1 /* 2 drbd_int.h 3 4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg. 5 6 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. 7 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. 8 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. 9 10 drbd is free software; you can redistribute it and/or modify 11 it under the terms of the GNU General Public License as published by 12 the Free Software Foundation; either version 2, or (at your option) 13 any later version. 14 15 drbd is distributed in the hope that it will be useful, 16 but WITHOUT ANY WARRANTY; without even the implied warranty of 17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 18 GNU General Public License for more details. 19 20 You should have received a copy of the GNU General Public License 21 along with drbd; see the file COPYING. If not, write to 22 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. 23 24 */ 25 26 #ifndef _DRBD_INT_H 27 #define _DRBD_INT_H 28 29 #include <linux/compiler.h> 30 #include <linux/types.h> 31 #include <linux/list.h> 32 #include <linux/sched.h> 33 #include <linux/bitops.h> 34 #include <linux/slab.h> 35 #include <linux/crypto.h> 36 #include <linux/ratelimit.h> 37 #include <linux/tcp.h> 38 #include <linux/mutex.h> 39 #include <linux/major.h> 40 #include <linux/blkdev.h> 41 #include <linux/genhd.h> 42 #include <linux/idr.h> 43 #include <net/tcp.h> 44 #include <linux/lru_cache.h> 45 #include <linux/prefetch.h> 46 #include <linux/drbd_genl_api.h> 47 #include <linux/drbd.h> 48 #include "drbd_strings.h" 49 #include "drbd_state.h" 50 #include "drbd_protocol.h" 51 52 #ifdef __CHECKER__ 53 # define __protected_by(x) __attribute__((require_context(x,1,999,"rdwr"))) 54 # define __protected_read_by(x) __attribute__((require_context(x,1,999,"read"))) 55 # define __protected_write_by(x) __attribute__((require_context(x,1,999,"write"))) 56 # define __must_hold(x) __attribute__((context(x,1,1), require_context(x,1,999,"call"))) 57 #else 58 # define __protected_by(x) 59 # define __protected_read_by(x) 60 # define __protected_write_by(x) 61 # define __must_hold(x) 62 #endif 63 64 #define __no_warn(lock, stmt) do { __acquire(lock); stmt; __release(lock); } while (0) 65 66 /* module parameter, defined in drbd_main.c */ 67 extern unsigned int minor_count; 68 extern bool disable_sendpage; 69 extern bool allow_oos; 70 void tl_abort_disk_io(struct drbd_device *device); 71 72 #ifdef CONFIG_DRBD_FAULT_INJECTION 73 extern int enable_faults; 74 extern int fault_rate; 75 extern int fault_devs; 76 #endif 77 78 extern char usermode_helper[]; 79 80 81 /* I don't remember why XCPU ... 82 * This is used to wake the asender, 83 * and to interrupt sending the sending task 84 * on disconnect. 85 */ 86 #define DRBD_SIG SIGXCPU 87 88 /* This is used to stop/restart our threads. 89 * Cannot use SIGTERM nor SIGKILL, since these 90 * are sent out by init on runlevel changes 91 * I choose SIGHUP for now. 92 */ 93 #define DRBD_SIGKILL SIGHUP 94 95 #define ID_IN_SYNC (4711ULL) 96 #define ID_OUT_OF_SYNC (4712ULL) 97 #define ID_SYNCER (-1ULL) 98 99 #define UUID_NEW_BM_OFFSET ((u64)0x0001000000000000ULL) 100 101 struct drbd_device; 102 struct drbd_connection; 103 104 #define __drbd_printk_device(level, device, fmt, args...) \ 105 dev_printk(level, disk_to_dev((device)->vdisk), fmt, ## args) 106 #define __drbd_printk_peer_device(level, peer_device, fmt, args...) \ 107 dev_printk(level, disk_to_dev((peer_device)->device->vdisk), fmt, ## args) 108 #define __drbd_printk_resource(level, resource, fmt, args...) \ 109 printk(level "drbd %s: " fmt, (resource)->name, ## args) 110 #define __drbd_printk_connection(level, connection, fmt, args...) \ 111 printk(level "drbd %s: " fmt, (connection)->resource->name, ## args) 112 113 void drbd_printk_with_wrong_object_type(void); 114 115 #define __drbd_printk_if_same_type(obj, type, func, level, fmt, args...) \ 116 (__builtin_types_compatible_p(typeof(obj), type) || \ 117 __builtin_types_compatible_p(typeof(obj), const type)), \ 118 func(level, (const type)(obj), fmt, ## args) 119 120 #define drbd_printk(level, obj, fmt, args...) \ 121 __builtin_choose_expr( \ 122 __drbd_printk_if_same_type(obj, struct drbd_device *, \ 123 __drbd_printk_device, level, fmt, ## args), \ 124 __builtin_choose_expr( \ 125 __drbd_printk_if_same_type(obj, struct drbd_resource *, \ 126 __drbd_printk_resource, level, fmt, ## args), \ 127 __builtin_choose_expr( \ 128 __drbd_printk_if_same_type(obj, struct drbd_connection *, \ 129 __drbd_printk_connection, level, fmt, ## args), \ 130 __builtin_choose_expr( \ 131 __drbd_printk_if_same_type(obj, struct drbd_peer_device *, \ 132 __drbd_printk_peer_device, level, fmt, ## args), \ 133 drbd_printk_with_wrong_object_type())))) 134 135 #define drbd_dbg(obj, fmt, args...) \ 136 drbd_printk(KERN_DEBUG, obj, fmt, ## args) 137 #define drbd_alert(obj, fmt, args...) \ 138 drbd_printk(KERN_ALERT, obj, fmt, ## args) 139 #define drbd_err(obj, fmt, args...) \ 140 drbd_printk(KERN_ERR, obj, fmt, ## args) 141 #define drbd_warn(obj, fmt, args...) \ 142 drbd_printk(KERN_WARNING, obj, fmt, ## args) 143 #define drbd_info(obj, fmt, args...) \ 144 drbd_printk(KERN_INFO, obj, fmt, ## args) 145 #define drbd_emerg(obj, fmt, args...) \ 146 drbd_printk(KERN_EMERG, obj, fmt, ## args) 147 148 #define dynamic_drbd_dbg(device, fmt, args...) \ 149 dynamic_dev_dbg(disk_to_dev(device->vdisk), fmt, ## args) 150 151 #define D_ASSERT(device, exp) do { \ 152 if (!(exp)) \ 153 drbd_err(device, "ASSERT( " #exp " ) in %s:%d\n", __FILE__, __LINE__); \ 154 } while (0) 155 156 /** 157 * expect - Make an assertion 158 * 159 * Unlike the assert macro, this macro returns a boolean result. 160 */ 161 #define expect(exp) ({ \ 162 bool _bool = (exp); \ 163 if (!_bool) \ 164 drbd_err(device, "ASSERTION %s FAILED in %s\n", \ 165 #exp, __func__); \ 166 _bool; \ 167 }) 168 169 /* Defines to control fault insertion */ 170 enum { 171 DRBD_FAULT_MD_WR = 0, /* meta data write */ 172 DRBD_FAULT_MD_RD = 1, /* read */ 173 DRBD_FAULT_RS_WR = 2, /* resync */ 174 DRBD_FAULT_RS_RD = 3, 175 DRBD_FAULT_DT_WR = 4, /* data */ 176 DRBD_FAULT_DT_RD = 5, 177 DRBD_FAULT_DT_RA = 6, /* data read ahead */ 178 DRBD_FAULT_BM_ALLOC = 7, /* bitmap allocation */ 179 DRBD_FAULT_AL_EE = 8, /* alloc ee */ 180 DRBD_FAULT_RECEIVE = 9, /* Changes some bytes upon receiving a [rs]data block */ 181 182 DRBD_FAULT_MAX, 183 }; 184 185 extern unsigned int 186 _drbd_insert_fault(struct drbd_device *device, unsigned int type); 187 188 static inline int 189 drbd_insert_fault(struct drbd_device *device, unsigned int type) { 190 #ifdef CONFIG_DRBD_FAULT_INJECTION 191 return fault_rate && 192 (enable_faults & (1<<type)) && 193 _drbd_insert_fault(device, type); 194 #else 195 return 0; 196 #endif 197 } 198 199 /* integer division, round _UP_ to the next integer */ 200 #define div_ceil(A, B) ((A)/(B) + ((A)%(B) ? 1 : 0)) 201 /* usual integer division */ 202 #define div_floor(A, B) ((A)/(B)) 203 204 extern struct ratelimit_state drbd_ratelimit_state; 205 extern struct idr drbd_devices; /* RCU, updates: genl_lock() */ 206 extern struct list_head drbd_resources; /* RCU, updates: genl_lock() */ 207 208 extern const char *cmdname(enum drbd_packet cmd); 209 210 /* for sending/receiving the bitmap, 211 * possibly in some encoding scheme */ 212 struct bm_xfer_ctx { 213 /* "const" 214 * stores total bits and long words 215 * of the bitmap, so we don't need to 216 * call the accessor functions over and again. */ 217 unsigned long bm_bits; 218 unsigned long bm_words; 219 /* during xfer, current position within the bitmap */ 220 unsigned long bit_offset; 221 unsigned long word_offset; 222 223 /* statistics; index: (h->command == P_BITMAP) */ 224 unsigned packets[2]; 225 unsigned bytes[2]; 226 }; 227 228 extern void INFO_bm_xfer_stats(struct drbd_device *device, 229 const char *direction, struct bm_xfer_ctx *c); 230 231 static inline void bm_xfer_ctx_bit_to_word_offset(struct bm_xfer_ctx *c) 232 { 233 /* word_offset counts "native long words" (32 or 64 bit), 234 * aligned at 64 bit. 235 * Encoded packet may end at an unaligned bit offset. 236 * In case a fallback clear text packet is transmitted in 237 * between, we adjust this offset back to the last 64bit 238 * aligned "native long word", which makes coding and decoding 239 * the plain text bitmap much more convenient. */ 240 #if BITS_PER_LONG == 64 241 c->word_offset = c->bit_offset >> 6; 242 #elif BITS_PER_LONG == 32 243 c->word_offset = c->bit_offset >> 5; 244 c->word_offset &= ~(1UL); 245 #else 246 # error "unsupported BITS_PER_LONG" 247 #endif 248 } 249 250 extern unsigned int drbd_header_size(struct drbd_connection *connection); 251 252 /**********************************************************************/ 253 enum drbd_thread_state { 254 NONE, 255 RUNNING, 256 EXITING, 257 RESTARTING 258 }; 259 260 struct drbd_thread { 261 spinlock_t t_lock; 262 struct task_struct *task; 263 struct completion stop; 264 enum drbd_thread_state t_state; 265 int (*function) (struct drbd_thread *); 266 struct drbd_resource *resource; 267 struct drbd_connection *connection; 268 int reset_cpu_mask; 269 const char *name; 270 }; 271 272 static inline enum drbd_thread_state get_t_state(struct drbd_thread *thi) 273 { 274 /* THINK testing the t_state seems to be uncritical in all cases 275 * (but thread_{start,stop}), so we can read it *without* the lock. 276 * --lge */ 277 278 smp_rmb(); 279 return thi->t_state; 280 } 281 282 struct drbd_work { 283 struct list_head list; 284 int (*cb)(struct drbd_work *, int cancel); 285 }; 286 287 struct drbd_device_work { 288 struct drbd_work w; 289 struct drbd_device *device; 290 }; 291 292 #include "drbd_interval.h" 293 294 extern int drbd_wait_misc(struct drbd_device *, struct drbd_interval *); 295 296 struct drbd_request { 297 struct drbd_work w; 298 struct drbd_device *device; 299 300 /* if local IO is not allowed, will be NULL. 301 * if local IO _is_ allowed, holds the locally submitted bio clone, 302 * or, after local IO completion, the ERR_PTR(error). 303 * see drbd_request_endio(). */ 304 struct bio *private_bio; 305 306 struct drbd_interval i; 307 308 /* epoch: used to check on "completion" whether this req was in 309 * the current epoch, and we therefore have to close it, 310 * causing a p_barrier packet to be send, starting a new epoch. 311 * 312 * This corresponds to "barrier" in struct p_barrier[_ack], 313 * and to "barrier_nr" in struct drbd_epoch (and various 314 * comments/function parameters/local variable names). 315 */ 316 unsigned int epoch; 317 318 struct list_head tl_requests; /* ring list in the transfer log */ 319 struct bio *master_bio; /* master bio pointer */ 320 321 /* see struct drbd_device */ 322 struct list_head req_pending_master_completion; 323 struct list_head req_pending_local; 324 325 /* for generic IO accounting */ 326 unsigned long start_jif; 327 328 /* for DRBD internal statistics */ 329 330 /* Minimal set of time stamps to determine if we wait for activity log 331 * transactions, local disk or peer. 32 bit "jiffies" are good enough, 332 * we don't expect a DRBD request to be stalled for several month. 333 */ 334 335 /* before actual request processing */ 336 unsigned long in_actlog_jif; 337 338 /* local disk */ 339 unsigned long pre_submit_jif; 340 341 /* per connection */ 342 unsigned long pre_send_jif; 343 unsigned long acked_jif; 344 unsigned long net_done_jif; 345 346 /* Possibly even more detail to track each phase: 347 * master_completion_jif 348 * how long did it take to complete the master bio 349 * (application visible latency) 350 * allocated_jif 351 * how long the master bio was blocked until we finally allocated 352 * a tracking struct 353 * in_actlog_jif 354 * how long did we wait for activity log transactions 355 * 356 * net_queued_jif 357 * when did we finally queue it for sending 358 * pre_send_jif 359 * when did we start sending it 360 * post_send_jif 361 * how long did we block in the network stack trying to send it 362 * acked_jif 363 * when did we receive (or fake, in protocol A) a remote ACK 364 * net_done_jif 365 * when did we receive final acknowledgement (P_BARRIER_ACK), 366 * or decide, e.g. on connection loss, that we do no longer expect 367 * anything from this peer for this request. 368 * 369 * pre_submit_jif 370 * post_sub_jif 371 * when did we start submiting to the lower level device, 372 * and how long did we block in that submit function 373 * local_completion_jif 374 * how long did it take the lower level device to complete this request 375 */ 376 377 378 /* once it hits 0, we may complete the master_bio */ 379 atomic_t completion_ref; 380 /* once it hits 0, we may destroy this drbd_request object */ 381 struct kref kref; 382 383 unsigned rq_state; /* see comments above _req_mod() */ 384 }; 385 386 struct drbd_epoch { 387 struct drbd_connection *connection; 388 struct list_head list; 389 unsigned int barrier_nr; 390 atomic_t epoch_size; /* increased on every request added. */ 391 atomic_t active; /* increased on every req. added, and dec on every finished. */ 392 unsigned long flags; 393 }; 394 395 /* Prototype declaration of function defined in drbd_receiver.c */ 396 int drbdd_init(struct drbd_thread *); 397 int drbd_asender(struct drbd_thread *); 398 399 /* drbd_epoch flag bits */ 400 enum { 401 DE_HAVE_BARRIER_NUMBER, 402 }; 403 404 enum epoch_event { 405 EV_PUT, 406 EV_GOT_BARRIER_NR, 407 EV_BECAME_LAST, 408 EV_CLEANUP = 32, /* used as flag */ 409 }; 410 411 struct digest_info { 412 int digest_size; 413 void *digest; 414 }; 415 416 struct drbd_peer_request { 417 struct drbd_work w; 418 struct drbd_peer_device *peer_device; 419 struct drbd_epoch *epoch; /* for writes */ 420 struct page *pages; 421 atomic_t pending_bios; 422 struct drbd_interval i; 423 /* see comments on ee flag bits below */ 424 unsigned long flags; 425 unsigned long submit_jif; 426 union { 427 u64 block_id; 428 struct digest_info *digest; 429 }; 430 }; 431 432 /* ee flag bits. 433 * While corresponding bios are in flight, the only modification will be 434 * set_bit WAS_ERROR, which has to be atomic. 435 * If no bios are in flight yet, or all have been completed, 436 * non-atomic modification to ee->flags is ok. 437 */ 438 enum { 439 __EE_CALL_AL_COMPLETE_IO, 440 __EE_MAY_SET_IN_SYNC, 441 442 /* is this a TRIM aka REQ_DISCARD? */ 443 __EE_IS_TRIM, 444 /* our lower level cannot handle trim, 445 * and we want to fall back to zeroout instead */ 446 __EE_IS_TRIM_USE_ZEROOUT, 447 448 /* In case a barrier failed, 449 * we need to resubmit without the barrier flag. */ 450 __EE_RESUBMITTED, 451 452 /* we may have several bios per peer request. 453 * if any of those fail, we set this flag atomically 454 * from the endio callback */ 455 __EE_WAS_ERROR, 456 457 /* This ee has a pointer to a digest instead of a block id */ 458 __EE_HAS_DIGEST, 459 460 /* Conflicting local requests need to be restarted after this request */ 461 __EE_RESTART_REQUESTS, 462 463 /* The peer wants a write ACK for this (wire proto C) */ 464 __EE_SEND_WRITE_ACK, 465 466 /* Is set when net_conf had two_primaries set while creating this peer_req */ 467 __EE_IN_INTERVAL_TREE, 468 469 /* for debugfs: */ 470 /* has this been submitted, or does it still wait for something else? */ 471 __EE_SUBMITTED, 472 473 /* this is/was a write request */ 474 __EE_WRITE, 475 476 /* this originates from application on peer 477 * (not some resync or verify or other DRBD internal request) */ 478 __EE_APPLICATION, 479 }; 480 #define EE_CALL_AL_COMPLETE_IO (1<<__EE_CALL_AL_COMPLETE_IO) 481 #define EE_MAY_SET_IN_SYNC (1<<__EE_MAY_SET_IN_SYNC) 482 #define EE_IS_TRIM (1<<__EE_IS_TRIM) 483 #define EE_IS_TRIM_USE_ZEROOUT (1<<__EE_IS_TRIM_USE_ZEROOUT) 484 #define EE_RESUBMITTED (1<<__EE_RESUBMITTED) 485 #define EE_WAS_ERROR (1<<__EE_WAS_ERROR) 486 #define EE_HAS_DIGEST (1<<__EE_HAS_DIGEST) 487 #define EE_RESTART_REQUESTS (1<<__EE_RESTART_REQUESTS) 488 #define EE_SEND_WRITE_ACK (1<<__EE_SEND_WRITE_ACK) 489 #define EE_IN_INTERVAL_TREE (1<<__EE_IN_INTERVAL_TREE) 490 #define EE_SUBMITTED (1<<__EE_SUBMITTED) 491 #define EE_WRITE (1<<__EE_WRITE) 492 #define EE_APPLICATION (1<<__EE_APPLICATION) 493 494 /* flag bits per device */ 495 enum { 496 UNPLUG_REMOTE, /* sending a "UnplugRemote" could help */ 497 MD_DIRTY, /* current uuids and flags not yet on disk */ 498 USE_DEGR_WFC_T, /* degr-wfc-timeout instead of wfc-timeout. */ 499 CL_ST_CHG_SUCCESS, 500 CL_ST_CHG_FAIL, 501 CRASHED_PRIMARY, /* This node was a crashed primary. 502 * Gets cleared when the state.conn 503 * goes into C_CONNECTED state. */ 504 CONSIDER_RESYNC, 505 506 MD_NO_FUA, /* Users wants us to not use FUA/FLUSH on meta data dev */ 507 508 SUSPEND_IO, /* suspend application io */ 509 BITMAP_IO, /* suspend application io; 510 once no more io in flight, start bitmap io */ 511 BITMAP_IO_QUEUED, /* Started bitmap IO */ 512 WAS_IO_ERROR, /* Local disk failed, returned IO error */ 513 WAS_READ_ERROR, /* Local disk READ failed (set additionally to the above) */ 514 FORCE_DETACH, /* Force-detach from local disk, aborting any pending local IO */ 515 RESYNC_AFTER_NEG, /* Resync after online grow after the attach&negotiate finished. */ 516 RESIZE_PENDING, /* Size change detected locally, waiting for the response from 517 * the peer, if it changed there as well. */ 518 NEW_CUR_UUID, /* Create new current UUID when thawing IO */ 519 AL_SUSPENDED, /* Activity logging is currently suspended. */ 520 AHEAD_TO_SYNC_SOURCE, /* Ahead -> SyncSource queued */ 521 B_RS_H_DONE, /* Before resync handler done (already executed) */ 522 DISCARD_MY_DATA, /* discard_my_data flag per volume */ 523 READ_BALANCE_RR, 524 525 FLUSH_PENDING, /* if set, device->flush_jif is when we submitted that flush 526 * from drbd_flush_after_epoch() */ 527 528 /* cleared only after backing device related structures have been destroyed. */ 529 GOING_DISKLESS, /* Disk is being detached, because of io-error, or admin request. */ 530 531 /* to be used in drbd_device_post_work() */ 532 GO_DISKLESS, /* tell worker to schedule cleanup before detach */ 533 DESTROY_DISK, /* tell worker to close backing devices and destroy related structures. */ 534 MD_SYNC, /* tell worker to call drbd_md_sync() */ 535 RS_START, /* tell worker to start resync/OV */ 536 RS_PROGRESS, /* tell worker that resync made significant progress */ 537 RS_DONE, /* tell worker that resync is done */ 538 }; 539 540 struct drbd_bitmap; /* opaque for drbd_device */ 541 542 /* definition of bits in bm_flags to be used in drbd_bm_lock 543 * and drbd_bitmap_io and friends. */ 544 enum bm_flag { 545 /* do we need to kfree, or vfree bm_pages? */ 546 BM_P_VMALLOCED = 0x10000, /* internal use only, will be masked out */ 547 548 /* currently locked for bulk operation */ 549 BM_LOCKED_MASK = 0xf, 550 551 /* in detail, that is: */ 552 BM_DONT_CLEAR = 0x1, 553 BM_DONT_SET = 0x2, 554 BM_DONT_TEST = 0x4, 555 556 /* so we can mark it locked for bulk operation, 557 * and still allow all non-bulk operations */ 558 BM_IS_LOCKED = 0x8, 559 560 /* (test bit, count bit) allowed (common case) */ 561 BM_LOCKED_TEST_ALLOWED = BM_DONT_CLEAR | BM_DONT_SET | BM_IS_LOCKED, 562 563 /* testing bits, as well as setting new bits allowed, but clearing bits 564 * would be unexpected. Used during bitmap receive. Setting new bits 565 * requires sending of "out-of-sync" information, though. */ 566 BM_LOCKED_SET_ALLOWED = BM_DONT_CLEAR | BM_IS_LOCKED, 567 568 /* for drbd_bm_write_copy_pages, everything is allowed, 569 * only concurrent bulk operations are locked out. */ 570 BM_LOCKED_CHANGE_ALLOWED = BM_IS_LOCKED, 571 }; 572 573 struct drbd_work_queue { 574 struct list_head q; 575 spinlock_t q_lock; /* to protect the list. */ 576 wait_queue_head_t q_wait; 577 }; 578 579 struct drbd_socket { 580 struct mutex mutex; 581 struct socket *socket; 582 /* this way we get our 583 * send/receive buffers off the stack */ 584 void *sbuf; 585 void *rbuf; 586 }; 587 588 struct drbd_md { 589 u64 md_offset; /* sector offset to 'super' block */ 590 591 u64 la_size_sect; /* last agreed size, unit sectors */ 592 spinlock_t uuid_lock; 593 u64 uuid[UI_SIZE]; 594 u64 device_uuid; 595 u32 flags; 596 u32 md_size_sect; 597 598 s32 al_offset; /* signed relative sector offset to activity log */ 599 s32 bm_offset; /* signed relative sector offset to bitmap */ 600 601 /* cached value of bdev->disk_conf->meta_dev_idx (see below) */ 602 s32 meta_dev_idx; 603 604 /* see al_tr_number_to_on_disk_sector() */ 605 u32 al_stripes; 606 u32 al_stripe_size_4k; 607 u32 al_size_4k; /* cached product of the above */ 608 }; 609 610 struct drbd_backing_dev { 611 struct block_device *backing_bdev; 612 struct block_device *md_bdev; 613 struct drbd_md md; 614 struct disk_conf *disk_conf; /* RCU, for updates: resource->conf_update */ 615 sector_t known_size; /* last known size of that backing device */ 616 }; 617 618 struct drbd_md_io { 619 struct page *page; 620 unsigned long start_jif; /* last call to drbd_md_get_buffer */ 621 unsigned long submit_jif; /* last _drbd_md_sync_page_io() submit */ 622 const char *current_use; 623 atomic_t in_use; 624 unsigned int done; 625 int error; 626 }; 627 628 struct bm_io_work { 629 struct drbd_work w; 630 char *why; 631 enum bm_flag flags; 632 int (*io_fn)(struct drbd_device *device); 633 void (*done)(struct drbd_device *device, int rv); 634 }; 635 636 enum write_ordering_e { 637 WO_none, 638 WO_drain_io, 639 WO_bdev_flush, 640 }; 641 642 struct fifo_buffer { 643 unsigned int head_index; 644 unsigned int size; 645 int total; /* sum of all values */ 646 int values[0]; 647 }; 648 extern struct fifo_buffer *fifo_alloc(int fifo_size); 649 650 /* flag bits per connection */ 651 enum { 652 NET_CONGESTED, /* The data socket is congested */ 653 RESOLVE_CONFLICTS, /* Set on one node, cleared on the peer! */ 654 SEND_PING, /* whether asender should send a ping asap */ 655 SIGNAL_ASENDER, /* whether asender wants to be interrupted */ 656 GOT_PING_ACK, /* set when we receive a ping_ack packet, ping_wait gets woken */ 657 CONN_WD_ST_CHG_REQ, /* A cluster wide state change on the connection is active */ 658 CONN_WD_ST_CHG_OKAY, 659 CONN_WD_ST_CHG_FAIL, 660 CONN_DRY_RUN, /* Expect disconnect after resync handshake. */ 661 CREATE_BARRIER, /* next P_DATA is preceded by a P_BARRIER */ 662 STATE_SENT, /* Do not change state/UUIDs while this is set */ 663 CALLBACK_PENDING, /* Whether we have a call_usermodehelper(, UMH_WAIT_PROC) 664 * pending, from drbd worker context. 665 * If set, bdi_write_congested() returns true, 666 * so shrink_page_list() would not recurse into, 667 * and potentially deadlock on, this drbd worker. 668 */ 669 DISCONNECT_SENT, 670 671 DEVICE_WORK_PENDING, /* tell worker that some device has pending work */ 672 }; 673 674 struct drbd_resource { 675 char *name; 676 #ifdef CONFIG_DEBUG_FS 677 struct dentry *debugfs_res; 678 struct dentry *debugfs_res_volumes; 679 struct dentry *debugfs_res_connections; 680 struct dentry *debugfs_res_in_flight_summary; 681 #endif 682 struct kref kref; 683 struct idr devices; /* volume number to device mapping */ 684 struct list_head connections; 685 struct list_head resources; 686 struct res_opts res_opts; 687 struct mutex conf_update; /* mutex for ready-copy-update of net_conf and disk_conf */ 688 struct mutex adm_mutex; /* mutex to serialize administrative requests */ 689 spinlock_t req_lock; 690 691 unsigned susp:1; /* IO suspended by user */ 692 unsigned susp_nod:1; /* IO suspended because no data */ 693 unsigned susp_fen:1; /* IO suspended because fence peer handler runs */ 694 695 enum write_ordering_e write_ordering; 696 697 cpumask_var_t cpu_mask; 698 }; 699 700 struct drbd_thread_timing_details 701 { 702 unsigned long start_jif; 703 void *cb_addr; 704 const char *caller_fn; 705 unsigned int line; 706 unsigned int cb_nr; 707 }; 708 709 struct drbd_connection { 710 struct list_head connections; 711 struct drbd_resource *resource; 712 #ifdef CONFIG_DEBUG_FS 713 struct dentry *debugfs_conn; 714 struct dentry *debugfs_conn_callback_history; 715 struct dentry *debugfs_conn_oldest_requests; 716 #endif 717 struct kref kref; 718 struct idr peer_devices; /* volume number to peer device mapping */ 719 enum drbd_conns cstate; /* Only C_STANDALONE to C_WF_REPORT_PARAMS */ 720 struct mutex cstate_mutex; /* Protects graceful disconnects */ 721 unsigned int connect_cnt; /* Inc each time a connection is established */ 722 723 unsigned long flags; 724 struct net_conf *net_conf; /* content protected by rcu */ 725 wait_queue_head_t ping_wait; /* Woken upon reception of a ping, and a state change */ 726 727 struct sockaddr_storage my_addr; 728 int my_addr_len; 729 struct sockaddr_storage peer_addr; 730 int peer_addr_len; 731 732 struct drbd_socket data; /* data/barrier/cstate/parameter packets */ 733 struct drbd_socket meta; /* ping/ack (metadata) packets */ 734 int agreed_pro_version; /* actually used protocol version */ 735 u32 agreed_features; 736 unsigned long last_received; /* in jiffies, either socket */ 737 unsigned int ko_count; 738 739 struct list_head transfer_log; /* all requests not yet fully processed */ 740 741 struct crypto_hash *cram_hmac_tfm; 742 struct crypto_hash *integrity_tfm; /* checksums we compute, updates protected by connection->data->mutex */ 743 struct crypto_hash *peer_integrity_tfm; /* checksums we verify, only accessed from receiver thread */ 744 struct crypto_hash *csums_tfm; 745 struct crypto_hash *verify_tfm; 746 void *int_dig_in; 747 void *int_dig_vv; 748 749 /* receiver side */ 750 struct drbd_epoch *current_epoch; 751 spinlock_t epoch_lock; 752 unsigned int epochs; 753 atomic_t current_tle_nr; /* transfer log epoch number */ 754 unsigned current_tle_writes; /* writes seen within this tl epoch */ 755 756 unsigned long last_reconnect_jif; 757 struct drbd_thread receiver; 758 struct drbd_thread worker; 759 struct drbd_thread asender; 760 761 /* cached pointers, 762 * so we can look up the oldest pending requests more quickly. 763 * protected by resource->req_lock */ 764 struct drbd_request *req_next; /* DRBD 9: todo.req_next */ 765 struct drbd_request *req_ack_pending; 766 struct drbd_request *req_not_net_done; 767 768 /* sender side */ 769 struct drbd_work_queue sender_work; 770 771 #define DRBD_THREAD_DETAILS_HIST 16 772 unsigned int w_cb_nr; /* keeps counting up */ 773 unsigned int r_cb_nr; /* keeps counting up */ 774 struct drbd_thread_timing_details w_timing_details[DRBD_THREAD_DETAILS_HIST]; 775 struct drbd_thread_timing_details r_timing_details[DRBD_THREAD_DETAILS_HIST]; 776 777 struct { 778 /* whether this sender thread 779 * has processed a single write yet. */ 780 bool seen_any_write_yet; 781 782 /* Which barrier number to send with the next P_BARRIER */ 783 int current_epoch_nr; 784 785 /* how many write requests have been sent 786 * with req->epoch == current_epoch_nr. 787 * If none, no P_BARRIER will be sent. */ 788 unsigned current_epoch_writes; 789 } send; 790 }; 791 792 void __update_timing_details( 793 struct drbd_thread_timing_details *tdp, 794 unsigned int *cb_nr, 795 void *cb, 796 const char *fn, const unsigned int line); 797 798 #define update_worker_timing_details(c, cb) \ 799 __update_timing_details(c->w_timing_details, &c->w_cb_nr, cb, __func__ , __LINE__ ) 800 #define update_receiver_timing_details(c, cb) \ 801 __update_timing_details(c->r_timing_details, &c->r_cb_nr, cb, __func__ , __LINE__ ) 802 803 struct submit_worker { 804 struct workqueue_struct *wq; 805 struct work_struct worker; 806 807 /* protected by ..->resource->req_lock */ 808 struct list_head writes; 809 }; 810 811 struct drbd_peer_device { 812 struct list_head peer_devices; 813 struct drbd_device *device; 814 struct drbd_connection *connection; 815 #ifdef CONFIG_DEBUG_FS 816 struct dentry *debugfs_peer_dev; 817 #endif 818 }; 819 820 struct drbd_device { 821 struct drbd_resource *resource; 822 struct list_head peer_devices; 823 struct list_head pending_bitmap_io; 824 825 unsigned long flush_jif; 826 #ifdef CONFIG_DEBUG_FS 827 struct dentry *debugfs_minor; 828 struct dentry *debugfs_vol; 829 struct dentry *debugfs_vol_oldest_requests; 830 struct dentry *debugfs_vol_act_log_extents; 831 struct dentry *debugfs_vol_resync_extents; 832 struct dentry *debugfs_vol_data_gen_id; 833 #endif 834 835 unsigned int vnr; /* volume number within the connection */ 836 unsigned int minor; /* device minor number */ 837 838 struct kref kref; 839 840 /* things that are stored as / read from meta data on disk */ 841 unsigned long flags; 842 843 /* configured by drbdsetup */ 844 struct drbd_backing_dev *ldev __protected_by(local); 845 846 sector_t p_size; /* partner's disk size */ 847 struct request_queue *rq_queue; 848 struct block_device *this_bdev; 849 struct gendisk *vdisk; 850 851 unsigned long last_reattach_jif; 852 struct drbd_work resync_work; 853 struct drbd_work unplug_work; 854 struct timer_list resync_timer; 855 struct timer_list md_sync_timer; 856 struct timer_list start_resync_timer; 857 struct timer_list request_timer; 858 859 /* Used after attach while negotiating new disk state. */ 860 union drbd_state new_state_tmp; 861 862 union drbd_dev_state state; 863 wait_queue_head_t misc_wait; 864 wait_queue_head_t state_wait; /* upon each state change. */ 865 unsigned int send_cnt; 866 unsigned int recv_cnt; 867 unsigned int read_cnt; 868 unsigned int writ_cnt; 869 unsigned int al_writ_cnt; 870 unsigned int bm_writ_cnt; 871 atomic_t ap_bio_cnt; /* Requests we need to complete */ 872 atomic_t ap_actlog_cnt; /* Requests waiting for activity log */ 873 atomic_t ap_pending_cnt; /* AP data packets on the wire, ack expected */ 874 atomic_t rs_pending_cnt; /* RS request/data packets on the wire */ 875 atomic_t unacked_cnt; /* Need to send replies for */ 876 atomic_t local_cnt; /* Waiting for local completion */ 877 878 /* Interval tree of pending local requests */ 879 struct rb_root read_requests; 880 struct rb_root write_requests; 881 882 /* for statistics and timeouts */ 883 /* [0] read, [1] write */ 884 struct list_head pending_master_completion[2]; 885 struct list_head pending_completion[2]; 886 887 /* use checksums for *this* resync */ 888 bool use_csums; 889 /* blocks to resync in this run [unit BM_BLOCK_SIZE] */ 890 unsigned long rs_total; 891 /* number of resync blocks that failed in this run */ 892 unsigned long rs_failed; 893 /* Syncer's start time [unit jiffies] */ 894 unsigned long rs_start; 895 /* cumulated time in PausedSyncX state [unit jiffies] */ 896 unsigned long rs_paused; 897 /* skipped because csum was equal [unit BM_BLOCK_SIZE] */ 898 unsigned long rs_same_csum; 899 #define DRBD_SYNC_MARKS 8 900 #define DRBD_SYNC_MARK_STEP (3*HZ) 901 /* block not up-to-date at mark [unit BM_BLOCK_SIZE] */ 902 unsigned long rs_mark_left[DRBD_SYNC_MARKS]; 903 /* marks's time [unit jiffies] */ 904 unsigned long rs_mark_time[DRBD_SYNC_MARKS]; 905 /* current index into rs_mark_{left,time} */ 906 int rs_last_mark; 907 unsigned long rs_last_bcast; /* [unit jiffies] */ 908 909 /* where does the admin want us to start? (sector) */ 910 sector_t ov_start_sector; 911 sector_t ov_stop_sector; 912 /* where are we now? (sector) */ 913 sector_t ov_position; 914 /* Start sector of out of sync range (to merge printk reporting). */ 915 sector_t ov_last_oos_start; 916 /* size of out-of-sync range in sectors. */ 917 sector_t ov_last_oos_size; 918 unsigned long ov_left; /* in bits */ 919 920 struct drbd_bitmap *bitmap; 921 unsigned long bm_resync_fo; /* bit offset for drbd_bm_find_next */ 922 923 /* Used to track operations of resync... */ 924 struct lru_cache *resync; 925 /* Number of locked elements in resync LRU */ 926 unsigned int resync_locked; 927 /* resync extent number waiting for application requests */ 928 unsigned int resync_wenr; 929 930 int open_cnt; 931 u64 *p_uuid; 932 933 struct list_head active_ee; /* IO in progress (P_DATA gets written to disk) */ 934 struct list_head sync_ee; /* IO in progress (P_RS_DATA_REPLY gets written to disk) */ 935 struct list_head done_ee; /* need to send P_WRITE_ACK */ 936 struct list_head read_ee; /* [RS]P_DATA_REQUEST being read */ 937 struct list_head net_ee; /* zero-copy network send in progress */ 938 939 int next_barrier_nr; 940 struct list_head resync_reads; 941 atomic_t pp_in_use; /* allocated from page pool */ 942 atomic_t pp_in_use_by_net; /* sendpage()d, still referenced by tcp */ 943 wait_queue_head_t ee_wait; 944 struct drbd_md_io md_io; 945 spinlock_t al_lock; 946 wait_queue_head_t al_wait; 947 struct lru_cache *act_log; /* activity log */ 948 unsigned int al_tr_number; 949 int al_tr_cycle; 950 wait_queue_head_t seq_wait; 951 atomic_t packet_seq; 952 unsigned int peer_seq; 953 spinlock_t peer_seq_lock; 954 unsigned long comm_bm_set; /* communicated number of set bits. */ 955 struct bm_io_work bm_io_work; 956 u64 ed_uuid; /* UUID of the exposed data */ 957 struct mutex own_state_mutex; 958 struct mutex *state_mutex; /* either own_state_mutex or first_peer_device(device)->connection->cstate_mutex */ 959 char congestion_reason; /* Why we where congested... */ 960 atomic_t rs_sect_in; /* for incoming resync data rate, SyncTarget */ 961 atomic_t rs_sect_ev; /* for submitted resync data rate, both */ 962 int rs_last_sect_ev; /* counter to compare with */ 963 int rs_last_events; /* counter of read or write "events" (unit sectors) 964 * on the lower level device when we last looked. */ 965 int c_sync_rate; /* current resync rate after syncer throttle magic */ 966 struct fifo_buffer *rs_plan_s; /* correction values of resync planer (RCU, connection->conn_update) */ 967 int rs_in_flight; /* resync sectors in flight (to proxy, in proxy and from proxy) */ 968 atomic_t ap_in_flight; /* App sectors in flight (waiting for ack) */ 969 unsigned int peer_max_bio_size; 970 unsigned int local_max_bio_size; 971 972 /* any requests that would block in drbd_make_request() 973 * are deferred to this single-threaded work queue */ 974 struct submit_worker submit; 975 }; 976 977 struct drbd_bm_aio_ctx { 978 struct drbd_device *device; 979 struct list_head list; /* on device->pending_bitmap_io */; 980 unsigned long start_jif; 981 atomic_t in_flight; 982 unsigned int done; 983 unsigned flags; 984 #define BM_AIO_COPY_PAGES 1 985 #define BM_AIO_WRITE_HINTED 2 986 #define BM_AIO_WRITE_ALL_PAGES 4 987 #define BM_AIO_READ 8 988 int error; 989 struct kref kref; 990 }; 991 992 struct drbd_config_context { 993 /* assigned from drbd_genlmsghdr */ 994 unsigned int minor; 995 /* assigned from request attributes, if present */ 996 unsigned int volume; 997 #define VOLUME_UNSPECIFIED (-1U) 998 /* pointer into the request skb, 999 * limited lifetime! */ 1000 char *resource_name; 1001 struct nlattr *my_addr; 1002 struct nlattr *peer_addr; 1003 1004 /* reply buffer */ 1005 struct sk_buff *reply_skb; 1006 /* pointer into reply buffer */ 1007 struct drbd_genlmsghdr *reply_dh; 1008 /* resolved from attributes, if possible */ 1009 struct drbd_device *device; 1010 struct drbd_resource *resource; 1011 struct drbd_connection *connection; 1012 }; 1013 1014 static inline struct drbd_device *minor_to_device(unsigned int minor) 1015 { 1016 return (struct drbd_device *)idr_find(&drbd_devices, minor); 1017 } 1018 1019 static inline struct drbd_peer_device *first_peer_device(struct drbd_device *device) 1020 { 1021 return list_first_entry_or_null(&device->peer_devices, struct drbd_peer_device, peer_devices); 1022 } 1023 1024 #define for_each_resource(resource, _resources) \ 1025 list_for_each_entry(resource, _resources, resources) 1026 1027 #define for_each_resource_rcu(resource, _resources) \ 1028 list_for_each_entry_rcu(resource, _resources, resources) 1029 1030 #define for_each_resource_safe(resource, tmp, _resources) \ 1031 list_for_each_entry_safe(resource, tmp, _resources, resources) 1032 1033 #define for_each_connection(connection, resource) \ 1034 list_for_each_entry(connection, &resource->connections, connections) 1035 1036 #define for_each_connection_rcu(connection, resource) \ 1037 list_for_each_entry_rcu(connection, &resource->connections, connections) 1038 1039 #define for_each_connection_safe(connection, tmp, resource) \ 1040 list_for_each_entry_safe(connection, tmp, &resource->connections, connections) 1041 1042 #define for_each_peer_device(peer_device, device) \ 1043 list_for_each_entry(peer_device, &device->peer_devices, peer_devices) 1044 1045 #define for_each_peer_device_rcu(peer_device, device) \ 1046 list_for_each_entry_rcu(peer_device, &device->peer_devices, peer_devices) 1047 1048 #define for_each_peer_device_safe(peer_device, tmp, device) \ 1049 list_for_each_entry_safe(peer_device, tmp, &device->peer_devices, peer_devices) 1050 1051 static inline unsigned int device_to_minor(struct drbd_device *device) 1052 { 1053 return device->minor; 1054 } 1055 1056 /* 1057 * function declarations 1058 *************************/ 1059 1060 /* drbd_main.c */ 1061 1062 enum dds_flags { 1063 DDSF_FORCED = 1, 1064 DDSF_NO_RESYNC = 2, /* Do not run a resync for the new space */ 1065 }; 1066 1067 extern void drbd_init_set_defaults(struct drbd_device *device); 1068 extern int drbd_thread_start(struct drbd_thread *thi); 1069 extern void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait); 1070 #ifdef CONFIG_SMP 1071 extern void drbd_thread_current_set_cpu(struct drbd_thread *thi); 1072 #else 1073 #define drbd_thread_current_set_cpu(A) ({}) 1074 #endif 1075 extern void tl_release(struct drbd_connection *, unsigned int barrier_nr, 1076 unsigned int set_size); 1077 extern void tl_clear(struct drbd_connection *); 1078 extern void drbd_free_sock(struct drbd_connection *connection); 1079 extern int drbd_send(struct drbd_connection *connection, struct socket *sock, 1080 void *buf, size_t size, unsigned msg_flags); 1081 extern int drbd_send_all(struct drbd_connection *, struct socket *, void *, size_t, 1082 unsigned); 1083 1084 extern int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd); 1085 extern int drbd_send_protocol(struct drbd_connection *connection); 1086 extern int drbd_send_uuids(struct drbd_peer_device *); 1087 extern int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *); 1088 extern void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *); 1089 extern int drbd_send_sizes(struct drbd_peer_device *, int trigger_reply, enum dds_flags flags); 1090 extern int drbd_send_state(struct drbd_peer_device *, union drbd_state s); 1091 extern int drbd_send_current_state(struct drbd_peer_device *); 1092 extern int drbd_send_sync_param(struct drbd_peer_device *); 1093 extern void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, 1094 u32 set_size); 1095 extern int drbd_send_ack(struct drbd_peer_device *, enum drbd_packet, 1096 struct drbd_peer_request *); 1097 extern void drbd_send_ack_rp(struct drbd_peer_device *, enum drbd_packet, 1098 struct p_block_req *rp); 1099 extern void drbd_send_ack_dp(struct drbd_peer_device *, enum drbd_packet, 1100 struct p_data *dp, int data_size); 1101 extern int drbd_send_ack_ex(struct drbd_peer_device *, enum drbd_packet, 1102 sector_t sector, int blksize, u64 block_id); 1103 extern int drbd_send_out_of_sync(struct drbd_peer_device *, struct drbd_request *); 1104 extern int drbd_send_block(struct drbd_peer_device *, enum drbd_packet, 1105 struct drbd_peer_request *); 1106 extern int drbd_send_dblock(struct drbd_peer_device *, struct drbd_request *req); 1107 extern int drbd_send_drequest(struct drbd_peer_device *, int cmd, 1108 sector_t sector, int size, u64 block_id); 1109 extern int drbd_send_drequest_csum(struct drbd_peer_device *, sector_t sector, 1110 int size, void *digest, int digest_size, 1111 enum drbd_packet cmd); 1112 extern int drbd_send_ov_request(struct drbd_peer_device *, sector_t sector, int size); 1113 1114 extern int drbd_send_bitmap(struct drbd_device *device); 1115 extern void drbd_send_sr_reply(struct drbd_peer_device *, enum drbd_state_rv retcode); 1116 extern void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode); 1117 extern void drbd_free_ldev(struct drbd_backing_dev *ldev); 1118 extern void drbd_device_cleanup(struct drbd_device *device); 1119 void drbd_print_uuids(struct drbd_device *device, const char *text); 1120 1121 extern void conn_md_sync(struct drbd_connection *connection); 1122 extern void drbd_md_write(struct drbd_device *device, void *buffer); 1123 extern void drbd_md_sync(struct drbd_device *device); 1124 extern int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev); 1125 extern void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local); 1126 extern void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local); 1127 extern void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local); 1128 extern void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local); 1129 extern void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local); 1130 extern void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local); 1131 extern void drbd_md_set_flag(struct drbd_device *device, int flags) __must_hold(local); 1132 extern void drbd_md_clear_flag(struct drbd_device *device, int flags)__must_hold(local); 1133 extern int drbd_md_test_flag(struct drbd_backing_dev *, int); 1134 extern void drbd_md_mark_dirty(struct drbd_device *device); 1135 extern void drbd_queue_bitmap_io(struct drbd_device *device, 1136 int (*io_fn)(struct drbd_device *), 1137 void (*done)(struct drbd_device *, int), 1138 char *why, enum bm_flag flags); 1139 extern int drbd_bitmap_io(struct drbd_device *device, 1140 int (*io_fn)(struct drbd_device *), 1141 char *why, enum bm_flag flags); 1142 extern int drbd_bitmap_io_from_worker(struct drbd_device *device, 1143 int (*io_fn)(struct drbd_device *), 1144 char *why, enum bm_flag flags); 1145 extern int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local); 1146 extern int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local); 1147 1148 /* Meta data layout 1149 * 1150 * We currently have two possible layouts. 1151 * Offsets in (512 byte) sectors. 1152 * external: 1153 * |----------- md_size_sect ------------------| 1154 * [ 4k superblock ][ activity log ][ Bitmap ] 1155 * | al_offset == 8 | 1156 * | bm_offset = al_offset + X | 1157 * ==> bitmap sectors = md_size_sect - bm_offset 1158 * 1159 * Variants: 1160 * old, indexed fixed size meta data: 1161 * 1162 * internal: 1163 * |----------- md_size_sect ------------------| 1164 * [data.....][ Bitmap ][ activity log ][ 4k superblock ][padding*] 1165 * | al_offset < 0 | 1166 * | bm_offset = al_offset - Y | 1167 * ==> bitmap sectors = Y = al_offset - bm_offset 1168 * 1169 * [padding*] are zero or up to 7 unused 512 Byte sectors to the 1170 * end of the device, so that the [4k superblock] will be 4k aligned. 1171 * 1172 * The activity log consists of 4k transaction blocks, 1173 * which are written in a ring-buffer, or striped ring-buffer like fashion, 1174 * which are writtensize used to be fixed 32kB, 1175 * but is about to become configurable. 1176 */ 1177 1178 /* Our old fixed size meta data layout 1179 * allows up to about 3.8TB, so if you want more, 1180 * you need to use the "flexible" meta data format. */ 1181 #define MD_128MB_SECT (128LLU << 11) /* 128 MB, unit sectors */ 1182 #define MD_4kB_SECT 8 1183 #define MD_32kB_SECT 64 1184 1185 /* One activity log extent represents 4M of storage */ 1186 #define AL_EXTENT_SHIFT 22 1187 #define AL_EXTENT_SIZE (1<<AL_EXTENT_SHIFT) 1188 1189 /* We could make these currently hardcoded constants configurable 1190 * variables at create-md time (or even re-configurable at runtime?). 1191 * Which will require some more changes to the DRBD "super block" 1192 * and attach code. 1193 * 1194 * updates per transaction: 1195 * This many changes to the active set can be logged with one transaction. 1196 * This number is arbitrary. 1197 * context per transaction: 1198 * This many context extent numbers are logged with each transaction. 1199 * This number is resulting from the transaction block size (4k), the layout 1200 * of the transaction header, and the number of updates per transaction. 1201 * See drbd_actlog.c:struct al_transaction_on_disk 1202 * */ 1203 #define AL_UPDATES_PER_TRANSACTION 64 // arbitrary 1204 #define AL_CONTEXT_PER_TRANSACTION 919 // (4096 - 36 - 6*64)/4 1205 1206 #if BITS_PER_LONG == 32 1207 #define LN2_BPL 5 1208 #define cpu_to_lel(A) cpu_to_le32(A) 1209 #define lel_to_cpu(A) le32_to_cpu(A) 1210 #elif BITS_PER_LONG == 64 1211 #define LN2_BPL 6 1212 #define cpu_to_lel(A) cpu_to_le64(A) 1213 #define lel_to_cpu(A) le64_to_cpu(A) 1214 #else 1215 #error "LN2 of BITS_PER_LONG unknown!" 1216 #endif 1217 1218 /* resync bitmap */ 1219 /* 16MB sized 'bitmap extent' to track syncer usage */ 1220 struct bm_extent { 1221 int rs_left; /* number of bits set (out of sync) in this extent. */ 1222 int rs_failed; /* number of failed resync requests in this extent. */ 1223 unsigned long flags; 1224 struct lc_element lce; 1225 }; 1226 1227 #define BME_NO_WRITES 0 /* bm_extent.flags: no more requests on this one! */ 1228 #define BME_LOCKED 1 /* bm_extent.flags: syncer active on this one. */ 1229 #define BME_PRIORITY 2 /* finish resync IO on this extent ASAP! App IO waiting! */ 1230 1231 /* drbd_bitmap.c */ 1232 /* 1233 * We need to store one bit for a block. 1234 * Example: 1GB disk @ 4096 byte blocks ==> we need 32 KB bitmap. 1235 * Bit 0 ==> local node thinks this block is binary identical on both nodes 1236 * Bit 1 ==> local node thinks this block needs to be synced. 1237 */ 1238 1239 #define SLEEP_TIME (HZ/10) 1240 1241 /* We do bitmap IO in units of 4k blocks. 1242 * We also still have a hardcoded 4k per bit relation. */ 1243 #define BM_BLOCK_SHIFT 12 /* 4k per bit */ 1244 #define BM_BLOCK_SIZE (1<<BM_BLOCK_SHIFT) 1245 /* mostly arbitrarily set the represented size of one bitmap extent, 1246 * aka resync extent, to 16 MiB (which is also 512 Byte worth of bitmap 1247 * at 4k per bit resolution) */ 1248 #define BM_EXT_SHIFT 24 /* 16 MiB per resync extent */ 1249 #define BM_EXT_SIZE (1<<BM_EXT_SHIFT) 1250 1251 #if (BM_EXT_SHIFT != 24) || (BM_BLOCK_SHIFT != 12) 1252 #error "HAVE YOU FIXED drbdmeta AS WELL??" 1253 #endif 1254 1255 /* thus many _storage_ sectors are described by one bit */ 1256 #define BM_SECT_TO_BIT(x) ((x)>>(BM_BLOCK_SHIFT-9)) 1257 #define BM_BIT_TO_SECT(x) ((sector_t)(x)<<(BM_BLOCK_SHIFT-9)) 1258 #define BM_SECT_PER_BIT BM_BIT_TO_SECT(1) 1259 1260 /* bit to represented kilo byte conversion */ 1261 #define Bit2KB(bits) ((bits)<<(BM_BLOCK_SHIFT-10)) 1262 1263 /* in which _bitmap_ extent (resp. sector) the bit for a certain 1264 * _storage_ sector is located in */ 1265 #define BM_SECT_TO_EXT(x) ((x)>>(BM_EXT_SHIFT-9)) 1266 #define BM_BIT_TO_EXT(x) ((x) >> (BM_EXT_SHIFT - BM_BLOCK_SHIFT)) 1267 1268 /* first storage sector a bitmap extent corresponds to */ 1269 #define BM_EXT_TO_SECT(x) ((sector_t)(x) << (BM_EXT_SHIFT-9)) 1270 /* how much _storage_ sectors we have per bitmap extent */ 1271 #define BM_SECT_PER_EXT BM_EXT_TO_SECT(1) 1272 /* how many bits are covered by one bitmap extent (resync extent) */ 1273 #define BM_BITS_PER_EXT (1UL << (BM_EXT_SHIFT - BM_BLOCK_SHIFT)) 1274 1275 #define BM_BLOCKS_PER_BM_EXT_MASK (BM_BITS_PER_EXT - 1) 1276 1277 1278 /* in one sector of the bitmap, we have this many activity_log extents. */ 1279 #define AL_EXT_PER_BM_SECT (1 << (BM_EXT_SHIFT - AL_EXTENT_SHIFT)) 1280 1281 /* the extent in "PER_EXTENT" below is an activity log extent 1282 * we need that many (long words/bytes) to store the bitmap 1283 * of one AL_EXTENT_SIZE chunk of storage. 1284 * we can store the bitmap for that many AL_EXTENTS within 1285 * one sector of the _on_disk_ bitmap: 1286 * bit 0 bit 37 bit 38 bit (512*8)-1 1287 * ...|........|........|.. // ..|........| 1288 * sect. 0 `296 `304 ^(512*8*8)-1 1289 * 1290 #define BM_WORDS_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / BITS_PER_LONG ) 1291 #define BM_BYTES_PER_EXT ( (AL_EXT_SIZE/BM_BLOCK_SIZE) / 8 ) // 128 1292 #define BM_EXT_PER_SECT ( 512 / BM_BYTES_PER_EXTENT ) // 4 1293 */ 1294 1295 #define DRBD_MAX_SECTORS_32 (0xffffffffLU) 1296 /* we have a certain meta data variant that has a fixed on-disk size of 128 1297 * MiB, of which 4k are our "superblock", and 32k are the fixed size activity 1298 * log, leaving this many sectors for the bitmap. 1299 */ 1300 1301 #define DRBD_MAX_SECTORS_FIXED_BM \ 1302 ((MD_128MB_SECT - MD_32kB_SECT - MD_4kB_SECT) * (1LL<<(BM_EXT_SHIFT-9))) 1303 #if !defined(CONFIG_LBDAF) && BITS_PER_LONG == 32 1304 #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_32 1305 #define DRBD_MAX_SECTORS_FLEX DRBD_MAX_SECTORS_32 1306 #else 1307 #define DRBD_MAX_SECTORS DRBD_MAX_SECTORS_FIXED_BM 1308 /* 16 TB in units of sectors */ 1309 #if BITS_PER_LONG == 32 1310 /* adjust by one page worth of bitmap, 1311 * so we won't wrap around in drbd_bm_find_next_bit. 1312 * you should use 64bit OS for that much storage, anyways. */ 1313 #define DRBD_MAX_SECTORS_FLEX BM_BIT_TO_SECT(0xffff7fff) 1314 #else 1315 /* we allow up to 1 PiB now on 64bit architecture with "flexible" meta data */ 1316 #define DRBD_MAX_SECTORS_FLEX (1UL << 51) 1317 /* corresponds to (1UL << 38) bits right now. */ 1318 #endif 1319 #endif 1320 1321 /* BIO_MAX_SIZE is 256 * PAGE_CACHE_SIZE, 1322 * so for typical PAGE_CACHE_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_SIZE 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 one activity log extent worth of data 1337 * to be discarded in one go. We may need to rework drbd_al_begin_io() 1338 * to allow for even larger discard ranges */ 1339 #define DRBD_MAX_DISCARD_SIZE AL_EXTENT_SIZE 1340 #define DRBD_MAX_DISCARD_SECTORS (DRBD_MAX_DISCARD_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 int drbd_bm_write_hinted(struct drbd_device *device) __must_hold(local); 1364 extern int drbd_bm_write_lazy(struct drbd_device *device, unsigned upper_idx) __must_hold(local); 1365 extern int drbd_bm_write_all(struct drbd_device *device) __must_hold(local); 1366 extern int drbd_bm_write_copy_pages(struct drbd_device *device) __must_hold(local); 1367 extern size_t drbd_bm_words(struct drbd_device *device); 1368 extern unsigned long drbd_bm_bits(struct drbd_device *device); 1369 extern sector_t drbd_bm_capacity(struct drbd_device *device); 1370 1371 #define DRBD_END_OF_BITMAP (~(unsigned long)0) 1372 extern unsigned long drbd_bm_find_next(struct drbd_device *device, unsigned long bm_fo); 1373 /* bm_find_next variants for use while you hold drbd_bm_lock() */ 1374 extern unsigned long _drbd_bm_find_next(struct drbd_device *device, unsigned long bm_fo); 1375 extern unsigned long _drbd_bm_find_next_zero(struct drbd_device *device, unsigned long bm_fo); 1376 extern unsigned long _drbd_bm_total_weight(struct drbd_device *device); 1377 extern unsigned long drbd_bm_total_weight(struct drbd_device *device); 1378 /* for receive_bitmap */ 1379 extern void drbd_bm_merge_lel(struct drbd_device *device, size_t offset, 1380 size_t number, unsigned long *buffer); 1381 /* for _drbd_send_bitmap */ 1382 extern void drbd_bm_get_lel(struct drbd_device *device, size_t offset, 1383 size_t number, unsigned long *buffer); 1384 1385 extern void drbd_bm_lock(struct drbd_device *device, char *why, enum bm_flag flags); 1386 extern void drbd_bm_unlock(struct drbd_device *device); 1387 /* drbd_main.c */ 1388 1389 extern struct kmem_cache *drbd_request_cache; 1390 extern struct kmem_cache *drbd_ee_cache; /* peer requests */ 1391 extern struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */ 1392 extern struct kmem_cache *drbd_al_ext_cache; /* activity log extents */ 1393 extern mempool_t *drbd_request_mempool; 1394 extern mempool_t *drbd_ee_mempool; 1395 1396 /* drbd's page pool, used to buffer data received from the peer, 1397 * or data requested by the peer. 1398 * 1399 * This does not have an emergency reserve. 1400 * 1401 * When allocating from this pool, it first takes pages from the pool. 1402 * Only if the pool is depleted will try to allocate from the system. 1403 * 1404 * The assumption is that pages taken from this pool will be processed, 1405 * and given back, "quickly", and then can be recycled, so we can avoid 1406 * frequent calls to alloc_page(), and still will be able to make progress even 1407 * under memory pressure. 1408 */ 1409 extern struct page *drbd_pp_pool; 1410 extern spinlock_t drbd_pp_lock; 1411 extern int drbd_pp_vacant; 1412 extern wait_queue_head_t drbd_pp_wait; 1413 1414 /* We also need a standard (emergency-reserve backed) page pool 1415 * for meta data IO (activity log, bitmap). 1416 * We can keep it global, as long as it is used as "N pages at a time". 1417 * 128 should be plenty, currently we probably can get away with as few as 1. 1418 */ 1419 #define DRBD_MIN_POOL_PAGES 128 1420 extern mempool_t *drbd_md_io_page_pool; 1421 1422 /* We also need to make sure we get a bio 1423 * when we need it for housekeeping purposes */ 1424 extern struct bio_set *drbd_md_io_bio_set; 1425 /* to allocate from that set */ 1426 extern struct bio *bio_alloc_drbd(gfp_t gfp_mask); 1427 1428 extern rwlock_t global_state_lock; 1429 1430 extern int conn_lowest_minor(struct drbd_connection *connection); 1431 extern enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor); 1432 extern void drbd_destroy_device(struct kref *kref); 1433 extern void drbd_delete_device(struct drbd_device *device); 1434 1435 extern struct drbd_resource *drbd_create_resource(const char *name); 1436 extern void drbd_free_resource(struct drbd_resource *resource); 1437 1438 extern int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts); 1439 extern struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts); 1440 extern void drbd_destroy_connection(struct kref *kref); 1441 extern struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len, 1442 void *peer_addr, int peer_addr_len); 1443 extern struct drbd_resource *drbd_find_resource(const char *name); 1444 extern void drbd_destroy_resource(struct kref *kref); 1445 extern void conn_free_crypto(struct drbd_connection *connection); 1446 1447 extern int proc_details; 1448 1449 /* drbd_req */ 1450 extern void do_submit(struct work_struct *ws); 1451 extern void __drbd_make_request(struct drbd_device *, struct bio *, unsigned long); 1452 extern void drbd_make_request(struct request_queue *q, struct bio *bio); 1453 extern int drbd_read_remote(struct drbd_device *device, struct drbd_request *req); 1454 extern int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct bio_vec *bvec); 1455 extern int is_valid_ar_handle(struct drbd_request *, sector_t); 1456 1457 1458 /* drbd_nl.c */ 1459 extern int drbd_msg_put_info(struct sk_buff *skb, const char *info); 1460 extern void drbd_suspend_io(struct drbd_device *device); 1461 extern void drbd_resume_io(struct drbd_device *device); 1462 extern char *ppsize(char *buf, unsigned long long size); 1463 extern sector_t drbd_new_dev_size(struct drbd_device *, struct drbd_backing_dev *, sector_t, int); 1464 enum determine_dev_size { 1465 DS_ERROR_SHRINK = -3, 1466 DS_ERROR_SPACE_MD = -2, 1467 DS_ERROR = -1, 1468 DS_UNCHANGED = 0, 1469 DS_SHRUNK = 1, 1470 DS_GREW = 2, 1471 DS_GREW_FROM_ZERO = 3, 1472 }; 1473 extern enum determine_dev_size 1474 drbd_determine_dev_size(struct drbd_device *, enum dds_flags, struct resize_parms *) __must_hold(local); 1475 extern void resync_after_online_grow(struct drbd_device *); 1476 extern void drbd_reconsider_max_bio_size(struct drbd_device *device, struct drbd_backing_dev *bdev); 1477 extern enum drbd_state_rv drbd_set_role(struct drbd_device *device, 1478 enum drbd_role new_role, 1479 int force); 1480 extern bool conn_try_outdate_peer(struct drbd_connection *connection); 1481 extern void conn_try_outdate_peer_async(struct drbd_connection *connection); 1482 extern int drbd_khelper(struct drbd_device *device, char *cmd); 1483 1484 /* drbd_worker.c */ 1485 /* bi_end_io handlers */ 1486 extern void drbd_md_io_complete(struct bio *bio, int error); 1487 extern void drbd_peer_request_endio(struct bio *bio, int error); 1488 extern void drbd_request_endio(struct bio *bio, int error); 1489 extern int drbd_worker(struct drbd_thread *thi); 1490 enum drbd_ret_code drbd_resync_after_valid(struct drbd_device *device, int o_minor); 1491 void drbd_resync_after_changed(struct drbd_device *device); 1492 extern void drbd_start_resync(struct drbd_device *device, enum drbd_conns side); 1493 extern void resume_next_sg(struct drbd_device *device); 1494 extern void suspend_other_sg(struct drbd_device *device); 1495 extern int drbd_resync_finished(struct drbd_device *device); 1496 /* maybe rather drbd_main.c ? */ 1497 extern void *drbd_md_get_buffer(struct drbd_device *device, const char *intent); 1498 extern void drbd_md_put_buffer(struct drbd_device *device); 1499 extern int drbd_md_sync_page_io(struct drbd_device *device, 1500 struct drbd_backing_dev *bdev, sector_t sector, int rw); 1501 extern void drbd_ov_out_of_sync_found(struct drbd_device *, sector_t, int); 1502 extern void wait_until_done_or_force_detached(struct drbd_device *device, 1503 struct drbd_backing_dev *bdev, unsigned int *done); 1504 extern void drbd_rs_controller_reset(struct drbd_device *device); 1505 1506 static inline void ov_out_of_sync_print(struct drbd_device *device) 1507 { 1508 if (device->ov_last_oos_size) { 1509 drbd_err(device, "Out of sync: start=%llu, size=%lu (sectors)\n", 1510 (unsigned long long)device->ov_last_oos_start, 1511 (unsigned long)device->ov_last_oos_size); 1512 } 1513 device->ov_last_oos_size = 0; 1514 } 1515 1516 1517 extern void drbd_csum_bio(struct crypto_hash *, struct bio *, void *); 1518 extern void drbd_csum_ee(struct crypto_hash *, struct drbd_peer_request *, void *); 1519 /* worker callbacks */ 1520 extern int w_e_end_data_req(struct drbd_work *, int); 1521 extern int w_e_end_rsdata_req(struct drbd_work *, int); 1522 extern int w_e_end_csum_rs_req(struct drbd_work *, int); 1523 extern int w_e_end_ov_reply(struct drbd_work *, int); 1524 extern int w_e_end_ov_req(struct drbd_work *, int); 1525 extern int w_ov_finished(struct drbd_work *, int); 1526 extern int w_resync_timer(struct drbd_work *, int); 1527 extern int w_send_write_hint(struct drbd_work *, int); 1528 extern int w_send_dblock(struct drbd_work *, int); 1529 extern int w_send_read_req(struct drbd_work *, int); 1530 extern int w_e_reissue(struct drbd_work *, int); 1531 extern int w_restart_disk_io(struct drbd_work *, int); 1532 extern int w_send_out_of_sync(struct drbd_work *, int); 1533 extern int w_start_resync(struct drbd_work *, int); 1534 1535 extern void resync_timer_fn(unsigned long data); 1536 extern void start_resync_timer_fn(unsigned long data); 1537 1538 extern void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req); 1539 1540 /* drbd_receiver.c */ 1541 extern int drbd_receiver(struct drbd_thread *thi); 1542 extern int drbd_asender(struct drbd_thread *thi); 1543 extern bool drbd_rs_c_min_rate_throttle(struct drbd_device *device); 1544 extern bool drbd_rs_should_slow_down(struct drbd_device *device, sector_t sector, 1545 bool throttle_if_app_is_waiting); 1546 extern int drbd_submit_peer_request(struct drbd_device *, 1547 struct drbd_peer_request *, const unsigned, 1548 const int); 1549 extern int drbd_free_peer_reqs(struct drbd_device *, struct list_head *); 1550 extern struct drbd_peer_request *drbd_alloc_peer_req(struct drbd_peer_device *, u64, 1551 sector_t, unsigned int, 1552 bool, 1553 gfp_t) __must_hold(local); 1554 extern void __drbd_free_peer_req(struct drbd_device *, struct drbd_peer_request *, 1555 int); 1556 #define drbd_free_peer_req(m,e) __drbd_free_peer_req(m, e, 0) 1557 #define drbd_free_net_peer_req(m,e) __drbd_free_peer_req(m, e, 1) 1558 extern struct page *drbd_alloc_pages(struct drbd_peer_device *, unsigned int, bool); 1559 extern void drbd_set_recv_tcq(struct drbd_device *device, int tcq_enabled); 1560 extern void _drbd_clear_done_ee(struct drbd_device *device, struct list_head *to_be_freed); 1561 extern int drbd_connected(struct drbd_peer_device *); 1562 1563 /* Yes, there is kernel_setsockopt, but only since 2.6.18. 1564 * So we have our own copy of it here. */ 1565 static inline int drbd_setsockopt(struct socket *sock, int level, int optname, 1566 char *optval, int optlen) 1567 { 1568 mm_segment_t oldfs = get_fs(); 1569 char __user *uoptval; 1570 int err; 1571 1572 uoptval = (char __user __force *)optval; 1573 1574 set_fs(KERNEL_DS); 1575 if (level == SOL_SOCKET) 1576 err = sock_setsockopt(sock, level, optname, uoptval, optlen); 1577 else 1578 err = sock->ops->setsockopt(sock, level, optname, uoptval, 1579 optlen); 1580 set_fs(oldfs); 1581 return err; 1582 } 1583 1584 static inline void drbd_tcp_cork(struct socket *sock) 1585 { 1586 int val = 1; 1587 (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK, 1588 (char*)&val, sizeof(val)); 1589 } 1590 1591 static inline void drbd_tcp_uncork(struct socket *sock) 1592 { 1593 int val = 0; 1594 (void) drbd_setsockopt(sock, SOL_TCP, TCP_CORK, 1595 (char*)&val, sizeof(val)); 1596 } 1597 1598 static inline void drbd_tcp_nodelay(struct socket *sock) 1599 { 1600 int val = 1; 1601 (void) drbd_setsockopt(sock, SOL_TCP, TCP_NODELAY, 1602 (char*)&val, sizeof(val)); 1603 } 1604 1605 static inline void drbd_tcp_quickack(struct socket *sock) 1606 { 1607 int val = 2; 1608 (void) drbd_setsockopt(sock, SOL_TCP, TCP_QUICKACK, 1609 (char*)&val, sizeof(val)); 1610 } 1611 1612 /* sets the number of 512 byte sectors of our virtual device */ 1613 static inline void drbd_set_my_capacity(struct drbd_device *device, 1614 sector_t size) 1615 { 1616 /* set_capacity(device->this_bdev->bd_disk, size); */ 1617 set_capacity(device->vdisk, size); 1618 device->this_bdev->bd_inode->i_size = (loff_t)size << 9; 1619 } 1620 1621 /* 1622 * used to submit our private bio 1623 */ 1624 static inline void drbd_generic_make_request(struct drbd_device *device, 1625 int fault_type, struct bio *bio) 1626 { 1627 __release(local); 1628 if (!bio->bi_bdev) { 1629 drbd_err(device, "drbd_generic_make_request: bio->bi_bdev == NULL\n"); 1630 bio_endio(bio, -ENODEV); 1631 return; 1632 } 1633 1634 if (drbd_insert_fault(device, fault_type)) 1635 bio_endio(bio, -EIO); 1636 else 1637 generic_make_request(bio); 1638 } 1639 1640 void drbd_bump_write_ordering(struct drbd_resource *resource, struct drbd_backing_dev *bdev, 1641 enum write_ordering_e wo); 1642 1643 /* drbd_proc.c */ 1644 extern struct proc_dir_entry *drbd_proc; 1645 extern const struct file_operations drbd_proc_fops; 1646 extern const char *drbd_conn_str(enum drbd_conns s); 1647 extern const char *drbd_role_str(enum drbd_role s); 1648 1649 /* drbd_actlog.c */ 1650 extern bool drbd_al_begin_io_prepare(struct drbd_device *device, struct drbd_interval *i); 1651 extern int drbd_al_begin_io_nonblock(struct drbd_device *device, struct drbd_interval *i); 1652 extern void drbd_al_begin_io_commit(struct drbd_device *device); 1653 extern bool drbd_al_begin_io_fastpath(struct drbd_device *device, struct drbd_interval *i); 1654 extern void drbd_al_begin_io(struct drbd_device *device, struct drbd_interval *i); 1655 extern void drbd_al_complete_io(struct drbd_device *device, struct drbd_interval *i); 1656 extern void drbd_rs_complete_io(struct drbd_device *device, sector_t sector); 1657 extern int drbd_rs_begin_io(struct drbd_device *device, sector_t sector); 1658 extern int drbd_try_rs_begin_io(struct drbd_device *device, sector_t sector); 1659 extern void drbd_rs_cancel_all(struct drbd_device *device); 1660 extern int drbd_rs_del_all(struct drbd_device *device); 1661 extern void drbd_rs_failed_io(struct drbd_device *device, 1662 sector_t sector, int size); 1663 extern void drbd_advance_rs_marks(struct drbd_device *device, unsigned long still_to_go); 1664 1665 enum update_sync_bits_mode { RECORD_RS_FAILED, SET_OUT_OF_SYNC, SET_IN_SYNC }; 1666 extern int __drbd_change_sync(struct drbd_device *device, sector_t sector, int size, 1667 enum update_sync_bits_mode mode, 1668 const char *file, const unsigned int line); 1669 #define drbd_set_in_sync(device, sector, size) \ 1670 __drbd_change_sync(device, sector, size, SET_IN_SYNC, __FILE__, __LINE__) 1671 #define drbd_set_out_of_sync(device, sector, size) \ 1672 __drbd_change_sync(device, sector, size, SET_OUT_OF_SYNC, __FILE__, __LINE__) 1673 #define drbd_rs_failed_io(device, sector, size) \ 1674 __drbd_change_sync(device, sector, size, RECORD_RS_FAILED, __FILE__, __LINE__) 1675 extern void drbd_al_shrink(struct drbd_device *device); 1676 extern int drbd_initialize_al(struct drbd_device *, void *); 1677 1678 /* drbd_nl.c */ 1679 /* state info broadcast */ 1680 struct sib_info { 1681 enum drbd_state_info_bcast_reason sib_reason; 1682 union { 1683 struct { 1684 char *helper_name; 1685 unsigned helper_exit_code; 1686 }; 1687 struct { 1688 union drbd_state os; 1689 union drbd_state ns; 1690 }; 1691 }; 1692 }; 1693 void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib); 1694 1695 /* 1696 * inline helper functions 1697 *************************/ 1698 1699 /* see also page_chain_add and friends in drbd_receiver.c */ 1700 static inline struct page *page_chain_next(struct page *page) 1701 { 1702 return (struct page *)page_private(page); 1703 } 1704 #define page_chain_for_each(page) \ 1705 for (; page && ({ prefetch(page_chain_next(page)); 1; }); \ 1706 page = page_chain_next(page)) 1707 #define page_chain_for_each_safe(page, n) \ 1708 for (; page && ({ n = page_chain_next(page); 1; }); page = n) 1709 1710 1711 static inline int drbd_peer_req_has_active_page(struct drbd_peer_request *peer_req) 1712 { 1713 struct page *page = peer_req->pages; 1714 page_chain_for_each(page) { 1715 if (page_count(page) > 1) 1716 return 1; 1717 } 1718 return 0; 1719 } 1720 1721 static inline enum drbd_state_rv 1722 _drbd_set_state(struct drbd_device *device, union drbd_state ns, 1723 enum chg_state_flags flags, struct completion *done) 1724 { 1725 enum drbd_state_rv rv; 1726 1727 read_lock(&global_state_lock); 1728 rv = __drbd_set_state(device, ns, flags, done); 1729 read_unlock(&global_state_lock); 1730 1731 return rv; 1732 } 1733 1734 static inline union drbd_state drbd_read_state(struct drbd_device *device) 1735 { 1736 struct drbd_resource *resource = device->resource; 1737 union drbd_state rv; 1738 1739 rv.i = device->state.i; 1740 rv.susp = resource->susp; 1741 rv.susp_nod = resource->susp_nod; 1742 rv.susp_fen = resource->susp_fen; 1743 1744 return rv; 1745 } 1746 1747 enum drbd_force_detach_flags { 1748 DRBD_READ_ERROR, 1749 DRBD_WRITE_ERROR, 1750 DRBD_META_IO_ERROR, 1751 DRBD_FORCE_DETACH, 1752 }; 1753 1754 #define __drbd_chk_io_error(m,f) __drbd_chk_io_error_(m,f, __func__) 1755 static inline void __drbd_chk_io_error_(struct drbd_device *device, 1756 enum drbd_force_detach_flags df, 1757 const char *where) 1758 { 1759 enum drbd_io_error_p ep; 1760 1761 rcu_read_lock(); 1762 ep = rcu_dereference(device->ldev->disk_conf)->on_io_error; 1763 rcu_read_unlock(); 1764 switch (ep) { 1765 case EP_PASS_ON: /* FIXME would this be better named "Ignore"? */ 1766 if (df == DRBD_READ_ERROR || df == DRBD_WRITE_ERROR) { 1767 if (__ratelimit(&drbd_ratelimit_state)) 1768 drbd_err(device, "Local IO failed in %s.\n", where); 1769 if (device->state.disk > D_INCONSISTENT) 1770 _drbd_set_state(_NS(device, disk, D_INCONSISTENT), CS_HARD, NULL); 1771 break; 1772 } 1773 /* NOTE fall through for DRBD_META_IO_ERROR or DRBD_FORCE_DETACH */ 1774 case EP_DETACH: 1775 case EP_CALL_HELPER: 1776 /* Remember whether we saw a READ or WRITE error. 1777 * 1778 * Recovery of the affected area for WRITE failure is covered 1779 * by the activity log. 1780 * READ errors may fall outside that area though. Certain READ 1781 * errors can be "healed" by writing good data to the affected 1782 * blocks, which triggers block re-allocation in lower layers. 1783 * 1784 * If we can not write the bitmap after a READ error, 1785 * we may need to trigger a full sync (see w_go_diskless()). 1786 * 1787 * Force-detach is not really an IO error, but rather a 1788 * desperate measure to try to deal with a completely 1789 * unresponsive lower level IO stack. 1790 * Still it should be treated as a WRITE error. 1791 * 1792 * Meta IO error is always WRITE error: 1793 * we read meta data only once during attach, 1794 * which will fail in case of errors. 1795 */ 1796 set_bit(WAS_IO_ERROR, &device->flags); 1797 if (df == DRBD_READ_ERROR) 1798 set_bit(WAS_READ_ERROR, &device->flags); 1799 if (df == DRBD_FORCE_DETACH) 1800 set_bit(FORCE_DETACH, &device->flags); 1801 if (device->state.disk > D_FAILED) { 1802 _drbd_set_state(_NS(device, disk, D_FAILED), CS_HARD, NULL); 1803 drbd_err(device, 1804 "Local IO failed in %s. Detaching...\n", where); 1805 } 1806 break; 1807 } 1808 } 1809 1810 /** 1811 * drbd_chk_io_error: Handle the on_io_error setting, should be called from all io completion handlers 1812 * @device: DRBD device. 1813 * @error: Error code passed to the IO completion callback 1814 * @forcedetach: Force detach. I.e. the error happened while accessing the meta data 1815 * 1816 * See also drbd_main.c:after_state_ch() if (os.disk > D_FAILED && ns.disk == D_FAILED) 1817 */ 1818 #define drbd_chk_io_error(m,e,f) drbd_chk_io_error_(m,e,f, __func__) 1819 static inline void drbd_chk_io_error_(struct drbd_device *device, 1820 int error, enum drbd_force_detach_flags forcedetach, const char *where) 1821 { 1822 if (error) { 1823 unsigned long flags; 1824 spin_lock_irqsave(&device->resource->req_lock, flags); 1825 __drbd_chk_io_error_(device, forcedetach, where); 1826 spin_unlock_irqrestore(&device->resource->req_lock, flags); 1827 } 1828 } 1829 1830 1831 /** 1832 * drbd_md_first_sector() - Returns the first sector number of the meta data area 1833 * @bdev: Meta data block device. 1834 * 1835 * BTW, for internal meta data, this happens to be the maximum capacity 1836 * we could agree upon with our peer node. 1837 */ 1838 static inline sector_t drbd_md_first_sector(struct drbd_backing_dev *bdev) 1839 { 1840 switch (bdev->md.meta_dev_idx) { 1841 case DRBD_MD_INDEX_INTERNAL: 1842 case DRBD_MD_INDEX_FLEX_INT: 1843 return bdev->md.md_offset + bdev->md.bm_offset; 1844 case DRBD_MD_INDEX_FLEX_EXT: 1845 default: 1846 return bdev->md.md_offset; 1847 } 1848 } 1849 1850 /** 1851 * drbd_md_last_sector() - Return the last sector number of the meta data area 1852 * @bdev: Meta data block device. 1853 */ 1854 static inline sector_t drbd_md_last_sector(struct drbd_backing_dev *bdev) 1855 { 1856 switch (bdev->md.meta_dev_idx) { 1857 case DRBD_MD_INDEX_INTERNAL: 1858 case DRBD_MD_INDEX_FLEX_INT: 1859 return bdev->md.md_offset + MD_4kB_SECT -1; 1860 case DRBD_MD_INDEX_FLEX_EXT: 1861 default: 1862 return bdev->md.md_offset + bdev->md.md_size_sect -1; 1863 } 1864 } 1865 1866 /* Returns the number of 512 byte sectors of the device */ 1867 static inline sector_t drbd_get_capacity(struct block_device *bdev) 1868 { 1869 /* return bdev ? get_capacity(bdev->bd_disk) : 0; */ 1870 return bdev ? i_size_read(bdev->bd_inode) >> 9 : 0; 1871 } 1872 1873 /** 1874 * drbd_get_max_capacity() - Returns the capacity we announce to out peer 1875 * @bdev: Meta data block device. 1876 * 1877 * returns the capacity we announce to out peer. we clip ourselves at the 1878 * various MAX_SECTORS, because if we don't, current implementation will 1879 * oops sooner or later 1880 */ 1881 static inline sector_t drbd_get_max_capacity(struct drbd_backing_dev *bdev) 1882 { 1883 sector_t s; 1884 1885 switch (bdev->md.meta_dev_idx) { 1886 case DRBD_MD_INDEX_INTERNAL: 1887 case DRBD_MD_INDEX_FLEX_INT: 1888 s = drbd_get_capacity(bdev->backing_bdev) 1889 ? min_t(sector_t, DRBD_MAX_SECTORS_FLEX, 1890 drbd_md_first_sector(bdev)) 1891 : 0; 1892 break; 1893 case DRBD_MD_INDEX_FLEX_EXT: 1894 s = min_t(sector_t, DRBD_MAX_SECTORS_FLEX, 1895 drbd_get_capacity(bdev->backing_bdev)); 1896 /* clip at maximum size the meta device can support */ 1897 s = min_t(sector_t, s, 1898 BM_EXT_TO_SECT(bdev->md.md_size_sect 1899 - bdev->md.bm_offset)); 1900 break; 1901 default: 1902 s = min_t(sector_t, DRBD_MAX_SECTORS, 1903 drbd_get_capacity(bdev->backing_bdev)); 1904 } 1905 return s; 1906 } 1907 1908 /** 1909 * drbd_md_ss() - Return the sector number of our meta data super block 1910 * @bdev: Meta data block device. 1911 */ 1912 static inline sector_t drbd_md_ss(struct drbd_backing_dev *bdev) 1913 { 1914 const int meta_dev_idx = bdev->md.meta_dev_idx; 1915 1916 if (meta_dev_idx == DRBD_MD_INDEX_FLEX_EXT) 1917 return 0; 1918 1919 /* Since drbd08, internal meta data is always "flexible". 1920 * position: last 4k aligned block of 4k size */ 1921 if (meta_dev_idx == DRBD_MD_INDEX_INTERNAL || 1922 meta_dev_idx == DRBD_MD_INDEX_FLEX_INT) 1923 return (drbd_get_capacity(bdev->backing_bdev) & ~7ULL) - 8; 1924 1925 /* external, some index; this is the old fixed size layout */ 1926 return MD_128MB_SECT * bdev->md.meta_dev_idx; 1927 } 1928 1929 static inline void 1930 drbd_queue_work(struct drbd_work_queue *q, struct drbd_work *w) 1931 { 1932 unsigned long flags; 1933 spin_lock_irqsave(&q->q_lock, flags); 1934 list_add_tail(&w->list, &q->q); 1935 spin_unlock_irqrestore(&q->q_lock, flags); 1936 wake_up(&q->q_wait); 1937 } 1938 1939 static inline void 1940 drbd_queue_work_if_unqueued(struct drbd_work_queue *q, struct drbd_work *w) 1941 { 1942 unsigned long flags; 1943 spin_lock_irqsave(&q->q_lock, flags); 1944 if (list_empty_careful(&w->list)) 1945 list_add_tail(&w->list, &q->q); 1946 spin_unlock_irqrestore(&q->q_lock, flags); 1947 wake_up(&q->q_wait); 1948 } 1949 1950 static inline void 1951 drbd_device_post_work(struct drbd_device *device, int work_bit) 1952 { 1953 if (!test_and_set_bit(work_bit, &device->flags)) { 1954 struct drbd_connection *connection = 1955 first_peer_device(device)->connection; 1956 struct drbd_work_queue *q = &connection->sender_work; 1957 if (!test_and_set_bit(DEVICE_WORK_PENDING, &connection->flags)) 1958 wake_up(&q->q_wait); 1959 } 1960 } 1961 1962 extern void drbd_flush_workqueue(struct drbd_work_queue *work_queue); 1963 1964 static inline void wake_asender(struct drbd_connection *connection) 1965 { 1966 if (test_bit(SIGNAL_ASENDER, &connection->flags)) 1967 force_sig(DRBD_SIG, connection->asender.task); 1968 } 1969 1970 static inline void request_ping(struct drbd_connection *connection) 1971 { 1972 set_bit(SEND_PING, &connection->flags); 1973 wake_asender(connection); 1974 } 1975 1976 extern void *conn_prepare_command(struct drbd_connection *, struct drbd_socket *); 1977 extern void *drbd_prepare_command(struct drbd_peer_device *, struct drbd_socket *); 1978 extern int conn_send_command(struct drbd_connection *, struct drbd_socket *, 1979 enum drbd_packet, unsigned int, void *, 1980 unsigned int); 1981 extern int drbd_send_command(struct drbd_peer_device *, struct drbd_socket *, 1982 enum drbd_packet, unsigned int, void *, 1983 unsigned int); 1984 1985 extern int drbd_send_ping(struct drbd_connection *connection); 1986 extern int drbd_send_ping_ack(struct drbd_connection *connection); 1987 extern int drbd_send_state_req(struct drbd_peer_device *, union drbd_state, union drbd_state); 1988 extern int conn_send_state_req(struct drbd_connection *, union drbd_state, union drbd_state); 1989 1990 static inline void drbd_thread_stop(struct drbd_thread *thi) 1991 { 1992 _drbd_thread_stop(thi, false, true); 1993 } 1994 1995 static inline void drbd_thread_stop_nowait(struct drbd_thread *thi) 1996 { 1997 _drbd_thread_stop(thi, false, false); 1998 } 1999 2000 static inline void drbd_thread_restart_nowait(struct drbd_thread *thi) 2001 { 2002 _drbd_thread_stop(thi, true, false); 2003 } 2004 2005 /* counts how many answer packets packets we expect from our peer, 2006 * for either explicit application requests, 2007 * or implicit barrier packets as necessary. 2008 * increased: 2009 * w_send_barrier 2010 * _req_mod(req, QUEUE_FOR_NET_WRITE or QUEUE_FOR_NET_READ); 2011 * it is much easier and equally valid to count what we queue for the 2012 * worker, even before it actually was queued or send. 2013 * (drbd_make_request_common; recovery path on read io-error) 2014 * decreased: 2015 * got_BarrierAck (respective tl_clear, tl_clear_barrier) 2016 * _req_mod(req, DATA_RECEIVED) 2017 * [from receive_DataReply] 2018 * _req_mod(req, WRITE_ACKED_BY_PEER or RECV_ACKED_BY_PEER or NEG_ACKED) 2019 * [from got_BlockAck (P_WRITE_ACK, P_RECV_ACK)] 2020 * for some reason it is NOT decreased in got_NegAck, 2021 * but in the resulting cleanup code from report_params. 2022 * we should try to remember the reason for that... 2023 * _req_mod(req, SEND_FAILED or SEND_CANCELED) 2024 * _req_mod(req, CONNECTION_LOST_WHILE_PENDING) 2025 * [from tl_clear_barrier] 2026 */ 2027 static inline void inc_ap_pending(struct drbd_device *device) 2028 { 2029 atomic_inc(&device->ap_pending_cnt); 2030 } 2031 2032 #define ERR_IF_CNT_IS_NEGATIVE(which, func, line) \ 2033 if (atomic_read(&device->which) < 0) \ 2034 drbd_err(device, "in %s:%d: " #which " = %d < 0 !\n", \ 2035 func, line, \ 2036 atomic_read(&device->which)) 2037 2038 #define dec_ap_pending(device) _dec_ap_pending(device, __func__, __LINE__) 2039 static inline void _dec_ap_pending(struct drbd_device *device, const char *func, int line) 2040 { 2041 if (atomic_dec_and_test(&device->ap_pending_cnt)) 2042 wake_up(&device->misc_wait); 2043 ERR_IF_CNT_IS_NEGATIVE(ap_pending_cnt, func, line); 2044 } 2045 2046 /* counts how many resync-related answers we still expect from the peer 2047 * increase decrease 2048 * C_SYNC_TARGET sends P_RS_DATA_REQUEST (and expects P_RS_DATA_REPLY) 2049 * C_SYNC_SOURCE sends P_RS_DATA_REPLY (and expects P_WRITE_ACK with ID_SYNCER) 2050 * (or P_NEG_ACK with ID_SYNCER) 2051 */ 2052 static inline void inc_rs_pending(struct drbd_device *device) 2053 { 2054 atomic_inc(&device->rs_pending_cnt); 2055 } 2056 2057 #define dec_rs_pending(device) _dec_rs_pending(device, __func__, __LINE__) 2058 static inline void _dec_rs_pending(struct drbd_device *device, const char *func, int line) 2059 { 2060 atomic_dec(&device->rs_pending_cnt); 2061 ERR_IF_CNT_IS_NEGATIVE(rs_pending_cnt, func, line); 2062 } 2063 2064 /* counts how many answers we still need to send to the peer. 2065 * increased on 2066 * receive_Data unless protocol A; 2067 * we need to send a P_RECV_ACK (proto B) 2068 * or P_WRITE_ACK (proto C) 2069 * receive_RSDataReply (recv_resync_read) we need to send a P_WRITE_ACK 2070 * receive_DataRequest (receive_RSDataRequest) we need to send back P_DATA 2071 * receive_Barrier_* we need to send a P_BARRIER_ACK 2072 */ 2073 static inline void inc_unacked(struct drbd_device *device) 2074 { 2075 atomic_inc(&device->unacked_cnt); 2076 } 2077 2078 #define dec_unacked(device) _dec_unacked(device, __func__, __LINE__) 2079 static inline void _dec_unacked(struct drbd_device *device, const char *func, int line) 2080 { 2081 atomic_dec(&device->unacked_cnt); 2082 ERR_IF_CNT_IS_NEGATIVE(unacked_cnt, func, line); 2083 } 2084 2085 #define sub_unacked(device, n) _sub_unacked(device, n, __func__, __LINE__) 2086 static inline void _sub_unacked(struct drbd_device *device, int n, const char *func, int line) 2087 { 2088 atomic_sub(n, &device->unacked_cnt); 2089 ERR_IF_CNT_IS_NEGATIVE(unacked_cnt, func, line); 2090 } 2091 2092 static inline bool is_sync_state(enum drbd_conns connection_state) 2093 { 2094 return 2095 (connection_state == C_SYNC_SOURCE 2096 || connection_state == C_SYNC_TARGET 2097 || connection_state == C_PAUSED_SYNC_S 2098 || connection_state == C_PAUSED_SYNC_T); 2099 } 2100 2101 /** 2102 * get_ldev() - Increase the ref count on device->ldev. Returns 0 if there is no ldev 2103 * @M: DRBD device. 2104 * 2105 * You have to call put_ldev() when finished working with device->ldev. 2106 */ 2107 #define get_ldev(M) __cond_lock(local, _get_ldev_if_state(M,D_INCONSISTENT)) 2108 #define get_ldev_if_state(M,MINS) __cond_lock(local, _get_ldev_if_state(M,MINS)) 2109 2110 static inline void put_ldev(struct drbd_device *device) 2111 { 2112 enum drbd_disk_state ds = device->state.disk; 2113 /* We must check the state *before* the atomic_dec becomes visible, 2114 * or we have a theoretical race where someone hitting zero, 2115 * while state still D_FAILED, will then see D_DISKLESS in the 2116 * condition below and calling into destroy, where he must not, yet. */ 2117 int i = atomic_dec_return(&device->local_cnt); 2118 2119 /* This may be called from some endio handler, 2120 * so we must not sleep here. */ 2121 2122 __release(local); 2123 D_ASSERT(device, i >= 0); 2124 if (i == 0) { 2125 if (ds == D_DISKLESS) 2126 /* even internal references gone, safe to destroy */ 2127 drbd_device_post_work(device, DESTROY_DISK); 2128 if (ds == D_FAILED) 2129 /* all application IO references gone. */ 2130 if (!test_and_set_bit(GOING_DISKLESS, &device->flags)) 2131 drbd_device_post_work(device, GO_DISKLESS); 2132 wake_up(&device->misc_wait); 2133 } 2134 } 2135 2136 #ifndef __CHECKER__ 2137 static inline int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins) 2138 { 2139 int io_allowed; 2140 2141 /* never get a reference while D_DISKLESS */ 2142 if (device->state.disk == D_DISKLESS) 2143 return 0; 2144 2145 atomic_inc(&device->local_cnt); 2146 io_allowed = (device->state.disk >= mins); 2147 if (!io_allowed) 2148 put_ldev(device); 2149 return io_allowed; 2150 } 2151 #else 2152 extern int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins); 2153 #endif 2154 2155 /* this throttles on-the-fly application requests 2156 * according to max_buffers settings; 2157 * maybe re-implement using semaphores? */ 2158 static inline int drbd_get_max_buffers(struct drbd_device *device) 2159 { 2160 struct net_conf *nc; 2161 int mxb; 2162 2163 rcu_read_lock(); 2164 nc = rcu_dereference(first_peer_device(device)->connection->net_conf); 2165 mxb = nc ? nc->max_buffers : 1000000; /* arbitrary limit on open requests */ 2166 rcu_read_unlock(); 2167 2168 return mxb; 2169 } 2170 2171 static inline int drbd_state_is_stable(struct drbd_device *device) 2172 { 2173 union drbd_dev_state s = device->state; 2174 2175 /* DO NOT add a default clause, we want the compiler to warn us 2176 * for any newly introduced state we may have forgotten to add here */ 2177 2178 switch ((enum drbd_conns)s.conn) { 2179 /* new io only accepted when there is no connection, ... */ 2180 case C_STANDALONE: 2181 case C_WF_CONNECTION: 2182 /* ... or there is a well established connection. */ 2183 case C_CONNECTED: 2184 case C_SYNC_SOURCE: 2185 case C_SYNC_TARGET: 2186 case C_VERIFY_S: 2187 case C_VERIFY_T: 2188 case C_PAUSED_SYNC_S: 2189 case C_PAUSED_SYNC_T: 2190 case C_AHEAD: 2191 case C_BEHIND: 2192 /* transitional states, IO allowed */ 2193 case C_DISCONNECTING: 2194 case C_UNCONNECTED: 2195 case C_TIMEOUT: 2196 case C_BROKEN_PIPE: 2197 case C_NETWORK_FAILURE: 2198 case C_PROTOCOL_ERROR: 2199 case C_TEAR_DOWN: 2200 case C_WF_REPORT_PARAMS: 2201 case C_STARTING_SYNC_S: 2202 case C_STARTING_SYNC_T: 2203 break; 2204 2205 /* Allow IO in BM exchange states with new protocols */ 2206 case C_WF_BITMAP_S: 2207 if (first_peer_device(device)->connection->agreed_pro_version < 96) 2208 return 0; 2209 break; 2210 2211 /* no new io accepted in these states */ 2212 case C_WF_BITMAP_T: 2213 case C_WF_SYNC_UUID: 2214 case C_MASK: 2215 /* not "stable" */ 2216 return 0; 2217 } 2218 2219 switch ((enum drbd_disk_state)s.disk) { 2220 case D_DISKLESS: 2221 case D_INCONSISTENT: 2222 case D_OUTDATED: 2223 case D_CONSISTENT: 2224 case D_UP_TO_DATE: 2225 case D_FAILED: 2226 /* disk state is stable as well. */ 2227 break; 2228 2229 /* no new io accepted during transitional states */ 2230 case D_ATTACHING: 2231 case D_NEGOTIATING: 2232 case D_UNKNOWN: 2233 case D_MASK: 2234 /* not "stable" */ 2235 return 0; 2236 } 2237 2238 return 1; 2239 } 2240 2241 static inline int drbd_suspended(struct drbd_device *device) 2242 { 2243 struct drbd_resource *resource = device->resource; 2244 2245 return resource->susp || resource->susp_fen || resource->susp_nod; 2246 } 2247 2248 static inline bool may_inc_ap_bio(struct drbd_device *device) 2249 { 2250 int mxb = drbd_get_max_buffers(device); 2251 2252 if (drbd_suspended(device)) 2253 return false; 2254 if (test_bit(SUSPEND_IO, &device->flags)) 2255 return false; 2256 2257 /* to avoid potential deadlock or bitmap corruption, 2258 * in various places, we only allow new application io 2259 * to start during "stable" states. */ 2260 2261 /* no new io accepted when attaching or detaching the disk */ 2262 if (!drbd_state_is_stable(device)) 2263 return false; 2264 2265 /* since some older kernels don't have atomic_add_unless, 2266 * and we are within the spinlock anyways, we have this workaround. */ 2267 if (atomic_read(&device->ap_bio_cnt) > mxb) 2268 return false; 2269 if (test_bit(BITMAP_IO, &device->flags)) 2270 return false; 2271 return true; 2272 } 2273 2274 static inline bool inc_ap_bio_cond(struct drbd_device *device) 2275 { 2276 bool rv = false; 2277 2278 spin_lock_irq(&device->resource->req_lock); 2279 rv = may_inc_ap_bio(device); 2280 if (rv) 2281 atomic_inc(&device->ap_bio_cnt); 2282 spin_unlock_irq(&device->resource->req_lock); 2283 2284 return rv; 2285 } 2286 2287 static inline void inc_ap_bio(struct drbd_device *device) 2288 { 2289 /* we wait here 2290 * as long as the device is suspended 2291 * until the bitmap is no longer on the fly during connection 2292 * handshake as long as we would exceed the max_buffer limit. 2293 * 2294 * to avoid races with the reconnect code, 2295 * we need to atomic_inc within the spinlock. */ 2296 2297 wait_event(device->misc_wait, inc_ap_bio_cond(device)); 2298 } 2299 2300 static inline void dec_ap_bio(struct drbd_device *device) 2301 { 2302 int mxb = drbd_get_max_buffers(device); 2303 int ap_bio = atomic_dec_return(&device->ap_bio_cnt); 2304 2305 D_ASSERT(device, ap_bio >= 0); 2306 2307 if (ap_bio == 0 && test_bit(BITMAP_IO, &device->flags)) { 2308 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags)) 2309 drbd_queue_work(&first_peer_device(device)-> 2310 connection->sender_work, 2311 &device->bm_io_work.w); 2312 } 2313 2314 /* this currently does wake_up for every dec_ap_bio! 2315 * maybe rather introduce some type of hysteresis? 2316 * e.g. (ap_bio == mxb/2 || ap_bio == 0) ? */ 2317 if (ap_bio < mxb) 2318 wake_up(&device->misc_wait); 2319 } 2320 2321 static inline bool verify_can_do_stop_sector(struct drbd_device *device) 2322 { 2323 return first_peer_device(device)->connection->agreed_pro_version >= 97 && 2324 first_peer_device(device)->connection->agreed_pro_version != 100; 2325 } 2326 2327 static inline int drbd_set_ed_uuid(struct drbd_device *device, u64 val) 2328 { 2329 int changed = device->ed_uuid != val; 2330 device->ed_uuid = val; 2331 return changed; 2332 } 2333 2334 static inline int drbd_queue_order_type(struct drbd_device *device) 2335 { 2336 /* sorry, we currently have no working implementation 2337 * of distributed TCQ stuff */ 2338 #ifndef QUEUE_ORDERED_NONE 2339 #define QUEUE_ORDERED_NONE 0 2340 #endif 2341 return QUEUE_ORDERED_NONE; 2342 } 2343 2344 static inline struct drbd_connection *first_connection(struct drbd_resource *resource) 2345 { 2346 return list_first_entry_or_null(&resource->connections, 2347 struct drbd_connection, connections); 2348 } 2349 2350 #endif 2351