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