_send_jif
353 * how long did we block in the network stack trying to send it
354 * acked_jif
355 * when did we receive (or fake, in protocol A) a remote ACK
356 * net_done_jif
357 * when did we receive final acknowledgement (P_BARRIER_ACK),
358 * or decide, e.g. on connection loss, that we do no longer expect
359 * anything from this peer for this request.
360 *
361 * pre_submit_jif
362 * post_sub_jif
363 * when did we start submiting to the lower level device,
364 * and how long did we block in that submit function
365 * local_completion_jif
366 * how long did it take the lower level device to complete this request
367 */
368
369
370 /* once it hits 0, we may complete the master_bio */
371 atomic_t completion_ref;
372 /* once it hits 0, we may destroy this drbd_request object */
373 struct kref kref;
374
375 unsigned rq_state; /* see comments above _req_mod() */
376};
377
378struct drbd_epoch {
379 struct drbd_connection *connection;
380 struct list_head list;
381 unsigned int barrier_nr;
382 atomic_t epoch_size; /* increased on every request added. */
383 atomic_t active; /* increased on every req. added, and dec on every finished. */
384 unsigned long flags;
385};
386
387/* Prototype declaration of function defined in drbd_receiver.c */
388int drbdd_init(struct drbd_thread *);
389int drbd_asender(struct drbd_thread *);
390
391/* drbd_epoch flag bits */
392enum {
393 DE_HAVE_BARRIER_NUMBER,
394};
395
396enum epoch_event {
397 EV_PUT,
398 EV_GOT_BARRIER_NR,
399 EV_BECAME_LAST,
400 EV_CLEANUP = 32, /* used as flag */
401};
402
403struct digest_info {
404 int digest_size;
405 void *digest;
406};
407
408struct drbd_peer_request {
409 struct drbd_work w;
410 struct drbd_peer_device *peer_device;
411 struct drbd_epoch *epoch; /* for writes */
412 struct page *pages;
413 atomic_t pending_bios;
414 struct drbd_interval i;
415 /* see comments on ee flag bits below */
416 unsigned long flags;
417 unsigned long submit_jif;
418 union {
419 u64 block_id;
420 struct digest_info *digest;
421 };
422};
423
424/* ee flag bits.
425 * While corresponding bios are in flight, the only modification will be
426 * set_bit WAS_ERROR, which has to be atomic.
427 * If no bios are in flight yet, or all have been completed,
428 * non-atomic modification to ee->flags is ok.
429 */
430enum {
431 __EE_CALL_AL_COMPLETE_IO,
432 __EE_MAY_SET_IN_SYNC,
433
434 /* is this a TRIM aka REQ_DISCARD? */
435 __EE_IS_TRIM,
436
437 /* In case a barrier failed,
438 * we need to resubmit without the barrier flag. */
439 __EE_RESUBMITTED,
440
441 /* we may have several bios per peer request.
442 * if any of those fail, we set this flag atomically
443 * from the endio callback */
444 __EE_WAS_ERROR,
445
446 /* This ee has a pointer to a digest instead of a block id */
447 __EE_HAS_DIGEST,
448
449 /* Conflicting local requests need to be restarted after this request */
450 __EE_RESTART_REQUESTS,
451
452 /* The peer wants a write ACK for this (wire proto C) */
453 __EE_SEND_WRITE_ACK,
454
455 /* Is set when net_conf had two_primaries set while creating this peer_req */
456 __EE_IN_INTERVAL_TREE,
457
458 /* for debugfs: */
459 /* has this been submitted, or does it still wait for something else? */
460 __EE_SUBMITTED,
461
462 /* this is/was a write request */
463 __EE_WRITE,
464
465 /* this is/was a write same request */
466 __EE_WRITE_SAME,
467
468 /* this originates from application on peer
469 * (not some resync or verify or other DRBD internal request) */
470 __EE_APPLICATION,
471
472 /* If it contains only 0 bytes, send back P_RS_DEALLOCATED */
473 __EE_RS_THIN_REQ,
474};
475#define EE_CALL_AL_COMPLETE_IO (1<<__EE_CALL_AL_COMPLETE_IO)
476#define EE_MAY_SET_IN_SYNC (1<<__EE_MAY_SET_IN_SYNC)
477#define EE_IS_TRIM (1<<__EE_IS_TRIM)
478#define EE_RESUBMITTED (1<<__EE_RESUBMITTED)
479#define EE_WAS_ERROR (1<<__EE_WAS_ERROR)
480#define EE_HAS_DIGEST (1<<__EE_HAS_DIGEST)
481#define EE_RESTART_REQUESTS (1<<__EE_RESTART_REQUESTS)
482#define EE_SEND_WRITE_ACK (1<<__EE_SEND_WRITE_ACK)
483#define EE_IN_INTERVAL_TREE (1<<__EE_IN_INTERVAL_TREE)
484#define EE_SUBMITTED (1<<__EE_SUBMITTED)
485#define EE_WRITE (1<<__EE_WRITE)
486#define EE_WRITE_SAME (1<<__EE_WRITE_SAME)
487#define EE_APPLICATION (1<<__EE_APPLICATION)
488#define EE_RS_THIN_REQ (1<<__EE_RS_THIN_REQ)
489
490/* flag bits per device */
491enum {
492 UNPLUG_REMOTE, /* sending a "UnplugRemote" could help */
493 MD_DIRTY, /* current uuids and flags not yet on disk */
494 USE_DEGR_WFC_T, /* degr-wfc-timeout instead of wfc-timeout. */
495 CL_ST_CHG_SUCCESS,
496 CL_ST_CHG_FAIL,
497 CRASHED_PRIMARY, /* This node was a crashed primary.
498 * Gets cleared when the state.conn
499 * goes into C_CONNECTED state. */
500 CONSIDER_RESYNC,
501
502 MD_NO_FUA, /* Users wants us to not use FUA/FLUSH on meta data dev */
503
504 BITMAP_IO, /* suspend application io;
505 once no more io in flight, start bitmap io */
506 BITMAP_IO_QUEUED, /* Started bitmap IO */
507 WAS_IO_ERROR, /* Local disk failed, returned IO error */
508 WAS_READ_ERROR, /* Local disk READ failed (set additionally to the above) */
509 FORCE_DETACH, /* Force-detach from local disk, aborting any pending local IO */
510 RESYNC_AFTER_NEG, /* Resync after online grow after the attach&negotiate finished. */
511 RESIZE_PENDING, /* Size change detected locally, waiting for the response from
512 * the peer, if it changed there as well. */
513 NEW_CUR_UUID, /* Create new current UUID when thawing IO */
514 AL_SUSPENDED, /* Activity logging is currently suspended. */
515 AHEAD_TO_SYNC_SOURCE, /* Ahead -> SyncSource queued */
516 B_RS_H_DONE, /* Before resync handler done (already executed) */
517 DISCARD_MY_DATA, /* discard_my_data flag per volume */
518 READ_BALANCE_RR,
519
520 FLUSH_PENDING, /* if set, device->flush_jif is when we submitted that flush
521 * from drbd_flush_after_epoch() */
522
523 /* cleared only after backing device related structures have been destroyed. */
524 GOING_DISKLESS, /* Disk is being detached, because of io-error, or admin request. */
525
526 /* to be used in drbd_device_post_work() */
527 GO_DISKLESS, /* tell worker to schedule cleanup before detach */
528 DESTROY_DISK, /* tell worker to close backing devices and destroy related structures. */
529 MD_SYNC, /* tell worker to call drbd_md_sync() */
530 RS_START, /* tell worker to start resync/OV */
531 RS_PROGRESS, /* tell worker that resync made significant progress */
532 RS_DONE, /* tell worker that resync is done */
533};
534
535struct drbd_bitmap; /* opaque for drbd_device */
536
537/* definition of bits in bm_flags to be used in drbd_bm_lock
538 * and drbd_bitmap_io and friends. */
539enum bm_flag {
540 /* currently locked for bulk operation */
541 BM_LOCKED_MASK = 0xf,
542
543 /* in detail, that is: */
544 BM_DONT_CLEAR = 0x1,
545 BM_DONT_SET = 0x2,
546 BM_DONT_TEST = 0x4,
547
548 /* so we can mark it locked for bulk operation,
549 * and still allow all non-bulk operations */
550 BM_IS_LOCKED = 0x8,
551
552 /* (test bit, count bit) allowed (common case) */
553 BM_LOCKED_TEST_ALLOWED = BM_DONT_CLEAR | BM_DONT_SET | BM_IS_LOCKED,
554
555 /* testing bits, as well as setting new bits allowed, but clearing bits
556 * would be unexpected. Used during bitmap receive. Setting new bits
557 * requires sending of "out-of-sync" information, though. */
558 BM_LOCKED_SET_ALLOWED = BM_DONT_CLEAR | BM_IS_LOCKED,
559
560 /* for drbd_bm_write_copy_pages, everything is allowed,
561 * only concurrent bulk operations are locked out. */
562 BM_LOCKED_CHANGE_ALLOWED = BM_IS_LOCKED,
563};
564
565struct drbd_work_queue {
566 struct list_head q;
567 spinlock_t q_lock; /* to protect the list. */
568 wait_queue_head_t q_wait;
569};
570
571struct drbd_socket {
572 struct mutex mutex;
573 struct socket *socket;
574 /* this way we get our
575 * send/receive buffers off the stack */
576 void *sbuf;
577 void *rbuf;
578};
579
580struct drbd_md {
581 u64 md_offset; /* sector offset to 'super' block */
582
583 u64 la_size_sect; /* last agreed size, unit sectors */
584 spinlock_t uuid_lock;
585 u64 uuid[UI_SIZE];
586 u64 device_uuid;
587 u32 flags;
588 u32 md_size_sect;
589
590 s32 al_offset; /* signed relative sector offset to activity log */
591 s32 bm_offset; /* signed relative sector offset to bitmap */
592
593 /* cached value of bdev->disk_conf->meta_dev_idx (see below) */
594 s32 meta_dev_idx;
595
596 /* see al_tr_number_to_on_disk_sector() */
597 u32 al_stripes;
598 u32 al_stripe_size_4k;
599 u32 al_size_4k; /* cached product of the above */
600};
601
602struct drbd_backing_dev {
603 struct block_device *backing_bdev;
604 struct block_device *md_bdev;
605 struct drbd_md md;
606 struct disk_conf *disk_conf; /* RCU, for updates: resource->conf_update */
607 sector_t known_size; /* last known size of that backing device */
608};
609
610struct drbd_md_io {
611 struct page *page;
612 unsigned long start_jif; /* last call to drbd_md_get_buffer */
613 unsigned long submit_jif; /* last _drbd_md_sync_page_io() submit */
614 const char *current_use;
615 atomic_t in_use;
616 unsigned int done;
617 int error;
618};
619
620struct bm_io_work {
621 struct drbd_work w;
622 char *why;
623 enum bm_flag flags;
624 int (*io_fn)(struct drbd_device *device);
625 void (*done)(struct drbd_device *device, int rv);
626};
627
628struct fifo_buffer {
629 unsigned int head_index;
630 unsigned int size;
631 int total; /* sum of all values */
632 int values[0];
633};
634extern struct fifo_buffer *fifo_alloc(int fifo_size);
635
636/* flag bits per connection */
637enum {
638 NET_CONGESTED, /* The data socket is congested */
639 RESOLVE_CONFLICTS, /* Set on one node, cleared on the peer! */
640 SEND_PING,
641 GOT_PING_ACK, /* set when we receive a ping_ack packet, ping_wait gets woken */
642 CONN_WD_ST_CHG_REQ, /* A cluster wide state change on the connection is active */
643 CONN_WD_ST_CHG_OKAY,
644 CONN_WD_ST_CHG_FAIL,
645 CONN_DRY_RUN, /* Expect disconnect after resync handshake. */
646 CREATE_BARRIER, /* next P_DATA is preceded by a P_BARRIER */
647 STATE_SENT, /* Do not change state/UUIDs while this is set */
648 CALLBACK_PENDING, /* Whether we have a call_usermodehelper(, UMH_WAIT_PROC)
649 * pending, from drbd worker context.
650 * If set, bdi_write_congested() returns true,
651 * so shrink_page_list() would not recurse into,
652 * and potentially deadlock on, this drbd worker.
653 */
654 DISCONNECT_SENT,
655
656 DEVICE_WORK_PENDING, /* tell worker that some device has pending work */
657};
658
659enum which_state { NOW, OLD = NOW, NEW };
660
661struct drbd_resource {
662 char *name;
663#ifdef CONFIG_DEBUG_FS
664 struct dentry *debugfs_res;
665 struct dentry *debugfs_res_volumes;
666 struct dentry *debugfs_res_connections;
667 struct dentry *debugfs_res_in_flight_summary;
668#endif
669 struct kref kref;
670 struct idr devices; /* volume number to device mapping */
671 struct list_head connections;
672 struct list_head resources;
673 struct res_opts res_opts;
674 struct mutex conf_update; /* mutex for ready-copy-update of net_conf and disk_conf */
675 struct mutex adm_mutex; /* mutex to serialize administrative requests */
676 spinlock_t req_lock;
677
678 unsigned susp:1; /* IO suspended by user */
679 unsigned susp_nod:1; /* IO suspended because no data */
680 unsigned susp_fen:1; /* IO suspended because fence peer handler runs */
681
682 enum write_ordering_e write_ordering;
683
684 cpumask_var_t cpu_mask;
685};
686
687struct drbd_thread_timing_details
688{
689 unsigned long start_jif;
690 void *cb_addr;
691 const char *caller_fn;
692 unsigned int line;
693 unsigned int cb_nr;
694};
695
696struct drbd_connection {
697 struct list_head connections;
698 struct drbd_resource *resource;
699#ifdef CONFIG_DEBUG_FS
700 struct dentry *debugfs_conn;
701 struct dentry *debugfs_conn_callback_history;
702 struct dentry *debugfs_conn_oldest_requests;
703#endif
704 struct kref kref;
705 struct idr peer_devices; /* volume number to peer device mapping */
706 enum drbd_conns cstate; /* Only C_STANDALONE to C_WF_REPORT_PARAMS */
707 struct mutex cstate_mutex; /* Protects graceful disconnects */
708 unsigned int connect_cnt; /* Inc each time a connection is established */
709
710 unsigned long flags;
711 struct net_conf *net_conf; /* content protected by rcu */
712 wait_queue_head_t ping_wait; /* Woken upon reception of a ping, and a state change */
713
714 struct sockaddr_storage my_addr;
715 int my_addr_len;
716 struct sockaddr_storage peer_addr;
717 int peer_addr_len;
718
719 struct drbd_socket data; /* data/barrier/cstate/parameter packets */
720 struct drbd_socket meta; /* ping/ack (metadata) packets */
721 int agreed_pro_version; /* actually used protocol version */
722 u32 agreed_features;
723 unsigned long last_received; /* in jiffies, either socket */
724 unsigned int ko_count;
725
726 struct list_head transfer_log; /* all requests not yet fully processed */
727
728 struct crypto_shash *cram_hmac_tfm;
729 struct crypto_ahash *integrity_tfm; /* checksums we compute, updates protected by connection->data->mutex */
730 struct crypto_ahash *peer_integrity_tfm; /* checksums we verify, only accessed from receiver thread */
731 struct crypto_ahash *csums_tfm;
732 struct crypto_ahash *verify_tfm;
733 void *int_dig_in;
734 void *int_dig_vv;
735
736 /* receiver side */
737 struct drbd_epoch *current_epoch;
738 spinlock_t epoch_lock;
739 unsigned int epochs;
740 atomic_t current_tle_nr; /* transfer log epoch number */
741 unsigned current_tle_writes; /* writes seen within this tl epoch */
742
743 unsigned long last_reconnect_jif;
744 /* empty member on older kernels without blk_start_plug() */
745 struct blk_plug receiver_plug;
746 struct drbd_thread receiver;
747 struct drbd_thread worker;
748 struct drbd_thread ack_receiver;
749 struct workqueue_struct *ack_sender;
750
751 /* cached pointers,
752 * so we can look up the oldest pending requests more quickly.
753 * protected by resource->req_lock */
754 struct drbd_request *req_next; /* DRBD 9: todo.req_next */
755 struct drbd_request *req_ack_pending;
756 struct drbd_request *req_not_net_done;
757
758 /* sender side */
759 struct drbd_work_queue sender_work;
760
761#define DRBD_THREAD_DETAILS_HIST 16
762 unsigned int w_cb_nr; /* keeps counting up */
763 unsigned int r_cb_nr; /* keeps counting up */
764 struct drbd_thread_timing_details w_timing_details[DRBD_THREAD_DETAILS_HIST];
765 struct drbd_thread_timing_details r_timing_details[DRBD_THREAD_DETAILS_HIST];
766
767 struct {
768 unsigned long last_sent_barrier_jif;
769
770 /* whether this sender thread
771 * has processed a single write yet. */
772 bool seen_any_write_yet;
773
774 /* Which barrier number to send with the next P_BARRIER */
775 int current_epoch_nr;
776
777 /* how many write requests have been sent
778 * with req->epoch == current_epoch_nr.
779 * If none, no P_BARRIER will be sent. */
780 unsigned current_epoch_writes;
781 } send;
782};
783
784static inline bool has_net_conf(struct drbd_connection *connection)
785{
786 bool has_net_conf;
787
788 rcu_read_lock();
789 has_net_conf = rcu_dereference(connection->net_conf);
790 rcu_read_unlock();
791
792 return has_net_conf;
793}
794
795void __update_timing_details(
796 struct drbd_thread_timing_details *tdp,
797 unsigned int *cb_nr,
798 void *cb,
799 const char *fn, const unsigned int line);
800
801#define update_worker_timing_details(c, cb) \
802 __update_timing_details(c->w_timing_details, &c->w_cb_nr, cb, __func__ , __LINE__ )
803#define update_receiver_timing_details(c, cb) \
804 __update_timing_details(c->r_timing_details, &c->r_cb_nr, cb, __func__ , __LINE__ )
805
806struct submit_worker {
807 struct workqueue_struct *wq;
808 struct work_struct worker;
809
810 /* protected by ..->resource->req_lock */
811 struct list_head writes;
812};
813
814struct drbd_peer_device {
815 struct list_head peer_devices;
816 struct drbd_device *device;
817 struct drbd_connection *connection;
818 struct work_struct send_acks_work;
819#ifdef CONFIG_DEBUG_FS
820 struct dentry *debugfs_peer_dev;
821#endif
822};
823
824struct drbd_device {
825 struct drbd_resource *resource;
826 struct list_head peer_devices;
827 struct list_head pending_bitmap_io;
828
829 unsigned long flush_jif;
830#ifdef CONFIG_DEBUG_FS
831 struct dentry *debugfs_minor;
832 struct dentry *debugfs_vol;
833 struct dentry *debugfs_vol_oldest_requests;
834 struct dentry *debugfs_vol_act_log_extents;
835 struct dentry *debugfs_vol_resync_extents;
836 struct dentry *debugfs_vol_data_gen_id;
837 struct dentry *debugfs_vol_ed_gen_id;
838#endif
839
840 unsigned int vnr; /* volume number within the connection */
841 unsigned int minor; /* device minor number */
842
843 struct kref kref;
844
845 /* things that are stored as / read from meta data on disk */
846 unsigned long flags;
847
848 /* configured by drbdsetup */
849 struct drbd_backing_dev *ldev __protected_by(local);
850
851 sector_t p_size; /* partner's disk size */
852 struct request_queue *rq_queue;
853 struct block_device *this_bdev;
854 struct gendisk *vdisk;
855
856 unsigned long last_reattach_jif;
857 struct drbd_work resync_work;
858 struct drbd_work unplug_work;
859 struct timer_list resync_timer;
860 struct timer_list md_sync_timer;
861 struct timer_list start_resync_timer;
862 struct timer_list request_timer;
863
864 /* Used after attach while negotiating new disk state. */
865 union drbd_state new_state_tmp;
866
867 union drbd_dev_state state;
868 wait_queue_head_t misc_wait;
869 wait_queue_head_t state_wait; /* upon each state change. */
870 unsigned int send_cnt;
871 unsigned int recv_cnt;
872 unsigned int read_cnt;
873 unsigned int writ_cnt;
874 unsigned int al_writ_cnt;
875 unsigned int bm_writ_cnt;
876 atomic_t ap_bio_cnt; /* Requests we need to complete */
877 atomic_t ap_actlog_cnt; /* Requests waiting for activity log */
878 atomic_t ap_pending_cnt; /* AP data packets on the wire, ack expected */
879 atomic_t rs_pending_cnt; /* RS request/data packets on the wire */
880 atomic_t unacked_cnt; /* Need to send replies for */
881 atomic_t local_cnt; /* Waiting for local completion */
882 atomic_t suspend_cnt;
883
884 /* Interval tree of pending local requests */
885 struct rb_root read_requests;
886 struct rb_root write_requests;
887
888 /* for statistics and timeouts */
889 /* [0] read, [1] write */
890 struct list_head pending_master_completion[2];
891 struct list_head pending_completion[2];
892
893 /* use checksums for *this* resync */
894 bool use_csums;
895 /* blocks to resync in this run [unit BM_BLOCK_SIZE] */
896 unsigned long rs_total;
897 /* number of resync blocks that failed in this run */
898 unsigned long rs_failed;
899 /* Syncer's start time [unit jiffies] */
900 unsigned long rs_start;
901 /* cumulated time in PausedSyncX state [unit jiffies] */
902 unsigned long rs_paused;
903 /* skipped because csum was equal [unit BM_BLOCK_SIZE] */
904 unsigned long rs_same_csum;
905#define DRBD_SYNC_MARKS 8
906#define DRBD_SYNC_MARK_STEP (3*HZ)
907 /* block not up-to-date at mark [unit BM_BLOCK_SIZE] */
908 unsigned long rs_mark_left[DRBD_SYNC_MARKS];
909 /* marks's time [unit jiffies] */
910 unsigned long rs_mark_time[DRBD_SYNC_MARKS];
911 /* current index into rs_mark_{left,time} */
912 int rs_last_mark;
913 unsigned long rs_last_bcast; /* [unit jiffies] */
914
915 /* where does the admin want us to start? (sector) */
916 sector_t ov_start_sector;
917 sector_t ov_stop_sector;
918 /* where are we now? (sector) */
919 sector_t ov_position;
920 /* Start sector of out of sync range (to merge printk reporting). */
921 sector_t ov_last_oos_start;
922 /* size of out-of-sync range in sectors. */
923 sector_t ov_last_oos_size;
924 unsigned long ov_left; /* in bits */
925
926 struct drbd_bitmap *bitmap;
927 unsigned long bm_resync_fo; /* bit offset for drbd_bm_find_next */
928
929 /* Used to track operations of resync... */
930 struct lru_cache *resync;
931 /* Number of locked elements in resync LRU */
932 unsigned int resync_locked;
933 /* resync extent number waiting for application requests */
934 unsigned int resync_wenr;
935
936 int open_cnt;
937 u64 *p_uuid;
938
939 struct list_head active_ee; /* IO in progress (P_DATA gets written to disk) */
940 struct list_head sync_ee; /* IO in progress (P_RS_DATA_REPLY gets written to disk) */
941 struct list_head done_ee; /* need to send P_WRITE_ACK */
942 struct list_head read_ee; /* [RS]P_DATA_REQUEST being read */
943 struct list_head net_ee; /* zero-copy network send in progress */
944
945 int next_barrier_nr;
946 struct list_head resync_reads;
947 atomic_t pp_in_use; /* allocated from page pool */
948 atomic_t pp_in_use_by_net; /* sendpage()d, still referenced by tcp */
949 wait_queue_head_t ee_wait;
950 struct drbd_md_io md_io;
951 spinlock_t al_lock;
952 wait_queue_head_t al_wait;
953 struct lru_cache *act_log; /* activity log */
954 unsigned int al_tr_number;
955 int al_tr_cycle;
956 wait_queue_head_t seq_wait;
957 atomic_t packet_seq;
958 unsigned int peer_seq;
959 spinlock_t peer_seq_lock;
960 unsigned long comm_bm_set; /* communicated number of set bits. */
961 struct bm_io_work bm_io_work;
962 u64 ed_uuid; /* UUID of the exposed data */
963 struct mutex own_state_mutex;
964 struct mutex *state_mutex; /* either own_state_mutex or first_peer_device(device)->connection->cstate_mutex */
965 char congestion_reason; /* Why we where congested... */
966 atomic_t rs_sect_in; /* for incoming resync data rate, SyncTarget */
967 atomic_t rs_sect_ev; /* for submitted resync data rate, both */
968 int rs_last_sect_ev; /* counter to compare with */
969 int rs_last_events; /* counter of read or write "events" (unit sectors)
970 * on the lower level device when we last looked. */
971 int c_sync_rate; /* current resync rate after syncer throttle magic */
972 struct fifo_buffer *rs_plan_s; /* correction values of resync planer (RCU, connection->conn_update) */
973 int rs_in_flight; /* resync sectors in flight (to proxy, in proxy and from proxy) */
974 atomic_t ap_in_flight; /* App sectors in flight (waiting for ack) */
975 unsigned int peer_max_bio_size;
976 unsigned int local_max_bio_size;
977
978 /* any requests that would block in drbd_make_request()
979 * are deferred to this single-threaded work queue */
980 struct submit_worker submit;
981};
982
983struct drbd_bm_aio_ctx {
984 struct drbd_device *device;
985 struct list_head list; /* on device->pending_bitmap_io */;
986 unsigned long start_jif;
987 atomic_t in_flight;
988 unsigned int done;
989 unsigned flags;
990#define BM_AIO_COPY_PAGES 1
991#define BM_AIO_WRITE_HINTED 2
992#define BM_AIO_WRITE_ALL_PAGES 4
993#define BM_AIO_READ 8
994 int error;
995 struct kref kref;
996};
997
998struct drbd_config_context {
999 /* assigned from drbd_genlmsghdr */
1000 unsigned int minor;
1001 /* assigned from request attributes, if present */
1002 unsigned int volume;
1003#define VOLUME_UNSPECIFIED (-1U)
1004 /* pointer into the request skb,
1005 * limited lifetime! */
1006 char *resource_name;
1007 struct nlattr *my_addr;
1008 struct nlattr *peer_addr;
1009
1010 /* reply buffer */
1011 struct sk_buff *reply_skb;
1012 /* pointer into reply buffer */
1013 struct drbd_genlmsghdr *reply_dh;
1014 /* resolved from attributes, if possible */
1015 struct drbd_device *device;
1016 struct drbd_resource *resource;
1017 struct drbd_connection *connection;
1018};
1019
1020static inline struct drbd_device *minor_to_device(unsigned int minor)
1021{
1022 return (struct drbd_device *)idr_find(&drbd_devices, minor);
1023}
1024
1025static inline struct drbd_peer_device *first_peer_device(struct drbd_device *device)
1026{
1027 return list_first_entry_or_null(&device->peer_devices, struct drbd_peer_device, peer_devices);
1028}
1029
1030static inline struct drbd_peer_device *
1031conn_peer_device(struct drbd_connection *connection, int volume_number)
1032{
1033 return idr_find(&connection->peer_devices, volume_number);
1034}
1035
1036#define for_each_resource(resource, _resources) \
1037 list_for_each_entry(resource, _resources, resources)
1038
1039#define for_each_resource_rcu(resource, _resources) \
1040 list_for_each_entry_rcu(resource, _resources, resources)
1041
1042#define for_each_resource_safe(resource, tmp, _resources) \
1043 list_for_each_entry_safe(resource, tmp, _resources, resources)
1044
1045#define for_each_connection(connection, resource) \
1046 list_for_each_entry(connection, &resource->connections, connections)
1047
1048#define for_each_connection_rcu(connection, resource) \
1049 list_for_each_entry_rcu(connection, &resource->connections, connections)
1050
1051#define for_each_connection_safe(connection, tmp, resource) \
1052 list_for_each_entry_safe(connection, tmp, &resource->connections, connections)
1053
1054#define for_each_peer_device(peer_device, device) \
1055 list_for_each_entry(peer_device, &device->peer_devices, peer_devices)
1056
1057#define for_each_peer_device_rcu(peer_device, device) \
1058 list_for_each_entry_rcu(peer_device, &device->peer_devices, peer_devices)
1059
1060#define for_each_peer_device_safe(peer_device, tmp, device) \
1061 list_for_each_entry_safe(peer_device, tmp, &device->peer_devices, peer_devices)
1062
1063static inline unsigned int device_to_minor(struct drbd_device *device)
1064{
1065 return device->minor;
1066}
1067
1068/*
1069 * function declarations
1070 *************************/
1071
1072/* drbd_main.c */
1073
1074enum dds_flags {
1075 DDSF_FORCED = 1,
1076 DDSF_NO_RESYNC = 2, /* Do not run a resync for the new space */
1077};
1078
1079extern void drbd_init_set_defaults(struct drbd_device *device);
1080extern int drbd_thread_start(struct drbd_thread *thi);
1081extern void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait);
1082#ifdef CONFIG_SMP
1083extern void drbd_thread_current_set_cpu(struct drbd_thread *thi);
1084#else
1085#define drbd_thread_current_set_cpu(A) ({})
1086#endif
1087extern void tl_release(struct drbd_connection *, unsigned int barrier_nr,
1088 unsigned int set_size);
1089extern void tl_clear(struct drbd_connection *);
1090extern void drbd_free_sock(struct drbd_connection *connection);
1091extern int drbd_send(struct drbd_connection *connection, struct socket *sock,
1092 void *buf, size_t size, unsigned msg_flags);
1093extern int drbd_send_all(struct drbd_connection *, struct socket *, void *, size_t,
1094 unsigned);
1095
1096extern int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd);
1097extern int drbd_send_protocol(struct drbd_connection *connection);
1098extern int drbd_send_uuids(struct drbd_peer_device *);
1099extern int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *);
1100extern void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *);
1101extern int drbd_send_sizes(struct drbd_peer_device *, int trigger_reply, enum dds_flags flags);
1102extern int drbd_send_state(struct drbd_peer_device *, union drbd_state s);
1103extern int drbd_send_current_state(struct drbd_peer_device *);
1104extern int drbd_send_sync_param(struct drbd_peer_device *);
1105extern void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr,
1106 u32 set_size);
1107extern int drbd_send_ack(struct drbd_peer_device *, enum drbd_packet,
1108 struct drbd_peer_request *);
1109extern void drbd_send_ack_rp(struct drbd_peer_device *, enum drbd_packet,
1110 struct p_block_req *rp);
1111extern void drbd_send_ack_dp(struct drbd_peer_device *, enum drbd_packet,
1112 struct p_data *dp, int data_size);
1113extern int drbd_send_ack_ex(struct drbd_peer_device *, enum drbd_packet,
1114 sector_t sector, int blksize, u64 block_id);
1115extern int drbd_send_out_of_sync(struct drbd_peer_device *, struct drbd_request *);
1116extern int drbd_send_block(struct drbd_peer_device *, enum drbd_packet,
1117 struct drbd_peer_request *);
1118extern int drbd_send_dblock(struct drbd_peer_device *, struct drbd_request *req);
1119extern int drbd_send_drequest(struct drbd_peer_device *, int cmd,
1120 sector_t sector, int size, u64 block_id);
1121extern int drbd_send_drequest_csum(struct drbd_peer_device *, sector_t sector,
1122 int size, void *digest, int digest_size,
1123 enum drbd_packet cmd);
1124extern int drbd_send_ov_request(struct drbd_peer_device *, sector_t sector, int size);
1125
1126extern int drbd_send_bitmap(struct drbd_device *device);
1127extern void drbd_send_sr_reply(struct drbd_peer_device *, enum drbd_state_rv retcode);
1128extern void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode);
1129extern int drbd_send_rs_deallocated(struct drbd_peer_device *, struct drbd_peer_request *);
1130extern void drbd_backing_dev_free(struct drbd_device *device, struct drbd_backing_dev *ldev);
1131extern void drbd_device_cleanup(struct drbd_device *device);
1132extern void drbd_print_uuids(struct drbd_device *device, const char *text);
1133extern void drbd_queue_unplug(struct drbd_device *device);
1134
1135extern void conn_md_sync(struct drbd_connection *connection);
1136extern void drbd_md_write(struct drbd_device *device, void *buffer);
1137extern void drbd_md_sync(struct drbd_device *device);
1138extern int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev);
1139extern void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local);
1140extern void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local);
1141extern void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local);
1142extern void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local);
1143extern void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local);
1144extern void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local);
1145extern void drbd_md_set_flag(struct drbd_device *device, int flags) __must_hold(local);
1146extern void drbd_md_clear_flag(struct drbd_device *device, int flags)__must_hold(local);
1147extern int drbd_md_test_flag(struct drbd_backing_dev *, int);
1148extern void drbd_md_mark_dirty(struct drbd_device *device);
1149extern void drbd_queue_bitmap_io(struct drbd_device *device,
1150 int (*io_fn)(struct drbd_device *),
1151 void (*done)(struct drbd_device *, int),
1152 char *why, enum bm_flag flags);
1153extern int drbd_bitmap_io(struct drbd_device *device,
1154 int (*io_fn)(struct drbd_device *),
1155 char *why, enum bm_flag flags);
1156extern int drbd_bitmap_io_from_worker(struct drbd_device *device,
1157 int (*io_fn)(struct drbd_device *),
1158 char *why, enum bm_flag flags);
1159extern int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local);
1160extern int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local);
1161
1162/* Meta data layout
1163 *
1164 * We currently have two possible layouts.
1165 * Offsets in (512 byte) sectors.
1166 * external:
1167 * |----------- md_size_sect ------------------|
1168 * [ 4k superblock ][ activity log ][ Bitmap ]
1169 * | al_offset == 8 |
1170 * | bm_offset = al_offset + X |
1171 * ==> bitmap sectors = md_size_sect - bm_offset
1172 *
1173 * Variants:
1174 * old, indexed fixed size meta data:
1175 *
1176 * internal:
1177 * |----------- md_size_sect ------------------|
1178 * [data.....][ Bitmap ][ activity log ][ 4k superblock ][padding*]
1179 * | al_offset < 0 |
1180 * | bm_offset = al_offset - Y |
1181 * ==> bitmap sectors = Y = al_offset - bm_offset
1182 *
1183 * [padding*] are zero or up to 7 unused 512 Byte sectors to the
1184 * end of the device, so that the [4k superblock] will be 4k aligned.
1185 *
1186 * The activity log consists of 4k transaction blocks,
1187 * which are written in a ring-buffer, or striped ring-buffer like fashion,
1188 * which are writtensize used to be fixed 32kB,
1189 * but is about to become configurable.
1190 */
1191
1192/* Our old fixed size meta data layout
1193 * allows up to about 3.8TB, so if you want more,
1194 * you need to use the "flexible" meta data format. */
1195#define MD_128MB_SECT (128LLU << 11) /* 128 MB, unit sectors */
1196#define MD_4kB_SECT 8
1197#define MD_32kB_SECT 64
1198
1199/* One activity log extent represents 4M of storage */
1200#define AL_EXTENT_SHIFT 22
1201#define AL_EXTENT_SIZE (1<<AL_EXTENT_SHIFT)
1202
1203/* We could make these currently hardcoded constants configurable
1204 * variables at create-md time (or even re-configurable at runtime?).
1205 * Which will require some more changes to the DRBD "super block"
1206 * and attach code.
1207 *
1208 * updates per transaction:
1209 * This many changes to the active set can be logged with one transaction.
1210 * This number is arbitrary.
1211 * context per transaction:
1212 * This many context extent numbers are logged with each transaction.
1213 * This number is resulting from the transaction block size (4k), the layout
1214 * of the transaction header, and the number of updates per transaction.
1215 * See drbd_actlog.c:struct al_transaction_on_disk
1216 * */
1217#define AL_UPDATES_PER_TRANSACTION 64 // arbitrary
1218#define AL_CONTEXT_PER_TRANSACTION 919 // (4096 - 36 - 6*64)/4
1219
1220#if BITS_PER_LONG == 32
1221#define LN2_BPL 5
1222#define cpu_to_lel(A) cpu_to_le32(A)
1223#define lel_to_cpu(A) le32_to_cpu(A)
1224#elif BITS_PER_LONG == 64
1225#define LN2_BPL 6
1226#define cpu_to_lel(A) cpu_to_le64(A)
1227#define lel_to_cpu(A) le64_to_cpu(A)
1228#else
1229#error "LN2 of BITS_PER_LONG unknown!"
1230#endif
1231
1232/* resync bitmap */
1233/* 16MB sized 'bitmap extent' to track syncer usage */
1234struct bm_extent {
1235 int rs_left; /* number of bits set (out of sync) in this extent. */
1236 int rs_failed; /* number of failed resync requests in this extent. */
1237 unsigned long flags;
1238 struct lc_element lce;
1239};
1240
1241#define BME_NO_WRITES 0 /* bm_extent.flags: no more requests on this one! */
1242#define BME_LOCKED 1 /* bm_extent.flags: syncer active on this one. */
1243#define BME_PRIORITY 2 /* finish resync IO on this extent ASAP! App IO waiting! */
1244
1245/* drbd_bitmap.c */
1246/*
1247 * We need to store one bit for a block.
1248 * Example: 1GB disk @ 4096 byte blocks ==> we need 32 KB bitmap.
1249 * Bit 0 ==> local node thinks this block is binary identical on both nodes
1250 * Bit 1 ==> local node thinks this block needs to be synced.
1251 */
1252
1253#define SLEEP_TIME (HZ/10)
1254
1255/* We do bitmap IO in units of 4k blocks.
1256 * We also still have a hardcoded 4k per bit relation. */
1257#define BM_BLOCK_SHIFT 12 /* 4k per bit */
1258#define BM_BLOCK_SIZE (1<<BM_BLOCK_SHIFT)
1259/* mostly arbitrarily set the represented size of one bitmap extent,
1260 * aka resync extent, to 16 MiB (which is also 512 Byte worth of bitmap
1261 * at 4k per bit resolution) */
1262#define BM_EXT_SHIFT 24 /* 16 MiB per resync extent */
1263#define BM_EXT_SIZE (1<<BM_EXT_SHIFT)
1264
1265#if (BM_EXT_SHIFT != 24) || (BM_BLOCK_SHIFT != 12)
1266#error "HAVE YOU FIXED drbdmeta AS WELL??"
1267#endif
1268
1269/* thus many _storage_ sectors are described by one bit */
1270#define BM_SECT_TO_BIT(x) ((x)>>(BM_BLOCK_SHIFT-9))
1271#define BM_BIT_TO_SECT(x) ((sector_t)(x)<<(BM_BLOCK_SHIFT-9))
1272#define BM_SECT_PER_BIT BM_BIT_TO_SECT(1)
1273
1274/* bit to represented kilo byte conversion */
1275#define Bit2KB(bits) ((bits)<<(BM_BLOCK_SHIFT-10))
1276
1277/* in which _bitmap_ extent (resp. sector) the bit for a certain
1278 * _storage_ sector is located in */
1279#define BM_SECT_TO_EXT(x) ((x)>>(BM_EXT_SHIFT-9))
1280#define BM_BIT_TO_EXT(x) ((x) >> (BM_EXT_SHIFT - BM_BLOCK_SHIFT))
1281
1282/* first storage sector a bitmap extent corresponds to */
1283#define BM_EXT_TO_SECT(x) ((sector_t)(x) << (BM_EXT_SHIFT-9))
1284/* how much _storage_ sectors we have per bitmap extent */
1285#define BM_SECT_PER_EXT BM_EXT_TO_SECT(1)
1286/* how many bits are covered by one bitmap extent (resync extent) */
1287#define BM_BITS_PER_EXT (1UL << (BM_EXT_SHIFT - BM_BLOCK_SHIFT))
1288
1289#define BM_BLOCKS_PER_BM_EXT_MASK (BM_BITS_PER_EXT - 1)
1290
1291
1292/* in one sector of the bitmap, we have this many activity_log extents. */
1293#define AL_EXT_PER_BM_SECT (1 << (BM_EXT_SHIFT - AL_EXTENT_SHIFT))
1294
1295/* the extent in "PER_EXTENT" below is an activity log extent
1296 * we need that many (long words/bytes) to store the bitmap
1297 * of one AL_EXTENT_SIZE chunk of storage.
1298 * we can store the bitmap for that many AL_EXTENTS within
1299 * one sector of the _on_disk_ bitmap:
1300 * bit 0 bit 37 bit 38 bit (512*8)-1
1301 * ...|........|........|.. // ..|........|
1302 * sect. 0 `296 `304 ^(512*8*8)-1
1303 *
1304#define BM_WORDS_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / BITS_PER_LONG )
1305#define BM_BYTES_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / 8 ) // 128
1306#define BM_EXT_PER_SECT ( 512 / BM_BYTES_PER_EXTENT ) // 4
1307 */
1308
1309#define DRBD_MAX_SECTORS_32 (0xffffffffLU)
1310/* we have a certain meta data variant that has a fixed on-disk size of 128
1311 * MiB, of which 4k are our "superblock", and 32k are the fixed size activity
1312 * log, leaving this many sectors for the bitmap.
1313 */
1314
1315#define DRBD_MAX_SECTORS_FIXED_BM \
1316 ((MD_128MB_SECT - MD_32kB_SECT - MD_4kB_SECT) * (1LL<<(BM_EXT_SHIFT-9)))
1317#if !defined(CONFIG_LBDAF) && BITS_PER_LONG == 32
1318#define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_32
1319#define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_32
1320#else
1321#define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_FIXED_BM
1322/* 16 TB in units of sectors */
1323#if BITS_PER_LONG == 32
1324/* adjust by one page worth of bitmap,
1325 * so we won't wrap around in drbd_bm_find_next_bit.
1326 * you should use 64bit OS for that much storage, anyways. */
1327#define DRBD_MAX_SECTORS_FLEX BM_BIT_TO_SECT(0xffff7fff)
1328#else
1329/* we allow up to 1 PiB now on 64bit architecture with "flexible" meta data */
1330#define DRBD_MAX_SECTORS_FLEX (1UL << 51)
1331/* corresponds to (1UL << 38) bits right now. */
1332#endif
1333#endif
1334
1335/* Estimate max bio size as 256 * PAGE_SIZE,
1336 * so for typical PAGE_SIZE of 4k, that is (1<<20) Byte.
1337 * Since we may live in a mixed-platform cluster,
1338 * we limit us to a platform agnostic constant here for now.
1339 * A followup commit may allow even bigger BIO sizes,
1340 * once we thought that through. */
1341#define DRBD_MAX_BIO_SIZE (1U << 20)
1342#if DRBD_MAX_BIO_SIZE > (BIO_MAX_PAGES << PAGE_SHIFT)
1343#error Architecture not supported: DRBD_MAX_BIO_SIZE > BIO_MAX_SIZE
1344#endif
1345#define DRBD_MAX_BIO_SIZE_SAFE (1U << 12) /* Works always = 4k */
1346
1347#define DRBD_MAX_SIZE_H80_PACKET (1U << 15) /* Header 80 only allows packets up to 32KiB data */
1348#define DRBD_MAX_BIO_SIZE_P95 (1U << 17) /* Protocol 95 to 99 allows bios up to 128KiB */
1349
1350/* For now, don't allow more than half of what we can "activate" in one
1351 * activity log transaction to be discarded in one go. We may need to rework
1352 * drbd_al_begin_io() to allow for even larger discard ranges */
1353#define DRBD_MAX_BATCH_BIO_SIZE (AL_UPDATES_PER_TRANSACTION/2*AL_EXTENT_SIZE)
1354#define DRBD_MAX_BBIO_SECTORS (DRBD_MAX_BATCH_BIO_SIZE >> 9)
1355
1356extern int drbd_bm_init(struct drbd_device *device);
1357extern int drbd_bm_resize(struct drbd_device *device, sector_t sectors, int set_new_bits);
1358extern void drbd_bm_cleanup(struct drbd_device *device);
1359extern void drbd_bm_set_all(struct drbd_device *device);
1360extern void drbd_bm_clear_all(struct drbd_device *device);
1361/* set/clear/test only a few bits at a time */
1362extern int drbd_bm_set_bits(
1363 struct drbd_device *device, unsigned long s, unsigned long e);
1364extern int drbd_bm_clear_bits(
1365 struct drbd_device *device, unsigned long s, unsigned long e);
1366extern int drbd_bm_count_bits(
1367 struct drbd_device *device, const unsigned long s, const unsigned long e);
1368/* bm_set_bits variant for use while holding drbd_bm_lock,
1369 * may process the whole bitmap in one go */
1370extern void _drbd_bm_set_bits(struct drbd_device *device,
1371 const unsigned long s, const unsigned long e);
1372extern int drbd_bm_test_bit(struct drbd_device *device, unsigned long bitnr);
1373extern int drbd_bm_e_weight(struct drbd_device *device, unsigned long enr);
1374extern int drbd_bm_read(struct drbd_device *device) __must_hold(local);
1375extern void drbd_bm_mark_for_writeout(struct drbd_device *device, int page_nr);
1376extern int drbd_bm_write(struct drbd_device *device) __must_hold(local);
1377extern void drbd_bm_reset_al_hints(struct drbd_device *device) __must_hold(local);
1378extern int drbd_bm_write_hinted(struct drbd_device *device) __must_hold(local);
1379extern int drbd_bm_write_lazy(struct drbd_device *device, unsigned upper_idx) __must_hold(local);
1380extern int drbd_bm_write_all(struct drbd_device *device) __must_hold(local);
1381extern int drbd_bm_write_copy_pages(struct drbd_device *device) __must_hold(local);
1382extern size_t drbd_bm_words(struct drbd_device *device);
1383extern unsigned long drbd_bm_bits(struct drbd_device *device);
1384extern sector_t drbd_bm_capacity(struct drbd_device *device);
1385
1386#define DRBD_END_OF_BITMAP (~(unsigned long)0)
1387extern unsigned long drbd_bm_find_next(struct drbd_device *device, unsigned long bm_fo);
1388/* bm_find_next variants for use while you hold drbd_bm_lock() */
1389extern unsigned long _drbd_bm_find_next(struct drbd_device *device, unsigned long bm_fo);
1390extern unsigned long _drbd_bm_find_next_zero(struct drbd_device *device, unsigned long bm_fo);
1391extern unsigned long _drbd_bm_total_weight(struct drbd_device *device);
1392extern unsigned long drbd_bm_total_weight(struct drbd_device *device);
1393/* for receive_bitmap */
1394extern void drbd_bm_merge_lel(struct drbd_device *device, size_t offset,
1395 size_t number, unsigned long *buffer);
1396/* for _drbd_send_bitmap */
1397extern void drbd_bm_get_lel(struct drbd_device *device, size_t offset,
1398 size_t number, unsigned long *buffer);
1399
1400extern void drbd_bm_lock(struct drbd_device *device, char *why, enum bm_flag flags);
1401extern void drbd_bm_unlock(struct drbd_device *device);
1402/* drbd_main.c */
1403
1404extern struct kmem_cache *drbd_request_cache;
1405extern struct kmem_cache *drbd_ee_cache; /* peer requests */
1406extern struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
1407extern struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
1408extern mempool_t *drbd_request_mempool;
1409extern mempool_t *drbd_ee_mempool;
1410
1411/* drbd's page pool, used to buffer data received from the peer,
1412 * or data requested by the peer.
1413 *
1414 * This does not have an emergency reserve.
1415 *
1416 * When allocating from this pool, it first takes pages from the pool.
1417 * Only if the pool is depleted will try to allocate from the system.
1418 *
1419 * The assumption is that pages taken from this pool will be processed,
1420 * and given back, "quickly", and then can be recycled, so we can avoid
1421 * frequent calls to alloc_page(), and still will be able to make progress even
1422 * under memory pressure.
1423 */
1424extern struct page *drbd_pp_pool;
1425extern spinlock_t drbd_pp_lock;
1426extern int drbd_pp_vacant;
1427extern wait_queue_head_t drbd_pp_wait;
1428
1429/* We also need a standard (emergency-reserve backed) page pool
1430 * for meta data IO (activity log, bitmap).
1431 * We can keep it global, as long as it is used as "N pages at a time".
1432 * 128 should be plenty, currently we probably can get away with as few as 1.
1433 */
1434#define DRBD_MIN_POOL_PAGES 128
1435extern mempool_t *drbd_md_io_page_pool;
1436
1437/* We also need to make sure we get a bio
1438 * when we need it for housekeeping purposes */
1439extern struct bio_set *drbd_md_io_bio_set;
1440/* to allocate from that set */
1441extern struct bio *bio_alloc_drbd(gfp_t gfp_mask);
1442
1443/* And a bio_set for cloning */
1444extern struct bio_set *drbd_io_bio_set;
1445
1446extern struct mutex resources_mutex;
1447
1448extern int conn_lowest_minor(struct drbd_connection *connection);
1449extern enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor);
1450extern void drbd_destroy_device(struct kref *kref);
1451extern void drbd_delete_device(struct drbd_device *device);
1452
1453extern struct drbd_resource *drbd_create_resource(const char *name);
1454extern void drbd_free_resource(struct drbd_resource *resource);
1455
1456extern int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts);
1457extern struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts);
1458extern void drbd_destroy_connection(struct kref *kref);
1459extern struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
1460 void *peer_addr, int peer_addr_len);
1461extern struct drbd_resource *drbd_find_resource(const char *name);
1462extern void drbd_destroy_resource(struct kref *kref);
1463extern void conn_free_crypto(struct drbd_connection *connection);
1464
1465/* drbd_req */
1466extern void do_submit(struct work_struct *ws);
1467extern void __drbd_make_request(struct drbd_device *, struct bio *, unsigned long);
1468extern blk_qc_t drbd_make_request(struct request_queue *q, struct bio *bio);
1469extern int drbd_read_remote(struct drbd_device *device, struct drbd_request *req);
1470extern int is_valid_ar_handle(struct drbd_request *, sector_t);
1471
1472
1473/* drbd_nl.c */
1474
1475extern struct mutex notification_mutex;
1476
1477extern void drbd_suspend_io(struct drbd_device *device);
1478extern void drbd_resume_io(struct drbd_device *device);
1479extern char *ppsize(char *buf, unsigned long long size);
1480extern sector_t drbd_new_dev_size(struct drbd_device *, struct drbd_backing_dev *, sector_t, int);
1481enum determine_dev_size {
1482 DS_ERROR_SHRINK = -3,
1483 DS_ERROR_SPACE_MD = -2,
1484 DS_ERROR = -1,
1485 DS_UNCHANGED = 0,
1486 DS_SHRUNK = 1,
1487 DS_GREW = 2,
1488 DS_GREW_FROM_ZERO = 3,
1489};
1490extern enum determine_dev_size
1491drbd_determine_dev_size(struct drbd_device *, enum dds_flags, struct resize_parms *) __must_hold(local);
1492extern void resync_after_online_grow(struct drbd_device *);
1493extern void drbd_reconsider_queue_parameters(struct drbd_device *device,
1494 struct drbd_backing_dev *bdev, struct o_qlim *o);
1495extern enum drbd_state_rv drbd_set_role(struct drbd_device *device,
1496 enum drbd_role new_role,
1497 int force);
1498extern bool conn_try_outdate_peer(struct drbd_connection *connection);
1499extern void conn_try_outdate_peer_async(struct drbd_connection *connection);
1500extern enum drbd_peer_state conn_khelper(struct drbd_connection *connection, char *cmd);
1501extern int drbd_khelper(struct drbd_device *device, char *cmd);
1502
1503/* drbd_worker.c */
1504/* bi_end_io handlers */
1505extern void drbd_md_endio(struct bio *bio);
1506extern void drbd_peer_request_endio(struct bio *bio);
1507extern void drbd_request_endio(struct bio *bio);
1508extern int drbd_worker(struct drbd_thread *thi);
1509enum drbd_ret_code drbd_resync_after_valid(struct drbd_device *device, int o_minor);
1510void drbd_resync_after_changed(struct drbd_device *device);
1511extern void drbd_start_resync(struct drbd_device *device, enum drbd_conns side);
1512extern void resume_next_sg(struct drbd_device *device);
1513extern void suspend_other_sg(struct drbd_device *device);
1514extern int drbd_resync_finished(struct drbd_device *device);
1515/* maybe rather drbd_main.c ? */
1516extern void *drbd_md_get_buffer(struct drbd_device *device, const char *intent);
1517extern void drbd_md_put_buffer(struct drbd_device *device);
1518extern int drbd_md_sync_page_io(struct drbd_device *device,
1519 struct drbd_backing_dev *bdev, sector_t sector, int op);
1520extern void drbd_ov_out_of_sync_found(struct drbd_device *, sector_t, int);
1521extern void wait_until_done_or_force_detached(struct drbd_device *device,
1522 struct drbd_backing_dev *bdev, unsigned int *done);
1523extern void drbd_rs_controller_reset(struct drbd_device *device);
1524
1525static inline void ov_out_of_sync_print(struct drbd_device *device)
1526{
1527 if (device->ov_last_oos_size) {
1528 drbd_err(device, "Out of sync: start=%llu, size=%lu (sectors)\n",
1529 (unsigned long long)device->ov_last_oos_start,
1530 (unsigned long)device->ov_last_oos_size);
1531 }
1532 device->ov_last_oos_size = 0;
1533}
1534
1535
1536extern void drbd_csum_bio(struct crypto_ahash *, struct bio *, void *);
1537extern void drbd_csum_ee(struct crypto_ahash *, struct drbd_peer_request *, void *);
1538/* worker callbacks */
1539extern int w_e_end_data_req(struct drbd_work *, int);
1540extern int w_e_end_rsdata_req(struct drbd_work *, int);
1541extern int w_e_end_csum_rs_req(struct drbd_work *, int);
1542extern int w_e_end_ov_reply(struct drbd_work *, int);
1543extern int w_e_end_ov_req(struct drbd_work *, int);
1544extern int w_ov_finished(struct drbd_work *, int);
1545extern int w_resync_timer(struct drbd_work *, int);
1546extern int w_send_write_hint(struct drbd_work *, int);
1547extern int w_send_dblock(struct drbd_work *, int);
1548extern int w_send_read_req(struct drbd_work *, int);
1549extern int w_e_reissue(struct drbd_work *, int);
1550extern int w_restart_disk_io(struct drbd_work *, int);
1551extern int w_send_out_of_sync(struct drbd_work *, int);
1552extern int w_start_resync(struct drbd_work *, int);
1553
1554extern void resync_timer_fn(struct timer_list *t);
1555extern void start_resync_timer_fn(struct timer_list *t);
1556
1557extern void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req);
1558
1559/* drbd_receiver.c */
1560extern int drbd_receiver(struct drbd_thread *thi);
1561extern int drbd_ack_receiver(struct drbd_thread *thi);
1562extern void drbd_send_ping_wf(struct work_struct *ws);
1563extern void drbd_send_acks_wf(struct work_struct *ws);
1564extern bool drbd_rs_c_min_rate_throttle(struct drbd_device *device);
1565extern bool drbd_rs_should_slow_down(struct drbd_device *device, sector_t sector,
1566 bool throttle_if_app_is_waiting);
1567extern int drbd_submit_peer_request(struct drbd_device *,
1568 struct drbd_peer_request *, const unsigned,
1569 const unsigned, const int);
1570extern int drbd_free_peer_reqs(struct drbd_device *, struct list_head *);
1571extern struct drbd_peer_request *drbd_alloc_peer_req(struct drbd_peer_device *, u64,
1572 sector_t, unsigned int,
1573 unsigned int,
1574 gfp_t) __must_hold(local);
1575extern void __drbd_free_peer_req(struct drbd_device *, struct drbd_peer_request *,
1576 int);
1577#define drbd_free_peer_req(m,e) __drbd_free_peer_req(m, e, 0)
1578#define drbd_free_net_peer_req(m,e) __drbd_free_peer_req(m, e, 1)
1579extern struct page *drbd_alloc_pages(struct drbd_peer_device *, unsigned int, bool);
1580extern void drbd_set_recv_tcq(struct drbd_device *device, int tcq_enabled);
1581extern void _drbd_clear_done_ee(struct drbd_device *device, struct list_head *to_be_freed);
1582extern int drbd_connected(struct drbd_peer_device *);
1583
1584static inline void drbd_tcp_cork(struct socket *sock)
1585{
1586 int val = 1;
1587 (void) kernel_setsockopt(sock, SOL_TCP, TCP_CORK,
1588 (char*)&val, sizeof(val));
1589}
1590
1591static inline void drbd_tcp_uncork(struct socket *sock)
1592{
1593 int val = 0;
1594 (void) kernel_setsockopt(sock, SOL_TCP, TCP_CORK,
1595 (char*)&val, sizeof(val));
1596}
1597
1598static inline void drbd_tcp_nodelay(struct socket *sock)
1599{
1600 int val = 1;
1601 (void) kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
1602 (char*)&val, sizeof(val));
1603}
1604
1605static inline void drbd_tcp_quickack(struct socket *sock)
1606{
1607 int val = 2;
1608 (void) kernel_setsockopt(sock, SOL_TCP, TCP_QUICKACK,
1609 (char*)&val, sizeof(val));
1610}
1611
1612/* sets the number of 512 byte sectors of our virtual device */
1613static inline void drbd_set_my_capacity(struct drbd_device *device,
1614 sector_t size)
1615{
1616 /* set_capacity(device->this_bdev->bd_disk, size); */
1617 set_capacity(device->vdisk, size);
1618 device->this_bdev->bd_inode->i_size = (loff_t)size << 9;
1619}
1620
1621/*
1622 * used to submit our private bio
1623 */
1624static inline void drbd_generic_make_request(struct drbd_device *device,
1625 int fault_type, struct bio *bio)
1626{
1627 __release(local);
1628 if (!bio->bi_disk) {
1629 drbd_err(device, "drbd_generic_make_request: bio->bi_disk == NULL\n");
1630 bio->bi_status = BLK_STS_IOERR;
1631 bio_endio(bio);
1632 return;
1633 }
1634
1635 if (drbd_insert_fault(device, fault_type))
1636 bio_io_error(bio);
1637 else
1638 generic_make_request(bio);
1639}
1640
1641void drbd_bump_write_ordering(struct drbd_resource *resource, struct drbd_backing_dev *bdev,
1642 enum write_ordering_e wo);
1643
1644/* drbd_proc.c */
1645extern struct proc_dir_entry *drbd_proc;
1646extern const struct file_operations drbd_proc_fops;
1647
1648/* drbd_actlog.c */
1649extern bool drbd_al_begin_io_prepare(struct drbd_device *device, struct drbd_interval *i);
1650extern int drbd_al_begin_io_nonblock(struct drbd_device *device, struct drbd_interval *i);
1651extern void drbd_al_begin_io_commit(struct drbd_device *device);
1652extern bool drbd_al_begin_io_fastpath(struct drbd_device *device, struct drbd_interval *i);
1653extern void drbd_al_begin_io(struct drbd_device *device, struct drbd_interval *i);
1654extern void drbd_al_complete_io(struct drbd_device *device, struct drbd_interval *i);
1655extern void drbd_rs_complete_io(struct drbd_device *device, sector_t sector);
1656extern int drbd_rs_begin_io(struct drbd_device *device, sector_t sector);
1657extern int drbd_try_rs_begin_io(struct drbd_device *device, sector_t sector);
1658extern void drbd_rs_cancel_all(struct drbd_device *device);
1659extern int drbd_rs_del_all(struct drbd_device *device);
1660extern void drbd_rs_failed_io(struct drbd_device *device,
1661 sector_t sector, int size);
1662extern void drbd_advance_rs_marks(struct drbd_device *device, unsigned long still_to_go);
1663
1664enum update_sync_bits_mode { RECORD_RS_FAILED, SET_OUT_OF_SYNC, SET_IN_SYNC };
1665extern int __drbd_change_sync(struct drbd_device *device, sector_t sector, int size,
1666 enum update_sync_bits_mode mode);
1667#define drbd_set_in_sync(device, sector, size) \
1668 __drbd_change_sync(device, sector, size, SET_IN_SYNC)
1669#define drbd_set_out_of_sync(device, sector, size) \
1670 __drbd_change_sync(device, sector, size, SET_OUT_OF_SYNC)
1671#define drbd_rs_failed_io(device, sector, size) \
1672 __drbd_change_sync(device, sector, size, RECORD_RS_FAILED)
1673extern void drbd_al_shrink(struct drbd_device *device);
1674extern int drbd_al_initialize(struct drbd_device *, void *);
1675
1676/* drbd_nl.c */
1677/* state info broadcast */
1678struct sib_info {
1679 enum drbd_state_info_bcast_reason sib_reason;
1680 union {
1681 struct {
1682 char *helper_name;
1683 unsigned helper_exit_code;
1684 };
1685 struct {
1686 union drbd_state os;
1687 union drbd_state ns;
1688 };
1689 };
1690};
1691void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib);
1692
1693extern void notify_resource_state(struct sk_buff *,
1694 unsigned int,
1695 struct drbd_resource *,
1696 struct resource_info *,
1697 enum drbd_notification_type);
1698extern void notify_device_state(struct sk_buff *,
1699 unsigned int,
1700 struct drbd_device *,
1701 struct device_info *,
1702 enum drbd_notification_type);
1703extern void notify_connection_state(struct sk_buff *,
1704 unsigned int,
1705 struct drbd_connection *,
1706 struct connection_info *,
1707 enum drbd_notification_type);
1708extern void notify_peer_device_state(struct sk_buff *,
1709 unsigned int,
1710 struct drbd_peer_device *,
1711 struct peer_device_info *,
1712 enum drbd_notification_type);
1713extern void notify_helper(enum drbd_notification_type, struct drbd_device *,
1714 struct drbd_connection *, const char *, int);
1715
1716/*
1717 * inline helper functions
1718 *************************/
1719
1720/* see also page_chain_add and friends in drbd_receiver.c */
1721static inline struct page *page_chain_next(struct page *page)
1722{
1723 return (struct page *)page_private(page);
1724}
1725#define page_chain_for_each(page) \
1726 for (; page && ({ prefetch(page_chain_next(page)); 1; }); \
1727 page = page_chain_next(page))
1728#define page_chain_for_each_safe(page, n) \
1729 for (; page && ({ n = page_chain_next(page); 1; }); page = n)
1730
1731
1732static inline int drbd_peer_req_has_active_page(struct drbd_peer_request *peer_req)
1733{
1734 struct page *page = peer_req->pages;
1735 page_chain_for_each(page) {
1736 if (page_count(page) > 1)
1737 return 1;
1738 }
1739 return 0;
1740}
1741
1742static inline union drbd_state drbd_read_state(struct drbd_device *device)
1743{
1744 struct drbd_resource *resource = device->resource;
1745 union drbd_state rv;
1746
1747 rv.i = device->state.i;
1748 rv.susp = resource->susp;
1749 rv.susp_nod = resource->susp_nod;
1750 rv.susp_fen = resource->susp_fen;
1751
1752 return rv;
1753}
1754
1755enum drbd_force_detach_flags {
1756 DRBD_READ_ERROR,
1757 DRBD_WRITE_ERROR,
1758 DRBD_META_IO_ERROR,
1759 DRBD_FORCE_DETACH,
1760};
1761
1762#define __drbd_chk_io_error(m,f) __drbd_chk_io_error_(m,f, __func__)
1763static inline void __drbd_chk_io_error_(struct drbd_device *device,
1764 enum drbd_force_detach_flags df,
1765 const char *where)
1766{
1767 enum drbd_io_error_p ep;
1768
1769 rcu_read_lock();
1770 ep = rcu_dereference(device->ldev->disk_conf)->on_io_error;
1771 rcu_read_unlock();
1772 switch (ep) {
1773 case EP_PASS_ON: /* FIXME would this be better named "Ignore"? */
1774 if (df == DRBD_READ_ERROR || df == DRBD_WRITE_ERROR) {
1775 if (__ratelimit(&drbd_ratelimit_state))
1776 drbd_err(device, "Local IO failed in %s.\n", where);
1777 if (device->state.disk > D_INCONSISTENT)
1778 _drbd_set_state(_NS(device, disk, D_INCONSISTENT), CS_HARD, NULL);
1779 break;
1780 }
1781 /* NOTE fall through for DRBD_META_IO_ERROR or DRBD_FORCE_DETACH */
1782 case EP_DETACH:
1783 case EP_CALL_HELPER:
1784 /* Remember whether we saw a READ or WRITE error.
1785 *
1786 * Recovery of the affected area for WRITE failure is covered
1787 * by the activity log.
1788 * READ errors may fall outside that area though. Certain READ
1789 * errors can be "healed" by writing good data to the affected
1790 * blocks, which triggers block re-allocation in lower layers.
1791 *
1792 * If we can not write the bitmap after a READ error,
1793 * we may need to trigger a full sync (see w_go_diskless()).
1794 *
1795 * Force-detach is not really an IO error, but rather a
1796 * desperate measure to try to deal with a completely
1797 * unresponsive lower level IO stack.
1798 * Still it should be treated as a WRITE error.
1799 *
1800 * Meta IO error is always WRITE error:
1801 * we read meta data only once during attach,
1802 * which will fail in case of errors.
1803 */
1804 set_bit(WAS_IO_ERROR, &device->flags);
1805 if (df == DRBD_READ_ERROR)
1806 set_bit(WAS_READ_ERROR, &device->flags);
1807 if (df == DRBD_FORCE_DETACH)
1808 set_bit(FORCE_DETACH, &device->flags);
1809 if (device->state.disk > D_FAILED) {
1810 _drbd_set_state(_NS(device, disk, D_FAILED), CS_HARD, NULL);
1811 drbd_err(device,
1812 "Local IO failed in %s. Detaching...\n", where);
1813 }
1814 break;
1815 }
1816}
1817
1818/**
1819 * drbd_chk_io_error: Handle the on_io_error setting, should be called from all io completion handlers
1820 * @device: DRBD device.
1821 * @error: Error code passed to the IO completion callback
1822 * @forcedetach: Force detach. I.e. the error happened while accessing the meta data
1823 *
1824 * See also drbd_main.c:after_state_ch() if (os.disk > D_FAILED && ns.disk == D_FAILED)
1825 */
1826#define drbd_chk_io_error(m,e,f) drbd_chk_io_error_(m,e,f, __func__)
1827static inline void drbd_chk_io_error_(struct drbd_device *device,
1828 int error, enum drbd_force_detach_flags forcedetach, const char *where)
1829{
1830 if (error) {
1831 unsigned long flags;
1832 spin_lock_irqsave(&device->resource->req_lock, flags);
1833 __drbd_chk_io_error_(device, forcedetach, where);
1834 spin_unlock_irqrestore(&device->resource->req_lock, flags);
1835 }
1836}
1837
1838
1839/**
1840 * drbd_md_first_sector() - Returns the first sector number of the meta data area
1841 * @bdev: Meta data block device.
1842 *
1843 * BTW, for internal meta data, this happens to be the maximum capacity
1844 * we could agree upon with our peer node.
1845 */
1846static inline sector_t drbd_md_first_sector(struct drbd_backing_dev *bdev)
1847{
1848 switch (bdev->md.meta_dev_idx) {
1849 case DRBD_MD_INDEX_INTERNAL:
1850 case DRBD_MD_INDEX_FLEX_INT:
1851 return bdev->md.md_offset + bdev->md.bm_offset;
1852 case DRBD_MD_INDEX_FLEX_EXT:
1853 default:
1854 return bdev->md.md_offset;
1855 }
1856}
1857
1858/**
1859 * drbd_md_last_sector() - Return the last sector number of the meta data area
1860 * @bdev: Meta data block device.
1861 */
1862static inline sector_t drbd_md_last_sector(struct drbd_backing_dev *bdev)
1863{
1864 switch (bdev->md.meta_dev_idx) {
1865 case DRBD_MD_INDEX_INTERNAL:
1866 case DRBD_MD_INDEX_FLEX_INT:
1867 return bdev->md.md_offset + MD_4kB_SECT -1;
1868 case DRBD_MD_INDEX_FLEX_EXT:
1869 default:
1870 return bdev->md.md_offset + bdev->md.md_size_sect -1;
1871 }
1872}
1873
1874/* Returns the number of 512 byte sectors of the device */
1875static inline sector_t drbd_get_capacity(struct block_device *bdev)
1876{
1877 /* return bdev ? get_capacity(bdev->bd_disk) : 0; */
1878 return bdev ? i_size_read(bdev->bd_inode) >> 9 : 0;
1879}
1880
1881/**
1882 * drbd_get_max_capacity() - Returns the capacity we announce to out peer
1883 * @bdev: Meta data block device.
1884 *
1885 * returns the capacity we announce to out peer. we clip ourselves at the
1886 * various MAX_SECTORS, because if we don't, current implementation will
1887 * oops sooner or later
1888 */
1889static inline sector_t drbd_get_max_capacity(struct drbd_backing_dev *bdev)
1890{
1891 sector_t s;
1892
1893 switch (bdev->md.meta_dev_idx) {
1894 case DRBD_MD_INDEX_INTERNAL:
1895 case DRBD_MD_INDEX_FLEX_INT:
1896 s = drbd_get_capacity(bdev->backing_bdev)
1897 ? min_t(sector_t, DRBD_MAX_SECTORS_FLEX,
1898 drbd_md_first_sector(bdev))
1899 : 0;
1900 break;
1901 case DRBD_MD_INDEX_FLEX_EXT:
1902 s = min_t(sector_t, DRBD_MAX_SECTORS_FLEX,
1903 drbd_get_capacity(bdev->backing_bdev));
1904 /* clip at maximum size the meta device can support */
1905 s = min_t(sector_t, s,
1906 BM_EXT_TO_SECT(bdev->md.md_size_sect
1907 - bdev->md.bm_offset));
1908 break;
1909 default:
1910 s = min_t(sector_t, DRBD_MAX_SECTORS,
1911 drbd_get_capacity(bdev->backing_bdev));
1912 }
1913 return s;
1914}
1915
1916/**
1917 * drbd_md_ss() - Return the sector number of our meta data super block
1918 * @bdev: Meta data block device.
1919 */
1920static inline sector_t drbd_md_ss(struct drbd_backing_dev *bdev)
1921{
1922 const int meta_dev_idx = bdev->md.meta_dev_idx;
1923
1924 if (meta_dev_idx == DRBD_MD_INDEX_FLEX_EXT)
1925 return 0;
1926
1927 /* Since drbd08, internal meta data is always "flexible".
1928 * position: last 4k aligned block of 4k size */
1929 if (meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1930 meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)
1931 return (drbd_get_capacity(bdev->backing_bdev) & ~7ULL) - 8;
1932
1933 /* external, some index; this is the old fixed size layout */
1934 return MD_128MB_SECT * bdev->md.meta_dev_idx;
1935}
1936
1937static inline void
1938drbd_queue_work(struct drbd_work_queue *q, struct drbd_work *w)
1939{
1940 unsigned long flags;
1941 spin_lock_irqsave(&q->q_lock, flags);
1942 list_add_tail(&w->list, &q->q);
1943 spin_unlock_irqrestore(&q->q_lock, flags);
1944 wake_up(&q->q_wait);
1945}
1946
1947static inline void
1948drbd_queue_work_if_unqueued(struct drbd_work_queue *q, struct drbd_work *w)
1949{
1950 unsigned long flags;
1951 spin_lock_irqsave(&q->q_lock, flags);
1952 if (list_empty_careful(&w->list))
1953 list_add_tail(&w->list, &q->q);
1954 spin_unlock_irqrestore(&q->q_lock, flags);
1955 wake_up(&q->q_wait);
1956}
1957
1958static inline void
1959drbd_device_post_work(struct drbd_device *device, int work_bit)
1960{
1961 if (!test_and_set_bit(work_bit, &device->flags)) {
1962 struct drbd_connection *connection =
1963 first_peer_device(device)->connection;
1964 struct drbd_work_queue *q = &connection->sender_work;
1965 if (!test_and_set_bit(DEVICE_WORK_PENDING, &connection->flags))
1966 wake_up(&q->q_wait);
1967 }
1968}
1969
1970extern void drbd_flush_workqueue(struct drbd_work_queue *work_queue);
1971
1972/* To get the ack_receiver out of the blocking network stack,
1973 * so it can change its sk_rcvtimeo from idle- to ping-timeout,
1974 * and send a ping, we need to send a signal.
1975 * Which signal we send is irrelevant. */
1976static inline void wake_ack_receiver(struct drbd_connection *connection)
1977{
1978 struct task_struct *task = connection->ack_receiver.task;
1979 if (task && get_t_state(&connection->ack_receiver) == RUNNING)
1980 force_sig(SIGXCPU, task);
1981}
1982
1983static inline void request_ping(struct drbd_connection *connection)
1984{
1985 set_bit(SEND_PING, &connection->flags);
1986 wake_ack_receiver(connection);
1987}
1988
1989extern void *conn_prepare_command(struct drbd_connection *, struct drbd_socket *);
1990extern void *drbd_prepare_command(struct drbd_peer_device *, struct drbd_socket *);
1991extern int conn_send_command(struct drbd_connection *, struct drbd_socket *,
1992 enum drbd_packet, unsigned int, void *,
1993 unsigned int);
1994extern int drbd_send_command(struct drbd_peer_device *, struct drbd_socket *,
1995 enum drbd_packet, unsigned int, void *,
1996 unsigned int);
1997
1998extern int drbd_send_ping(struct drbd_connection *connection);
1999extern int drbd_send_ping_ack(struct drbd_connection *connection);
2000extern int drbd_send_state_req(struct drbd_peer_device *, union drbd_state, union drbd_state);
2001extern int conn_send_state_req(struct drbd_connection *, union drbd_state, union drbd_state);
2002
2003static inline void drbd_thread_stop(struct drbd_thread *thi)
2004{
2005 _drbd_thread_stop(thi, false, true);
2006}
2007
2008static inline void drbd_thread_stop_nowait(struct drbd_thread *thi)
2009{
2010 _drbd_thread_stop(thi, false, false);
2011}
2012
2013static inline void drbd_thread_restart_nowait(struct drbd_thread *thi)
2014{
2015 _drbd_thread_stop(thi, true, false);
2016}
2017
2018/* counts how many answer packets packets we expect from our peer,
2019 * for either explicit application requests,
2020 * or implicit barrier packets as necessary.
2021 * increased:
2022 * w_send_barrier
2023 * _req_mod(req, QUEUE_FOR_NET_WRITE or QUEUE_FOR_NET_READ);
2024 * it is much easier and equally valid to count what we queue for the
2025 * worker, even before it actually was queued or send.
2026 * (drbd_make_request_common; recovery path on read io-error)
2027 * decreased:
2028 * got_BarrierAck (respective tl_clear, tl_clear_barrier)
2029 * _req_mod(req, DATA_RECEIVED)
2030 * [from receive_DataReply]
2031 * _req_mod(req, WRITE_ACKED_BY_PEER or RECV_ACKED_BY_PEER or NEG_ACKED)
2032 * [from got_BlockAck (P_WRITE_ACK, P_RECV_ACK)]
2033 * for some reason it is NOT decreased in got_NegAck,
2034 * but in the resulting cleanup code from report_params.
2035 * we should try to remember the reason for that...
2036 * _req_mod(req, SEND_FAILED or SEND_CANCELED)
2037 * _req_mod(req, CONNECTION_LOST_WHILE_PENDING)
2038 * [from tl_clear_barrier]
2039 */
2040static inline void inc_ap_pending(struct drbd_device *device)
2041{
2042 atomic_inc(&device->ap_pending_cnt);
2043}
2044
2045#define ERR_IF_CNT_IS_NEGATIVE(which, func, line) \
2046 if (atomic_read(&device->which) < 0) \
2047 drbd_err(device, "in %s:%d: " #which " = %d < 0 !\n", \
2048 func, line, \
2049 atomic_read(&device->which))
2050
2051#define dec_ap_pending(device) _dec_ap_pending(device, __func__, __LINE__)
2052static inline void _dec_ap_pending(struct drbd_device *device, const char *func, int line)
2053{
2054 if (atomic_dec_and_test(&device->ap_pending_cnt))
2055 wake_up(&device->misc_wait);
2056 ERR_IF_CNT_IS_NEGATIVE(ap_pending_cnt, func, line);
2057}
2058
2059/* counts how many resync-related answers we still expect from the peer
2060 * increase decrease
2061 * C_SYNC_TARGET sends P_RS_DATA_REQUEST (and expects P_RS_DATA_REPLY)
2062 * C_SYNC_SOURCE sends P_RS_DATA_REPLY (and expects P_WRITE_ACK with ID_SYNCER)
2063 * (or P_NEG_ACK with ID_SYNCER)
2064 */
2065static inline void inc_rs_pending(struct drbd_device *device)
2066{
2067 atomic_inc(&device->rs_pending_cnt);
2068}
2069
2070#define dec_rs_pending(device) _dec_rs_pending(device, __func__, __LINE__)
2071static inline void _dec_rs_pending(struct drbd_device *device, const char *func, int line)
2072{
2073 atomic_dec(&device->rs_pending_cnt);
2074 ERR_IF_CNT_IS_NEGATIVE(rs_pending_cnt, func, line);
2075}
2076
2077/* counts how many answers we still need to send to the peer.
2078 * increased on
2079 * receive_Data unless protocol A;
2080 * we need to send a P_RECV_ACK (proto B)
2081 * or P_WRITE_ACK (proto C)
2082 * receive_RSDataReply (recv_resync_read) we need to send a P_WRITE_ACK
2083 * receive_DataRequest (receive_RSDataRequest) we need to send back P_DATA
2084 * receive_Barrier_* we need to send a P_BARRIER_ACK
2085 */
2086static inline void inc_unacked(struct drbd_device *device)
2087{
2088 atomic_inc(&device->unacked_cnt);
2089}
2090
2091#define dec_unacked(device) _dec_unacked(device, __func__, __LINE__)
2092static inline void _dec_unacked(struct drbd_device *device, const char *func, int line)
2093{
2094 atomic_dec(&device->unacked_cnt);
2095 ERR_IF_CNT_IS_NEGATIVE(unacked_cnt, func, line);
2096}
2097
2098#define sub_unacked(device, n) _sub_unacked(device, n, __func__, __LINE__)
2099static inline void _sub_unacked(struct drbd_device *device, int n, const char *func, int line)
2100{
2101 atomic_sub(n, &device->unacked_cnt);
2102 ERR_IF_CNT_IS_NEGATIVE(unacked_cnt, func, line);
2103}
2104
2105static inline bool is_sync_target_state(enum drbd_conns connection_state)
2106{
2107 return connection_state == C_SYNC_TARGET ||
2108 connection_state == C_PAUSED_SYNC_T;
2109}
2110
2111static inline bool is_sync_source_state(enum drbd_conns connection_state)
2112{
2113 return connection_state == C_SYNC_SOURCE ||
2114 connection_state == C_PAUSED_SYNC_S;
2115}
2116
2117static inline bool is_sync_state(enum drbd_conns connection_state)
2118{
2119 return is_sync_source_state(connection_state) ||
2120 is_sync_target_state(connection_state);
2121}
2122
2123/**
2124 * get_ldev() - Increase the ref count on device->ldev. Returns 0 if there is no ldev
2125 * @_device: DRBD device.
2126 * @_min_state: Minimum device state required for success.
2127 *
2128 * You have to call put_ldev() when finished working with device->ldev.
2129 */
2130#define get_ldev_if_state(_device, _min_state) \
2131 (_get_ldev_if_state((_device), (_min_state)) ? \
2132 ({ __acquire(x); true; }) : false)
2133#define get_ldev(_device) get_ldev_if_state(_device, D_INCONSISTENT)
2134
2135static inline void put_ldev(struct drbd_device *device)
2136{
2137 enum drbd_disk_state disk_state = device->state.disk;
2138 /* We must check the state *before* the atomic_dec becomes visible,
2139 * or we have a theoretical race where someone hitting zero,
2140 * while state still D_FAILED, will then see D_DISKLESS in the
2141 * condition below and calling into destroy, where he must not, yet. */
2142 int i = atomic_dec_return(&device->local_cnt);
2143
2144 /* This may be called from some endio handler,
2145 * so we must not sleep here. */
2146
2147 __release(local);
2148 D_ASSERT(device, i >= 0);
2149 if (i == 0) {
2150 if (disk_state == D_DISKLESS)
2151 /* even internal references gone, safe to destroy */
2152 drbd_device_post_work(device, DESTROY_DISK);
2153 if (disk_state == D_FAILED)
2154 /* all application IO references gone. */
2155 if (!test_and_set_bit(GOING_DISKLESS, &device->flags))
2156 drbd_device_post_work(device, GO_DISKLESS);
2157 wake_up(&device->misc_wait);
2158 }
2159}
2160
2161#ifndef __CHECKER__
2162static inline int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
2163{
2164 int io_allowed;
2165
2166 /* never get a reference while D_DISKLESS */
2167 if (device->state.disk == D_DISKLESS)
2168 return 0;
2169
2170 atomic_inc(&device->local_cnt);
2171 io_allowed = (device->state.disk >= mins);
2172 if (!io_allowed)
2173 put_ldev(device);
2174 return io_allowed;
2175}
2176#else
2177extern int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins);
2178#endif
2179
2180/* this throttles on-the-fly application requests
2181 * according to max_buffers settings;
2182 * maybe re-implement using semaphores? */
2183static inline int drbd_get_max_buffers(struct drbd_device *device)
2184{
2185 struct net_conf *nc;
2186 int mxb;
2187
2188 rcu_read_lock();
2189 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
2190 mxb = nc ? nc->max_buffers : 1000000; /* arbitrary limit on open requests */
2191 rcu_read_unlock();
2192
2193 return mxb;
2194}
2195
2196static inline int drbd_state_is_stable(struct drbd_device *device)
2197{
2198 union drbd_dev_state s = device->state;
2199
2200 /* DO NOT add a default clause, we want the compiler to warn us
2201 * for any newly introduced state we may have forgotten to add here */
2202
2203 switch ((enum drbd_conns)s.conn) {
2204 /* new io only accepted when there is no connection, ... */
2205 case C_STANDALONE:
2206 case C_WF_CONNECTION:
2207 /* ... or there is a well established connection. */
2208 case C_CONNECTED:
2209 case C_SYNC_SOURCE:
2210 case C_SYNC_TARGET:
2211 case C_VERIFY_S:
2212 case C_VERIFY_T:
2213 case C_PAUSED_SYNC_S:
2214 case C_PAUSED_SYNC_T:
2215 case C_AHEAD:
2216 case C_BEHIND:
2217 /* transitional states, IO allowed */
2218 case C_DISCONNECTING:
2219 case C_UNCONNECTED:
2220 case C_TIMEOUT:
2221 case C_BROKEN_PIPE:
2222 case C_NETWORK_FAILURE:
2223 case C_PROTOCOL_ERROR:
2224 case C_TEAR_DOWN:
2225 case C_WF_REPORT_PARAMS:
2226 case C_STARTING_SYNC_S:
2227 case C_STARTING_SYNC_T:
2228 break;
2229
2230 /* Allow IO in BM exchange states with new protocols */
2231 case C_WF_BITMAP_S:
2232 if (first_peer_device(device)->connection->agreed_pro_version < 96)
2233 return 0;
2234 break;
2235
2236 /* no new io accepted in these states */
2237 case C_WF_BITMAP_T:
2238 case C_WF_SYNC_UUID:
2239 case C_MASK:
2240 /* not "stable" */
2241 return 0;
2242 }
2243
2244 switch ((enum drbd_disk_state)s.disk) {
2245 case D_DISKLESS:
2246 case D_INCONSISTENT:
2247 case D_OUTDATED:
2248 case D_CONSISTENT:
2249 case D_UP_TO_DATE:
2250 case D_FAILED:
2251 /* disk state is stable as well. */
2252 break;
2253
2254 /* no new io accepted during transitional states */
2255 case D_ATTACHING:
2256 case D_NEGOTIATING:
2257 case D_UNKNOWN:
2258 case D_MASK:
2259 /* not "stable" */
2260 return 0;
2261 }
2262
2263 return 1;
2264}
2265
2266static inline int drbd_suspended(struct drbd_device *device)
2267{
2268 struct drbd_resource *resource = device->resource;
2269
2270 return resource->susp || resource->susp_fen || resource->susp_nod;
2271}
2272
2273static inline bool may_inc_ap_bio(struct drbd_device *device)
2274{
2275 int mxb = drbd_get_max_buffers(device);
2276
2277 if (drbd_suspended(device))
2278 return false;
2279 if (atomic_read(&device->suspend_cnt))
2280 return false;
2281
2282 /* to avoid potential deadlock or bitmap corruption,
2283 * in various places, we only allow new application io
2284 * to start during "stable" states. */
2285
2286 /* no new io accepted when attaching or detaching the disk */
2287 if (!drbd_state_is_stable(device))
2288 return false;
2289
2290 /* since some older kernels don't have atomic_add_unless,
2291 * and we are within the spinlock anyways, we have this workaround. */
2292 if (atomic_read(&device->ap_bio_cnt) > mxb)
2293 return false;
2294 if (test_bit(BITMAP_IO, &device->flags))
2295 return false;
2296 return true;
2297}
2298
2299static inline bool inc_ap_bio_cond(struct drbd_device *device)
2300{
2301 bool rv = false;
2302
2303 spin_lock_irq(&device->resource->req_lock);
2304 rv = may_inc_ap_bio(device);
2305 if (rv)
2306 atomic_inc(&device->ap_bio_cnt);
2307 spin_unlock_irq(&device->resource->req_lock);
2308
2309 return rv;
2310}
2311
2312static inline void inc_ap_bio(struct drbd_device *device)
2313{
2314 /* we wait here
2315 * as long as the device is suspended
2316 * until the bitmap is no longer on the fly during connection
2317 * handshake as long as we would exceed the max_buffer limit.
2318 *
2319 * to avoid races with the reconnect code,
2320 * we need to atomic_inc within the spinlock. */
2321
2322 wait_event(device->misc_wait, inc_ap_bio_cond(device));
2323}
2324
2325static inline void dec_ap_bio(struct drbd_device *device)
2326{
2327 int mxb = drbd_get_max_buffers(device);
2328 int ap_bio = atomic_dec_return(&device->ap_bio_cnt);
2329
2330 D_ASSERT(device, ap_bio >= 0);
2331
2332 if (ap_bio == 0 && test_bit(BITMAP_IO, &device->flags)) {
2333 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
2334 drbd_queue_work(&first_peer_device(device)->
2335 connection->sender_work,
2336 &device->bm_io_work.w);
2337 }
2338
2339 /* this currently does wake_up for every dec_ap_bio!
2340 * maybe rather introduce some type of hysteresis?
2341 * e.g. (ap_bio == mxb/2 || ap_bio == 0) ? */
2342 if (ap_bio < mxb)
2343 wake_up(&device->misc_wait);
2344}
2345
2346static inline bool verify_can_do_stop_sector(struct drbd_device *device)
2347{
2348 return first_peer_device(device)->connection->agreed_pro_version >= 97 &&
2349 first_peer_device(device)->connection->agreed_pro_version != 100;
2350}
2351
2352static inline int drbd_set_ed_uuid(struct drbd_device *device, u64 val)
2353{
2354 int changed = device->ed_uuid != val;
2355 device->ed_uuid = val;
2356 return changed;
2357}
2358
2359static inline int drbd_queue_order_type(struct drbd_device *device)
2360{
2361 /* sorry, we currently have no working implementation
2362 * of distributed TCQ stuff */
2363#ifndef QUEUE_ORDERED_NONE
2364#define QUEUE_ORDERED_NONE 0
2365#endif
2366 return QUEUE_ORDERED_NONE;
2367}
2368
2369static inline struct drbd_connection *first_connection(struct drbd_resource *resource)
2370{
2371 return list_first_entry_or_null(&resource->connections,
2372 struct drbd_connection, connections);
2373}
2374
2375#endif