4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
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>.
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)
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.
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.
29 #include <linux/compiler.h>
30 #include <linux/types.h>
31 #include <linux/list.h>
32 #include <linux/sched.h>
33 #include <linux/bitops.h>
34 #include <linux/slab.h>
35 #include <linux/crypto.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/genhd.h>
43 #include <linux/lru_cache.h>
44 #include <linux/prefetch.h>
47 # define __protected_by(x) __attribute__((require_context(x,1,999,"rdwr")))
48 # define __protected_read_by(x) __attribute__((require_context(x,1,999,"read")))
49 # define __protected_write_by(x) __attribute__((require_context(x,1,999,"write")))
50 # define __must_hold(x) __attribute__((context(x,1,1), require_context(x,1,999,"call")))
52 # define __protected_by(x)
53 # define __protected_read_by(x)
54 # define __protected_write_by(x)
55 # define __must_hold(x)
58 #define __no_warn(lock, stmt) do { __acquire(lock); stmt; __release(lock); } while (0)
60 /* module parameter, defined in drbd_main.c */
61 extern unsigned int minor_count;
62 extern bool disable_sendpage;
63 extern bool allow_oos;
64 extern unsigned int cn_idx;
66 #ifdef CONFIG_DRBD_FAULT_INJECTION
67 extern int enable_faults;
68 extern int fault_rate;
69 extern int fault_devs;
72 extern char usermode_helper[];
75 /* I don't remember why XCPU ...
76 * This is used to wake the asender,
77 * and to interrupt sending the sending task
80 #define DRBD_SIG SIGXCPU
82 /* This is used to stop/restart our threads.
83 * Cannot use SIGTERM nor SIGKILL, since these
84 * are sent out by init on runlevel changes
85 * I choose SIGHUP for now.
87 #define DRBD_SIGKILL SIGHUP
89 /* All EEs on the free list should have ID_VACANT (== 0)
90 * freshly allocated EEs get !ID_VACANT (== 1)
91 * so if it says "cannot dereference null pointer at address 0x00000001",
92 * it is most likely one of these :( */
94 #define ID_IN_SYNC (4711ULL)
95 #define ID_OUT_OF_SYNC (4712ULL)
97 #define ID_SYNCER (-1ULL)
99 #define is_syncer_block_id(id) ((id) == ID_SYNCER)
100 #define UUID_NEW_BM_OFFSET ((u64)0x0001000000000000ULL)
105 /* to shorten dev_warn(DEV, "msg"); and relatives statements */
106 #define DEV (disk_to_dev(mdev->vdisk))
108 #define D_ASSERT(exp) if (!(exp)) \
109 dev_err(DEV, "ASSERT( " #exp " ) in %s:%d\n", __FILE__, __LINE__)
111 #define ERR_IF(exp) if (({ \
112 int _b = (exp) != 0; \
113 if (_b) dev_err(DEV, "ASSERT FAILED: %s: (%s) in %s:%d\n", \
114 __func__, #exp, __FILE__, __LINE__); \
118 /* Defines to control fault insertion */
120 DRBD_FAULT_MD_WR = 0, /* meta data write */
121 DRBD_FAULT_MD_RD = 1, /* read */
122 DRBD_FAULT_RS_WR = 2, /* resync */
123 DRBD_FAULT_RS_RD = 3,
124 DRBD_FAULT_DT_WR = 4, /* data */
125 DRBD_FAULT_DT_RD = 5,
126 DRBD_FAULT_DT_RA = 6, /* data read ahead */
127 DRBD_FAULT_BM_ALLOC = 7, /* bitmap allocation */
128 DRBD_FAULT_AL_EE = 8, /* alloc ee */
129 DRBD_FAULT_RECEIVE = 9, /* Changes some bytes upon receiving a [rs]data block */
135 _drbd_insert_fault(struct drbd_conf *mdev, unsigned int type);
138 drbd_insert_fault(struct drbd_conf *mdev, unsigned int type) {
139 #ifdef CONFIG_DRBD_FAULT_INJECTION
141 (enable_faults & (1<<type)) &&
142 _drbd_insert_fault(mdev, type);
148 /* integer division, round _UP_ to the next integer */
149 #define div_ceil(A, B) ((A)/(B) + ((A)%(B) ? 1 : 0))
150 /* usual integer division */
151 #define div_floor(A, B) ((A)/(B))
153 /* drbd_meta-data.c (still in drbd_main.c) */
154 /* 4th incarnation of the disk layout. */
155 #define DRBD_MD_MAGIC (DRBD_MAGIC+4)
157 extern struct drbd_conf **minor_table;
158 extern struct ratelimit_state drbd_ratelimit_state;
162 /* receiver (data socket) */
164 P_DATA_REPLY = 0x01, /* Response to P_DATA_REQUEST */
165 P_RS_DATA_REPLY = 0x02, /* Response to P_RS_DATA_REQUEST */
168 P_BECOME_SYNC_TARGET = 0x05,
169 P_BECOME_SYNC_SOURCE = 0x06,
170 P_UNPLUG_REMOTE = 0x07, /* Used at various times to hint the peer */
171 P_DATA_REQUEST = 0x08, /* Used to ask for a data block */
172 P_RS_DATA_REQUEST = 0x09, /* Used to ask for a data block for resync */
179 P_AUTH_CHALLENGE = 0x10,
180 P_AUTH_RESPONSE = 0x11,
181 P_STATE_CHG_REQ = 0x12,
183 /* asender (meta socket */
186 P_RECV_ACK = 0x15, /* Used in protocol B */
187 P_WRITE_ACK = 0x16, /* Used in protocol C */
188 P_RS_WRITE_ACK = 0x17, /* Is a P_WRITE_ACK, additionally call set_in_sync(). */
189 P_DISCARD_ACK = 0x18, /* Used in proto C, two-primaries conflict detection */
190 P_NEG_ACK = 0x19, /* Sent if local disk is unusable */
191 P_NEG_DREPLY = 0x1a, /* Local disk is broken... */
192 P_NEG_RS_DREPLY = 0x1b, /* Local disk is broken... */
193 P_BARRIER_ACK = 0x1c,
194 P_STATE_CHG_REPLY = 0x1d,
196 /* "new" commands, no longer fitting into the ordering scheme above */
198 P_OV_REQUEST = 0x1e, /* data socket */
200 P_OV_RESULT = 0x20, /* meta socket */
201 P_CSUM_RS_REQUEST = 0x21, /* data socket */
202 P_RS_IS_IN_SYNC = 0x22, /* meta socket */
203 P_SYNC_PARAM89 = 0x23, /* data socket, protocol version 89 replacement for P_SYNC_PARAM */
204 P_COMPRESSED_BITMAP = 0x24, /* compressed or otherwise encoded bitmap transfer */
205 /* P_CKPT_FENCE_REQ = 0x25, * currently reserved for protocol D */
206 /* P_CKPT_DISABLE_REQ = 0x26, * currently reserved for protocol D */
207 P_DELAY_PROBE = 0x27, /* is used on BOTH sockets */
208 P_OUT_OF_SYNC = 0x28, /* Mark as out of sync (Outrunning), data socket */
209 P_RS_CANCEL = 0x29, /* meta: Used to cancel RS_DATA_REQUEST packet by SyncSource */
212 P_MAY_IGNORE = 0x100, /* Flag to test if (cmd > P_MAY_IGNORE) ... */
213 P_MAX_OPT_CMD = 0x101,
215 /* special command ids for handshake */
217 P_HAND_SHAKE_M = 0xfff1, /* First Packet on the MetaSock */
218 P_HAND_SHAKE_S = 0xfff2, /* First Packet on the Socket */
220 P_HAND_SHAKE = 0xfffe /* FIXED for the next century! */
223 static inline const char *cmdname(enum drbd_packets cmd)
225 /* THINK may need to become several global tables
226 * when we want to support more than
228 static const char *cmdnames[] = {
230 [P_DATA_REPLY] = "DataReply",
231 [P_RS_DATA_REPLY] = "RSDataReply",
232 [P_BARRIER] = "Barrier",
233 [P_BITMAP] = "ReportBitMap",
234 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
235 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
236 [P_UNPLUG_REMOTE] = "UnplugRemote",
237 [P_DATA_REQUEST] = "DataRequest",
238 [P_RS_DATA_REQUEST] = "RSDataRequest",
239 [P_SYNC_PARAM] = "SyncParam",
240 [P_SYNC_PARAM89] = "SyncParam89",
241 [P_PROTOCOL] = "ReportProtocol",
242 [P_UUIDS] = "ReportUUIDs",
243 [P_SIZES] = "ReportSizes",
244 [P_STATE] = "ReportState",
245 [P_SYNC_UUID] = "ReportSyncUUID",
246 [P_AUTH_CHALLENGE] = "AuthChallenge",
247 [P_AUTH_RESPONSE] = "AuthResponse",
249 [P_PING_ACK] = "PingAck",
250 [P_RECV_ACK] = "RecvAck",
251 [P_WRITE_ACK] = "WriteAck",
252 [P_RS_WRITE_ACK] = "RSWriteAck",
253 [P_DISCARD_ACK] = "DiscardAck",
254 [P_NEG_ACK] = "NegAck",
255 [P_NEG_DREPLY] = "NegDReply",
256 [P_NEG_RS_DREPLY] = "NegRSDReply",
257 [P_BARRIER_ACK] = "BarrierAck",
258 [P_STATE_CHG_REQ] = "StateChgRequest",
259 [P_STATE_CHG_REPLY] = "StateChgReply",
260 [P_OV_REQUEST] = "OVRequest",
261 [P_OV_REPLY] = "OVReply",
262 [P_OV_RESULT] = "OVResult",
263 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
264 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
265 [P_COMPRESSED_BITMAP] = "CBitmap",
266 [P_DELAY_PROBE] = "DelayProbe",
267 [P_OUT_OF_SYNC] = "OutOfSync",
271 if (cmd == P_HAND_SHAKE_M)
273 if (cmd == P_HAND_SHAKE_S)
275 if (cmd == P_HAND_SHAKE)
277 if (cmd >= P_MAX_CMD)
279 return cmdnames[cmd];
282 /* for sending/receiving the bitmap,
283 * possibly in some encoding scheme */
286 * stores total bits and long words
287 * of the bitmap, so we don't need to
288 * call the accessor functions over and again. */
289 unsigned long bm_bits;
290 unsigned long bm_words;
291 /* during xfer, current position within the bitmap */
292 unsigned long bit_offset;
293 unsigned long word_offset;
295 /* statistics; index: (h->command == P_BITMAP) */
300 extern void INFO_bm_xfer_stats(struct drbd_conf *mdev,
301 const char *direction, struct bm_xfer_ctx *c);
303 static inline void bm_xfer_ctx_bit_to_word_offset(struct bm_xfer_ctx *c)
305 /* word_offset counts "native long words" (32 or 64 bit),
307 * Encoded packet may end at an unaligned bit offset.
308 * In case a fallback clear text packet is transmitted in
309 * between, we adjust this offset back to the last 64bit
310 * aligned "native long word", which makes coding and decoding
311 * the plain text bitmap much more convenient. */
312 #if BITS_PER_LONG == 64
313 c->word_offset = c->bit_offset >> 6;
314 #elif BITS_PER_LONG == 32
315 c->word_offset = c->bit_offset >> 5;
316 c->word_offset &= ~(1UL);
318 # error "unsupported BITS_PER_LONG"
323 #define __packed __attribute__((packed))
326 /* This is the layout for a packet on the wire.
327 * The byteorder is the network byte order.
328 * (except block_id and barrier fields.
329 * these are pointers to local structs
330 * and have no relevance for the partner,
331 * which just echoes them as received.)
333 * NOTE that the payload starts at a long aligned offset,
334 * regardless of 32 or 64 bit arch!
339 u16 length; /* bytes of data after this header */
343 /* Header for big packets, Used for data packets exceeding 64kB */
345 u16 magic; /* use DRBD_MAGIC_BIG here */
347 u32 length; /* Use only 24 bits of that. Ignore the highest 8 bit. */
352 struct p_header80 h80;
353 struct p_header95 h95;
357 * short commands, packets without payload, plain p_header:
360 * P_BECOME_SYNC_TARGET
361 * P_BECOME_SYNC_SOURCE
366 * commands with out-of-struct payload:
367 * P_BITMAP (no additional fields)
368 * P_DATA, P_DATA_REPLY (see p_data)
369 * P_COMPRESSED_BITMAP (see receive_compressed_bitmap)
372 /* these defines must not be changed without changing the protocol version */
373 #define DP_HARDBARRIER 1 /* depricated */
374 #define DP_RW_SYNC 2 /* equals REQ_SYNC */
375 #define DP_MAY_SET_IN_SYNC 4
376 #define DP_UNPLUG 8 /* not used anymore */
377 #define DP_FUA 16 /* equals REQ_FUA */
378 #define DP_FLUSH 32 /* equals REQ_FLUSH */
379 #define DP_DISCARD 64 /* equals REQ_DISCARD */
383 u64 sector; /* 64 bits sector number */
384 u64 block_id; /* to identify the request in protocol B&C */
390 * commands which share a struct:
392 * P_RECV_ACK (proto B), P_WRITE_ACK (proto C),
393 * P_DISCARD_ACK (proto C, two-primaries conflict detection)
395 * P_DATA_REQUEST, P_RS_DATA_REQUEST
398 struct p_header80 head;
407 struct p_header80 head;
411 u32 pad; /* to multiple of 8 Byte */
415 * commands with their own struct for additional fields:
424 struct p_header80 head; /* 8 bytes */
429 /* should be more than enough for future enhancements
430 * for now, feature_flags and the reserverd array shall be zero.
436 /* 80 bytes, FIXED for the next century */
439 struct p_header80 head;
440 u32 barrier; /* barrier number _handle_ only */
441 u32 pad; /* to multiple of 8 Byte */
444 struct p_barrier_ack {
445 struct p_header80 head;
451 struct p_header80 head;
454 /* Since protocol version 88 and higher. */
458 struct p_rs_param_89 {
459 struct p_header80 head;
461 /* protocol version 89: */
462 char verify_alg[SHARED_SECRET_MAX];
463 char csums_alg[SHARED_SECRET_MAX];
466 struct p_rs_param_95 {
467 struct p_header80 head;
469 char verify_alg[SHARED_SECRET_MAX];
470 char csums_alg[SHARED_SECRET_MAX];
477 enum drbd_conn_flags {
483 struct p_header80 head;
491 /* Since protocol version 87 and higher. */
492 char integrity_alg[0];
497 struct p_header80 head;
498 u64 uuid[UI_EXTENDED_SIZE];
502 struct p_header80 head;
507 struct p_header80 head;
508 u64 d_size; /* size of disk */
509 u64 u_size; /* user requested size */
510 u64 c_size; /* current exported size */
511 u32 max_bio_size; /* Maximal size of a BIO */
512 u16 queue_order_type; /* not yet implemented in DRBD*/
513 u16 dds_flags; /* use enum dds_flags here. */
517 struct p_header80 head;
522 struct p_header80 head;
527 struct p_req_state_reply {
528 struct p_header80 head;
532 struct p_drbd06_param {
543 struct p_header80 head;
549 struct p_block_desc {
550 struct p_header80 head;
553 u32 pad; /* to multiple of 8 Byte */
556 /* Valid values for the encoding field.
557 * Bump proto version when changing this. */
558 enum drbd_bitmap_code {
559 /* RLE_VLI_Bytes = 0,
560 * and other bit variants had been defined during
561 * algorithm evaluation. */
565 struct p_compressed_bm {
566 struct p_header80 head;
567 /* (encoding & 0x0f): actual encoding, see enum drbd_bitmap_code
568 * (encoding & 0x80): polarity (set/unset) of first runlength
569 * ((encoding >> 4) & 0x07): pad_bits, number of trailing zero bits
570 * used to pad up to head.length bytes
577 struct p_delay_probe93 {
578 struct p_header80 head;
579 u32 seq_num; /* sequence number to match the two probe packets */
580 u32 offset; /* usecs the probe got sent after the reference time point */
583 /* DCBP: Drbd Compressed Bitmap Packet ... */
584 static inline enum drbd_bitmap_code
585 DCBP_get_code(struct p_compressed_bm *p)
587 return (enum drbd_bitmap_code)(p->encoding & 0x0f);
591 DCBP_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
594 p->encoding = (p->encoding & ~0xf) | code;
598 DCBP_get_start(struct p_compressed_bm *p)
600 return (p->encoding & 0x80) != 0;
604 DCBP_set_start(struct p_compressed_bm *p, int set)
606 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
610 DCBP_get_pad_bits(struct p_compressed_bm *p)
612 return (p->encoding >> 4) & 0x7;
616 DCBP_set_pad_bits(struct p_compressed_bm *p, int n)
619 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
622 /* one bitmap packet, including the p_header,
623 * should fit within one _architecture independend_ page.
624 * so we need to use the fixed size 4KiB page size
625 * most architectures have used for a long time.
627 #define BM_PACKET_PAYLOAD_BYTES (4096 - sizeof(struct p_header80))
628 #define BM_PACKET_WORDS (BM_PACKET_PAYLOAD_BYTES/sizeof(long))
629 #define BM_PACKET_VLI_BYTES_MAX (4096 - sizeof(struct p_compressed_bm))
630 #if (PAGE_SIZE < 4096)
631 /* drbd_send_bitmap / receive_bitmap would break horribly */
632 #error "PAGE_SIZE too small"
636 union p_header header;
637 struct p_handshake handshake;
639 struct p_block_ack block_ack;
640 struct p_barrier barrier;
641 struct p_barrier_ack barrier_ack;
642 struct p_rs_param_89 rs_param_89;
643 struct p_rs_param_95 rs_param_95;
644 struct p_protocol protocol;
645 struct p_sizes sizes;
646 struct p_uuids uuids;
647 struct p_state state;
648 struct p_req_state req_state;
649 struct p_req_state_reply req_state_reply;
650 struct p_block_req block_req;
651 struct p_delay_probe93 delay_probe93;
652 struct p_rs_uuid rs_uuid;
653 struct p_block_desc block_desc;
656 /**********************************************************************/
657 enum drbd_thread_state {
666 struct task_struct *task;
667 struct completion stop;
668 enum drbd_thread_state t_state;
669 int (*function) (struct drbd_thread *);
670 struct drbd_conf *mdev;
674 static inline enum drbd_thread_state get_t_state(struct drbd_thread *thi)
676 /* THINK testing the t_state seems to be uncritical in all cases
677 * (but thread_{start,stop}), so we can read it *without* the lock.
685 typedef int (*drbd_work_cb)(struct drbd_conf *, struct drbd_work *, int cancel);
687 struct list_head list;
691 struct drbd_tl_epoch;
692 struct drbd_request {
694 struct drbd_conf *mdev;
696 /* if local IO is not allowed, will be NULL.
697 * if local IO _is_ allowed, holds the locally submitted bio clone,
698 * or, after local IO completion, the ERR_PTR(error).
699 * see drbd_endio_pri(). */
700 struct bio *private_bio;
702 struct hlist_node collision;
705 unsigned int epoch; /* barrier_nr */
707 /* barrier_nr: used to check on "completion" whether this req was in
708 * the current epoch, and we therefore have to close it,
709 * starting a new epoch...
712 struct list_head tl_requests; /* ring list in the transfer log */
713 struct bio *master_bio; /* master bio pointer */
714 unsigned long rq_state; /* see comments above _req_mod() */
715 unsigned long start_time;
718 struct drbd_tl_epoch {
720 struct list_head requests; /* requests before */
721 struct drbd_tl_epoch *next; /* pointer to the next barrier */
722 unsigned int br_number; /* the barriers identifier. */
723 int n_writes; /* number of requests attached before this barrier */
728 /* These Tl_epoch_entries may be in one of 6 lists:
729 active_ee .. data packet being written
730 sync_ee .. syncer block being written
731 done_ee .. block written, need to send P_WRITE_ACK
732 read_ee .. [RS]P_DATA_REQUEST being read
736 struct list_head list;
737 unsigned int barrier_nr;
738 atomic_t epoch_size; /* increased on every request added. */
739 atomic_t active; /* increased on every req. added, and dec on every finished. */
743 /* drbd_epoch flag bits */
745 DE_HAVE_BARRIER_NUMBER,
752 EV_CLEANUP = 32, /* used as flag */
755 struct drbd_wq_barrier {
757 struct completion done;
765 struct drbd_epoch_entry {
767 struct hlist_node collision;
768 struct drbd_epoch *epoch; /* for writes */
769 struct drbd_conf *mdev;
771 atomic_t pending_bios;
773 /* see comments on ee flag bits below */
778 struct digest_info *digest;
783 * While corresponding bios are in flight, the only modification will be
784 * set_bit WAS_ERROR, which has to be atomic.
785 * If no bios are in flight yet, or all have been completed,
786 * non-atomic modification to ee->flags is ok.
789 __EE_CALL_AL_COMPLETE_IO,
790 __EE_MAY_SET_IN_SYNC,
792 /* In case a barrier failed,
793 * we need to resubmit without the barrier flag. */
796 /* we may have several bios per epoch entry.
797 * if any of those fail, we set this flag atomically
798 * from the endio callback */
801 /* This ee has a pointer to a digest instead of a block id */
804 #define EE_CALL_AL_COMPLETE_IO (1<<__EE_CALL_AL_COMPLETE_IO)
805 #define EE_MAY_SET_IN_SYNC (1<<__EE_MAY_SET_IN_SYNC)
806 #define EE_RESUBMITTED (1<<__EE_RESUBMITTED)
807 #define EE_WAS_ERROR (1<<__EE_WAS_ERROR)
808 #define EE_HAS_DIGEST (1<<__EE_HAS_DIGEST)
810 /* global flag bits */
812 CREATE_BARRIER, /* next P_DATA is preceded by a P_BARRIER */
813 SIGNAL_ASENDER, /* whether asender wants to be interrupted */
814 SEND_PING, /* whether asender should send a ping asap */
816 UNPLUG_REMOTE, /* sending a "UnplugRemote" could help */
817 MD_DIRTY, /* current uuids and flags not yet on disk */
818 DISCARD_CONCURRENT, /* Set on one node, cleared on the peer! */
819 USE_DEGR_WFC_T, /* degr-wfc-timeout instead of wfc-timeout. */
820 CLUSTER_ST_CHANGE, /* Cluster wide state change going on... */
823 CRASHED_PRIMARY, /* This node was a crashed primary.
824 * Gets cleared when the state.conn
825 * goes into C_CONNECTED state. */
828 MD_NO_FUA, /* Users wants us to not use FUA/FLUSH on meta data dev */
829 SUSPEND_IO, /* suspend application io */
830 BITMAP_IO, /* suspend application io;
831 once no more io in flight, start bitmap io */
832 BITMAP_IO_QUEUED, /* Started bitmap IO */
833 GO_DISKLESS, /* Disk is being detached, on io-error or admin request. */
834 WAS_IO_ERROR, /* Local disk failed returned IO error */
835 FORCE_DETACH, /* Force-detach from local disk, aborting any pending local IO */
836 RESYNC_AFTER_NEG, /* Resync after online grow after the attach&negotiate finished. */
837 NET_CONGESTED, /* The data socket is congested */
839 CONFIG_PENDING, /* serialization of (re)configuration requests.
840 * if set, also prevents the device from dying */
841 DEVICE_DYING, /* device became unconfigured,
842 * but worker thread is still handling the cleanup.
843 * reconfiguring (nl_disk_conf, nl_net_conf) is dissalowed,
844 * while this is set. */
845 RESIZE_PENDING, /* Size change detected locally, waiting for the response from
846 * the peer, if it changed there as well. */
847 CONN_DRY_RUN, /* Expect disconnect after resync handshake. */
848 GOT_PING_ACK, /* set when we receive a ping_ack packet, misc wait gets woken */
849 NEW_CUR_UUID, /* Create new current UUID when thawing IO */
850 AL_SUSPENDED, /* Activity logging is currently suspended. */
851 AHEAD_TO_SYNC_SOURCE, /* Ahead -> SyncSource queued */
852 STATE_SENT, /* Do not change state/UUIDs while this is set */
855 struct drbd_bitmap; /* opaque for drbd_conf */
857 /* definition of bits in bm_flags to be used in drbd_bm_lock
858 * and drbd_bitmap_io and friends. */
860 /* do we need to kfree, or vfree bm_pages? */
861 BM_P_VMALLOCED = 0x10000, /* internal use only, will be masked out */
863 /* currently locked for bulk operation */
864 BM_LOCKED_MASK = 0xf,
866 /* in detail, that is: */
871 /* so we can mark it locked for bulk operation,
872 * and still allow all non-bulk operations */
875 /* (test bit, count bit) allowed (common case) */
876 BM_LOCKED_TEST_ALLOWED = BM_DONT_CLEAR | BM_DONT_SET | BM_IS_LOCKED,
878 /* testing bits, as well as setting new bits allowed, but clearing bits
879 * would be unexpected. Used during bitmap receive. Setting new bits
880 * requires sending of "out-of-sync" information, though. */
881 BM_LOCKED_SET_ALLOWED = BM_DONT_CLEAR | BM_IS_LOCKED,
883 /* for drbd_bm_write_copy_pages, everything is allowed,
884 * only concurrent bulk operations are locked out. */
885 BM_LOCKED_CHANGE_ALLOWED = BM_IS_LOCKED,
888 struct drbd_work_queue {
890 struct semaphore s; /* producers up it, worker down()s it */
891 spinlock_t q_lock; /* to protect the list. */
895 struct drbd_work_queue work;
897 struct socket *socket;
898 /* this way we get our
899 * send/receive buffers off the stack */
900 union p_polymorph sbuf;
901 union p_polymorph rbuf;
905 u64 md_offset; /* sector offset to 'super' block */
907 u64 la_size_sect; /* last agreed size, unit sectors */
913 s32 al_offset; /* signed relative sector offset to al area */
914 s32 bm_offset; /* signed relative sector offset to bitmap */
916 /* u32 al_nr_extents; important for restoring the AL
917 * is stored into sync_conf.al_extents, which in turn
918 * gets applied to act_log->nr_elements
922 /* for sync_conf and other types... */
923 #define NL_PACKET(name, number, fields) struct name { fields };
924 #define NL_INTEGER(pn,pr,member) int member;
925 #define NL_INT64(pn,pr,member) __u64 member;
926 #define NL_BIT(pn,pr,member) unsigned member:1;
927 #define NL_STRING(pn,pr,member,len) unsigned char member[len]; int member ## _len;
928 #include <linux/drbd_nl.h>
930 struct drbd_backing_dev {
931 struct block_device *backing_bdev;
932 struct block_device *md_bdev;
934 struct disk_conf dc; /* The user provided config... */
935 sector_t known_size; /* last known size of that backing device */
947 int (*io_fn)(struct drbd_conf *mdev);
948 void (*done)(struct drbd_conf *mdev, int rv);
951 enum write_ordering_e {
959 unsigned int head_index;
964 /* things that are stored as / read from meta data on disk */
967 /* configured by drbdsetup */
968 struct net_conf *net_conf; /* protected by get_net_conf() and put_net_conf() */
969 struct syncer_conf sync_conf;
970 struct drbd_backing_dev *ldev __protected_by(local);
972 sector_t p_size; /* partner's disk size */
973 struct request_queue *rq_queue;
974 struct block_device *this_bdev;
975 struct gendisk *vdisk;
977 struct drbd_socket data; /* data/barrier/cstate/parameter packets */
978 struct drbd_socket meta; /* ping/ack (metadata) packets */
979 int agreed_pro_version; /* actually used protocol version */
980 unsigned long last_received; /* in jiffies, either socket */
981 unsigned int ko_count;
982 struct drbd_work resync_work,
987 struct timer_list resync_timer;
988 struct timer_list md_sync_timer;
989 struct timer_list start_resync_timer;
990 struct timer_list request_timer;
991 #ifdef DRBD_DEBUG_MD_SYNC
995 } last_md_mark_dirty;
998 /* Used after attach while negotiating new disk state. */
999 union drbd_state new_state_tmp;
1001 union drbd_state state;
1002 wait_queue_head_t misc_wait;
1003 wait_queue_head_t state_wait; /* upon each state change. */
1004 wait_queue_head_t net_cnt_wait;
1005 unsigned int send_cnt;
1006 unsigned int recv_cnt;
1007 unsigned int read_cnt;
1008 unsigned int writ_cnt;
1009 unsigned int al_writ_cnt;
1010 unsigned int bm_writ_cnt;
1011 atomic_t ap_bio_cnt; /* Requests we need to complete */
1012 atomic_t ap_pending_cnt; /* AP data packets on the wire, ack expected */
1013 atomic_t rs_pending_cnt; /* RS request/data packets on the wire */
1014 atomic_t unacked_cnt; /* Need to send replys for */
1015 atomic_t local_cnt; /* Waiting for local completion */
1016 atomic_t net_cnt; /* Users of net_conf */
1017 spinlock_t req_lock;
1018 struct drbd_tl_epoch *unused_spare_tle; /* for pre-allocation */
1019 struct drbd_tl_epoch *newest_tle;
1020 struct drbd_tl_epoch *oldest_tle;
1021 struct list_head out_of_sequence_requests;
1022 struct list_head barrier_acked_requests;
1023 struct hlist_head *tl_hash;
1024 unsigned int tl_hash_s;
1026 /* blocks to resync in this run [unit BM_BLOCK_SIZE] */
1027 unsigned long rs_total;
1028 /* number of resync blocks that failed in this run */
1029 unsigned long rs_failed;
1030 /* Syncer's start time [unit jiffies] */
1031 unsigned long rs_start;
1032 /* cumulated time in PausedSyncX state [unit jiffies] */
1033 unsigned long rs_paused;
1034 /* skipped because csum was equal [unit BM_BLOCK_SIZE] */
1035 unsigned long rs_same_csum;
1036 #define DRBD_SYNC_MARKS 8
1037 #define DRBD_SYNC_MARK_STEP (3*HZ)
1038 /* block not up-to-date at mark [unit BM_BLOCK_SIZE] */
1039 unsigned long rs_mark_left[DRBD_SYNC_MARKS];
1040 /* marks's time [unit jiffies] */
1041 unsigned long rs_mark_time[DRBD_SYNC_MARKS];
1042 /* current index into rs_mark_{left,time} */
1045 /* where does the admin want us to start? (sector) */
1046 sector_t ov_start_sector;
1047 /* where are we now? (sector) */
1048 sector_t ov_position;
1049 /* Start sector of out of sync range (to merge printk reporting). */
1050 sector_t ov_last_oos_start;
1051 /* size of out-of-sync range in sectors. */
1052 sector_t ov_last_oos_size;
1053 unsigned long ov_left; /* in bits */
1054 struct crypto_hash *csums_tfm;
1055 struct crypto_hash *verify_tfm;
1057 unsigned long last_reattach_jif;
1058 unsigned long last_reconnect_jif;
1059 struct drbd_thread receiver;
1060 struct drbd_thread worker;
1061 struct drbd_thread asender;
1062 struct drbd_bitmap *bitmap;
1063 unsigned long bm_resync_fo; /* bit offset for drbd_bm_find_next */
1065 /* Used to track operations of resync... */
1066 struct lru_cache *resync;
1067 /* Number of locked elements in resync LRU */
1068 unsigned int resync_locked;
1069 /* resync extent number waiting for application requests */
1070 unsigned int resync_wenr;
1074 struct drbd_epoch *current_epoch;
1075 spinlock_t epoch_lock;
1076 unsigned int epochs;
1077 enum write_ordering_e write_ordering;
1078 struct list_head active_ee; /* IO in progress (P_DATA gets written to disk) */
1079 struct list_head sync_ee; /* IO in progress (P_RS_DATA_REPLY gets written to disk) */
1080 struct list_head done_ee; /* send ack */
1081 struct list_head read_ee; /* IO in progress (any read) */
1082 struct list_head net_ee; /* zero-copy network send in progress */
1083 struct hlist_head *ee_hash; /* is proteced by req_lock! */
1084 unsigned int ee_hash_s;
1086 /* this one is protected by ee_lock, single thread */
1087 struct drbd_epoch_entry *last_write_w_barrier;
1089 int next_barrier_nr;
1090 struct hlist_head *app_reads_hash; /* is proteced by req_lock */
1091 struct list_head resync_reads;
1092 atomic_t pp_in_use; /* allocated from page pool */
1093 atomic_t pp_in_use_by_net; /* sendpage()d, still referenced by tcp */
1094 wait_queue_head_t ee_wait;
1095 struct page *md_io_page; /* one page buffer for md_io */
1096 struct page *md_io_tmpp; /* for logical_block_size != 512 */
1097 struct drbd_md_io md_io;
1098 atomic_t md_io_in_use; /* protects the md_io, md_io_page and md_io_tmpp */
1100 wait_queue_head_t al_wait;
1101 struct lru_cache *act_log; /* activity log */
1102 unsigned int al_tr_number;
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 */
1111 wait_queue_head_t seq_wait;
1112 atomic_t packet_seq;
1113 unsigned int peer_seq;
1114 spinlock_t peer_seq_lock;
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;
1136 static inline struct drbd_conf *minor_to_mdev(unsigned int minor)
1138 struct drbd_conf *mdev;
1140 mdev = minor < minor_count ? minor_table[minor] : NULL;
1145 static inline unsigned int mdev_to_minor(struct drbd_conf *mdev)
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))
1156 * drbd_put_data_sock(mdev);
1158 static inline int drbd_get_data_sock(struct drbd_conf *mdev)
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);
1170 static inline void drbd_put_data_sock(struct drbd_conf *mdev)
1172 mutex_unlock(&mdev->data.mutex);
1176 * function declarations
1177 *************************/
1181 enum chg_state_flags {
1184 CS_WAIT_COMPLETE = 4,
1186 CS_ORDERED = CS_WAIT_COMPLETE + CS_SERIALIZE,
1191 DDSF_NO_RESYNC = 2, /* Do not run a resync for the new space */
1194 extern void drbd_init_set_defaults(struct drbd_conf *mdev);
1195 extern 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);
1199 extern void drbd_force_state(struct drbd_conf *, union drbd_state,
1201 extern enum drbd_state_rv _drbd_request_state(struct drbd_conf *,
1204 enum chg_state_flags);
1205 extern enum drbd_state_rv __drbd_set_state(struct drbd_conf *, union drbd_state,
1206 enum chg_state_flags,
1207 struct completion *done);
1208 extern void print_st_err(struct drbd_conf *, union drbd_state,
1209 union drbd_state, int);
1210 extern int drbd_thread_start(struct drbd_thread *thi);
1211 extern void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait);
1213 extern void drbd_thread_current_set_cpu(struct drbd_conf *mdev);
1214 extern void drbd_calc_cpu_mask(struct drbd_conf *mdev);
1216 #define drbd_thread_current_set_cpu(A) ({})
1217 #define drbd_calc_cpu_mask(A) ({})
1219 extern void drbd_free_resources(struct drbd_conf *mdev);
1220 extern void tl_release(struct drbd_conf *mdev, unsigned int barrier_nr,
1221 unsigned int set_size);
1222 extern void tl_clear(struct drbd_conf *mdev);
1223 extern void _tl_add_barrier(struct drbd_conf *, struct drbd_tl_epoch *);
1224 extern void drbd_free_sock(struct drbd_conf *mdev);
1225 extern int drbd_send(struct drbd_conf *mdev, struct socket *sock,
1226 void *buf, size_t size, unsigned msg_flags);
1227 extern int drbd_send_protocol(struct drbd_conf *mdev);
1228 extern int drbd_send_uuids(struct drbd_conf *mdev);
1229 extern int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev);
1230 extern int drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev);
1231 extern int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags);
1232 extern int drbd_send_state(struct drbd_conf *mdev, union drbd_state s);
1233 extern int drbd_send_current_state(struct drbd_conf *mdev);
1234 extern 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
1239 extern int drbd_send_cmd(struct drbd_conf *mdev, int use_data_socket,
1240 enum drbd_packets cmd, struct p_header80 *h,
1242 extern int drbd_send_cmd2(struct drbd_conf *mdev, enum drbd_packets cmd,
1243 char *data, size_t size);
1244 extern int drbd_send_sync_param(struct drbd_conf *mdev, struct syncer_conf *sc);
1245 extern int drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr,
1247 extern int drbd_send_ack(struct drbd_conf *mdev, enum drbd_packets cmd,
1248 struct drbd_epoch_entry *e);
1249 extern int drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packets cmd,
1250 struct p_block_req *rp);
1251 extern int drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packets cmd,
1252 struct p_data *dp, int data_size);
1253 extern int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packets cmd,
1254 sector_t sector, int blksize, u64 block_id);
1255 extern int drbd_send_oos(struct drbd_conf *mdev, struct drbd_request *req);
1256 extern int drbd_send_block(struct drbd_conf *mdev, enum drbd_packets cmd,
1257 struct drbd_epoch_entry *e);
1258 extern int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req);
1259 extern int drbd_send_drequest(struct drbd_conf *mdev, int cmd,
1260 sector_t sector, int size, u64 block_id);
1261 extern 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);
1265 extern int drbd_send_ov_request(struct drbd_conf *mdev,sector_t sector,int size);
1267 extern int drbd_send_bitmap(struct drbd_conf *mdev);
1268 extern int _drbd_send_bitmap(struct drbd_conf *mdev);
1269 extern int drbd_send_sr_reply(struct drbd_conf *mdev, enum drbd_state_rv retcode);
1270 extern void drbd_free_bc(struct drbd_backing_dev *ldev);
1271 extern void drbd_mdev_cleanup(struct drbd_conf *mdev);
1272 void drbd_print_uuids(struct drbd_conf *mdev, const char *text);
1274 extern void drbd_md_sync(struct drbd_conf *mdev);
1275 extern int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev);
1276 extern void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local);
1277 extern void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local);
1278 extern void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local);
1279 extern void _drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local);
1280 extern void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local);
1281 extern void drbd_md_set_flag(struct drbd_conf *mdev, int flags) __must_hold(local);
1282 extern void drbd_md_clear_flag(struct drbd_conf *mdev, int flags)__must_hold(local);
1283 extern int drbd_md_test_flag(struct drbd_backing_dev *, int);
1284 #ifndef DRBD_DEBUG_MD_SYNC
1285 extern void drbd_md_mark_dirty(struct drbd_conf *mdev);
1287 #define drbd_md_mark_dirty(m) drbd_md_mark_dirty_(m, __LINE__ , __func__ )
1288 extern void drbd_md_mark_dirty_(struct drbd_conf *mdev,
1289 unsigned int line, const char *func);
1291 extern 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);
1295 extern int drbd_bitmap_io(struct drbd_conf *mdev,
1296 int (*io_fn)(struct drbd_conf *),
1297 char *why, enum bm_flag flags);
1298 extern int drbd_bmio_set_n_write(struct drbd_conf *mdev);
1299 extern int drbd_bmio_clear_n_write(struct drbd_conf *mdev);
1300 extern void drbd_go_diskless(struct drbd_conf *mdev);
1301 extern void drbd_ldev_destroy(struct drbd_conf *mdev);
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. */
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)
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)
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)
1325 #if BITS_PER_LONG == 32
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
1331 #define cpu_to_lel(A) cpu_to_le64(A)
1332 #define lel_to_cpu(A) le64_to_cpu(A)
1334 #error "LN2 of BITS_PER_LONG unknown!"
1338 /* 16MB sized 'bitmap extent' to track syncer usage */
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;
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! */
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.
1358 #define SLEEP_TIME (HZ/10)
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)
1367 #if (BM_EXT_SHIFT != 24) || (BM_BLOCK_SHIFT != 12)
1368 #error "HAVE YOU FIXED drbdmeta AS WELL??"
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)
1376 /* bit to represented kilo byte conversion */
1377 #define Bit2KB(bits) ((bits)<<(BM_BLOCK_SHIFT-10))
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))
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)
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))
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)
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
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
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
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)
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. */
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
1436 #define DRBD_MAX_BIO_SIZE (1U<<(9+HT_SHIFT))
1437 #define DRBD_MAX_BIO_SIZE_SAFE (1 << 12) /* Works always = 4k */
1439 #define DRBD_MAX_SIZE_H80_PACKET (1 << 15) /* The old header only allows packets up to 32Kib data */
1441 /* Number of elements in the app_reads_hash */
1442 #define APP_R_HSIZE 15
1444 extern int drbd_bm_init(struct drbd_conf *mdev);
1445 extern int drbd_bm_resize(struct drbd_conf *mdev, sector_t sectors, int set_new_bits);
1446 extern void drbd_bm_cleanup(struct drbd_conf *mdev);
1447 extern void drbd_bm_set_all(struct drbd_conf *mdev);
1448 extern void drbd_bm_clear_all(struct drbd_conf *mdev);
1449 /* set/clear/test only a few bits at a time */
1450 extern int drbd_bm_set_bits(
1451 struct drbd_conf *mdev, unsigned long s, unsigned long e);
1452 extern int drbd_bm_clear_bits(
1453 struct drbd_conf *mdev, unsigned long s, unsigned long e);
1454 extern 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 */
1458 extern void _drbd_bm_set_bits(struct drbd_conf *mdev,
1459 const unsigned long s, const unsigned long e);
1460 extern int drbd_bm_test_bit(struct drbd_conf *mdev, unsigned long bitnr);
1461 extern int drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr);
1462 extern int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(local);
1463 extern int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local);
1464 extern int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local);
1465 extern int drbd_bm_write_copy_pages(struct drbd_conf *mdev) __must_hold(local);
1466 extern unsigned long drbd_bm_ALe_set_all(struct drbd_conf *mdev,
1467 unsigned long al_enr);
1468 extern size_t drbd_bm_words(struct drbd_conf *mdev);
1469 extern unsigned long drbd_bm_bits(struct drbd_conf *mdev);
1470 extern sector_t drbd_bm_capacity(struct drbd_conf *mdev);
1472 #define DRBD_END_OF_BITMAP (~(unsigned long)0)
1473 extern unsigned long drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo);
1474 /* bm_find_next variants for use while you hold drbd_bm_lock() */
1475 extern unsigned long _drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo);
1476 extern unsigned long _drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo);
1477 extern unsigned long _drbd_bm_total_weight(struct drbd_conf *mdev);
1478 extern unsigned long drbd_bm_total_weight(struct drbd_conf *mdev);
1479 extern int drbd_bm_rs_done(struct drbd_conf *mdev);
1480 /* for receive_bitmap */
1481 extern void drbd_bm_merge_lel(struct drbd_conf *mdev, size_t offset,
1482 size_t number, unsigned long *buffer);
1483 /* for _drbd_send_bitmap */
1484 extern void drbd_bm_get_lel(struct drbd_conf *mdev, size_t offset,
1485 size_t number, unsigned long *buffer);
1487 extern void drbd_bm_lock(struct drbd_conf *mdev, char *why, enum bm_flag flags);
1488 extern void drbd_bm_unlock(struct drbd_conf *mdev);
1491 extern struct kmem_cache *drbd_request_cache;
1492 extern struct kmem_cache *drbd_ee_cache; /* epoch entries */
1493 extern struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
1494 extern struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
1495 extern mempool_t *drbd_request_mempool;
1496 extern mempool_t *drbd_ee_mempool;
1498 /* drbd's page pool, used to buffer data received from the peer,
1499 * or data requested by the peer.
1501 * This does not have an emergency reserve.
1503 * When allocating from this pool, it first takes pages from the pool.
1504 * Only if the pool is depleted will try to allocate from the system.
1506 * The assumption is that pages taken from this pool will be processed,
1507 * and given back, "quickly", and then can be recycled, so we can avoid
1508 * frequent calls to alloc_page(), and still will be able to make progress even
1509 * under memory pressure.
1511 extern struct page *drbd_pp_pool;
1512 extern spinlock_t drbd_pp_lock;
1513 extern int drbd_pp_vacant;
1514 extern wait_queue_head_t drbd_pp_wait;
1516 /* We also need a standard (emergency-reserve backed) page pool
1517 * for meta data IO (activity log, bitmap).
1518 * We can keep it global, as long as it is used as "N pages at a time".
1519 * 128 should be plenty, currently we probably can get away with as few as 1.
1521 #define DRBD_MIN_POOL_PAGES 128
1522 extern mempool_t *drbd_md_io_page_pool;
1524 /* We also need to make sure we get a bio
1525 * when we need it for housekeeping purposes */
1526 extern struct bio_set *drbd_md_io_bio_set;
1527 /* to allocate from that set */
1528 extern struct bio *bio_alloc_drbd(gfp_t gfp_mask);
1530 extern rwlock_t global_state_lock;
1532 extern struct drbd_conf *drbd_new_device(unsigned int minor);
1533 extern void drbd_free_mdev(struct drbd_conf *mdev);
1535 extern int proc_details;
1538 extern void drbd_make_request(struct request_queue *q, struct bio *bio);
1539 extern int drbd_read_remote(struct drbd_conf *mdev, struct drbd_request *req);
1540 extern int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct bio_vec *bvec);
1541 extern int is_valid_ar_handle(struct drbd_request *, sector_t);
1545 extern void drbd_suspend_io(struct drbd_conf *mdev);
1546 extern void drbd_resume_io(struct drbd_conf *mdev);
1547 extern char *ppsize(char *buf, unsigned long long size);
1548 extern sector_t drbd_new_dev_size(struct drbd_conf *, struct drbd_backing_dev *, int);
1549 enum determine_dev_size { dev_size_error = -1, unchanged = 0, shrunk = 1, grew = 2 };
1550 extern enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *, enum dds_flags) __must_hold(local);
1551 extern void resync_after_online_grow(struct drbd_conf *);
1552 extern void drbd_reconsider_max_bio_size(struct drbd_conf *mdev);
1553 extern enum drbd_state_rv drbd_set_role(struct drbd_conf *mdev,
1554 enum drbd_role new_role,
1556 extern enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev);
1557 extern void drbd_try_outdate_peer_async(struct drbd_conf *mdev);
1558 extern int drbd_khelper(struct drbd_conf *mdev, char *cmd);
1561 extern int drbd_worker(struct drbd_thread *thi);
1562 extern int drbd_alter_sa(struct drbd_conf *mdev, int na);
1563 extern void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side);
1564 extern void resume_next_sg(struct drbd_conf *mdev);
1565 extern void suspend_other_sg(struct drbd_conf *mdev);
1566 extern int drbd_resync_finished(struct drbd_conf *mdev);
1567 /* maybe rather drbd_main.c ? */
1568 extern void *drbd_md_get_buffer(struct drbd_conf *mdev);
1569 extern void drbd_md_put_buffer(struct drbd_conf *mdev);
1570 extern int drbd_md_sync_page_io(struct drbd_conf *mdev,
1571 struct drbd_backing_dev *bdev, sector_t sector, int rw);
1572 extern void wait_until_done_or_disk_failure(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
1573 unsigned int *done);
1574 extern void drbd_ov_oos_found(struct drbd_conf*, sector_t, int);
1575 extern void drbd_rs_controller_reset(struct drbd_conf *mdev);
1577 static inline void ov_oos_print(struct drbd_conf *mdev)
1579 if (mdev->ov_last_oos_size) {
1580 dev_err(DEV, "Out of sync: start=%llu, size=%lu (sectors)\n",
1581 (unsigned long long)mdev->ov_last_oos_start,
1582 (unsigned long)mdev->ov_last_oos_size);
1584 mdev->ov_last_oos_size=0;
1588 extern void drbd_csum_bio(struct drbd_conf *, struct crypto_hash *, struct bio *, void *);
1589 extern void drbd_csum_ee(struct drbd_conf *, struct crypto_hash *, struct drbd_epoch_entry *, void *);
1590 /* worker callbacks */
1591 extern int w_req_cancel_conflict(struct drbd_conf *, struct drbd_work *, int);
1592 extern int w_read_retry_remote(struct drbd_conf *, struct drbd_work *, int);
1593 extern int w_e_end_data_req(struct drbd_conf *, struct drbd_work *, int);
1594 extern int w_e_end_rsdata_req(struct drbd_conf *, struct drbd_work *, int);
1595 extern int w_e_end_csum_rs_req(struct drbd_conf *, struct drbd_work *, int);
1596 extern int w_e_end_ov_reply(struct drbd_conf *, struct drbd_work *, int);
1597 extern int w_e_end_ov_req(struct drbd_conf *, struct drbd_work *, int);
1598 extern int w_ov_finished(struct drbd_conf *, struct drbd_work *, int);
1599 extern int w_resync_timer(struct drbd_conf *, struct drbd_work *, int);
1600 extern int w_resume_next_sg(struct drbd_conf *, struct drbd_work *, int);
1601 extern int w_send_write_hint(struct drbd_conf *, struct drbd_work *, int);
1602 extern int w_send_dblock(struct drbd_conf *, struct drbd_work *, int);
1603 extern int w_send_barrier(struct drbd_conf *, struct drbd_work *, int);
1604 extern int w_send_read_req(struct drbd_conf *, struct drbd_work *, int);
1605 extern int w_prev_work_done(struct drbd_conf *, struct drbd_work *, int);
1606 extern int w_e_reissue(struct drbd_conf *, struct drbd_work *, int);
1607 extern int w_restart_disk_io(struct drbd_conf *, struct drbd_work *, int);
1608 extern int w_send_oos(struct drbd_conf *, struct drbd_work *, int);
1609 extern int w_start_resync(struct drbd_conf *, struct drbd_work *, int);
1611 extern void resync_timer_fn(unsigned long data);
1612 extern void start_resync_timer_fn(unsigned long data);
1614 /* drbd_receiver.c */
1615 extern int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector);
1616 extern int drbd_submit_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e,
1617 const unsigned rw, const int fault_type);
1618 extern int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list);
1619 extern struct drbd_epoch_entry *drbd_alloc_ee(struct drbd_conf *mdev,
1622 unsigned int data_size,
1623 gfp_t gfp_mask) __must_hold(local);
1624 extern void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e,
1626 #define drbd_free_ee(m,e) drbd_free_some_ee(m, e, 0)
1627 #define drbd_free_net_ee(m,e) drbd_free_some_ee(m, e, 1)
1628 extern void drbd_wait_ee_list_empty(struct drbd_conf *mdev,
1629 struct list_head *head);
1630 extern void _drbd_wait_ee_list_empty(struct drbd_conf *mdev,
1631 struct list_head *head);
1632 extern void drbd_set_recv_tcq(struct drbd_conf *mdev, int tcq_enabled);
1633 extern void _drbd_clear_done_ee(struct drbd_conf *mdev, struct list_head *to_be_freed);
1634 extern void drbd_flush_workqueue(struct drbd_conf *mdev);
1635 extern void drbd_free_tl_hash(struct drbd_conf *mdev);
1637 /* yes, there is kernel_setsockopt, but only since 2.6.18. we don't need to
1638 * mess with get_fs/set_fs, we know we are KERNEL_DS always. */
1639 static inline int drbd_setsockopt(struct socket *sock, int level, int optname,
1640 char __user *optval, int optlen)
1643 if (level == SOL_SOCKET)
1644 err = sock_setsockopt(sock, level, optname, optval, optlen);
1646 err = sock->ops->setsockopt(sock, level, optname, optval,
1651 static inline void drbd_tcp_cork(struct socket *sock)
1654 (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK,
1655 (char __user *)&val, sizeof(val));
1658 static inline void drbd_tcp_uncork(struct socket *sock)
1661 (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK,
1662 (char __user *)&val, sizeof(val));
1665 static inline void drbd_tcp_nodelay(struct socket *sock)
1668 (void) drbd_setsockopt(sock, SOL_TCP, TCP_NODELAY,
1669 (char __user *)&val, sizeof(val));
1672 static inline void drbd_tcp_quickack(struct socket *sock)
1675 (void) drbd_setsockopt(sock, SOL_TCP, TCP_QUICKACK,
1676 (char __user *)&val, sizeof(val));
1679 void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo);
1682 extern struct proc_dir_entry *drbd_proc;
1683 extern const struct file_operations drbd_proc_fops;
1684 extern const char *drbd_conn_str(enum drbd_conns s);
1685 extern const char *drbd_role_str(enum drbd_role s);
1688 extern void drbd_al_begin_io(struct drbd_conf *mdev, sector_t sector);
1689 extern void drbd_al_complete_io(struct drbd_conf *mdev, sector_t sector);
1690 extern void drbd_rs_complete_io(struct drbd_conf *mdev, sector_t sector);
1691 extern int drbd_rs_begin_io(struct drbd_conf *mdev, sector_t sector);
1692 extern int drbd_try_rs_begin_io(struct drbd_conf *mdev, sector_t sector);
1693 extern void drbd_rs_cancel_all(struct drbd_conf *mdev);
1694 extern int drbd_rs_del_all(struct drbd_conf *mdev);
1695 extern void drbd_rs_failed_io(struct drbd_conf *mdev,
1696 sector_t sector, int size);
1697 extern int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *);
1698 extern void drbd_advance_rs_marks(struct drbd_conf *mdev, unsigned long still_to_go);
1699 extern void __drbd_set_in_sync(struct drbd_conf *mdev, sector_t sector,
1700 int size, const char *file, const unsigned int line);
1701 #define drbd_set_in_sync(mdev, sector, size) \
1702 __drbd_set_in_sync(mdev, sector, size, __FILE__, __LINE__)
1703 extern int __drbd_set_out_of_sync(struct drbd_conf *mdev, sector_t sector,
1704 int size, const char *file, const unsigned int line);
1705 #define drbd_set_out_of_sync(mdev, sector, size) \
1706 __drbd_set_out_of_sync(mdev, sector, size, __FILE__, __LINE__)
1707 extern void drbd_al_apply_to_bm(struct drbd_conf *mdev);
1708 extern void drbd_al_shrink(struct drbd_conf *mdev);
1713 void drbd_nl_cleanup(void);
1714 int __init drbd_nl_init(void);
1715 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state);
1716 void drbd_bcast_sync_progress(struct drbd_conf *mdev);
1717 void drbd_bcast_ee(struct drbd_conf *mdev,
1718 const char *reason, const int dgs,
1719 const char* seen_hash, const char* calc_hash,
1720 const struct drbd_epoch_entry* e);
1724 * DOC: DRBD State macros
1726 * These macros are used to express state changes in easily readable form.
1728 * The NS macros expand to a mask and a value, that can be bit ored onto the
1729 * current state as soon as the spinlock (req_lock) was taken.
1731 * The _NS macros are used for state functions that get called with the
1732 * spinlock. These macros expand directly to the new state value.
1734 * Besides the basic forms NS() and _NS() additional _?NS[23] are defined
1735 * to express state changes that affect more than one aspect of the state.
1737 * E.g. NS2(conn, C_CONNECTED, peer, R_SECONDARY)
1738 * Means that the network connection was established and that the peer
1739 * is in secondary role.
1741 #define role_MASK R_MASK
1742 #define peer_MASK R_MASK
1743 #define disk_MASK D_MASK
1744 #define pdsk_MASK D_MASK
1745 #define conn_MASK C_MASK
1747 #define user_isp_MASK 1
1748 #define aftr_isp_MASK 1
1749 #define susp_nod_MASK 1
1750 #define susp_fen_MASK 1
1753 ({ union drbd_state mask; mask.i = 0; mask.T = T##_MASK; mask; }), \
1754 ({ union drbd_state val; val.i = 0; val.T = (S); val; })
1755 #define NS2(T1, S1, T2, S2) \
1756 ({ union drbd_state mask; mask.i = 0; mask.T1 = T1##_MASK; \
1757 mask.T2 = T2##_MASK; mask; }), \
1758 ({ union drbd_state val; val.i = 0; val.T1 = (S1); \
1759 val.T2 = (S2); val; })
1760 #define NS3(T1, S1, T2, S2, T3, S3) \
1761 ({ union drbd_state mask; mask.i = 0; mask.T1 = T1##_MASK; \
1762 mask.T2 = T2##_MASK; mask.T3 = T3##_MASK; mask; }), \
1763 ({ union drbd_state val; val.i = 0; val.T1 = (S1); \
1764 val.T2 = (S2); val.T3 = (S3); val; })
1766 #define _NS(D, T, S) \
1767 D, ({ union drbd_state __ns; __ns.i = D->state.i; __ns.T = (S); __ns; })
1768 #define _NS2(D, T1, S1, T2, S2) \
1769 D, ({ union drbd_state __ns; __ns.i = D->state.i; __ns.T1 = (S1); \
1770 __ns.T2 = (S2); __ns; })
1771 #define _NS3(D, T1, S1, T2, S2, T3, S3) \
1772 D, ({ union drbd_state __ns; __ns.i = D->state.i; __ns.T1 = (S1); \
1773 __ns.T2 = (S2); __ns.T3 = (S3); __ns; })
1776 * inline helper functions
1777 *************************/
1779 /* see also page_chain_add and friends in drbd_receiver.c */
1780 static inline struct page *page_chain_next(struct page *page)
1782 return (struct page *)page_private(page);
1784 #define page_chain_for_each(page) \
1785 for (; page && ({ prefetch(page_chain_next(page)); 1; }); \
1786 page = page_chain_next(page))
1787 #define page_chain_for_each_safe(page, n) \
1788 for (; page && ({ n = page_chain_next(page); 1; }); page = n)
1790 static inline int drbd_ee_has_active_page(struct drbd_epoch_entry *e)
1792 struct page *page = e->pages;
1793 page_chain_for_each(page) {
1794 if (page_count(page) > 1)
1800 static inline void drbd_state_lock(struct drbd_conf *mdev)
1802 wait_event(mdev->misc_wait,
1803 !test_and_set_bit(CLUSTER_ST_CHANGE, &mdev->flags));
1806 static inline void drbd_state_unlock(struct drbd_conf *mdev)
1808 clear_bit(CLUSTER_ST_CHANGE, &mdev->flags);
1809 wake_up(&mdev->misc_wait);
1812 static inline enum drbd_state_rv
1813 _drbd_set_state(struct drbd_conf *mdev, union drbd_state ns,
1814 enum chg_state_flags flags, struct completion *done)
1816 enum drbd_state_rv rv;
1818 read_lock(&global_state_lock);
1819 rv = __drbd_set_state(mdev, ns, flags, done);
1820 read_unlock(&global_state_lock);
1826 * drbd_request_state() - Reqest a state change
1827 * @mdev: DRBD device.
1828 * @mask: mask of state bits to change.
1829 * @val: value of new state bits.
1831 * This is the most graceful way of requesting a state change. It is verbose
1832 * quite verbose in case the state change is not possible, and all those
1833 * state changes are globally serialized.
1835 static inline int drbd_request_state(struct drbd_conf *mdev,
1836 union drbd_state mask,
1837 union drbd_state val)
1839 return _drbd_request_state(mdev, mask, val, CS_VERBOSE + CS_ORDERED);
1842 enum drbd_force_detach_flags {
1848 #define __drbd_chk_io_error(m,f) __drbd_chk_io_error_(m,f, __func__)
1849 static inline void __drbd_chk_io_error_(struct drbd_conf *mdev,
1850 enum drbd_force_detach_flags forcedetach,
1853 switch (mdev->ldev->dc.on_io_error) {
1855 if (forcedetach == DRBD_IO_ERROR) {
1856 if (__ratelimit(&drbd_ratelimit_state))
1857 dev_err(DEV, "Local IO failed in %s.\n", where);
1858 if (mdev->state.disk > D_INCONSISTENT)
1859 _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_HARD, NULL);
1862 /* NOTE fall through to detach case if forcedetach set */
1864 case EP_CALL_HELPER:
1865 set_bit(WAS_IO_ERROR, &mdev->flags);
1866 if (forcedetach == DRBD_FORCE_DETACH)
1867 set_bit(FORCE_DETACH, &mdev->flags);
1868 if (mdev->state.disk > D_FAILED) {
1869 _drbd_set_state(_NS(mdev, disk, D_FAILED), CS_HARD, NULL);
1871 "Local IO failed in %s. Detaching...\n", where);
1878 * drbd_chk_io_error: Handle the on_io_error setting, should be called from all io completion handlers
1879 * @mdev: DRBD device.
1880 * @error: Error code passed to the IO completion callback
1881 * @forcedetach: Force detach. I.e. the error happened while accessing the meta data
1883 * See also drbd_main.c:after_state_ch() if (os.disk > D_FAILED && ns.disk == D_FAILED)
1885 #define drbd_chk_io_error(m,e,f) drbd_chk_io_error_(m,e,f, __func__)
1886 static inline void drbd_chk_io_error_(struct drbd_conf *mdev,
1887 int error, enum drbd_force_detach_flags forcedetach, const char *where)
1890 unsigned long flags;
1891 spin_lock_irqsave(&mdev->req_lock, flags);
1892 __drbd_chk_io_error_(mdev, forcedetach, where);
1893 spin_unlock_irqrestore(&mdev->req_lock, flags);
1899 * drbd_md_first_sector() - Returns the first sector number of the meta data area
1900 * @bdev: Meta data block device.
1902 * BTW, for internal meta data, this happens to be the maximum capacity
1903 * we could agree upon with our peer node.
1905 static inline sector_t drbd_md_first_sector(struct drbd_backing_dev *bdev)
1907 switch (bdev->dc.meta_dev_idx) {
1908 case DRBD_MD_INDEX_INTERNAL:
1909 case DRBD_MD_INDEX_FLEX_INT:
1910 return bdev->md.md_offset + bdev->md.bm_offset;
1911 case DRBD_MD_INDEX_FLEX_EXT:
1913 return bdev->md.md_offset;
1918 * drbd_md_last_sector() - Return the last sector number of the meta data area
1919 * @bdev: Meta data block device.
1921 static inline sector_t drbd_md_last_sector(struct drbd_backing_dev *bdev)
1923 switch (bdev->dc.meta_dev_idx) {
1924 case DRBD_MD_INDEX_INTERNAL:
1925 case DRBD_MD_INDEX_FLEX_INT:
1926 return bdev->md.md_offset + MD_AL_OFFSET - 1;
1927 case DRBD_MD_INDEX_FLEX_EXT:
1929 return bdev->md.md_offset + bdev->md.md_size_sect;
1933 /* Returns the number of 512 byte sectors of the device */
1934 static inline sector_t drbd_get_capacity(struct block_device *bdev)
1936 /* return bdev ? get_capacity(bdev->bd_disk) : 0; */
1937 return bdev ? i_size_read(bdev->bd_inode) >> 9 : 0;
1941 * drbd_get_max_capacity() - Returns the capacity we announce to out peer
1942 * @bdev: Meta data block device.
1944 * returns the capacity we announce to out peer. we clip ourselves at the
1945 * various MAX_SECTORS, because if we don't, current implementation will
1946 * oops sooner or later
1948 static inline sector_t drbd_get_max_capacity(struct drbd_backing_dev *bdev)
1951 switch (bdev->dc.meta_dev_idx) {
1952 case DRBD_MD_INDEX_INTERNAL:
1953 case DRBD_MD_INDEX_FLEX_INT:
1954 s = drbd_get_capacity(bdev->backing_bdev)
1955 ? min_t(sector_t, DRBD_MAX_SECTORS_FLEX,
1956 drbd_md_first_sector(bdev))
1959 case DRBD_MD_INDEX_FLEX_EXT:
1960 s = min_t(sector_t, DRBD_MAX_SECTORS_FLEX,
1961 drbd_get_capacity(bdev->backing_bdev));
1962 /* clip at maximum size the meta device can support */
1963 s = min_t(sector_t, s,
1964 BM_EXT_TO_SECT(bdev->md.md_size_sect
1965 - bdev->md.bm_offset));
1968 s = min_t(sector_t, DRBD_MAX_SECTORS,
1969 drbd_get_capacity(bdev->backing_bdev));
1975 * drbd_md_ss__() - Return the sector number of our meta data super block
1976 * @mdev: DRBD device.
1977 * @bdev: Meta data block device.
1979 static inline sector_t drbd_md_ss__(struct drbd_conf *mdev,
1980 struct drbd_backing_dev *bdev)
1982 switch (bdev->dc.meta_dev_idx) {
1983 default: /* external, some index */
1984 return MD_RESERVED_SECT * bdev->dc.meta_dev_idx;
1985 case DRBD_MD_INDEX_INTERNAL:
1986 /* with drbd08, internal meta data is always "flexible" */
1987 case DRBD_MD_INDEX_FLEX_INT:
1988 /* sizeof(struct md_on_disk_07) == 4k
1989 * position: last 4k aligned block of 4k size */
1990 if (!bdev->backing_bdev) {
1991 if (__ratelimit(&drbd_ratelimit_state)) {
1992 dev_err(DEV, "bdev->backing_bdev==NULL\n");
1997 return (drbd_get_capacity(bdev->backing_bdev) & ~7ULL)
1999 case DRBD_MD_INDEX_FLEX_EXT:
2005 drbd_queue_work_front(struct drbd_work_queue *q, struct drbd_work *w)
2007 unsigned long flags;
2008 spin_lock_irqsave(&q->q_lock, flags);
2009 list_add(&w->list, &q->q);
2010 up(&q->s); /* within the spinlock,
2011 see comment near end of drbd_worker() */
2012 spin_unlock_irqrestore(&q->q_lock, flags);
2016 drbd_queue_work(struct drbd_work_queue *q, struct drbd_work *w)
2018 unsigned long flags;
2019 spin_lock_irqsave(&q->q_lock, flags);
2020 list_add_tail(&w->list, &q->q);
2021 up(&q->s); /* within the spinlock,
2022 see comment near end of drbd_worker() */
2023 spin_unlock_irqrestore(&q->q_lock, flags);
2026 static inline void wake_asender(struct drbd_conf *mdev)
2028 if (test_bit(SIGNAL_ASENDER, &mdev->flags))
2029 force_sig(DRBD_SIG, mdev->asender.task);
2032 static inline void request_ping(struct drbd_conf *mdev)
2034 set_bit(SEND_PING, &mdev->flags);
2038 static inline int drbd_send_short_cmd(struct drbd_conf *mdev,
2039 enum drbd_packets cmd)
2041 struct p_header80 h;
2042 return drbd_send_cmd(mdev, USE_DATA_SOCKET, cmd, &h, sizeof(h));
2045 static inline int drbd_send_ping(struct drbd_conf *mdev)
2047 struct p_header80 h;
2048 return drbd_send_cmd(mdev, USE_META_SOCKET, P_PING, &h, sizeof(h));
2051 static inline int drbd_send_ping_ack(struct drbd_conf *mdev)
2053 struct p_header80 h;
2054 return drbd_send_cmd(mdev, USE_META_SOCKET, P_PING_ACK, &h, sizeof(h));
2057 static inline void drbd_thread_stop(struct drbd_thread *thi)
2059 _drbd_thread_stop(thi, false, true);
2062 static inline void drbd_thread_stop_nowait(struct drbd_thread *thi)
2064 _drbd_thread_stop(thi, false, false);
2067 static inline void drbd_thread_restart_nowait(struct drbd_thread *thi)
2069 _drbd_thread_stop(thi, true, false);
2072 /* counts how many answer packets packets we expect from our peer,
2073 * for either explicit application requests,
2074 * or implicit barrier packets as necessary.
2077 * _req_mod(req, queue_for_net_write or queue_for_net_read);
2078 * it is much easier and equally valid to count what we queue for the
2079 * worker, even before it actually was queued or send.
2080 * (drbd_make_request_common; recovery path on read io-error)
2082 * got_BarrierAck (respective tl_clear, tl_clear_barrier)
2083 * _req_mod(req, data_received)
2084 * [from receive_DataReply]
2085 * _req_mod(req, write_acked_by_peer or recv_acked_by_peer or neg_acked)
2086 * [from got_BlockAck (P_WRITE_ACK, P_RECV_ACK)]
2087 * for some reason it is NOT decreased in got_NegAck,
2088 * but in the resulting cleanup code from report_params.
2089 * we should try to remember the reason for that...
2090 * _req_mod(req, send_failed or send_canceled)
2091 * _req_mod(req, connection_lost_while_pending)
2092 * [from tl_clear_barrier]
2094 static inline void inc_ap_pending(struct drbd_conf *mdev)
2096 atomic_inc(&mdev->ap_pending_cnt);
2099 #define ERR_IF_CNT_IS_NEGATIVE(which) \
2100 if (atomic_read(&mdev->which) < 0) \
2101 dev_err(DEV, "in %s:%d: " #which " = %d < 0 !\n", \
2102 __func__ , __LINE__ , \
2103 atomic_read(&mdev->which))
2105 #define dec_ap_pending(mdev) do { \
2106 typecheck(struct drbd_conf *, mdev); \
2107 if (atomic_dec_and_test(&mdev->ap_pending_cnt)) \
2108 wake_up(&mdev->misc_wait); \
2109 ERR_IF_CNT_IS_NEGATIVE(ap_pending_cnt); } while (0)
2111 /* counts how many resync-related answers we still expect from the peer
2113 * C_SYNC_TARGET sends P_RS_DATA_REQUEST (and expects P_RS_DATA_REPLY)
2114 * C_SYNC_SOURCE sends P_RS_DATA_REPLY (and expects P_WRITE_ACK with ID_SYNCER)
2115 * (or P_NEG_ACK with ID_SYNCER)
2117 static inline void inc_rs_pending(struct drbd_conf *mdev)
2119 atomic_inc(&mdev->rs_pending_cnt);
2122 #define dec_rs_pending(mdev) do { \
2123 typecheck(struct drbd_conf *, mdev); \
2124 atomic_dec(&mdev->rs_pending_cnt); \
2125 ERR_IF_CNT_IS_NEGATIVE(rs_pending_cnt); } while (0)
2127 /* counts how many answers we still need to send to the peer.
2129 * receive_Data unless protocol A;
2130 * we need to send a P_RECV_ACK (proto B)
2131 * or P_WRITE_ACK (proto C)
2132 * receive_RSDataReply (recv_resync_read) we need to send a P_WRITE_ACK
2133 * receive_DataRequest (receive_RSDataRequest) we need to send back P_DATA
2134 * receive_Barrier_* we need to send a P_BARRIER_ACK
2136 static inline void inc_unacked(struct drbd_conf *mdev)
2138 atomic_inc(&mdev->unacked_cnt);
2141 #define dec_unacked(mdev) do { \
2142 typecheck(struct drbd_conf *, mdev); \
2143 atomic_dec(&mdev->unacked_cnt); \
2144 ERR_IF_CNT_IS_NEGATIVE(unacked_cnt); } while (0)
2146 #define sub_unacked(mdev, n) do { \
2147 typecheck(struct drbd_conf *, mdev); \
2148 atomic_sub(n, &mdev->unacked_cnt); \
2149 ERR_IF_CNT_IS_NEGATIVE(unacked_cnt); } while (0)
2152 static inline void put_net_conf(struct drbd_conf *mdev)
2154 if (atomic_dec_and_test(&mdev->net_cnt))
2155 wake_up(&mdev->net_cnt_wait);
2159 * get_net_conf() - Increase ref count on mdev->net_conf; Returns 0 if nothing there
2160 * @mdev: DRBD device.
2162 * You have to call put_net_conf() when finished working with mdev->net_conf.
2164 static inline int get_net_conf(struct drbd_conf *mdev)
2168 atomic_inc(&mdev->net_cnt);
2169 have_net_conf = mdev->state.conn >= C_UNCONNECTED;
2172 return have_net_conf;
2176 * get_ldev() - Increase the ref count on mdev->ldev. Returns 0 if there is no ldev
2179 * You have to call put_ldev() when finished working with mdev->ldev.
2181 #define get_ldev(M) __cond_lock(local, _get_ldev_if_state(M,D_INCONSISTENT))
2182 #define get_ldev_if_state(M,MINS) __cond_lock(local, _get_ldev_if_state(M,MINS))
2184 static inline void put_ldev(struct drbd_conf *mdev)
2186 int i = atomic_dec_return(&mdev->local_cnt);
2188 /* This may be called from some endio handler,
2189 * so we must not sleep here. */
2194 if (mdev->state.disk == D_DISKLESS)
2195 /* even internal references gone, safe to destroy */
2196 drbd_ldev_destroy(mdev);
2197 if (mdev->state.disk == D_FAILED)
2198 /* all application IO references gone. */
2199 drbd_go_diskless(mdev);
2200 wake_up(&mdev->misc_wait);
2205 static inline int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins)
2209 /* never get a reference while D_DISKLESS */
2210 if (mdev->state.disk == D_DISKLESS)
2213 atomic_inc(&mdev->local_cnt);
2214 io_allowed = (mdev->state.disk >= mins);
2220 extern int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins);
2223 /* you must have an "get_ldev" reference */
2224 static inline void drbd_get_syncer_progress(struct drbd_conf *mdev,
2225 unsigned long *bits_left, unsigned int *per_mil_done)
2227 /* this is to break it at compile time when we change that, in case we
2228 * want to support more than (1<<32) bits on a 32bit arch. */
2229 typecheck(unsigned long, mdev->rs_total);
2231 /* note: both rs_total and rs_left are in bits, i.e. in
2232 * units of BM_BLOCK_SIZE.
2233 * for the percentage, we don't care. */
2235 if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T)
2236 *bits_left = mdev->ov_left;
2238 *bits_left = drbd_bm_total_weight(mdev) - mdev->rs_failed;
2239 /* >> 10 to prevent overflow,
2240 * +1 to prevent division by zero */
2241 if (*bits_left > mdev->rs_total) {
2242 /* doh. maybe a logic bug somewhere.
2243 * may also be just a race condition
2244 * between this and a disconnect during sync.
2245 * for now, just prevent in-kernel buffer overflow.
2248 dev_warn(DEV, "cs:%s rs_left=%lu > rs_total=%lu (rs_failed %lu)\n",
2249 drbd_conn_str(mdev->state.conn),
2250 *bits_left, mdev->rs_total, mdev->rs_failed);
2253 /* Make sure the division happens in long context.
2254 * We allow up to one petabyte storage right now,
2255 * at a granularity of 4k per bit that is 2**38 bits.
2256 * After shift right and multiplication by 1000,
2257 * this should still fit easily into a 32bit long,
2258 * so we don't need a 64bit division on 32bit arch.
2259 * Note: currently we don't support such large bitmaps on 32bit
2260 * arch anyways, but no harm done to be prepared for it here.
2262 unsigned int shift = mdev->rs_total > UINT_MAX ? 16 : 10;
2263 unsigned long left = *bits_left >> shift;
2264 unsigned long total = 1UL + (mdev->rs_total >> shift);
2265 unsigned long tmp = 1000UL - left * 1000UL/total;
2266 *per_mil_done = tmp;
2271 /* this throttles on-the-fly application requests
2272 * according to max_buffers settings;
2273 * maybe re-implement using semaphores? */
2274 static inline int drbd_get_max_buffers(struct drbd_conf *mdev)
2276 int mxb = 1000000; /* arbitrary limit on open requests */
2277 if (get_net_conf(mdev)) {
2278 mxb = mdev->net_conf->max_buffers;
2284 static inline int drbd_state_is_stable(struct drbd_conf *mdev)
2286 union drbd_state s = mdev->state;
2288 /* DO NOT add a default clause, we want the compiler to warn us
2289 * for any newly introduced state we may have forgotten to add here */
2291 switch ((enum drbd_conns)s.conn) {
2292 /* new io only accepted when there is no connection, ... */
2294 case C_WF_CONNECTION:
2295 /* ... or there is a well established connection. */
2301 case C_PAUSED_SYNC_S:
2302 case C_PAUSED_SYNC_T:
2305 /* transitional states, IO allowed */
2306 case C_DISCONNECTING:
2310 case C_NETWORK_FAILURE:
2311 case C_PROTOCOL_ERROR:
2313 case C_WF_REPORT_PARAMS:
2314 case C_STARTING_SYNC_S:
2315 case C_STARTING_SYNC_T:
2318 /* Allow IO in BM exchange states with new protocols */
2320 if (mdev->agreed_pro_version < 96)
2324 /* no new io accepted in these states */
2326 case C_WF_SYNC_UUID:
2332 switch ((enum drbd_disk_state)s.disk) {
2334 case D_INCONSISTENT:
2339 /* disk state is stable as well. */
2342 /* no new io accepted during tansitional states */
2354 static inline int is_susp(union drbd_state s)
2356 return s.susp || s.susp_nod || s.susp_fen;
2359 static inline bool may_inc_ap_bio(struct drbd_conf *mdev)
2361 int mxb = drbd_get_max_buffers(mdev);
2363 if (is_susp(mdev->state))
2365 if (test_bit(SUSPEND_IO, &mdev->flags))
2368 /* to avoid potential deadlock or bitmap corruption,
2369 * in various places, we only allow new application io
2370 * to start during "stable" states. */
2372 /* no new io accepted when attaching or detaching the disk */
2373 if (!drbd_state_is_stable(mdev))
2376 /* since some older kernels don't have atomic_add_unless,
2377 * and we are within the spinlock anyways, we have this workaround. */
2378 if (atomic_read(&mdev->ap_bio_cnt) > mxb)
2380 if (test_bit(BITMAP_IO, &mdev->flags))
2385 static inline bool inc_ap_bio_cond(struct drbd_conf *mdev, int count)
2389 spin_lock_irq(&mdev->req_lock);
2390 rv = may_inc_ap_bio(mdev);
2392 atomic_add(count, &mdev->ap_bio_cnt);
2393 spin_unlock_irq(&mdev->req_lock);
2398 static inline void inc_ap_bio(struct drbd_conf *mdev, int count)
2401 * as long as the device is suspended
2402 * until the bitmap is no longer on the fly during connection
2403 * handshake as long as we would exeed the max_buffer limit.
2405 * to avoid races with the reconnect code,
2406 * we need to atomic_inc within the spinlock. */
2408 wait_event(mdev->misc_wait, inc_ap_bio_cond(mdev, count));
2411 static inline void dec_ap_bio(struct drbd_conf *mdev)
2413 int mxb = drbd_get_max_buffers(mdev);
2414 int ap_bio = atomic_dec_return(&mdev->ap_bio_cnt);
2416 D_ASSERT(ap_bio >= 0);
2417 /* this currently does wake_up for every dec_ap_bio!
2418 * maybe rather introduce some type of hysteresis?
2419 * e.g. (ap_bio == mxb/2 || ap_bio == 0) ? */
2421 wake_up(&mdev->misc_wait);
2422 if (ap_bio == 0 && test_bit(BITMAP_IO, &mdev->flags)) {
2423 if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags))
2424 drbd_queue_work(&mdev->data.work, &mdev->bm_io_work.w);
2428 static inline int drbd_set_ed_uuid(struct drbd_conf *mdev, u64 val)
2430 int changed = mdev->ed_uuid != val;
2431 mdev->ed_uuid = val;
2435 static inline int seq_cmp(u32 a, u32 b)
2437 /* we assume wrap around at 32bit.
2438 * for wrap around at 24bit (old atomic_t),
2442 return (s32)(a) - (s32)(b);
2444 #define seq_lt(a, b) (seq_cmp((a), (b)) < 0)
2445 #define seq_gt(a, b) (seq_cmp((a), (b)) > 0)
2446 #define seq_ge(a, b) (seq_cmp((a), (b)) >= 0)
2447 #define seq_le(a, b) (seq_cmp((a), (b)) <= 0)
2448 /* CAUTION: please no side effects in arguments! */
2449 #define seq_max(a, b) ((u32)(seq_gt((a), (b)) ? (a) : (b)))
2451 static inline void update_peer_seq(struct drbd_conf *mdev, unsigned int new_seq)
2454 spin_lock(&mdev->peer_seq_lock);
2455 m = seq_max(mdev->peer_seq, new_seq);
2457 spin_unlock(&mdev->peer_seq_lock);
2459 wake_up(&mdev->seq_wait);
2462 static inline void drbd_update_congested(struct drbd_conf *mdev)
2464 struct sock *sk = mdev->data.socket->sk;
2465 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
2466 set_bit(NET_CONGESTED, &mdev->flags);
2469 static inline int drbd_queue_order_type(struct drbd_conf *mdev)
2471 /* sorry, we currently have no working implementation
2472 * of distributed TCQ stuff */
2473 #ifndef QUEUE_ORDERED_NONE
2474 #define QUEUE_ORDERED_NONE 0
2476 return QUEUE_ORDERED_NONE;
2479 static inline void drbd_md_flush(struct drbd_conf *mdev)
2483 if (test_bit(MD_NO_FUA, &mdev->flags))
2486 r = blkdev_issue_flush(mdev->ldev->md_bdev, GFP_KERNEL, NULL);
2488 set_bit(MD_NO_FUA, &mdev->flags);
2489 dev_err(DEV, "meta data flush failed with status %d, disabling md-flushes\n", r);