1b411b363SPhilipp Reisner /* 2b411b363SPhilipp Reisner drbd_worker.c 3b411b363SPhilipp Reisner 4b411b363SPhilipp Reisner This file is part of DRBD by Philipp Reisner and Lars Ellenberg. 5b411b363SPhilipp Reisner 6b411b363SPhilipp Reisner Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. 7b411b363SPhilipp Reisner Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. 8b411b363SPhilipp Reisner Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. 9b411b363SPhilipp Reisner 10b411b363SPhilipp Reisner drbd is free software; you can redistribute it and/or modify 11b411b363SPhilipp Reisner it under the terms of the GNU General Public License as published by 12b411b363SPhilipp Reisner the Free Software Foundation; either version 2, or (at your option) 13b411b363SPhilipp Reisner any later version. 14b411b363SPhilipp Reisner 15b411b363SPhilipp Reisner drbd is distributed in the hope that it will be useful, 16b411b363SPhilipp Reisner but WITHOUT ANY WARRANTY; without even the implied warranty of 17b411b363SPhilipp Reisner MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 18b411b363SPhilipp Reisner GNU General Public License for more details. 19b411b363SPhilipp Reisner 20b411b363SPhilipp Reisner You should have received a copy of the GNU General Public License 21b411b363SPhilipp Reisner along with drbd; see the file COPYING. If not, write to 22b411b363SPhilipp Reisner the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. 23b411b363SPhilipp Reisner 24b411b363SPhilipp Reisner */ 25b411b363SPhilipp Reisner 26b411b363SPhilipp Reisner #include <linux/module.h> 27b411b363SPhilipp Reisner #include <linux/drbd.h> 28b411b363SPhilipp Reisner #include <linux/sched.h> 29b411b363SPhilipp Reisner #include <linux/wait.h> 30b411b363SPhilipp Reisner #include <linux/mm.h> 31b411b363SPhilipp Reisner #include <linux/memcontrol.h> 32b411b363SPhilipp Reisner #include <linux/mm_inline.h> 33b411b363SPhilipp Reisner #include <linux/slab.h> 34b411b363SPhilipp Reisner #include <linux/random.h> 35b411b363SPhilipp Reisner #include <linux/string.h> 36b411b363SPhilipp Reisner #include <linux/scatterlist.h> 37b411b363SPhilipp Reisner 38b411b363SPhilipp Reisner #include "drbd_int.h" 39b411b363SPhilipp Reisner #include "drbd_req.h" 40b411b363SPhilipp Reisner 41b411b363SPhilipp Reisner static int w_make_ov_request(struct drbd_conf *mdev, struct drbd_work *w, int cancel); 42b411b363SPhilipp Reisner 43b411b363SPhilipp Reisner 44b411b363SPhilipp Reisner 45b411b363SPhilipp Reisner /* defined here: 46b411b363SPhilipp Reisner drbd_md_io_complete 4745bb912bSLars Ellenberg drbd_endio_sec 48b411b363SPhilipp Reisner drbd_endio_pri 49b411b363SPhilipp Reisner 50b411b363SPhilipp Reisner * more endio handlers: 51b411b363SPhilipp Reisner atodb_endio in drbd_actlog.c 52b411b363SPhilipp Reisner drbd_bm_async_io_complete in drbd_bitmap.c 53b411b363SPhilipp Reisner 54b411b363SPhilipp Reisner * For all these callbacks, note the following: 55b411b363SPhilipp Reisner * The callbacks will be called in irq context by the IDE drivers, 56b411b363SPhilipp Reisner * and in Softirqs/Tasklets/BH context by the SCSI drivers. 57b411b363SPhilipp Reisner * Try to get the locking right :) 58b411b363SPhilipp Reisner * 59b411b363SPhilipp Reisner */ 60b411b363SPhilipp Reisner 61b411b363SPhilipp Reisner 62b411b363SPhilipp Reisner /* About the global_state_lock 63b411b363SPhilipp Reisner Each state transition on an device holds a read lock. In case we have 64b411b363SPhilipp Reisner to evaluate the sync after dependencies, we grab a write lock, because 65b411b363SPhilipp Reisner we need stable states on all devices for that. */ 66b411b363SPhilipp Reisner rwlock_t global_state_lock; 67b411b363SPhilipp Reisner 68b411b363SPhilipp Reisner /* used for synchronous meta data and bitmap IO 69b411b363SPhilipp Reisner * submitted by drbd_md_sync_page_io() 70b411b363SPhilipp Reisner */ 71b411b363SPhilipp Reisner void drbd_md_io_complete(struct bio *bio, int error) 72b411b363SPhilipp Reisner { 73b411b363SPhilipp Reisner struct drbd_md_io *md_io; 74b411b363SPhilipp Reisner 75b411b363SPhilipp Reisner md_io = (struct drbd_md_io *)bio->bi_private; 76b411b363SPhilipp Reisner md_io->error = error; 77b411b363SPhilipp Reisner 78b411b363SPhilipp Reisner complete(&md_io->event); 79b411b363SPhilipp Reisner } 80b411b363SPhilipp Reisner 81b411b363SPhilipp Reisner /* reads on behalf of the partner, 82b411b363SPhilipp Reisner * "submitted" by the receiver 83b411b363SPhilipp Reisner */ 8445bb912bSLars Ellenberg void drbd_endio_read_sec_final(struct drbd_epoch_entry *e) __releases(local) 85b411b363SPhilipp Reisner { 86b411b363SPhilipp Reisner unsigned long flags = 0; 8745bb912bSLars Ellenberg struct drbd_conf *mdev = e->mdev; 88b411b363SPhilipp Reisner 89b411b363SPhilipp Reisner D_ASSERT(e->block_id != ID_VACANT); 90b411b363SPhilipp Reisner 91b411b363SPhilipp Reisner spin_lock_irqsave(&mdev->req_lock, flags); 92b411b363SPhilipp Reisner mdev->read_cnt += e->size >> 9; 93b411b363SPhilipp Reisner list_del(&e->w.list); 94b411b363SPhilipp Reisner if (list_empty(&mdev->read_ee)) 95b411b363SPhilipp Reisner wake_up(&mdev->ee_wait); 9645bb912bSLars Ellenberg if (test_bit(__EE_WAS_ERROR, &e->flags)) 9745bb912bSLars Ellenberg __drbd_chk_io_error(mdev, FALSE); 98b411b363SPhilipp Reisner spin_unlock_irqrestore(&mdev->req_lock, flags); 99b411b363SPhilipp Reisner 100b411b363SPhilipp Reisner drbd_queue_work(&mdev->data.work, &e->w); 101b411b363SPhilipp Reisner put_ldev(mdev); 102b411b363SPhilipp Reisner } 103b411b363SPhilipp Reisner 104b411b363SPhilipp Reisner /* writes on behalf of the partner, or resync writes, 10545bb912bSLars Ellenberg * "submitted" by the receiver, final stage. */ 10645bb912bSLars Ellenberg static void drbd_endio_write_sec_final(struct drbd_epoch_entry *e) __releases(local) 107b411b363SPhilipp Reisner { 108b411b363SPhilipp Reisner unsigned long flags = 0; 10945bb912bSLars Ellenberg struct drbd_conf *mdev = e->mdev; 110b411b363SPhilipp Reisner sector_t e_sector; 111b411b363SPhilipp Reisner int do_wake; 112b411b363SPhilipp Reisner int is_syncer_req; 113b411b363SPhilipp Reisner int do_al_complete_io; 114b411b363SPhilipp Reisner 115b411b363SPhilipp Reisner D_ASSERT(e->block_id != ID_VACANT); 116b411b363SPhilipp Reisner 117b411b363SPhilipp Reisner /* after we moved e to done_ee, 118b411b363SPhilipp Reisner * we may no longer access it, 119b411b363SPhilipp Reisner * it may be freed/reused already! 120b411b363SPhilipp Reisner * (as soon as we release the req_lock) */ 121b411b363SPhilipp Reisner e_sector = e->sector; 122b411b363SPhilipp Reisner do_al_complete_io = e->flags & EE_CALL_AL_COMPLETE_IO; 12345bb912bSLars Ellenberg is_syncer_req = is_syncer_block_id(e->block_id); 124b411b363SPhilipp Reisner 12545bb912bSLars Ellenberg spin_lock_irqsave(&mdev->req_lock, flags); 12645bb912bSLars Ellenberg mdev->writ_cnt += e->size >> 9; 127b411b363SPhilipp Reisner list_del(&e->w.list); /* has been on active_ee or sync_ee */ 128b411b363SPhilipp Reisner list_add_tail(&e->w.list, &mdev->done_ee); 129b411b363SPhilipp Reisner 130b411b363SPhilipp Reisner /* No hlist_del_init(&e->colision) here, we did not send the Ack yet, 131b411b363SPhilipp Reisner * neither did we wake possibly waiting conflicting requests. 132b411b363SPhilipp Reisner * done from "drbd_process_done_ee" within the appropriate w.cb 133b411b363SPhilipp Reisner * (e_end_block/e_end_resync_block) or from _drbd_clear_done_ee */ 134b411b363SPhilipp Reisner 135b411b363SPhilipp Reisner do_wake = is_syncer_req 136b411b363SPhilipp Reisner ? list_empty(&mdev->sync_ee) 137b411b363SPhilipp Reisner : list_empty(&mdev->active_ee); 138b411b363SPhilipp Reisner 13945bb912bSLars Ellenberg if (test_bit(__EE_WAS_ERROR, &e->flags)) 140b411b363SPhilipp Reisner __drbd_chk_io_error(mdev, FALSE); 141b411b363SPhilipp Reisner spin_unlock_irqrestore(&mdev->req_lock, flags); 142b411b363SPhilipp Reisner 143b411b363SPhilipp Reisner if (is_syncer_req) 144b411b363SPhilipp Reisner drbd_rs_complete_io(mdev, e_sector); 145b411b363SPhilipp Reisner 146b411b363SPhilipp Reisner if (do_wake) 147b411b363SPhilipp Reisner wake_up(&mdev->ee_wait); 148b411b363SPhilipp Reisner 149b411b363SPhilipp Reisner if (do_al_complete_io) 150b411b363SPhilipp Reisner drbd_al_complete_io(mdev, e_sector); 151b411b363SPhilipp Reisner 152b411b363SPhilipp Reisner wake_asender(mdev); 153b411b363SPhilipp Reisner put_ldev(mdev); 15445bb912bSLars Ellenberg } 155b411b363SPhilipp Reisner 15645bb912bSLars Ellenberg /* writes on behalf of the partner, or resync writes, 15745bb912bSLars Ellenberg * "submitted" by the receiver. 15845bb912bSLars Ellenberg */ 15945bb912bSLars Ellenberg void drbd_endio_sec(struct bio *bio, int error) 16045bb912bSLars Ellenberg { 16145bb912bSLars Ellenberg struct drbd_epoch_entry *e = bio->bi_private; 16245bb912bSLars Ellenberg struct drbd_conf *mdev = e->mdev; 16345bb912bSLars Ellenberg int uptodate = bio_flagged(bio, BIO_UPTODATE); 16445bb912bSLars Ellenberg int is_write = bio_data_dir(bio) == WRITE; 16545bb912bSLars Ellenberg 16645bb912bSLars Ellenberg if (error) 16745bb912bSLars Ellenberg dev_warn(DEV, "%s: error=%d s=%llus\n", 16845bb912bSLars Ellenberg is_write ? "write" : "read", error, 16945bb912bSLars Ellenberg (unsigned long long)e->sector); 17045bb912bSLars Ellenberg if (!error && !uptodate) { 17145bb912bSLars Ellenberg dev_warn(DEV, "%s: setting error to -EIO s=%llus\n", 17245bb912bSLars Ellenberg is_write ? "write" : "read", 17345bb912bSLars Ellenberg (unsigned long long)e->sector); 17445bb912bSLars Ellenberg /* strange behavior of some lower level drivers... 17545bb912bSLars Ellenberg * fail the request by clearing the uptodate flag, 17645bb912bSLars Ellenberg * but do not return any error?! */ 17745bb912bSLars Ellenberg error = -EIO; 17845bb912bSLars Ellenberg } 17945bb912bSLars Ellenberg 18045bb912bSLars Ellenberg if (error) 18145bb912bSLars Ellenberg set_bit(__EE_WAS_ERROR, &e->flags); 18245bb912bSLars Ellenberg 18345bb912bSLars Ellenberg bio_put(bio); /* no need for the bio anymore */ 18445bb912bSLars Ellenberg if (atomic_dec_and_test(&e->pending_bios)) { 18545bb912bSLars Ellenberg if (is_write) 18645bb912bSLars Ellenberg drbd_endio_write_sec_final(e); 18745bb912bSLars Ellenberg else 18845bb912bSLars Ellenberg drbd_endio_read_sec_final(e); 18945bb912bSLars Ellenberg } 190b411b363SPhilipp Reisner } 191b411b363SPhilipp Reisner 192b411b363SPhilipp Reisner /* read, readA or write requests on R_PRIMARY coming from drbd_make_request 193b411b363SPhilipp Reisner */ 194b411b363SPhilipp Reisner void drbd_endio_pri(struct bio *bio, int error) 195b411b363SPhilipp Reisner { 196a115413dSLars Ellenberg unsigned long flags; 197b411b363SPhilipp Reisner struct drbd_request *req = bio->bi_private; 198b411b363SPhilipp Reisner struct drbd_conf *mdev = req->mdev; 199a115413dSLars Ellenberg struct bio_and_error m; 200b411b363SPhilipp Reisner enum drbd_req_event what; 201b411b363SPhilipp Reisner int uptodate = bio_flagged(bio, BIO_UPTODATE); 202b411b363SPhilipp Reisner 203b411b363SPhilipp Reisner if (!error && !uptodate) { 204b411b363SPhilipp Reisner dev_warn(DEV, "p %s: setting error to -EIO\n", 205b411b363SPhilipp Reisner bio_data_dir(bio) == WRITE ? "write" : "read"); 206b411b363SPhilipp Reisner /* strange behavior of some lower level drivers... 207b411b363SPhilipp Reisner * fail the request by clearing the uptodate flag, 208b411b363SPhilipp Reisner * but do not return any error?! */ 209b411b363SPhilipp Reisner error = -EIO; 210b411b363SPhilipp Reisner } 211b411b363SPhilipp Reisner 212b411b363SPhilipp Reisner /* to avoid recursion in __req_mod */ 213b411b363SPhilipp Reisner if (unlikely(error)) { 214b411b363SPhilipp Reisner what = (bio_data_dir(bio) == WRITE) 215b411b363SPhilipp Reisner ? write_completed_with_error 2165c3c7e64SLars Ellenberg : (bio_rw(bio) == READ) 217b411b363SPhilipp Reisner ? read_completed_with_error 218b411b363SPhilipp Reisner : read_ahead_completed_with_error; 219b411b363SPhilipp Reisner } else 220b411b363SPhilipp Reisner what = completed_ok; 221b411b363SPhilipp Reisner 222b411b363SPhilipp Reisner bio_put(req->private_bio); 223b411b363SPhilipp Reisner req->private_bio = ERR_PTR(error); 224b411b363SPhilipp Reisner 225a115413dSLars Ellenberg /* not req_mod(), we need irqsave here! */ 226a115413dSLars Ellenberg spin_lock_irqsave(&mdev->req_lock, flags); 227a115413dSLars Ellenberg __req_mod(req, what, &m); 228a115413dSLars Ellenberg spin_unlock_irqrestore(&mdev->req_lock, flags); 229a115413dSLars Ellenberg 230a115413dSLars Ellenberg if (m.bio) 231a115413dSLars Ellenberg complete_master_bio(mdev, &m); 232b411b363SPhilipp Reisner } 233b411b363SPhilipp Reisner 234b411b363SPhilipp Reisner int w_read_retry_remote(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 235b411b363SPhilipp Reisner { 236b411b363SPhilipp Reisner struct drbd_request *req = container_of(w, struct drbd_request, w); 237b411b363SPhilipp Reisner 238b411b363SPhilipp Reisner /* We should not detach for read io-error, 239b411b363SPhilipp Reisner * but try to WRITE the P_DATA_REPLY to the failed location, 240b411b363SPhilipp Reisner * to give the disk the chance to relocate that block */ 241b411b363SPhilipp Reisner 242b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 243d255e5ffSLars Ellenberg if (cancel || mdev->state.pdsk != D_UP_TO_DATE) { 244d255e5ffSLars Ellenberg _req_mod(req, read_retry_remote_canceled); 245b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 246b411b363SPhilipp Reisner return 1; 247b411b363SPhilipp Reisner } 248b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 249b411b363SPhilipp Reisner 250b411b363SPhilipp Reisner return w_send_read_req(mdev, w, 0); 251b411b363SPhilipp Reisner } 252b411b363SPhilipp Reisner 253b411b363SPhilipp Reisner int w_resync_inactive(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 254b411b363SPhilipp Reisner { 255b411b363SPhilipp Reisner ERR_IF(cancel) return 1; 256b411b363SPhilipp Reisner dev_err(DEV, "resync inactive, but callback triggered??\n"); 257b411b363SPhilipp Reisner return 1; /* Simply ignore this! */ 258b411b363SPhilipp Reisner } 259b411b363SPhilipp Reisner 26045bb912bSLars Ellenberg void drbd_csum_ee(struct drbd_conf *mdev, struct crypto_hash *tfm, struct drbd_epoch_entry *e, void *digest) 26145bb912bSLars Ellenberg { 26245bb912bSLars Ellenberg struct hash_desc desc; 26345bb912bSLars Ellenberg struct scatterlist sg; 26445bb912bSLars Ellenberg struct page *page = e->pages; 26545bb912bSLars Ellenberg struct page *tmp; 26645bb912bSLars Ellenberg unsigned len; 26745bb912bSLars Ellenberg 26845bb912bSLars Ellenberg desc.tfm = tfm; 26945bb912bSLars Ellenberg desc.flags = 0; 27045bb912bSLars Ellenberg 27145bb912bSLars Ellenberg sg_init_table(&sg, 1); 27245bb912bSLars Ellenberg crypto_hash_init(&desc); 27345bb912bSLars Ellenberg 27445bb912bSLars Ellenberg while ((tmp = page_chain_next(page))) { 27545bb912bSLars Ellenberg /* all but the last page will be fully used */ 27645bb912bSLars Ellenberg sg_set_page(&sg, page, PAGE_SIZE, 0); 27745bb912bSLars Ellenberg crypto_hash_update(&desc, &sg, sg.length); 27845bb912bSLars Ellenberg page = tmp; 27945bb912bSLars Ellenberg } 28045bb912bSLars Ellenberg /* and now the last, possibly only partially used page */ 28145bb912bSLars Ellenberg len = e->size & (PAGE_SIZE - 1); 28245bb912bSLars Ellenberg sg_set_page(&sg, page, len ?: PAGE_SIZE, 0); 28345bb912bSLars Ellenberg crypto_hash_update(&desc, &sg, sg.length); 28445bb912bSLars Ellenberg crypto_hash_final(&desc, digest); 28545bb912bSLars Ellenberg } 28645bb912bSLars Ellenberg 28745bb912bSLars Ellenberg void drbd_csum_bio(struct drbd_conf *mdev, struct crypto_hash *tfm, struct bio *bio, void *digest) 288b411b363SPhilipp Reisner { 289b411b363SPhilipp Reisner struct hash_desc desc; 290b411b363SPhilipp Reisner struct scatterlist sg; 291b411b363SPhilipp Reisner struct bio_vec *bvec; 292b411b363SPhilipp Reisner int i; 293b411b363SPhilipp Reisner 294b411b363SPhilipp Reisner desc.tfm = tfm; 295b411b363SPhilipp Reisner desc.flags = 0; 296b411b363SPhilipp Reisner 297b411b363SPhilipp Reisner sg_init_table(&sg, 1); 298b411b363SPhilipp Reisner crypto_hash_init(&desc); 299b411b363SPhilipp Reisner 300b411b363SPhilipp Reisner __bio_for_each_segment(bvec, bio, i, 0) { 301b411b363SPhilipp Reisner sg_set_page(&sg, bvec->bv_page, bvec->bv_len, bvec->bv_offset); 302b411b363SPhilipp Reisner crypto_hash_update(&desc, &sg, sg.length); 303b411b363SPhilipp Reisner } 304b411b363SPhilipp Reisner crypto_hash_final(&desc, digest); 305b411b363SPhilipp Reisner } 306b411b363SPhilipp Reisner 307b411b363SPhilipp Reisner static int w_e_send_csum(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 308b411b363SPhilipp Reisner { 309b411b363SPhilipp Reisner struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); 310b411b363SPhilipp Reisner int digest_size; 311b411b363SPhilipp Reisner void *digest; 312b411b363SPhilipp Reisner int ok; 313b411b363SPhilipp Reisner 314b411b363SPhilipp Reisner D_ASSERT(e->block_id == DRBD_MAGIC + 0xbeef); 315b411b363SPhilipp Reisner 316b411b363SPhilipp Reisner if (unlikely(cancel)) { 317b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 318b411b363SPhilipp Reisner return 1; 319b411b363SPhilipp Reisner } 320b411b363SPhilipp Reisner 32145bb912bSLars Ellenberg if (likely((e->flags & EE_WAS_ERROR) == 0)) { 322b411b363SPhilipp Reisner digest_size = crypto_hash_digestsize(mdev->csums_tfm); 323b411b363SPhilipp Reisner digest = kmalloc(digest_size, GFP_NOIO); 324b411b363SPhilipp Reisner if (digest) { 32545bb912bSLars Ellenberg drbd_csum_ee(mdev, mdev->csums_tfm, e, digest); 326b411b363SPhilipp Reisner 327b411b363SPhilipp Reisner inc_rs_pending(mdev); 328b411b363SPhilipp Reisner ok = drbd_send_drequest_csum(mdev, 329b411b363SPhilipp Reisner e->sector, 330b411b363SPhilipp Reisner e->size, 331b411b363SPhilipp Reisner digest, 332b411b363SPhilipp Reisner digest_size, 333b411b363SPhilipp Reisner P_CSUM_RS_REQUEST); 334b411b363SPhilipp Reisner kfree(digest); 335b411b363SPhilipp Reisner } else { 336b411b363SPhilipp Reisner dev_err(DEV, "kmalloc() of digest failed.\n"); 337b411b363SPhilipp Reisner ok = 0; 338b411b363SPhilipp Reisner } 339b411b363SPhilipp Reisner } else 340b411b363SPhilipp Reisner ok = 1; 341b411b363SPhilipp Reisner 342b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 343b411b363SPhilipp Reisner 344b411b363SPhilipp Reisner if (unlikely(!ok)) 345b411b363SPhilipp Reisner dev_err(DEV, "drbd_send_drequest(..., csum) failed\n"); 346b411b363SPhilipp Reisner return ok; 347b411b363SPhilipp Reisner } 348b411b363SPhilipp Reisner 349b411b363SPhilipp Reisner #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) 350b411b363SPhilipp Reisner 351b411b363SPhilipp Reisner static int read_for_csum(struct drbd_conf *mdev, sector_t sector, int size) 352b411b363SPhilipp Reisner { 353b411b363SPhilipp Reisner struct drbd_epoch_entry *e; 354b411b363SPhilipp Reisner 355b411b363SPhilipp Reisner if (!get_ldev(mdev)) 35680a40e43SLars Ellenberg return -EIO; 357b411b363SPhilipp Reisner 3580f0601f4SLars Ellenberg if (drbd_rs_should_slow_down(mdev)) 3590f0601f4SLars Ellenberg goto defer; 3600f0601f4SLars Ellenberg 361b411b363SPhilipp Reisner /* GFP_TRY, because if there is no memory available right now, this may 362b411b363SPhilipp Reisner * be rescheduled for later. It is "only" background resync, after all. */ 363b411b363SPhilipp Reisner e = drbd_alloc_ee(mdev, DRBD_MAGIC+0xbeef, sector, size, GFP_TRY); 36445bb912bSLars Ellenberg if (!e) 36580a40e43SLars Ellenberg goto defer; 366b411b363SPhilipp Reisner 36780a40e43SLars Ellenberg e->w.cb = w_e_send_csum; 368b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 369b411b363SPhilipp Reisner list_add(&e->w.list, &mdev->read_ee); 370b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 371b411b363SPhilipp Reisner 3720f0601f4SLars Ellenberg atomic_add(size >> 9, &mdev->rs_sect_ev); 37345bb912bSLars Ellenberg if (drbd_submit_ee(mdev, e, READ, DRBD_FAULT_RS_RD) == 0) 37480a40e43SLars Ellenberg return 0; 37545bb912bSLars Ellenberg 37622cc37a9SLars Ellenberg /* drbd_submit_ee currently fails for one reason only: 37722cc37a9SLars Ellenberg * not being able to allocate enough bios. 37822cc37a9SLars Ellenberg * Is dropping the connection going to help? */ 37922cc37a9SLars Ellenberg spin_lock_irq(&mdev->req_lock); 38022cc37a9SLars Ellenberg list_del(&e->w.list); 38122cc37a9SLars Ellenberg spin_unlock_irq(&mdev->req_lock); 38222cc37a9SLars Ellenberg 38345bb912bSLars Ellenberg drbd_free_ee(mdev, e); 38480a40e43SLars Ellenberg defer: 38545bb912bSLars Ellenberg put_ldev(mdev); 38680a40e43SLars Ellenberg return -EAGAIN; 387b411b363SPhilipp Reisner } 388b411b363SPhilipp Reisner 389b411b363SPhilipp Reisner void resync_timer_fn(unsigned long data) 390b411b363SPhilipp Reisner { 391b411b363SPhilipp Reisner struct drbd_conf *mdev = (struct drbd_conf *) data; 392b411b363SPhilipp Reisner int queue; 393b411b363SPhilipp Reisner 394b411b363SPhilipp Reisner queue = 1; 39563106d3cSPhilipp Reisner switch (mdev->state.conn) { 39663106d3cSPhilipp Reisner case C_VERIFY_S: 397b411b363SPhilipp Reisner mdev->resync_work.cb = w_make_ov_request; 39863106d3cSPhilipp Reisner break; 39963106d3cSPhilipp Reisner case C_SYNC_TARGET: 400b411b363SPhilipp Reisner mdev->resync_work.cb = w_make_resync_request; 40163106d3cSPhilipp Reisner break; 40263106d3cSPhilipp Reisner default: 403b411b363SPhilipp Reisner queue = 0; 404b411b363SPhilipp Reisner mdev->resync_work.cb = w_resync_inactive; 405b411b363SPhilipp Reisner } 406b411b363SPhilipp Reisner 407b411b363SPhilipp Reisner /* harmless race: list_empty outside data.work.q_lock */ 408b411b363SPhilipp Reisner if (list_empty(&mdev->resync_work.list) && queue) 409b411b363SPhilipp Reisner drbd_queue_work(&mdev->data.work, &mdev->resync_work); 410b411b363SPhilipp Reisner } 411b411b363SPhilipp Reisner 412778f271dSPhilipp Reisner static void fifo_set(struct fifo_buffer *fb, int value) 413778f271dSPhilipp Reisner { 414778f271dSPhilipp Reisner int i; 415778f271dSPhilipp Reisner 416778f271dSPhilipp Reisner for (i = 0; i < fb->size; i++) 417f10f2623SPhilipp Reisner fb->values[i] = value; 418778f271dSPhilipp Reisner } 419778f271dSPhilipp Reisner 420778f271dSPhilipp Reisner static int fifo_push(struct fifo_buffer *fb, int value) 421778f271dSPhilipp Reisner { 422778f271dSPhilipp Reisner int ov; 423778f271dSPhilipp Reisner 424778f271dSPhilipp Reisner ov = fb->values[fb->head_index]; 425778f271dSPhilipp Reisner fb->values[fb->head_index++] = value; 426778f271dSPhilipp Reisner 427778f271dSPhilipp Reisner if (fb->head_index >= fb->size) 428778f271dSPhilipp Reisner fb->head_index = 0; 429778f271dSPhilipp Reisner 430778f271dSPhilipp Reisner return ov; 431778f271dSPhilipp Reisner } 432778f271dSPhilipp Reisner 433778f271dSPhilipp Reisner static void fifo_add_val(struct fifo_buffer *fb, int value) 434778f271dSPhilipp Reisner { 435778f271dSPhilipp Reisner int i; 436778f271dSPhilipp Reisner 437778f271dSPhilipp Reisner for (i = 0; i < fb->size; i++) 438778f271dSPhilipp Reisner fb->values[i] += value; 439778f271dSPhilipp Reisner } 440778f271dSPhilipp Reisner 441778f271dSPhilipp Reisner int drbd_rs_controller(struct drbd_conf *mdev) 442778f271dSPhilipp Reisner { 443778f271dSPhilipp Reisner unsigned int sect_in; /* Number of sectors that came in since the last turn */ 444778f271dSPhilipp Reisner unsigned int want; /* The number of sectors we want in the proxy */ 445778f271dSPhilipp Reisner int req_sect; /* Number of sectors to request in this turn */ 446778f271dSPhilipp Reisner int correction; /* Number of sectors more we need in the proxy*/ 447778f271dSPhilipp Reisner int cps; /* correction per invocation of drbd_rs_controller() */ 448778f271dSPhilipp Reisner int steps; /* Number of time steps to plan ahead */ 449778f271dSPhilipp Reisner int curr_corr; 450778f271dSPhilipp Reisner int max_sect; 451778f271dSPhilipp Reisner 452778f271dSPhilipp Reisner sect_in = atomic_xchg(&mdev->rs_sect_in, 0); /* Number of sectors that came in */ 453778f271dSPhilipp Reisner mdev->rs_in_flight -= sect_in; 454778f271dSPhilipp Reisner 455778f271dSPhilipp Reisner spin_lock(&mdev->peer_seq_lock); /* get an atomic view on mdev->rs_plan_s */ 456778f271dSPhilipp Reisner 457778f271dSPhilipp Reisner steps = mdev->rs_plan_s.size; /* (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ; */ 458778f271dSPhilipp Reisner 459778f271dSPhilipp Reisner if (mdev->rs_in_flight + sect_in == 0) { /* At start of resync */ 460778f271dSPhilipp Reisner want = ((mdev->sync_conf.rate * 2 * SLEEP_TIME) / HZ) * steps; 461778f271dSPhilipp Reisner } else { /* normal path */ 462778f271dSPhilipp Reisner want = mdev->sync_conf.c_fill_target ? mdev->sync_conf.c_fill_target : 463778f271dSPhilipp Reisner sect_in * mdev->sync_conf.c_delay_target * HZ / (SLEEP_TIME * 10); 464778f271dSPhilipp Reisner } 465778f271dSPhilipp Reisner 466778f271dSPhilipp Reisner correction = want - mdev->rs_in_flight - mdev->rs_planed; 467778f271dSPhilipp Reisner 468778f271dSPhilipp Reisner /* Plan ahead */ 469778f271dSPhilipp Reisner cps = correction / steps; 470778f271dSPhilipp Reisner fifo_add_val(&mdev->rs_plan_s, cps); 471778f271dSPhilipp Reisner mdev->rs_planed += cps * steps; 472778f271dSPhilipp Reisner 473778f271dSPhilipp Reisner /* What we do in this step */ 474778f271dSPhilipp Reisner curr_corr = fifo_push(&mdev->rs_plan_s, 0); 475778f271dSPhilipp Reisner spin_unlock(&mdev->peer_seq_lock); 476778f271dSPhilipp Reisner mdev->rs_planed -= curr_corr; 477778f271dSPhilipp Reisner 478778f271dSPhilipp Reisner req_sect = sect_in + curr_corr; 479778f271dSPhilipp Reisner if (req_sect < 0) 480778f271dSPhilipp Reisner req_sect = 0; 481778f271dSPhilipp Reisner 482778f271dSPhilipp Reisner max_sect = (mdev->sync_conf.c_max_rate * 2 * SLEEP_TIME) / HZ; 483778f271dSPhilipp Reisner if (req_sect > max_sect) 484778f271dSPhilipp Reisner req_sect = max_sect; 485778f271dSPhilipp Reisner 486778f271dSPhilipp Reisner /* 487778f271dSPhilipp Reisner dev_warn(DEV, "si=%u if=%d wa=%u co=%d st=%d cps=%d pl=%d cc=%d rs=%d\n", 488778f271dSPhilipp Reisner sect_in, mdev->rs_in_flight, want, correction, 489778f271dSPhilipp Reisner steps, cps, mdev->rs_planed, curr_corr, req_sect); 490778f271dSPhilipp Reisner */ 491778f271dSPhilipp Reisner 492778f271dSPhilipp Reisner return req_sect; 493778f271dSPhilipp Reisner } 494778f271dSPhilipp Reisner 495e65f440dSLars Ellenberg int drbd_rs_number_requests(struct drbd_conf *mdev) 496e65f440dSLars Ellenberg { 497e65f440dSLars Ellenberg int number; 498e65f440dSLars Ellenberg if (mdev->rs_plan_s.size) { /* mdev->sync_conf.c_plan_ahead */ 499e65f440dSLars Ellenberg number = drbd_rs_controller(mdev) >> (BM_BLOCK_SHIFT - 9); 500e65f440dSLars Ellenberg mdev->c_sync_rate = number * HZ * (BM_BLOCK_SIZE / 1024) / SLEEP_TIME; 501e65f440dSLars Ellenberg } else { 502e65f440dSLars Ellenberg mdev->c_sync_rate = mdev->sync_conf.rate; 503e65f440dSLars Ellenberg number = SLEEP_TIME * mdev->c_sync_rate / ((BM_BLOCK_SIZE / 1024) * HZ); 504e65f440dSLars Ellenberg } 505e65f440dSLars Ellenberg 506e65f440dSLars Ellenberg /* Throttle resync on lower level disk activity, which may also be 507e65f440dSLars Ellenberg * caused by application IO on Primary/SyncTarget. 508e65f440dSLars Ellenberg * Keep this after the call to drbd_rs_controller, as that assumes 509e65f440dSLars Ellenberg * to be called as precisely as possible every SLEEP_TIME, 510e65f440dSLars Ellenberg * and would be confused otherwise. */ 511e65f440dSLars Ellenberg if (number && drbd_rs_should_slow_down(mdev)) { 512e65f440dSLars Ellenberg mdev->c_sync_rate = 1; 513e65f440dSLars Ellenberg number = 0; 514e65f440dSLars Ellenberg } 515e65f440dSLars Ellenberg 516e65f440dSLars Ellenberg /* ignore the amount of pending requests, the resync controller should 517e65f440dSLars Ellenberg * throttle down to incoming reply rate soon enough anyways. */ 518e65f440dSLars Ellenberg return number; 519e65f440dSLars Ellenberg } 520e65f440dSLars Ellenberg 521b411b363SPhilipp Reisner int w_make_resync_request(struct drbd_conf *mdev, 522b411b363SPhilipp Reisner struct drbd_work *w, int cancel) 523b411b363SPhilipp Reisner { 524b411b363SPhilipp Reisner unsigned long bit; 525b411b363SPhilipp Reisner sector_t sector; 526b411b363SPhilipp Reisner const sector_t capacity = drbd_get_capacity(mdev->this_bdev); 5271816a2b4SLars Ellenberg int max_bio_size; 528e65f440dSLars Ellenberg int number, rollback_i, size; 529b411b363SPhilipp Reisner int align, queued, sndbuf; 5300f0601f4SLars Ellenberg int i = 0; 531b411b363SPhilipp Reisner 532b411b363SPhilipp Reisner if (unlikely(cancel)) 533b411b363SPhilipp Reisner return 1; 534b411b363SPhilipp Reisner 535b411b363SPhilipp Reisner if (unlikely(mdev->state.conn < C_CONNECTED)) { 536b411b363SPhilipp Reisner dev_err(DEV, "Confused in w_make_resync_request()! cstate < Connected"); 537b411b363SPhilipp Reisner return 0; 538b411b363SPhilipp Reisner } 539b411b363SPhilipp Reisner 540b411b363SPhilipp Reisner if (mdev->state.conn != C_SYNC_TARGET) 541b411b363SPhilipp Reisner dev_err(DEV, "%s in w_make_resync_request\n", 542b411b363SPhilipp Reisner drbd_conn_str(mdev->state.conn)); 543b411b363SPhilipp Reisner 544af85e8e8SLars Ellenberg if (mdev->rs_total == 0) { 545af85e8e8SLars Ellenberg /* empty resync? */ 546af85e8e8SLars Ellenberg drbd_resync_finished(mdev); 547af85e8e8SLars Ellenberg return 1; 548af85e8e8SLars Ellenberg } 549af85e8e8SLars Ellenberg 550b411b363SPhilipp Reisner if (!get_ldev(mdev)) { 551b411b363SPhilipp Reisner /* Since we only need to access mdev->rsync a 552b411b363SPhilipp Reisner get_ldev_if_state(mdev,D_FAILED) would be sufficient, but 553b411b363SPhilipp Reisner to continue resync with a broken disk makes no sense at 554b411b363SPhilipp Reisner all */ 555b411b363SPhilipp Reisner dev_err(DEV, "Disk broke down during resync!\n"); 556b411b363SPhilipp Reisner mdev->resync_work.cb = w_resync_inactive; 557b411b363SPhilipp Reisner return 1; 558b411b363SPhilipp Reisner } 559b411b363SPhilipp Reisner 560bb3d000cSLars Ellenberg /* starting with drbd 8.3.8, we can handle multi-bio EEs, 561bb3d000cSLars Ellenberg * if it should be necessary */ 5621816a2b4SLars Ellenberg max_bio_size = 5631816a2b4SLars Ellenberg mdev->agreed_pro_version < 94 ? queue_max_hw_sectors(mdev->rq_queue) << 9 : 5641816a2b4SLars Ellenberg mdev->agreed_pro_version < 95 ? DRBD_MAX_SIZE_H80_PACKET : DRBD_MAX_BIO_SIZE; 565bb3d000cSLars Ellenberg 566e65f440dSLars Ellenberg number = drbd_rs_number_requests(mdev); 567e65f440dSLars Ellenberg if (number == 0) 5680f0601f4SLars Ellenberg goto requeue; 569b411b363SPhilipp Reisner 570b411b363SPhilipp Reisner for (i = 0; i < number; i++) { 571b411b363SPhilipp Reisner /* Stop generating RS requests, when half of the send buffer is filled */ 572b411b363SPhilipp Reisner mutex_lock(&mdev->data.mutex); 573b411b363SPhilipp Reisner if (mdev->data.socket) { 574b411b363SPhilipp Reisner queued = mdev->data.socket->sk->sk_wmem_queued; 575b411b363SPhilipp Reisner sndbuf = mdev->data.socket->sk->sk_sndbuf; 576b411b363SPhilipp Reisner } else { 577b411b363SPhilipp Reisner queued = 1; 578b411b363SPhilipp Reisner sndbuf = 0; 579b411b363SPhilipp Reisner } 580b411b363SPhilipp Reisner mutex_unlock(&mdev->data.mutex); 581b411b363SPhilipp Reisner if (queued > sndbuf / 2) 582b411b363SPhilipp Reisner goto requeue; 583b411b363SPhilipp Reisner 584b411b363SPhilipp Reisner next_sector: 585b411b363SPhilipp Reisner size = BM_BLOCK_SIZE; 586b411b363SPhilipp Reisner bit = drbd_bm_find_next(mdev, mdev->bm_resync_fo); 587b411b363SPhilipp Reisner 588b411b363SPhilipp Reisner if (bit == -1UL) { 589b411b363SPhilipp Reisner mdev->bm_resync_fo = drbd_bm_bits(mdev); 590b411b363SPhilipp Reisner mdev->resync_work.cb = w_resync_inactive; 591b411b363SPhilipp Reisner put_ldev(mdev); 592b411b363SPhilipp Reisner return 1; 593b411b363SPhilipp Reisner } 594b411b363SPhilipp Reisner 595b411b363SPhilipp Reisner sector = BM_BIT_TO_SECT(bit); 596b411b363SPhilipp Reisner 597b411b363SPhilipp Reisner if (drbd_try_rs_begin_io(mdev, sector)) { 598b411b363SPhilipp Reisner mdev->bm_resync_fo = bit; 599b411b363SPhilipp Reisner goto requeue; 600b411b363SPhilipp Reisner } 601b411b363SPhilipp Reisner mdev->bm_resync_fo = bit + 1; 602b411b363SPhilipp Reisner 603b411b363SPhilipp Reisner if (unlikely(drbd_bm_test_bit(mdev, bit) == 0)) { 604b411b363SPhilipp Reisner drbd_rs_complete_io(mdev, sector); 605b411b363SPhilipp Reisner goto next_sector; 606b411b363SPhilipp Reisner } 607b411b363SPhilipp Reisner 6081816a2b4SLars Ellenberg #if DRBD_MAX_BIO_SIZE > BM_BLOCK_SIZE 609b411b363SPhilipp Reisner /* try to find some adjacent bits. 610b411b363SPhilipp Reisner * we stop if we have already the maximum req size. 611b411b363SPhilipp Reisner * 612b411b363SPhilipp Reisner * Additionally always align bigger requests, in order to 613b411b363SPhilipp Reisner * be prepared for all stripe sizes of software RAIDs. 614b411b363SPhilipp Reisner */ 615b411b363SPhilipp Reisner align = 1; 616d207450cSPhilipp Reisner rollback_i = i; 617b411b363SPhilipp Reisner for (;;) { 6181816a2b4SLars Ellenberg if (size + BM_BLOCK_SIZE > max_bio_size) 619b411b363SPhilipp Reisner break; 620b411b363SPhilipp Reisner 621b411b363SPhilipp Reisner /* Be always aligned */ 622b411b363SPhilipp Reisner if (sector & ((1<<(align+3))-1)) 623b411b363SPhilipp Reisner break; 624b411b363SPhilipp Reisner 625b411b363SPhilipp Reisner /* do not cross extent boundaries */ 626b411b363SPhilipp Reisner if (((bit+1) & BM_BLOCKS_PER_BM_EXT_MASK) == 0) 627b411b363SPhilipp Reisner break; 628b411b363SPhilipp Reisner /* now, is it actually dirty, after all? 629b411b363SPhilipp Reisner * caution, drbd_bm_test_bit is tri-state for some 630b411b363SPhilipp Reisner * obscure reason; ( b == 0 ) would get the out-of-band 631b411b363SPhilipp Reisner * only accidentally right because of the "oddly sized" 632b411b363SPhilipp Reisner * adjustment below */ 633b411b363SPhilipp Reisner if (drbd_bm_test_bit(mdev, bit+1) != 1) 634b411b363SPhilipp Reisner break; 635b411b363SPhilipp Reisner bit++; 636b411b363SPhilipp Reisner size += BM_BLOCK_SIZE; 637b411b363SPhilipp Reisner if ((BM_BLOCK_SIZE << align) <= size) 638b411b363SPhilipp Reisner align++; 639b411b363SPhilipp Reisner i++; 640b411b363SPhilipp Reisner } 641b411b363SPhilipp Reisner /* if we merged some, 642b411b363SPhilipp Reisner * reset the offset to start the next drbd_bm_find_next from */ 643b411b363SPhilipp Reisner if (size > BM_BLOCK_SIZE) 644b411b363SPhilipp Reisner mdev->bm_resync_fo = bit + 1; 645b411b363SPhilipp Reisner #endif 646b411b363SPhilipp Reisner 647b411b363SPhilipp Reisner /* adjust very last sectors, in case we are oddly sized */ 648b411b363SPhilipp Reisner if (sector + (size>>9) > capacity) 649b411b363SPhilipp Reisner size = (capacity-sector)<<9; 650b411b363SPhilipp Reisner if (mdev->agreed_pro_version >= 89 && mdev->csums_tfm) { 651b411b363SPhilipp Reisner switch (read_for_csum(mdev, sector, size)) { 65280a40e43SLars Ellenberg case -EIO: /* Disk failure */ 653b411b363SPhilipp Reisner put_ldev(mdev); 654b411b363SPhilipp Reisner return 0; 65580a40e43SLars Ellenberg case -EAGAIN: /* allocation failed, or ldev busy */ 656b411b363SPhilipp Reisner drbd_rs_complete_io(mdev, sector); 657b411b363SPhilipp Reisner mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); 658d207450cSPhilipp Reisner i = rollback_i; 659b411b363SPhilipp Reisner goto requeue; 66080a40e43SLars Ellenberg case 0: 66180a40e43SLars Ellenberg /* everything ok */ 66280a40e43SLars Ellenberg break; 66380a40e43SLars Ellenberg default: 66480a40e43SLars Ellenberg BUG(); 665b411b363SPhilipp Reisner } 666b411b363SPhilipp Reisner } else { 667b411b363SPhilipp Reisner inc_rs_pending(mdev); 668b411b363SPhilipp Reisner if (!drbd_send_drequest(mdev, P_RS_DATA_REQUEST, 669b411b363SPhilipp Reisner sector, size, ID_SYNCER)) { 670b411b363SPhilipp Reisner dev_err(DEV, "drbd_send_drequest() failed, aborting...\n"); 671b411b363SPhilipp Reisner dec_rs_pending(mdev); 672b411b363SPhilipp Reisner put_ldev(mdev); 673b411b363SPhilipp Reisner return 0; 674b411b363SPhilipp Reisner } 675b411b363SPhilipp Reisner } 676b411b363SPhilipp Reisner } 677b411b363SPhilipp Reisner 678b411b363SPhilipp Reisner if (mdev->bm_resync_fo >= drbd_bm_bits(mdev)) { 679b411b363SPhilipp Reisner /* last syncer _request_ was sent, 680b411b363SPhilipp Reisner * but the P_RS_DATA_REPLY not yet received. sync will end (and 681b411b363SPhilipp Reisner * next sync group will resume), as soon as we receive the last 682b411b363SPhilipp Reisner * resync data block, and the last bit is cleared. 683b411b363SPhilipp Reisner * until then resync "work" is "inactive" ... 684b411b363SPhilipp Reisner */ 685b411b363SPhilipp Reisner mdev->resync_work.cb = w_resync_inactive; 686b411b363SPhilipp Reisner put_ldev(mdev); 687b411b363SPhilipp Reisner return 1; 688b411b363SPhilipp Reisner } 689b411b363SPhilipp Reisner 690b411b363SPhilipp Reisner requeue: 691778f271dSPhilipp Reisner mdev->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); 692b411b363SPhilipp Reisner mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME); 693b411b363SPhilipp Reisner put_ldev(mdev); 694b411b363SPhilipp Reisner return 1; 695b411b363SPhilipp Reisner } 696b411b363SPhilipp Reisner 697b411b363SPhilipp Reisner static int w_make_ov_request(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 698b411b363SPhilipp Reisner { 699b411b363SPhilipp Reisner int number, i, size; 700b411b363SPhilipp Reisner sector_t sector; 701b411b363SPhilipp Reisner const sector_t capacity = drbd_get_capacity(mdev->this_bdev); 702b411b363SPhilipp Reisner 703b411b363SPhilipp Reisner if (unlikely(cancel)) 704b411b363SPhilipp Reisner return 1; 705b411b363SPhilipp Reisner 706b411b363SPhilipp Reisner if (unlikely(mdev->state.conn < C_CONNECTED)) { 707b411b363SPhilipp Reisner dev_err(DEV, "Confused in w_make_ov_request()! cstate < Connected"); 708b411b363SPhilipp Reisner return 0; 709b411b363SPhilipp Reisner } 710b411b363SPhilipp Reisner 7112649f080SLars Ellenberg number = drbd_rs_number_requests(mdev); 712b411b363SPhilipp Reisner 713b411b363SPhilipp Reisner sector = mdev->ov_position; 714b411b363SPhilipp Reisner for (i = 0; i < number; i++) { 715b411b363SPhilipp Reisner if (sector >= capacity) { 716b411b363SPhilipp Reisner mdev->resync_work.cb = w_resync_inactive; 717b411b363SPhilipp Reisner return 1; 718b411b363SPhilipp Reisner } 719b411b363SPhilipp Reisner 720b411b363SPhilipp Reisner size = BM_BLOCK_SIZE; 721b411b363SPhilipp Reisner 722b411b363SPhilipp Reisner if (drbd_try_rs_begin_io(mdev, sector)) { 723b411b363SPhilipp Reisner mdev->ov_position = sector; 724b411b363SPhilipp Reisner goto requeue; 725b411b363SPhilipp Reisner } 726b411b363SPhilipp Reisner 727b411b363SPhilipp Reisner if (sector + (size>>9) > capacity) 728b411b363SPhilipp Reisner size = (capacity-sector)<<9; 729b411b363SPhilipp Reisner 730b411b363SPhilipp Reisner inc_rs_pending(mdev); 731b411b363SPhilipp Reisner if (!drbd_send_ov_request(mdev, sector, size)) { 732b411b363SPhilipp Reisner dec_rs_pending(mdev); 733b411b363SPhilipp Reisner return 0; 734b411b363SPhilipp Reisner } 735b411b363SPhilipp Reisner sector += BM_SECT_PER_BIT; 736b411b363SPhilipp Reisner } 737b411b363SPhilipp Reisner mdev->ov_position = sector; 738b411b363SPhilipp Reisner 739b411b363SPhilipp Reisner requeue: 7402649f080SLars Ellenberg mdev->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); 741b411b363SPhilipp Reisner mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME); 742b411b363SPhilipp Reisner return 1; 743b411b363SPhilipp Reisner } 744b411b363SPhilipp Reisner 745c4752ef1SPhilipp Reisner 746c4752ef1SPhilipp Reisner int w_start_resync(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 747c4752ef1SPhilipp Reisner { 748c4752ef1SPhilipp Reisner drbd_start_resync(mdev, C_SYNC_SOURCE); 749c4752ef1SPhilipp Reisner 750c4752ef1SPhilipp Reisner return 1; 751c4752ef1SPhilipp Reisner } 752c4752ef1SPhilipp Reisner 753b411b363SPhilipp Reisner int w_ov_finished(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 754b411b363SPhilipp Reisner { 755b411b363SPhilipp Reisner kfree(w); 756b411b363SPhilipp Reisner ov_oos_print(mdev); 757b411b363SPhilipp Reisner drbd_resync_finished(mdev); 758b411b363SPhilipp Reisner 759b411b363SPhilipp Reisner return 1; 760b411b363SPhilipp Reisner } 761b411b363SPhilipp Reisner 762b411b363SPhilipp Reisner static int w_resync_finished(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 763b411b363SPhilipp Reisner { 764b411b363SPhilipp Reisner kfree(w); 765b411b363SPhilipp Reisner 766b411b363SPhilipp Reisner drbd_resync_finished(mdev); 767b411b363SPhilipp Reisner 768b411b363SPhilipp Reisner return 1; 769b411b363SPhilipp Reisner } 770b411b363SPhilipp Reisner 771af85e8e8SLars Ellenberg static void ping_peer(struct drbd_conf *mdev) 772af85e8e8SLars Ellenberg { 773af85e8e8SLars Ellenberg clear_bit(GOT_PING_ACK, &mdev->flags); 774af85e8e8SLars Ellenberg request_ping(mdev); 775af85e8e8SLars Ellenberg wait_event(mdev->misc_wait, 776af85e8e8SLars Ellenberg test_bit(GOT_PING_ACK, &mdev->flags) || mdev->state.conn < C_CONNECTED); 777af85e8e8SLars Ellenberg } 778af85e8e8SLars Ellenberg 779b411b363SPhilipp Reisner int drbd_resync_finished(struct drbd_conf *mdev) 780b411b363SPhilipp Reisner { 781b411b363SPhilipp Reisner unsigned long db, dt, dbdt; 782b411b363SPhilipp Reisner unsigned long n_oos; 783b411b363SPhilipp Reisner union drbd_state os, ns; 784b411b363SPhilipp Reisner struct drbd_work *w; 785b411b363SPhilipp Reisner char *khelper_cmd = NULL; 78626525618SLars Ellenberg int verify_done = 0; 787b411b363SPhilipp Reisner 788b411b363SPhilipp Reisner /* Remove all elements from the resync LRU. Since future actions 789b411b363SPhilipp Reisner * might set bits in the (main) bitmap, then the entries in the 790b411b363SPhilipp Reisner * resync LRU would be wrong. */ 791b411b363SPhilipp Reisner if (drbd_rs_del_all(mdev)) { 792b411b363SPhilipp Reisner /* In case this is not possible now, most probably because 793b411b363SPhilipp Reisner * there are P_RS_DATA_REPLY Packets lingering on the worker's 794b411b363SPhilipp Reisner * queue (or even the read operations for those packets 795b411b363SPhilipp Reisner * is not finished by now). Retry in 100ms. */ 796b411b363SPhilipp Reisner 797b411b363SPhilipp Reisner __set_current_state(TASK_INTERRUPTIBLE); 798b411b363SPhilipp Reisner schedule_timeout(HZ / 10); 799b411b363SPhilipp Reisner w = kmalloc(sizeof(struct drbd_work), GFP_ATOMIC); 800b411b363SPhilipp Reisner if (w) { 801b411b363SPhilipp Reisner w->cb = w_resync_finished; 802b411b363SPhilipp Reisner drbd_queue_work(&mdev->data.work, w); 803b411b363SPhilipp Reisner return 1; 804b411b363SPhilipp Reisner } 805b411b363SPhilipp Reisner dev_err(DEV, "Warn failed to drbd_rs_del_all() and to kmalloc(w).\n"); 806b411b363SPhilipp Reisner } 807b411b363SPhilipp Reisner 808b411b363SPhilipp Reisner dt = (jiffies - mdev->rs_start - mdev->rs_paused) / HZ; 809b411b363SPhilipp Reisner if (dt <= 0) 810b411b363SPhilipp Reisner dt = 1; 811b411b363SPhilipp Reisner db = mdev->rs_total; 812b411b363SPhilipp Reisner dbdt = Bit2KB(db/dt); 813b411b363SPhilipp Reisner mdev->rs_paused /= HZ; 814b411b363SPhilipp Reisner 815b411b363SPhilipp Reisner if (!get_ldev(mdev)) 816b411b363SPhilipp Reisner goto out; 817b411b363SPhilipp Reisner 818af85e8e8SLars Ellenberg ping_peer(mdev); 819af85e8e8SLars Ellenberg 820b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 821b411b363SPhilipp Reisner os = mdev->state; 822b411b363SPhilipp Reisner 82326525618SLars Ellenberg verify_done = (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T); 82426525618SLars Ellenberg 825b411b363SPhilipp Reisner /* This protects us against multiple calls (that can happen in the presence 826b411b363SPhilipp Reisner of application IO), and against connectivity loss just before we arrive here. */ 827b411b363SPhilipp Reisner if (os.conn <= C_CONNECTED) 828b411b363SPhilipp Reisner goto out_unlock; 829b411b363SPhilipp Reisner 830b411b363SPhilipp Reisner ns = os; 831b411b363SPhilipp Reisner ns.conn = C_CONNECTED; 832b411b363SPhilipp Reisner 833b411b363SPhilipp Reisner dev_info(DEV, "%s done (total %lu sec; paused %lu sec; %lu K/sec)\n", 83426525618SLars Ellenberg verify_done ? "Online verify " : "Resync", 835b411b363SPhilipp Reisner dt + mdev->rs_paused, mdev->rs_paused, dbdt); 836b411b363SPhilipp Reisner 837b411b363SPhilipp Reisner n_oos = drbd_bm_total_weight(mdev); 838b411b363SPhilipp Reisner 839b411b363SPhilipp Reisner if (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) { 840b411b363SPhilipp Reisner if (n_oos) { 841b411b363SPhilipp Reisner dev_alert(DEV, "Online verify found %lu %dk block out of sync!\n", 842b411b363SPhilipp Reisner n_oos, Bit2KB(1)); 843b411b363SPhilipp Reisner khelper_cmd = "out-of-sync"; 844b411b363SPhilipp Reisner } 845b411b363SPhilipp Reisner } else { 846b411b363SPhilipp Reisner D_ASSERT((n_oos - mdev->rs_failed) == 0); 847b411b363SPhilipp Reisner 848b411b363SPhilipp Reisner if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) 849b411b363SPhilipp Reisner khelper_cmd = "after-resync-target"; 850b411b363SPhilipp Reisner 851b411b363SPhilipp Reisner if (mdev->csums_tfm && mdev->rs_total) { 852b411b363SPhilipp Reisner const unsigned long s = mdev->rs_same_csum; 853b411b363SPhilipp Reisner const unsigned long t = mdev->rs_total; 854b411b363SPhilipp Reisner const int ratio = 855b411b363SPhilipp Reisner (t == 0) ? 0 : 856b411b363SPhilipp Reisner (t < 100000) ? ((s*100)/t) : (s/(t/100)); 857b411b363SPhilipp Reisner dev_info(DEV, "%u %% had equal check sums, eliminated: %luK; " 858b411b363SPhilipp Reisner "transferred %luK total %luK\n", 859b411b363SPhilipp Reisner ratio, 860b411b363SPhilipp Reisner Bit2KB(mdev->rs_same_csum), 861b411b363SPhilipp Reisner Bit2KB(mdev->rs_total - mdev->rs_same_csum), 862b411b363SPhilipp Reisner Bit2KB(mdev->rs_total)); 863b411b363SPhilipp Reisner } 864b411b363SPhilipp Reisner } 865b411b363SPhilipp Reisner 866b411b363SPhilipp Reisner if (mdev->rs_failed) { 867b411b363SPhilipp Reisner dev_info(DEV, " %lu failed blocks\n", mdev->rs_failed); 868b411b363SPhilipp Reisner 869b411b363SPhilipp Reisner if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { 870b411b363SPhilipp Reisner ns.disk = D_INCONSISTENT; 871b411b363SPhilipp Reisner ns.pdsk = D_UP_TO_DATE; 872b411b363SPhilipp Reisner } else { 873b411b363SPhilipp Reisner ns.disk = D_UP_TO_DATE; 874b411b363SPhilipp Reisner ns.pdsk = D_INCONSISTENT; 875b411b363SPhilipp Reisner } 876b411b363SPhilipp Reisner } else { 877b411b363SPhilipp Reisner ns.disk = D_UP_TO_DATE; 878b411b363SPhilipp Reisner ns.pdsk = D_UP_TO_DATE; 879b411b363SPhilipp Reisner 880b411b363SPhilipp Reisner if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { 881b411b363SPhilipp Reisner if (mdev->p_uuid) { 882b411b363SPhilipp Reisner int i; 883b411b363SPhilipp Reisner for (i = UI_BITMAP ; i <= UI_HISTORY_END ; i++) 884b411b363SPhilipp Reisner _drbd_uuid_set(mdev, i, mdev->p_uuid[i]); 885b411b363SPhilipp Reisner drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_CURRENT]); 886b411b363SPhilipp Reisner _drbd_uuid_set(mdev, UI_CURRENT, mdev->p_uuid[UI_CURRENT]); 887b411b363SPhilipp Reisner } else { 888b411b363SPhilipp Reisner dev_err(DEV, "mdev->p_uuid is NULL! BUG\n"); 889b411b363SPhilipp Reisner } 890b411b363SPhilipp Reisner } 891b411b363SPhilipp Reisner 892b411b363SPhilipp Reisner drbd_uuid_set_bm(mdev, 0UL); 893b411b363SPhilipp Reisner 894b411b363SPhilipp Reisner if (mdev->p_uuid) { 895b411b363SPhilipp Reisner /* Now the two UUID sets are equal, update what we 896b411b363SPhilipp Reisner * know of the peer. */ 897b411b363SPhilipp Reisner int i; 898b411b363SPhilipp Reisner for (i = UI_CURRENT ; i <= UI_HISTORY_END ; i++) 899b411b363SPhilipp Reisner mdev->p_uuid[i] = mdev->ldev->md.uuid[i]; 900b411b363SPhilipp Reisner } 901b411b363SPhilipp Reisner } 902b411b363SPhilipp Reisner 903b411b363SPhilipp Reisner _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); 904b411b363SPhilipp Reisner out_unlock: 905b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 906b411b363SPhilipp Reisner put_ldev(mdev); 907b411b363SPhilipp Reisner out: 908b411b363SPhilipp Reisner mdev->rs_total = 0; 909b411b363SPhilipp Reisner mdev->rs_failed = 0; 910b411b363SPhilipp Reisner mdev->rs_paused = 0; 91126525618SLars Ellenberg if (verify_done) 912b411b363SPhilipp Reisner mdev->ov_start_sector = 0; 913b411b363SPhilipp Reisner 91413d42685SLars Ellenberg drbd_md_sync(mdev); 91513d42685SLars Ellenberg 916b411b363SPhilipp Reisner if (test_and_clear_bit(WRITE_BM_AFTER_RESYNC, &mdev->flags)) { 917bc571b8cSLars Ellenberg dev_info(DEV, "Writing the whole bitmap\n"); 918b411b363SPhilipp Reisner drbd_queue_bitmap_io(mdev, &drbd_bm_write, NULL, "write from resync_finished"); 919b411b363SPhilipp Reisner } 920b411b363SPhilipp Reisner 921b411b363SPhilipp Reisner if (khelper_cmd) 922b411b363SPhilipp Reisner drbd_khelper(mdev, khelper_cmd); 923b411b363SPhilipp Reisner 924b411b363SPhilipp Reisner return 1; 925b411b363SPhilipp Reisner } 926b411b363SPhilipp Reisner 927b411b363SPhilipp Reisner /* helper */ 928b411b363SPhilipp Reisner static void move_to_net_ee_or_free(struct drbd_conf *mdev, struct drbd_epoch_entry *e) 929b411b363SPhilipp Reisner { 93045bb912bSLars Ellenberg if (drbd_ee_has_active_page(e)) { 931b411b363SPhilipp Reisner /* This might happen if sendpage() has not finished */ 93278db8928SLars Ellenberg int i = (e->size + PAGE_SIZE -1) >> PAGE_SHIFT; 933435f0740SLars Ellenberg atomic_add(i, &mdev->pp_in_use_by_net); 934435f0740SLars Ellenberg atomic_sub(i, &mdev->pp_in_use); 935b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 936b411b363SPhilipp Reisner list_add_tail(&e->w.list, &mdev->net_ee); 937b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 938435f0740SLars Ellenberg wake_up(&drbd_pp_wait); 939b411b363SPhilipp Reisner } else 940b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 941b411b363SPhilipp Reisner } 942b411b363SPhilipp Reisner 943b411b363SPhilipp Reisner /** 944b411b363SPhilipp Reisner * w_e_end_data_req() - Worker callback, to send a P_DATA_REPLY packet in response to a P_DATA_REQUEST 945b411b363SPhilipp Reisner * @mdev: DRBD device. 946b411b363SPhilipp Reisner * @w: work object. 947b411b363SPhilipp Reisner * @cancel: The connection will be closed anyways 948b411b363SPhilipp Reisner */ 949b411b363SPhilipp Reisner int w_e_end_data_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 950b411b363SPhilipp Reisner { 951b411b363SPhilipp Reisner struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); 952b411b363SPhilipp Reisner int ok; 953b411b363SPhilipp Reisner 954b411b363SPhilipp Reisner if (unlikely(cancel)) { 955b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 956b411b363SPhilipp Reisner dec_unacked(mdev); 957b411b363SPhilipp Reisner return 1; 958b411b363SPhilipp Reisner } 959b411b363SPhilipp Reisner 96045bb912bSLars Ellenberg if (likely((e->flags & EE_WAS_ERROR) == 0)) { 961b411b363SPhilipp Reisner ok = drbd_send_block(mdev, P_DATA_REPLY, e); 962b411b363SPhilipp Reisner } else { 963b411b363SPhilipp Reisner if (__ratelimit(&drbd_ratelimit_state)) 964b411b363SPhilipp Reisner dev_err(DEV, "Sending NegDReply. sector=%llus.\n", 965b411b363SPhilipp Reisner (unsigned long long)e->sector); 966b411b363SPhilipp Reisner 967b411b363SPhilipp Reisner ok = drbd_send_ack(mdev, P_NEG_DREPLY, e); 968b411b363SPhilipp Reisner } 969b411b363SPhilipp Reisner 970b411b363SPhilipp Reisner dec_unacked(mdev); 971b411b363SPhilipp Reisner 972b411b363SPhilipp Reisner move_to_net_ee_or_free(mdev, e); 973b411b363SPhilipp Reisner 974b411b363SPhilipp Reisner if (unlikely(!ok)) 975b411b363SPhilipp Reisner dev_err(DEV, "drbd_send_block() failed\n"); 976b411b363SPhilipp Reisner return ok; 977b411b363SPhilipp Reisner } 978b411b363SPhilipp Reisner 979b411b363SPhilipp Reisner /** 980b411b363SPhilipp Reisner * w_e_end_rsdata_req() - Worker callback to send a P_RS_DATA_REPLY packet in response to a P_RS_DATA_REQUESTRS 981b411b363SPhilipp Reisner * @mdev: DRBD device. 982b411b363SPhilipp Reisner * @w: work object. 983b411b363SPhilipp Reisner * @cancel: The connection will be closed anyways 984b411b363SPhilipp Reisner */ 985b411b363SPhilipp Reisner int w_e_end_rsdata_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 986b411b363SPhilipp Reisner { 987b411b363SPhilipp Reisner struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); 988b411b363SPhilipp Reisner int ok; 989b411b363SPhilipp Reisner 990b411b363SPhilipp Reisner if (unlikely(cancel)) { 991b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 992b411b363SPhilipp Reisner dec_unacked(mdev); 993b411b363SPhilipp Reisner return 1; 994b411b363SPhilipp Reisner } 995b411b363SPhilipp Reisner 996b411b363SPhilipp Reisner if (get_ldev_if_state(mdev, D_FAILED)) { 997b411b363SPhilipp Reisner drbd_rs_complete_io(mdev, e->sector); 998b411b363SPhilipp Reisner put_ldev(mdev); 999b411b363SPhilipp Reisner } 1000b411b363SPhilipp Reisner 100145bb912bSLars Ellenberg if (likely((e->flags & EE_WAS_ERROR) == 0)) { 1002b411b363SPhilipp Reisner if (likely(mdev->state.pdsk >= D_INCONSISTENT)) { 1003b411b363SPhilipp Reisner inc_rs_pending(mdev); 1004b411b363SPhilipp Reisner ok = drbd_send_block(mdev, P_RS_DATA_REPLY, e); 1005b411b363SPhilipp Reisner } else { 1006b411b363SPhilipp Reisner if (__ratelimit(&drbd_ratelimit_state)) 1007b411b363SPhilipp Reisner dev_err(DEV, "Not sending RSDataReply, " 1008b411b363SPhilipp Reisner "partner DISKLESS!\n"); 1009b411b363SPhilipp Reisner ok = 1; 1010b411b363SPhilipp Reisner } 1011b411b363SPhilipp Reisner } else { 1012b411b363SPhilipp Reisner if (__ratelimit(&drbd_ratelimit_state)) 1013b411b363SPhilipp Reisner dev_err(DEV, "Sending NegRSDReply. sector %llus.\n", 1014b411b363SPhilipp Reisner (unsigned long long)e->sector); 1015b411b363SPhilipp Reisner 1016b411b363SPhilipp Reisner ok = drbd_send_ack(mdev, P_NEG_RS_DREPLY, e); 1017b411b363SPhilipp Reisner 1018b411b363SPhilipp Reisner /* update resync data with failure */ 1019b411b363SPhilipp Reisner drbd_rs_failed_io(mdev, e->sector, e->size); 1020b411b363SPhilipp Reisner } 1021b411b363SPhilipp Reisner 1022b411b363SPhilipp Reisner dec_unacked(mdev); 1023b411b363SPhilipp Reisner 1024b411b363SPhilipp Reisner move_to_net_ee_or_free(mdev, e); 1025b411b363SPhilipp Reisner 1026b411b363SPhilipp Reisner if (unlikely(!ok)) 1027b411b363SPhilipp Reisner dev_err(DEV, "drbd_send_block() failed\n"); 1028b411b363SPhilipp Reisner return ok; 1029b411b363SPhilipp Reisner } 1030b411b363SPhilipp Reisner 1031b411b363SPhilipp Reisner int w_e_end_csum_rs_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1032b411b363SPhilipp Reisner { 1033b411b363SPhilipp Reisner struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); 1034b411b363SPhilipp Reisner struct digest_info *di; 1035b411b363SPhilipp Reisner int digest_size; 1036b411b363SPhilipp Reisner void *digest = NULL; 1037b411b363SPhilipp Reisner int ok, eq = 0; 1038b411b363SPhilipp Reisner 1039b411b363SPhilipp Reisner if (unlikely(cancel)) { 1040b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 1041b411b363SPhilipp Reisner dec_unacked(mdev); 1042b411b363SPhilipp Reisner return 1; 1043b411b363SPhilipp Reisner } 1044b411b363SPhilipp Reisner 10451d53f09eSLars Ellenberg if (get_ldev(mdev)) { 1046b411b363SPhilipp Reisner drbd_rs_complete_io(mdev, e->sector); 10471d53f09eSLars Ellenberg put_ldev(mdev); 10481d53f09eSLars Ellenberg } 1049b411b363SPhilipp Reisner 105085719573SPhilipp Reisner di = e->digest; 1051b411b363SPhilipp Reisner 105245bb912bSLars Ellenberg if (likely((e->flags & EE_WAS_ERROR) == 0)) { 1053b411b363SPhilipp Reisner /* quick hack to try to avoid a race against reconfiguration. 1054b411b363SPhilipp Reisner * a real fix would be much more involved, 1055b411b363SPhilipp Reisner * introducing more locking mechanisms */ 1056b411b363SPhilipp Reisner if (mdev->csums_tfm) { 1057b411b363SPhilipp Reisner digest_size = crypto_hash_digestsize(mdev->csums_tfm); 1058b411b363SPhilipp Reisner D_ASSERT(digest_size == di->digest_size); 1059b411b363SPhilipp Reisner digest = kmalloc(digest_size, GFP_NOIO); 1060b411b363SPhilipp Reisner } 1061b411b363SPhilipp Reisner if (digest) { 106245bb912bSLars Ellenberg drbd_csum_ee(mdev, mdev->csums_tfm, e, digest); 1063b411b363SPhilipp Reisner eq = !memcmp(digest, di->digest, digest_size); 1064b411b363SPhilipp Reisner kfree(digest); 1065b411b363SPhilipp Reisner } 1066b411b363SPhilipp Reisner 1067b411b363SPhilipp Reisner if (eq) { 1068b411b363SPhilipp Reisner drbd_set_in_sync(mdev, e->sector, e->size); 1069676396d5SLars Ellenberg /* rs_same_csums unit is BM_BLOCK_SIZE */ 1070676396d5SLars Ellenberg mdev->rs_same_csum += e->size >> BM_BLOCK_SHIFT; 1071b411b363SPhilipp Reisner ok = drbd_send_ack(mdev, P_RS_IS_IN_SYNC, e); 1072b411b363SPhilipp Reisner } else { 1073b411b363SPhilipp Reisner inc_rs_pending(mdev); 1074204bba99SPhilipp Reisner e->block_id = ID_SYNCER; /* By setting block_id, digest pointer becomes invalid! */ 1075204bba99SPhilipp Reisner e->flags &= ~EE_HAS_DIGEST; /* This e no longer has a digest pointer */ 1076204bba99SPhilipp Reisner kfree(di); 1077b411b363SPhilipp Reisner ok = drbd_send_block(mdev, P_RS_DATA_REPLY, e); 1078b411b363SPhilipp Reisner } 1079b411b363SPhilipp Reisner } else { 1080b411b363SPhilipp Reisner ok = drbd_send_ack(mdev, P_NEG_RS_DREPLY, e); 1081b411b363SPhilipp Reisner if (__ratelimit(&drbd_ratelimit_state)) 1082b411b363SPhilipp Reisner dev_err(DEV, "Sending NegDReply. I guess it gets messy.\n"); 1083b411b363SPhilipp Reisner } 1084b411b363SPhilipp Reisner 1085b411b363SPhilipp Reisner dec_unacked(mdev); 1086b411b363SPhilipp Reisner move_to_net_ee_or_free(mdev, e); 1087b411b363SPhilipp Reisner 1088b411b363SPhilipp Reisner if (unlikely(!ok)) 1089b411b363SPhilipp Reisner dev_err(DEV, "drbd_send_block/ack() failed\n"); 1090b411b363SPhilipp Reisner return ok; 1091b411b363SPhilipp Reisner } 1092b411b363SPhilipp Reisner 1093b411b363SPhilipp Reisner int w_e_end_ov_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1094b411b363SPhilipp Reisner { 1095b411b363SPhilipp Reisner struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); 1096b411b363SPhilipp Reisner int digest_size; 1097b411b363SPhilipp Reisner void *digest; 1098b411b363SPhilipp Reisner int ok = 1; 1099b411b363SPhilipp Reisner 1100b411b363SPhilipp Reisner if (unlikely(cancel)) 1101b411b363SPhilipp Reisner goto out; 1102b411b363SPhilipp Reisner 110345bb912bSLars Ellenberg if (unlikely((e->flags & EE_WAS_ERROR) != 0)) 1104b411b363SPhilipp Reisner goto out; 1105b411b363SPhilipp Reisner 1106b411b363SPhilipp Reisner digest_size = crypto_hash_digestsize(mdev->verify_tfm); 1107b411b363SPhilipp Reisner /* FIXME if this allocation fails, online verify will not terminate! */ 1108b411b363SPhilipp Reisner digest = kmalloc(digest_size, GFP_NOIO); 1109b411b363SPhilipp Reisner if (digest) { 111045bb912bSLars Ellenberg drbd_csum_ee(mdev, mdev->verify_tfm, e, digest); 1111b411b363SPhilipp Reisner inc_rs_pending(mdev); 1112b411b363SPhilipp Reisner ok = drbd_send_drequest_csum(mdev, e->sector, e->size, 1113b411b363SPhilipp Reisner digest, digest_size, P_OV_REPLY); 1114b411b363SPhilipp Reisner if (!ok) 1115b411b363SPhilipp Reisner dec_rs_pending(mdev); 1116b411b363SPhilipp Reisner kfree(digest); 1117b411b363SPhilipp Reisner } 1118b411b363SPhilipp Reisner 1119b411b363SPhilipp Reisner out: 1120b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 1121b411b363SPhilipp Reisner 1122b411b363SPhilipp Reisner dec_unacked(mdev); 1123b411b363SPhilipp Reisner 1124b411b363SPhilipp Reisner return ok; 1125b411b363SPhilipp Reisner } 1126b411b363SPhilipp Reisner 1127b411b363SPhilipp Reisner void drbd_ov_oos_found(struct drbd_conf *mdev, sector_t sector, int size) 1128b411b363SPhilipp Reisner { 1129b411b363SPhilipp Reisner if (mdev->ov_last_oos_start + mdev->ov_last_oos_size == sector) { 1130b411b363SPhilipp Reisner mdev->ov_last_oos_size += size>>9; 1131b411b363SPhilipp Reisner } else { 1132b411b363SPhilipp Reisner mdev->ov_last_oos_start = sector; 1133b411b363SPhilipp Reisner mdev->ov_last_oos_size = size>>9; 1134b411b363SPhilipp Reisner } 1135b411b363SPhilipp Reisner drbd_set_out_of_sync(mdev, sector, size); 1136b411b363SPhilipp Reisner set_bit(WRITE_BM_AFTER_RESYNC, &mdev->flags); 1137b411b363SPhilipp Reisner } 1138b411b363SPhilipp Reisner 1139b411b363SPhilipp Reisner int w_e_end_ov_reply(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1140b411b363SPhilipp Reisner { 1141b411b363SPhilipp Reisner struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); 1142b411b363SPhilipp Reisner struct digest_info *di; 1143b411b363SPhilipp Reisner int digest_size; 1144b411b363SPhilipp Reisner void *digest; 1145b411b363SPhilipp Reisner int ok, eq = 0; 1146b411b363SPhilipp Reisner 1147b411b363SPhilipp Reisner if (unlikely(cancel)) { 1148b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 1149b411b363SPhilipp Reisner dec_unacked(mdev); 1150b411b363SPhilipp Reisner return 1; 1151b411b363SPhilipp Reisner } 1152b411b363SPhilipp Reisner 1153b411b363SPhilipp Reisner /* after "cancel", because after drbd_disconnect/drbd_rs_cancel_all 1154b411b363SPhilipp Reisner * the resync lru has been cleaned up already */ 11551d53f09eSLars Ellenberg if (get_ldev(mdev)) { 1156b411b363SPhilipp Reisner drbd_rs_complete_io(mdev, e->sector); 11571d53f09eSLars Ellenberg put_ldev(mdev); 11581d53f09eSLars Ellenberg } 1159b411b363SPhilipp Reisner 116085719573SPhilipp Reisner di = e->digest; 1161b411b363SPhilipp Reisner 116245bb912bSLars Ellenberg if (likely((e->flags & EE_WAS_ERROR) == 0)) { 1163b411b363SPhilipp Reisner digest_size = crypto_hash_digestsize(mdev->verify_tfm); 1164b411b363SPhilipp Reisner digest = kmalloc(digest_size, GFP_NOIO); 1165b411b363SPhilipp Reisner if (digest) { 116645bb912bSLars Ellenberg drbd_csum_ee(mdev, mdev->verify_tfm, e, digest); 1167b411b363SPhilipp Reisner 1168b411b363SPhilipp Reisner D_ASSERT(digest_size == di->digest_size); 1169b411b363SPhilipp Reisner eq = !memcmp(digest, di->digest, digest_size); 1170b411b363SPhilipp Reisner kfree(digest); 1171b411b363SPhilipp Reisner } 1172b411b363SPhilipp Reisner } else { 1173b411b363SPhilipp Reisner ok = drbd_send_ack(mdev, P_NEG_RS_DREPLY, e); 1174b411b363SPhilipp Reisner if (__ratelimit(&drbd_ratelimit_state)) 1175b411b363SPhilipp Reisner dev_err(DEV, "Sending NegDReply. I guess it gets messy.\n"); 1176b411b363SPhilipp Reisner } 1177b411b363SPhilipp Reisner 1178b411b363SPhilipp Reisner dec_unacked(mdev); 1179b411b363SPhilipp Reisner if (!eq) 1180b411b363SPhilipp Reisner drbd_ov_oos_found(mdev, e->sector, e->size); 1181b411b363SPhilipp Reisner else 1182b411b363SPhilipp Reisner ov_oos_print(mdev); 1183b411b363SPhilipp Reisner 1184b411b363SPhilipp Reisner ok = drbd_send_ack_ex(mdev, P_OV_RESULT, e->sector, e->size, 1185b411b363SPhilipp Reisner eq ? ID_IN_SYNC : ID_OUT_OF_SYNC); 1186b411b363SPhilipp Reisner 1187b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 1188b411b363SPhilipp Reisner 1189ea5442afSLars Ellenberg --mdev->ov_left; 1190ea5442afSLars Ellenberg 1191ea5442afSLars Ellenberg /* let's advance progress step marks only for every other megabyte */ 1192ea5442afSLars Ellenberg if ((mdev->ov_left & 0x200) == 0x200) 1193ea5442afSLars Ellenberg drbd_advance_rs_marks(mdev, mdev->ov_left); 1194ea5442afSLars Ellenberg 1195ea5442afSLars Ellenberg if (mdev->ov_left == 0) { 1196b411b363SPhilipp Reisner ov_oos_print(mdev); 1197b411b363SPhilipp Reisner drbd_resync_finished(mdev); 1198b411b363SPhilipp Reisner } 1199b411b363SPhilipp Reisner 1200b411b363SPhilipp Reisner return ok; 1201b411b363SPhilipp Reisner } 1202b411b363SPhilipp Reisner 1203b411b363SPhilipp Reisner int w_prev_work_done(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1204b411b363SPhilipp Reisner { 1205b411b363SPhilipp Reisner struct drbd_wq_barrier *b = container_of(w, struct drbd_wq_barrier, w); 1206b411b363SPhilipp Reisner complete(&b->done); 1207b411b363SPhilipp Reisner return 1; 1208b411b363SPhilipp Reisner } 1209b411b363SPhilipp Reisner 1210b411b363SPhilipp Reisner int w_send_barrier(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1211b411b363SPhilipp Reisner { 1212b411b363SPhilipp Reisner struct drbd_tl_epoch *b = container_of(w, struct drbd_tl_epoch, w); 1213b411b363SPhilipp Reisner struct p_barrier *p = &mdev->data.sbuf.barrier; 1214b411b363SPhilipp Reisner int ok = 1; 1215b411b363SPhilipp Reisner 1216b411b363SPhilipp Reisner /* really avoid racing with tl_clear. w.cb may have been referenced 1217b411b363SPhilipp Reisner * just before it was reassigned and re-queued, so double check that. 1218b411b363SPhilipp Reisner * actually, this race was harmless, since we only try to send the 1219b411b363SPhilipp Reisner * barrier packet here, and otherwise do nothing with the object. 1220b411b363SPhilipp Reisner * but compare with the head of w_clear_epoch */ 1221b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 1222b411b363SPhilipp Reisner if (w->cb != w_send_barrier || mdev->state.conn < C_CONNECTED) 1223b411b363SPhilipp Reisner cancel = 1; 1224b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1225b411b363SPhilipp Reisner if (cancel) 1226b411b363SPhilipp Reisner return 1; 1227b411b363SPhilipp Reisner 1228b411b363SPhilipp Reisner if (!drbd_get_data_sock(mdev)) 1229b411b363SPhilipp Reisner return 0; 1230b411b363SPhilipp Reisner p->barrier = b->br_number; 1231b411b363SPhilipp Reisner /* inc_ap_pending was done where this was queued. 1232b411b363SPhilipp Reisner * dec_ap_pending will be done in got_BarrierAck 1233b411b363SPhilipp Reisner * or (on connection loss) in w_clear_epoch. */ 1234b411b363SPhilipp Reisner ok = _drbd_send_cmd(mdev, mdev->data.socket, P_BARRIER, 12350b70a13dSPhilipp Reisner (struct p_header80 *)p, sizeof(*p), 0); 1236b411b363SPhilipp Reisner drbd_put_data_sock(mdev); 1237b411b363SPhilipp Reisner 1238b411b363SPhilipp Reisner return ok; 1239b411b363SPhilipp Reisner } 1240b411b363SPhilipp Reisner 1241b411b363SPhilipp Reisner int w_send_write_hint(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1242b411b363SPhilipp Reisner { 1243b411b363SPhilipp Reisner if (cancel) 1244b411b363SPhilipp Reisner return 1; 1245b411b363SPhilipp Reisner return drbd_send_short_cmd(mdev, P_UNPLUG_REMOTE); 1246b411b363SPhilipp Reisner } 1247b411b363SPhilipp Reisner 124873a01a18SPhilipp Reisner int w_send_oos(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 124973a01a18SPhilipp Reisner { 125073a01a18SPhilipp Reisner struct drbd_request *req = container_of(w, struct drbd_request, w); 125173a01a18SPhilipp Reisner int ok; 125273a01a18SPhilipp Reisner 125373a01a18SPhilipp Reisner if (unlikely(cancel)) { 125473a01a18SPhilipp Reisner req_mod(req, send_canceled); 125573a01a18SPhilipp Reisner return 1; 125673a01a18SPhilipp Reisner } 125773a01a18SPhilipp Reisner 125873a01a18SPhilipp Reisner ok = drbd_send_oos(mdev, req); 125973a01a18SPhilipp Reisner req_mod(req, oos_handed_to_network); 126073a01a18SPhilipp Reisner 126173a01a18SPhilipp Reisner return ok; 126273a01a18SPhilipp Reisner } 126373a01a18SPhilipp Reisner 1264b411b363SPhilipp Reisner /** 1265b411b363SPhilipp Reisner * w_send_dblock() - Worker callback to send a P_DATA packet in order to mirror a write request 1266b411b363SPhilipp Reisner * @mdev: DRBD device. 1267b411b363SPhilipp Reisner * @w: work object. 1268b411b363SPhilipp Reisner * @cancel: The connection will be closed anyways 1269b411b363SPhilipp Reisner */ 1270b411b363SPhilipp Reisner int w_send_dblock(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1271b411b363SPhilipp Reisner { 1272b411b363SPhilipp Reisner struct drbd_request *req = container_of(w, struct drbd_request, w); 1273b411b363SPhilipp Reisner int ok; 1274b411b363SPhilipp Reisner 1275b411b363SPhilipp Reisner if (unlikely(cancel)) { 1276b411b363SPhilipp Reisner req_mod(req, send_canceled); 1277b411b363SPhilipp Reisner return 1; 1278b411b363SPhilipp Reisner } 1279b411b363SPhilipp Reisner 1280b411b363SPhilipp Reisner ok = drbd_send_dblock(mdev, req); 1281b411b363SPhilipp Reisner req_mod(req, ok ? handed_over_to_network : send_failed); 1282b411b363SPhilipp Reisner 1283b411b363SPhilipp Reisner return ok; 1284b411b363SPhilipp Reisner } 1285b411b363SPhilipp Reisner 1286b411b363SPhilipp Reisner /** 1287b411b363SPhilipp Reisner * w_send_read_req() - Worker callback to send a read request (P_DATA_REQUEST) packet 1288b411b363SPhilipp Reisner * @mdev: DRBD device. 1289b411b363SPhilipp Reisner * @w: work object. 1290b411b363SPhilipp Reisner * @cancel: The connection will be closed anyways 1291b411b363SPhilipp Reisner */ 1292b411b363SPhilipp Reisner int w_send_read_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1293b411b363SPhilipp Reisner { 1294b411b363SPhilipp Reisner struct drbd_request *req = container_of(w, struct drbd_request, w); 1295b411b363SPhilipp Reisner int ok; 1296b411b363SPhilipp Reisner 1297b411b363SPhilipp Reisner if (unlikely(cancel)) { 1298b411b363SPhilipp Reisner req_mod(req, send_canceled); 1299b411b363SPhilipp Reisner return 1; 1300b411b363SPhilipp Reisner } 1301b411b363SPhilipp Reisner 1302b411b363SPhilipp Reisner ok = drbd_send_drequest(mdev, P_DATA_REQUEST, req->sector, req->size, 1303b411b363SPhilipp Reisner (unsigned long)req); 1304b411b363SPhilipp Reisner 1305b411b363SPhilipp Reisner if (!ok) { 1306b411b363SPhilipp Reisner /* ?? we set C_TIMEOUT or C_BROKEN_PIPE in drbd_send(); 1307b411b363SPhilipp Reisner * so this is probably redundant */ 1308b411b363SPhilipp Reisner if (mdev->state.conn >= C_CONNECTED) 1309b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE)); 1310b411b363SPhilipp Reisner } 1311b411b363SPhilipp Reisner req_mod(req, ok ? handed_over_to_network : send_failed); 1312b411b363SPhilipp Reisner 1313b411b363SPhilipp Reisner return ok; 1314b411b363SPhilipp Reisner } 1315b411b363SPhilipp Reisner 1316265be2d0SPhilipp Reisner int w_restart_disk_io(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1317265be2d0SPhilipp Reisner { 1318265be2d0SPhilipp Reisner struct drbd_request *req = container_of(w, struct drbd_request, w); 1319265be2d0SPhilipp Reisner 13200778286aSPhilipp Reisner if (bio_data_dir(req->master_bio) == WRITE && req->rq_state & RQ_IN_ACT_LOG) 1321265be2d0SPhilipp Reisner drbd_al_begin_io(mdev, req->sector); 1322265be2d0SPhilipp Reisner /* Calling drbd_al_begin_io() out of the worker might deadlocks 1323265be2d0SPhilipp Reisner theoretically. Practically it can not deadlock, since this is 1324265be2d0SPhilipp Reisner only used when unfreezing IOs. All the extents of the requests 1325265be2d0SPhilipp Reisner that made it into the TL are already active */ 1326265be2d0SPhilipp Reisner 1327265be2d0SPhilipp Reisner drbd_req_make_private_bio(req, req->master_bio); 1328265be2d0SPhilipp Reisner req->private_bio->bi_bdev = mdev->ldev->backing_bdev; 1329265be2d0SPhilipp Reisner generic_make_request(req->private_bio); 1330265be2d0SPhilipp Reisner 1331265be2d0SPhilipp Reisner return 1; 1332265be2d0SPhilipp Reisner } 1333265be2d0SPhilipp Reisner 1334b411b363SPhilipp Reisner static int _drbd_may_sync_now(struct drbd_conf *mdev) 1335b411b363SPhilipp Reisner { 1336b411b363SPhilipp Reisner struct drbd_conf *odev = mdev; 1337b411b363SPhilipp Reisner 1338b411b363SPhilipp Reisner while (1) { 1339b411b363SPhilipp Reisner if (odev->sync_conf.after == -1) 1340b411b363SPhilipp Reisner return 1; 1341b411b363SPhilipp Reisner odev = minor_to_mdev(odev->sync_conf.after); 1342b411b363SPhilipp Reisner ERR_IF(!odev) return 1; 1343b411b363SPhilipp Reisner if ((odev->state.conn >= C_SYNC_SOURCE && 1344b411b363SPhilipp Reisner odev->state.conn <= C_PAUSED_SYNC_T) || 1345b411b363SPhilipp Reisner odev->state.aftr_isp || odev->state.peer_isp || 1346b411b363SPhilipp Reisner odev->state.user_isp) 1347b411b363SPhilipp Reisner return 0; 1348b411b363SPhilipp Reisner } 1349b411b363SPhilipp Reisner } 1350b411b363SPhilipp Reisner 1351b411b363SPhilipp Reisner /** 1352b411b363SPhilipp Reisner * _drbd_pause_after() - Pause resync on all devices that may not resync now 1353b411b363SPhilipp Reisner * @mdev: DRBD device. 1354b411b363SPhilipp Reisner * 1355b411b363SPhilipp Reisner * Called from process context only (admin command and after_state_ch). 1356b411b363SPhilipp Reisner */ 1357b411b363SPhilipp Reisner static int _drbd_pause_after(struct drbd_conf *mdev) 1358b411b363SPhilipp Reisner { 1359b411b363SPhilipp Reisner struct drbd_conf *odev; 1360b411b363SPhilipp Reisner int i, rv = 0; 1361b411b363SPhilipp Reisner 1362b411b363SPhilipp Reisner for (i = 0; i < minor_count; i++) { 1363b411b363SPhilipp Reisner odev = minor_to_mdev(i); 1364b411b363SPhilipp Reisner if (!odev) 1365b411b363SPhilipp Reisner continue; 1366b411b363SPhilipp Reisner if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) 1367b411b363SPhilipp Reisner continue; 1368b411b363SPhilipp Reisner if (!_drbd_may_sync_now(odev)) 1369b411b363SPhilipp Reisner rv |= (__drbd_set_state(_NS(odev, aftr_isp, 1), CS_HARD, NULL) 1370b411b363SPhilipp Reisner != SS_NOTHING_TO_DO); 1371b411b363SPhilipp Reisner } 1372b411b363SPhilipp Reisner 1373b411b363SPhilipp Reisner return rv; 1374b411b363SPhilipp Reisner } 1375b411b363SPhilipp Reisner 1376b411b363SPhilipp Reisner /** 1377b411b363SPhilipp Reisner * _drbd_resume_next() - Resume resync on all devices that may resync now 1378b411b363SPhilipp Reisner * @mdev: DRBD device. 1379b411b363SPhilipp Reisner * 1380b411b363SPhilipp Reisner * Called from process context only (admin command and worker). 1381b411b363SPhilipp Reisner */ 1382b411b363SPhilipp Reisner static int _drbd_resume_next(struct drbd_conf *mdev) 1383b411b363SPhilipp Reisner { 1384b411b363SPhilipp Reisner struct drbd_conf *odev; 1385b411b363SPhilipp Reisner int i, rv = 0; 1386b411b363SPhilipp Reisner 1387b411b363SPhilipp Reisner for (i = 0; i < minor_count; i++) { 1388b411b363SPhilipp Reisner odev = minor_to_mdev(i); 1389b411b363SPhilipp Reisner if (!odev) 1390b411b363SPhilipp Reisner continue; 1391b411b363SPhilipp Reisner if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) 1392b411b363SPhilipp Reisner continue; 1393b411b363SPhilipp Reisner if (odev->state.aftr_isp) { 1394b411b363SPhilipp Reisner if (_drbd_may_sync_now(odev)) 1395b411b363SPhilipp Reisner rv |= (__drbd_set_state(_NS(odev, aftr_isp, 0), 1396b411b363SPhilipp Reisner CS_HARD, NULL) 1397b411b363SPhilipp Reisner != SS_NOTHING_TO_DO) ; 1398b411b363SPhilipp Reisner } 1399b411b363SPhilipp Reisner } 1400b411b363SPhilipp Reisner return rv; 1401b411b363SPhilipp Reisner } 1402b411b363SPhilipp Reisner 1403b411b363SPhilipp Reisner void resume_next_sg(struct drbd_conf *mdev) 1404b411b363SPhilipp Reisner { 1405b411b363SPhilipp Reisner write_lock_irq(&global_state_lock); 1406b411b363SPhilipp Reisner _drbd_resume_next(mdev); 1407b411b363SPhilipp Reisner write_unlock_irq(&global_state_lock); 1408b411b363SPhilipp Reisner } 1409b411b363SPhilipp Reisner 1410b411b363SPhilipp Reisner void suspend_other_sg(struct drbd_conf *mdev) 1411b411b363SPhilipp Reisner { 1412b411b363SPhilipp Reisner write_lock_irq(&global_state_lock); 1413b411b363SPhilipp Reisner _drbd_pause_after(mdev); 1414b411b363SPhilipp Reisner write_unlock_irq(&global_state_lock); 1415b411b363SPhilipp Reisner } 1416b411b363SPhilipp Reisner 1417b411b363SPhilipp Reisner static int sync_after_error(struct drbd_conf *mdev, int o_minor) 1418b411b363SPhilipp Reisner { 1419b411b363SPhilipp Reisner struct drbd_conf *odev; 1420b411b363SPhilipp Reisner 1421b411b363SPhilipp Reisner if (o_minor == -1) 1422b411b363SPhilipp Reisner return NO_ERROR; 1423b411b363SPhilipp Reisner if (o_minor < -1 || minor_to_mdev(o_minor) == NULL) 1424b411b363SPhilipp Reisner return ERR_SYNC_AFTER; 1425b411b363SPhilipp Reisner 1426b411b363SPhilipp Reisner /* check for loops */ 1427b411b363SPhilipp Reisner odev = minor_to_mdev(o_minor); 1428b411b363SPhilipp Reisner while (1) { 1429b411b363SPhilipp Reisner if (odev == mdev) 1430b411b363SPhilipp Reisner return ERR_SYNC_AFTER_CYCLE; 1431b411b363SPhilipp Reisner 1432b411b363SPhilipp Reisner /* dependency chain ends here, no cycles. */ 1433b411b363SPhilipp Reisner if (odev->sync_conf.after == -1) 1434b411b363SPhilipp Reisner return NO_ERROR; 1435b411b363SPhilipp Reisner 1436b411b363SPhilipp Reisner /* follow the dependency chain */ 1437b411b363SPhilipp Reisner odev = minor_to_mdev(odev->sync_conf.after); 1438b411b363SPhilipp Reisner } 1439b411b363SPhilipp Reisner } 1440b411b363SPhilipp Reisner 1441b411b363SPhilipp Reisner int drbd_alter_sa(struct drbd_conf *mdev, int na) 1442b411b363SPhilipp Reisner { 1443b411b363SPhilipp Reisner int changes; 1444b411b363SPhilipp Reisner int retcode; 1445b411b363SPhilipp Reisner 1446b411b363SPhilipp Reisner write_lock_irq(&global_state_lock); 1447b411b363SPhilipp Reisner retcode = sync_after_error(mdev, na); 1448b411b363SPhilipp Reisner if (retcode == NO_ERROR) { 1449b411b363SPhilipp Reisner mdev->sync_conf.after = na; 1450b411b363SPhilipp Reisner do { 1451b411b363SPhilipp Reisner changes = _drbd_pause_after(mdev); 1452b411b363SPhilipp Reisner changes |= _drbd_resume_next(mdev); 1453b411b363SPhilipp Reisner } while (changes); 1454b411b363SPhilipp Reisner } 1455b411b363SPhilipp Reisner write_unlock_irq(&global_state_lock); 1456b411b363SPhilipp Reisner return retcode; 1457b411b363SPhilipp Reisner } 1458b411b363SPhilipp Reisner 14599bd28d3cSLars Ellenberg void drbd_rs_controller_reset(struct drbd_conf *mdev) 14609bd28d3cSLars Ellenberg { 14619bd28d3cSLars Ellenberg atomic_set(&mdev->rs_sect_in, 0); 14629bd28d3cSLars Ellenberg atomic_set(&mdev->rs_sect_ev, 0); 14639bd28d3cSLars Ellenberg mdev->rs_in_flight = 0; 14649bd28d3cSLars Ellenberg mdev->rs_planed = 0; 14659bd28d3cSLars Ellenberg spin_lock(&mdev->peer_seq_lock); 14669bd28d3cSLars Ellenberg fifo_set(&mdev->rs_plan_s, 0); 14679bd28d3cSLars Ellenberg spin_unlock(&mdev->peer_seq_lock); 14689bd28d3cSLars Ellenberg } 14699bd28d3cSLars Ellenberg 1470b411b363SPhilipp Reisner /** 1471b411b363SPhilipp Reisner * drbd_start_resync() - Start the resync process 1472b411b363SPhilipp Reisner * @mdev: DRBD device. 1473b411b363SPhilipp Reisner * @side: Either C_SYNC_SOURCE or C_SYNC_TARGET 1474b411b363SPhilipp Reisner * 1475b411b363SPhilipp Reisner * This function might bring you directly into one of the 1476b411b363SPhilipp Reisner * C_PAUSED_SYNC_* states. 1477b411b363SPhilipp Reisner */ 1478b411b363SPhilipp Reisner void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side) 1479b411b363SPhilipp Reisner { 1480b411b363SPhilipp Reisner union drbd_state ns; 1481b411b363SPhilipp Reisner int r; 1482b411b363SPhilipp Reisner 1483c4752ef1SPhilipp Reisner if (mdev->state.conn >= C_SYNC_SOURCE && mdev->state.conn < C_AHEAD) { 1484b411b363SPhilipp Reisner dev_err(DEV, "Resync already running!\n"); 1485b411b363SPhilipp Reisner return; 1486b411b363SPhilipp Reisner } 1487b411b363SPhilipp Reisner 148859817f4fSPhilipp Reisner if (mdev->state.conn < C_AHEAD) { 1489b411b363SPhilipp Reisner /* In case a previous resync run was aborted by an IO error/detach on the peer. */ 1490b411b363SPhilipp Reisner drbd_rs_cancel_all(mdev); 149159817f4fSPhilipp Reisner /* This should be done when we abort the resync. We definitely do not 149259817f4fSPhilipp Reisner want to have this for connections going back and forth between 149359817f4fSPhilipp Reisner Ahead/Behind and SyncSource/SyncTarget */ 149459817f4fSPhilipp Reisner } 1495b411b363SPhilipp Reisner 1496b411b363SPhilipp Reisner if (side == C_SYNC_TARGET) { 1497b411b363SPhilipp Reisner /* Since application IO was locked out during C_WF_BITMAP_T and 1498b411b363SPhilipp Reisner C_WF_SYNC_UUID we are still unmodified. Before going to C_SYNC_TARGET 1499b411b363SPhilipp Reisner we check that we might make the data inconsistent. */ 1500b411b363SPhilipp Reisner r = drbd_khelper(mdev, "before-resync-target"); 1501b411b363SPhilipp Reisner r = (r >> 8) & 0xff; 1502b411b363SPhilipp Reisner if (r > 0) { 1503b411b363SPhilipp Reisner dev_info(DEV, "before-resync-target handler returned %d, " 1504b411b363SPhilipp Reisner "dropping connection.\n", r); 1505b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 1506b411b363SPhilipp Reisner return; 1507b411b363SPhilipp Reisner } 1508b411b363SPhilipp Reisner } 1509b411b363SPhilipp Reisner 1510b411b363SPhilipp Reisner drbd_state_lock(mdev); 1511b411b363SPhilipp Reisner 1512b411b363SPhilipp Reisner if (!get_ldev_if_state(mdev, D_NEGOTIATING)) { 1513b411b363SPhilipp Reisner drbd_state_unlock(mdev); 1514b411b363SPhilipp Reisner return; 1515b411b363SPhilipp Reisner } 1516b411b363SPhilipp Reisner 1517b411b363SPhilipp Reisner if (side == C_SYNC_TARGET) { 1518b411b363SPhilipp Reisner mdev->bm_resync_fo = 0; 1519b411b363SPhilipp Reisner } else /* side == C_SYNC_SOURCE */ { 1520b411b363SPhilipp Reisner u64 uuid; 1521b411b363SPhilipp Reisner 1522b411b363SPhilipp Reisner get_random_bytes(&uuid, sizeof(u64)); 1523b411b363SPhilipp Reisner drbd_uuid_set(mdev, UI_BITMAP, uuid); 1524b411b363SPhilipp Reisner drbd_send_sync_uuid(mdev, uuid); 1525b411b363SPhilipp Reisner 1526b411b363SPhilipp Reisner D_ASSERT(mdev->state.disk == D_UP_TO_DATE); 1527b411b363SPhilipp Reisner } 1528b411b363SPhilipp Reisner 1529b411b363SPhilipp Reisner write_lock_irq(&global_state_lock); 1530b411b363SPhilipp Reisner ns = mdev->state; 1531b411b363SPhilipp Reisner 1532b411b363SPhilipp Reisner ns.aftr_isp = !_drbd_may_sync_now(mdev); 1533b411b363SPhilipp Reisner 1534b411b363SPhilipp Reisner ns.conn = side; 1535b411b363SPhilipp Reisner 1536b411b363SPhilipp Reisner if (side == C_SYNC_TARGET) 1537b411b363SPhilipp Reisner ns.disk = D_INCONSISTENT; 1538b411b363SPhilipp Reisner else /* side == C_SYNC_SOURCE */ 1539b411b363SPhilipp Reisner ns.pdsk = D_INCONSISTENT; 1540b411b363SPhilipp Reisner 1541b411b363SPhilipp Reisner r = __drbd_set_state(mdev, ns, CS_VERBOSE, NULL); 1542b411b363SPhilipp Reisner ns = mdev->state; 1543b411b363SPhilipp Reisner 1544b411b363SPhilipp Reisner if (ns.conn < C_CONNECTED) 1545b411b363SPhilipp Reisner r = SS_UNKNOWN_ERROR; 1546b411b363SPhilipp Reisner 1547b411b363SPhilipp Reisner if (r == SS_SUCCESS) { 15481d7734a0SLars Ellenberg unsigned long tw = drbd_bm_total_weight(mdev); 15491d7734a0SLars Ellenberg unsigned long now = jiffies; 15501d7734a0SLars Ellenberg int i; 15511d7734a0SLars Ellenberg 1552b411b363SPhilipp Reisner mdev->rs_failed = 0; 1553b411b363SPhilipp Reisner mdev->rs_paused = 0; 1554b411b363SPhilipp Reisner mdev->rs_same_csum = 0; 15550f0601f4SLars Ellenberg mdev->rs_last_events = 0; 15560f0601f4SLars Ellenberg mdev->rs_last_sect_ev = 0; 15571d7734a0SLars Ellenberg mdev->rs_total = tw; 15581d7734a0SLars Ellenberg mdev->rs_start = now; 15591d7734a0SLars Ellenberg for (i = 0; i < DRBD_SYNC_MARKS; i++) { 15601d7734a0SLars Ellenberg mdev->rs_mark_left[i] = tw; 15611d7734a0SLars Ellenberg mdev->rs_mark_time[i] = now; 15621d7734a0SLars Ellenberg } 1563b411b363SPhilipp Reisner _drbd_pause_after(mdev); 1564b411b363SPhilipp Reisner } 1565b411b363SPhilipp Reisner write_unlock_irq(&global_state_lock); 1566b411b363SPhilipp Reisner put_ldev(mdev); 1567b411b363SPhilipp Reisner 1568b411b363SPhilipp Reisner if (r == SS_SUCCESS) { 1569b411b363SPhilipp Reisner dev_info(DEV, "Began resync as %s (will sync %lu KB [%lu bits set]).\n", 1570b411b363SPhilipp Reisner drbd_conn_str(ns.conn), 1571b411b363SPhilipp Reisner (unsigned long) mdev->rs_total << (BM_BLOCK_SHIFT-10), 1572b411b363SPhilipp Reisner (unsigned long) mdev->rs_total); 1573b411b363SPhilipp Reisner 1574af85e8e8SLars Ellenberg if (mdev->agreed_pro_version < 95 && mdev->rs_total == 0) { 1575af85e8e8SLars Ellenberg /* This still has a race (about when exactly the peers 1576af85e8e8SLars Ellenberg * detect connection loss) that can lead to a full sync 1577af85e8e8SLars Ellenberg * on next handshake. In 8.3.9 we fixed this with explicit 1578af85e8e8SLars Ellenberg * resync-finished notifications, but the fix 1579af85e8e8SLars Ellenberg * introduces a protocol change. Sleeping for some 1580af85e8e8SLars Ellenberg * time longer than the ping interval + timeout on the 1581af85e8e8SLars Ellenberg * SyncSource, to give the SyncTarget the chance to 1582af85e8e8SLars Ellenberg * detect connection loss, then waiting for a ping 1583af85e8e8SLars Ellenberg * response (implicit in drbd_resync_finished) reduces 1584af85e8e8SLars Ellenberg * the race considerably, but does not solve it. */ 1585af85e8e8SLars Ellenberg if (side == C_SYNC_SOURCE) 1586af85e8e8SLars Ellenberg schedule_timeout_interruptible( 1587af85e8e8SLars Ellenberg mdev->net_conf->ping_int * HZ + 1588af85e8e8SLars Ellenberg mdev->net_conf->ping_timeo*HZ/9); 1589b411b363SPhilipp Reisner drbd_resync_finished(mdev); 1590b411b363SPhilipp Reisner } 1591b411b363SPhilipp Reisner 15929bd28d3cSLars Ellenberg drbd_rs_controller_reset(mdev); 1593b411b363SPhilipp Reisner /* ns.conn may already be != mdev->state.conn, 1594b411b363SPhilipp Reisner * we may have been paused in between, or become paused until 1595b411b363SPhilipp Reisner * the timer triggers. 1596b411b363SPhilipp Reisner * No matter, that is handled in resync_timer_fn() */ 1597b411b363SPhilipp Reisner if (ns.conn == C_SYNC_TARGET) 1598b411b363SPhilipp Reisner mod_timer(&mdev->resync_timer, jiffies); 1599b411b363SPhilipp Reisner 1600b411b363SPhilipp Reisner drbd_md_sync(mdev); 1601b411b363SPhilipp Reisner } 1602d0c3f60fSPhilipp Reisner drbd_state_unlock(mdev); 1603b411b363SPhilipp Reisner } 1604b411b363SPhilipp Reisner 1605b411b363SPhilipp Reisner int drbd_worker(struct drbd_thread *thi) 1606b411b363SPhilipp Reisner { 1607b411b363SPhilipp Reisner struct drbd_conf *mdev = thi->mdev; 1608b411b363SPhilipp Reisner struct drbd_work *w = NULL; 1609b411b363SPhilipp Reisner LIST_HEAD(work_list); 1610b411b363SPhilipp Reisner int intr = 0, i; 1611b411b363SPhilipp Reisner 1612b411b363SPhilipp Reisner sprintf(current->comm, "drbd%d_worker", mdev_to_minor(mdev)); 1613b411b363SPhilipp Reisner 1614b411b363SPhilipp Reisner while (get_t_state(thi) == Running) { 1615b411b363SPhilipp Reisner drbd_thread_current_set_cpu(mdev); 1616b411b363SPhilipp Reisner 1617b411b363SPhilipp Reisner if (down_trylock(&mdev->data.work.s)) { 1618b411b363SPhilipp Reisner mutex_lock(&mdev->data.mutex); 1619b411b363SPhilipp Reisner if (mdev->data.socket && !mdev->net_conf->no_cork) 1620b411b363SPhilipp Reisner drbd_tcp_uncork(mdev->data.socket); 1621b411b363SPhilipp Reisner mutex_unlock(&mdev->data.mutex); 1622b411b363SPhilipp Reisner 1623b411b363SPhilipp Reisner intr = down_interruptible(&mdev->data.work.s); 1624b411b363SPhilipp Reisner 1625b411b363SPhilipp Reisner mutex_lock(&mdev->data.mutex); 1626b411b363SPhilipp Reisner if (mdev->data.socket && !mdev->net_conf->no_cork) 1627b411b363SPhilipp Reisner drbd_tcp_cork(mdev->data.socket); 1628b411b363SPhilipp Reisner mutex_unlock(&mdev->data.mutex); 1629b411b363SPhilipp Reisner } 1630b411b363SPhilipp Reisner 1631b411b363SPhilipp Reisner if (intr) { 1632b411b363SPhilipp Reisner D_ASSERT(intr == -EINTR); 1633b411b363SPhilipp Reisner flush_signals(current); 1634b411b363SPhilipp Reisner ERR_IF (get_t_state(thi) == Running) 1635b411b363SPhilipp Reisner continue; 1636b411b363SPhilipp Reisner break; 1637b411b363SPhilipp Reisner } 1638b411b363SPhilipp Reisner 1639b411b363SPhilipp Reisner if (get_t_state(thi) != Running) 1640b411b363SPhilipp Reisner break; 1641b411b363SPhilipp Reisner /* With this break, we have done a down() but not consumed 1642b411b363SPhilipp Reisner the entry from the list. The cleanup code takes care of 1643b411b363SPhilipp Reisner this... */ 1644b411b363SPhilipp Reisner 1645b411b363SPhilipp Reisner w = NULL; 1646b411b363SPhilipp Reisner spin_lock_irq(&mdev->data.work.q_lock); 1647b411b363SPhilipp Reisner ERR_IF(list_empty(&mdev->data.work.q)) { 1648b411b363SPhilipp Reisner /* something terribly wrong in our logic. 1649b411b363SPhilipp Reisner * we were able to down() the semaphore, 1650b411b363SPhilipp Reisner * but the list is empty... doh. 1651b411b363SPhilipp Reisner * 1652b411b363SPhilipp Reisner * what is the best thing to do now? 1653b411b363SPhilipp Reisner * try again from scratch, restarting the receiver, 1654b411b363SPhilipp Reisner * asender, whatnot? could break even more ugly, 1655b411b363SPhilipp Reisner * e.g. when we are primary, but no good local data. 1656b411b363SPhilipp Reisner * 1657b411b363SPhilipp Reisner * I'll try to get away just starting over this loop. 1658b411b363SPhilipp Reisner */ 1659b411b363SPhilipp Reisner spin_unlock_irq(&mdev->data.work.q_lock); 1660b411b363SPhilipp Reisner continue; 1661b411b363SPhilipp Reisner } 1662b411b363SPhilipp Reisner w = list_entry(mdev->data.work.q.next, struct drbd_work, list); 1663b411b363SPhilipp Reisner list_del_init(&w->list); 1664b411b363SPhilipp Reisner spin_unlock_irq(&mdev->data.work.q_lock); 1665b411b363SPhilipp Reisner 1666b411b363SPhilipp Reisner if (!w->cb(mdev, w, mdev->state.conn < C_CONNECTED)) { 1667b411b363SPhilipp Reisner /* dev_warn(DEV, "worker: a callback failed! \n"); */ 1668b411b363SPhilipp Reisner if (mdev->state.conn >= C_CONNECTED) 1669b411b363SPhilipp Reisner drbd_force_state(mdev, 1670b411b363SPhilipp Reisner NS(conn, C_NETWORK_FAILURE)); 1671b411b363SPhilipp Reisner } 1672b411b363SPhilipp Reisner } 1673b411b363SPhilipp Reisner D_ASSERT(test_bit(DEVICE_DYING, &mdev->flags)); 1674b411b363SPhilipp Reisner D_ASSERT(test_bit(CONFIG_PENDING, &mdev->flags)); 1675b411b363SPhilipp Reisner 1676b411b363SPhilipp Reisner spin_lock_irq(&mdev->data.work.q_lock); 1677b411b363SPhilipp Reisner i = 0; 1678b411b363SPhilipp Reisner while (!list_empty(&mdev->data.work.q)) { 1679b411b363SPhilipp Reisner list_splice_init(&mdev->data.work.q, &work_list); 1680b411b363SPhilipp Reisner spin_unlock_irq(&mdev->data.work.q_lock); 1681b411b363SPhilipp Reisner 1682b411b363SPhilipp Reisner while (!list_empty(&work_list)) { 1683b411b363SPhilipp Reisner w = list_entry(work_list.next, struct drbd_work, list); 1684b411b363SPhilipp Reisner list_del_init(&w->list); 1685b411b363SPhilipp Reisner w->cb(mdev, w, 1); 1686b411b363SPhilipp Reisner i++; /* dead debugging code */ 1687b411b363SPhilipp Reisner } 1688b411b363SPhilipp Reisner 1689b411b363SPhilipp Reisner spin_lock_irq(&mdev->data.work.q_lock); 1690b411b363SPhilipp Reisner } 1691b411b363SPhilipp Reisner sema_init(&mdev->data.work.s, 0); 1692b411b363SPhilipp Reisner /* DANGEROUS race: if someone did queue his work within the spinlock, 1693b411b363SPhilipp Reisner * but up() ed outside the spinlock, we could get an up() on the 1694b411b363SPhilipp Reisner * semaphore without corresponding list entry. 1695b411b363SPhilipp Reisner * So don't do that. 1696b411b363SPhilipp Reisner */ 1697b411b363SPhilipp Reisner spin_unlock_irq(&mdev->data.work.q_lock); 1698b411b363SPhilipp Reisner 1699b411b363SPhilipp Reisner D_ASSERT(mdev->state.disk == D_DISKLESS && mdev->state.conn == C_STANDALONE); 1700b411b363SPhilipp Reisner /* _drbd_set_state only uses stop_nowait. 1701b411b363SPhilipp Reisner * wait here for the Exiting receiver. */ 1702b411b363SPhilipp Reisner drbd_thread_stop(&mdev->receiver); 1703b411b363SPhilipp Reisner drbd_mdev_cleanup(mdev); 1704b411b363SPhilipp Reisner 1705b411b363SPhilipp Reisner dev_info(DEV, "worker terminated\n"); 1706b411b363SPhilipp Reisner 1707b411b363SPhilipp Reisner clear_bit(DEVICE_DYING, &mdev->flags); 1708b411b363SPhilipp Reisner clear_bit(CONFIG_PENDING, &mdev->flags); 1709b411b363SPhilipp Reisner wake_up(&mdev->state_wait); 1710b411b363SPhilipp Reisner 1711b411b363SPhilipp Reisner return 0; 1712b411b363SPhilipp Reisner } 1713