1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 drbd_int.h
4
5 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
6
7 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
8 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
9 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
10
11
12*/
13
14#ifndef _DRBD_INT_H
15#define _DRBD_INT_H
16
17#include <crypto/hash.h>
18#include <linux/compiler.h>
19#include <linux/types.h>
20#include <linux/list.h>
21#include <linux/sched/signal.h>
22#include <linux/bitops.h>
23#include <linux/slab.h>
24#include <linux/ratelimit.h>
25#include <linux/tcp.h>
26#include <linux/mutex.h>
27#include <linux/major.h>
28#include <linux/blkdev.h>
29#include <linux/backing-dev.h>
30#include <linux/idr.h>
31#include <linux/dynamic_debug.h>
32#include <net/tcp.h>
33#include <linux/lru_cache.h>
34#include <linux/prefetch.h>
35#include <linux/drbd_genl_api.h>
36#include <linux/drbd.h>
37#include <linux/drbd_config.h>
38#include "drbd_strings.h"
39#include "drbd_state.h"
40#include "drbd_protocol.h"
41#include "drbd_polymorph_printk.h"
42
43/* shared module parameters, defined in drbd_main.c */
44#ifdef CONFIG_DRBD_FAULT_INJECTION
45extern int drbd_enable_faults;
46extern int drbd_fault_rate;
47#endif
48
49extern unsigned int drbd_minor_count;
50extern char drbd_usermode_helper[];
51extern int drbd_proc_details;
52
53
54/* This is used to stop/restart our threads.
55 * Cannot use SIGTERM nor SIGKILL, since these
56 * are sent out by init on runlevel changes
57 * I choose SIGHUP for now.
58 */
59#define DRBD_SIGKILL SIGHUP
60
61#define ID_IN_SYNC (4711ULL)
62#define ID_OUT_OF_SYNC (4712ULL)
63#define ID_SYNCER (-1ULL)
64
65#define UUID_NEW_BM_OFFSET ((u64)0x0001000000000000ULL)
66
67struct drbd_device;
68struct drbd_connection;
69struct drbd_peer_device;
70
71/* Defines to control fault insertion */
72enum {
73 DRBD_FAULT_MD_WR = 0, /* meta data write */
74 DRBD_FAULT_MD_RD = 1, /* read */
75 DRBD_FAULT_RS_WR = 2, /* resync */
76 DRBD_FAULT_RS_RD = 3,
77 DRBD_FAULT_DT_WR = 4, /* data */
78 DRBD_FAULT_DT_RD = 5,
79 DRBD_FAULT_DT_RA = 6, /* data read ahead */
80 DRBD_FAULT_BM_ALLOC = 7, /* bitmap allocation */
81 DRBD_FAULT_AL_EE = 8, /* alloc ee */
82 DRBD_FAULT_RECEIVE = 9, /* Changes some bytes upon receiving a [rs]data block */
83
84 DRBD_FAULT_MAX,
85};
86
87extern unsigned int
88_drbd_insert_fault(struct drbd_device *device, unsigned int type);
89
90static inline int
91drbd_insert_fault(struct drbd_device *device, unsigned int type) {
92#ifdef CONFIG_DRBD_FAULT_INJECTION
93 return drbd_fault_rate &&
94 (drbd_enable_faults & (1<<type)) &&
95 _drbd_insert_fault(device, type);
96#else
97 return 0;
98#endif
99}
100
101/* integer division, round _UP_ to the next integer */
102#define div_ceil(A, B) ((A)/(B) + ((A)%(B) ? 1 : 0))
103/* usual integer division */
104#define div_floor(A, B) ((A)/(B))
105
106extern struct ratelimit_state drbd_ratelimit_state;
107extern struct idr drbd_devices; /* RCU, updates: genl_lock() */
108extern struct list_head drbd_resources; /* RCU, updates: genl_lock() */
109
110extern const char *cmdname(enum drbd_packet cmd);
111
112/* for sending/receiving the bitmap,
113 * possibly in some encoding scheme */
114struct bm_xfer_ctx {
115 /* "const"
116 * stores total bits and long words
117 * of the bitmap, so we don't need to
118 * call the accessor functions over and again. */
119 unsigned long bm_bits;
120 unsigned long bm_words;
121 /* during xfer, current position within the bitmap */
122 unsigned long bit_offset;
123 unsigned long word_offset;
124
125 /* statistics; index: (h->command == P_BITMAP) */
126 unsigned packets[2];
127 unsigned bytes[2];
128};
129
130extern void INFO_bm_xfer_stats(struct drbd_peer_device *peer_device,
131 const char *direction, struct bm_xfer_ctx *c);
132
133static inline void bm_xfer_ctx_bit_to_word_offset(struct bm_xfer_ctx *c)
134{
135 /* word_offset counts "native long words" (32 or 64 bit),
136 * aligned at 64 bit.
137 * Encoded packet may end at an unaligned bit offset.
138 * In case a fallback clear text packet is transmitted in
139 * between, we adjust this offset back to the last 64bit
140 * aligned "native long word", which makes coding and decoding
141 * the plain text bitmap much more convenient. */
142#if BITS_PER_LONG == 64
143 c->word_offset = c->bit_offset >> 6;
144#elif BITS_PER_LONG == 32
145 c->word_offset = c->bit_offset >> 5;
146 c->word_offset &= ~(1UL);
147#else
148# error "unsupported BITS_PER_LONG"
149#endif
150}
151
152extern unsigned int drbd_header_size(struct drbd_connection *connection);
153
154/**********************************************************************/
155enum drbd_thread_state {
156 NONE,
157 RUNNING,
158 EXITING,
159 RESTARTING
160};
161
162struct drbd_thread {
163 spinlock_t t_lock;
164 struct task_struct *task;
165 struct completion stop;
166 enum drbd_thread_state t_state;
167 int (*function) (struct drbd_thread *);
168 struct drbd_resource *resource;
169 struct drbd_connection *connection;
170 int reset_cpu_mask;
171 const char *name;
172};
173
174static inline enum drbd_thread_state get_t_state(struct drbd_thread *thi)
175{
176 /* THINK testing the t_state seems to be uncritical in all cases
177 * (but thread_{start,stop}), so we can read it *without* the lock.
178 * --lge */
179
180 smp_rmb();
181 return thi->t_state;
182}
183
184struct drbd_work {
185 struct list_head list;
186 int (*cb)(struct drbd_work *, int cancel);
187};
188
189struct drbd_device_work {
190 struct drbd_work w;
191 struct drbd_device *device;
192};
193
194#include "drbd_interval.h"
195
196extern int drbd_wait_misc(struct drbd_device *, struct drbd_interval *);
197
198extern void lock_all_resources(void);
199extern void unlock_all_resources(void);
200
201struct drbd_request {
202 struct drbd_work w;
203 struct drbd_device *device;
204
205 /* if local IO is not allowed, will be NULL.
206 * if local IO _is_ allowed, holds the locally submitted bio clone,
207 * or, after local IO completion, the ERR_PTR(error).
208 * see drbd_request_endio(). */
209 struct bio *private_bio;
210
211 struct drbd_interval i;
212
213 /* epoch: used to check on "completion" whether this req was in
214 * the current epoch, and we therefore have to close it,
215 * causing a p_barrier packet to be send, starting a new epoch.
216 *
217 * This corresponds to "barrier" in struct p_barrier[_ack],
218 * and to "barrier_nr" in struct drbd_epoch (and various
219 * comments/function parameters/local variable names).
220 */
221 unsigned int epoch;
222
223 struct list_head tl_requests; /* ring list in the transfer log */
224 struct bio *master_bio; /* master bio pointer */
225
226 /* see struct drbd_device */
227 struct list_head req_pending_master_completion;
228 struct list_head req_pending_local;
229
230 /* for generic IO accounting */
231 unsigned long start_jif;
232
233 /* for DRBD internal statistics */
234
235 /* Minimal set of time stamps to determine if we wait for activity log
236 * transactions, local disk or peer. 32 bit "jiffies" are good enough,
237 * we don't expect a DRBD request to be stalled for several month.
238 */
239
240 /* before actual request processing */
241 unsigned long in_actlog_jif;
242
243 /* local disk */
244 unsigned long pre_submit_jif;
245
246 /* per connection */
247 unsigned long pre_send_jif;
248 unsigned long acked_jif;
249 unsigned long net_done_jif;
250
251 /* Possibly even more detail to track each phase:
252 * master_completion_jif
253 * how long did it take to complete the master bio
254 * (application visible latency)
255 * allocated_jif
256 * how long the master bio was blocked until we finally allocated
257 * a tracking struct
258 * in_actlog_jif
259 * how long did we wait for activity log transactions
260 *
261 * net_queued_jif
262 * when did we finally queue it for sending
263 * pre_send_jif
264 * when did we start sending it
265 * post_send_jif
266 * how long did we block in the network stack trying to send it
267 * acked_jif
268 * when did we receive (or fake, in protocol A) a remote ACK
269 * net_done_jif
270 * when did we receive final acknowledgement (P_BARRIER_ACK),
271 * or decide, e.g. on connection loss, that we do no longer expect
272 * anything from this peer for this request.
273 *
274 * pre_submit_jif
275 * post_sub_jif
276 * when did we start submiting to the lower level device,
277 * and how long did we block in that submit function
278 * local_completion_jif
279 * how long did it take the lower level device to complete this request
280 */
281
282
283 /* once it hits 0, we may complete the master_bio */
284 atomic_t completion_ref;
285 /* once it hits 0, we may destroy this drbd_request object */
286 struct kref kref;
287
288 unsigned rq_state; /* see comments above _req_mod() */
289};
290
291struct drbd_epoch {
292 struct drbd_connection *connection;
293 struct list_head list;
294 unsigned int barrier_nr;
295 atomic_t epoch_size; /* increased on every request added. */
296 atomic_t active; /* increased on every req. added, and dec on every finished. */
297 unsigned long flags;
298};
299
300/* Prototype declaration of function defined in drbd_receiver.c */
301int drbdd_init(struct drbd_thread *);
302int drbd_asender(struct drbd_thread *);
303
304/* drbd_epoch flag bits */
305enum {
306 DE_HAVE_BARRIER_NUMBER,
307};
308
309enum epoch_event {
310 EV_PUT,
311 EV_GOT_BARRIER_NR,
312 EV_BECAME_LAST,
313 EV_CLEANUP = 32, /* used as flag */
314};
315
316struct digest_info {
317 int digest_size;
318 void *digest;
319};
320
321struct drbd_peer_request {
322 struct drbd_work w;
323 struct drbd_peer_device *peer_device;
324 struct drbd_epoch *epoch; /* for writes */
325 struct page *pages;
326 blk_opf_t opf;
327 atomic_t pending_bios;
328 struct drbd_interval i;
329 /* see comments on ee flag bits below */
330 unsigned long flags;
331 unsigned long submit_jif;
332 union {
333 u64 block_id;
334 struct digest_info *digest;
335 };
336};
337
338/* Equivalent to bio_op and req_op. */
339#define peer_req_op(peer_req) \
340 ((peer_req)->opf & REQ_OP_MASK)
341
342/* ee flag bits.
343 * While corresponding bios are in flight, the only modification will be
344 * set_bit WAS_ERROR, which has to be atomic.
345 * If no bios are in flight yet, or all have been completed,
346 * non-atomic modification to ee->flags is ok.
347 */
348enum {
349 __EE_CALL_AL_COMPLETE_IO,
350 __EE_MAY_SET_IN_SYNC,
351
352 /* is this a TRIM aka REQ_OP_DISCARD? */
353 __EE_TRIM,
354 /* explicit zero-out requested, or
355 * our lower level cannot handle trim,
356 * and we want to fall back to zeroout instead */
357 __EE_ZEROOUT,
358
359 /* In case a barrier failed,
360 * we need to resubmit without the barrier flag. */
361 __EE_RESUBMITTED,
362
363 /* we may have several bios per peer request.
364 * if any of those fail, we set this flag atomically
365 * from the endio callback */
366 __EE_WAS_ERROR,
367
368 /* This ee has a pointer to a digest instead of a block id */
369 __EE_HAS_DIGEST,
370
371 /* Conflicting local requests need to be restarted after this request */
372 __EE_RESTART_REQUESTS,
373
374 /* The peer wants a write ACK for this (wire proto C) */
375 __EE_SEND_WRITE_ACK,
376
377 /* Is set when net_conf had two_primaries set while creating this peer_req */
378 __EE_IN_INTERVAL_TREE,
379
380 /* for debugfs: */
381 /* has this been submitted, or does it still wait for something else? */
382 __EE_SUBMITTED,
383
384 /* this is/was a write request */
385 __EE_WRITE,
386
387 /* this is/was a write same request */
388 __EE_WRITE_SAME,
389
390 /* this originates from application on peer
391 * (not some resync or verify or other DRBD internal request) */
392 __EE_APPLICATION,
393
394 /* If it contains only 0 bytes, send back P_RS_DEALLOCATED */
395 __EE_RS_THIN_REQ,
396};
397#define EE_CALL_AL_COMPLETE_IO (1<<__EE_CALL_AL_COMPLETE_IO)
398#define EE_MAY_SET_IN_SYNC (1<<__EE_MAY_SET_IN_SYNC)
399#define EE_TRIM (1<<__EE_TRIM)
400#define EE_ZEROOUT (1<<__EE_ZEROOUT)
401#define EE_RESUBMITTED (1<<__EE_RESUBMITTED)
402#define EE_WAS_ERROR (1<<__EE_WAS_ERROR)
403#define EE_HAS_DIGEST (1<<__EE_HAS_DIGEST)
404#define EE_RESTART_REQUESTS (1<<__EE_RESTART_REQUESTS)
405#define EE_SEND_WRITE_ACK (1<<__EE_SEND_WRITE_ACK)
406#define EE_IN_INTERVAL_TREE (1<<__EE_IN_INTERVAL_TREE)
407#define EE_SUBMITTED (1<<__EE_SUBMITTED)
408#define EE_WRITE (1<<__EE_WRITE)
409#define EE_WRITE_SAME (1<<__EE_WRITE_SAME)
410#define EE_APPLICATION (1<<__EE_APPLICATION)
411#define EE_RS_THIN_REQ (1<<__EE_RS_THIN_REQ)
412
413/* flag bits per device */
414enum {
415 UNPLUG_REMOTE, /* sending a "UnplugRemote" could help */
416 MD_DIRTY, /* current uuids and flags not yet on disk */
417 USE_DEGR_WFC_T, /* degr-wfc-timeout instead of wfc-timeout. */
418 CL_ST_CHG_SUCCESS,
419 CL_ST_CHG_FAIL,
420 CRASHED_PRIMARY, /* This node was a crashed primary.
421 * Gets cleared when the state.conn
422 * goes into C_CONNECTED state. */
423 CONSIDER_RESYNC,
424
425 MD_NO_FUA, /* Users wants us to not use FUA/FLUSH on meta data dev */
426
427 BITMAP_IO, /* suspend application io;
428 once no more io in flight, start bitmap io */
429 BITMAP_IO_QUEUED, /* Started bitmap IO */
430 WAS_IO_ERROR, /* Local disk failed, returned IO error */
431 WAS_READ_ERROR, /* Local disk READ failed (set additionally to the above) */
432 FORCE_DETACH, /* Force-detach from local disk, aborting any pending local IO */
433 RESYNC_AFTER_NEG, /* Resync after online grow after the attach&negotiate finished. */
434 RESIZE_PENDING, /* Size change detected locally, waiting for the response from
435 * the peer, if it changed there as well. */
436 NEW_CUR_UUID, /* Create new current UUID when thawing IO */
437 AL_SUSPENDED, /* Activity logging is currently suspended. */
438 AHEAD_TO_SYNC_SOURCE, /* Ahead -> SyncSource queued */
439 B_RS_H_DONE, /* Before resync handler done (already executed) */
440 DISCARD_MY_DATA, /* discard_my_data flag per volume */
441 READ_BALANCE_RR,
442
443 FLUSH_PENDING, /* if set, device->flush_jif is when we submitted that flush
444 * from drbd_flush_after_epoch() */
445
446 /* cleared only after backing device related structures have been destroyed. */
447 GOING_DISKLESS, /* Disk is being detached, because of io-error, or admin request. */
448
449 /* to be used in drbd_device_post_work() */
450 GO_DISKLESS, /* tell worker to schedule cleanup before detach */
451 DESTROY_DISK, /* tell worker to close backing devices and destroy related structures. */
452 MD_SYNC, /* tell worker to call drbd_md_sync() */
453 RS_START, /* tell worker to start resync/OV */
454 RS_PROGRESS, /* tell worker that resync made significant progress */
455 RS_DONE, /* tell worker that resync is done */
456};
457
458struct drbd_bitmap; /* opaque for drbd_device */
459
460/* definition of bits in bm_flags to be used in drbd_bm_lock
461 * and drbd_bitmap_io and friends. */
462enum bm_flag {
463 /* currently locked for bulk operation */
464 BM_LOCKED_MASK = 0xf,
465
466 /* in detail, that is: */
467 BM_DONT_CLEAR = 0x1,
468 BM_DONT_SET = 0x2,
469 BM_DONT_TEST = 0x4,
470
471 /* so we can mark it locked for bulk operation,
472 * and still allow all non-bulk operations */
473 BM_IS_LOCKED = 0x8,
474
475 /* (test bit, count bit) allowed (common case) */
476 BM_LOCKED_TEST_ALLOWED = BM_DONT_CLEAR | BM_DONT_SET | BM_IS_LOCKED,
477
478 /* testing bits, as well as setting new bits allowed, but clearing bits
479 * would be unexpected. Used during bitmap receive. Setting new bits
480 * requires sending of "out-of-sync" information, though. */
481 BM_LOCKED_SET_ALLOWED = BM_DONT_CLEAR | BM_IS_LOCKED,
482
483 /* for drbd_bm_write_copy_pages, everything is allowed,
484 * only concurrent bulk operations are locked out. */
485 BM_LOCKED_CHANGE_ALLOWED = BM_IS_LOCKED,
486};
487
488struct drbd_work_queue {
489 struct list_head q;
490 spinlock_t q_lock; /* to protect the list. */
491 wait_queue_head_t q_wait;
492};
493
494struct drbd_socket {
495 struct mutex mutex;
496 struct socket *socket;
497 /* this way we get our
498 * send/receive buffers off the stack */
499 void *sbuf;
500 void *rbuf;
501};
502
503struct drbd_md {
504 u64 md_offset; /* sector offset to 'super' block */
505
506 u64 la_size_sect; /* last agreed size, unit sectors */
507 spinlock_t uuid_lock;
508 u64 uuid[UI_SIZE];
509 u64 device_uuid;
510 u32 flags;
511 u32 md_size_sect;
512
513 s32 al_offset; /* signed relative sector offset to activity log */
514 s32 bm_offset; /* signed relative sector offset to bitmap */
515
516 /* cached value of bdev->disk_conf->meta_dev_idx (see below) */
517 s32 meta_dev_idx;
518
519 /* see al_tr_number_to_on_disk_sector() */
520 u32 al_stripes;
521 u32 al_stripe_size_4k;
522 u32 al_size_4k; /* cached product of the above */
523};
524
525struct drbd_backing_dev {
526 struct block_device *backing_bdev;
527 struct bdev_handle *backing_bdev_handle;
528 struct block_device *md_bdev;
529 struct bdev_handle *md_bdev_handle;
530 struct drbd_md md;
531 struct disk_conf *disk_conf; /* RCU, for updates: resource->conf_update */
532 sector_t known_size; /* last known size of that backing device */
533};
534
535struct drbd_md_io {
536 struct page *page;
537 unsigned long start_jif; /* last call to drbd_md_get_buffer */
538 unsigned long submit_jif; /* last _drbd_md_sync_page_io() submit */
539 const char *current_use;
540 atomic_t in_use;
541 unsigned int done;
542 int error;
543};
544
545struct bm_io_work {
546 struct drbd_work w;
547 struct drbd_peer_device *peer_device;
548 char *why;
549 enum bm_flag flags;
550 int (*io_fn)(struct drbd_device *device, struct drbd_peer_device *peer_device);
551 void (*done)(struct drbd_device *device, int rv);
552};
553
554struct fifo_buffer {
555 unsigned int head_index;
556 unsigned int size;
557 int total; /* sum of all values */
558 int values[] __counted_by(size);
559};
560extern struct fifo_buffer *fifo_alloc(unsigned int fifo_size);
561
562/* flag bits per connection */
563enum {
564 NET_CONGESTED, /* The data socket is congested */
565 RESOLVE_CONFLICTS, /* Set on one node, cleared on the peer! */
566 SEND_PING,
567 GOT_PING_ACK, /* set when we receive a ping_ack packet, ping_wait gets woken */
568 CONN_WD_ST_CHG_REQ, /* A cluster wide state change on the connection is active */
569 CONN_WD_ST_CHG_OKAY,
570 CONN_WD_ST_CHG_FAIL,
571 CONN_DRY_RUN, /* Expect disconnect after resync handshake. */
572 CREATE_BARRIER, /* next P_DATA is preceded by a P_BARRIER */
573 STATE_SENT, /* Do not change state/UUIDs while this is set */
574 CALLBACK_PENDING, /* Whether we have a call_usermodehelper(, UMH_WAIT_PROC)
575 * pending, from drbd worker context.
576 */
577 DISCONNECT_SENT,
578
579 DEVICE_WORK_PENDING, /* tell worker that some device has pending work */
580};
581
582enum which_state { NOW, OLD = NOW, NEW };
583
584struct drbd_resource {
585 char *name;
586#ifdef CONFIG_DEBUG_FS
587 struct dentry *debugfs_res;
588 struct dentry *debugfs_res_volumes;
589 struct dentry *debugfs_res_connections;
590 struct dentry *debugfs_res_in_flight_summary;
591#endif
592 struct kref kref;
593 struct idr devices; /* volume number to device mapping */
594 struct list_head connections;
595 struct list_head resources;
596 struct res_opts res_opts;
597 struct mutex conf_update; /* mutex for ready-copy-update of net_conf and disk_conf */
598 struct mutex adm_mutex; /* mutex to serialize administrative requests */
599 spinlock_t req_lock;
600
601 unsigned susp:1; /* IO suspended by user */
602 unsigned susp_nod:1; /* IO suspended because no data */
603 unsigned susp_fen:1; /* IO suspended because fence peer handler runs */
604
605 enum write_ordering_e write_ordering;
606
607 cpumask_var_t cpu_mask;
608};
609
610struct drbd_thread_timing_details
611{
612 unsigned long start_jif;
613 void *cb_addr;
614 const char *caller_fn;
615 unsigned int line;
616 unsigned int cb_nr;
617};
618
619struct drbd_connection {
620 struct list_head connections;
621 struct drbd_resource *resource;
622#ifdef CONFIG_DEBUG_FS
623 struct dentry *debugfs_conn;
624 struct dentry *debugfs_conn_callback_history;
625 struct dentry *debugfs_conn_oldest_requests;
626#endif
627 struct kref kref;
628 struct idr peer_devices; /* volume number to peer device mapping */
629 enum drbd_conns cstate; /* Only C_STANDALONE to C_WF_REPORT_PARAMS */
630 struct mutex cstate_mutex; /* Protects graceful disconnects */
631 unsigned int connect_cnt; /* Inc each time a connection is established */
632
633 unsigned long flags;
634 struct net_conf *net_conf; /* content protected by rcu */
635 wait_queue_head_t ping_wait; /* Woken upon reception of a ping, and a state change */
636
637 struct sockaddr_storage my_addr;
638 int my_addr_len;
639 struct sockaddr_storage peer_addr;
640 int peer_addr_len;
641
642 struct drbd_socket data; /* data/barrier/cstate/parameter packets */
643 struct drbd_socket meta; /* ping/ack (metadata) packets */
644 int agreed_pro_version; /* actually used protocol version */
645 u32 agreed_features;
646 unsigned long last_received; /* in jiffies, either socket */
647 unsigned int ko_count;
648
649 struct list_head transfer_log; /* all requests not yet fully processed */
650
651 struct crypto_shash *cram_hmac_tfm;
652 struct crypto_shash *integrity_tfm; /* checksums we compute, updates protected by connection->data->mutex */
653 struct crypto_shash *peer_integrity_tfm; /* checksums we verify, only accessed from receiver thread */
654 struct crypto_shash *csums_tfm;
655 struct crypto_shash *verify_tfm;
656 void *int_dig_in;
657 void *int_dig_vv;
658
659 /* receiver side */
660 struct drbd_epoch *current_epoch;
661 spinlock_t epoch_lock;
662 unsigned int epochs;
663 atomic_t current_tle_nr; /* transfer log epoch number */
664 unsigned current_tle_writes; /* writes seen within this tl epoch */
665
666 unsigned long last_reconnect_jif;
667 /* empty member on older kernels without blk_start_plug() */
668 struct blk_plug receiver_plug;
669 struct drbd_thread receiver;
670 struct drbd_thread worker;
671 struct drbd_thread ack_receiver;
672 struct workqueue_struct *ack_sender;
673
674 /* cached pointers,
675 * so we can look up the oldest pending requests more quickly.
676 * protected by resource->req_lock */
677 struct drbd_request *req_next; /* DRBD 9: todo.req_next */
678 struct drbd_request *req_ack_pending;
679 struct drbd_request *req_not_net_done;
680
681 /* sender side */
682 struct drbd_work_queue sender_work;
683
684#define DRBD_THREAD_DETAILS_HIST 16
685 unsigned int w_cb_nr; /* keeps counting up */
686 unsigned int r_cb_nr; /* keeps counting up */
687 struct drbd_thread_timing_details w_timing_details[DRBD_THREAD_DETAILS_HIST];
688 struct drbd_thread_timing_details r_timing_details[DRBD_THREAD_DETAILS_HIST];
689
690 struct {
691 unsigned long last_sent_barrier_jif;
692
693 /* whether this sender thread
694 * has processed a single write yet. */
695 bool seen_any_write_yet;
696
697 /* Which barrier number to send with the next P_BARRIER */
698 int current_epoch_nr;
699
700 /* how many write requests have been sent
701 * with req->epoch == current_epoch_nr.
702 * If none, no P_BARRIER will be sent. */
703 unsigned current_epoch_writes;
704 } send;
705};
706
707static inline bool has_net_conf(struct drbd_connection *connection)
708{
709 bool has_net_conf;
710
711 rcu_read_lock();
712 has_net_conf = rcu_dereference(connection->net_conf);
713 rcu_read_unlock();
714
715 return has_net_conf;
716}
717
718void __update_timing_details(
719 struct drbd_thread_timing_details *tdp,
720 unsigned int *cb_nr,
721 void *cb,
722 const char *fn, const unsigned int line);
723
724#define update_worker_timing_details(c, cb) \
725 __update_timing_details(c->w_timing_details, &c->w_cb_nr, cb, __func__ , __LINE__ )
726#define update_receiver_timing_details(c, cb) \
727 __update_timing_details(c->r_timing_details, &c->r_cb_nr, cb, __func__ , __LINE__ )
728
729struct submit_worker {
730 struct workqueue_struct *wq;
731 struct work_struct worker;
732
733 /* protected by ..->resource->req_lock */
734 struct list_head writes;
735};
736
737struct drbd_peer_device {
738 struct list_head peer_devices;
739 struct drbd_device *device;
740 struct drbd_connection *connection;
741 struct work_struct send_acks_work;
742#ifdef CONFIG_DEBUG_FS
743 struct dentry *debugfs_peer_dev;
744#endif
745};
746
747struct drbd_device {
748 struct drbd_resource *resource;
749 struct list_head peer_devices;
750 struct list_head pending_bitmap_io;
751
752 unsigned long flush_jif;
753#ifdef CONFIG_DEBUG_FS
754 struct dentry *debugfs_minor;
755 struct dentry *debugfs_vol;
756 struct dentry *debugfs_vol_oldest_requests;
757 struct dentry *debugfs_vol_act_log_extents;
758 struct dentry *debugfs_vol_resync_extents;
759 struct dentry *debugfs_vol_data_gen_id;
760 struct dentry *debugfs_vol_ed_gen_id;
761#endif
762
763 unsigned int vnr; /* volume number within the connection */
764 unsigned int minor; /* device minor number */
765
766 struct kref kref;
767
768 /* things that are stored as / read from meta data on disk */
769 unsigned long flags;
770
771 /* configured by drbdsetup */
772 struct drbd_backing_dev *ldev;
773
774 sector_t p_size; /* partner's disk size */
775 struct request_queue *rq_queue;
776 struct gendisk *vdisk;
777
778 unsigned long last_reattach_jif;
779 struct drbd_work resync_work;
780 struct drbd_work unplug_work;
781 struct timer_list resync_timer;
782 struct timer_list md_sync_timer;
783 struct timer_list start_resync_timer;
784 struct timer_list request_timer;
785
786 /* Used after attach while negotiating new disk state. */
787 union drbd_state new_state_tmp;
788
789 union drbd_dev_state state;
790 wait_queue_head_t misc_wait;
791 wait_queue_head_t state_wait; /* upon each state change. */
792 unsigned int send_cnt;
793 unsigned int recv_cnt;
794 unsigned int read_cnt;
795 unsigned int writ_cnt;
796 unsigned int al_writ_cnt;
797 unsigned int bm_writ_cnt;
798 atomic_t ap_bio_cnt; /* Requests we need to complete */
799 atomic_t ap_actlog_cnt; /* Requests waiting for activity log */
800 atomic_t ap_pending_cnt; /* AP data packets on the wire, ack expected */
801 atomic_t rs_pending_cnt; /* RS request/data packets on the wire */
802 atomic_t unacked_cnt; /* Need to send replies for */
803 atomic_t local_cnt; /* Waiting for local completion */
804 atomic_t suspend_cnt;
805
806 /* Interval tree of pending local requests */
807 struct rb_root read_requests;
808 struct rb_root write_requests;
809
810 /* for statistics and timeouts */
811 /* [0] read, [1] write */
812 struct list_head pending_master_completion[2];
813 struct list_head pending_completion[2];
814
815 /* use checksums for *this* resync */
816 bool use_csums;
817 /* blocks to resync in this run [unit BM_BLOCK_SIZE] */
818 unsigned long rs_total;
819 /* number of resync blocks that failed in this run */
820 unsigned long rs_failed;
821 /* Syncer's start time [unit jiffies] */
822 unsigned long rs_start;
823 /* cumulated time in PausedSyncX state [unit jiffies] */
824 unsigned long rs_paused;
825 /* skipped because csum was equal [unit BM_BLOCK_SIZE] */
826 unsigned long rs_same_csum;
827#define DRBD_SYNC_MARKS 8
828#define DRBD_SYNC_MARK_STEP (3*HZ)
829 /* block not up-to-date at mark [unit BM_BLOCK_SIZE] */
830 unsigned long rs_mark_left[DRBD_SYNC_MARKS];
831 /* marks's time [unit jiffies] */
832 unsigned long rs_mark_time[DRBD_SYNC_MARKS];
833 /* current index into rs_mark_{left,time} */
834 int rs_last_mark;
835 unsigned long rs_last_bcast; /* [unit jiffies] */
836
837 /* where does the admin want us to start? (sector) */
838 sector_t ov_start_sector;
839 sector_t ov_stop_sector;
840 /* where are we now? (sector) */
841 sector_t ov_position;
842 /* Start sector of out of sync range (to merge printk reporting). */
843 sector_t ov_last_oos_start;
844 /* size of out-of-sync range in sectors. */
845 sector_t ov_last_oos_size;
846 unsigned long ov_left; /* in bits */
847
848 struct drbd_bitmap *bitmap;
849 unsigned long bm_resync_fo; /* bit offset for drbd_bm_find_next */
850
851 /* Used to track operations of resync... */
852 struct lru_cache *resync;
853 /* Number of locked elements in resync LRU */
854 unsigned int resync_locked;
855 /* resync extent number waiting for application requests */
856 unsigned int resync_wenr;
857
858 int open_cnt;
859 u64 *p_uuid;
860
861 struct list_head active_ee; /* IO in progress (P_DATA gets written to disk) */
862 struct list_head sync_ee; /* IO in progress (P_RS_DATA_REPLY gets written to disk) */
863 struct list_head done_ee; /* need to send P_WRITE_ACK */
864 struct list_head read_ee; /* [RS]P_DATA_REQUEST being read */
865 struct list_head net_ee; /* zero-copy network send in progress */
866
867 int next_barrier_nr;
868 struct list_head resync_reads;
869 atomic_t pp_in_use; /* allocated from page pool */
870 atomic_t pp_in_use_by_net; /* sendpage()d, still referenced by tcp */
871 wait_queue_head_t ee_wait;
872 struct drbd_md_io md_io;
873 spinlock_t al_lock;
874 wait_queue_head_t al_wait;
875 struct lru_cache *act_log; /* activity log */
876 unsigned int al_tr_number;
877 int al_tr_cycle;
878 wait_queue_head_t seq_wait;
879 atomic_t packet_seq;
880 unsigned int peer_seq;
881 spinlock_t peer_seq_lock;
882 unsigned long comm_bm_set; /* communicated number of set bits. */
883 struct bm_io_work bm_io_work;
884 u64 ed_uuid; /* UUID of the exposed data */
885 struct mutex own_state_mutex;
886 struct mutex *state_mutex; /* either own_state_mutex or first_peer_device(device)->connection->cstate_mutex */
887 char congestion_reason; /* Why we where congested... */
888 atomic_t rs_sect_in; /* for incoming resync data rate, SyncTarget */
889 atomic_t rs_sect_ev; /* for submitted resync data rate, both */
890 int rs_last_sect_ev; /* counter to compare with */
891 int rs_last_events; /* counter of read or write "events" (unit sectors)
892 * on the lower level device when we last looked. */
893 int c_sync_rate; /* current resync rate after syncer throttle magic */
894 struct fifo_buffer *rs_plan_s; /* correction values of resync planer (RCU, connection->conn_update) */
895 int rs_in_flight; /* resync sectors in flight (to proxy, in proxy and from proxy) */
896 atomic_t ap_in_flight; /* App sectors in flight (waiting for ack) */
897 unsigned int peer_max_bio_size;
898 unsigned int local_max_bio_size;
899
900 /* any requests that would block in drbd_make_request()
901 * are deferred to this single-threaded work queue */
902 struct submit_worker submit;
903};
904
905struct drbd_bm_aio_ctx {
906 struct drbd_device *device;
907 struct list_head list; /* on device->pending_bitmap_io */;
908 unsigned long start_jif;
909 atomic_t in_flight;
910 unsigned int done;
911 unsigned flags;
912#define BM_AIO_COPY_PAGES 1
913#define BM_AIO_WRITE_HINTED 2
914#define BM_AIO_WRITE_ALL_PAGES 4
915#define BM_AIO_READ 8
916 int error;
917 struct kref kref;
918};
919
920struct drbd_config_context {
921 /* assigned from drbd_genlmsghdr */
922 unsigned int minor;
923 /* assigned from request attributes, if present */
924 unsigned int volume;
925#define VOLUME_UNSPECIFIED (-1U)
926 /* pointer into the request skb,
927 * limited lifetime! */
928 char *resource_name;
929 struct nlattr *my_addr;
930 struct nlattr *peer_addr;
931
932 /* reply buffer */
933 struct sk_buff *reply_skb;
934 /* pointer into reply buffer */
935 struct drbd_genlmsghdr *reply_dh;
936 /* resolved from attributes, if possible */
937 struct drbd_device *device;
938 struct drbd_resource *resource;
939 struct drbd_connection *connection;
940};
941
942static inline struct drbd_device *minor_to_device(unsigned int minor)
943{
944 return (struct drbd_device *)idr_find(&drbd_devices, id: minor);
945}
946
947static inline struct drbd_peer_device *first_peer_device(struct drbd_device *device)
948{
949 return list_first_entry_or_null(&device->peer_devices, struct drbd_peer_device, peer_devices);
950}
951
952static inline struct drbd_peer_device *
953conn_peer_device(struct drbd_connection *connection, int volume_number)
954{
955 return idr_find(&connection->peer_devices, id: volume_number);
956}
957
958#define for_each_resource(resource, _resources) \
959 list_for_each_entry(resource, _resources, resources)
960
961#define for_each_resource_rcu(resource, _resources) \
962 list_for_each_entry_rcu(resource, _resources, resources)
963
964#define for_each_resource_safe(resource, tmp, _resources) \
965 list_for_each_entry_safe(resource, tmp, _resources, resources)
966
967#define for_each_connection(connection, resource) \
968 list_for_each_entry(connection, &resource->connections, connections)
969
970#define for_each_connection_rcu(connection, resource) \
971 list_for_each_entry_rcu(connection, &resource->connections, connections)
972
973#define for_each_connection_safe(connection, tmp, resource) \
974 list_for_each_entry_safe(connection, tmp, &resource->connections, connections)
975
976#define for_each_peer_device(peer_device, device) \
977 list_for_each_entry(peer_device, &device->peer_devices, peer_devices)
978
979#define for_each_peer_device_rcu(peer_device, device) \
980 list_for_each_entry_rcu(peer_device, &device->peer_devices, peer_devices)
981
982#define for_each_peer_device_safe(peer_device, tmp, device) \
983 list_for_each_entry_safe(peer_device, tmp, &device->peer_devices, peer_devices)
984
985static inline unsigned int device_to_minor(struct drbd_device *device)
986{
987 return device->minor;
988}
989
990/*
991 * function declarations
992 *************************/
993
994/* drbd_main.c */
995
996enum dds_flags {
997 DDSF_FORCED = 1,
998 DDSF_NO_RESYNC = 2, /* Do not run a resync for the new space */
999};
1000
1001extern void drbd_init_set_defaults(struct drbd_device *device);
1002extern int drbd_thread_start(struct drbd_thread *thi);
1003extern void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait);
1004#ifdef CONFIG_SMP
1005extern void drbd_thread_current_set_cpu(struct drbd_thread *thi);
1006#else
1007#define drbd_thread_current_set_cpu(A) ({})
1008#endif
1009extern void tl_release(struct drbd_connection *, unsigned int barrier_nr,
1010 unsigned int set_size);
1011extern void tl_clear(struct drbd_connection *);
1012extern void drbd_free_sock(struct drbd_connection *connection);
1013extern int drbd_send(struct drbd_connection *connection, struct socket *sock,
1014 void *buf, size_t size, unsigned msg_flags);
1015extern int drbd_send_all(struct drbd_connection *, struct socket *, void *, size_t,
1016 unsigned);
1017
1018extern int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd);
1019extern int drbd_send_protocol(struct drbd_connection *connection);
1020extern int drbd_send_uuids(struct drbd_peer_device *);
1021extern int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *);
1022extern void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *);
1023extern int drbd_send_sizes(struct drbd_peer_device *, int trigger_reply, enum dds_flags flags);
1024extern int drbd_send_state(struct drbd_peer_device *, union drbd_state s);
1025extern int drbd_send_current_state(struct drbd_peer_device *);
1026extern int drbd_send_sync_param(struct drbd_peer_device *);
1027extern void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr,
1028 u32 set_size);
1029extern int drbd_send_ack(struct drbd_peer_device *, enum drbd_packet,
1030 struct drbd_peer_request *);
1031extern void drbd_send_ack_rp(struct drbd_peer_device *, enum drbd_packet,
1032 struct p_block_req *rp);
1033extern void drbd_send_ack_dp(struct drbd_peer_device *, enum drbd_packet,
1034 struct p_data *dp, int data_size);
1035extern int drbd_send_ack_ex(struct drbd_peer_device *, enum drbd_packet,
1036 sector_t sector, int blksize, u64 block_id);
1037extern int drbd_send_out_of_sync(struct drbd_peer_device *, struct drbd_request *);
1038extern int drbd_send_block(struct drbd_peer_device *, enum drbd_packet,
1039 struct drbd_peer_request *);
1040extern int drbd_send_dblock(struct drbd_peer_device *, struct drbd_request *req);
1041extern int drbd_send_drequest(struct drbd_peer_device *, int cmd,
1042 sector_t sector, int size, u64 block_id);
1043extern int drbd_send_drequest_csum(struct drbd_peer_device *, sector_t sector,
1044 int size, void *digest, int digest_size,
1045 enum drbd_packet cmd);
1046extern int drbd_send_ov_request(struct drbd_peer_device *, sector_t sector, int size);
1047
1048extern int drbd_send_bitmap(struct drbd_device *device, struct drbd_peer_device *peer_device);
1049extern void drbd_send_sr_reply(struct drbd_peer_device *, enum drbd_state_rv retcode);
1050extern void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode);
1051extern int drbd_send_rs_deallocated(struct drbd_peer_device *, struct drbd_peer_request *);
1052extern void drbd_backing_dev_free(struct drbd_device *device, struct drbd_backing_dev *ldev);
1053extern void drbd_device_cleanup(struct drbd_device *device);
1054extern void drbd_print_uuids(struct drbd_device *device, const char *text);
1055extern void drbd_queue_unplug(struct drbd_device *device);
1056
1057extern void conn_md_sync(struct drbd_connection *connection);
1058extern void drbd_md_write(struct drbd_device *device, void *buffer);
1059extern void drbd_md_sync(struct drbd_device *device);
1060extern int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev);
1061extern void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local);
1062extern void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local);
1063extern void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local);
1064extern void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local);
1065extern void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local);
1066extern void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local);
1067extern void drbd_md_set_flag(struct drbd_device *device, int flags) __must_hold(local);
1068extern void drbd_md_clear_flag(struct drbd_device *device, int flags)__must_hold(local);
1069extern int drbd_md_test_flag(struct drbd_backing_dev *, int);
1070extern void drbd_md_mark_dirty(struct drbd_device *device);
1071extern void drbd_queue_bitmap_io(struct drbd_device *device,
1072 int (*io_fn)(struct drbd_device *, struct drbd_peer_device *),
1073 void (*done)(struct drbd_device *, int),
1074 char *why, enum bm_flag flags,
1075 struct drbd_peer_device *peer_device);
1076extern int drbd_bitmap_io(struct drbd_device *device,
1077 int (*io_fn)(struct drbd_device *, struct drbd_peer_device *),
1078 char *why, enum bm_flag flags,
1079 struct drbd_peer_device *peer_device);
1080extern int drbd_bitmap_io_from_worker(struct drbd_device *device,
1081 int (*io_fn)(struct drbd_device *, struct drbd_peer_device *),
1082 char *why, enum bm_flag flags,
1083 struct drbd_peer_device *peer_device);
1084extern int drbd_bmio_set_n_write(struct drbd_device *device,
1085 struct drbd_peer_device *peer_device) __must_hold(local);
1086extern int drbd_bmio_clear_n_write(struct drbd_device *device,
1087 struct drbd_peer_device *peer_device) __must_hold(local);
1088
1089/* Meta data layout
1090 *
1091 * We currently have two possible layouts.
1092 * Offsets in (512 byte) sectors.
1093 * external:
1094 * |----------- md_size_sect ------------------|
1095 * [ 4k superblock ][ activity log ][ Bitmap ]
1096 * | al_offset == 8 |
1097 * | bm_offset = al_offset + X |
1098 * ==> bitmap sectors = md_size_sect - bm_offset
1099 *
1100 * Variants:
1101 * old, indexed fixed size meta data:
1102 *
1103 * internal:
1104 * |----------- md_size_sect ------------------|
1105 * [data.....][ Bitmap ][ activity log ][ 4k superblock ][padding*]
1106 * | al_offset < 0 |
1107 * | bm_offset = al_offset - Y |
1108 * ==> bitmap sectors = Y = al_offset - bm_offset
1109 *
1110 * [padding*] are zero or up to 7 unused 512 Byte sectors to the
1111 * end of the device, so that the [4k superblock] will be 4k aligned.
1112 *
1113 * The activity log consists of 4k transaction blocks,
1114 * which are written in a ring-buffer, or striped ring-buffer like fashion,
1115 * which are writtensize used to be fixed 32kB,
1116 * but is about to become configurable.
1117 */
1118
1119/* Our old fixed size meta data layout
1120 * allows up to about 3.8TB, so if you want more,
1121 * you need to use the "flexible" meta data format. */
1122#define MD_128MB_SECT (128LLU << 11) /* 128 MB, unit sectors */
1123#define MD_4kB_SECT 8
1124#define MD_32kB_SECT 64
1125
1126/* One activity log extent represents 4M of storage */
1127#define AL_EXTENT_SHIFT 22
1128#define AL_EXTENT_SIZE (1<<AL_EXTENT_SHIFT)
1129
1130/* We could make these currently hardcoded constants configurable
1131 * variables at create-md time (or even re-configurable at runtime?).
1132 * Which will require some more changes to the DRBD "super block"
1133 * and attach code.
1134 *
1135 * updates per transaction:
1136 * This many changes to the active set can be logged with one transaction.
1137 * This number is arbitrary.
1138 * context per transaction:
1139 * This many context extent numbers are logged with each transaction.
1140 * This number is resulting from the transaction block size (4k), the layout
1141 * of the transaction header, and the number of updates per transaction.
1142 * See drbd_actlog.c:struct al_transaction_on_disk
1143 * */
1144#define AL_UPDATES_PER_TRANSACTION 64 // arbitrary
1145#define AL_CONTEXT_PER_TRANSACTION 919 // (4096 - 36 - 6*64)/4
1146
1147#if BITS_PER_LONG == 32
1148#define LN2_BPL 5
1149#define cpu_to_lel(A) cpu_to_le32(A)
1150#define lel_to_cpu(A) le32_to_cpu(A)
1151#elif BITS_PER_LONG == 64
1152#define LN2_BPL 6
1153#define cpu_to_lel(A) cpu_to_le64(A)
1154#define lel_to_cpu(A) le64_to_cpu(A)
1155#else
1156#error "LN2 of BITS_PER_LONG unknown!"
1157#endif
1158
1159/* resync bitmap */
1160/* 16MB sized 'bitmap extent' to track syncer usage */
1161struct bm_extent {
1162 int rs_left; /* number of bits set (out of sync) in this extent. */
1163 int rs_failed; /* number of failed resync requests in this extent. */
1164 unsigned long flags;
1165 struct lc_element lce;
1166};
1167
1168#define BME_NO_WRITES 0 /* bm_extent.flags: no more requests on this one! */
1169#define BME_LOCKED 1 /* bm_extent.flags: syncer active on this one. */
1170#define BME_PRIORITY 2 /* finish resync IO on this extent ASAP! App IO waiting! */
1171
1172/* drbd_bitmap.c */
1173/*
1174 * We need to store one bit for a block.
1175 * Example: 1GB disk @ 4096 byte blocks ==> we need 32 KB bitmap.
1176 * Bit 0 ==> local node thinks this block is binary identical on both nodes
1177 * Bit 1 ==> local node thinks this block needs to be synced.
1178 */
1179
1180#define SLEEP_TIME (HZ/10)
1181
1182/* We do bitmap IO in units of 4k blocks.
1183 * We also still have a hardcoded 4k per bit relation. */
1184#define BM_BLOCK_SHIFT 12 /* 4k per bit */
1185#define BM_BLOCK_SIZE (1<<BM_BLOCK_SHIFT)
1186/* mostly arbitrarily set the represented size of one bitmap extent,
1187 * aka resync extent, to 16 MiB (which is also 512 Byte worth of bitmap
1188 * at 4k per bit resolution) */
1189#define BM_EXT_SHIFT 24 /* 16 MiB per resync extent */
1190#define BM_EXT_SIZE (1<<BM_EXT_SHIFT)
1191
1192#if (BM_EXT_SHIFT != 24) || (BM_BLOCK_SHIFT != 12)
1193#error "HAVE YOU FIXED drbdmeta AS WELL??"
1194#endif
1195
1196/* thus many _storage_ sectors are described by one bit */
1197#define BM_SECT_TO_BIT(x) ((x)>>(BM_BLOCK_SHIFT-9))
1198#define BM_BIT_TO_SECT(x) ((sector_t)(x)<<(BM_BLOCK_SHIFT-9))
1199#define BM_SECT_PER_BIT BM_BIT_TO_SECT(1)
1200
1201/* bit to represented kilo byte conversion */
1202#define Bit2KB(bits) ((bits)<<(BM_BLOCK_SHIFT-10))
1203
1204/* in which _bitmap_ extent (resp. sector) the bit for a certain
1205 * _storage_ sector is located in */
1206#define BM_SECT_TO_EXT(x) ((x)>>(BM_EXT_SHIFT-9))
1207#define BM_BIT_TO_EXT(x) ((x) >> (BM_EXT_SHIFT - BM_BLOCK_SHIFT))
1208
1209/* first storage sector a bitmap extent corresponds to */
1210#define BM_EXT_TO_SECT(x) ((sector_t)(x) << (BM_EXT_SHIFT-9))
1211/* how much _storage_ sectors we have per bitmap extent */
1212#define BM_SECT_PER_EXT BM_EXT_TO_SECT(1)
1213/* how many bits are covered by one bitmap extent (resync extent) */
1214#define BM_BITS_PER_EXT (1UL << (BM_EXT_SHIFT - BM_BLOCK_SHIFT))
1215
1216#define BM_BLOCKS_PER_BM_EXT_MASK (BM_BITS_PER_EXT - 1)
1217
1218
1219/* in one sector of the bitmap, we have this many activity_log extents. */
1220#define AL_EXT_PER_BM_SECT (1 << (BM_EXT_SHIFT - AL_EXTENT_SHIFT))
1221
1222/* the extent in "PER_EXTENT" below is an activity log extent
1223 * we need that many (long words/bytes) to store the bitmap
1224 * of one AL_EXTENT_SIZE chunk of storage.
1225 * we can store the bitmap for that many AL_EXTENTS within
1226 * one sector of the _on_disk_ bitmap:
1227 * bit 0 bit 37 bit 38 bit (512*8)-1
1228 * ...|........|........|.. // ..|........|
1229 * sect. 0 `296 `304 ^(512*8*8)-1
1230 *
1231#define BM_WORDS_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / BITS_PER_LONG )
1232#define BM_BYTES_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / 8 ) // 128
1233#define BM_EXT_PER_SECT ( 512 / BM_BYTES_PER_EXTENT ) // 4
1234 */
1235
1236#define DRBD_MAX_SECTORS_32 (0xffffffffLU)
1237/* we have a certain meta data variant that has a fixed on-disk size of 128
1238 * MiB, of which 4k are our "superblock", and 32k are the fixed size activity
1239 * log, leaving this many sectors for the bitmap.
1240 */
1241
1242#define DRBD_MAX_SECTORS_FIXED_BM \
1243 ((MD_128MB_SECT - MD_32kB_SECT - MD_4kB_SECT) * (1LL<<(BM_EXT_SHIFT-9)))
1244#define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_FIXED_BM
1245/* 16 TB in units of sectors */
1246#if BITS_PER_LONG == 32
1247/* adjust by one page worth of bitmap,
1248 * so we won't wrap around in drbd_bm_find_next_bit.
1249 * you should use 64bit OS for that much storage, anyways. */
1250#define DRBD_MAX_SECTORS_FLEX BM_BIT_TO_SECT(0xffff7fff)
1251#else
1252/* we allow up to 1 PiB now on 64bit architecture with "flexible" meta data */
1253#define DRBD_MAX_SECTORS_FLEX (1UL << 51)
1254/* corresponds to (1UL << 38) bits right now. */
1255#endif
1256
1257/* Estimate max bio size as 256 * PAGE_SIZE,
1258 * so for typical PAGE_SIZE of 4k, that is (1<<20) Byte.
1259 * Since we may live in a mixed-platform cluster,
1260 * we limit us to a platform agnostic constant here for now.
1261 * A followup commit may allow even bigger BIO sizes,
1262 * once we thought that through. */
1263#define DRBD_MAX_BIO_SIZE (1U << 20)
1264#if DRBD_MAX_BIO_SIZE > (BIO_MAX_VECS << PAGE_SHIFT)
1265#error Architecture not supported: DRBD_MAX_BIO_SIZE > BIO_MAX_SIZE
1266#endif
1267#define DRBD_MAX_BIO_SIZE_SAFE (1U << 12) /* Works always = 4k */
1268
1269#define DRBD_MAX_SIZE_H80_PACKET (1U << 15) /* Header 80 only allows packets up to 32KiB data */
1270#define DRBD_MAX_BIO_SIZE_P95 (1U << 17) /* Protocol 95 to 99 allows bios up to 128KiB */
1271
1272/* For now, don't allow more than half of what we can "activate" in one
1273 * activity log transaction to be discarded in one go. We may need to rework
1274 * drbd_al_begin_io() to allow for even larger discard ranges */
1275#define DRBD_MAX_BATCH_BIO_SIZE (AL_UPDATES_PER_TRANSACTION/2*AL_EXTENT_SIZE)
1276#define DRBD_MAX_BBIO_SECTORS (DRBD_MAX_BATCH_BIO_SIZE >> 9)
1277
1278extern int drbd_bm_init(struct drbd_device *device);
1279extern int drbd_bm_resize(struct drbd_device *device, sector_t sectors, int set_new_bits);
1280extern void drbd_bm_cleanup(struct drbd_device *device);
1281extern void drbd_bm_set_all(struct drbd_device *device);
1282extern void drbd_bm_clear_all(struct drbd_device *device);
1283/* set/clear/test only a few bits at a time */
1284extern int drbd_bm_set_bits(
1285 struct drbd_device *device, unsigned long s, unsigned long e);
1286extern int drbd_bm_clear_bits(
1287 struct drbd_device *device, unsigned long s, unsigned long e);
1288extern int drbd_bm_count_bits(
1289 struct drbd_device *device, const unsigned long s, const unsigned long e);
1290/* bm_set_bits variant for use while holding drbd_bm_lock,
1291 * may process the whole bitmap in one go */
1292extern void _drbd_bm_set_bits(struct drbd_device *device,
1293 const unsigned long s, const unsigned long e);
1294extern int drbd_bm_test_bit(struct drbd_device *device, unsigned long bitnr);
1295extern int drbd_bm_e_weight(struct drbd_device *device, unsigned long enr);
1296extern int drbd_bm_read(struct drbd_device *device,
1297 struct drbd_peer_device *peer_device) __must_hold(local);
1298extern void drbd_bm_mark_for_writeout(struct drbd_device *device, int page_nr);
1299extern int drbd_bm_write(struct drbd_device *device,
1300 struct drbd_peer_device *peer_device) __must_hold(local);
1301extern void drbd_bm_reset_al_hints(struct drbd_device *device) __must_hold(local);
1302extern int drbd_bm_write_hinted(struct drbd_device *device) __must_hold(local);
1303extern int drbd_bm_write_lazy(struct drbd_device *device, unsigned upper_idx) __must_hold(local);
1304extern int drbd_bm_write_all(struct drbd_device *device,
1305 struct drbd_peer_device *peer_device) __must_hold(local);
1306extern int drbd_bm_write_copy_pages(struct drbd_device *device,
1307 struct drbd_peer_device *peer_device) __must_hold(local);
1308extern size_t drbd_bm_words(struct drbd_device *device);
1309extern unsigned long drbd_bm_bits(struct drbd_device *device);
1310extern sector_t drbd_bm_capacity(struct drbd_device *device);
1311
1312#define DRBD_END_OF_BITMAP (~(unsigned long)0)
1313extern unsigned long drbd_bm_find_next(struct drbd_device *device, unsigned long bm_fo);
1314/* bm_find_next variants for use while you hold drbd_bm_lock() */
1315extern unsigned long _drbd_bm_find_next(struct drbd_device *device, unsigned long bm_fo);
1316extern unsigned long _drbd_bm_find_next_zero(struct drbd_device *device, unsigned long bm_fo);
1317extern unsigned long _drbd_bm_total_weight(struct drbd_device *device);
1318extern unsigned long drbd_bm_total_weight(struct drbd_device *device);
1319/* for receive_bitmap */
1320extern void drbd_bm_merge_lel(struct drbd_device *device, size_t offset,
1321 size_t number, unsigned long *buffer);
1322/* for _drbd_send_bitmap */
1323extern void drbd_bm_get_lel(struct drbd_device *device, size_t offset,
1324 size_t number, unsigned long *buffer);
1325
1326extern void drbd_bm_lock(struct drbd_device *device, char *why, enum bm_flag flags);
1327extern void drbd_bm_unlock(struct drbd_device *device);
1328/* drbd_main.c */
1329
1330extern struct kmem_cache *drbd_request_cache;
1331extern struct kmem_cache *drbd_ee_cache; /* peer requests */
1332extern struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
1333extern struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
1334extern mempool_t drbd_request_mempool;
1335extern mempool_t drbd_ee_mempool;
1336
1337/* drbd's page pool, used to buffer data received from the peer,
1338 * or data requested by the peer.
1339 *
1340 * This does not have an emergency reserve.
1341 *
1342 * When allocating from this pool, it first takes pages from the pool.
1343 * Only if the pool is depleted will try to allocate from the system.
1344 *
1345 * The assumption is that pages taken from this pool will be processed,
1346 * and given back, "quickly", and then can be recycled, so we can avoid
1347 * frequent calls to alloc_page(), and still will be able to make progress even
1348 * under memory pressure.
1349 */
1350extern struct page *drbd_pp_pool;
1351extern spinlock_t drbd_pp_lock;
1352extern int drbd_pp_vacant;
1353extern wait_queue_head_t drbd_pp_wait;
1354
1355/* We also need a standard (emergency-reserve backed) page pool
1356 * for meta data IO (activity log, bitmap).
1357 * We can keep it global, as long as it is used as "N pages at a time".
1358 * 128 should be plenty, currently we probably can get away with as few as 1.
1359 */
1360#define DRBD_MIN_POOL_PAGES 128
1361extern mempool_t drbd_md_io_page_pool;
1362
1363/* We also need to make sure we get a bio
1364 * when we need it for housekeeping purposes */
1365extern struct bio_set drbd_md_io_bio_set;
1366
1367/* And a bio_set for cloning */
1368extern struct bio_set drbd_io_bio_set;
1369
1370extern struct mutex resources_mutex;
1371
1372extern int conn_lowest_minor(struct drbd_connection *connection);
1373extern enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor);
1374extern void drbd_destroy_device(struct kref *kref);
1375extern void drbd_delete_device(struct drbd_device *device);
1376
1377extern struct drbd_resource *drbd_create_resource(const char *name);
1378extern void drbd_free_resource(struct drbd_resource *resource);
1379
1380extern int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts);
1381extern struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts);
1382extern void drbd_destroy_connection(struct kref *kref);
1383extern struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
1384 void *peer_addr, int peer_addr_len);
1385extern struct drbd_resource *drbd_find_resource(const char *name);
1386extern void drbd_destroy_resource(struct kref *kref);
1387extern void conn_free_crypto(struct drbd_connection *connection);
1388
1389/* drbd_req */
1390extern void do_submit(struct work_struct *ws);
1391extern void __drbd_make_request(struct drbd_device *, struct bio *);
1392void drbd_submit_bio(struct bio *bio);
1393extern int drbd_read_remote(struct drbd_device *device, struct drbd_request *req);
1394extern int is_valid_ar_handle(struct drbd_request *, sector_t);
1395
1396
1397/* drbd_nl.c */
1398
1399extern struct mutex notification_mutex;
1400
1401extern void drbd_suspend_io(struct drbd_device *device);
1402extern void drbd_resume_io(struct drbd_device *device);
1403extern char *ppsize(char *buf, unsigned long long size);
1404extern sector_t drbd_new_dev_size(struct drbd_device *, struct drbd_backing_dev *, sector_t, int);
1405enum determine_dev_size {
1406 DS_ERROR_SHRINK = -3,
1407 DS_ERROR_SPACE_MD = -2,
1408 DS_ERROR = -1,
1409 DS_UNCHANGED = 0,
1410 DS_SHRUNK = 1,
1411 DS_GREW = 2,
1412 DS_GREW_FROM_ZERO = 3,
1413};
1414extern enum determine_dev_size
1415drbd_determine_dev_size(struct drbd_device *, enum dds_flags, struct resize_parms *) __must_hold(local);
1416extern void resync_after_online_grow(struct drbd_device *);
1417extern void drbd_reconsider_queue_parameters(struct drbd_device *device,
1418 struct drbd_backing_dev *bdev, struct o_qlim *o);
1419extern enum drbd_state_rv drbd_set_role(struct drbd_device *device,
1420 enum drbd_role new_role,
1421 int force);
1422extern bool conn_try_outdate_peer(struct drbd_connection *connection);
1423extern void conn_try_outdate_peer_async(struct drbd_connection *connection);
1424extern enum drbd_peer_state conn_khelper(struct drbd_connection *connection, char *cmd);
1425extern int drbd_khelper(struct drbd_device *device, char *cmd);
1426
1427/* drbd_worker.c */
1428/* bi_end_io handlers */
1429extern void drbd_md_endio(struct bio *bio);
1430extern void drbd_peer_request_endio(struct bio *bio);
1431extern void drbd_request_endio(struct bio *bio);
1432extern int drbd_worker(struct drbd_thread *thi);
1433enum drbd_ret_code drbd_resync_after_valid(struct drbd_device *device, int o_minor);
1434void drbd_resync_after_changed(struct drbd_device *device);
1435extern void drbd_start_resync(struct drbd_device *device, enum drbd_conns side);
1436extern void resume_next_sg(struct drbd_device *device);
1437extern void suspend_other_sg(struct drbd_device *device);
1438extern int drbd_resync_finished(struct drbd_peer_device *peer_device);
1439/* maybe rather drbd_main.c ? */
1440extern void *drbd_md_get_buffer(struct drbd_device *device, const char *intent);
1441extern void drbd_md_put_buffer(struct drbd_device *device);
1442extern int drbd_md_sync_page_io(struct drbd_device *device,
1443 struct drbd_backing_dev *bdev, sector_t sector, enum req_op op);
1444extern void drbd_ov_out_of_sync_found(struct drbd_peer_device *peer_device,
1445 sector_t sector, int size);
1446extern void wait_until_done_or_force_detached(struct drbd_device *device,
1447 struct drbd_backing_dev *bdev, unsigned int *done);
1448extern void drbd_rs_controller_reset(struct drbd_peer_device *peer_device);
1449
1450static inline void ov_out_of_sync_print(struct drbd_peer_device *peer_device)
1451{
1452 struct drbd_device *device = peer_device->device;
1453
1454 if (device->ov_last_oos_size) {
1455 drbd_err(peer_device, "Out of sync: start=%llu, size=%lu (sectors)\n",
1456 (unsigned long long)device->ov_last_oos_start,
1457 (unsigned long)device->ov_last_oos_size);
1458 }
1459 device->ov_last_oos_size = 0;
1460}
1461
1462
1463extern void drbd_csum_bio(struct crypto_shash *, struct bio *, void *);
1464extern void drbd_csum_ee(struct crypto_shash *, struct drbd_peer_request *,
1465 void *);
1466/* worker callbacks */
1467extern int w_e_end_data_req(struct drbd_work *, int);
1468extern int w_e_end_rsdata_req(struct drbd_work *, int);
1469extern int w_e_end_csum_rs_req(struct drbd_work *, int);
1470extern int w_e_end_ov_reply(struct drbd_work *, int);
1471extern int w_e_end_ov_req(struct drbd_work *, int);
1472extern int w_ov_finished(struct drbd_work *, int);
1473extern int w_resync_timer(struct drbd_work *, int);
1474extern int w_send_write_hint(struct drbd_work *, int);
1475extern int w_send_dblock(struct drbd_work *, int);
1476extern int w_send_read_req(struct drbd_work *, int);
1477extern int w_e_reissue(struct drbd_work *, int);
1478extern int w_restart_disk_io(struct drbd_work *, int);
1479extern int w_send_out_of_sync(struct drbd_work *, int);
1480
1481extern void resync_timer_fn(struct timer_list *t);
1482extern void start_resync_timer_fn(struct timer_list *t);
1483
1484extern void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req);
1485
1486/* drbd_receiver.c */
1487extern int drbd_issue_discard_or_zero_out(struct drbd_device *device,
1488 sector_t start, unsigned int nr_sectors, int flags);
1489extern int drbd_receiver(struct drbd_thread *thi);
1490extern int drbd_ack_receiver(struct drbd_thread *thi);
1491extern void drbd_send_ping_wf(struct work_struct *ws);
1492extern void drbd_send_acks_wf(struct work_struct *ws);
1493extern bool drbd_rs_c_min_rate_throttle(struct drbd_device *device);
1494extern bool drbd_rs_should_slow_down(struct drbd_peer_device *peer_device, sector_t sector,
1495 bool throttle_if_app_is_waiting);
1496extern int drbd_submit_peer_request(struct drbd_peer_request *peer_req);
1497extern int drbd_free_peer_reqs(struct drbd_device *, struct list_head *);
1498extern struct drbd_peer_request *drbd_alloc_peer_req(struct drbd_peer_device *, u64,
1499 sector_t, unsigned int,
1500 unsigned int,
1501 gfp_t) __must_hold(local);
1502extern void __drbd_free_peer_req(struct drbd_device *, struct drbd_peer_request *,
1503 int);
1504#define drbd_free_peer_req(m,e) __drbd_free_peer_req(m, e, 0)
1505#define drbd_free_net_peer_req(m,e) __drbd_free_peer_req(m, e, 1)
1506extern struct page *drbd_alloc_pages(struct drbd_peer_device *, unsigned int, bool);
1507extern void drbd_set_recv_tcq(struct drbd_device *device, int tcq_enabled);
1508extern void _drbd_clear_done_ee(struct drbd_device *device, struct list_head *to_be_freed);
1509extern int drbd_connected(struct drbd_peer_device *);
1510
1511/* sets the number of 512 byte sectors of our virtual device */
1512void drbd_set_my_capacity(struct drbd_device *device, sector_t size);
1513
1514/*
1515 * used to submit our private bio
1516 */
1517static inline void drbd_submit_bio_noacct(struct drbd_device *device,
1518 int fault_type, struct bio *bio)
1519{
1520 __release(local);
1521 if (!bio->bi_bdev) {
1522 drbd_err(device, "drbd_submit_bio_noacct: bio->bi_bdev == NULL\n");
1523 bio->bi_status = BLK_STS_IOERR;
1524 bio_endio(bio);
1525 return;
1526 }
1527
1528 if (drbd_insert_fault(device, type: fault_type))
1529 bio_io_error(bio);
1530 else
1531 submit_bio_noacct(bio);
1532}
1533
1534void drbd_bump_write_ordering(struct drbd_resource *resource, struct drbd_backing_dev *bdev,
1535 enum write_ordering_e wo);
1536
1537/* drbd_proc.c */
1538extern struct proc_dir_entry *drbd_proc;
1539int drbd_seq_show(struct seq_file *seq, void *v);
1540
1541/* drbd_actlog.c */
1542extern bool drbd_al_begin_io_prepare(struct drbd_device *device, struct drbd_interval *i);
1543extern int drbd_al_begin_io_nonblock(struct drbd_device *device, struct drbd_interval *i);
1544extern void drbd_al_begin_io_commit(struct drbd_device *device);
1545extern bool drbd_al_begin_io_fastpath(struct drbd_device *device, struct drbd_interval *i);
1546extern void drbd_al_begin_io(struct drbd_device *device, struct drbd_interval *i);
1547extern void drbd_al_complete_io(struct drbd_device *device, struct drbd_interval *i);
1548extern void drbd_rs_complete_io(struct drbd_device *device, sector_t sector);
1549extern int drbd_rs_begin_io(struct drbd_device *device, sector_t sector);
1550extern int drbd_try_rs_begin_io(struct drbd_peer_device *peer_device, sector_t sector);
1551extern void drbd_rs_cancel_all(struct drbd_device *device);
1552extern int drbd_rs_del_all(struct drbd_device *device);
1553extern void drbd_rs_failed_io(struct drbd_peer_device *peer_device,
1554 sector_t sector, int size);
1555extern void drbd_advance_rs_marks(struct drbd_peer_device *peer_device, unsigned long still_to_go);
1556
1557enum update_sync_bits_mode { RECORD_RS_FAILED, SET_OUT_OF_SYNC, SET_IN_SYNC };
1558extern int __drbd_change_sync(struct drbd_peer_device *peer_device, sector_t sector, int size,
1559 enum update_sync_bits_mode mode);
1560#define drbd_set_in_sync(peer_device, sector, size) \
1561 __drbd_change_sync(peer_device, sector, size, SET_IN_SYNC)
1562#define drbd_set_out_of_sync(peer_device, sector, size) \
1563 __drbd_change_sync(peer_device, sector, size, SET_OUT_OF_SYNC)
1564#define drbd_rs_failed_io(peer_device, sector, size) \
1565 __drbd_change_sync(peer_device, sector, size, RECORD_RS_FAILED)
1566extern void drbd_al_shrink(struct drbd_device *device);
1567extern int drbd_al_initialize(struct drbd_device *, void *);
1568
1569/* drbd_nl.c */
1570/* state info broadcast */
1571struct sib_info {
1572 enum drbd_state_info_bcast_reason sib_reason;
1573 union {
1574 struct {
1575 char *helper_name;
1576 unsigned helper_exit_code;
1577 };
1578 struct {
1579 union drbd_state os;
1580 union drbd_state ns;
1581 };
1582 };
1583};
1584void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib);
1585
1586extern int notify_resource_state(struct sk_buff *,
1587 unsigned int,
1588 struct drbd_resource *,
1589 struct resource_info *,
1590 enum drbd_notification_type);
1591extern int notify_device_state(struct sk_buff *,
1592 unsigned int,
1593 struct drbd_device *,
1594 struct device_info *,
1595 enum drbd_notification_type);
1596extern int notify_connection_state(struct sk_buff *,
1597 unsigned int,
1598 struct drbd_connection *,
1599 struct connection_info *,
1600 enum drbd_notification_type);
1601extern int notify_peer_device_state(struct sk_buff *,
1602 unsigned int,
1603 struct drbd_peer_device *,
1604 struct peer_device_info *,
1605 enum drbd_notification_type);
1606extern void notify_helper(enum drbd_notification_type, struct drbd_device *,
1607 struct drbd_connection *, const char *, int);
1608
1609/*
1610 * inline helper functions
1611 *************************/
1612
1613/* see also page_chain_add and friends in drbd_receiver.c */
1614static inline struct page *page_chain_next(struct page *page)
1615{
1616 return (struct page *)page_private(page);
1617}
1618#define page_chain_for_each(page) \
1619 for (; page && ({ prefetch(page_chain_next(page)); 1; }); \
1620 page = page_chain_next(page))
1621#define page_chain_for_each_safe(page, n) \
1622 for (; page && ({ n = page_chain_next(page); 1; }); page = n)
1623
1624
1625static inline int drbd_peer_req_has_active_page(struct drbd_peer_request *peer_req)
1626{
1627 struct page *page = peer_req->pages;
1628 page_chain_for_each(page) {
1629 if (page_count(page) > 1)
1630 return 1;
1631 }
1632 return 0;
1633}
1634
1635static inline union drbd_state drbd_read_state(struct drbd_device *device)
1636{
1637 struct drbd_resource *resource = device->resource;
1638 union drbd_state rv;
1639
1640 rv.i = device->state.i;
1641 rv.susp = resource->susp;
1642 rv.susp_nod = resource->susp_nod;
1643 rv.susp_fen = resource->susp_fen;
1644
1645 return rv;
1646}
1647
1648enum drbd_force_detach_flags {
1649 DRBD_READ_ERROR,
1650 DRBD_WRITE_ERROR,
1651 DRBD_META_IO_ERROR,
1652 DRBD_FORCE_DETACH,
1653};
1654
1655#define __drbd_chk_io_error(m,f) __drbd_chk_io_error_(m,f, __func__)
1656static inline void __drbd_chk_io_error_(struct drbd_device *device,
1657 enum drbd_force_detach_flags df,
1658 const char *where)
1659{
1660 enum drbd_io_error_p ep;
1661
1662 rcu_read_lock();
1663 ep = rcu_dereference(device->ldev->disk_conf)->on_io_error;
1664 rcu_read_unlock();
1665 switch (ep) {
1666 case EP_PASS_ON: /* FIXME would this be better named "Ignore"? */
1667 if (df == DRBD_READ_ERROR || df == DRBD_WRITE_ERROR) {
1668 if (drbd_ratelimit())
1669 drbd_err(device, "Local IO failed in %s.\n", where);
1670 if (device->state.disk > D_INCONSISTENT)
1671 _drbd_set_state(_NS(device, disk, D_INCONSISTENT), CS_HARD, NULL);
1672 break;
1673 }
1674 fallthrough; /* for DRBD_META_IO_ERROR or DRBD_FORCE_DETACH */
1675 case EP_DETACH:
1676 case EP_CALL_HELPER:
1677 /* Remember whether we saw a READ or WRITE error.
1678 *
1679 * Recovery of the affected area for WRITE failure is covered
1680 * by the activity log.
1681 * READ errors may fall outside that area though. Certain READ
1682 * errors can be "healed" by writing good data to the affected
1683 * blocks, which triggers block re-allocation in lower layers.
1684 *
1685 * If we can not write the bitmap after a READ error,
1686 * we may need to trigger a full sync (see w_go_diskless()).
1687 *
1688 * Force-detach is not really an IO error, but rather a
1689 * desperate measure to try to deal with a completely
1690 * unresponsive lower level IO stack.
1691 * Still it should be treated as a WRITE error.
1692 *
1693 * Meta IO error is always WRITE error:
1694 * we read meta data only once during attach,
1695 * which will fail in case of errors.
1696 */
1697 set_bit(nr: WAS_IO_ERROR, addr: &device->flags);
1698 if (df == DRBD_READ_ERROR)
1699 set_bit(nr: WAS_READ_ERROR, addr: &device->flags);
1700 if (df == DRBD_FORCE_DETACH)
1701 set_bit(nr: FORCE_DETACH, addr: &device->flags);
1702 if (device->state.disk > D_FAILED) {
1703 _drbd_set_state(_NS(device, disk, D_FAILED), CS_HARD, NULL);
1704 drbd_err(device,
1705 "Local IO failed in %s. Detaching...\n", where);
1706 }
1707 break;
1708 }
1709}
1710
1711/**
1712 * drbd_chk_io_error: Handle the on_io_error setting, should be called from all io completion handlers
1713 * @device: DRBD device.
1714 * @error: Error code passed to the IO completion callback
1715 * @forcedetach: Force detach. I.e. the error happened while accessing the meta data
1716 *
1717 * See also drbd_main.c:after_state_ch() if (os.disk > D_FAILED && ns.disk == D_FAILED)
1718 */
1719#define drbd_chk_io_error(m,e,f) drbd_chk_io_error_(m,e,f, __func__)
1720static inline void drbd_chk_io_error_(struct drbd_device *device,
1721 int error, enum drbd_force_detach_flags forcedetach, const char *where)
1722{
1723 if (error) {
1724 unsigned long flags;
1725 spin_lock_irqsave(&device->resource->req_lock, flags);
1726 __drbd_chk_io_error_(device, df: forcedetach, where);
1727 spin_unlock_irqrestore(lock: &device->resource->req_lock, flags);
1728 }
1729}
1730
1731
1732/**
1733 * drbd_md_first_sector() - Returns the first sector number of the meta data area
1734 * @bdev: Meta data block device.
1735 *
1736 * BTW, for internal meta data, this happens to be the maximum capacity
1737 * we could agree upon with our peer node.
1738 */
1739static inline sector_t drbd_md_first_sector(struct drbd_backing_dev *bdev)
1740{
1741 switch (bdev->md.meta_dev_idx) {
1742 case DRBD_MD_INDEX_INTERNAL:
1743 case DRBD_MD_INDEX_FLEX_INT:
1744 return bdev->md.md_offset + bdev->md.bm_offset;
1745 case DRBD_MD_INDEX_FLEX_EXT:
1746 default:
1747 return bdev->md.md_offset;
1748 }
1749}
1750
1751/**
1752 * drbd_md_last_sector() - Return the last sector number of the meta data area
1753 * @bdev: Meta data block device.
1754 */
1755static inline sector_t drbd_md_last_sector(struct drbd_backing_dev *bdev)
1756{
1757 switch (bdev->md.meta_dev_idx) {
1758 case DRBD_MD_INDEX_INTERNAL:
1759 case DRBD_MD_INDEX_FLEX_INT:
1760 return bdev->md.md_offset + MD_4kB_SECT -1;
1761 case DRBD_MD_INDEX_FLEX_EXT:
1762 default:
1763 return bdev->md.md_offset + bdev->md.md_size_sect -1;
1764 }
1765}
1766
1767/* Returns the number of 512 byte sectors of the device */
1768static inline sector_t drbd_get_capacity(struct block_device *bdev)
1769{
1770 return bdev ? bdev_nr_sectors(bdev) : 0;
1771}
1772
1773/**
1774 * drbd_get_max_capacity() - Returns the capacity we announce to out peer
1775 * @bdev: Meta data block device.
1776 *
1777 * returns the capacity we announce to out peer. we clip ourselves at the
1778 * various MAX_SECTORS, because if we don't, current implementation will
1779 * oops sooner or later
1780 */
1781static inline sector_t drbd_get_max_capacity(struct drbd_backing_dev *bdev)
1782{
1783 sector_t s;
1784
1785 switch (bdev->md.meta_dev_idx) {
1786 case DRBD_MD_INDEX_INTERNAL:
1787 case DRBD_MD_INDEX_FLEX_INT:
1788 s = drbd_get_capacity(bdev: bdev->backing_bdev)
1789 ? min_t(sector_t, DRBD_MAX_SECTORS_FLEX,
1790 drbd_md_first_sector(bdev))
1791 : 0;
1792 break;
1793 case DRBD_MD_INDEX_FLEX_EXT:
1794 s = min_t(sector_t, DRBD_MAX_SECTORS_FLEX,
1795 drbd_get_capacity(bdev->backing_bdev));
1796 /* clip at maximum size the meta device can support */
1797 s = min_t(sector_t, s,
1798 BM_EXT_TO_SECT(bdev->md.md_size_sect
1799 - bdev->md.bm_offset));
1800 break;
1801 default:
1802 s = min_t(sector_t, DRBD_MAX_SECTORS,
1803 drbd_get_capacity(bdev->backing_bdev));
1804 }
1805 return s;
1806}
1807
1808/**
1809 * drbd_md_ss() - Return the sector number of our meta data super block
1810 * @bdev: Meta data block device.
1811 */
1812static inline sector_t drbd_md_ss(struct drbd_backing_dev *bdev)
1813{
1814 const int meta_dev_idx = bdev->md.meta_dev_idx;
1815
1816 if (meta_dev_idx == DRBD_MD_INDEX_FLEX_EXT)
1817 return 0;
1818
1819 /* Since drbd08, internal meta data is always "flexible".
1820 * position: last 4k aligned block of 4k size */
1821 if (meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1822 meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)
1823 return (drbd_get_capacity(bdev: bdev->backing_bdev) & ~7ULL) - 8;
1824
1825 /* external, some index; this is the old fixed size layout */
1826 return MD_128MB_SECT * bdev->md.meta_dev_idx;
1827}
1828
1829static inline void
1830drbd_queue_work(struct drbd_work_queue *q, struct drbd_work *w)
1831{
1832 unsigned long flags;
1833 spin_lock_irqsave(&q->q_lock, flags);
1834 list_add_tail(new: &w->list, head: &q->q);
1835 spin_unlock_irqrestore(lock: &q->q_lock, flags);
1836 wake_up(&q->q_wait);
1837}
1838
1839static inline void
1840drbd_queue_work_if_unqueued(struct drbd_work_queue *q, struct drbd_work *w)
1841{
1842 unsigned long flags;
1843 spin_lock_irqsave(&q->q_lock, flags);
1844 if (list_empty_careful(head: &w->list))
1845 list_add_tail(new: &w->list, head: &q->q);
1846 spin_unlock_irqrestore(lock: &q->q_lock, flags);
1847 wake_up(&q->q_wait);
1848}
1849
1850static inline void
1851drbd_device_post_work(struct drbd_device *device, int work_bit)
1852{
1853 if (!test_and_set_bit(nr: work_bit, addr: &device->flags)) {
1854 struct drbd_connection *connection =
1855 first_peer_device(device)->connection;
1856 struct drbd_work_queue *q = &connection->sender_work;
1857 if (!test_and_set_bit(nr: DEVICE_WORK_PENDING, addr: &connection->flags))
1858 wake_up(&q->q_wait);
1859 }
1860}
1861
1862extern void drbd_flush_workqueue(struct drbd_work_queue *work_queue);
1863
1864/* To get the ack_receiver out of the blocking network stack,
1865 * so it can change its sk_rcvtimeo from idle- to ping-timeout,
1866 * and send a ping, we need to send a signal.
1867 * Which signal we send is irrelevant. */
1868static inline void wake_ack_receiver(struct drbd_connection *connection)
1869{
1870 struct task_struct *task = connection->ack_receiver.task;
1871 if (task && get_t_state(thi: &connection->ack_receiver) == RUNNING)
1872 send_sig(SIGXCPU, task, 1);
1873}
1874
1875static inline void request_ping(struct drbd_connection *connection)
1876{
1877 set_bit(nr: SEND_PING, addr: &connection->flags);
1878 wake_ack_receiver(connection);
1879}
1880
1881extern void *conn_prepare_command(struct drbd_connection *, struct drbd_socket *);
1882extern void *drbd_prepare_command(struct drbd_peer_device *, struct drbd_socket *);
1883extern int conn_send_command(struct drbd_connection *, struct drbd_socket *,
1884 enum drbd_packet, unsigned int, void *,
1885 unsigned int);
1886extern int drbd_send_command(struct drbd_peer_device *, struct drbd_socket *,
1887 enum drbd_packet, unsigned int, void *,
1888 unsigned int);
1889
1890extern int drbd_send_ping(struct drbd_connection *connection);
1891extern int drbd_send_ping_ack(struct drbd_connection *connection);
1892extern int drbd_send_state_req(struct drbd_peer_device *, union drbd_state, union drbd_state);
1893extern int conn_send_state_req(struct drbd_connection *, union drbd_state, union drbd_state);
1894
1895static inline void drbd_thread_stop(struct drbd_thread *thi)
1896{
1897 _drbd_thread_stop(thi, restart: false, wait: true);
1898}
1899
1900static inline void drbd_thread_stop_nowait(struct drbd_thread *thi)
1901{
1902 _drbd_thread_stop(thi, restart: false, wait: false);
1903}
1904
1905static inline void drbd_thread_restart_nowait(struct drbd_thread *thi)
1906{
1907 _drbd_thread_stop(thi, restart: true, wait: false);
1908}
1909
1910/* counts how many answer packets packets we expect from our peer,
1911 * for either explicit application requests,
1912 * or implicit barrier packets as necessary.
1913 * increased:
1914 * w_send_barrier
1915 * _req_mod(req, QUEUE_FOR_NET_WRITE or QUEUE_FOR_NET_READ);
1916 * it is much easier and equally valid to count what we queue for the
1917 * worker, even before it actually was queued or send.
1918 * (drbd_make_request_common; recovery path on read io-error)
1919 * decreased:
1920 * got_BarrierAck (respective tl_clear, tl_clear_barrier)
1921 * _req_mod(req, DATA_RECEIVED)
1922 * [from receive_DataReply]
1923 * _req_mod(req, WRITE_ACKED_BY_PEER or RECV_ACKED_BY_PEER or NEG_ACKED)
1924 * [from got_BlockAck (P_WRITE_ACK, P_RECV_ACK)]
1925 * for some reason it is NOT decreased in got_NegAck,
1926 * but in the resulting cleanup code from report_params.
1927 * we should try to remember the reason for that...
1928 * _req_mod(req, SEND_FAILED or SEND_CANCELED)
1929 * _req_mod(req, CONNECTION_LOST_WHILE_PENDING)
1930 * [from tl_clear_barrier]
1931 */
1932static inline void inc_ap_pending(struct drbd_device *device)
1933{
1934 atomic_inc(v: &device->ap_pending_cnt);
1935}
1936
1937#define dec_ap_pending(device) ((void)expect((device), __dec_ap_pending(device) >= 0))
1938static inline int __dec_ap_pending(struct drbd_device *device)
1939{
1940 int ap_pending_cnt = atomic_dec_return(v: &device->ap_pending_cnt);
1941
1942 if (ap_pending_cnt == 0)
1943 wake_up(&device->misc_wait);
1944 return ap_pending_cnt;
1945}
1946
1947/* counts how many resync-related answers we still expect from the peer
1948 * increase decrease
1949 * C_SYNC_TARGET sends P_RS_DATA_REQUEST (and expects P_RS_DATA_REPLY)
1950 * C_SYNC_SOURCE sends P_RS_DATA_REPLY (and expects P_WRITE_ACK with ID_SYNCER)
1951 * (or P_NEG_ACK with ID_SYNCER)
1952 */
1953static inline void inc_rs_pending(struct drbd_peer_device *peer_device)
1954{
1955 atomic_inc(v: &peer_device->device->rs_pending_cnt);
1956}
1957
1958#define dec_rs_pending(peer_device) \
1959 ((void)expect((peer_device), __dec_rs_pending(peer_device) >= 0))
1960static inline int __dec_rs_pending(struct drbd_peer_device *peer_device)
1961{
1962 return atomic_dec_return(v: &peer_device->device->rs_pending_cnt);
1963}
1964
1965/* counts how many answers we still need to send to the peer.
1966 * increased on
1967 * receive_Data unless protocol A;
1968 * we need to send a P_RECV_ACK (proto B)
1969 * or P_WRITE_ACK (proto C)
1970 * receive_RSDataReply (recv_resync_read) we need to send a P_WRITE_ACK
1971 * receive_DataRequest (receive_RSDataRequest) we need to send back P_DATA
1972 * receive_Barrier_* we need to send a P_BARRIER_ACK
1973 */
1974static inline void inc_unacked(struct drbd_device *device)
1975{
1976 atomic_inc(v: &device->unacked_cnt);
1977}
1978
1979#define dec_unacked(device) ((void)expect(device, __dec_unacked(device) >= 0))
1980static inline int __dec_unacked(struct drbd_device *device)
1981{
1982 return atomic_dec_return(v: &device->unacked_cnt);
1983}
1984
1985#define sub_unacked(device, n) ((void)expect(device, __sub_unacked(device) >= 0))
1986static inline int __sub_unacked(struct drbd_device *device, int n)
1987{
1988 return atomic_sub_return(i: n, v: &device->unacked_cnt);
1989}
1990
1991static inline bool is_sync_target_state(enum drbd_conns connection_state)
1992{
1993 return connection_state == C_SYNC_TARGET ||
1994 connection_state == C_PAUSED_SYNC_T;
1995}
1996
1997static inline bool is_sync_source_state(enum drbd_conns connection_state)
1998{
1999 return connection_state == C_SYNC_SOURCE ||
2000 connection_state == C_PAUSED_SYNC_S;
2001}
2002
2003static inline bool is_sync_state(enum drbd_conns connection_state)
2004{
2005 return is_sync_source_state(connection_state) ||
2006 is_sync_target_state(connection_state);
2007}
2008
2009/**
2010 * get_ldev() - Increase the ref count on device->ldev. Returns 0 if there is no ldev
2011 * @_device: DRBD device.
2012 * @_min_state: Minimum device state required for success.
2013 *
2014 * You have to call put_ldev() when finished working with device->ldev.
2015 */
2016#define get_ldev_if_state(_device, _min_state) \
2017 (_get_ldev_if_state((_device), (_min_state)) ? \
2018 ({ __acquire(x); true; }) : false)
2019#define get_ldev(_device) get_ldev_if_state(_device, D_INCONSISTENT)
2020
2021static inline void put_ldev(struct drbd_device *device)
2022{
2023 enum drbd_disk_state disk_state = device->state.disk;
2024 /* We must check the state *before* the atomic_dec becomes visible,
2025 * or we have a theoretical race where someone hitting zero,
2026 * while state still D_FAILED, will then see D_DISKLESS in the
2027 * condition below and calling into destroy, where he must not, yet. */
2028 int i = atomic_dec_return(v: &device->local_cnt);
2029
2030 /* This may be called from some endio handler,
2031 * so we must not sleep here. */
2032
2033 __release(local);
2034 D_ASSERT(device, i >= 0);
2035 if (i == 0) {
2036 if (disk_state == D_DISKLESS)
2037 /* even internal references gone, safe to destroy */
2038 drbd_device_post_work(device, work_bit: DESTROY_DISK);
2039 if (disk_state == D_FAILED)
2040 /* all application IO references gone. */
2041 if (!test_and_set_bit(nr: GOING_DISKLESS, addr: &device->flags))
2042 drbd_device_post_work(device, work_bit: GO_DISKLESS);
2043 wake_up(&device->misc_wait);
2044 }
2045}
2046
2047#ifndef __CHECKER__
2048static inline int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
2049{
2050 int io_allowed;
2051
2052 /* never get a reference while D_DISKLESS */
2053 if (device->state.disk == D_DISKLESS)
2054 return 0;
2055
2056 atomic_inc(v: &device->local_cnt);
2057 io_allowed = (device->state.disk >= mins);
2058 if (!io_allowed)
2059 put_ldev(device);
2060 return io_allowed;
2061}
2062#else
2063extern int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins);
2064#endif
2065
2066/* this throttles on-the-fly application requests
2067 * according to max_buffers settings;
2068 * maybe re-implement using semaphores? */
2069static inline int drbd_get_max_buffers(struct drbd_device *device)
2070{
2071 struct net_conf *nc;
2072 int mxb;
2073
2074 rcu_read_lock();
2075 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
2076 mxb = nc ? nc->max_buffers : 1000000; /* arbitrary limit on open requests */
2077 rcu_read_unlock();
2078
2079 return mxb;
2080}
2081
2082static inline int drbd_state_is_stable(struct drbd_device *device)
2083{
2084 union drbd_dev_state s = device->state;
2085
2086 /* DO NOT add a default clause, we want the compiler to warn us
2087 * for any newly introduced state we may have forgotten to add here */
2088
2089 switch ((enum drbd_conns)s.conn) {
2090 /* new io only accepted when there is no connection, ... */
2091 case C_STANDALONE:
2092 case C_WF_CONNECTION:
2093 /* ... or there is a well established connection. */
2094 case C_CONNECTED:
2095 case C_SYNC_SOURCE:
2096 case C_SYNC_TARGET:
2097 case C_VERIFY_S:
2098 case C_VERIFY_T:
2099 case C_PAUSED_SYNC_S:
2100 case C_PAUSED_SYNC_T:
2101 case C_AHEAD:
2102 case C_BEHIND:
2103 /* transitional states, IO allowed */
2104 case C_DISCONNECTING:
2105 case C_UNCONNECTED:
2106 case C_TIMEOUT:
2107 case C_BROKEN_PIPE:
2108 case C_NETWORK_FAILURE:
2109 case C_PROTOCOL_ERROR:
2110 case C_TEAR_DOWN:
2111 case C_WF_REPORT_PARAMS:
2112 case C_STARTING_SYNC_S:
2113 case C_STARTING_SYNC_T:
2114 break;
2115
2116 /* Allow IO in BM exchange states with new protocols */
2117 case C_WF_BITMAP_S:
2118 if (first_peer_device(device)->connection->agreed_pro_version < 96)
2119 return 0;
2120 break;
2121
2122 /* no new io accepted in these states */
2123 case C_WF_BITMAP_T:
2124 case C_WF_SYNC_UUID:
2125 case C_MASK:
2126 /* not "stable" */
2127 return 0;
2128 }
2129
2130 switch ((enum drbd_disk_state)s.disk) {
2131 case D_DISKLESS:
2132 case D_INCONSISTENT:
2133 case D_OUTDATED:
2134 case D_CONSISTENT:
2135 case D_UP_TO_DATE:
2136 case D_FAILED:
2137 /* disk state is stable as well. */
2138 break;
2139
2140 /* no new io accepted during transitional states */
2141 case D_ATTACHING:
2142 case D_NEGOTIATING:
2143 case D_UNKNOWN:
2144 case D_MASK:
2145 /* not "stable" */
2146 return 0;
2147 }
2148
2149 return 1;
2150}
2151
2152static inline int drbd_suspended(struct drbd_device *device)
2153{
2154 struct drbd_resource *resource = device->resource;
2155
2156 return resource->susp || resource->susp_fen || resource->susp_nod;
2157}
2158
2159static inline bool may_inc_ap_bio(struct drbd_device *device)
2160{
2161 int mxb = drbd_get_max_buffers(device);
2162
2163 if (drbd_suspended(device))
2164 return false;
2165 if (atomic_read(v: &device->suspend_cnt))
2166 return false;
2167
2168 /* to avoid potential deadlock or bitmap corruption,
2169 * in various places, we only allow new application io
2170 * to start during "stable" states. */
2171
2172 /* no new io accepted when attaching or detaching the disk */
2173 if (!drbd_state_is_stable(device))
2174 return false;
2175
2176 /* since some older kernels don't have atomic_add_unless,
2177 * and we are within the spinlock anyways, we have this workaround. */
2178 if (atomic_read(v: &device->ap_bio_cnt) > mxb)
2179 return false;
2180 if (test_bit(BITMAP_IO, &device->flags))
2181 return false;
2182 return true;
2183}
2184
2185static inline bool inc_ap_bio_cond(struct drbd_device *device)
2186{
2187 bool rv = false;
2188
2189 spin_lock_irq(lock: &device->resource->req_lock);
2190 rv = may_inc_ap_bio(device);
2191 if (rv)
2192 atomic_inc(v: &device->ap_bio_cnt);
2193 spin_unlock_irq(lock: &device->resource->req_lock);
2194
2195 return rv;
2196}
2197
2198static inline void inc_ap_bio(struct drbd_device *device)
2199{
2200 /* we wait here
2201 * as long as the device is suspended
2202 * until the bitmap is no longer on the fly during connection
2203 * handshake as long as we would exceed the max_buffer limit.
2204 *
2205 * to avoid races with the reconnect code,
2206 * we need to atomic_inc within the spinlock. */
2207
2208 wait_event(device->misc_wait, inc_ap_bio_cond(device));
2209}
2210
2211static inline void dec_ap_bio(struct drbd_device *device)
2212{
2213 int mxb = drbd_get_max_buffers(device);
2214 int ap_bio = atomic_dec_return(v: &device->ap_bio_cnt);
2215
2216 D_ASSERT(device, ap_bio >= 0);
2217
2218 if (ap_bio == 0 && test_bit(BITMAP_IO, &device->flags)) {
2219 if (!test_and_set_bit(nr: BITMAP_IO_QUEUED, addr: &device->flags))
2220 drbd_queue_work(q: &first_peer_device(device)->
2221 connection->sender_work,
2222 w: &device->bm_io_work.w);
2223 }
2224
2225 /* this currently does wake_up for every dec_ap_bio!
2226 * maybe rather introduce some type of hysteresis?
2227 * e.g. (ap_bio == mxb/2 || ap_bio == 0) ? */
2228 if (ap_bio < mxb)
2229 wake_up(&device->misc_wait);
2230}
2231
2232static inline bool verify_can_do_stop_sector(struct drbd_device *device)
2233{
2234 return first_peer_device(device)->connection->agreed_pro_version >= 97 &&
2235 first_peer_device(device)->connection->agreed_pro_version != 100;
2236}
2237
2238static inline int drbd_set_ed_uuid(struct drbd_device *device, u64 val)
2239{
2240 int changed = device->ed_uuid != val;
2241 device->ed_uuid = val;
2242 return changed;
2243}
2244
2245static inline int drbd_queue_order_type(struct drbd_device *device)
2246{
2247 /* sorry, we currently have no working implementation
2248 * of distributed TCQ stuff */
2249#ifndef QUEUE_ORDERED_NONE
2250#define QUEUE_ORDERED_NONE 0
2251#endif
2252 return QUEUE_ORDERED_NONE;
2253}
2254
2255static inline struct drbd_connection *first_connection(struct drbd_resource *resource)
2256{
2257 return list_first_entry_or_null(&resource->connections,
2258 struct drbd_connection, connections);
2259}
2260
2261#endif
2262

source code of linux/drivers/block/drbd/drbd_int.h