1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 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/genhd.h> 31 #include <linux/idr.h> 32 #include <linux/dynamic_debug.h> 33 #include <net/tcp.h> 34 #include <linux/lru_cache.h> 35 #include <linux/prefetch.h> 36 #include <linux/drbd_genl_api.h> 37 #include <linux/drbd.h> 38 #include "drbd_strings.h" 39 #include "drbd_state.h" 40 #include "drbd_protocol.h" 41 42 #ifdef __CHECKER__ 43 # define __protected_by(x) __attribute__((require_context(x,1,999,"rdwr"))) 44 # define __protected_read_by(x) __attribute__((require_context(x,1,999,"read"))) 45 # define __protected_write_by(x) __attribute__((require_context(x,1,999,"write"))) 46 #else 47 # define __protected_by(x) 48 # define __protected_read_by(x) 49 # define __protected_write_by(x) 50 #endif 51 52 /* shared module parameters, defined in drbd_main.c */ 53 #ifdef CONFIG_DRBD_FAULT_INJECTION 54 extern int drbd_enable_faults; 55 extern int drbd_fault_rate; 56 #endif 57 58 extern unsigned int drbd_minor_count; 59 extern char drbd_usermode_helper[]; 60 extern int drbd_proc_details; 61 62 63 /* This is used to stop/restart our threads. 64 * Cannot use SIGTERM nor SIGKILL, since these 65 * are sent out by init on runlevel changes 66 * I choose SIGHUP for now. 67 */ 68 #define DRBD_SIGKILL SIGHUP 69 70 #define ID_IN_SYNC (4711ULL) 71 #define ID_OUT_OF_SYNC (4712ULL) 72 #define ID_SYNCER (-1ULL) 73 74 #define UUID_NEW_BM_OFFSET ((u64)0x0001000000000000ULL) 75 76 struct drbd_device; 77 struct drbd_connection; 78 79 #define __drbd_printk_device(level, device, fmt, args...) \ 80 dev_printk(level, disk_to_dev((device)->vdisk), fmt, ## args) 81 #define __drbd_printk_peer_device(level, peer_device, fmt, args...) \ 82 dev_printk(level, disk_to_dev((peer_device)->device->vdisk), fmt, ## args) 83 #define __drbd_printk_resource(level, resource, fmt, args...) \ 84 printk(level "drbd %s: " fmt, (resource)->name, ## args) 85 #define __drbd_printk_connection(level, connection, fmt, args...) \ 86 printk(level "drbd %s: " fmt, (connection)->resource->name, ## args) 87 88 void drbd_printk_with_wrong_object_type(void); 89 90 #define __drbd_printk_if_same_type(obj, type, func, level, fmt, args...) \ 91 (__builtin_types_compatible_p(typeof(obj), type) || \ 92 __builtin_types_compatible_p(typeof(obj), const type)), \ 93 func(level, (const type)(obj), fmt, ## args) 94 95 #define drbd_printk(level, obj, fmt, args...) \ 96 __builtin_choose_expr( \ 97 __drbd_printk_if_same_type(obj, struct drbd_device *, \ 98 __drbd_printk_device, level, fmt, ## args), \ 99 __builtin_choose_expr( \ 100 __drbd_printk_if_same_type(obj, struct drbd_resource *, \ 101 __drbd_printk_resource, level, fmt, ## args), \ 102 __builtin_choose_expr( \ 103 __drbd_printk_if_same_type(obj, struct drbd_connection *, \ 104 __drbd_printk_connection, level, fmt, ## args), \ 105 __builtin_choose_expr( \ 106 __drbd_printk_if_same_type(obj, struct drbd_peer_device *, \ 107 __drbd_printk_peer_device, level, fmt, ## args), \ 108 drbd_printk_with_wrong_object_type())))) 109 110 #define drbd_dbg(obj, fmt, args...) \ 111 drbd_printk(KERN_DEBUG, obj, fmt, ## args) 112 #define drbd_alert(obj, fmt, args...) \ 113 drbd_printk(KERN_ALERT, obj, fmt, ## args) 114 #define drbd_err(obj, fmt, args...) \ 115 drbd_printk(KERN_ERR, obj, fmt, ## args) 116 #define drbd_warn(obj, fmt, args...) \ 117 drbd_printk(KERN_WARNING, obj, fmt, ## args) 118 #define drbd_info(obj, fmt, args...) \ 119 drbd_printk(KERN_INFO, obj, fmt, ## args) 120 #define drbd_emerg(obj, fmt, args...) \ 121 drbd_printk(KERN_EMERG, obj, fmt, ## args) 122 123 #define dynamic_drbd_dbg(device, fmt, args...) \ 124 dynamic_dev_dbg(disk_to_dev(device->vdisk), fmt, ## args) 125 126 #define D_ASSERT(device, exp) do { \ 127 if (!(exp)) \ 128 drbd_err(device, "ASSERT( " #exp " ) in %s:%d\n", __FILE__, __LINE__); \ 129 } while (0) 130 131 /** 132 * expect - Make an assertion 133 * 134 * Unlike the assert macro, this macro returns a boolean result. 135 */ 136 #define expect(exp) ({ \ 137 bool _bool = (exp); \ 138 if (!_bool) \ 139 drbd_err(device, "ASSERTION %s FAILED in %s\n", \ 140 #exp, __func__); \ 141 _bool; \ 142 }) 143 144 /* Defines to control fault insertion */ 145 enum { 146 DRBD_FAULT_MD_WR = 0, /* meta data write */ 147 DRBD_FAULT_MD_RD = 1, /* read */ 148 DRBD_FAULT_RS_WR = 2, /* resync */ 149 DRBD_FAULT_RS_RD = 3, 150 DRBD_FAULT_DT_WR = 4, /* data */ 151 DRBD_FAULT_DT_RD = 5, 152 DRBD_FAULT_DT_RA = 6, /* data read ahead */ 153 DRBD_FAULT_BM_ALLOC = 7, /* bitmap allocation */ 154 DRBD_FAULT_AL_EE = 8, /* alloc ee */ 155 DRBD_FAULT_RECEIVE = 9, /* Changes some bytes upon receiving a [rs]data block */ 156 157 DRBD_FAULT_MAX, 158 }; 159 160 extern unsigned int 161 _drbd_insert_fault(struct drbd_device *device, unsigned int type); 162 163 static inline int 164 drbd_insert_fault(struct drbd_device *device, unsigned int type) { 165 #ifdef CONFIG_DRBD_FAULT_INJECTION 166 return drbd_fault_rate && 167 (drbd_enable_faults & (1<<type)) && 168 _drbd_insert_fault(device, type); 169 #else 170 return 0; 171 #endif 172 } 173 174 /* integer division, round _UP_ to the next integer */ 175 #define div_ceil(A, B) ((A)/(B) + ((A)%(B) ? 1 : 0)) 176 /* usual integer division */ 177 #define div_floor(A, B) ((A)/(B)) 178 179 extern struct ratelimit_state drbd_ratelimit_state; 180 extern struct idr drbd_devices; /* RCU, updates: genl_lock() */ 181 extern struct list_head drbd_resources; /* RCU, updates: genl_lock() */ 182 183 extern const char *cmdname(enum drbd_packet cmd); 184 185 /* for sending/receiving the bitmap, 186 * possibly in some encoding scheme */ 187 struct bm_xfer_ctx { 188 /* "const" 189 * stores total bits and long words 190 * of the bitmap, so we don't need to 191 * call the accessor functions over and again. */ 192 unsigned long bm_bits; 193 unsigned long bm_words; 194 /* during xfer, current position within the bitmap */ 195 unsigned long bit_offset; 196 unsigned long word_offset; 197 198 /* statistics; index: (h->command == P_BITMAP) */ 199 unsigned packets[2]; 200 unsigned bytes[2]; 201 }; 202 203 extern void INFO_bm_xfer_stats(struct drbd_device *device, 204 const char *direction, struct bm_xfer_ctx *c); 205 206 static inline void bm_xfer_ctx_bit_to_word_offset(struct bm_xfer_ctx *c) 207 { 208 /* word_offset counts "native long words" (32 or 64 bit), 209 * aligned at 64 bit. 210 * Encoded packet may end at an unaligned bit offset. 211 * In case a fallback clear text packet is transmitted in 212 * between, we adjust this offset back to the last 64bit 213 * aligned "native long word", which makes coding and decoding 214 * the plain text bitmap much more convenient. */ 215 #if BITS_PER_LONG == 64 216 c->word_offset = c->bit_offset >> 6; 217 #elif BITS_PER_LONG == 32 218 c->word_offset = c->bit_offset >> 5; 219 c->word_offset &= ~(1UL); 220 #else 221 # error "unsupported BITS_PER_LONG" 222 #endif 223 } 224 225 extern unsigned int drbd_header_size(struct drbd_connection *connection); 226 227 /**********************************************************************/ 228 enum drbd_thread_state { 229 NONE, 230 RUNNING, 231 EXITING, 232 RESTARTING 233 }; 234 235 struct drbd_thread { 236 spinlock_t t_lock; 237 struct task_struct *task; 238 struct completion stop; 239 enum drbd_thread_state t_state; 240 int (*function) (struct drbd_thread *); 241 struct drbd_resource *resource; 242 struct drbd_connection *connection; 243 int reset_cpu_mask; 244 const char *name; 245 }; 246 247 static inline enum drbd_thread_state get_t_state(struct drbd_thread *thi) 248 { 249 /* THINK testing the t_state seems to be uncritical in all cases 250 * (but thread_{start,stop}), so we can read it *without* the lock. 251 * --lge */ 252 253 smp_rmb(); 254 return thi->t_state; 255 } 256 257 struct drbd_work { 258 struct list_head list; 259 int (*cb)(struct drbd_work *, int cancel); 260 }; 261 262 struct drbd_device_work { 263 struct drbd_work w; 264 struct drbd_device *device; 265 }; 266 267 #include "drbd_interval.h" 268 269 extern int drbd_wait_misc(struct drbd_device *, struct drbd_interval *); 270 271 extern void lock_all_resources(void); 272 extern void unlock_all_resources(void); 273 274 struct drbd_request { 275 struct drbd_work w; 276 struct drbd_device *device; 277 278 /* if local IO is not allowed, will be NULL. 279 * if local IO _is_ allowed, holds the locally submitted bio clone, 280 * or, after local IO completion, the ERR_PTR(error). 281 * see drbd_request_endio(). */ 282 struct bio *private_bio; 283 284 struct drbd_interval i; 285 286 /* epoch: used to check on "completion" whether this req was in 287 * the current epoch, and we therefore have to close it, 288 * causing a p_barrier packet to be send, starting a new epoch. 289 * 290 * This corresponds to "barrier" in struct p_barrier[_ack], 291 * and to "barrier_nr" in struct drbd_epoch (and various 292 * comments/function parameters/local variable names). 293 */ 294 unsigned int epoch; 295 296 struct list_head tl_requests; /* ring list in the transfer log */ 297 struct bio *master_bio; /* master bio pointer */ 298 299 /* see struct drbd_device */ 300 struct list_head req_pending_master_completion; 301 struct list_head req_pending_local; 302 303 /* for generic IO accounting */ 304 unsigned long start_jif; 305 306 /* for DRBD internal statistics */ 307 308 /* Minimal set of time stamps to determine if we wait for activity log 309 * transactions, local disk or peer. 32 bit "jiffies" are good enough, 310 * we don't expect a DRBD request to be stalled for several month. 311 */ 312 313 /* before actual request processing */ 314 unsigned long in_actlog_jif; 315 316 /* local disk */ 317 unsigned long pre_submit_jif; 318 319 /* per connection */ 320 unsigned long pre_send_jif; 321 unsigned long acked_jif; 322 unsigned long net_done_jif; 323 324 /* Possibly even more detail to track each phase: 325 * master_completion_jif 326 * how long did it take to complete the master bio 327 * (application visible latency) 328 * allocated_jif 329 * how long the master bio was blocked until we finally allocated 330 * a tracking struct 331 * in_actlog_jif 332 * how long did we wait for activity log transactions 333 * 334 * net_queued_jif 335 * when did we finally queue it for sending 336 * pre_send_jif 337 * when did we start sending it 338 * post_send_jif 339 * how long did we block in the network stack trying to send it 340 * acked_jif 341 * when did we receive (or fake, in protocol A) a remote ACK 342 * net_done_jif 343 * when did we receive final acknowledgement (P_BARRIER_ACK), 344 * or decide, e.g. on connection loss, that we do no longer expect 345 * anything from this peer for this request. 346 * 347 * pre_submit_jif 348 * post_sub_jif 349 * when did we start submiting to the lower level device, 350 * and how long did we block in that submit function 351 * local_completion_jif 352 * how long did it take the lower level device to complete this request 353 */ 354 355 356 /* once it hits 0, we may complete the master_bio */ 357 atomic_t completion_ref; 358 /* once it hits 0, we may destroy this drbd_request object */ 359 struct kref kref; 360 361 unsigned rq_state; /* see comments above _req_mod() */ 362 }; 363 364 struct drbd_epoch { 365 struct drbd_connection *connection; 366 struct list_head list; 367 unsigned int barrier_nr; 368 atomic_t epoch_size; /* increased on every request added. */ 369 atomic_t active; /* increased on every req. added, and dec on every finished. */ 370 unsigned long flags; 371 }; 372 373 /* Prototype declaration of function defined in drbd_receiver.c */ 374 int drbdd_init(struct drbd_thread *); 375 int drbd_asender(struct drbd_thread *); 376 377 /* drbd_epoch flag bits */ 378 enum { 379 DE_HAVE_BARRIER_NUMBER, 380 }; 381 382 enum epoch_event { 383 EV_PUT, 384 EV_GOT_BARRIER_NR, 385 EV_BECAME_LAST, 386 EV_CLEANUP = 32, /* used as flag */ 387 }; 388 389 struct digest_info { 390 int digest_size; 391 void *digest; 392 }; 393 394 struct drbd_peer_request { 395 struct drbd_work w; 396 struct drbd_peer_device *peer_device; 397 struct drbd_epoch *epoch; /* for writes */ 398 struct page *pages; 399 atomic_t pending_bios; 400 struct drbd_interval i; 401 /* see comments on ee flag bits below */ 402 unsigned long flags; 403 unsigned long submit_jif; 404 union { 405 u64 block_id; 406 struct digest_info *digest; 407 }; 408 }; 409 410 /* ee flag bits. 411 * While corresponding bios are in flight, the only modification will be 412 * set_bit WAS_ERROR, which has to be atomic. 413 * If no bios are in flight yet, or all have been completed, 414 * non-atomic modification to ee->flags is ok. 415 */ 416 enum { 417 __EE_CALL_AL_COMPLETE_IO, 418 __EE_MAY_SET_IN_SYNC, 419 420 /* is this a TRIM aka REQ_OP_DISCARD? */ 421 __EE_TRIM, 422 /* explicit zero-out requested, or 423 * our lower level cannot handle trim, 424 * and we want to fall back to zeroout instead */ 425 __EE_ZEROOUT, 426 427 /* In case a barrier failed, 428 * we need to resubmit without the barrier flag. */ 429 __EE_RESUBMITTED, 430 431 /* we may have several bios per peer request. 432 * if any of those fail, we set this flag atomically 433 * from the endio callback */ 434 __EE_WAS_ERROR, 435 436 /* This ee has a pointer to a digest instead of a block id */ 437 __EE_HAS_DIGEST, 438 439 /* Conflicting local requests need to be restarted after this request */ 440 __EE_RESTART_REQUESTS, 441 442 /* The peer wants a write ACK for this (wire proto C) */ 443 __EE_SEND_WRITE_ACK, 444 445 /* Is set when net_conf had two_primaries set while creating this peer_req */ 446 __EE_IN_INTERVAL_TREE, 447 448 /* for debugfs: */ 449 /* has this been submitted, or does it still wait for something else? */ 450 __EE_SUBMITTED, 451 452 /* this is/was a write request */ 453 __EE_WRITE, 454 455 /* this is/was a write same request */ 456 __EE_WRITE_SAME, 457 458 /* this originates from application on peer 459 * (not some resync or verify or other DRBD internal request) */ 460 __EE_APPLICATION, 461 462 /* If it contains only 0 bytes, send back P_RS_DEALLOCATED */ 463 __EE_RS_THIN_REQ, 464 }; 465 #define EE_CALL_AL_COMPLETE_IO (1<<__EE_CALL_AL_COMPLETE_IO) 466 #define EE_MAY_SET_IN_SYNC (1<<__EE_MAY_SET_IN_SYNC) 467 #define EE_TRIM (1<<__EE_TRIM) 468 #define EE_ZEROOUT (1<<__EE_ZEROOUT) 469 #define EE_RESUBMITTED (1<<__EE_RESUBMITTED) 470 #define EE_WAS_ERROR (1<<__EE_WAS_ERROR) 471 #define EE_HAS_DIGEST (1<<__EE_HAS_DIGEST) 472 #define EE_RESTART_REQUESTS (1<<__EE_RESTART_REQUESTS) 473 #define EE_SEND_WRITE_ACK (1<<__EE_SEND_WRITE_ACK) 474 #define EE_IN_INTERVAL_TREE (1<<__EE_IN_INTERVAL_TREE) 475 #define EE_SUBMITTED (1<<__EE_SUBMITTED) 476 #define EE_WRITE (1<<__EE_WRITE) 477 #define EE_WRITE_SAME (1<<__EE_WRITE_SAME) 478 #define EE_APPLICATION (1<<__EE_APPLICATION) 479 #define EE_RS_THIN_REQ (1<<__EE_RS_THIN_REQ) 480 481 /* flag bits per device */ 482 enum { 483 UNPLUG_REMOTE, /* sending a "UnplugRemote" could help */ 484 MD_DIRTY, /* current uuids and flags not yet on disk */ 485 USE_DEGR_WFC_T, /* degr-wfc-timeout instead of wfc-timeout. */ 486 CL_ST_CHG_SUCCESS, 487 CL_ST_CHG_FAIL, 488 CRASHED_PRIMARY, /* This node was a crashed primary. 489 * Gets cleared when the state.conn 490 * goes into C_CONNECTED state. */ 491 CONSIDER_RESYNC, 492 493 MD_NO_FUA, /* Users wants us to not use FUA/FLUSH on meta data dev */ 494 495 BITMAP_IO, /* suspend application io; 496 once no more io in flight, start bitmap io */ 497 BITMAP_IO_QUEUED, /* Started bitmap IO */ 498 WAS_IO_ERROR, /* Local disk failed, returned IO error */ 499 WAS_READ_ERROR, /* Local disk READ failed (set additionally to the above) */ 500 FORCE_DETACH, /* Force-detach from local disk, aborting any pending local IO */ 501 RESYNC_AFTER_NEG, /* Resync after online grow after the attach&negotiate finished. */ 502 RESIZE_PENDING, /* Size change detected locally, waiting for the response from 503 * the peer, if it changed there as well. */ 504 NEW_CUR_UUID, /* Create new current UUID when thawing IO */ 505 AL_SUSPENDED, /* Activity logging is currently suspended. */ 506 AHEAD_TO_SYNC_SOURCE, /* Ahead -> SyncSource queued */ 507 B_RS_H_DONE, /* Before resync handler done (already executed) */ 508 DISCARD_MY_DATA, /* discard_my_data flag per volume */ 509 READ_BALANCE_RR, 510 511 FLUSH_PENDING, /* if set, device->flush_jif is when we submitted that flush 512 * from drbd_flush_after_epoch() */ 513 514 /* cleared only after backing device related structures have been destroyed. */ 515 GOING_DISKLESS, /* Disk is being detached, because of io-error, or admin request. */ 516 517 /* to be used in drbd_device_post_work() */ 518 GO_DISKLESS, /* tell worker to schedule cleanup before detach */ 519 DESTROY_DISK, /* tell worker to close backing devices and destroy related structures. */ 520 MD_SYNC, /* tell worker to call drbd_md_sync() */ 521 RS_START, /* tell worker to start resync/OV */ 522 RS_PROGRESS, /* tell worker that resync made significant progress */ 523 RS_DONE, /* tell worker that resync is done */ 524 }; 525 526 struct drbd_bitmap; /* opaque for drbd_device */ 527 528 /* definition of bits in bm_flags to be used in drbd_bm_lock 529 * and drbd_bitmap_io and friends. */ 530 enum bm_flag { 531 /* currently locked for bulk operation */ 532 BM_LOCKED_MASK = 0xf, 533 534 /* in detail, that is: */ 535 BM_DONT_CLEAR = 0x1, 536 BM_DONT_SET = 0x2, 537 BM_DONT_TEST = 0x4, 538 539 /* so we can mark it locked for bulk operation, 540 * and still allow all non-bulk operations */ 541 BM_IS_LOCKED = 0x8, 542 543 /* (test bit, count bit) allowed (common case) */ 544 BM_LOCKED_TEST_ALLOWED = BM_DONT_CLEAR | BM_DONT_SET | BM_IS_LOCKED, 545 546 /* testing bits, as well as setting new bits allowed, but clearing bits 547 * would be unexpected. Used during bitmap receive. Setting new bits 548 * requires sending of "out-of-sync" information, though. */ 549 BM_LOCKED_SET_ALLOWED = BM_DONT_CLEAR | BM_IS_LOCKED, 550 551 /* for drbd_bm_write_copy_pages, everything is allowed, 552 * only concurrent bulk operations are locked out. */ 553 BM_LOCKED_CHANGE_ALLOWED = BM_IS_LOCKED, 554 }; 555 556 struct drbd_work_queue { 557 struct list_head q; 558 spinlock_t q_lock; /* to protect the list. */ 559 wait_queue_head_t q_wait; 560 }; 561 562 struct drbd_socket { 563 struct mutex mutex; 564 struct socket *socket; 565 /* this way we get our 566 * send/receive buffers off the stack */ 567 void *sbuf; 568 void *rbuf; 569 }; 570 571 struct drbd_md { 572 u64 md_offset; /* sector offset to 'super' block */ 573 574 u64 la_size_sect; /* last agreed size, unit sectors */ 575 spinlock_t uuid_lock; 576 u64 uuid[UI_SIZE]; 577 u64 device_uuid; 578 u32 flags; 579 u32 md_size_sect; 580 581 s32 al_offset; /* signed relative sector offset to activity log */ 582 s32 bm_offset; /* signed relative sector offset to bitmap */ 583 584 /* cached value of bdev->disk_conf->meta_dev_idx (see below) */ 585 s32 meta_dev_idx; 586 587 /* see al_tr_number_to_on_disk_sector() */ 588 u32 al_stripes; 589 u32 al_stripe_size_4k; 590 u32 al_size_4k; /* cached product of the above */ 591 }; 592 593 struct drbd_backing_dev { 594 struct block_device *backing_bdev; 595 struct block_device *md_bdev; 596 struct drbd_md md; 597 struct disk_conf *disk_conf; /* RCU, for updates: resource->conf_update */ 598 sector_t known_size; /* last known size of that backing device */ 599 }; 600 601 struct drbd_md_io { 602 struct page *page; 603 unsigned long start_jif; /* last call to drbd_md_get_buffer */ 604 unsigned long submit_jif; /* last _drbd_md_sync_page_io() submit */ 605 const char *current_use; 606 atomic_t in_use; 607 unsigned int done; 608 int error; 609 }; 610 611 struct bm_io_work { 612 struct drbd_work w; 613 char *why; 614 enum bm_flag flags; 615 int (*io_fn)(struct drbd_device *device); 616 void (*done)(struct drbd_device *device, int rv); 617 }; 618 619 struct fifo_buffer { 620 unsigned int head_index; 621 unsigned int size; 622 int total; /* sum of all values */ 623 int values[0]; 624 }; 625 extern struct fifo_buffer *fifo_alloc(int fifo_size); 626 627 /* flag bits per connection */ 628 enum { 629 NET_CONGESTED, /* The data socket is congested */ 630 RESOLVE_CONFLICTS, /* Set on one node, cleared on the peer! */ 631 SEND_PING, 632 GOT_PING_ACK, /* set when we receive a ping_ack packet, ping_wait gets woken */ 633 CONN_WD_ST_CHG_REQ, /* A cluster wide state change on the connection is active */ 634 CONN_WD_ST_CHG_OKAY, 635 CONN_WD_ST_CHG_FAIL, 636 CONN_DRY_RUN, /* Expect disconnect after resync handshake. */ 637 CREATE_BARRIER, /* next P_DATA is preceded by a P_BARRIER */ 638 STATE_SENT, /* Do not change state/UUIDs while this is set */ 639 CALLBACK_PENDING, /* Whether we have a call_usermodehelper(, UMH_WAIT_PROC) 640 * pending, from drbd worker context. 641 * If set, bdi_write_congested() returns true, 642 * so shrink_page_list() would not recurse into, 643 * and potentially deadlock on, this drbd worker. 644 */ 645 DISCONNECT_SENT, 646 647 DEVICE_WORK_PENDING, /* tell worker that some device has pending work */ 648 }; 649 650 enum which_state { NOW, OLD = NOW, NEW }; 651 652 struct drbd_resource { 653 char *name; 654 #ifdef CONFIG_DEBUG_FS 655 struct dentry *debugfs_res; 656 struct dentry *debugfs_res_volumes; 657 struct dentry *debugfs_res_connections; 658 struct dentry *debugfs_res_in_flight_summary; 659 #endif 660 struct kref kref; 661 struct idr devices; /* volume number to device mapping */ 662 struct list_head connections; 663 struct list_head resources; 664 struct res_opts res_opts; 665 struct mutex conf_update; /* mutex for ready-copy-update of net_conf and disk_conf */ 666 struct mutex adm_mutex; /* mutex to serialize administrative requests */ 667 spinlock_t req_lock; 668 669 unsigned susp:1; /* IO suspended by user */ 670 unsigned susp_nod:1; /* IO suspended because no data */ 671 unsigned susp_fen:1; /* IO suspended because fence peer handler runs */ 672 673 enum write_ordering_e write_ordering; 674 675 cpumask_var_t cpu_mask; 676 }; 677 678 struct drbd_thread_timing_details 679 { 680 unsigned long start_jif; 681 void *cb_addr; 682 const char *caller_fn; 683 unsigned int line; 684 unsigned int cb_nr; 685 }; 686 687 struct drbd_connection { 688 struct list_head connections; 689 struct drbd_resource *resource; 690 #ifdef CONFIG_DEBUG_FS 691 struct dentry *debugfs_conn; 692 struct dentry *debugfs_conn_callback_history; 693 struct dentry *debugfs_conn_oldest_requests; 694 #endif 695 struct kref kref; 696 struct idr peer_devices; /* volume number to peer device mapping */ 697 enum drbd_conns cstate; /* Only C_STANDALONE to C_WF_REPORT_PARAMS */ 698 struct mutex cstate_mutex; /* Protects graceful disconnects */ 699 unsigned int connect_cnt; /* Inc each time a connection is established */ 700 701 unsigned long flags; 702 struct net_conf *net_conf; /* content protected by rcu */ 703 wait_queue_head_t ping_wait; /* Woken upon reception of a ping, and a state change */ 704 705 struct sockaddr_storage my_addr; 706 int my_addr_len; 707 struct sockaddr_storage peer_addr; 708 int peer_addr_len; 709 710 struct drbd_socket data; /* data/barrier/cstate/parameter packets */ 711 struct drbd_socket meta; /* ping/ack (metadata) packets */ 712 int agreed_pro_version; /* actually used protocol version */ 713 u32 agreed_features; 714 unsigned long last_received; /* in jiffies, either socket */ 715 unsigned int ko_count; 716 717 struct list_head transfer_log; /* all requests not yet fully processed */ 718 719 struct crypto_shash *cram_hmac_tfm; 720 struct crypto_shash *integrity_tfm; /* checksums we compute, updates protected by connection->data->mutex */ 721 struct crypto_shash *peer_integrity_tfm; /* checksums we verify, only accessed from receiver thread */ 722 struct crypto_shash *csums_tfm; 723 struct crypto_shash *verify_tfm; 724 void *int_dig_in; 725 void *int_dig_vv; 726 727 /* receiver side */ 728 struct drbd_epoch *current_epoch; 729 spinlock_t epoch_lock; 730 unsigned int epochs; 731 atomic_t current_tle_nr; /* transfer log epoch number */ 732 unsigned current_tle_writes; /* writes seen within this tl epoch */ 733 734 unsigned long last_reconnect_jif; 735 /* empty member on older kernels without blk_start_plug() */ 736 struct blk_plug receiver_plug; 737 struct drbd_thread receiver; 738 struct drbd_thread worker; 739 struct drbd_thread ack_receiver; 740 struct workqueue_struct *ack_sender; 741 742 /* cached pointers, 743 * so we can look up the oldest pending requests more quickly. 744 * protected by resource->req_lock */ 745 struct drbd_request *req_next; /* DRBD 9: todo.req_next */ 746 struct drbd_request *req_ack_pending; 747 struct drbd_request *req_not_net_done; 748 749 /* sender side */ 750 struct drbd_work_queue sender_work; 751 752 #define DRBD_THREAD_DETAILS_HIST 16 753 unsigned int w_cb_nr; /* keeps counting up */ 754 unsigned int r_cb_nr; /* keeps counting up */ 755 struct drbd_thread_timing_details w_timing_details[DRBD_THREAD_DETAILS_HIST]; 756 struct drbd_thread_timing_details r_timing_details[DRBD_THREAD_DETAILS_HIST]; 757 758 struct { 759 unsigned long last_sent_barrier_jif; 760 761 /* whether this sender thread 762 * has processed a single write yet. */ 763 bool seen_any_write_yet; 764 765 /* Which barrier number to send with the next P_BARRIER */ 766 int current_epoch_nr; 767 768 /* how many write requests have been sent 769 * with req->epoch == current_epoch_nr. 770 * If none, no P_BARRIER will be sent. */ 771 unsigned current_epoch_writes; 772 } send; 773 }; 774 775 static inline bool has_net_conf(struct drbd_connection *connection) 776 { 777 bool has_net_conf; 778 779 rcu_read_lock(); 780 has_net_conf = rcu_dereference(connection->net_conf); 781 rcu_read_unlock(); 782 783 return has_net_conf; 784 } 785 786 void __update_timing_details( 787 struct drbd_thread_timing_details *tdp, 788 unsigned int *cb_nr, 789 void *cb, 790 const char *fn, const unsigned int line); 791 792 #define update_worker_timing_details(c, cb) \ 793 __update_timing_details(c->w_timing_details, &c->w_cb_nr, cb, __func__ , __LINE__ ) 794 #define update_receiver_timing_details(c, cb) \ 795 __update_timing_details(c->r_timing_details, &c->r_cb_nr, cb, __func__ , __LINE__ ) 796 797 struct submit_worker { 798 struct workqueue_struct *wq; 799 struct work_struct worker; 800 801 /* protected by ..->resource->req_lock */ 802 struct list_head writes; 803 }; 804 805 struct drbd_peer_device { 806 struct list_head peer_devices; 807 struct drbd_device *device; 808 struct drbd_connection *connection; 809 struct work_struct send_acks_work; 810 #ifdef CONFIG_DEBUG_FS 811 struct dentry *debugfs_peer_dev; 812 #endif 813 }; 814 815 struct drbd_device { 816 struct drbd_resource *resource; 817 struct list_head peer_devices; 818 struct list_head pending_bitmap_io; 819 820 unsigned long flush_jif; 821 #ifdef CONFIG_DEBUG_FS 822 struct dentry *debugfs_minor; 823 struct dentry *debugfs_vol; 824 struct dentry *debugfs_vol_oldest_requests; 825 struct dentry *debugfs_vol_act_log_extents; 826 struct dentry *debugfs_vol_resync_extents; 827 struct dentry *debugfs_vol_data_gen_id; 828 struct dentry *debugfs_vol_ed_gen_id; 829 #endif 830 831 unsigned int vnr; /* volume number within the connection */ 832 unsigned int minor; /* device minor number */ 833 834 struct kref kref; 835 836 /* things that are stored as / read from meta data on disk */ 837 unsigned long flags; 838 839 /* configured by drbdsetup */ 840 struct drbd_backing_dev *ldev __protected_by(local); 841 842 sector_t p_size; /* partner's disk size */ 843 struct request_queue *rq_queue; 844 struct block_device *this_bdev; 845 struct gendisk *vdisk; 846 847 unsigned long last_reattach_jif; 848 struct drbd_work resync_work; 849 struct drbd_work unplug_work; 850 struct timer_list resync_timer; 851 struct timer_list md_sync_timer; 852 struct timer_list start_resync_timer; 853 struct timer_list request_timer; 854 855 /* Used after attach while negotiating new disk state. */ 856 union drbd_state new_state_tmp; 857 858 union drbd_dev_state state; 859 wait_queue_head_t misc_wait; 860 wait_queue_head_t state_wait; /* upon each state change. */ 861 unsigned int send_cnt; 862 unsigned int recv_cnt; 863 unsigned int read_cnt; 864 unsigned int writ_cnt; 865 unsigned int al_writ_cnt; 866 unsigned int bm_writ_cnt; 867 atomic_t ap_bio_cnt; /* Requests we need to complete */ 868 atomic_t ap_actlog_cnt; /* Requests waiting for activity log */ 869 atomic_t ap_pending_cnt; /* AP data packets on the wire, ack expected */ 870 atomic_t rs_pending_cnt; /* RS request/data packets on the wire */ 871 atomic_t unacked_cnt; /* Need to send replies for */ 872 atomic_t local_cnt; /* Waiting for local completion */ 873 atomic_t suspend_cnt; 874 875 /* Interval tree of pending local requests */ 876 struct rb_root read_requests; 877 struct rb_root write_requests; 878 879 /* for statistics and timeouts */ 880 /* [0] read, [1] write */ 881 struct list_head pending_master_completion[2]; 882 struct list_head pending_completion[2]; 883 884 /* use checksums for *this* resync */ 885 bool use_csums; 886 /* blocks to resync in this run [unit BM_BLOCK_SIZE] */ 887 unsigned long rs_total; 888 /* number of resync blocks that failed in this run */ 889 unsigned long rs_failed; 890 /* Syncer's start time [unit jiffies] */ 891 unsigned long rs_start; 892 /* cumulated time in PausedSyncX state [unit jiffies] */ 893 unsigned long rs_paused; 894 /* skipped because csum was equal [unit BM_BLOCK_SIZE] */ 895 unsigned long rs_same_csum; 896 #define DRBD_SYNC_MARKS 8 897 #define DRBD_SYNC_MARK_STEP (3*HZ) 898 /* block not up-to-date at mark [unit BM_BLOCK_SIZE] */ 899 unsigned long rs_mark_left[DRBD_SYNC_MARKS]; 900 /* marks's time [unit jiffies] */ 901 unsigned long rs_mark_time[DRBD_SYNC_MARKS]; 902 /* current index into rs_mark_{left,time} */ 903 int rs_last_mark; 904 unsigned long rs_last_bcast; /* [unit jiffies] */ 905 906 /* where does the admin want us to start? (sector) */ 907 sector_t ov_start_sector; 908 sector_t ov_stop_sector; 909 /* where are we now? (sector) */ 910 sector_t ov_position; 911 /* Start sector of out of sync range (to merge printk reporting). */ 912 sector_t ov_last_oos_start; 913 /* size of out-of-sync range in sectors. */ 914 sector_t ov_last_oos_size; 915 unsigned long ov_left; /* in bits */ 916 917 struct drbd_bitmap *bitmap; 918 unsigned long bm_resync_fo; /* bit offset for drbd_bm_find_next */ 919 920 /* Used to track operations of resync... */ 921 struct lru_cache *resync; 922 /* Number of locked elements in resync LRU */ 923 unsigned int resync_locked; 924 /* resync extent number waiting for application requests */ 925 unsigned int resync_wenr; 926 927 int open_cnt; 928 u64 *p_uuid; 929 930 struct list_head active_ee; /* IO in progress (P_DATA gets written to disk) */ 931 struct list_head sync_ee; /* IO in progress (P_RS_DATA_REPLY gets written to disk) */ 932 struct list_head done_ee; /* need to send P_WRITE_ACK */ 933 struct list_head read_ee; /* [RS]P_DATA_REQUEST being read */ 934 struct list_head net_ee; /* zero-copy network send in progress */ 935 936 int next_barrier_nr; 937 struct list_head resync_reads; 938 atomic_t pp_in_use; /* allocated from page pool */ 939 atomic_t pp_in_use_by_net; /* sendpage()d, still referenced by tcp */ 940 wait_queue_head_t ee_wait; 941 struct drbd_md_io md_io; 942 spinlock_t al_lock; 943 wait_queue_head_t al_wait; 944 struct lru_cache *act_log; /* activity log */ 945 unsigned int al_tr_number; 946 int al_tr_cycle; 947 wait_queue_head_t seq_wait; 948 atomic_t packet_seq; 949 unsigned int peer_seq; 950 spinlock_t peer_seq_lock; 951 unsigned long comm_bm_set; /* communicated number of set bits. */ 952 struct bm_io_work bm_io_work; 953 u64 ed_uuid; /* UUID of the exposed data */ 954 struct mutex own_state_mutex; 955 struct mutex *state_mutex; /* either own_state_mutex or first_peer_device(device)->connection->cstate_mutex */ 956 char congestion_reason; /* Why we where congested... */ 957 atomic_t rs_sect_in; /* for incoming resync data rate, SyncTarget */ 958 atomic_t rs_sect_ev; /* for submitted resync data rate, both */ 959 int rs_last_sect_ev; /* counter to compare with */ 960 int rs_last_events; /* counter of read or write "events" (unit sectors) 961 * on the lower level device when we last looked. */ 962 int c_sync_rate; /* current resync rate after syncer throttle magic */ 963 struct fifo_buffer *rs_plan_s; /* correction values of resync planer (RCU, connection->conn_update) */ 964 int rs_in_flight; /* resync sectors in flight (to proxy, in proxy and from proxy) */ 965 atomic_t ap_in_flight; /* App sectors in flight (waiting for ack) */ 966 unsigned int peer_max_bio_size; 967 unsigned int local_max_bio_size; 968 969 /* any requests that would block in drbd_make_request() 970 * are deferred to this single-threaded work queue */ 971 struct submit_worker submit; 972 }; 973 974 struct drbd_bm_aio_ctx { 975 struct drbd_device *device; 976 struct list_head list; /* on device->pending_bitmap_io */; 977 unsigned long start_jif; 978 atomic_t in_flight; 979 unsigned int done; 980 unsigned flags; 981 #define BM_AIO_COPY_PAGES 1 982 #define BM_AIO_WRITE_HINTED 2 983 #define BM_AIO_WRITE_ALL_PAGES 4 984 #define BM_AIO_READ 8 985 int error; 986 struct kref kref; 987 }; 988 989 struct drbd_config_context { 990 /* assigned from drbd_genlmsghdr */ 991 unsigned int minor; 992 /* assigned from request attributes, if present */ 993 unsigned int volume; 994 #define VOLUME_UNSPECIFIED (-1U) 995 /* pointer into the request skb, 996 * limited lifetime! */ 997 char *resource_name; 998 struct nlattr *my_addr; 999 struct nlattr *peer_addr; 1000 1001 /* reply buffer */ 1002 struct sk_buff *reply_skb; 1003 /* pointer into reply buffer */ 1004 struct drbd_genlmsghdr *reply_dh; 1005 /* resolved from attributes, if possible */ 1006 struct drbd_device *device; 1007 struct drbd_resource *resource; 1008 struct drbd_connection *connection; 1009 }; 1010 1011 static inline struct drbd_device *minor_to_device(unsigned int minor) 1012 { 1013 return (struct drbd_device *)idr_find(&drbd_devices, minor); 1014 } 1015 1016 static inline struct drbd_peer_device *first_peer_device(struct drbd_device *device) 1017 { 1018 return list_first_entry_or_null(&device->peer_devices, struct drbd_peer_device, peer_devices); 1019 } 1020 1021 static inline struct drbd_peer_device * 1022 conn_peer_device(struct drbd_connection *connection, int volume_number) 1023 { 1024 return idr_find(&connection->peer_devices, volume_number); 1025 } 1026 1027 #define for_each_resource(resource, _resources) \ 1028 list_for_each_entry(resource, _resources, resources) 1029 1030 #define for_each_resource_rcu(resource, _resources) \ 1031 list_for_each_entry_rcu(resource, _resources, resources) 1032 1033 #define for_each_resource_safe(resource, tmp, _resources) \ 1034 list_for_each_entry_safe(resource, tmp, _resources, resources) 1035 1036 #define for_each_connection(connection, resource) \ 1037 list_for_each_entry(connection, &resource->connections, connections) 1038 1039 #define for_each_connection_rcu(connection, resource) \ 1040 list_for_each_entry_rcu(connection, &resource->connections, connections) 1041 1042 #define for_each_connection_safe(connection, tmp, resource) \ 1043 list_for_each_entry_safe(connection, tmp, &resource->connections, connections) 1044 1045 #define for_each_peer_device(peer_device, device) \ 1046 list_for_each_entry(peer_device, &device->peer_devices, peer_devices) 1047 1048 #define for_each_peer_device_rcu(peer_device, device) \ 1049 list_for_each_entry_rcu(peer_device, &device->peer_devices, peer_devices) 1050 1051 #define for_each_peer_device_safe(peer_device, tmp, device) \ 1052 list_for_each_entry_safe(peer_device, tmp, &device->peer_devices, peer_devices) 1053 1054 static inline unsigned int device_to_minor(struct drbd_device *device) 1055 { 1056 return device->minor; 1057 } 1058 1059 /* 1060 * function declarations 1061 *************************/ 1062 1063 /* drbd_main.c */ 1064 1065 enum dds_flags { 1066 DDSF_FORCED = 1, 1067 DDSF_NO_RESYNC = 2, /* Do not run a resync for the new space */ 1068 }; 1069 1070 extern void drbd_init_set_defaults(struct drbd_device *device); 1071 extern int drbd_thread_start(struct drbd_thread *thi); 1072 extern void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait); 1073 #ifdef CONFIG_SMP 1074 extern void drbd_thread_current_set_cpu(struct drbd_thread *thi); 1075 #else 1076 #define drbd_thread_current_set_cpu(A) ({}) 1077 #endif 1078 extern void tl_release(struct drbd_connection *, unsigned int barrier_nr, 1079 unsigned int set_size); 1080 extern void tl_clear(struct drbd_connection *); 1081 extern void drbd_free_sock(struct drbd_connection *connection); 1082 extern int drbd_send(struct drbd_connection *connection, struct socket *sock, 1083 void *buf, size_t size, unsigned msg_flags); 1084 extern int drbd_send_all(struct drbd_connection *, struct socket *, void *, size_t, 1085 unsigned); 1086 1087 extern int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd); 1088 extern int drbd_send_protocol(struct drbd_connection *connection); 1089 extern int drbd_send_uuids(struct drbd_peer_device *); 1090 extern int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *); 1091 extern void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *); 1092 extern int drbd_send_sizes(struct drbd_peer_device *, int trigger_reply, enum dds_flags flags); 1093 extern int drbd_send_state(struct drbd_peer_device *, union drbd_state s); 1094 extern int drbd_send_current_state(struct drbd_peer_device *); 1095 extern int drbd_send_sync_param(struct drbd_peer_device *); 1096 extern void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, 1097 u32 set_size); 1098 extern int drbd_send_ack(struct drbd_peer_device *, enum drbd_packet, 1099 struct drbd_peer_request *); 1100 extern void drbd_send_ack_rp(struct drbd_peer_device *, enum drbd_packet, 1101 struct p_block_req *rp); 1102 extern void drbd_send_ack_dp(struct drbd_peer_device *, enum drbd_packet, 1103 struct p_data *dp, int data_size); 1104 extern int drbd_send_ack_ex(struct drbd_peer_device *, enum drbd_packet, 1105 sector_t sector, int blksize, u64 block_id); 1106 extern int drbd_send_out_of_sync(struct drbd_peer_device *, struct drbd_request *); 1107 extern int drbd_send_block(struct drbd_peer_device *, enum drbd_packet, 1108 struct drbd_peer_request *); 1109 extern int drbd_send_dblock(struct drbd_peer_device *, struct drbd_request *req); 1110 extern int drbd_send_drequest(struct drbd_peer_device *, int cmd, 1111 sector_t sector, int size, u64 block_id); 1112 extern int drbd_send_drequest_csum(struct drbd_peer_device *, sector_t sector, 1113 int size, void *digest, int digest_size, 1114 enum drbd_packet cmd); 1115 extern int drbd_send_ov_request(struct drbd_peer_device *, sector_t sector, int size); 1116 1117 extern int drbd_send_bitmap(struct drbd_device *device); 1118 extern void drbd_send_sr_reply(struct drbd_peer_device *, enum drbd_state_rv retcode); 1119 extern void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode); 1120 extern int drbd_send_rs_deallocated(struct drbd_peer_device *, struct drbd_peer_request *); 1121 extern void drbd_backing_dev_free(struct drbd_device *device, struct drbd_backing_dev *ldev); 1122 extern void drbd_device_cleanup(struct drbd_device *device); 1123 extern void drbd_print_uuids(struct drbd_device *device, const char *text); 1124 extern void drbd_queue_unplug(struct drbd_device *device); 1125 1126 extern void conn_md_sync(struct drbd_connection *connection); 1127 extern void drbd_md_write(struct drbd_device *device, void *buffer); 1128 extern void drbd_md_sync(struct drbd_device *device); 1129 extern int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev); 1130 extern void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local); 1131 extern void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local); 1132 extern void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local); 1133 extern void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local); 1134 extern void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local); 1135 extern void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local); 1136 extern void drbd_md_set_flag(struct drbd_device *device, int flags) __must_hold(local); 1137 extern void drbd_md_clear_flag(struct drbd_device *device, int flags)__must_hold(local); 1138 extern int drbd_md_test_flag(struct drbd_backing_dev *, int); 1139 extern void drbd_md_mark_dirty(struct drbd_device *device); 1140 extern void drbd_queue_bitmap_io(struct drbd_device *device, 1141 int (*io_fn)(struct drbd_device *), 1142 void (*done)(struct drbd_device *, int), 1143 char *why, enum bm_flag flags); 1144 extern int drbd_bitmap_io(struct drbd_device *device, 1145 int (*io_fn)(struct drbd_device *), 1146 char *why, enum bm_flag flags); 1147 extern int drbd_bitmap_io_from_worker(struct drbd_device *device, 1148 int (*io_fn)(struct drbd_device *), 1149 char *why, enum bm_flag flags); 1150 extern int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local); 1151 extern int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local); 1152 1153 /* Meta data layout 1154 * 1155 * We currently have two possible layouts. 1156 * Offsets in (512 byte) sectors. 1157 * external: 1158 * |----------- md_size_sect ------------------| 1159 * [ 4k superblock ][ activity log ][ Bitmap ] 1160 * | al_offset == 8 | 1161 * | bm_offset = al_offset + X | 1162 * ==> bitmap sectors = md_size_sect - bm_offset 1163 * 1164 * Variants: 1165 * old, indexed fixed size meta data: 1166 * 1167 * internal: 1168 * |----------- md_size_sect ------------------| 1169 * [data.....][ Bitmap ][ activity log ][ 4k superblock ][padding*] 1170 * | al_offset < 0 | 1171 * | bm_offset = al_offset - Y | 1172 * ==> bitmap sectors = Y = al_offset - bm_offset 1173 * 1174 * [padding*] are zero or up to 7 unused 512 Byte sectors to the 1175 * end of the device, so that the [4k superblock] will be 4k aligned. 1176 * 1177 * The activity log consists of 4k transaction blocks, 1178 * which are written in a ring-buffer, or striped ring-buffer like fashion, 1179 * which are writtensize used to be fixed 32kB, 1180 * but is about to become configurable. 1181 */ 1182 1183 /* Our old fixed size meta data layout 1184 * allows up to about 3.8TB, so if you want more, 1185 * you need to use the "flexible" meta data format. */ 1186 #define MD_128MB_SECT (128LLU << 11) /* 128 MB, unit sectors */ 1187 #define MD_4kB_SECT 8 1188 #define MD_32kB_SECT 64 1189 1190 /* One activity log extent represents 4M of storage */ 1191 #define AL_EXTENT_SHIFT 22 1192 #define AL_EXTENT_SIZE (1<<AL_EXTENT_SHIFT) 1193 1194 /* We could make these currently hardcoded constants configurable 1195 * variables at create-md time (or even re-configurable at runtime?). 1196 * Which will require some more changes to the DRBD "super block" 1197 * and attach code. 1198 * 1199 * updates per transaction: 1200 * This many changes to the active set can be logged with one transaction. 1201 * This number is arbitrary. 1202 * context per transaction: 1203 * This many context extent numbers are logged with each transaction. 1204 * This number is resulting from the transaction block size (4k), the layout 1205 * of the transaction header, and the number of updates per transaction. 1206 * See drbd_actlog.c:struct al_transaction_on_disk 1207 * */ 1208 #define AL_UPDATES_PER_TRANSACTION 64 // arbitrary 1209 #define AL_CONTEXT_PER_TRANSACTION 919 // (4096 - 36 - 6*64)/4 1210 1211 #if BITS_PER_LONG == 32 1212 #define LN2_BPL 5 1213 #define cpu_to_lel(A) cpu_to_le32(A) 1214 #define lel_to_cpu(A) le32_to_cpu(A) 1215 #elif BITS_PER_LONG == 64 1216 #define LN2_BPL 6 1217 #define cpu_to_lel(A) cpu_to_le64(A) 1218 #define lel_to_cpu(A) le64_to_cpu(A) 1219 #else 1220 #error "LN2 of BITS_PER_LONG unknown!" 1221 #endif 1222 1223 /* resync bitmap */ 1224 /* 16MB sized 'bitmap extent' to track syncer usage */ 1225 struct bm_extent { 1226 int rs_left; /* number of bits set (out of sync) in this extent. */ 1227 int rs_failed; /* number of failed resync requests in this extent. */ 1228 unsigned long flags; 1229 struct lc_element lce; 1230 }; 1231 1232 #define BME_NO_WRITES 0 /* bm_extent.flags: no more requests on this one! */ 1233 #define BME_LOCKED 1 /* bm_extent.flags: syncer active on this one. */ 1234 #define BME_PRIORITY 2 /* finish resync IO on this extent ASAP! App IO waiting! */ 1235 1236 /* drbd_bitmap.c */ 1237 /* 1238 * We need to store one bit for a block. 1239 * Example: 1GB disk @ 4096 byte blocks ==> we need 32 KB bitmap. 1240 * Bit 0 ==> local node thinks this block is binary identical on both nodes 1241 * Bit 1 ==> local node thinks this block needs to be synced. 1242 */ 1243 1244 #define SLEEP_TIME (HZ/10) 1245 1246 /* We do bitmap IO in units of 4k blocks. 1247 * We also still have a hardcoded 4k per bit relation. */ 1248 #define BM_BLOCK_SHIFT 12 /* 4k per bit */ 1249 #define BM_BLOCK_SIZE (1<<BM_BLOCK_SHIFT) 1250 /* mostly arbitrarily set the represented size of one bitmap extent, 1251 * aka resync extent, to 16 MiB (which is also 512 Byte worth of bitmap 1252 * at 4k per bit resolution) */ 1253 #define BM_EXT_SHIFT 24 /* 16 MiB per resync extent */ 1254 #define BM_EXT_SIZE (1<<BM_EXT_SHIFT) 1255 1256 #if (BM_EXT_SHIFT != 24) || (BM_BLOCK_SHIFT != 12) 1257 #error "HAVE YOU FIXED drbdmeta AS WELL??" 1258 #endif 1259 1260 /* thus many _storage_ sectors are described by one bit */ 1261 #define BM_SECT_TO_BIT(x) ((x)>>(BM_BLOCK_SHIFT-9)) 1262 #define BM_BIT_TO_SECT(x) ((sector_t)(x)<<(BM_BLOCK_SHIFT-9)) 1263 #define BM_SECT_PER_BIT BM_BIT_TO_SECT(1) 1264 1265 /* bit to represented kilo byte conversion */ 1266 #define Bit2KB(bits) ((bits)<<(BM_BLOCK_SHIFT-10)) 1267 1268 /* in which _bitmap_ extent (resp. sector) the bit for a certain 1269 * _storage_ sector is located in */ 1270 #define BM_SECT_TO_EXT(x) ((x)>>(BM_EXT_SHIFT-9)) 1271 #define BM_BIT_TO_EXT(x) ((x) >> (BM_EXT_SHIFT - BM_BLOCK_SHIFT)) 1272 1273 /* first storage sector a bitmap extent corresponds to */ 1274 #define BM_EXT_TO_SECT(x) ((sector_t)(x) << (BM_EXT_SHIFT-9)) 1275 /* how much _storage_ sectors we have per bitmap extent */ 1276 #define BM_SECT_PER_EXT BM_EXT_TO_SECT(1) 1277 /* how many bits are covered by one bitmap extent (resync extent) */ 1278 #define BM_BITS_PER_EXT (1UL << (BM_EXT_SHIFT - BM_BLOCK_SHIFT)) 1279 1280 #define BM_BLOCKS_PER_BM_EXT_MASK (BM_BITS_PER_EXT - 1) 1281 1282 1283 /* in one sector of the bitmap, we have this many activity_log extents. */ 1284 #define AL_EXT_PER_BM_SECT (1 << (BM_EXT_SHIFT - AL_EXTENT_SHIFT)) 1285 1286 /* the extent in "PER_EXTENT" below is an activity log extent 1287 * we need that many (long words/bytes) to store the bitmap 1288 * of one AL_EXTENT_SIZE chunk of storage. 1289 * we can store the bitmap for that many AL_EXTENTS within 1290 * one sector of the _on_disk_ bitmap: 1291 * bit 0 bit 37 bit 38 bit (512*8)-1 1292 * ...|........|........|.. // ..|........| 1293 * sect. 0 `296 `304 ^(512*8*8)-1 1294 * 1295 #define BM_WORDS_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / BITS_PER_LONG ) 1296 #define BM_BYTES_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / 8 ) // 128 1297 #define BM_EXT_PER_SECT ( 512 / BM_BYTES_PER_EXTENT ) // 4 1298 */ 1299 1300 #define DRBD_MAX_SECTORS_32 (0xffffffffLU) 1301 /* we have a certain meta data variant that has a fixed on-disk size of 128 1302 * MiB, of which 4k are our "superblock", and 32k are the fixed size activity 1303 * log, leaving this many sectors for the bitmap. 1304 */ 1305 1306 #define DRBD_MAX_SECTORS_FIXED_BM \ 1307 ((MD_128MB_SECT - MD_32kB_SECT - MD_4kB_SECT) * (1LL<<(BM_EXT_SHIFT-9))) 1308 #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_FIXED_BM 1309 /* 16 TB in units of sectors */ 1310 #if BITS_PER_LONG == 32 1311 /* adjust by one page worth of bitmap, 1312 * so we won't wrap around in drbd_bm_find_next_bit. 1313 * you should use 64bit OS for that much storage, anyways. */ 1314 #define DRBD_MAX_SECTORS_FLEX BM_BIT_TO_SECT(0xffff7fff) 1315 #else 1316 /* we allow up to 1 PiB now on 64bit architecture with "flexible" meta data */ 1317 #define DRBD_MAX_SECTORS_FLEX (1UL << 51) 1318 /* corresponds to (1UL << 38) bits right now. */ 1319 #endif 1320 1321 /* Estimate max bio size as 256 * PAGE_SIZE, 1322 * so for typical PAGE_SIZE of 4k, that is (1<<20) Byte. 1323 * Since we may live in a mixed-platform cluster, 1324 * we limit us to a platform agnostic constant here for now. 1325 * A followup commit may allow even bigger BIO sizes, 1326 * once we thought that through. */ 1327 #define DRBD_MAX_BIO_SIZE (1U << 20) 1328 #if DRBD_MAX_BIO_SIZE > (BIO_MAX_PAGES << PAGE_SHIFT) 1329 #error Architecture not supported: DRBD_MAX_BIO_SIZE > BIO_MAX_SIZE 1330 #endif 1331 #define DRBD_MAX_BIO_SIZE_SAFE (1U << 12) /* Works always = 4k */ 1332 1333 #define DRBD_MAX_SIZE_H80_PACKET (1U << 15) /* Header 80 only allows packets up to 32KiB data */ 1334 #define DRBD_MAX_BIO_SIZE_P95 (1U << 17) /* Protocol 95 to 99 allows bios up to 128KiB */ 1335 1336 /* For now, don't allow more than half of what we can "activate" in one 1337 * activity log transaction to be discarded in one go. We may need to rework 1338 * drbd_al_begin_io() to allow for even larger discard ranges */ 1339 #define DRBD_MAX_BATCH_BIO_SIZE (AL_UPDATES_PER_TRANSACTION/2*AL_EXTENT_SIZE) 1340 #define DRBD_MAX_BBIO_SECTORS (DRBD_MAX_BATCH_BIO_SIZE >> 9) 1341 1342 extern int drbd_bm_init(struct drbd_device *device); 1343 extern int drbd_bm_resize(struct drbd_device *device, sector_t sectors, int set_new_bits); 1344 extern void drbd_bm_cleanup(struct drbd_device *device); 1345 extern void drbd_bm_set_all(struct drbd_device *device); 1346 extern void drbd_bm_clear_all(struct drbd_device *device); 1347 /* set/clear/test only a few bits at a time */ 1348 extern int drbd_bm_set_bits( 1349 struct drbd_device *device, unsigned long s, unsigned long e); 1350 extern int drbd_bm_clear_bits( 1351 struct drbd_device *device, unsigned long s, unsigned long e); 1352 extern int drbd_bm_count_bits( 1353 struct drbd_device *device, const unsigned long s, const unsigned long e); 1354 /* bm_set_bits variant for use while holding drbd_bm_lock, 1355 * may process the whole bitmap in one go */ 1356 extern void _drbd_bm_set_bits(struct drbd_device *device, 1357 const unsigned long s, const unsigned long e); 1358 extern int drbd_bm_test_bit(struct drbd_device *device, unsigned long bitnr); 1359 extern int drbd_bm_e_weight(struct drbd_device *device, unsigned long enr); 1360 extern int drbd_bm_read(struct drbd_device *device) __must_hold(local); 1361 extern void drbd_bm_mark_for_writeout(struct drbd_device *device, int page_nr); 1362 extern int drbd_bm_write(struct drbd_device *device) __must_hold(local); 1363 extern void drbd_bm_reset_al_hints(struct drbd_device *device) __must_hold(local); 1364 extern int drbd_bm_write_hinted(struct drbd_device *device) __must_hold(local); 1365 extern int drbd_bm_write_lazy(struct drbd_device *device, unsigned upper_idx) __must_hold(local); 1366 extern int drbd_bm_write_all(struct drbd_device *device) __must_hold(local); 1367 extern int drbd_bm_write_copy_pages(struct drbd_device *device) __must_hold(local); 1368 extern size_t drbd_bm_words(struct drbd_device *device); 1369 extern unsigned long drbd_bm_bits(struct drbd_device *device); 1370 extern sector_t drbd_bm_capacity(struct drbd_device *device); 1371 1372 #define DRBD_END_OF_BITMAP (~(unsigned long)0) 1373 extern unsigned long drbd_bm_find_next(struct drbd_device *device, unsigned long bm_fo); 1374 /* bm_find_next variants for use while you hold drbd_bm_lock() */ 1375 extern unsigned long _drbd_bm_find_next(struct drbd_device *device, unsigned long bm_fo); 1376 extern unsigned long _drbd_bm_find_next_zero(struct drbd_device *device, unsigned long bm_fo); 1377 extern unsigned long _drbd_bm_total_weight(struct drbd_device *device); 1378 extern unsigned long drbd_bm_total_weight(struct drbd_device *device); 1379 /* for receive_bitmap */ 1380 extern void drbd_bm_merge_lel(struct drbd_device *device, size_t offset, 1381 size_t number, unsigned long *buffer); 1382 /* for _drbd_send_bitmap */ 1383 extern void drbd_bm_get_lel(struct drbd_device *device, size_t offset, 1384 size_t number, unsigned long *buffer); 1385 1386 extern void drbd_bm_lock(struct drbd_device *device, char *why, enum bm_flag flags); 1387 extern void drbd_bm_unlock(struct drbd_device *device); 1388 /* drbd_main.c */ 1389 1390 extern struct kmem_cache *drbd_request_cache; 1391 extern struct kmem_cache *drbd_ee_cache; /* peer requests */ 1392 extern struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */ 1393 extern struct kmem_cache *drbd_al_ext_cache; /* activity log extents */ 1394 extern mempool_t drbd_request_mempool; 1395 extern mempool_t drbd_ee_mempool; 1396 1397 /* drbd's page pool, used to buffer data received from the peer, 1398 * or data requested by the peer. 1399 * 1400 * This does not have an emergency reserve. 1401 * 1402 * When allocating from this pool, it first takes pages from the pool. 1403 * Only if the pool is depleted will try to allocate from the system. 1404 * 1405 * The assumption is that pages taken from this pool will be processed, 1406 * and given back, "quickly", and then can be recycled, so we can avoid 1407 * frequent calls to alloc_page(), and still will be able to make progress even 1408 * under memory pressure. 1409 */ 1410 extern struct page *drbd_pp_pool; 1411 extern spinlock_t drbd_pp_lock; 1412 extern int drbd_pp_vacant; 1413 extern wait_queue_head_t drbd_pp_wait; 1414 1415 /* We also need a standard (emergency-reserve backed) page pool 1416 * for meta data IO (activity log, bitmap). 1417 * We can keep it global, as long as it is used as "N pages at a time". 1418 * 128 should be plenty, currently we probably can get away with as few as 1. 1419 */ 1420 #define DRBD_MIN_POOL_PAGES 128 1421 extern mempool_t drbd_md_io_page_pool; 1422 1423 /* We also need to make sure we get a bio 1424 * when we need it for housekeeping purposes */ 1425 extern struct bio_set drbd_md_io_bio_set; 1426 /* to allocate from that set */ 1427 extern struct bio *bio_alloc_drbd(gfp_t gfp_mask); 1428 1429 /* And a bio_set for cloning */ 1430 extern struct bio_set drbd_io_bio_set; 1431 1432 extern struct mutex resources_mutex; 1433 1434 extern int conn_lowest_minor(struct drbd_connection *connection); 1435 extern enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor); 1436 extern void drbd_destroy_device(struct kref *kref); 1437 extern void drbd_delete_device(struct drbd_device *device); 1438 1439 extern struct drbd_resource *drbd_create_resource(const char *name); 1440 extern void drbd_free_resource(struct drbd_resource *resource); 1441 1442 extern int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts); 1443 extern struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts); 1444 extern void drbd_destroy_connection(struct kref *kref); 1445 extern struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len, 1446 void *peer_addr, int peer_addr_len); 1447 extern struct drbd_resource *drbd_find_resource(const char *name); 1448 extern void drbd_destroy_resource(struct kref *kref); 1449 extern void conn_free_crypto(struct drbd_connection *connection); 1450 1451 /* drbd_req */ 1452 extern void do_submit(struct work_struct *ws); 1453 extern void __drbd_make_request(struct drbd_device *, struct bio *, unsigned long); 1454 extern blk_qc_t drbd_make_request(struct request_queue *q, struct bio *bio); 1455 extern int drbd_read_remote(struct drbd_device *device, struct drbd_request *req); 1456 extern int is_valid_ar_handle(struct drbd_request *, sector_t); 1457 1458 1459 /* drbd_nl.c */ 1460 1461 extern struct mutex notification_mutex; 1462 1463 extern void drbd_suspend_io(struct drbd_device *device); 1464 extern void drbd_resume_io(struct drbd_device *device); 1465 extern char *ppsize(char *buf, unsigned long long size); 1466 extern sector_t drbd_new_dev_size(struct drbd_device *, struct drbd_backing_dev *, sector_t, int); 1467 enum determine_dev_size { 1468 DS_ERROR_SHRINK = -3, 1469 DS_ERROR_SPACE_MD = -2, 1470 DS_ERROR = -1, 1471 DS_UNCHANGED = 0, 1472 DS_SHRUNK = 1, 1473 DS_GREW = 2, 1474 DS_GREW_FROM_ZERO = 3, 1475 }; 1476 extern enum determine_dev_size 1477 drbd_determine_dev_size(struct drbd_device *, enum dds_flags, struct resize_parms *) __must_hold(local); 1478 extern void resync_after_online_grow(struct drbd_device *); 1479 extern void drbd_reconsider_queue_parameters(struct drbd_device *device, 1480 struct drbd_backing_dev *bdev, struct o_qlim *o); 1481 extern enum drbd_state_rv drbd_set_role(struct drbd_device *device, 1482 enum drbd_role new_role, 1483 int force); 1484 extern bool conn_try_outdate_peer(struct drbd_connection *connection); 1485 extern void conn_try_outdate_peer_async(struct drbd_connection *connection); 1486 extern enum drbd_peer_state conn_khelper(struct drbd_connection *connection, char *cmd); 1487 extern int drbd_khelper(struct drbd_device *device, char *cmd); 1488 1489 /* drbd_worker.c */ 1490 /* bi_end_io handlers */ 1491 extern void drbd_md_endio(struct bio *bio); 1492 extern void drbd_peer_request_endio(struct bio *bio); 1493 extern void drbd_request_endio(struct bio *bio); 1494 extern int drbd_worker(struct drbd_thread *thi); 1495 enum drbd_ret_code drbd_resync_after_valid(struct drbd_device *device, int o_minor); 1496 void drbd_resync_after_changed(struct drbd_device *device); 1497 extern void drbd_start_resync(struct drbd_device *device, enum drbd_conns side); 1498 extern void resume_next_sg(struct drbd_device *device); 1499 extern void suspend_other_sg(struct drbd_device *device); 1500 extern int drbd_resync_finished(struct drbd_device *device); 1501 /* maybe rather drbd_main.c ? */ 1502 extern void *drbd_md_get_buffer(struct drbd_device *device, const char *intent); 1503 extern void drbd_md_put_buffer(struct drbd_device *device); 1504 extern int drbd_md_sync_page_io(struct drbd_device *device, 1505 struct drbd_backing_dev *bdev, sector_t sector, int op); 1506 extern void drbd_ov_out_of_sync_found(struct drbd_device *, sector_t, int); 1507 extern void wait_until_done_or_force_detached(struct drbd_device *device, 1508 struct drbd_backing_dev *bdev, unsigned int *done); 1509 extern void drbd_rs_controller_reset(struct drbd_device *device); 1510 1511 static inline void ov_out_of_sync_print(struct drbd_device *device) 1512 { 1513 if (device->ov_last_oos_size) { 1514 drbd_err(device, "Out of sync: start=%llu, size=%lu (sectors)\n", 1515 (unsigned long long)device->ov_last_oos_start, 1516 (unsigned long)device->ov_last_oos_size); 1517 } 1518 device->ov_last_oos_size = 0; 1519 } 1520 1521 1522 extern void drbd_csum_bio(struct crypto_shash *, struct bio *, void *); 1523 extern void drbd_csum_ee(struct crypto_shash *, struct drbd_peer_request *, 1524 void *); 1525 /* worker callbacks */ 1526 extern int w_e_end_data_req(struct drbd_work *, int); 1527 extern int w_e_end_rsdata_req(struct drbd_work *, int); 1528 extern int w_e_end_csum_rs_req(struct drbd_work *, int); 1529 extern int w_e_end_ov_reply(struct drbd_work *, int); 1530 extern int w_e_end_ov_req(struct drbd_work *, int); 1531 extern int w_ov_finished(struct drbd_work *, int); 1532 extern int w_resync_timer(struct drbd_work *, int); 1533 extern int w_send_write_hint(struct drbd_work *, int); 1534 extern int w_send_dblock(struct drbd_work *, int); 1535 extern int w_send_read_req(struct drbd_work *, int); 1536 extern int w_e_reissue(struct drbd_work *, int); 1537 extern int w_restart_disk_io(struct drbd_work *, int); 1538 extern int w_send_out_of_sync(struct drbd_work *, int); 1539 extern int w_start_resync(struct drbd_work *, int); 1540 1541 extern void resync_timer_fn(struct timer_list *t); 1542 extern void start_resync_timer_fn(struct timer_list *t); 1543 1544 extern void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req); 1545 1546 /* drbd_receiver.c */ 1547 extern int drbd_issue_discard_or_zero_out(struct drbd_device *device, 1548 sector_t start, unsigned int nr_sectors, int flags); 1549 extern int drbd_receiver(struct drbd_thread *thi); 1550 extern int drbd_ack_receiver(struct drbd_thread *thi); 1551 extern void drbd_send_ping_wf(struct work_struct *ws); 1552 extern void drbd_send_acks_wf(struct work_struct *ws); 1553 extern bool drbd_rs_c_min_rate_throttle(struct drbd_device *device); 1554 extern bool drbd_rs_should_slow_down(struct drbd_device *device, sector_t sector, 1555 bool throttle_if_app_is_waiting); 1556 extern int drbd_submit_peer_request(struct drbd_device *, 1557 struct drbd_peer_request *, const unsigned, 1558 const unsigned, const int); 1559 extern int drbd_free_peer_reqs(struct drbd_device *, struct list_head *); 1560 extern struct drbd_peer_request *drbd_alloc_peer_req(struct drbd_peer_device *, u64, 1561 sector_t, unsigned int, 1562 unsigned int, 1563 gfp_t) __must_hold(local); 1564 extern void __drbd_free_peer_req(struct drbd_device *, struct drbd_peer_request *, 1565 int); 1566 #define drbd_free_peer_req(m,e) __drbd_free_peer_req(m, e, 0) 1567 #define drbd_free_net_peer_req(m,e) __drbd_free_peer_req(m, e, 1) 1568 extern struct page *drbd_alloc_pages(struct drbd_peer_device *, unsigned int, bool); 1569 extern void drbd_set_recv_tcq(struct drbd_device *device, int tcq_enabled); 1570 extern void _drbd_clear_done_ee(struct drbd_device *device, struct list_head *to_be_freed); 1571 extern int drbd_connected(struct drbd_peer_device *); 1572 1573 static inline void drbd_tcp_cork(struct socket *sock) 1574 { 1575 int val = 1; 1576 (void) kernel_setsockopt(sock, SOL_TCP, TCP_CORK, 1577 (char*)&val, sizeof(val)); 1578 } 1579 1580 static inline void drbd_tcp_uncork(struct socket *sock) 1581 { 1582 int val = 0; 1583 (void) kernel_setsockopt(sock, SOL_TCP, TCP_CORK, 1584 (char*)&val, sizeof(val)); 1585 } 1586 1587 static inline void drbd_tcp_nodelay(struct socket *sock) 1588 { 1589 int val = 1; 1590 (void) kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, 1591 (char*)&val, sizeof(val)); 1592 } 1593 1594 static inline void drbd_tcp_quickack(struct socket *sock) 1595 { 1596 int val = 2; 1597 (void) kernel_setsockopt(sock, SOL_TCP, TCP_QUICKACK, 1598 (char*)&val, sizeof(val)); 1599 } 1600 1601 /* sets the number of 512 byte sectors of our virtual device */ 1602 void drbd_set_my_capacity(struct drbd_device *device, sector_t size); 1603 1604 /* 1605 * used to submit our private bio 1606 */ 1607 static inline void drbd_generic_make_request(struct drbd_device *device, 1608 int fault_type, struct bio *bio) 1609 { 1610 __release(local); 1611 if (!bio->bi_disk) { 1612 drbd_err(device, "drbd_generic_make_request: bio->bi_disk == NULL\n"); 1613 bio->bi_status = BLK_STS_IOERR; 1614 bio_endio(bio); 1615 return; 1616 } 1617 1618 if (drbd_insert_fault(device, fault_type)) 1619 bio_io_error(bio); 1620 else 1621 generic_make_request(bio); 1622 } 1623 1624 void drbd_bump_write_ordering(struct drbd_resource *resource, struct drbd_backing_dev *bdev, 1625 enum write_ordering_e wo); 1626 1627 /* drbd_proc.c */ 1628 extern struct proc_dir_entry *drbd_proc; 1629 int drbd_seq_show(struct seq_file *seq, void *v); 1630 1631 /* drbd_actlog.c */ 1632 extern bool drbd_al_begin_io_prepare(struct drbd_device *device, struct drbd_interval *i); 1633 extern int drbd_al_begin_io_nonblock(struct drbd_device *device, struct drbd_interval *i); 1634 extern void drbd_al_begin_io_commit(struct drbd_device *device); 1635 extern bool drbd_al_begin_io_fastpath(struct drbd_device *device, struct drbd_interval *i); 1636 extern void drbd_al_begin_io(struct drbd_device *device, struct drbd_interval *i); 1637 extern void drbd_al_complete_io(struct drbd_device *device, struct drbd_interval *i); 1638 extern void drbd_rs_complete_io(struct drbd_device *device, sector_t sector); 1639 extern int drbd_rs_begin_io(struct drbd_device *device, sector_t sector); 1640 extern int drbd_try_rs_begin_io(struct drbd_device *device, sector_t sector); 1641 extern void drbd_rs_cancel_all(struct drbd_device *device); 1642 extern int drbd_rs_del_all(struct drbd_device *device); 1643 extern void drbd_rs_failed_io(struct drbd_device *device, 1644 sector_t sector, int size); 1645 extern void drbd_advance_rs_marks(struct drbd_device *device, unsigned long still_to_go); 1646 1647 enum update_sync_bits_mode { RECORD_RS_FAILED, SET_OUT_OF_SYNC, SET_IN_SYNC }; 1648 extern int __drbd_change_sync(struct drbd_device *device, sector_t sector, int size, 1649 enum update_sync_bits_mode mode); 1650 #define drbd_set_in_sync(device, sector, size) \ 1651 __drbd_change_sync(device, sector, size, SET_IN_SYNC) 1652 #define drbd_set_out_of_sync(device, sector, size) \ 1653 __drbd_change_sync(device, sector, size, SET_OUT_OF_SYNC) 1654 #define drbd_rs_failed_io(device, sector, size) \ 1655 __drbd_change_sync(device, sector, size, RECORD_RS_FAILED) 1656 extern void drbd_al_shrink(struct drbd_device *device); 1657 extern int drbd_al_initialize(struct drbd_device *, void *); 1658 1659 /* drbd_nl.c */ 1660 /* state info broadcast */ 1661 struct sib_info { 1662 enum drbd_state_info_bcast_reason sib_reason; 1663 union { 1664 struct { 1665 char *helper_name; 1666 unsigned helper_exit_code; 1667 }; 1668 struct { 1669 union drbd_state os; 1670 union drbd_state ns; 1671 }; 1672 }; 1673 }; 1674 void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib); 1675 1676 extern void notify_resource_state(struct sk_buff *, 1677 unsigned int, 1678 struct drbd_resource *, 1679 struct resource_info *, 1680 enum drbd_notification_type); 1681 extern void notify_device_state(struct sk_buff *, 1682 unsigned int, 1683 struct drbd_device *, 1684 struct device_info *, 1685 enum drbd_notification_type); 1686 extern void notify_connection_state(struct sk_buff *, 1687 unsigned int, 1688 struct drbd_connection *, 1689 struct connection_info *, 1690 enum drbd_notification_type); 1691 extern void notify_peer_device_state(struct sk_buff *, 1692 unsigned int, 1693 struct drbd_peer_device *, 1694 struct peer_device_info *, 1695 enum drbd_notification_type); 1696 extern void notify_helper(enum drbd_notification_type, struct drbd_device *, 1697 struct drbd_connection *, const char *, int); 1698 1699 /* 1700 * inline helper functions 1701 *************************/ 1702 1703 /* see also page_chain_add and friends in drbd_receiver.c */ 1704 static inline struct page *page_chain_next(struct page *page) 1705 { 1706 return (struct page *)page_private(page); 1707 } 1708 #define page_chain_for_each(page) \ 1709 for (; page && ({ prefetch(page_chain_next(page)); 1; }); \ 1710 page = page_chain_next(page)) 1711 #define page_chain_for_each_safe(page, n) \ 1712 for (; page && ({ n = page_chain_next(page); 1; }); page = n) 1713 1714 1715 static inline int drbd_peer_req_has_active_page(struct drbd_peer_request *peer_req) 1716 { 1717 struct page *page = peer_req->pages; 1718 page_chain_for_each(page) { 1719 if (page_count(page) > 1) 1720 return 1; 1721 } 1722 return 0; 1723 } 1724 1725 static inline union drbd_state drbd_read_state(struct drbd_device *device) 1726 { 1727 struct drbd_resource *resource = device->resource; 1728 union drbd_state rv; 1729 1730 rv.i = device->state.i; 1731 rv.susp = resource->susp; 1732 rv.susp_nod = resource->susp_nod; 1733 rv.susp_fen = resource->susp_fen; 1734 1735 return rv; 1736 } 1737 1738 enum drbd_force_detach_flags { 1739 DRBD_READ_ERROR, 1740 DRBD_WRITE_ERROR, 1741 DRBD_META_IO_ERROR, 1742 DRBD_FORCE_DETACH, 1743 }; 1744 1745 #define __drbd_chk_io_error(m,f) __drbd_chk_io_error_(m,f, __func__) 1746 static inline void __drbd_chk_io_error_(struct drbd_device *device, 1747 enum drbd_force_detach_flags df, 1748 const char *where) 1749 { 1750 enum drbd_io_error_p ep; 1751 1752 rcu_read_lock(); 1753 ep = rcu_dereference(device->ldev->disk_conf)->on_io_error; 1754 rcu_read_unlock(); 1755 switch (ep) { 1756 case EP_PASS_ON: /* FIXME would this be better named "Ignore"? */ 1757 if (df == DRBD_READ_ERROR || df == DRBD_WRITE_ERROR) { 1758 if (__ratelimit(&drbd_ratelimit_state)) 1759 drbd_err(device, "Local IO failed in %s.\n", where); 1760 if (device->state.disk > D_INCONSISTENT) 1761 _drbd_set_state(_NS(device, disk, D_INCONSISTENT), CS_HARD, NULL); 1762 break; 1763 } 1764 /* fall through - for DRBD_META_IO_ERROR or DRBD_FORCE_DETACH */ 1765 case EP_DETACH: 1766 case EP_CALL_HELPER: 1767 /* Remember whether we saw a READ or WRITE error. 1768 * 1769 * Recovery of the affected area for WRITE failure is covered 1770 * by the activity log. 1771 * READ errors may fall outside that area though. Certain READ 1772 * errors can be "healed" by writing good data to the affected 1773 * blocks, which triggers block re-allocation in lower layers. 1774 * 1775 * If we can not write the bitmap after a READ error, 1776 * we may need to trigger a full sync (see w_go_diskless()). 1777 * 1778 * Force-detach is not really an IO error, but rather a 1779 * desperate measure to try to deal with a completely 1780 * unresponsive lower level IO stack. 1781 * Still it should be treated as a WRITE error. 1782 * 1783 * Meta IO error is always WRITE error: 1784 * we read meta data only once during attach, 1785 * which will fail in case of errors. 1786 */ 1787 set_bit(WAS_IO_ERROR, &device->flags); 1788 if (df == DRBD_READ_ERROR) 1789 set_bit(WAS_READ_ERROR, &device->flags); 1790 if (df == DRBD_FORCE_DETACH) 1791 set_bit(FORCE_DETACH, &device->flags); 1792 if (device->state.disk > D_FAILED) { 1793 _drbd_set_state(_NS(device, disk, D_FAILED), CS_HARD, NULL); 1794 drbd_err(device, 1795 "Local IO failed in %s. Detaching...\n", where); 1796 } 1797 break; 1798 } 1799 } 1800 1801 /** 1802 * drbd_chk_io_error: Handle the on_io_error setting, should be called from all io completion handlers 1803 * @device: DRBD device. 1804 * @error: Error code passed to the IO completion callback 1805 * @forcedetach: Force detach. I.e. the error happened while accessing the meta data 1806 * 1807 * See also drbd_main.c:after_state_ch() if (os.disk > D_FAILED && ns.disk == D_FAILED) 1808 */ 1809 #define drbd_chk_io_error(m,e,f) drbd_chk_io_error_(m,e,f, __func__) 1810 static inline void drbd_chk_io_error_(struct drbd_device *device, 1811 int error, enum drbd_force_detach_flags forcedetach, const char *where) 1812 { 1813 if (error) { 1814 unsigned long flags; 1815 spin_lock_irqsave(&device->resource->req_lock, flags); 1816 __drbd_chk_io_error_(device, forcedetach, where); 1817 spin_unlock_irqrestore(&device->resource->req_lock, flags); 1818 } 1819 } 1820 1821 1822 /** 1823 * drbd_md_first_sector() - Returns the first sector number of the meta data area 1824 * @bdev: Meta data block device. 1825 * 1826 * BTW, for internal meta data, this happens to be the maximum capacity 1827 * we could agree upon with our peer node. 1828 */ 1829 static inline sector_t drbd_md_first_sector(struct drbd_backing_dev *bdev) 1830 { 1831 switch (bdev->md.meta_dev_idx) { 1832 case DRBD_MD_INDEX_INTERNAL: 1833 case DRBD_MD_INDEX_FLEX_INT: 1834 return bdev->md.md_offset + bdev->md.bm_offset; 1835 case DRBD_MD_INDEX_FLEX_EXT: 1836 default: 1837 return bdev->md.md_offset; 1838 } 1839 } 1840 1841 /** 1842 * drbd_md_last_sector() - Return the last sector number of the meta data area 1843 * @bdev: Meta data block device. 1844 */ 1845 static inline sector_t drbd_md_last_sector(struct drbd_backing_dev *bdev) 1846 { 1847 switch (bdev->md.meta_dev_idx) { 1848 case DRBD_MD_INDEX_INTERNAL: 1849 case DRBD_MD_INDEX_FLEX_INT: 1850 return bdev->md.md_offset + MD_4kB_SECT -1; 1851 case DRBD_MD_INDEX_FLEX_EXT: 1852 default: 1853 return bdev->md.md_offset + bdev->md.md_size_sect -1; 1854 } 1855 } 1856 1857 /* Returns the number of 512 byte sectors of the device */ 1858 static inline sector_t drbd_get_capacity(struct block_device *bdev) 1859 { 1860 /* return bdev ? get_capacity(bdev->bd_disk) : 0; */ 1861 return bdev ? i_size_read(bdev->bd_inode) >> 9 : 0; 1862 } 1863 1864 /** 1865 * drbd_get_max_capacity() - Returns the capacity we announce to out peer 1866 * @bdev: Meta data block device. 1867 * 1868 * returns the capacity we announce to out peer. we clip ourselves at the 1869 * various MAX_SECTORS, because if we don't, current implementation will 1870 * oops sooner or later 1871 */ 1872 static inline sector_t drbd_get_max_capacity(struct drbd_backing_dev *bdev) 1873 { 1874 sector_t s; 1875 1876 switch (bdev->md.meta_dev_idx) { 1877 case DRBD_MD_INDEX_INTERNAL: 1878 case DRBD_MD_INDEX_FLEX_INT: 1879 s = drbd_get_capacity(bdev->backing_bdev) 1880 ? min_t(sector_t, DRBD_MAX_SECTORS_FLEX, 1881 drbd_md_first_sector(bdev)) 1882 : 0; 1883 break; 1884 case DRBD_MD_INDEX_FLEX_EXT: 1885 s = min_t(sector_t, DRBD_MAX_SECTORS_FLEX, 1886 drbd_get_capacity(bdev->backing_bdev)); 1887 /* clip at maximum size the meta device can support */ 1888 s = min_t(sector_t, s, 1889 BM_EXT_TO_SECT(bdev->md.md_size_sect 1890 - bdev->md.bm_offset)); 1891 break; 1892 default: 1893 s = min_t(sector_t, DRBD_MAX_SECTORS, 1894 drbd_get_capacity(bdev->backing_bdev)); 1895 } 1896 return s; 1897 } 1898 1899 /** 1900 * drbd_md_ss() - Return the sector number of our meta data super block 1901 * @bdev: Meta data block device. 1902 */ 1903 static inline sector_t drbd_md_ss(struct drbd_backing_dev *bdev) 1904 { 1905 const int meta_dev_idx = bdev->md.meta_dev_idx; 1906 1907 if (meta_dev_idx == DRBD_MD_INDEX_FLEX_EXT) 1908 return 0; 1909 1910 /* Since drbd08, internal meta data is always "flexible". 1911 * position: last 4k aligned block of 4k size */ 1912 if (meta_dev_idx == DRBD_MD_INDEX_INTERNAL || 1913 meta_dev_idx == DRBD_MD_INDEX_FLEX_INT) 1914 return (drbd_get_capacity(bdev->backing_bdev) & ~7ULL) - 8; 1915 1916 /* external, some index; this is the old fixed size layout */ 1917 return MD_128MB_SECT * bdev->md.meta_dev_idx; 1918 } 1919 1920 static inline void 1921 drbd_queue_work(struct drbd_work_queue *q, struct drbd_work *w) 1922 { 1923 unsigned long flags; 1924 spin_lock_irqsave(&q->q_lock, flags); 1925 list_add_tail(&w->list, &q->q); 1926 spin_unlock_irqrestore(&q->q_lock, flags); 1927 wake_up(&q->q_wait); 1928 } 1929 1930 static inline void 1931 drbd_queue_work_if_unqueued(struct drbd_work_queue *q, struct drbd_work *w) 1932 { 1933 unsigned long flags; 1934 spin_lock_irqsave(&q->q_lock, flags); 1935 if (list_empty_careful(&w->list)) 1936 list_add_tail(&w->list, &q->q); 1937 spin_unlock_irqrestore(&q->q_lock, flags); 1938 wake_up(&q->q_wait); 1939 } 1940 1941 static inline void 1942 drbd_device_post_work(struct drbd_device *device, int work_bit) 1943 { 1944 if (!test_and_set_bit(work_bit, &device->flags)) { 1945 struct drbd_connection *connection = 1946 first_peer_device(device)->connection; 1947 struct drbd_work_queue *q = &connection->sender_work; 1948 if (!test_and_set_bit(DEVICE_WORK_PENDING, &connection->flags)) 1949 wake_up(&q->q_wait); 1950 } 1951 } 1952 1953 extern void drbd_flush_workqueue(struct drbd_work_queue *work_queue); 1954 1955 /* To get the ack_receiver out of the blocking network stack, 1956 * so it can change its sk_rcvtimeo from idle- to ping-timeout, 1957 * and send a ping, we need to send a signal. 1958 * Which signal we send is irrelevant. */ 1959 static inline void wake_ack_receiver(struct drbd_connection *connection) 1960 { 1961 struct task_struct *task = connection->ack_receiver.task; 1962 if (task && get_t_state(&connection->ack_receiver) == RUNNING) 1963 force_sig(SIGXCPU, task); 1964 } 1965 1966 static inline void request_ping(struct drbd_connection *connection) 1967 { 1968 set_bit(SEND_PING, &connection->flags); 1969 wake_ack_receiver(connection); 1970 } 1971 1972 extern void *conn_prepare_command(struct drbd_connection *, struct drbd_socket *); 1973 extern void *drbd_prepare_command(struct drbd_peer_device *, struct drbd_socket *); 1974 extern int conn_send_command(struct drbd_connection *, struct drbd_socket *, 1975 enum drbd_packet, unsigned int, void *, 1976 unsigned int); 1977 extern int drbd_send_command(struct drbd_peer_device *, struct drbd_socket *, 1978 enum drbd_packet, unsigned int, void *, 1979 unsigned int); 1980 1981 extern int drbd_send_ping(struct drbd_connection *connection); 1982 extern int drbd_send_ping_ack(struct drbd_connection *connection); 1983 extern int drbd_send_state_req(struct drbd_peer_device *, union drbd_state, union drbd_state); 1984 extern int conn_send_state_req(struct drbd_connection *, union drbd_state, union drbd_state); 1985 1986 static inline void drbd_thread_stop(struct drbd_thread *thi) 1987 { 1988 _drbd_thread_stop(thi, false, true); 1989 } 1990 1991 static inline void drbd_thread_stop_nowait(struct drbd_thread *thi) 1992 { 1993 _drbd_thread_stop(thi, false, false); 1994 } 1995 1996 static inline void drbd_thread_restart_nowait(struct drbd_thread *thi) 1997 { 1998 _drbd_thread_stop(thi, true, false); 1999 } 2000 2001 /* counts how many answer packets packets we expect from our peer, 2002 * for either explicit application requests, 2003 * or implicit barrier packets as necessary. 2004 * increased: 2005 * w_send_barrier 2006 * _req_mod(req, QUEUE_FOR_NET_WRITE or QUEUE_FOR_NET_READ); 2007 * it is much easier and equally valid to count what we queue for the 2008 * worker, even before it actually was queued or send. 2009 * (drbd_make_request_common; recovery path on read io-error) 2010 * decreased: 2011 * got_BarrierAck (respective tl_clear, tl_clear_barrier) 2012 * _req_mod(req, DATA_RECEIVED) 2013 * [from receive_DataReply] 2014 * _req_mod(req, WRITE_ACKED_BY_PEER or RECV_ACKED_BY_PEER or NEG_ACKED) 2015 * [from got_BlockAck (P_WRITE_ACK, P_RECV_ACK)] 2016 * for some reason it is NOT decreased in got_NegAck, 2017 * but in the resulting cleanup code from report_params. 2018 * we should try to remember the reason for that... 2019 * _req_mod(req, SEND_FAILED or SEND_CANCELED) 2020 * _req_mod(req, CONNECTION_LOST_WHILE_PENDING) 2021 * [from tl_clear_barrier] 2022 */ 2023 static inline void inc_ap_pending(struct drbd_device *device) 2024 { 2025 atomic_inc(&device->ap_pending_cnt); 2026 } 2027 2028 #define ERR_IF_CNT_IS_NEGATIVE(which, func, line) \ 2029 if (atomic_read(&device->which) < 0) \ 2030 drbd_err(device, "in %s:%d: " #which " = %d < 0 !\n", \ 2031 func, line, \ 2032 atomic_read(&device->which)) 2033 2034 #define dec_ap_pending(device) _dec_ap_pending(device, __func__, __LINE__) 2035 static inline void _dec_ap_pending(struct drbd_device *device, const char *func, int line) 2036 { 2037 if (atomic_dec_and_test(&device->ap_pending_cnt)) 2038 wake_up(&device->misc_wait); 2039 ERR_IF_CNT_IS_NEGATIVE(ap_pending_cnt, func, line); 2040 } 2041 2042 /* counts how many resync-related answers we still expect from the peer 2043 * increase decrease 2044 * C_SYNC_TARGET sends P_RS_DATA_REQUEST (and expects P_RS_DATA_REPLY) 2045 * C_SYNC_SOURCE sends P_RS_DATA_REPLY (and expects P_WRITE_ACK with ID_SYNCER) 2046 * (or P_NEG_ACK with ID_SYNCER) 2047 */ 2048 static inline void inc_rs_pending(struct drbd_device *device) 2049 { 2050 atomic_inc(&device->rs_pending_cnt); 2051 } 2052 2053 #define dec_rs_pending(device) _dec_rs_pending(device, __func__, __LINE__) 2054 static inline void _dec_rs_pending(struct drbd_device *device, const char *func, int line) 2055 { 2056 atomic_dec(&device->rs_pending_cnt); 2057 ERR_IF_CNT_IS_NEGATIVE(rs_pending_cnt, func, line); 2058 } 2059 2060 /* counts how many answers we still need to send to the peer. 2061 * increased on 2062 * receive_Data unless protocol A; 2063 * we need to send a P_RECV_ACK (proto B) 2064 * or P_WRITE_ACK (proto C) 2065 * receive_RSDataReply (recv_resync_read) we need to send a P_WRITE_ACK 2066 * receive_DataRequest (receive_RSDataRequest) we need to send back P_DATA 2067 * receive_Barrier_* we need to send a P_BARRIER_ACK 2068 */ 2069 static inline void inc_unacked(struct drbd_device *device) 2070 { 2071 atomic_inc(&device->unacked_cnt); 2072 } 2073 2074 #define dec_unacked(device) _dec_unacked(device, __func__, __LINE__) 2075 static inline void _dec_unacked(struct drbd_device *device, const char *func, int line) 2076 { 2077 atomic_dec(&device->unacked_cnt); 2078 ERR_IF_CNT_IS_NEGATIVE(unacked_cnt, func, line); 2079 } 2080 2081 #define sub_unacked(device, n) _sub_unacked(device, n, __func__, __LINE__) 2082 static inline void _sub_unacked(struct drbd_device *device, int n, const char *func, int line) 2083 { 2084 atomic_sub(n, &device->unacked_cnt); 2085 ERR_IF_CNT_IS_NEGATIVE(unacked_cnt, func, line); 2086 } 2087 2088 static inline bool is_sync_target_state(enum drbd_conns connection_state) 2089 { 2090 return connection_state == C_SYNC_TARGET || 2091 connection_state == C_PAUSED_SYNC_T; 2092 } 2093 2094 static inline bool is_sync_source_state(enum drbd_conns connection_state) 2095 { 2096 return connection_state == C_SYNC_SOURCE || 2097 connection_state == C_PAUSED_SYNC_S; 2098 } 2099 2100 static inline bool is_sync_state(enum drbd_conns connection_state) 2101 { 2102 return is_sync_source_state(connection_state) || 2103 is_sync_target_state(connection_state); 2104 } 2105 2106 /** 2107 * get_ldev() - Increase the ref count on device->ldev. Returns 0 if there is no ldev 2108 * @_device: DRBD device. 2109 * @_min_state: Minimum device state required for success. 2110 * 2111 * You have to call put_ldev() when finished working with device->ldev. 2112 */ 2113 #define get_ldev_if_state(_device, _min_state) \ 2114 (_get_ldev_if_state((_device), (_min_state)) ? \ 2115 ({ __acquire(x); true; }) : false) 2116 #define get_ldev(_device) get_ldev_if_state(_device, D_INCONSISTENT) 2117 2118 static inline void put_ldev(struct drbd_device *device) 2119 { 2120 enum drbd_disk_state disk_state = device->state.disk; 2121 /* We must check the state *before* the atomic_dec becomes visible, 2122 * or we have a theoretical race where someone hitting zero, 2123 * while state still D_FAILED, will then see D_DISKLESS in the 2124 * condition below and calling into destroy, where he must not, yet. */ 2125 int i = atomic_dec_return(&device->local_cnt); 2126 2127 /* This may be called from some endio handler, 2128 * so we must not sleep here. */ 2129 2130 __release(local); 2131 D_ASSERT(device, i >= 0); 2132 if (i == 0) { 2133 if (disk_state == D_DISKLESS) 2134 /* even internal references gone, safe to destroy */ 2135 drbd_device_post_work(device, DESTROY_DISK); 2136 if (disk_state == D_FAILED) 2137 /* all application IO references gone. */ 2138 if (!test_and_set_bit(GOING_DISKLESS, &device->flags)) 2139 drbd_device_post_work(device, GO_DISKLESS); 2140 wake_up(&device->misc_wait); 2141 } 2142 } 2143 2144 #ifndef __CHECKER__ 2145 static inline int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins) 2146 { 2147 int io_allowed; 2148 2149 /* never get a reference while D_DISKLESS */ 2150 if (device->state.disk == D_DISKLESS) 2151 return 0; 2152 2153 atomic_inc(&device->local_cnt); 2154 io_allowed = (device->state.disk >= mins); 2155 if (!io_allowed) 2156 put_ldev(device); 2157 return io_allowed; 2158 } 2159 #else 2160 extern int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins); 2161 #endif 2162 2163 /* this throttles on-the-fly application requests 2164 * according to max_buffers settings; 2165 * maybe re-implement using semaphores? */ 2166 static inline int drbd_get_max_buffers(struct drbd_device *device) 2167 { 2168 struct net_conf *nc; 2169 int mxb; 2170 2171 rcu_read_lock(); 2172 nc = rcu_dereference(first_peer_device(device)->connection->net_conf); 2173 mxb = nc ? nc->max_buffers : 1000000; /* arbitrary limit on open requests */ 2174 rcu_read_unlock(); 2175 2176 return mxb; 2177 } 2178 2179 static inline int drbd_state_is_stable(struct drbd_device *device) 2180 { 2181 union drbd_dev_state s = device->state; 2182 2183 /* DO NOT add a default clause, we want the compiler to warn us 2184 * for any newly introduced state we may have forgotten to add here */ 2185 2186 switch ((enum drbd_conns)s.conn) { 2187 /* new io only accepted when there is no connection, ... */ 2188 case C_STANDALONE: 2189 case C_WF_CONNECTION: 2190 /* ... or there is a well established connection. */ 2191 case C_CONNECTED: 2192 case C_SYNC_SOURCE: 2193 case C_SYNC_TARGET: 2194 case C_VERIFY_S: 2195 case C_VERIFY_T: 2196 case C_PAUSED_SYNC_S: 2197 case C_PAUSED_SYNC_T: 2198 case C_AHEAD: 2199 case C_BEHIND: 2200 /* transitional states, IO allowed */ 2201 case C_DISCONNECTING: 2202 case C_UNCONNECTED: 2203 case C_TIMEOUT: 2204 case C_BROKEN_PIPE: 2205 case C_NETWORK_FAILURE: 2206 case C_PROTOCOL_ERROR: 2207 case C_TEAR_DOWN: 2208 case C_WF_REPORT_PARAMS: 2209 case C_STARTING_SYNC_S: 2210 case C_STARTING_SYNC_T: 2211 break; 2212 2213 /* Allow IO in BM exchange states with new protocols */ 2214 case C_WF_BITMAP_S: 2215 if (first_peer_device(device)->connection->agreed_pro_version < 96) 2216 return 0; 2217 break; 2218 2219 /* no new io accepted in these states */ 2220 case C_WF_BITMAP_T: 2221 case C_WF_SYNC_UUID: 2222 case C_MASK: 2223 /* not "stable" */ 2224 return 0; 2225 } 2226 2227 switch ((enum drbd_disk_state)s.disk) { 2228 case D_DISKLESS: 2229 case D_INCONSISTENT: 2230 case D_OUTDATED: 2231 case D_CONSISTENT: 2232 case D_UP_TO_DATE: 2233 case D_FAILED: 2234 /* disk state is stable as well. */ 2235 break; 2236 2237 /* no new io accepted during transitional states */ 2238 case D_ATTACHING: 2239 case D_NEGOTIATING: 2240 case D_UNKNOWN: 2241 case D_MASK: 2242 /* not "stable" */ 2243 return 0; 2244 } 2245 2246 return 1; 2247 } 2248 2249 static inline int drbd_suspended(struct drbd_device *device) 2250 { 2251 struct drbd_resource *resource = device->resource; 2252 2253 return resource->susp || resource->susp_fen || resource->susp_nod; 2254 } 2255 2256 static inline bool may_inc_ap_bio(struct drbd_device *device) 2257 { 2258 int mxb = drbd_get_max_buffers(device); 2259 2260 if (drbd_suspended(device)) 2261 return false; 2262 if (atomic_read(&device->suspend_cnt)) 2263 return false; 2264 2265 /* to avoid potential deadlock or bitmap corruption, 2266 * in various places, we only allow new application io 2267 * to start during "stable" states. */ 2268 2269 /* no new io accepted when attaching or detaching the disk */ 2270 if (!drbd_state_is_stable(device)) 2271 return false; 2272 2273 /* since some older kernels don't have atomic_add_unless, 2274 * and we are within the spinlock anyways, we have this workaround. */ 2275 if (atomic_read(&device->ap_bio_cnt) > mxb) 2276 return false; 2277 if (test_bit(BITMAP_IO, &device->flags)) 2278 return false; 2279 return true; 2280 } 2281 2282 static inline bool inc_ap_bio_cond(struct drbd_device *device) 2283 { 2284 bool rv = false; 2285 2286 spin_lock_irq(&device->resource->req_lock); 2287 rv = may_inc_ap_bio(device); 2288 if (rv) 2289 atomic_inc(&device->ap_bio_cnt); 2290 spin_unlock_irq(&device->resource->req_lock); 2291 2292 return rv; 2293 } 2294 2295 static inline void inc_ap_bio(struct drbd_device *device) 2296 { 2297 /* we wait here 2298 * as long as the device is suspended 2299 * until the bitmap is no longer on the fly during connection 2300 * handshake as long as we would exceed the max_buffer limit. 2301 * 2302 * to avoid races with the reconnect code, 2303 * we need to atomic_inc within the spinlock. */ 2304 2305 wait_event(device->misc_wait, inc_ap_bio_cond(device)); 2306 } 2307 2308 static inline void dec_ap_bio(struct drbd_device *device) 2309 { 2310 int mxb = drbd_get_max_buffers(device); 2311 int ap_bio = atomic_dec_return(&device->ap_bio_cnt); 2312 2313 D_ASSERT(device, ap_bio >= 0); 2314 2315 if (ap_bio == 0 && test_bit(BITMAP_IO, &device->flags)) { 2316 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags)) 2317 drbd_queue_work(&first_peer_device(device)-> 2318 connection->sender_work, 2319 &device->bm_io_work.w); 2320 } 2321 2322 /* this currently does wake_up for every dec_ap_bio! 2323 * maybe rather introduce some type of hysteresis? 2324 * e.g. (ap_bio == mxb/2 || ap_bio == 0) ? */ 2325 if (ap_bio < mxb) 2326 wake_up(&device->misc_wait); 2327 } 2328 2329 static inline bool verify_can_do_stop_sector(struct drbd_device *device) 2330 { 2331 return first_peer_device(device)->connection->agreed_pro_version >= 97 && 2332 first_peer_device(device)->connection->agreed_pro_version != 100; 2333 } 2334 2335 static inline int drbd_set_ed_uuid(struct drbd_device *device, u64 val) 2336 { 2337 int changed = device->ed_uuid != val; 2338 device->ed_uuid = val; 2339 return changed; 2340 } 2341 2342 static inline int drbd_queue_order_type(struct drbd_device *device) 2343 { 2344 /* sorry, we currently have no working implementation 2345 * of distributed TCQ stuff */ 2346 #ifndef QUEUE_ORDERED_NONE 2347 #define QUEUE_ORDERED_NONE 0 2348 #endif 2349 return QUEUE_ORDERED_NONE; 2350 } 2351 2352 static inline struct drbd_connection *first_connection(struct drbd_resource *resource) 2353 { 2354 return list_first_entry_or_null(&resource->connections, 2355 struct drbd_connection, connections); 2356 } 2357 2358 #endif 2359