1 /* 2 drbd_worker.c 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 #include <linux/module.h> 27 #include <linux/drbd.h> 28 #include <linux/sched.h> 29 #include <linux/smp_lock.h> 30 #include <linux/wait.h> 31 #include <linux/mm.h> 32 #include <linux/memcontrol.h> 33 #include <linux/mm_inline.h> 34 #include <linux/slab.h> 35 #include <linux/random.h> 36 #include <linux/string.h> 37 #include <linux/scatterlist.h> 38 39 #include "drbd_int.h" 40 #include "drbd_req.h" 41 42 #define SLEEP_TIME (HZ/10) 43 44 static int w_make_ov_request(struct drbd_conf *mdev, struct drbd_work *w, int cancel); 45 46 47 48 /* defined here: 49 drbd_md_io_complete 50 drbd_endio_write_sec 51 drbd_endio_read_sec 52 drbd_endio_pri 53 54 * more endio handlers: 55 atodb_endio in drbd_actlog.c 56 drbd_bm_async_io_complete in drbd_bitmap.c 57 58 * For all these callbacks, note the following: 59 * The callbacks will be called in irq context by the IDE drivers, 60 * and in Softirqs/Tasklets/BH context by the SCSI drivers. 61 * Try to get the locking right :) 62 * 63 */ 64 65 66 /* About the global_state_lock 67 Each state transition on an device holds a read lock. In case we have 68 to evaluate the sync after dependencies, we grab a write lock, because 69 we need stable states on all devices for that. */ 70 rwlock_t global_state_lock; 71 72 /* used for synchronous meta data and bitmap IO 73 * submitted by drbd_md_sync_page_io() 74 */ 75 void drbd_md_io_complete(struct bio *bio, int error) 76 { 77 struct drbd_md_io *md_io; 78 79 md_io = (struct drbd_md_io *)bio->bi_private; 80 md_io->error = error; 81 82 complete(&md_io->event); 83 } 84 85 /* reads on behalf of the partner, 86 * "submitted" by the receiver 87 */ 88 void drbd_endio_read_sec(struct bio *bio, int error) __releases(local) 89 { 90 unsigned long flags = 0; 91 struct drbd_epoch_entry *e = NULL; 92 struct drbd_conf *mdev; 93 int uptodate = bio_flagged(bio, BIO_UPTODATE); 94 95 e = bio->bi_private; 96 mdev = e->mdev; 97 98 if (error) 99 dev_warn(DEV, "read: error=%d s=%llus\n", error, 100 (unsigned long long)e->sector); 101 if (!error && !uptodate) { 102 dev_warn(DEV, "read: setting error to -EIO s=%llus\n", 103 (unsigned long long)e->sector); 104 /* strange behavior of some lower level drivers... 105 * fail the request by clearing the uptodate flag, 106 * but do not return any error?! */ 107 error = -EIO; 108 } 109 110 D_ASSERT(e->block_id != ID_VACANT); 111 112 spin_lock_irqsave(&mdev->req_lock, flags); 113 mdev->read_cnt += e->size >> 9; 114 list_del(&e->w.list); 115 if (list_empty(&mdev->read_ee)) 116 wake_up(&mdev->ee_wait); 117 spin_unlock_irqrestore(&mdev->req_lock, flags); 118 119 drbd_chk_io_error(mdev, error, FALSE); 120 drbd_queue_work(&mdev->data.work, &e->w); 121 put_ldev(mdev); 122 } 123 124 /* writes on behalf of the partner, or resync writes, 125 * "submitted" by the receiver. 126 */ 127 void drbd_endio_write_sec(struct bio *bio, int error) __releases(local) 128 { 129 unsigned long flags = 0; 130 struct drbd_epoch_entry *e = NULL; 131 struct drbd_conf *mdev; 132 sector_t e_sector; 133 int do_wake; 134 int is_syncer_req; 135 int do_al_complete_io; 136 int uptodate = bio_flagged(bio, BIO_UPTODATE); 137 int is_barrier = bio_rw_flagged(bio, BIO_RW_BARRIER); 138 139 e = bio->bi_private; 140 mdev = e->mdev; 141 142 if (error) 143 dev_warn(DEV, "write: error=%d s=%llus\n", error, 144 (unsigned long long)e->sector); 145 if (!error && !uptodate) { 146 dev_warn(DEV, "write: setting error to -EIO s=%llus\n", 147 (unsigned long long)e->sector); 148 /* strange behavior of some lower level drivers... 149 * fail the request by clearing the uptodate flag, 150 * but do not return any error?! */ 151 error = -EIO; 152 } 153 154 /* error == -ENOTSUPP would be a better test, 155 * alas it is not reliable */ 156 if (error && is_barrier && e->flags & EE_IS_BARRIER) { 157 drbd_bump_write_ordering(mdev, WO_bdev_flush); 158 spin_lock_irqsave(&mdev->req_lock, flags); 159 list_del(&e->w.list); 160 e->w.cb = w_e_reissue; 161 /* put_ldev actually happens below, once we come here again. */ 162 __release(local); 163 spin_unlock_irqrestore(&mdev->req_lock, flags); 164 drbd_queue_work(&mdev->data.work, &e->w); 165 return; 166 } 167 168 D_ASSERT(e->block_id != ID_VACANT); 169 170 spin_lock_irqsave(&mdev->req_lock, flags); 171 mdev->writ_cnt += e->size >> 9; 172 is_syncer_req = is_syncer_block_id(e->block_id); 173 174 /* after we moved e to done_ee, 175 * we may no longer access it, 176 * it may be freed/reused already! 177 * (as soon as we release the req_lock) */ 178 e_sector = e->sector; 179 do_al_complete_io = e->flags & EE_CALL_AL_COMPLETE_IO; 180 181 list_del(&e->w.list); /* has been on active_ee or sync_ee */ 182 list_add_tail(&e->w.list, &mdev->done_ee); 183 184 /* No hlist_del_init(&e->colision) here, we did not send the Ack yet, 185 * neither did we wake possibly waiting conflicting requests. 186 * done from "drbd_process_done_ee" within the appropriate w.cb 187 * (e_end_block/e_end_resync_block) or from _drbd_clear_done_ee */ 188 189 do_wake = is_syncer_req 190 ? list_empty(&mdev->sync_ee) 191 : list_empty(&mdev->active_ee); 192 193 if (error) 194 __drbd_chk_io_error(mdev, FALSE); 195 spin_unlock_irqrestore(&mdev->req_lock, flags); 196 197 if (is_syncer_req) 198 drbd_rs_complete_io(mdev, e_sector); 199 200 if (do_wake) 201 wake_up(&mdev->ee_wait); 202 203 if (do_al_complete_io) 204 drbd_al_complete_io(mdev, e_sector); 205 206 wake_asender(mdev); 207 put_ldev(mdev); 208 209 } 210 211 /* read, readA or write requests on R_PRIMARY coming from drbd_make_request 212 */ 213 void drbd_endio_pri(struct bio *bio, int error) 214 { 215 unsigned long flags; 216 struct drbd_request *req = bio->bi_private; 217 struct drbd_conf *mdev = req->mdev; 218 struct bio_and_error m; 219 enum drbd_req_event what; 220 int uptodate = bio_flagged(bio, BIO_UPTODATE); 221 222 if (error) 223 dev_warn(DEV, "p %s: error=%d\n", 224 bio_data_dir(bio) == WRITE ? "write" : "read", error); 225 if (!error && !uptodate) { 226 dev_warn(DEV, "p %s: setting error to -EIO\n", 227 bio_data_dir(bio) == WRITE ? "write" : "read"); 228 /* strange behavior of some lower level drivers... 229 * fail the request by clearing the uptodate flag, 230 * but do not return any error?! */ 231 error = -EIO; 232 } 233 234 /* to avoid recursion in __req_mod */ 235 if (unlikely(error)) { 236 what = (bio_data_dir(bio) == WRITE) 237 ? write_completed_with_error 238 : (bio_rw(bio) == READA) 239 ? read_completed_with_error 240 : read_ahead_completed_with_error; 241 } else 242 what = completed_ok; 243 244 bio_put(req->private_bio); 245 req->private_bio = ERR_PTR(error); 246 247 spin_lock_irqsave(&mdev->req_lock, flags); 248 __req_mod(req, what, &m); 249 spin_unlock_irqrestore(&mdev->req_lock, flags); 250 251 if (m.bio) 252 complete_master_bio(mdev, &m); 253 } 254 255 int w_io_error(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 256 { 257 struct drbd_request *req = container_of(w, struct drbd_request, w); 258 259 /* NOTE: mdev->ldev can be NULL by the time we get here! */ 260 /* D_ASSERT(mdev->ldev->dc.on_io_error != EP_PASS_ON); */ 261 262 /* the only way this callback is scheduled is from _req_may_be_done, 263 * when it is done and had a local write error, see comments there */ 264 drbd_req_free(req); 265 266 return TRUE; 267 } 268 269 int w_read_retry_remote(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 270 { 271 struct drbd_request *req = container_of(w, struct drbd_request, w); 272 273 /* We should not detach for read io-error, 274 * but try to WRITE the P_DATA_REPLY to the failed location, 275 * to give the disk the chance to relocate that block */ 276 277 spin_lock_irq(&mdev->req_lock); 278 if (cancel || 279 mdev->state.conn < C_CONNECTED || 280 mdev->state.pdsk <= D_INCONSISTENT) { 281 _req_mod(req, send_canceled); 282 spin_unlock_irq(&mdev->req_lock); 283 dev_alert(DEV, "WE ARE LOST. Local IO failure, no peer.\n"); 284 return 1; 285 } 286 spin_unlock_irq(&mdev->req_lock); 287 288 return w_send_read_req(mdev, w, 0); 289 } 290 291 int w_resync_inactive(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 292 { 293 ERR_IF(cancel) return 1; 294 dev_err(DEV, "resync inactive, but callback triggered??\n"); 295 return 1; /* Simply ignore this! */ 296 } 297 298 void drbd_csum(struct drbd_conf *mdev, struct crypto_hash *tfm, struct bio *bio, void *digest) 299 { 300 struct hash_desc desc; 301 struct scatterlist sg; 302 struct bio_vec *bvec; 303 int i; 304 305 desc.tfm = tfm; 306 desc.flags = 0; 307 308 sg_init_table(&sg, 1); 309 crypto_hash_init(&desc); 310 311 __bio_for_each_segment(bvec, bio, i, 0) { 312 sg_set_page(&sg, bvec->bv_page, bvec->bv_len, bvec->bv_offset); 313 crypto_hash_update(&desc, &sg, sg.length); 314 } 315 crypto_hash_final(&desc, digest); 316 } 317 318 static int w_e_send_csum(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 319 { 320 struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); 321 int digest_size; 322 void *digest; 323 int ok; 324 325 D_ASSERT(e->block_id == DRBD_MAGIC + 0xbeef); 326 327 if (unlikely(cancel)) { 328 drbd_free_ee(mdev, e); 329 return 1; 330 } 331 332 if (likely(drbd_bio_uptodate(e->private_bio))) { 333 digest_size = crypto_hash_digestsize(mdev->csums_tfm); 334 digest = kmalloc(digest_size, GFP_NOIO); 335 if (digest) { 336 drbd_csum(mdev, mdev->csums_tfm, e->private_bio, digest); 337 338 inc_rs_pending(mdev); 339 ok = drbd_send_drequest_csum(mdev, 340 e->sector, 341 e->size, 342 digest, 343 digest_size, 344 P_CSUM_RS_REQUEST); 345 kfree(digest); 346 } else { 347 dev_err(DEV, "kmalloc() of digest failed.\n"); 348 ok = 0; 349 } 350 } else 351 ok = 1; 352 353 drbd_free_ee(mdev, e); 354 355 if (unlikely(!ok)) 356 dev_err(DEV, "drbd_send_drequest(..., csum) failed\n"); 357 return ok; 358 } 359 360 #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) 361 362 static int read_for_csum(struct drbd_conf *mdev, sector_t sector, int size) 363 { 364 struct drbd_epoch_entry *e; 365 366 if (!get_ldev(mdev)) 367 return 0; 368 369 /* GFP_TRY, because if there is no memory available right now, this may 370 * be rescheduled for later. It is "only" background resync, after all. */ 371 e = drbd_alloc_ee(mdev, DRBD_MAGIC+0xbeef, sector, size, GFP_TRY); 372 if (!e) { 373 put_ldev(mdev); 374 return 2; 375 } 376 377 spin_lock_irq(&mdev->req_lock); 378 list_add(&e->w.list, &mdev->read_ee); 379 spin_unlock_irq(&mdev->req_lock); 380 381 e->private_bio->bi_end_io = drbd_endio_read_sec; 382 e->private_bio->bi_rw = READ; 383 e->w.cb = w_e_send_csum; 384 385 mdev->read_cnt += size >> 9; 386 drbd_generic_make_request(mdev, DRBD_FAULT_RS_RD, e->private_bio); 387 388 return 1; 389 } 390 391 void resync_timer_fn(unsigned long data) 392 { 393 unsigned long flags; 394 struct drbd_conf *mdev = (struct drbd_conf *) data; 395 int queue; 396 397 spin_lock_irqsave(&mdev->req_lock, flags); 398 399 if (likely(!test_and_clear_bit(STOP_SYNC_TIMER, &mdev->flags))) { 400 queue = 1; 401 if (mdev->state.conn == C_VERIFY_S) 402 mdev->resync_work.cb = w_make_ov_request; 403 else 404 mdev->resync_work.cb = w_make_resync_request; 405 } else { 406 queue = 0; 407 mdev->resync_work.cb = w_resync_inactive; 408 } 409 410 spin_unlock_irqrestore(&mdev->req_lock, flags); 411 412 /* harmless race: list_empty outside data.work.q_lock */ 413 if (list_empty(&mdev->resync_work.list) && queue) 414 drbd_queue_work(&mdev->data.work, &mdev->resync_work); 415 } 416 417 int w_make_resync_request(struct drbd_conf *mdev, 418 struct drbd_work *w, int cancel) 419 { 420 unsigned long bit; 421 sector_t sector; 422 const sector_t capacity = drbd_get_capacity(mdev->this_bdev); 423 int max_segment_size = queue_max_segment_size(mdev->rq_queue); 424 int number, i, size, pe, mx; 425 int align, queued, sndbuf; 426 427 if (unlikely(cancel)) 428 return 1; 429 430 if (unlikely(mdev->state.conn < C_CONNECTED)) { 431 dev_err(DEV, "Confused in w_make_resync_request()! cstate < Connected"); 432 return 0; 433 } 434 435 if (mdev->state.conn != C_SYNC_TARGET) 436 dev_err(DEV, "%s in w_make_resync_request\n", 437 drbd_conn_str(mdev->state.conn)); 438 439 if (!get_ldev(mdev)) { 440 /* Since we only need to access mdev->rsync a 441 get_ldev_if_state(mdev,D_FAILED) would be sufficient, but 442 to continue resync with a broken disk makes no sense at 443 all */ 444 dev_err(DEV, "Disk broke down during resync!\n"); 445 mdev->resync_work.cb = w_resync_inactive; 446 return 1; 447 } 448 449 number = SLEEP_TIME * mdev->sync_conf.rate / ((BM_BLOCK_SIZE/1024)*HZ); 450 pe = atomic_read(&mdev->rs_pending_cnt); 451 452 mutex_lock(&mdev->data.mutex); 453 if (mdev->data.socket) 454 mx = mdev->data.socket->sk->sk_rcvbuf / sizeof(struct p_block_req); 455 else 456 mx = 1; 457 mutex_unlock(&mdev->data.mutex); 458 459 /* For resync rates >160MB/sec, allow more pending RS requests */ 460 if (number > mx) 461 mx = number; 462 463 /* Limit the number of pending RS requests to no more than the peer's receive buffer */ 464 if ((pe + number) > mx) { 465 number = mx - pe; 466 } 467 468 for (i = 0; i < number; i++) { 469 /* Stop generating RS requests, when half of the send buffer is filled */ 470 mutex_lock(&mdev->data.mutex); 471 if (mdev->data.socket) { 472 queued = mdev->data.socket->sk->sk_wmem_queued; 473 sndbuf = mdev->data.socket->sk->sk_sndbuf; 474 } else { 475 queued = 1; 476 sndbuf = 0; 477 } 478 mutex_unlock(&mdev->data.mutex); 479 if (queued > sndbuf / 2) 480 goto requeue; 481 482 next_sector: 483 size = BM_BLOCK_SIZE; 484 bit = drbd_bm_find_next(mdev, mdev->bm_resync_fo); 485 486 if (bit == -1UL) { 487 mdev->bm_resync_fo = drbd_bm_bits(mdev); 488 mdev->resync_work.cb = w_resync_inactive; 489 put_ldev(mdev); 490 return 1; 491 } 492 493 sector = BM_BIT_TO_SECT(bit); 494 495 if (drbd_try_rs_begin_io(mdev, sector)) { 496 mdev->bm_resync_fo = bit; 497 goto requeue; 498 } 499 mdev->bm_resync_fo = bit + 1; 500 501 if (unlikely(drbd_bm_test_bit(mdev, bit) == 0)) { 502 drbd_rs_complete_io(mdev, sector); 503 goto next_sector; 504 } 505 506 #if DRBD_MAX_SEGMENT_SIZE > BM_BLOCK_SIZE 507 /* try to find some adjacent bits. 508 * we stop if we have already the maximum req size. 509 * 510 * Additionally always align bigger requests, in order to 511 * be prepared for all stripe sizes of software RAIDs. 512 * 513 * we _do_ care about the agreed-upon q->max_segment_size 514 * here, as splitting up the requests on the other side is more 515 * difficult. the consequence is, that on lvm and md and other 516 * "indirect" devices, this is dead code, since 517 * q->max_segment_size will be PAGE_SIZE. 518 */ 519 align = 1; 520 for (;;) { 521 if (size + BM_BLOCK_SIZE > max_segment_size) 522 break; 523 524 /* Be always aligned */ 525 if (sector & ((1<<(align+3))-1)) 526 break; 527 528 /* do not cross extent boundaries */ 529 if (((bit+1) & BM_BLOCKS_PER_BM_EXT_MASK) == 0) 530 break; 531 /* now, is it actually dirty, after all? 532 * caution, drbd_bm_test_bit is tri-state for some 533 * obscure reason; ( b == 0 ) would get the out-of-band 534 * only accidentally right because of the "oddly sized" 535 * adjustment below */ 536 if (drbd_bm_test_bit(mdev, bit+1) != 1) 537 break; 538 bit++; 539 size += BM_BLOCK_SIZE; 540 if ((BM_BLOCK_SIZE << align) <= size) 541 align++; 542 i++; 543 } 544 /* if we merged some, 545 * reset the offset to start the next drbd_bm_find_next from */ 546 if (size > BM_BLOCK_SIZE) 547 mdev->bm_resync_fo = bit + 1; 548 #endif 549 550 /* adjust very last sectors, in case we are oddly sized */ 551 if (sector + (size>>9) > capacity) 552 size = (capacity-sector)<<9; 553 if (mdev->agreed_pro_version >= 89 && mdev->csums_tfm) { 554 switch (read_for_csum(mdev, sector, size)) { 555 case 0: /* Disk failure*/ 556 put_ldev(mdev); 557 return 0; 558 case 2: /* Allocation failed */ 559 drbd_rs_complete_io(mdev, sector); 560 mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); 561 goto requeue; 562 /* case 1: everything ok */ 563 } 564 } else { 565 inc_rs_pending(mdev); 566 if (!drbd_send_drequest(mdev, P_RS_DATA_REQUEST, 567 sector, size, ID_SYNCER)) { 568 dev_err(DEV, "drbd_send_drequest() failed, aborting...\n"); 569 dec_rs_pending(mdev); 570 put_ldev(mdev); 571 return 0; 572 } 573 } 574 } 575 576 if (mdev->bm_resync_fo >= drbd_bm_bits(mdev)) { 577 /* last syncer _request_ was sent, 578 * but the P_RS_DATA_REPLY not yet received. sync will end (and 579 * next sync group will resume), as soon as we receive the last 580 * resync data block, and the last bit is cleared. 581 * until then resync "work" is "inactive" ... 582 */ 583 mdev->resync_work.cb = w_resync_inactive; 584 put_ldev(mdev); 585 return 1; 586 } 587 588 requeue: 589 mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME); 590 put_ldev(mdev); 591 return 1; 592 } 593 594 static int w_make_ov_request(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 595 { 596 int number, i, size; 597 sector_t sector; 598 const sector_t capacity = drbd_get_capacity(mdev->this_bdev); 599 600 if (unlikely(cancel)) 601 return 1; 602 603 if (unlikely(mdev->state.conn < C_CONNECTED)) { 604 dev_err(DEV, "Confused in w_make_ov_request()! cstate < Connected"); 605 return 0; 606 } 607 608 number = SLEEP_TIME*mdev->sync_conf.rate / ((BM_BLOCK_SIZE/1024)*HZ); 609 if (atomic_read(&mdev->rs_pending_cnt) > number) 610 goto requeue; 611 612 number -= atomic_read(&mdev->rs_pending_cnt); 613 614 sector = mdev->ov_position; 615 for (i = 0; i < number; i++) { 616 if (sector >= capacity) { 617 mdev->resync_work.cb = w_resync_inactive; 618 return 1; 619 } 620 621 size = BM_BLOCK_SIZE; 622 623 if (drbd_try_rs_begin_io(mdev, sector)) { 624 mdev->ov_position = sector; 625 goto requeue; 626 } 627 628 if (sector + (size>>9) > capacity) 629 size = (capacity-sector)<<9; 630 631 inc_rs_pending(mdev); 632 if (!drbd_send_ov_request(mdev, sector, size)) { 633 dec_rs_pending(mdev); 634 return 0; 635 } 636 sector += BM_SECT_PER_BIT; 637 } 638 mdev->ov_position = sector; 639 640 requeue: 641 mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME); 642 return 1; 643 } 644 645 646 int w_ov_finished(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 647 { 648 kfree(w); 649 ov_oos_print(mdev); 650 drbd_resync_finished(mdev); 651 652 return 1; 653 } 654 655 static int w_resync_finished(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 656 { 657 kfree(w); 658 659 drbd_resync_finished(mdev); 660 661 return 1; 662 } 663 664 int drbd_resync_finished(struct drbd_conf *mdev) 665 { 666 unsigned long db, dt, dbdt; 667 unsigned long n_oos; 668 union drbd_state os, ns; 669 struct drbd_work *w; 670 char *khelper_cmd = NULL; 671 672 /* Remove all elements from the resync LRU. Since future actions 673 * might set bits in the (main) bitmap, then the entries in the 674 * resync LRU would be wrong. */ 675 if (drbd_rs_del_all(mdev)) { 676 /* In case this is not possible now, most probably because 677 * there are P_RS_DATA_REPLY Packets lingering on the worker's 678 * queue (or even the read operations for those packets 679 * is not finished by now). Retry in 100ms. */ 680 681 drbd_kick_lo(mdev); 682 __set_current_state(TASK_INTERRUPTIBLE); 683 schedule_timeout(HZ / 10); 684 w = kmalloc(sizeof(struct drbd_work), GFP_ATOMIC); 685 if (w) { 686 w->cb = w_resync_finished; 687 drbd_queue_work(&mdev->data.work, w); 688 return 1; 689 } 690 dev_err(DEV, "Warn failed to drbd_rs_del_all() and to kmalloc(w).\n"); 691 } 692 693 dt = (jiffies - mdev->rs_start - mdev->rs_paused) / HZ; 694 if (dt <= 0) 695 dt = 1; 696 db = mdev->rs_total; 697 dbdt = Bit2KB(db/dt); 698 mdev->rs_paused /= HZ; 699 700 if (!get_ldev(mdev)) 701 goto out; 702 703 spin_lock_irq(&mdev->req_lock); 704 os = mdev->state; 705 706 /* This protects us against multiple calls (that can happen in the presence 707 of application IO), and against connectivity loss just before we arrive here. */ 708 if (os.conn <= C_CONNECTED) 709 goto out_unlock; 710 711 ns = os; 712 ns.conn = C_CONNECTED; 713 714 dev_info(DEV, "%s done (total %lu sec; paused %lu sec; %lu K/sec)\n", 715 (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) ? 716 "Online verify " : "Resync", 717 dt + mdev->rs_paused, mdev->rs_paused, dbdt); 718 719 n_oos = drbd_bm_total_weight(mdev); 720 721 if (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) { 722 if (n_oos) { 723 dev_alert(DEV, "Online verify found %lu %dk block out of sync!\n", 724 n_oos, Bit2KB(1)); 725 khelper_cmd = "out-of-sync"; 726 } 727 } else { 728 D_ASSERT((n_oos - mdev->rs_failed) == 0); 729 730 if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) 731 khelper_cmd = "after-resync-target"; 732 733 if (mdev->csums_tfm && mdev->rs_total) { 734 const unsigned long s = mdev->rs_same_csum; 735 const unsigned long t = mdev->rs_total; 736 const int ratio = 737 (t == 0) ? 0 : 738 (t < 100000) ? ((s*100)/t) : (s/(t/100)); 739 dev_info(DEV, "%u %% had equal check sums, eliminated: %luK; " 740 "transferred %luK total %luK\n", 741 ratio, 742 Bit2KB(mdev->rs_same_csum), 743 Bit2KB(mdev->rs_total - mdev->rs_same_csum), 744 Bit2KB(mdev->rs_total)); 745 } 746 } 747 748 if (mdev->rs_failed) { 749 dev_info(DEV, " %lu failed blocks\n", mdev->rs_failed); 750 751 if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { 752 ns.disk = D_INCONSISTENT; 753 ns.pdsk = D_UP_TO_DATE; 754 } else { 755 ns.disk = D_UP_TO_DATE; 756 ns.pdsk = D_INCONSISTENT; 757 } 758 } else { 759 ns.disk = D_UP_TO_DATE; 760 ns.pdsk = D_UP_TO_DATE; 761 762 if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { 763 if (mdev->p_uuid) { 764 int i; 765 for (i = UI_BITMAP ; i <= UI_HISTORY_END ; i++) 766 _drbd_uuid_set(mdev, i, mdev->p_uuid[i]); 767 drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_CURRENT]); 768 _drbd_uuid_set(mdev, UI_CURRENT, mdev->p_uuid[UI_CURRENT]); 769 } else { 770 dev_err(DEV, "mdev->p_uuid is NULL! BUG\n"); 771 } 772 } 773 774 drbd_uuid_set_bm(mdev, 0UL); 775 776 if (mdev->p_uuid) { 777 /* Now the two UUID sets are equal, update what we 778 * know of the peer. */ 779 int i; 780 for (i = UI_CURRENT ; i <= UI_HISTORY_END ; i++) 781 mdev->p_uuid[i] = mdev->ldev->md.uuid[i]; 782 } 783 } 784 785 _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); 786 out_unlock: 787 spin_unlock_irq(&mdev->req_lock); 788 put_ldev(mdev); 789 out: 790 mdev->rs_total = 0; 791 mdev->rs_failed = 0; 792 mdev->rs_paused = 0; 793 mdev->ov_start_sector = 0; 794 795 if (test_and_clear_bit(WRITE_BM_AFTER_RESYNC, &mdev->flags)) { 796 dev_warn(DEV, "Writing the whole bitmap, due to failed kmalloc\n"); 797 drbd_queue_bitmap_io(mdev, &drbd_bm_write, NULL, "write from resync_finished"); 798 } 799 800 if (khelper_cmd) 801 drbd_khelper(mdev, khelper_cmd); 802 803 return 1; 804 } 805 806 /* helper */ 807 static void move_to_net_ee_or_free(struct drbd_conf *mdev, struct drbd_epoch_entry *e) 808 { 809 if (drbd_bio_has_active_page(e->private_bio)) { 810 /* This might happen if sendpage() has not finished */ 811 spin_lock_irq(&mdev->req_lock); 812 list_add_tail(&e->w.list, &mdev->net_ee); 813 spin_unlock_irq(&mdev->req_lock); 814 } else 815 drbd_free_ee(mdev, e); 816 } 817 818 /** 819 * w_e_end_data_req() - Worker callback, to send a P_DATA_REPLY packet in response to a P_DATA_REQUEST 820 * @mdev: DRBD device. 821 * @w: work object. 822 * @cancel: The connection will be closed anyways 823 */ 824 int w_e_end_data_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 825 { 826 struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); 827 int ok; 828 829 if (unlikely(cancel)) { 830 drbd_free_ee(mdev, e); 831 dec_unacked(mdev); 832 return 1; 833 } 834 835 if (likely(drbd_bio_uptodate(e->private_bio))) { 836 ok = drbd_send_block(mdev, P_DATA_REPLY, e); 837 } else { 838 if (__ratelimit(&drbd_ratelimit_state)) 839 dev_err(DEV, "Sending NegDReply. sector=%llus.\n", 840 (unsigned long long)e->sector); 841 842 ok = drbd_send_ack(mdev, P_NEG_DREPLY, e); 843 } 844 845 dec_unacked(mdev); 846 847 move_to_net_ee_or_free(mdev, e); 848 849 if (unlikely(!ok)) 850 dev_err(DEV, "drbd_send_block() failed\n"); 851 return ok; 852 } 853 854 /** 855 * w_e_end_rsdata_req() - Worker callback to send a P_RS_DATA_REPLY packet in response to a P_RS_DATA_REQUESTRS 856 * @mdev: DRBD device. 857 * @w: work object. 858 * @cancel: The connection will be closed anyways 859 */ 860 int w_e_end_rsdata_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 861 { 862 struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); 863 int ok; 864 865 if (unlikely(cancel)) { 866 drbd_free_ee(mdev, e); 867 dec_unacked(mdev); 868 return 1; 869 } 870 871 if (get_ldev_if_state(mdev, D_FAILED)) { 872 drbd_rs_complete_io(mdev, e->sector); 873 put_ldev(mdev); 874 } 875 876 if (likely(drbd_bio_uptodate(e->private_bio))) { 877 if (likely(mdev->state.pdsk >= D_INCONSISTENT)) { 878 inc_rs_pending(mdev); 879 ok = drbd_send_block(mdev, P_RS_DATA_REPLY, e); 880 } else { 881 if (__ratelimit(&drbd_ratelimit_state)) 882 dev_err(DEV, "Not sending RSDataReply, " 883 "partner DISKLESS!\n"); 884 ok = 1; 885 } 886 } else { 887 if (__ratelimit(&drbd_ratelimit_state)) 888 dev_err(DEV, "Sending NegRSDReply. sector %llus.\n", 889 (unsigned long long)e->sector); 890 891 ok = drbd_send_ack(mdev, P_NEG_RS_DREPLY, e); 892 893 /* update resync data with failure */ 894 drbd_rs_failed_io(mdev, e->sector, e->size); 895 } 896 897 dec_unacked(mdev); 898 899 move_to_net_ee_or_free(mdev, e); 900 901 if (unlikely(!ok)) 902 dev_err(DEV, "drbd_send_block() failed\n"); 903 return ok; 904 } 905 906 int w_e_end_csum_rs_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 907 { 908 struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); 909 struct digest_info *di; 910 int digest_size; 911 void *digest = NULL; 912 int ok, eq = 0; 913 914 if (unlikely(cancel)) { 915 drbd_free_ee(mdev, e); 916 dec_unacked(mdev); 917 return 1; 918 } 919 920 drbd_rs_complete_io(mdev, e->sector); 921 922 di = (struct digest_info *)(unsigned long)e->block_id; 923 924 if (likely(drbd_bio_uptodate(e->private_bio))) { 925 /* quick hack to try to avoid a race against reconfiguration. 926 * a real fix would be much more involved, 927 * introducing more locking mechanisms */ 928 if (mdev->csums_tfm) { 929 digest_size = crypto_hash_digestsize(mdev->csums_tfm); 930 D_ASSERT(digest_size == di->digest_size); 931 digest = kmalloc(digest_size, GFP_NOIO); 932 } 933 if (digest) { 934 drbd_csum(mdev, mdev->csums_tfm, e->private_bio, digest); 935 eq = !memcmp(digest, di->digest, digest_size); 936 kfree(digest); 937 } 938 939 if (eq) { 940 drbd_set_in_sync(mdev, e->sector, e->size); 941 /* rs_same_csums unit is BM_BLOCK_SIZE */ 942 mdev->rs_same_csum += e->size >> BM_BLOCK_SHIFT; 943 ok = drbd_send_ack(mdev, P_RS_IS_IN_SYNC, e); 944 } else { 945 inc_rs_pending(mdev); 946 e->block_id = ID_SYNCER; 947 ok = drbd_send_block(mdev, P_RS_DATA_REPLY, e); 948 } 949 } else { 950 ok = drbd_send_ack(mdev, P_NEG_RS_DREPLY, e); 951 if (__ratelimit(&drbd_ratelimit_state)) 952 dev_err(DEV, "Sending NegDReply. I guess it gets messy.\n"); 953 } 954 955 dec_unacked(mdev); 956 957 kfree(di); 958 959 move_to_net_ee_or_free(mdev, e); 960 961 if (unlikely(!ok)) 962 dev_err(DEV, "drbd_send_block/ack() failed\n"); 963 return ok; 964 } 965 966 int w_e_end_ov_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 967 { 968 struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); 969 int digest_size; 970 void *digest; 971 int ok = 1; 972 973 if (unlikely(cancel)) 974 goto out; 975 976 if (unlikely(!drbd_bio_uptodate(e->private_bio))) 977 goto out; 978 979 digest_size = crypto_hash_digestsize(mdev->verify_tfm); 980 /* FIXME if this allocation fails, online verify will not terminate! */ 981 digest = kmalloc(digest_size, GFP_NOIO); 982 if (digest) { 983 drbd_csum(mdev, mdev->verify_tfm, e->private_bio, digest); 984 inc_rs_pending(mdev); 985 ok = drbd_send_drequest_csum(mdev, e->sector, e->size, 986 digest, digest_size, P_OV_REPLY); 987 if (!ok) 988 dec_rs_pending(mdev); 989 kfree(digest); 990 } 991 992 out: 993 drbd_free_ee(mdev, e); 994 995 dec_unacked(mdev); 996 997 return ok; 998 } 999 1000 void drbd_ov_oos_found(struct drbd_conf *mdev, sector_t sector, int size) 1001 { 1002 if (mdev->ov_last_oos_start + mdev->ov_last_oos_size == sector) { 1003 mdev->ov_last_oos_size += size>>9; 1004 } else { 1005 mdev->ov_last_oos_start = sector; 1006 mdev->ov_last_oos_size = size>>9; 1007 } 1008 drbd_set_out_of_sync(mdev, sector, size); 1009 set_bit(WRITE_BM_AFTER_RESYNC, &mdev->flags); 1010 } 1011 1012 int w_e_end_ov_reply(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1013 { 1014 struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); 1015 struct digest_info *di; 1016 int digest_size; 1017 void *digest; 1018 int ok, eq = 0; 1019 1020 if (unlikely(cancel)) { 1021 drbd_free_ee(mdev, e); 1022 dec_unacked(mdev); 1023 return 1; 1024 } 1025 1026 /* after "cancel", because after drbd_disconnect/drbd_rs_cancel_all 1027 * the resync lru has been cleaned up already */ 1028 drbd_rs_complete_io(mdev, e->sector); 1029 1030 di = (struct digest_info *)(unsigned long)e->block_id; 1031 1032 if (likely(drbd_bio_uptodate(e->private_bio))) { 1033 digest_size = crypto_hash_digestsize(mdev->verify_tfm); 1034 digest = kmalloc(digest_size, GFP_NOIO); 1035 if (digest) { 1036 drbd_csum(mdev, mdev->verify_tfm, e->private_bio, digest); 1037 1038 D_ASSERT(digest_size == di->digest_size); 1039 eq = !memcmp(digest, di->digest, digest_size); 1040 kfree(digest); 1041 } 1042 } else { 1043 ok = drbd_send_ack(mdev, P_NEG_RS_DREPLY, e); 1044 if (__ratelimit(&drbd_ratelimit_state)) 1045 dev_err(DEV, "Sending NegDReply. I guess it gets messy.\n"); 1046 } 1047 1048 dec_unacked(mdev); 1049 1050 kfree(di); 1051 1052 if (!eq) 1053 drbd_ov_oos_found(mdev, e->sector, e->size); 1054 else 1055 ov_oos_print(mdev); 1056 1057 ok = drbd_send_ack_ex(mdev, P_OV_RESULT, e->sector, e->size, 1058 eq ? ID_IN_SYNC : ID_OUT_OF_SYNC); 1059 1060 drbd_free_ee(mdev, e); 1061 1062 if (--mdev->ov_left == 0) { 1063 ov_oos_print(mdev); 1064 drbd_resync_finished(mdev); 1065 } 1066 1067 return ok; 1068 } 1069 1070 int w_prev_work_done(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1071 { 1072 struct drbd_wq_barrier *b = container_of(w, struct drbd_wq_barrier, w); 1073 complete(&b->done); 1074 return 1; 1075 } 1076 1077 int w_send_barrier(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1078 { 1079 struct drbd_tl_epoch *b = container_of(w, struct drbd_tl_epoch, w); 1080 struct p_barrier *p = &mdev->data.sbuf.barrier; 1081 int ok = 1; 1082 1083 /* really avoid racing with tl_clear. w.cb may have been referenced 1084 * just before it was reassigned and re-queued, so double check that. 1085 * actually, this race was harmless, since we only try to send the 1086 * barrier packet here, and otherwise do nothing with the object. 1087 * but compare with the head of w_clear_epoch */ 1088 spin_lock_irq(&mdev->req_lock); 1089 if (w->cb != w_send_barrier || mdev->state.conn < C_CONNECTED) 1090 cancel = 1; 1091 spin_unlock_irq(&mdev->req_lock); 1092 if (cancel) 1093 return 1; 1094 1095 if (!drbd_get_data_sock(mdev)) 1096 return 0; 1097 p->barrier = b->br_number; 1098 /* inc_ap_pending was done where this was queued. 1099 * dec_ap_pending will be done in got_BarrierAck 1100 * or (on connection loss) in w_clear_epoch. */ 1101 ok = _drbd_send_cmd(mdev, mdev->data.socket, P_BARRIER, 1102 (struct p_header *)p, sizeof(*p), 0); 1103 drbd_put_data_sock(mdev); 1104 1105 return ok; 1106 } 1107 1108 int w_send_write_hint(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1109 { 1110 if (cancel) 1111 return 1; 1112 return drbd_send_short_cmd(mdev, P_UNPLUG_REMOTE); 1113 } 1114 1115 /** 1116 * w_send_dblock() - Worker callback to send a P_DATA packet in order to mirror a write request 1117 * @mdev: DRBD device. 1118 * @w: work object. 1119 * @cancel: The connection will be closed anyways 1120 */ 1121 int w_send_dblock(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1122 { 1123 struct drbd_request *req = container_of(w, struct drbd_request, w); 1124 int ok; 1125 1126 if (unlikely(cancel)) { 1127 req_mod(req, send_canceled); 1128 return 1; 1129 } 1130 1131 ok = drbd_send_dblock(mdev, req); 1132 req_mod(req, ok ? handed_over_to_network : send_failed); 1133 1134 return ok; 1135 } 1136 1137 /** 1138 * w_send_read_req() - Worker callback to send a read request (P_DATA_REQUEST) packet 1139 * @mdev: DRBD device. 1140 * @w: work object. 1141 * @cancel: The connection will be closed anyways 1142 */ 1143 int w_send_read_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1144 { 1145 struct drbd_request *req = container_of(w, struct drbd_request, w); 1146 int ok; 1147 1148 if (unlikely(cancel)) { 1149 req_mod(req, send_canceled); 1150 return 1; 1151 } 1152 1153 ok = drbd_send_drequest(mdev, P_DATA_REQUEST, req->sector, req->size, 1154 (unsigned long)req); 1155 1156 if (!ok) { 1157 /* ?? we set C_TIMEOUT or C_BROKEN_PIPE in drbd_send(); 1158 * so this is probably redundant */ 1159 if (mdev->state.conn >= C_CONNECTED) 1160 drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE)); 1161 } 1162 req_mod(req, ok ? handed_over_to_network : send_failed); 1163 1164 return ok; 1165 } 1166 1167 static int _drbd_may_sync_now(struct drbd_conf *mdev) 1168 { 1169 struct drbd_conf *odev = mdev; 1170 1171 while (1) { 1172 if (odev->sync_conf.after == -1) 1173 return 1; 1174 odev = minor_to_mdev(odev->sync_conf.after); 1175 ERR_IF(!odev) return 1; 1176 if ((odev->state.conn >= C_SYNC_SOURCE && 1177 odev->state.conn <= C_PAUSED_SYNC_T) || 1178 odev->state.aftr_isp || odev->state.peer_isp || 1179 odev->state.user_isp) 1180 return 0; 1181 } 1182 } 1183 1184 /** 1185 * _drbd_pause_after() - Pause resync on all devices that may not resync now 1186 * @mdev: DRBD device. 1187 * 1188 * Called from process context only (admin command and after_state_ch). 1189 */ 1190 static int _drbd_pause_after(struct drbd_conf *mdev) 1191 { 1192 struct drbd_conf *odev; 1193 int i, rv = 0; 1194 1195 for (i = 0; i < minor_count; i++) { 1196 odev = minor_to_mdev(i); 1197 if (!odev) 1198 continue; 1199 if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) 1200 continue; 1201 if (!_drbd_may_sync_now(odev)) 1202 rv |= (__drbd_set_state(_NS(odev, aftr_isp, 1), CS_HARD, NULL) 1203 != SS_NOTHING_TO_DO); 1204 } 1205 1206 return rv; 1207 } 1208 1209 /** 1210 * _drbd_resume_next() - Resume resync on all devices that may resync now 1211 * @mdev: DRBD device. 1212 * 1213 * Called from process context only (admin command and worker). 1214 */ 1215 static int _drbd_resume_next(struct drbd_conf *mdev) 1216 { 1217 struct drbd_conf *odev; 1218 int i, rv = 0; 1219 1220 for (i = 0; i < minor_count; i++) { 1221 odev = minor_to_mdev(i); 1222 if (!odev) 1223 continue; 1224 if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) 1225 continue; 1226 if (odev->state.aftr_isp) { 1227 if (_drbd_may_sync_now(odev)) 1228 rv |= (__drbd_set_state(_NS(odev, aftr_isp, 0), 1229 CS_HARD, NULL) 1230 != SS_NOTHING_TO_DO) ; 1231 } 1232 } 1233 return rv; 1234 } 1235 1236 void resume_next_sg(struct drbd_conf *mdev) 1237 { 1238 write_lock_irq(&global_state_lock); 1239 _drbd_resume_next(mdev); 1240 write_unlock_irq(&global_state_lock); 1241 } 1242 1243 void suspend_other_sg(struct drbd_conf *mdev) 1244 { 1245 write_lock_irq(&global_state_lock); 1246 _drbd_pause_after(mdev); 1247 write_unlock_irq(&global_state_lock); 1248 } 1249 1250 static int sync_after_error(struct drbd_conf *mdev, int o_minor) 1251 { 1252 struct drbd_conf *odev; 1253 1254 if (o_minor == -1) 1255 return NO_ERROR; 1256 if (o_minor < -1 || minor_to_mdev(o_minor) == NULL) 1257 return ERR_SYNC_AFTER; 1258 1259 /* check for loops */ 1260 odev = minor_to_mdev(o_minor); 1261 while (1) { 1262 if (odev == mdev) 1263 return ERR_SYNC_AFTER_CYCLE; 1264 1265 /* dependency chain ends here, no cycles. */ 1266 if (odev->sync_conf.after == -1) 1267 return NO_ERROR; 1268 1269 /* follow the dependency chain */ 1270 odev = minor_to_mdev(odev->sync_conf.after); 1271 } 1272 } 1273 1274 int drbd_alter_sa(struct drbd_conf *mdev, int na) 1275 { 1276 int changes; 1277 int retcode; 1278 1279 write_lock_irq(&global_state_lock); 1280 retcode = sync_after_error(mdev, na); 1281 if (retcode == NO_ERROR) { 1282 mdev->sync_conf.after = na; 1283 do { 1284 changes = _drbd_pause_after(mdev); 1285 changes |= _drbd_resume_next(mdev); 1286 } while (changes); 1287 } 1288 write_unlock_irq(&global_state_lock); 1289 return retcode; 1290 } 1291 1292 static void ping_peer(struct drbd_conf *mdev) 1293 { 1294 clear_bit(GOT_PING_ACK, &mdev->flags); 1295 request_ping(mdev); 1296 wait_event(mdev->misc_wait, 1297 test_bit(GOT_PING_ACK, &mdev->flags) || mdev->state.conn < C_CONNECTED); 1298 } 1299 1300 /** 1301 * drbd_start_resync() - Start the resync process 1302 * @mdev: DRBD device. 1303 * @side: Either C_SYNC_SOURCE or C_SYNC_TARGET 1304 * 1305 * This function might bring you directly into one of the 1306 * C_PAUSED_SYNC_* states. 1307 */ 1308 void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side) 1309 { 1310 union drbd_state ns; 1311 int r; 1312 1313 if (mdev->state.conn >= C_SYNC_SOURCE) { 1314 dev_err(DEV, "Resync already running!\n"); 1315 return; 1316 } 1317 1318 /* In case a previous resync run was aborted by an IO error/detach on the peer. */ 1319 drbd_rs_cancel_all(mdev); 1320 1321 if (side == C_SYNC_TARGET) { 1322 /* Since application IO was locked out during C_WF_BITMAP_T and 1323 C_WF_SYNC_UUID we are still unmodified. Before going to C_SYNC_TARGET 1324 we check that we might make the data inconsistent. */ 1325 r = drbd_khelper(mdev, "before-resync-target"); 1326 r = (r >> 8) & 0xff; 1327 if (r > 0) { 1328 dev_info(DEV, "before-resync-target handler returned %d, " 1329 "dropping connection.\n", r); 1330 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 1331 return; 1332 } 1333 } 1334 1335 drbd_state_lock(mdev); 1336 1337 if (!get_ldev_if_state(mdev, D_NEGOTIATING)) { 1338 drbd_state_unlock(mdev); 1339 return; 1340 } 1341 1342 if (side == C_SYNC_TARGET) { 1343 mdev->bm_resync_fo = 0; 1344 } else /* side == C_SYNC_SOURCE */ { 1345 u64 uuid; 1346 1347 get_random_bytes(&uuid, sizeof(u64)); 1348 drbd_uuid_set(mdev, UI_BITMAP, uuid); 1349 drbd_send_sync_uuid(mdev, uuid); 1350 1351 D_ASSERT(mdev->state.disk == D_UP_TO_DATE); 1352 } 1353 1354 write_lock_irq(&global_state_lock); 1355 ns = mdev->state; 1356 1357 ns.aftr_isp = !_drbd_may_sync_now(mdev); 1358 1359 ns.conn = side; 1360 1361 if (side == C_SYNC_TARGET) 1362 ns.disk = D_INCONSISTENT; 1363 else /* side == C_SYNC_SOURCE */ 1364 ns.pdsk = D_INCONSISTENT; 1365 1366 r = __drbd_set_state(mdev, ns, CS_VERBOSE, NULL); 1367 ns = mdev->state; 1368 1369 if (ns.conn < C_CONNECTED) 1370 r = SS_UNKNOWN_ERROR; 1371 1372 if (r == SS_SUCCESS) { 1373 mdev->rs_total = 1374 mdev->rs_mark_left = drbd_bm_total_weight(mdev); 1375 mdev->rs_failed = 0; 1376 mdev->rs_paused = 0; 1377 mdev->rs_start = 1378 mdev->rs_mark_time = jiffies; 1379 mdev->rs_same_csum = 0; 1380 _drbd_pause_after(mdev); 1381 } 1382 write_unlock_irq(&global_state_lock); 1383 put_ldev(mdev); 1384 1385 if (r == SS_SUCCESS) { 1386 dev_info(DEV, "Began resync as %s (will sync %lu KB [%lu bits set]).\n", 1387 drbd_conn_str(ns.conn), 1388 (unsigned long) mdev->rs_total << (BM_BLOCK_SHIFT-10), 1389 (unsigned long) mdev->rs_total); 1390 1391 if (mdev->rs_total == 0) { 1392 /* Peer still reachable? Beware of failing before-resync-target handlers! */ 1393 ping_peer(mdev); 1394 drbd_resync_finished(mdev); 1395 } 1396 1397 /* ns.conn may already be != mdev->state.conn, 1398 * we may have been paused in between, or become paused until 1399 * the timer triggers. 1400 * No matter, that is handled in resync_timer_fn() */ 1401 if (ns.conn == C_SYNC_TARGET) 1402 mod_timer(&mdev->resync_timer, jiffies); 1403 1404 drbd_md_sync(mdev); 1405 } 1406 drbd_state_unlock(mdev); 1407 } 1408 1409 int drbd_worker(struct drbd_thread *thi) 1410 { 1411 struct drbd_conf *mdev = thi->mdev; 1412 struct drbd_work *w = NULL; 1413 LIST_HEAD(work_list); 1414 int intr = 0, i; 1415 1416 sprintf(current->comm, "drbd%d_worker", mdev_to_minor(mdev)); 1417 1418 while (get_t_state(thi) == Running) { 1419 drbd_thread_current_set_cpu(mdev); 1420 1421 if (down_trylock(&mdev->data.work.s)) { 1422 mutex_lock(&mdev->data.mutex); 1423 if (mdev->data.socket && !mdev->net_conf->no_cork) 1424 drbd_tcp_uncork(mdev->data.socket); 1425 mutex_unlock(&mdev->data.mutex); 1426 1427 intr = down_interruptible(&mdev->data.work.s); 1428 1429 mutex_lock(&mdev->data.mutex); 1430 if (mdev->data.socket && !mdev->net_conf->no_cork) 1431 drbd_tcp_cork(mdev->data.socket); 1432 mutex_unlock(&mdev->data.mutex); 1433 } 1434 1435 if (intr) { 1436 D_ASSERT(intr == -EINTR); 1437 flush_signals(current); 1438 ERR_IF (get_t_state(thi) == Running) 1439 continue; 1440 break; 1441 } 1442 1443 if (get_t_state(thi) != Running) 1444 break; 1445 /* With this break, we have done a down() but not consumed 1446 the entry from the list. The cleanup code takes care of 1447 this... */ 1448 1449 w = NULL; 1450 spin_lock_irq(&mdev->data.work.q_lock); 1451 ERR_IF(list_empty(&mdev->data.work.q)) { 1452 /* something terribly wrong in our logic. 1453 * we were able to down() the semaphore, 1454 * but the list is empty... doh. 1455 * 1456 * what is the best thing to do now? 1457 * try again from scratch, restarting the receiver, 1458 * asender, whatnot? could break even more ugly, 1459 * e.g. when we are primary, but no good local data. 1460 * 1461 * I'll try to get away just starting over this loop. 1462 */ 1463 spin_unlock_irq(&mdev->data.work.q_lock); 1464 continue; 1465 } 1466 w = list_entry(mdev->data.work.q.next, struct drbd_work, list); 1467 list_del_init(&w->list); 1468 spin_unlock_irq(&mdev->data.work.q_lock); 1469 1470 if (!w->cb(mdev, w, mdev->state.conn < C_CONNECTED)) { 1471 /* dev_warn(DEV, "worker: a callback failed! \n"); */ 1472 if (mdev->state.conn >= C_CONNECTED) 1473 drbd_force_state(mdev, 1474 NS(conn, C_NETWORK_FAILURE)); 1475 } 1476 } 1477 D_ASSERT(test_bit(DEVICE_DYING, &mdev->flags)); 1478 D_ASSERT(test_bit(CONFIG_PENDING, &mdev->flags)); 1479 1480 spin_lock_irq(&mdev->data.work.q_lock); 1481 i = 0; 1482 while (!list_empty(&mdev->data.work.q)) { 1483 list_splice_init(&mdev->data.work.q, &work_list); 1484 spin_unlock_irq(&mdev->data.work.q_lock); 1485 1486 while (!list_empty(&work_list)) { 1487 w = list_entry(work_list.next, struct drbd_work, list); 1488 list_del_init(&w->list); 1489 w->cb(mdev, w, 1); 1490 i++; /* dead debugging code */ 1491 } 1492 1493 spin_lock_irq(&mdev->data.work.q_lock); 1494 } 1495 sema_init(&mdev->data.work.s, 0); 1496 /* DANGEROUS race: if someone did queue his work within the spinlock, 1497 * but up() ed outside the spinlock, we could get an up() on the 1498 * semaphore without corresponding list entry. 1499 * So don't do that. 1500 */ 1501 spin_unlock_irq(&mdev->data.work.q_lock); 1502 1503 D_ASSERT(mdev->state.disk == D_DISKLESS && mdev->state.conn == C_STANDALONE); 1504 /* _drbd_set_state only uses stop_nowait. 1505 * wait here for the Exiting receiver. */ 1506 drbd_thread_stop(&mdev->receiver); 1507 drbd_mdev_cleanup(mdev); 1508 1509 dev_info(DEV, "worker terminated\n"); 1510 1511 clear_bit(DEVICE_DYING, &mdev->flags); 1512 clear_bit(CONFIG_PENDING, &mdev->flags); 1513 wake_up(&mdev->state_wait); 1514 1515 return 0; 1516 } 1517