1b411b363SPhilipp Reisner /* 2b411b363SPhilipp Reisner drbd_receiver.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 28b411b363SPhilipp Reisner #include <asm/uaccess.h> 29b411b363SPhilipp Reisner #include <net/sock.h> 30b411b363SPhilipp Reisner 31b411b363SPhilipp Reisner #include <linux/drbd.h> 32b411b363SPhilipp Reisner #include <linux/fs.h> 33b411b363SPhilipp Reisner #include <linux/file.h> 34b411b363SPhilipp Reisner #include <linux/in.h> 35b411b363SPhilipp Reisner #include <linux/mm.h> 36b411b363SPhilipp Reisner #include <linux/memcontrol.h> 37b411b363SPhilipp Reisner #include <linux/mm_inline.h> 38b411b363SPhilipp Reisner #include <linux/slab.h> 39b411b363SPhilipp Reisner #include <linux/smp_lock.h> 40b411b363SPhilipp Reisner #include <linux/pkt_sched.h> 41b411b363SPhilipp Reisner #define __KERNEL_SYSCALLS__ 42b411b363SPhilipp Reisner #include <linux/unistd.h> 43b411b363SPhilipp Reisner #include <linux/vmalloc.h> 44b411b363SPhilipp Reisner #include <linux/random.h> 45b411b363SPhilipp Reisner #include <linux/mm.h> 46b411b363SPhilipp Reisner #include <linux/string.h> 47b411b363SPhilipp Reisner #include <linux/scatterlist.h> 48b411b363SPhilipp Reisner #include "drbd_int.h" 49b411b363SPhilipp Reisner #include "drbd_req.h" 50b411b363SPhilipp Reisner 51b411b363SPhilipp Reisner #include "drbd_vli.h" 52b411b363SPhilipp Reisner 53b411b363SPhilipp Reisner struct flush_work { 54b411b363SPhilipp Reisner struct drbd_work w; 55b411b363SPhilipp Reisner struct drbd_epoch *epoch; 56b411b363SPhilipp Reisner }; 57b411b363SPhilipp Reisner 58b411b363SPhilipp Reisner enum finish_epoch { 59b411b363SPhilipp Reisner FE_STILL_LIVE, 60b411b363SPhilipp Reisner FE_DESTROYED, 61b411b363SPhilipp Reisner FE_RECYCLED, 62b411b363SPhilipp Reisner }; 63b411b363SPhilipp Reisner 64b411b363SPhilipp Reisner static int drbd_do_handshake(struct drbd_conf *mdev); 65b411b363SPhilipp Reisner static int drbd_do_auth(struct drbd_conf *mdev); 66b411b363SPhilipp Reisner 67b411b363SPhilipp Reisner static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *, struct drbd_epoch *, enum epoch_event); 68b411b363SPhilipp Reisner static int e_end_block(struct drbd_conf *, struct drbd_work *, int); 69b411b363SPhilipp Reisner 70b411b363SPhilipp Reisner static struct drbd_epoch *previous_epoch(struct drbd_conf *mdev, struct drbd_epoch *epoch) 71b411b363SPhilipp Reisner { 72b411b363SPhilipp Reisner struct drbd_epoch *prev; 73b411b363SPhilipp Reisner spin_lock(&mdev->epoch_lock); 74b411b363SPhilipp Reisner prev = list_entry(epoch->list.prev, struct drbd_epoch, list); 75b411b363SPhilipp Reisner if (prev == epoch || prev == mdev->current_epoch) 76b411b363SPhilipp Reisner prev = NULL; 77b411b363SPhilipp Reisner spin_unlock(&mdev->epoch_lock); 78b411b363SPhilipp Reisner return prev; 79b411b363SPhilipp Reisner } 80b411b363SPhilipp Reisner 81b411b363SPhilipp Reisner #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) 82b411b363SPhilipp Reisner 83b411b363SPhilipp Reisner static struct page *drbd_pp_first_page_or_try_alloc(struct drbd_conf *mdev) 84b411b363SPhilipp Reisner { 85b411b363SPhilipp Reisner struct page *page = NULL; 86b411b363SPhilipp Reisner 87b411b363SPhilipp Reisner /* Yes, testing drbd_pp_vacant outside the lock is racy. 88b411b363SPhilipp Reisner * So what. It saves a spin_lock. */ 89b411b363SPhilipp Reisner if (drbd_pp_vacant > 0) { 90b411b363SPhilipp Reisner spin_lock(&drbd_pp_lock); 91b411b363SPhilipp Reisner page = drbd_pp_pool; 92b411b363SPhilipp Reisner if (page) { 93b411b363SPhilipp Reisner drbd_pp_pool = (struct page *)page_private(page); 94b411b363SPhilipp Reisner set_page_private(page, 0); /* just to be polite */ 95b411b363SPhilipp Reisner drbd_pp_vacant--; 96b411b363SPhilipp Reisner } 97b411b363SPhilipp Reisner spin_unlock(&drbd_pp_lock); 98b411b363SPhilipp Reisner } 99b411b363SPhilipp Reisner /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD 100b411b363SPhilipp Reisner * "criss-cross" setup, that might cause write-out on some other DRBD, 101b411b363SPhilipp Reisner * which in turn might block on the other node at this very place. */ 102b411b363SPhilipp Reisner if (!page) 103b411b363SPhilipp Reisner page = alloc_page(GFP_TRY); 104b411b363SPhilipp Reisner if (page) 105b411b363SPhilipp Reisner atomic_inc(&mdev->pp_in_use); 106b411b363SPhilipp Reisner return page; 107b411b363SPhilipp Reisner } 108b411b363SPhilipp Reisner 109b411b363SPhilipp Reisner /* kick lower level device, if we have more than (arbitrary number) 110b411b363SPhilipp Reisner * reference counts on it, which typically are locally submitted io 111b411b363SPhilipp Reisner * requests. don't use unacked_cnt, so we speed up proto A and B, too. */ 112b411b363SPhilipp Reisner static void maybe_kick_lo(struct drbd_conf *mdev) 113b411b363SPhilipp Reisner { 114b411b363SPhilipp Reisner if (atomic_read(&mdev->local_cnt) >= mdev->net_conf->unplug_watermark) 115b411b363SPhilipp Reisner drbd_kick_lo(mdev); 116b411b363SPhilipp Reisner } 117b411b363SPhilipp Reisner 118b411b363SPhilipp Reisner static void reclaim_net_ee(struct drbd_conf *mdev, struct list_head *to_be_freed) 119b411b363SPhilipp Reisner { 120b411b363SPhilipp Reisner struct drbd_epoch_entry *e; 121b411b363SPhilipp Reisner struct list_head *le, *tle; 122b411b363SPhilipp Reisner 123b411b363SPhilipp Reisner /* The EEs are always appended to the end of the list. Since 124b411b363SPhilipp Reisner they are sent in order over the wire, they have to finish 125b411b363SPhilipp Reisner in order. As soon as we see the first not finished we can 126b411b363SPhilipp Reisner stop to examine the list... */ 127b411b363SPhilipp Reisner 128b411b363SPhilipp Reisner list_for_each_safe(le, tle, &mdev->net_ee) { 129b411b363SPhilipp Reisner e = list_entry(le, struct drbd_epoch_entry, w.list); 130b411b363SPhilipp Reisner if (drbd_bio_has_active_page(e->private_bio)) 131b411b363SPhilipp Reisner break; 132b411b363SPhilipp Reisner list_move(le, to_be_freed); 133b411b363SPhilipp Reisner } 134b411b363SPhilipp Reisner } 135b411b363SPhilipp Reisner 136b411b363SPhilipp Reisner static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev) 137b411b363SPhilipp Reisner { 138b411b363SPhilipp Reisner LIST_HEAD(reclaimed); 139b411b363SPhilipp Reisner struct drbd_epoch_entry *e, *t; 140b411b363SPhilipp Reisner 141b411b363SPhilipp Reisner maybe_kick_lo(mdev); 142b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 143b411b363SPhilipp Reisner reclaim_net_ee(mdev, &reclaimed); 144b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 145b411b363SPhilipp Reisner 146b411b363SPhilipp Reisner list_for_each_entry_safe(e, t, &reclaimed, w.list) 147b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 148b411b363SPhilipp Reisner } 149b411b363SPhilipp Reisner 150b411b363SPhilipp Reisner /** 151b411b363SPhilipp Reisner * drbd_pp_alloc() - Returns a page, fails only if a signal comes in 152b411b363SPhilipp Reisner * @mdev: DRBD device. 153b411b363SPhilipp Reisner * @retry: whether or not to retry allocation forever (or until signalled) 154b411b363SPhilipp Reisner * 155b411b363SPhilipp Reisner * Tries to allocate a page, first from our own page pool, then from the 156b411b363SPhilipp Reisner * kernel, unless this allocation would exceed the max_buffers setting. 157b411b363SPhilipp Reisner * If @retry is non-zero, retry until DRBD frees a page somewhere else. 158b411b363SPhilipp Reisner */ 159b411b363SPhilipp Reisner static struct page *drbd_pp_alloc(struct drbd_conf *mdev, int retry) 160b411b363SPhilipp Reisner { 161b411b363SPhilipp Reisner struct page *page = NULL; 162b411b363SPhilipp Reisner DEFINE_WAIT(wait); 163b411b363SPhilipp Reisner 164b411b363SPhilipp Reisner if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) { 165b411b363SPhilipp Reisner page = drbd_pp_first_page_or_try_alloc(mdev); 166b411b363SPhilipp Reisner if (page) 167b411b363SPhilipp Reisner return page; 168b411b363SPhilipp Reisner } 169b411b363SPhilipp Reisner 170b411b363SPhilipp Reisner for (;;) { 171b411b363SPhilipp Reisner prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); 172b411b363SPhilipp Reisner 173b411b363SPhilipp Reisner drbd_kick_lo_and_reclaim_net(mdev); 174b411b363SPhilipp Reisner 175b411b363SPhilipp Reisner if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) { 176b411b363SPhilipp Reisner page = drbd_pp_first_page_or_try_alloc(mdev); 177b411b363SPhilipp Reisner if (page) 178b411b363SPhilipp Reisner break; 179b411b363SPhilipp Reisner } 180b411b363SPhilipp Reisner 181b411b363SPhilipp Reisner if (!retry) 182b411b363SPhilipp Reisner break; 183b411b363SPhilipp Reisner 184b411b363SPhilipp Reisner if (signal_pending(current)) { 185b411b363SPhilipp Reisner dev_warn(DEV, "drbd_pp_alloc interrupted!\n"); 186b411b363SPhilipp Reisner break; 187b411b363SPhilipp Reisner } 188b411b363SPhilipp Reisner 189b411b363SPhilipp Reisner schedule(); 190b411b363SPhilipp Reisner } 191b411b363SPhilipp Reisner finish_wait(&drbd_pp_wait, &wait); 192b411b363SPhilipp Reisner 193b411b363SPhilipp Reisner return page; 194b411b363SPhilipp Reisner } 195b411b363SPhilipp Reisner 196b411b363SPhilipp Reisner /* Must not be used from irq, as that may deadlock: see drbd_pp_alloc. 197b411b363SPhilipp Reisner * Is also used from inside an other spin_lock_irq(&mdev->req_lock) */ 198b411b363SPhilipp Reisner static void drbd_pp_free(struct drbd_conf *mdev, struct page *page) 199b411b363SPhilipp Reisner { 200b411b363SPhilipp Reisner int free_it; 201b411b363SPhilipp Reisner 202b411b363SPhilipp Reisner spin_lock(&drbd_pp_lock); 203b411b363SPhilipp Reisner if (drbd_pp_vacant > (DRBD_MAX_SEGMENT_SIZE/PAGE_SIZE)*minor_count) { 204b411b363SPhilipp Reisner free_it = 1; 205b411b363SPhilipp Reisner } else { 206b411b363SPhilipp Reisner set_page_private(page, (unsigned long)drbd_pp_pool); 207b411b363SPhilipp Reisner drbd_pp_pool = page; 208b411b363SPhilipp Reisner drbd_pp_vacant++; 209b411b363SPhilipp Reisner free_it = 0; 210b411b363SPhilipp Reisner } 211b411b363SPhilipp Reisner spin_unlock(&drbd_pp_lock); 212b411b363SPhilipp Reisner 213b411b363SPhilipp Reisner atomic_dec(&mdev->pp_in_use); 214b411b363SPhilipp Reisner 215b411b363SPhilipp Reisner if (free_it) 216b411b363SPhilipp Reisner __free_page(page); 217b411b363SPhilipp Reisner 218b411b363SPhilipp Reisner wake_up(&drbd_pp_wait); 219b411b363SPhilipp Reisner } 220b411b363SPhilipp Reisner 221b411b363SPhilipp Reisner static void drbd_pp_free_bio_pages(struct drbd_conf *mdev, struct bio *bio) 222b411b363SPhilipp Reisner { 223b411b363SPhilipp Reisner struct page *p_to_be_freed = NULL; 224b411b363SPhilipp Reisner struct page *page; 225b411b363SPhilipp Reisner struct bio_vec *bvec; 226b411b363SPhilipp Reisner int i; 227b411b363SPhilipp Reisner 228b411b363SPhilipp Reisner spin_lock(&drbd_pp_lock); 229b411b363SPhilipp Reisner __bio_for_each_segment(bvec, bio, i, 0) { 230b411b363SPhilipp Reisner if (drbd_pp_vacant > (DRBD_MAX_SEGMENT_SIZE/PAGE_SIZE)*minor_count) { 231b411b363SPhilipp Reisner set_page_private(bvec->bv_page, (unsigned long)p_to_be_freed); 232b411b363SPhilipp Reisner p_to_be_freed = bvec->bv_page; 233b411b363SPhilipp Reisner } else { 234b411b363SPhilipp Reisner set_page_private(bvec->bv_page, (unsigned long)drbd_pp_pool); 235b411b363SPhilipp Reisner drbd_pp_pool = bvec->bv_page; 236b411b363SPhilipp Reisner drbd_pp_vacant++; 237b411b363SPhilipp Reisner } 238b411b363SPhilipp Reisner } 239b411b363SPhilipp Reisner spin_unlock(&drbd_pp_lock); 240b411b363SPhilipp Reisner atomic_sub(bio->bi_vcnt, &mdev->pp_in_use); 241b411b363SPhilipp Reisner 242b411b363SPhilipp Reisner while (p_to_be_freed) { 243b411b363SPhilipp Reisner page = p_to_be_freed; 244b411b363SPhilipp Reisner p_to_be_freed = (struct page *)page_private(page); 245b411b363SPhilipp Reisner set_page_private(page, 0); /* just to be polite */ 246b411b363SPhilipp Reisner put_page(page); 247b411b363SPhilipp Reisner } 248b411b363SPhilipp Reisner 249b411b363SPhilipp Reisner wake_up(&drbd_pp_wait); 250b411b363SPhilipp Reisner } 251b411b363SPhilipp Reisner 252b411b363SPhilipp Reisner /* 253b411b363SPhilipp Reisner You need to hold the req_lock: 254b411b363SPhilipp Reisner _drbd_wait_ee_list_empty() 255b411b363SPhilipp Reisner 256b411b363SPhilipp Reisner You must not have the req_lock: 257b411b363SPhilipp Reisner drbd_free_ee() 258b411b363SPhilipp Reisner drbd_alloc_ee() 259b411b363SPhilipp Reisner drbd_init_ee() 260b411b363SPhilipp Reisner drbd_release_ee() 261b411b363SPhilipp Reisner drbd_ee_fix_bhs() 262b411b363SPhilipp Reisner drbd_process_done_ee() 263b411b363SPhilipp Reisner drbd_clear_done_ee() 264b411b363SPhilipp Reisner drbd_wait_ee_list_empty() 265b411b363SPhilipp Reisner */ 266b411b363SPhilipp Reisner 267b411b363SPhilipp Reisner struct drbd_epoch_entry *drbd_alloc_ee(struct drbd_conf *mdev, 268b411b363SPhilipp Reisner u64 id, 269b411b363SPhilipp Reisner sector_t sector, 270b411b363SPhilipp Reisner unsigned int data_size, 271b411b363SPhilipp Reisner gfp_t gfp_mask) __must_hold(local) 272b411b363SPhilipp Reisner { 273b411b363SPhilipp Reisner struct request_queue *q; 274b411b363SPhilipp Reisner struct drbd_epoch_entry *e; 275b411b363SPhilipp Reisner struct page *page; 276b411b363SPhilipp Reisner struct bio *bio; 277b411b363SPhilipp Reisner unsigned int ds; 278b411b363SPhilipp Reisner 279b411b363SPhilipp Reisner if (FAULT_ACTIVE(mdev, DRBD_FAULT_AL_EE)) 280b411b363SPhilipp Reisner return NULL; 281b411b363SPhilipp Reisner 282b411b363SPhilipp Reisner e = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); 283b411b363SPhilipp Reisner if (!e) { 284b411b363SPhilipp Reisner if (!(gfp_mask & __GFP_NOWARN)) 285b411b363SPhilipp Reisner dev_err(DEV, "alloc_ee: Allocation of an EE failed\n"); 286b411b363SPhilipp Reisner return NULL; 287b411b363SPhilipp Reisner } 288b411b363SPhilipp Reisner 289b411b363SPhilipp Reisner bio = bio_alloc(gfp_mask & ~__GFP_HIGHMEM, div_ceil(data_size, PAGE_SIZE)); 290b411b363SPhilipp Reisner if (!bio) { 291b411b363SPhilipp Reisner if (!(gfp_mask & __GFP_NOWARN)) 292b411b363SPhilipp Reisner dev_err(DEV, "alloc_ee: Allocation of a bio failed\n"); 293b411b363SPhilipp Reisner goto fail1; 294b411b363SPhilipp Reisner } 295b411b363SPhilipp Reisner 296b411b363SPhilipp Reisner bio->bi_bdev = mdev->ldev->backing_bdev; 297b411b363SPhilipp Reisner bio->bi_sector = sector; 298b411b363SPhilipp Reisner 299b411b363SPhilipp Reisner ds = data_size; 300b411b363SPhilipp Reisner while (ds) { 301b411b363SPhilipp Reisner page = drbd_pp_alloc(mdev, (gfp_mask & __GFP_WAIT)); 302b411b363SPhilipp Reisner if (!page) { 303b411b363SPhilipp Reisner if (!(gfp_mask & __GFP_NOWARN)) 304b411b363SPhilipp Reisner dev_err(DEV, "alloc_ee: Allocation of a page failed\n"); 305b411b363SPhilipp Reisner goto fail2; 306b411b363SPhilipp Reisner } 307b411b363SPhilipp Reisner if (!bio_add_page(bio, page, min_t(int, ds, PAGE_SIZE), 0)) { 308b411b363SPhilipp Reisner drbd_pp_free(mdev, page); 309b411b363SPhilipp Reisner dev_err(DEV, "alloc_ee: bio_add_page(s=%llu," 310b411b363SPhilipp Reisner "data_size=%u,ds=%u) failed\n", 311b411b363SPhilipp Reisner (unsigned long long)sector, data_size, ds); 312b411b363SPhilipp Reisner 313b411b363SPhilipp Reisner q = bdev_get_queue(bio->bi_bdev); 314b411b363SPhilipp Reisner if (q->merge_bvec_fn) { 315b411b363SPhilipp Reisner struct bvec_merge_data bvm = { 316b411b363SPhilipp Reisner .bi_bdev = bio->bi_bdev, 317b411b363SPhilipp Reisner .bi_sector = bio->bi_sector, 318b411b363SPhilipp Reisner .bi_size = bio->bi_size, 319b411b363SPhilipp Reisner .bi_rw = bio->bi_rw, 320b411b363SPhilipp Reisner }; 321b411b363SPhilipp Reisner int l = q->merge_bvec_fn(q, &bvm, 322b411b363SPhilipp Reisner &bio->bi_io_vec[bio->bi_vcnt]); 323b411b363SPhilipp Reisner dev_err(DEV, "merge_bvec_fn() = %d\n", l); 324b411b363SPhilipp Reisner } 325b411b363SPhilipp Reisner 326b411b363SPhilipp Reisner /* dump more of the bio. */ 327b411b363SPhilipp Reisner dev_err(DEV, "bio->bi_max_vecs = %d\n", bio->bi_max_vecs); 328b411b363SPhilipp Reisner dev_err(DEV, "bio->bi_vcnt = %d\n", bio->bi_vcnt); 329b411b363SPhilipp Reisner dev_err(DEV, "bio->bi_size = %d\n", bio->bi_size); 330b411b363SPhilipp Reisner dev_err(DEV, "bio->bi_phys_segments = %d\n", bio->bi_phys_segments); 331b411b363SPhilipp Reisner 332b411b363SPhilipp Reisner goto fail2; 333b411b363SPhilipp Reisner break; 334b411b363SPhilipp Reisner } 335b411b363SPhilipp Reisner ds -= min_t(int, ds, PAGE_SIZE); 336b411b363SPhilipp Reisner } 337b411b363SPhilipp Reisner 338b411b363SPhilipp Reisner D_ASSERT(data_size == bio->bi_size); 339b411b363SPhilipp Reisner 340b411b363SPhilipp Reisner bio->bi_private = e; 341b411b363SPhilipp Reisner e->mdev = mdev; 342b411b363SPhilipp Reisner e->sector = sector; 343b411b363SPhilipp Reisner e->size = bio->bi_size; 344b411b363SPhilipp Reisner 345b411b363SPhilipp Reisner e->private_bio = bio; 346b411b363SPhilipp Reisner e->block_id = id; 347b411b363SPhilipp Reisner INIT_HLIST_NODE(&e->colision); 348b411b363SPhilipp Reisner e->epoch = NULL; 349b411b363SPhilipp Reisner e->flags = 0; 350b411b363SPhilipp Reisner 351b411b363SPhilipp Reisner return e; 352b411b363SPhilipp Reisner 353b411b363SPhilipp Reisner fail2: 354b411b363SPhilipp Reisner drbd_pp_free_bio_pages(mdev, bio); 355b411b363SPhilipp Reisner bio_put(bio); 356b411b363SPhilipp Reisner fail1: 357b411b363SPhilipp Reisner mempool_free(e, drbd_ee_mempool); 358b411b363SPhilipp Reisner 359b411b363SPhilipp Reisner return NULL; 360b411b363SPhilipp Reisner } 361b411b363SPhilipp Reisner 362b411b363SPhilipp Reisner void drbd_free_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e) 363b411b363SPhilipp Reisner { 364b411b363SPhilipp Reisner struct bio *bio = e->private_bio; 365b411b363SPhilipp Reisner drbd_pp_free_bio_pages(mdev, bio); 366b411b363SPhilipp Reisner bio_put(bio); 367b411b363SPhilipp Reisner D_ASSERT(hlist_unhashed(&e->colision)); 368b411b363SPhilipp Reisner mempool_free(e, drbd_ee_mempool); 369b411b363SPhilipp Reisner } 370b411b363SPhilipp Reisner 371b411b363SPhilipp Reisner int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list) 372b411b363SPhilipp Reisner { 373b411b363SPhilipp Reisner LIST_HEAD(work_list); 374b411b363SPhilipp Reisner struct drbd_epoch_entry *e, *t; 375b411b363SPhilipp Reisner int count = 0; 376b411b363SPhilipp Reisner 377b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 378b411b363SPhilipp Reisner list_splice_init(list, &work_list); 379b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 380b411b363SPhilipp Reisner 381b411b363SPhilipp Reisner list_for_each_entry_safe(e, t, &work_list, w.list) { 382b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 383b411b363SPhilipp Reisner count++; 384b411b363SPhilipp Reisner } 385b411b363SPhilipp Reisner return count; 386b411b363SPhilipp Reisner } 387b411b363SPhilipp Reisner 388b411b363SPhilipp Reisner 389b411b363SPhilipp Reisner /* 390b411b363SPhilipp Reisner * This function is called from _asender only_ 391b411b363SPhilipp Reisner * but see also comments in _req_mod(,barrier_acked) 392b411b363SPhilipp Reisner * and receive_Barrier. 393b411b363SPhilipp Reisner * 394b411b363SPhilipp Reisner * Move entries from net_ee to done_ee, if ready. 395b411b363SPhilipp Reisner * Grab done_ee, call all callbacks, free the entries. 396b411b363SPhilipp Reisner * The callbacks typically send out ACKs. 397b411b363SPhilipp Reisner */ 398b411b363SPhilipp Reisner static int drbd_process_done_ee(struct drbd_conf *mdev) 399b411b363SPhilipp Reisner { 400b411b363SPhilipp Reisner LIST_HEAD(work_list); 401b411b363SPhilipp Reisner LIST_HEAD(reclaimed); 402b411b363SPhilipp Reisner struct drbd_epoch_entry *e, *t; 403b411b363SPhilipp Reisner int ok = (mdev->state.conn >= C_WF_REPORT_PARAMS); 404b411b363SPhilipp Reisner 405b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 406b411b363SPhilipp Reisner reclaim_net_ee(mdev, &reclaimed); 407b411b363SPhilipp Reisner list_splice_init(&mdev->done_ee, &work_list); 408b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 409b411b363SPhilipp Reisner 410b411b363SPhilipp Reisner list_for_each_entry_safe(e, t, &reclaimed, w.list) 411b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 412b411b363SPhilipp Reisner 413b411b363SPhilipp Reisner /* possible callbacks here: 414b411b363SPhilipp Reisner * e_end_block, and e_end_resync_block, e_send_discard_ack. 415b411b363SPhilipp Reisner * all ignore the last argument. 416b411b363SPhilipp Reisner */ 417b411b363SPhilipp Reisner list_for_each_entry_safe(e, t, &work_list, w.list) { 418b411b363SPhilipp Reisner /* list_del not necessary, next/prev members not touched */ 419b411b363SPhilipp Reisner ok = e->w.cb(mdev, &e->w, !ok) && ok; 420b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 421b411b363SPhilipp Reisner } 422b411b363SPhilipp Reisner wake_up(&mdev->ee_wait); 423b411b363SPhilipp Reisner 424b411b363SPhilipp Reisner return ok; 425b411b363SPhilipp Reisner } 426b411b363SPhilipp Reisner 427b411b363SPhilipp Reisner void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) 428b411b363SPhilipp Reisner { 429b411b363SPhilipp Reisner DEFINE_WAIT(wait); 430b411b363SPhilipp Reisner 431b411b363SPhilipp Reisner /* avoids spin_lock/unlock 432b411b363SPhilipp Reisner * and calling prepare_to_wait in the fast path */ 433b411b363SPhilipp Reisner while (!list_empty(head)) { 434b411b363SPhilipp Reisner prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE); 435b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 436b411b363SPhilipp Reisner drbd_kick_lo(mdev); 437b411b363SPhilipp Reisner schedule(); 438b411b363SPhilipp Reisner finish_wait(&mdev->ee_wait, &wait); 439b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 440b411b363SPhilipp Reisner } 441b411b363SPhilipp Reisner } 442b411b363SPhilipp Reisner 443b411b363SPhilipp Reisner void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) 444b411b363SPhilipp Reisner { 445b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 446b411b363SPhilipp Reisner _drbd_wait_ee_list_empty(mdev, head); 447b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 448b411b363SPhilipp Reisner } 449b411b363SPhilipp Reisner 450b411b363SPhilipp Reisner /* see also kernel_accept; which is only present since 2.6.18. 451b411b363SPhilipp Reisner * also we want to log which part of it failed, exactly */ 452b411b363SPhilipp Reisner static int drbd_accept(struct drbd_conf *mdev, const char **what, 453b411b363SPhilipp Reisner struct socket *sock, struct socket **newsock) 454b411b363SPhilipp Reisner { 455b411b363SPhilipp Reisner struct sock *sk = sock->sk; 456b411b363SPhilipp Reisner int err = 0; 457b411b363SPhilipp Reisner 458b411b363SPhilipp Reisner *what = "listen"; 459b411b363SPhilipp Reisner err = sock->ops->listen(sock, 5); 460b411b363SPhilipp Reisner if (err < 0) 461b411b363SPhilipp Reisner goto out; 462b411b363SPhilipp Reisner 463b411b363SPhilipp Reisner *what = "sock_create_lite"; 464b411b363SPhilipp Reisner err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol, 465b411b363SPhilipp Reisner newsock); 466b411b363SPhilipp Reisner if (err < 0) 467b411b363SPhilipp Reisner goto out; 468b411b363SPhilipp Reisner 469b411b363SPhilipp Reisner *what = "accept"; 470b411b363SPhilipp Reisner err = sock->ops->accept(sock, *newsock, 0); 471b411b363SPhilipp Reisner if (err < 0) { 472b411b363SPhilipp Reisner sock_release(*newsock); 473b411b363SPhilipp Reisner *newsock = NULL; 474b411b363SPhilipp Reisner goto out; 475b411b363SPhilipp Reisner } 476b411b363SPhilipp Reisner (*newsock)->ops = sock->ops; 477b411b363SPhilipp Reisner 478b411b363SPhilipp Reisner out: 479b411b363SPhilipp Reisner return err; 480b411b363SPhilipp Reisner } 481b411b363SPhilipp Reisner 482b411b363SPhilipp Reisner static int drbd_recv_short(struct drbd_conf *mdev, struct socket *sock, 483b411b363SPhilipp Reisner void *buf, size_t size, int flags) 484b411b363SPhilipp Reisner { 485b411b363SPhilipp Reisner mm_segment_t oldfs; 486b411b363SPhilipp Reisner struct kvec iov = { 487b411b363SPhilipp Reisner .iov_base = buf, 488b411b363SPhilipp Reisner .iov_len = size, 489b411b363SPhilipp Reisner }; 490b411b363SPhilipp Reisner struct msghdr msg = { 491b411b363SPhilipp Reisner .msg_iovlen = 1, 492b411b363SPhilipp Reisner .msg_iov = (struct iovec *)&iov, 493b411b363SPhilipp Reisner .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) 494b411b363SPhilipp Reisner }; 495b411b363SPhilipp Reisner int rv; 496b411b363SPhilipp Reisner 497b411b363SPhilipp Reisner oldfs = get_fs(); 498b411b363SPhilipp Reisner set_fs(KERNEL_DS); 499b411b363SPhilipp Reisner rv = sock_recvmsg(sock, &msg, size, msg.msg_flags); 500b411b363SPhilipp Reisner set_fs(oldfs); 501b411b363SPhilipp Reisner 502b411b363SPhilipp Reisner return rv; 503b411b363SPhilipp Reisner } 504b411b363SPhilipp Reisner 505b411b363SPhilipp Reisner static int drbd_recv(struct drbd_conf *mdev, void *buf, size_t size) 506b411b363SPhilipp Reisner { 507b411b363SPhilipp Reisner mm_segment_t oldfs; 508b411b363SPhilipp Reisner struct kvec iov = { 509b411b363SPhilipp Reisner .iov_base = buf, 510b411b363SPhilipp Reisner .iov_len = size, 511b411b363SPhilipp Reisner }; 512b411b363SPhilipp Reisner struct msghdr msg = { 513b411b363SPhilipp Reisner .msg_iovlen = 1, 514b411b363SPhilipp Reisner .msg_iov = (struct iovec *)&iov, 515b411b363SPhilipp Reisner .msg_flags = MSG_WAITALL | MSG_NOSIGNAL 516b411b363SPhilipp Reisner }; 517b411b363SPhilipp Reisner int rv; 518b411b363SPhilipp Reisner 519b411b363SPhilipp Reisner oldfs = get_fs(); 520b411b363SPhilipp Reisner set_fs(KERNEL_DS); 521b411b363SPhilipp Reisner 522b411b363SPhilipp Reisner for (;;) { 523b411b363SPhilipp Reisner rv = sock_recvmsg(mdev->data.socket, &msg, size, msg.msg_flags); 524b411b363SPhilipp Reisner if (rv == size) 525b411b363SPhilipp Reisner break; 526b411b363SPhilipp Reisner 527b411b363SPhilipp Reisner /* Note: 528b411b363SPhilipp Reisner * ECONNRESET other side closed the connection 529b411b363SPhilipp Reisner * ERESTARTSYS (on sock) we got a signal 530b411b363SPhilipp Reisner */ 531b411b363SPhilipp Reisner 532b411b363SPhilipp Reisner if (rv < 0) { 533b411b363SPhilipp Reisner if (rv == -ECONNRESET) 534b411b363SPhilipp Reisner dev_info(DEV, "sock was reset by peer\n"); 535b411b363SPhilipp Reisner else if (rv != -ERESTARTSYS) 536b411b363SPhilipp Reisner dev_err(DEV, "sock_recvmsg returned %d\n", rv); 537b411b363SPhilipp Reisner break; 538b411b363SPhilipp Reisner } else if (rv == 0) { 539b411b363SPhilipp Reisner dev_info(DEV, "sock was shut down by peer\n"); 540b411b363SPhilipp Reisner break; 541b411b363SPhilipp Reisner } else { 542b411b363SPhilipp Reisner /* signal came in, or peer/link went down, 543b411b363SPhilipp Reisner * after we read a partial message 544b411b363SPhilipp Reisner */ 545b411b363SPhilipp Reisner /* D_ASSERT(signal_pending(current)); */ 546b411b363SPhilipp Reisner break; 547b411b363SPhilipp Reisner } 548b411b363SPhilipp Reisner }; 549b411b363SPhilipp Reisner 550b411b363SPhilipp Reisner set_fs(oldfs); 551b411b363SPhilipp Reisner 552b411b363SPhilipp Reisner if (rv != size) 553b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_BROKEN_PIPE)); 554b411b363SPhilipp Reisner 555b411b363SPhilipp Reisner return rv; 556b411b363SPhilipp Reisner } 557b411b363SPhilipp Reisner 558b411b363SPhilipp Reisner static struct socket *drbd_try_connect(struct drbd_conf *mdev) 559b411b363SPhilipp Reisner { 560b411b363SPhilipp Reisner const char *what; 561b411b363SPhilipp Reisner struct socket *sock; 562b411b363SPhilipp Reisner struct sockaddr_in6 src_in6; 563b411b363SPhilipp Reisner int err; 564b411b363SPhilipp Reisner int disconnect_on_error = 1; 565b411b363SPhilipp Reisner 566b411b363SPhilipp Reisner if (!get_net_conf(mdev)) 567b411b363SPhilipp Reisner return NULL; 568b411b363SPhilipp Reisner 569b411b363SPhilipp Reisner what = "sock_create_kern"; 570b411b363SPhilipp Reisner err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family, 571b411b363SPhilipp Reisner SOCK_STREAM, IPPROTO_TCP, &sock); 572b411b363SPhilipp Reisner if (err < 0) { 573b411b363SPhilipp Reisner sock = NULL; 574b411b363SPhilipp Reisner goto out; 575b411b363SPhilipp Reisner } 576b411b363SPhilipp Reisner 577b411b363SPhilipp Reisner sock->sk->sk_rcvtimeo = 578b411b363SPhilipp Reisner sock->sk->sk_sndtimeo = mdev->net_conf->try_connect_int*HZ; 579b411b363SPhilipp Reisner 580b411b363SPhilipp Reisner /* explicitly bind to the configured IP as source IP 581b411b363SPhilipp Reisner * for the outgoing connections. 582b411b363SPhilipp Reisner * This is needed for multihomed hosts and to be 583b411b363SPhilipp Reisner * able to use lo: interfaces for drbd. 584b411b363SPhilipp Reisner * Make sure to use 0 as port number, so linux selects 585b411b363SPhilipp Reisner * a free one dynamically. 586b411b363SPhilipp Reisner */ 587b411b363SPhilipp Reisner memcpy(&src_in6, mdev->net_conf->my_addr, 588b411b363SPhilipp Reisner min_t(int, mdev->net_conf->my_addr_len, sizeof(src_in6))); 589b411b363SPhilipp Reisner if (((struct sockaddr *)mdev->net_conf->my_addr)->sa_family == AF_INET6) 590b411b363SPhilipp Reisner src_in6.sin6_port = 0; 591b411b363SPhilipp Reisner else 592b411b363SPhilipp Reisner ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ 593b411b363SPhilipp Reisner 594b411b363SPhilipp Reisner what = "bind before connect"; 595b411b363SPhilipp Reisner err = sock->ops->bind(sock, 596b411b363SPhilipp Reisner (struct sockaddr *) &src_in6, 597b411b363SPhilipp Reisner mdev->net_conf->my_addr_len); 598b411b363SPhilipp Reisner if (err < 0) 599b411b363SPhilipp Reisner goto out; 600b411b363SPhilipp Reisner 601b411b363SPhilipp Reisner /* connect may fail, peer not yet available. 602b411b363SPhilipp Reisner * stay C_WF_CONNECTION, don't go Disconnecting! */ 603b411b363SPhilipp Reisner disconnect_on_error = 0; 604b411b363SPhilipp Reisner what = "connect"; 605b411b363SPhilipp Reisner err = sock->ops->connect(sock, 606b411b363SPhilipp Reisner (struct sockaddr *)mdev->net_conf->peer_addr, 607b411b363SPhilipp Reisner mdev->net_conf->peer_addr_len, 0); 608b411b363SPhilipp Reisner 609b411b363SPhilipp Reisner out: 610b411b363SPhilipp Reisner if (err < 0) { 611b411b363SPhilipp Reisner if (sock) { 612b411b363SPhilipp Reisner sock_release(sock); 613b411b363SPhilipp Reisner sock = NULL; 614b411b363SPhilipp Reisner } 615b411b363SPhilipp Reisner switch (-err) { 616b411b363SPhilipp Reisner /* timeout, busy, signal pending */ 617b411b363SPhilipp Reisner case ETIMEDOUT: case EAGAIN: case EINPROGRESS: 618b411b363SPhilipp Reisner case EINTR: case ERESTARTSYS: 619b411b363SPhilipp Reisner /* peer not (yet) available, network problem */ 620b411b363SPhilipp Reisner case ECONNREFUSED: case ENETUNREACH: 621b411b363SPhilipp Reisner case EHOSTDOWN: case EHOSTUNREACH: 622b411b363SPhilipp Reisner disconnect_on_error = 0; 623b411b363SPhilipp Reisner break; 624b411b363SPhilipp Reisner default: 625b411b363SPhilipp Reisner dev_err(DEV, "%s failed, err = %d\n", what, err); 626b411b363SPhilipp Reisner } 627b411b363SPhilipp Reisner if (disconnect_on_error) 628b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 629b411b363SPhilipp Reisner } 630b411b363SPhilipp Reisner put_net_conf(mdev); 631b411b363SPhilipp Reisner return sock; 632b411b363SPhilipp Reisner } 633b411b363SPhilipp Reisner 634b411b363SPhilipp Reisner static struct socket *drbd_wait_for_connect(struct drbd_conf *mdev) 635b411b363SPhilipp Reisner { 636b411b363SPhilipp Reisner int timeo, err; 637b411b363SPhilipp Reisner struct socket *s_estab = NULL, *s_listen; 638b411b363SPhilipp Reisner const char *what; 639b411b363SPhilipp Reisner 640b411b363SPhilipp Reisner if (!get_net_conf(mdev)) 641b411b363SPhilipp Reisner return NULL; 642b411b363SPhilipp Reisner 643b411b363SPhilipp Reisner what = "sock_create_kern"; 644b411b363SPhilipp Reisner err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family, 645b411b363SPhilipp Reisner SOCK_STREAM, IPPROTO_TCP, &s_listen); 646b411b363SPhilipp Reisner if (err) { 647b411b363SPhilipp Reisner s_listen = NULL; 648b411b363SPhilipp Reisner goto out; 649b411b363SPhilipp Reisner } 650b411b363SPhilipp Reisner 651b411b363SPhilipp Reisner timeo = mdev->net_conf->try_connect_int * HZ; 652b411b363SPhilipp Reisner timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */ 653b411b363SPhilipp Reisner 654b411b363SPhilipp Reisner s_listen->sk->sk_reuse = 1; /* SO_REUSEADDR */ 655b411b363SPhilipp Reisner s_listen->sk->sk_rcvtimeo = timeo; 656b411b363SPhilipp Reisner s_listen->sk->sk_sndtimeo = timeo; 657b411b363SPhilipp Reisner 658b411b363SPhilipp Reisner what = "bind before listen"; 659b411b363SPhilipp Reisner err = s_listen->ops->bind(s_listen, 660b411b363SPhilipp Reisner (struct sockaddr *) mdev->net_conf->my_addr, 661b411b363SPhilipp Reisner mdev->net_conf->my_addr_len); 662b411b363SPhilipp Reisner if (err < 0) 663b411b363SPhilipp Reisner goto out; 664b411b363SPhilipp Reisner 665b411b363SPhilipp Reisner err = drbd_accept(mdev, &what, s_listen, &s_estab); 666b411b363SPhilipp Reisner 667b411b363SPhilipp Reisner out: 668b411b363SPhilipp Reisner if (s_listen) 669b411b363SPhilipp Reisner sock_release(s_listen); 670b411b363SPhilipp Reisner if (err < 0) { 671b411b363SPhilipp Reisner if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { 672b411b363SPhilipp Reisner dev_err(DEV, "%s failed, err = %d\n", what, err); 673b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 674b411b363SPhilipp Reisner } 675b411b363SPhilipp Reisner } 676b411b363SPhilipp Reisner put_net_conf(mdev); 677b411b363SPhilipp Reisner 678b411b363SPhilipp Reisner return s_estab; 679b411b363SPhilipp Reisner } 680b411b363SPhilipp Reisner 681b411b363SPhilipp Reisner static int drbd_send_fp(struct drbd_conf *mdev, 682b411b363SPhilipp Reisner struct socket *sock, enum drbd_packets cmd) 683b411b363SPhilipp Reisner { 684b411b363SPhilipp Reisner struct p_header *h = (struct p_header *) &mdev->data.sbuf.header; 685b411b363SPhilipp Reisner 686b411b363SPhilipp Reisner return _drbd_send_cmd(mdev, sock, cmd, h, sizeof(*h), 0); 687b411b363SPhilipp Reisner } 688b411b363SPhilipp Reisner 689b411b363SPhilipp Reisner static enum drbd_packets drbd_recv_fp(struct drbd_conf *mdev, struct socket *sock) 690b411b363SPhilipp Reisner { 691b411b363SPhilipp Reisner struct p_header *h = (struct p_header *) &mdev->data.sbuf.header; 692b411b363SPhilipp Reisner int rr; 693b411b363SPhilipp Reisner 694b411b363SPhilipp Reisner rr = drbd_recv_short(mdev, sock, h, sizeof(*h), 0); 695b411b363SPhilipp Reisner 696b411b363SPhilipp Reisner if (rr == sizeof(*h) && h->magic == BE_DRBD_MAGIC) 697b411b363SPhilipp Reisner return be16_to_cpu(h->command); 698b411b363SPhilipp Reisner 699b411b363SPhilipp Reisner return 0xffff; 700b411b363SPhilipp Reisner } 701b411b363SPhilipp Reisner 702b411b363SPhilipp Reisner /** 703b411b363SPhilipp Reisner * drbd_socket_okay() - Free the socket if its connection is not okay 704b411b363SPhilipp Reisner * @mdev: DRBD device. 705b411b363SPhilipp Reisner * @sock: pointer to the pointer to the socket. 706b411b363SPhilipp Reisner */ 707b411b363SPhilipp Reisner static int drbd_socket_okay(struct drbd_conf *mdev, struct socket **sock) 708b411b363SPhilipp Reisner { 709b411b363SPhilipp Reisner int rr; 710b411b363SPhilipp Reisner char tb[4]; 711b411b363SPhilipp Reisner 712b411b363SPhilipp Reisner if (!*sock) 713b411b363SPhilipp Reisner return FALSE; 714b411b363SPhilipp Reisner 715b411b363SPhilipp Reisner rr = drbd_recv_short(mdev, *sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); 716b411b363SPhilipp Reisner 717b411b363SPhilipp Reisner if (rr > 0 || rr == -EAGAIN) { 718b411b363SPhilipp Reisner return TRUE; 719b411b363SPhilipp Reisner } else { 720b411b363SPhilipp Reisner sock_release(*sock); 721b411b363SPhilipp Reisner *sock = NULL; 722b411b363SPhilipp Reisner return FALSE; 723b411b363SPhilipp Reisner } 724b411b363SPhilipp Reisner } 725b411b363SPhilipp Reisner 726b411b363SPhilipp Reisner /* 727b411b363SPhilipp Reisner * return values: 728b411b363SPhilipp Reisner * 1 yes, we have a valid connection 729b411b363SPhilipp Reisner * 0 oops, did not work out, please try again 730b411b363SPhilipp Reisner * -1 peer talks different language, 731b411b363SPhilipp Reisner * no point in trying again, please go standalone. 732b411b363SPhilipp Reisner * -2 We do not have a network config... 733b411b363SPhilipp Reisner */ 734b411b363SPhilipp Reisner static int drbd_connect(struct drbd_conf *mdev) 735b411b363SPhilipp Reisner { 736b411b363SPhilipp Reisner struct socket *s, *sock, *msock; 737b411b363SPhilipp Reisner int try, h, ok; 738b411b363SPhilipp Reisner 739b411b363SPhilipp Reisner D_ASSERT(!mdev->data.socket); 740b411b363SPhilipp Reisner 741b411b363SPhilipp Reisner if (test_and_clear_bit(CREATE_BARRIER, &mdev->flags)) 742b411b363SPhilipp Reisner dev_err(DEV, "CREATE_BARRIER flag was set in drbd_connect - now cleared!\n"); 743b411b363SPhilipp Reisner 744b411b363SPhilipp Reisner if (drbd_request_state(mdev, NS(conn, C_WF_CONNECTION)) < SS_SUCCESS) 745b411b363SPhilipp Reisner return -2; 746b411b363SPhilipp Reisner 747b411b363SPhilipp Reisner clear_bit(DISCARD_CONCURRENT, &mdev->flags); 748b411b363SPhilipp Reisner 749b411b363SPhilipp Reisner sock = NULL; 750b411b363SPhilipp Reisner msock = NULL; 751b411b363SPhilipp Reisner 752b411b363SPhilipp Reisner do { 753b411b363SPhilipp Reisner for (try = 0;;) { 754b411b363SPhilipp Reisner /* 3 tries, this should take less than a second! */ 755b411b363SPhilipp Reisner s = drbd_try_connect(mdev); 756b411b363SPhilipp Reisner if (s || ++try >= 3) 757b411b363SPhilipp Reisner break; 758b411b363SPhilipp Reisner /* give the other side time to call bind() & listen() */ 759b411b363SPhilipp Reisner __set_current_state(TASK_INTERRUPTIBLE); 760b411b363SPhilipp Reisner schedule_timeout(HZ / 10); 761b411b363SPhilipp Reisner } 762b411b363SPhilipp Reisner 763b411b363SPhilipp Reisner if (s) { 764b411b363SPhilipp Reisner if (!sock) { 765b411b363SPhilipp Reisner drbd_send_fp(mdev, s, P_HAND_SHAKE_S); 766b411b363SPhilipp Reisner sock = s; 767b411b363SPhilipp Reisner s = NULL; 768b411b363SPhilipp Reisner } else if (!msock) { 769b411b363SPhilipp Reisner drbd_send_fp(mdev, s, P_HAND_SHAKE_M); 770b411b363SPhilipp Reisner msock = s; 771b411b363SPhilipp Reisner s = NULL; 772b411b363SPhilipp Reisner } else { 773b411b363SPhilipp Reisner dev_err(DEV, "Logic error in drbd_connect()\n"); 774b411b363SPhilipp Reisner goto out_release_sockets; 775b411b363SPhilipp Reisner } 776b411b363SPhilipp Reisner } 777b411b363SPhilipp Reisner 778b411b363SPhilipp Reisner if (sock && msock) { 779b411b363SPhilipp Reisner __set_current_state(TASK_INTERRUPTIBLE); 780b411b363SPhilipp Reisner schedule_timeout(HZ / 10); 781b411b363SPhilipp Reisner ok = drbd_socket_okay(mdev, &sock); 782b411b363SPhilipp Reisner ok = drbd_socket_okay(mdev, &msock) && ok; 783b411b363SPhilipp Reisner if (ok) 784b411b363SPhilipp Reisner break; 785b411b363SPhilipp Reisner } 786b411b363SPhilipp Reisner 787b411b363SPhilipp Reisner retry: 788b411b363SPhilipp Reisner s = drbd_wait_for_connect(mdev); 789b411b363SPhilipp Reisner if (s) { 790b411b363SPhilipp Reisner try = drbd_recv_fp(mdev, s); 791b411b363SPhilipp Reisner drbd_socket_okay(mdev, &sock); 792b411b363SPhilipp Reisner drbd_socket_okay(mdev, &msock); 793b411b363SPhilipp Reisner switch (try) { 794b411b363SPhilipp Reisner case P_HAND_SHAKE_S: 795b411b363SPhilipp Reisner if (sock) { 796b411b363SPhilipp Reisner dev_warn(DEV, "initial packet S crossed\n"); 797b411b363SPhilipp Reisner sock_release(sock); 798b411b363SPhilipp Reisner } 799b411b363SPhilipp Reisner sock = s; 800b411b363SPhilipp Reisner break; 801b411b363SPhilipp Reisner case P_HAND_SHAKE_M: 802b411b363SPhilipp Reisner if (msock) { 803b411b363SPhilipp Reisner dev_warn(DEV, "initial packet M crossed\n"); 804b411b363SPhilipp Reisner sock_release(msock); 805b411b363SPhilipp Reisner } 806b411b363SPhilipp Reisner msock = s; 807b411b363SPhilipp Reisner set_bit(DISCARD_CONCURRENT, &mdev->flags); 808b411b363SPhilipp Reisner break; 809b411b363SPhilipp Reisner default: 810b411b363SPhilipp Reisner dev_warn(DEV, "Error receiving initial packet\n"); 811b411b363SPhilipp Reisner sock_release(s); 812b411b363SPhilipp Reisner if (random32() & 1) 813b411b363SPhilipp Reisner goto retry; 814b411b363SPhilipp Reisner } 815b411b363SPhilipp Reisner } 816b411b363SPhilipp Reisner 817b411b363SPhilipp Reisner if (mdev->state.conn <= C_DISCONNECTING) 818b411b363SPhilipp Reisner goto out_release_sockets; 819b411b363SPhilipp Reisner if (signal_pending(current)) { 820b411b363SPhilipp Reisner flush_signals(current); 821b411b363SPhilipp Reisner smp_rmb(); 822b411b363SPhilipp Reisner if (get_t_state(&mdev->receiver) == Exiting) 823b411b363SPhilipp Reisner goto out_release_sockets; 824b411b363SPhilipp Reisner } 825b411b363SPhilipp Reisner 826b411b363SPhilipp Reisner if (sock && msock) { 827b411b363SPhilipp Reisner ok = drbd_socket_okay(mdev, &sock); 828b411b363SPhilipp Reisner ok = drbd_socket_okay(mdev, &msock) && ok; 829b411b363SPhilipp Reisner if (ok) 830b411b363SPhilipp Reisner break; 831b411b363SPhilipp Reisner } 832b411b363SPhilipp Reisner } while (1); 833b411b363SPhilipp Reisner 834b411b363SPhilipp Reisner msock->sk->sk_reuse = 1; /* SO_REUSEADDR */ 835b411b363SPhilipp Reisner sock->sk->sk_reuse = 1; /* SO_REUSEADDR */ 836b411b363SPhilipp Reisner 837b411b363SPhilipp Reisner sock->sk->sk_allocation = GFP_NOIO; 838b411b363SPhilipp Reisner msock->sk->sk_allocation = GFP_NOIO; 839b411b363SPhilipp Reisner 840b411b363SPhilipp Reisner sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; 841b411b363SPhilipp Reisner msock->sk->sk_priority = TC_PRIO_INTERACTIVE; 842b411b363SPhilipp Reisner 843b411b363SPhilipp Reisner if (mdev->net_conf->sndbuf_size) { 844b411b363SPhilipp Reisner sock->sk->sk_sndbuf = mdev->net_conf->sndbuf_size; 845b411b363SPhilipp Reisner sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 846b411b363SPhilipp Reisner } 847b411b363SPhilipp Reisner 848b411b363SPhilipp Reisner if (mdev->net_conf->rcvbuf_size) { 849b411b363SPhilipp Reisner sock->sk->sk_rcvbuf = mdev->net_conf->rcvbuf_size; 850b411b363SPhilipp Reisner sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 851b411b363SPhilipp Reisner } 852b411b363SPhilipp Reisner 853b411b363SPhilipp Reisner /* NOT YET ... 854b411b363SPhilipp Reisner * sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; 855b411b363SPhilipp Reisner * sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 856b411b363SPhilipp Reisner * first set it to the P_HAND_SHAKE timeout, 857b411b363SPhilipp Reisner * which we set to 4x the configured ping_timeout. */ 858b411b363SPhilipp Reisner sock->sk->sk_sndtimeo = 859b411b363SPhilipp Reisner sock->sk->sk_rcvtimeo = mdev->net_conf->ping_timeo*4*HZ/10; 860b411b363SPhilipp Reisner 861b411b363SPhilipp Reisner msock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; 862b411b363SPhilipp Reisner msock->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ; 863b411b363SPhilipp Reisner 864b411b363SPhilipp Reisner /* we don't want delays. 865b411b363SPhilipp Reisner * we use TCP_CORK where apropriate, though */ 866b411b363SPhilipp Reisner drbd_tcp_nodelay(sock); 867b411b363SPhilipp Reisner drbd_tcp_nodelay(msock); 868b411b363SPhilipp Reisner 869b411b363SPhilipp Reisner mdev->data.socket = sock; 870b411b363SPhilipp Reisner mdev->meta.socket = msock; 871b411b363SPhilipp Reisner mdev->last_received = jiffies; 872b411b363SPhilipp Reisner 873b411b363SPhilipp Reisner D_ASSERT(mdev->asender.task == NULL); 874b411b363SPhilipp Reisner 875b411b363SPhilipp Reisner h = drbd_do_handshake(mdev); 876b411b363SPhilipp Reisner if (h <= 0) 877b411b363SPhilipp Reisner return h; 878b411b363SPhilipp Reisner 879b411b363SPhilipp Reisner if (mdev->cram_hmac_tfm) { 880b411b363SPhilipp Reisner /* drbd_request_state(mdev, NS(conn, WFAuth)); */ 881b10d96cbSJohannes Thoma switch (drbd_do_auth(mdev)) { 882b10d96cbSJohannes Thoma case -1: 883b411b363SPhilipp Reisner dev_err(DEV, "Authentication of peer failed\n"); 884b411b363SPhilipp Reisner return -1; 885b10d96cbSJohannes Thoma case 0: 886b10d96cbSJohannes Thoma dev_err(DEV, "Authentication of peer failed, trying again.\n"); 887b10d96cbSJohannes Thoma return 0; 888b411b363SPhilipp Reisner } 889b411b363SPhilipp Reisner } 890b411b363SPhilipp Reisner 891b411b363SPhilipp Reisner if (drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS)) < SS_SUCCESS) 892b411b363SPhilipp Reisner return 0; 893b411b363SPhilipp Reisner 894b411b363SPhilipp Reisner sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10; 895b411b363SPhilipp Reisner sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 896b411b363SPhilipp Reisner 897b411b363SPhilipp Reisner atomic_set(&mdev->packet_seq, 0); 898b411b363SPhilipp Reisner mdev->peer_seq = 0; 899b411b363SPhilipp Reisner 900b411b363SPhilipp Reisner drbd_thread_start(&mdev->asender); 901b411b363SPhilipp Reisner 902b411b363SPhilipp Reisner drbd_send_protocol(mdev); 903b411b363SPhilipp Reisner drbd_send_sync_param(mdev, &mdev->sync_conf); 904b411b363SPhilipp Reisner drbd_send_sizes(mdev, 0); 905b411b363SPhilipp Reisner drbd_send_uuids(mdev); 906b411b363SPhilipp Reisner drbd_send_state(mdev); 907b411b363SPhilipp Reisner clear_bit(USE_DEGR_WFC_T, &mdev->flags); 908b411b363SPhilipp Reisner clear_bit(RESIZE_PENDING, &mdev->flags); 909b411b363SPhilipp Reisner 910b411b363SPhilipp Reisner return 1; 911b411b363SPhilipp Reisner 912b411b363SPhilipp Reisner out_release_sockets: 913b411b363SPhilipp Reisner if (sock) 914b411b363SPhilipp Reisner sock_release(sock); 915b411b363SPhilipp Reisner if (msock) 916b411b363SPhilipp Reisner sock_release(msock); 917b411b363SPhilipp Reisner return -1; 918b411b363SPhilipp Reisner } 919b411b363SPhilipp Reisner 920b411b363SPhilipp Reisner static int drbd_recv_header(struct drbd_conf *mdev, struct p_header *h) 921b411b363SPhilipp Reisner { 922b411b363SPhilipp Reisner int r; 923b411b363SPhilipp Reisner 924b411b363SPhilipp Reisner r = drbd_recv(mdev, h, sizeof(*h)); 925b411b363SPhilipp Reisner 926b411b363SPhilipp Reisner if (unlikely(r != sizeof(*h))) { 927b411b363SPhilipp Reisner dev_err(DEV, "short read expecting header on sock: r=%d\n", r); 928b411b363SPhilipp Reisner return FALSE; 929b411b363SPhilipp Reisner }; 930b411b363SPhilipp Reisner h->command = be16_to_cpu(h->command); 931b411b363SPhilipp Reisner h->length = be16_to_cpu(h->length); 932b411b363SPhilipp Reisner if (unlikely(h->magic != BE_DRBD_MAGIC)) { 933b411b363SPhilipp Reisner dev_err(DEV, "magic?? on data m: 0x%lx c: %d l: %d\n", 934b411b363SPhilipp Reisner (long)be32_to_cpu(h->magic), 935b411b363SPhilipp Reisner h->command, h->length); 936b411b363SPhilipp Reisner return FALSE; 937b411b363SPhilipp Reisner } 938b411b363SPhilipp Reisner mdev->last_received = jiffies; 939b411b363SPhilipp Reisner 940b411b363SPhilipp Reisner return TRUE; 941b411b363SPhilipp Reisner } 942b411b363SPhilipp Reisner 943b411b363SPhilipp Reisner static enum finish_epoch drbd_flush_after_epoch(struct drbd_conf *mdev, struct drbd_epoch *epoch) 944b411b363SPhilipp Reisner { 945b411b363SPhilipp Reisner int rv; 946b411b363SPhilipp Reisner 947b411b363SPhilipp Reisner if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) { 948b411b363SPhilipp Reisner rv = blkdev_issue_flush(mdev->ldev->backing_bdev, NULL); 949b411b363SPhilipp Reisner if (rv) { 950b411b363SPhilipp Reisner dev_err(DEV, "local disk flush failed with status %d\n", rv); 951b411b363SPhilipp Reisner /* would rather check on EOPNOTSUPP, but that is not reliable. 952b411b363SPhilipp Reisner * don't try again for ANY return value != 0 953b411b363SPhilipp Reisner * if (rv == -EOPNOTSUPP) */ 954b411b363SPhilipp Reisner drbd_bump_write_ordering(mdev, WO_drain_io); 955b411b363SPhilipp Reisner } 956b411b363SPhilipp Reisner put_ldev(mdev); 957b411b363SPhilipp Reisner } 958b411b363SPhilipp Reisner 959b411b363SPhilipp Reisner return drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE); 960b411b363SPhilipp Reisner } 961b411b363SPhilipp Reisner 962b411b363SPhilipp Reisner static int w_flush(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 963b411b363SPhilipp Reisner { 964b411b363SPhilipp Reisner struct flush_work *fw = (struct flush_work *)w; 965b411b363SPhilipp Reisner struct drbd_epoch *epoch = fw->epoch; 966b411b363SPhilipp Reisner 967b411b363SPhilipp Reisner kfree(w); 968b411b363SPhilipp Reisner 969b411b363SPhilipp Reisner if (!test_and_set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags)) 970b411b363SPhilipp Reisner drbd_flush_after_epoch(mdev, epoch); 971b411b363SPhilipp Reisner 972b411b363SPhilipp Reisner drbd_may_finish_epoch(mdev, epoch, EV_PUT | 973b411b363SPhilipp Reisner (mdev->state.conn < C_CONNECTED ? EV_CLEANUP : 0)); 974b411b363SPhilipp Reisner 975b411b363SPhilipp Reisner return 1; 976b411b363SPhilipp Reisner } 977b411b363SPhilipp Reisner 978b411b363SPhilipp Reisner /** 979b411b363SPhilipp Reisner * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. 980b411b363SPhilipp Reisner * @mdev: DRBD device. 981b411b363SPhilipp Reisner * @epoch: Epoch object. 982b411b363SPhilipp Reisner * @ev: Epoch event. 983b411b363SPhilipp Reisner */ 984b411b363SPhilipp Reisner static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev, 985b411b363SPhilipp Reisner struct drbd_epoch *epoch, 986b411b363SPhilipp Reisner enum epoch_event ev) 987b411b363SPhilipp Reisner { 988b411b363SPhilipp Reisner int finish, epoch_size; 989b411b363SPhilipp Reisner struct drbd_epoch *next_epoch; 990b411b363SPhilipp Reisner int schedule_flush = 0; 991b411b363SPhilipp Reisner enum finish_epoch rv = FE_STILL_LIVE; 992b411b363SPhilipp Reisner 993b411b363SPhilipp Reisner spin_lock(&mdev->epoch_lock); 994b411b363SPhilipp Reisner do { 995b411b363SPhilipp Reisner next_epoch = NULL; 996b411b363SPhilipp Reisner finish = 0; 997b411b363SPhilipp Reisner 998b411b363SPhilipp Reisner epoch_size = atomic_read(&epoch->epoch_size); 999b411b363SPhilipp Reisner 1000b411b363SPhilipp Reisner switch (ev & ~EV_CLEANUP) { 1001b411b363SPhilipp Reisner case EV_PUT: 1002b411b363SPhilipp Reisner atomic_dec(&epoch->active); 1003b411b363SPhilipp Reisner break; 1004b411b363SPhilipp Reisner case EV_GOT_BARRIER_NR: 1005b411b363SPhilipp Reisner set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); 1006b411b363SPhilipp Reisner 1007b411b363SPhilipp Reisner /* Special case: If we just switched from WO_bio_barrier to 1008b411b363SPhilipp Reisner WO_bdev_flush we should not finish the current epoch */ 1009b411b363SPhilipp Reisner if (test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags) && epoch_size == 1 && 1010b411b363SPhilipp Reisner mdev->write_ordering != WO_bio_barrier && 1011b411b363SPhilipp Reisner epoch == mdev->current_epoch) 1012b411b363SPhilipp Reisner clear_bit(DE_CONTAINS_A_BARRIER, &epoch->flags); 1013b411b363SPhilipp Reisner break; 1014b411b363SPhilipp Reisner case EV_BARRIER_DONE: 1015b411b363SPhilipp Reisner set_bit(DE_BARRIER_IN_NEXT_EPOCH_DONE, &epoch->flags); 1016b411b363SPhilipp Reisner break; 1017b411b363SPhilipp Reisner case EV_BECAME_LAST: 1018b411b363SPhilipp Reisner /* nothing to do*/ 1019b411b363SPhilipp Reisner break; 1020b411b363SPhilipp Reisner } 1021b411b363SPhilipp Reisner 1022b411b363SPhilipp Reisner if (epoch_size != 0 && 1023b411b363SPhilipp Reisner atomic_read(&epoch->active) == 0 && 1024b411b363SPhilipp Reisner test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) && 1025b411b363SPhilipp Reisner epoch->list.prev == &mdev->current_epoch->list && 1026b411b363SPhilipp Reisner !test_bit(DE_IS_FINISHING, &epoch->flags)) { 1027b411b363SPhilipp Reisner /* Nearly all conditions are met to finish that epoch... */ 1028b411b363SPhilipp Reisner if (test_bit(DE_BARRIER_IN_NEXT_EPOCH_DONE, &epoch->flags) || 1029b411b363SPhilipp Reisner mdev->write_ordering == WO_none || 1030b411b363SPhilipp Reisner (epoch_size == 1 && test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags)) || 1031b411b363SPhilipp Reisner ev & EV_CLEANUP) { 1032b411b363SPhilipp Reisner finish = 1; 1033b411b363SPhilipp Reisner set_bit(DE_IS_FINISHING, &epoch->flags); 1034b411b363SPhilipp Reisner } else if (!test_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags) && 1035b411b363SPhilipp Reisner mdev->write_ordering == WO_bio_barrier) { 1036b411b363SPhilipp Reisner atomic_inc(&epoch->active); 1037b411b363SPhilipp Reisner schedule_flush = 1; 1038b411b363SPhilipp Reisner } 1039b411b363SPhilipp Reisner } 1040b411b363SPhilipp Reisner if (finish) { 1041b411b363SPhilipp Reisner if (!(ev & EV_CLEANUP)) { 1042b411b363SPhilipp Reisner spin_unlock(&mdev->epoch_lock); 1043b411b363SPhilipp Reisner drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size); 1044b411b363SPhilipp Reisner spin_lock(&mdev->epoch_lock); 1045b411b363SPhilipp Reisner } 1046b411b363SPhilipp Reisner dec_unacked(mdev); 1047b411b363SPhilipp Reisner 1048b411b363SPhilipp Reisner if (mdev->current_epoch != epoch) { 1049b411b363SPhilipp Reisner next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); 1050b411b363SPhilipp Reisner list_del(&epoch->list); 1051b411b363SPhilipp Reisner ev = EV_BECAME_LAST | (ev & EV_CLEANUP); 1052b411b363SPhilipp Reisner mdev->epochs--; 1053b411b363SPhilipp Reisner kfree(epoch); 1054b411b363SPhilipp Reisner 1055b411b363SPhilipp Reisner if (rv == FE_STILL_LIVE) 1056b411b363SPhilipp Reisner rv = FE_DESTROYED; 1057b411b363SPhilipp Reisner } else { 1058b411b363SPhilipp Reisner epoch->flags = 0; 1059b411b363SPhilipp Reisner atomic_set(&epoch->epoch_size, 0); 1060b411b363SPhilipp Reisner /* atomic_set(&epoch->active, 0); is alrady zero */ 1061b411b363SPhilipp Reisner if (rv == FE_STILL_LIVE) 1062b411b363SPhilipp Reisner rv = FE_RECYCLED; 1063b411b363SPhilipp Reisner } 1064b411b363SPhilipp Reisner } 1065b411b363SPhilipp Reisner 1066b411b363SPhilipp Reisner if (!next_epoch) 1067b411b363SPhilipp Reisner break; 1068b411b363SPhilipp Reisner 1069b411b363SPhilipp Reisner epoch = next_epoch; 1070b411b363SPhilipp Reisner } while (1); 1071b411b363SPhilipp Reisner 1072b411b363SPhilipp Reisner spin_unlock(&mdev->epoch_lock); 1073b411b363SPhilipp Reisner 1074b411b363SPhilipp Reisner if (schedule_flush) { 1075b411b363SPhilipp Reisner struct flush_work *fw; 1076b411b363SPhilipp Reisner fw = kmalloc(sizeof(*fw), GFP_ATOMIC); 1077b411b363SPhilipp Reisner if (fw) { 1078b411b363SPhilipp Reisner fw->w.cb = w_flush; 1079b411b363SPhilipp Reisner fw->epoch = epoch; 1080b411b363SPhilipp Reisner drbd_queue_work(&mdev->data.work, &fw->w); 1081b411b363SPhilipp Reisner } else { 1082b411b363SPhilipp Reisner dev_warn(DEV, "Could not kmalloc a flush_work obj\n"); 1083b411b363SPhilipp Reisner set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags); 1084b411b363SPhilipp Reisner /* That is not a recursion, only one level */ 1085b411b363SPhilipp Reisner drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE); 1086b411b363SPhilipp Reisner drbd_may_finish_epoch(mdev, epoch, EV_PUT); 1087b411b363SPhilipp Reisner } 1088b411b363SPhilipp Reisner } 1089b411b363SPhilipp Reisner 1090b411b363SPhilipp Reisner return rv; 1091b411b363SPhilipp Reisner } 1092b411b363SPhilipp Reisner 1093b411b363SPhilipp Reisner /** 1094b411b363SPhilipp Reisner * drbd_bump_write_ordering() - Fall back to an other write ordering method 1095b411b363SPhilipp Reisner * @mdev: DRBD device. 1096b411b363SPhilipp Reisner * @wo: Write ordering method to try. 1097b411b363SPhilipp Reisner */ 1098b411b363SPhilipp Reisner void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local) 1099b411b363SPhilipp Reisner { 1100b411b363SPhilipp Reisner enum write_ordering_e pwo; 1101b411b363SPhilipp Reisner static char *write_ordering_str[] = { 1102b411b363SPhilipp Reisner [WO_none] = "none", 1103b411b363SPhilipp Reisner [WO_drain_io] = "drain", 1104b411b363SPhilipp Reisner [WO_bdev_flush] = "flush", 1105b411b363SPhilipp Reisner [WO_bio_barrier] = "barrier", 1106b411b363SPhilipp Reisner }; 1107b411b363SPhilipp Reisner 1108b411b363SPhilipp Reisner pwo = mdev->write_ordering; 1109b411b363SPhilipp Reisner wo = min(pwo, wo); 1110b411b363SPhilipp Reisner if (wo == WO_bio_barrier && mdev->ldev->dc.no_disk_barrier) 1111b411b363SPhilipp Reisner wo = WO_bdev_flush; 1112b411b363SPhilipp Reisner if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush) 1113b411b363SPhilipp Reisner wo = WO_drain_io; 1114b411b363SPhilipp Reisner if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain) 1115b411b363SPhilipp Reisner wo = WO_none; 1116b411b363SPhilipp Reisner mdev->write_ordering = wo; 1117b411b363SPhilipp Reisner if (pwo != mdev->write_ordering || wo == WO_bio_barrier) 1118b411b363SPhilipp Reisner dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]); 1119b411b363SPhilipp Reisner } 1120b411b363SPhilipp Reisner 1121b411b363SPhilipp Reisner /** 1122b411b363SPhilipp Reisner * w_e_reissue() - Worker callback; Resubmit a bio, without BIO_RW_BARRIER set 1123b411b363SPhilipp Reisner * @mdev: DRBD device. 1124b411b363SPhilipp Reisner * @w: work object. 1125b411b363SPhilipp Reisner * @cancel: The connection will be closed anyways (unused in this callback) 1126b411b363SPhilipp Reisner */ 1127b411b363SPhilipp Reisner int w_e_reissue(struct drbd_conf *mdev, struct drbd_work *w, int cancel) __releases(local) 1128b411b363SPhilipp Reisner { 1129b411b363SPhilipp Reisner struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; 1130b411b363SPhilipp Reisner struct bio *bio = e->private_bio; 1131b411b363SPhilipp Reisner 1132b411b363SPhilipp Reisner /* We leave DE_CONTAINS_A_BARRIER and EE_IS_BARRIER in place, 1133b411b363SPhilipp Reisner (and DE_BARRIER_IN_NEXT_EPOCH_ISSUED in the previous Epoch) 1134b411b363SPhilipp Reisner so that we can finish that epoch in drbd_may_finish_epoch(). 1135b411b363SPhilipp Reisner That is necessary if we already have a long chain of Epochs, before 1136b411b363SPhilipp Reisner we realize that BIO_RW_BARRIER is actually not supported */ 1137b411b363SPhilipp Reisner 1138b411b363SPhilipp Reisner /* As long as the -ENOTSUPP on the barrier is reported immediately 1139b411b363SPhilipp Reisner that will never trigger. If it is reported late, we will just 1140b411b363SPhilipp Reisner print that warning and continue correctly for all future requests 1141b411b363SPhilipp Reisner with WO_bdev_flush */ 1142b411b363SPhilipp Reisner if (previous_epoch(mdev, e->epoch)) 1143b411b363SPhilipp Reisner dev_warn(DEV, "Write ordering was not enforced (one time event)\n"); 1144b411b363SPhilipp Reisner 1145b411b363SPhilipp Reisner /* prepare bio for re-submit, 1146b411b363SPhilipp Reisner * re-init volatile members */ 1147b411b363SPhilipp Reisner /* we still have a local reference, 1148b411b363SPhilipp Reisner * get_ldev was done in receive_Data. */ 1149b411b363SPhilipp Reisner bio->bi_bdev = mdev->ldev->backing_bdev; 1150b411b363SPhilipp Reisner bio->bi_sector = e->sector; 1151b411b363SPhilipp Reisner bio->bi_size = e->size; 1152b411b363SPhilipp Reisner bio->bi_idx = 0; 1153b411b363SPhilipp Reisner 1154b411b363SPhilipp Reisner bio->bi_flags &= ~(BIO_POOL_MASK - 1); 1155b411b363SPhilipp Reisner bio->bi_flags |= 1 << BIO_UPTODATE; 1156b411b363SPhilipp Reisner 1157b411b363SPhilipp Reisner /* don't know whether this is necessary: */ 1158b411b363SPhilipp Reisner bio->bi_phys_segments = 0; 1159b411b363SPhilipp Reisner bio->bi_next = NULL; 1160b411b363SPhilipp Reisner 1161b411b363SPhilipp Reisner /* these should be unchanged: */ 1162b411b363SPhilipp Reisner /* bio->bi_end_io = drbd_endio_write_sec; */ 1163b411b363SPhilipp Reisner /* bio->bi_vcnt = whatever; */ 1164b411b363SPhilipp Reisner 1165b411b363SPhilipp Reisner e->w.cb = e_end_block; 1166b411b363SPhilipp Reisner 1167b411b363SPhilipp Reisner /* This is no longer a barrier request. */ 1168b411b363SPhilipp Reisner bio->bi_rw &= ~(1UL << BIO_RW_BARRIER); 1169b411b363SPhilipp Reisner 1170b411b363SPhilipp Reisner drbd_generic_make_request(mdev, DRBD_FAULT_DT_WR, bio); 1171b411b363SPhilipp Reisner 1172b411b363SPhilipp Reisner return 1; 1173b411b363SPhilipp Reisner } 1174b411b363SPhilipp Reisner 1175b411b363SPhilipp Reisner static int receive_Barrier(struct drbd_conf *mdev, struct p_header *h) 1176b411b363SPhilipp Reisner { 1177b411b363SPhilipp Reisner int rv, issue_flush; 1178b411b363SPhilipp Reisner struct p_barrier *p = (struct p_barrier *)h; 1179b411b363SPhilipp Reisner struct drbd_epoch *epoch; 1180b411b363SPhilipp Reisner 1181b411b363SPhilipp Reisner ERR_IF(h->length != (sizeof(*p)-sizeof(*h))) return FALSE; 1182b411b363SPhilipp Reisner 1183b411b363SPhilipp Reisner rv = drbd_recv(mdev, h->payload, h->length); 1184b411b363SPhilipp Reisner ERR_IF(rv != h->length) return FALSE; 1185b411b363SPhilipp Reisner 1186b411b363SPhilipp Reisner inc_unacked(mdev); 1187b411b363SPhilipp Reisner 1188b411b363SPhilipp Reisner if (mdev->net_conf->wire_protocol != DRBD_PROT_C) 1189b411b363SPhilipp Reisner drbd_kick_lo(mdev); 1190b411b363SPhilipp Reisner 1191b411b363SPhilipp Reisner mdev->current_epoch->barrier_nr = p->barrier; 1192b411b363SPhilipp Reisner rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR); 1193b411b363SPhilipp Reisner 1194b411b363SPhilipp Reisner /* P_BARRIER_ACK may imply that the corresponding extent is dropped from 1195b411b363SPhilipp Reisner * the activity log, which means it would not be resynced in case the 1196b411b363SPhilipp Reisner * R_PRIMARY crashes now. 1197b411b363SPhilipp Reisner * Therefore we must send the barrier_ack after the barrier request was 1198b411b363SPhilipp Reisner * completed. */ 1199b411b363SPhilipp Reisner switch (mdev->write_ordering) { 1200b411b363SPhilipp Reisner case WO_bio_barrier: 1201b411b363SPhilipp Reisner case WO_none: 1202b411b363SPhilipp Reisner if (rv == FE_RECYCLED) 1203b411b363SPhilipp Reisner return TRUE; 1204b411b363SPhilipp Reisner break; 1205b411b363SPhilipp Reisner 1206b411b363SPhilipp Reisner case WO_bdev_flush: 1207b411b363SPhilipp Reisner case WO_drain_io: 1208367a8d73SPhilipp Reisner if (rv == FE_STILL_LIVE) { 1209b411b363SPhilipp Reisner set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &mdev->current_epoch->flags); 1210b411b363SPhilipp Reisner drbd_wait_ee_list_empty(mdev, &mdev->active_ee); 1211b411b363SPhilipp Reisner rv = drbd_flush_after_epoch(mdev, mdev->current_epoch); 1212367a8d73SPhilipp Reisner } 1213b411b363SPhilipp Reisner if (rv == FE_RECYCLED) 1214b411b363SPhilipp Reisner return TRUE; 1215b411b363SPhilipp Reisner 1216b411b363SPhilipp Reisner /* The asender will send all the ACKs and barrier ACKs out, since 1217b411b363SPhilipp Reisner all EEs moved from the active_ee to the done_ee. We need to 1218b411b363SPhilipp Reisner provide a new epoch object for the EEs that come in soon */ 1219b411b363SPhilipp Reisner break; 1220b411b363SPhilipp Reisner } 1221b411b363SPhilipp Reisner 1222b411b363SPhilipp Reisner /* receiver context, in the writeout path of the other node. 1223b411b363SPhilipp Reisner * avoid potential distributed deadlock */ 1224b411b363SPhilipp Reisner epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); 1225b411b363SPhilipp Reisner if (!epoch) { 1226b411b363SPhilipp Reisner dev_warn(DEV, "Allocation of an epoch failed, slowing down\n"); 1227d3db7b48SDan Carpenter issue_flush = !test_and_set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &mdev->current_epoch->flags); 1228b411b363SPhilipp Reisner drbd_wait_ee_list_empty(mdev, &mdev->active_ee); 1229b411b363SPhilipp Reisner if (issue_flush) { 1230b411b363SPhilipp Reisner rv = drbd_flush_after_epoch(mdev, mdev->current_epoch); 1231b411b363SPhilipp Reisner if (rv == FE_RECYCLED) 1232b411b363SPhilipp Reisner return TRUE; 1233b411b363SPhilipp Reisner } 1234b411b363SPhilipp Reisner 1235b411b363SPhilipp Reisner drbd_wait_ee_list_empty(mdev, &mdev->done_ee); 1236b411b363SPhilipp Reisner 1237b411b363SPhilipp Reisner return TRUE; 1238b411b363SPhilipp Reisner } 1239b411b363SPhilipp Reisner 1240b411b363SPhilipp Reisner epoch->flags = 0; 1241b411b363SPhilipp Reisner atomic_set(&epoch->epoch_size, 0); 1242b411b363SPhilipp Reisner atomic_set(&epoch->active, 0); 1243b411b363SPhilipp Reisner 1244b411b363SPhilipp Reisner spin_lock(&mdev->epoch_lock); 1245b411b363SPhilipp Reisner if (atomic_read(&mdev->current_epoch->epoch_size)) { 1246b411b363SPhilipp Reisner list_add(&epoch->list, &mdev->current_epoch->list); 1247b411b363SPhilipp Reisner mdev->current_epoch = epoch; 1248b411b363SPhilipp Reisner mdev->epochs++; 1249b411b363SPhilipp Reisner } else { 1250b411b363SPhilipp Reisner /* The current_epoch got recycled while we allocated this one... */ 1251b411b363SPhilipp Reisner kfree(epoch); 1252b411b363SPhilipp Reisner } 1253b411b363SPhilipp Reisner spin_unlock(&mdev->epoch_lock); 1254b411b363SPhilipp Reisner 1255b411b363SPhilipp Reisner return TRUE; 1256b411b363SPhilipp Reisner } 1257b411b363SPhilipp Reisner 1258b411b363SPhilipp Reisner /* used from receive_RSDataReply (recv_resync_read) 1259b411b363SPhilipp Reisner * and from receive_Data */ 1260b411b363SPhilipp Reisner static struct drbd_epoch_entry * 1261b411b363SPhilipp Reisner read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, int data_size) __must_hold(local) 1262b411b363SPhilipp Reisner { 1263b411b363SPhilipp Reisner struct drbd_epoch_entry *e; 1264b411b363SPhilipp Reisner struct bio_vec *bvec; 1265b411b363SPhilipp Reisner struct page *page; 1266b411b363SPhilipp Reisner struct bio *bio; 1267b411b363SPhilipp Reisner int dgs, ds, i, rr; 1268b411b363SPhilipp Reisner void *dig_in = mdev->int_dig_in; 1269b411b363SPhilipp Reisner void *dig_vv = mdev->int_dig_vv; 1270b411b363SPhilipp Reisner 1271b411b363SPhilipp Reisner dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ? 1272b411b363SPhilipp Reisner crypto_hash_digestsize(mdev->integrity_r_tfm) : 0; 1273b411b363SPhilipp Reisner 1274b411b363SPhilipp Reisner if (dgs) { 1275b411b363SPhilipp Reisner rr = drbd_recv(mdev, dig_in, dgs); 1276b411b363SPhilipp Reisner if (rr != dgs) { 1277b411b363SPhilipp Reisner dev_warn(DEV, "short read receiving data digest: read %d expected %d\n", 1278b411b363SPhilipp Reisner rr, dgs); 1279b411b363SPhilipp Reisner return NULL; 1280b411b363SPhilipp Reisner } 1281b411b363SPhilipp Reisner } 1282b411b363SPhilipp Reisner 1283b411b363SPhilipp Reisner data_size -= dgs; 1284b411b363SPhilipp Reisner 1285b411b363SPhilipp Reisner ERR_IF(data_size & 0x1ff) return NULL; 1286b411b363SPhilipp Reisner ERR_IF(data_size > DRBD_MAX_SEGMENT_SIZE) return NULL; 1287b411b363SPhilipp Reisner 1288b411b363SPhilipp Reisner /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD 1289b411b363SPhilipp Reisner * "criss-cross" setup, that might cause write-out on some other DRBD, 1290b411b363SPhilipp Reisner * which in turn might block on the other node at this very place. */ 1291b411b363SPhilipp Reisner e = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO); 1292b411b363SPhilipp Reisner if (!e) 1293b411b363SPhilipp Reisner return NULL; 1294b411b363SPhilipp Reisner bio = e->private_bio; 1295b411b363SPhilipp Reisner ds = data_size; 1296b411b363SPhilipp Reisner bio_for_each_segment(bvec, bio, i) { 1297b411b363SPhilipp Reisner page = bvec->bv_page; 1298b411b363SPhilipp Reisner rr = drbd_recv(mdev, kmap(page), min_t(int, ds, PAGE_SIZE)); 1299b411b363SPhilipp Reisner kunmap(page); 1300b411b363SPhilipp Reisner if (rr != min_t(int, ds, PAGE_SIZE)) { 1301b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 1302b411b363SPhilipp Reisner dev_warn(DEV, "short read receiving data: read %d expected %d\n", 1303b411b363SPhilipp Reisner rr, min_t(int, ds, PAGE_SIZE)); 1304b411b363SPhilipp Reisner return NULL; 1305b411b363SPhilipp Reisner } 1306b411b363SPhilipp Reisner ds -= rr; 1307b411b363SPhilipp Reisner } 1308b411b363SPhilipp Reisner 1309b411b363SPhilipp Reisner if (dgs) { 1310b411b363SPhilipp Reisner drbd_csum(mdev, mdev->integrity_r_tfm, bio, dig_vv); 1311b411b363SPhilipp Reisner if (memcmp(dig_in, dig_vv, dgs)) { 1312b411b363SPhilipp Reisner dev_err(DEV, "Digest integrity check FAILED.\n"); 1313b411b363SPhilipp Reisner drbd_bcast_ee(mdev, "digest failed", 1314b411b363SPhilipp Reisner dgs, dig_in, dig_vv, e); 1315b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 1316b411b363SPhilipp Reisner return NULL; 1317b411b363SPhilipp Reisner } 1318b411b363SPhilipp Reisner } 1319b411b363SPhilipp Reisner mdev->recv_cnt += data_size>>9; 1320b411b363SPhilipp Reisner return e; 1321b411b363SPhilipp Reisner } 1322b411b363SPhilipp Reisner 1323b411b363SPhilipp Reisner /* drbd_drain_block() just takes a data block 1324b411b363SPhilipp Reisner * out of the socket input buffer, and discards it. 1325b411b363SPhilipp Reisner */ 1326b411b363SPhilipp Reisner static int drbd_drain_block(struct drbd_conf *mdev, int data_size) 1327b411b363SPhilipp Reisner { 1328b411b363SPhilipp Reisner struct page *page; 1329b411b363SPhilipp Reisner int rr, rv = 1; 1330b411b363SPhilipp Reisner void *data; 1331b411b363SPhilipp Reisner 1332b411b363SPhilipp Reisner page = drbd_pp_alloc(mdev, 1); 1333b411b363SPhilipp Reisner 1334b411b363SPhilipp Reisner data = kmap(page); 1335b411b363SPhilipp Reisner while (data_size) { 1336b411b363SPhilipp Reisner rr = drbd_recv(mdev, data, min_t(int, data_size, PAGE_SIZE)); 1337b411b363SPhilipp Reisner if (rr != min_t(int, data_size, PAGE_SIZE)) { 1338b411b363SPhilipp Reisner rv = 0; 1339b411b363SPhilipp Reisner dev_warn(DEV, "short read receiving data: read %d expected %d\n", 1340b411b363SPhilipp Reisner rr, min_t(int, data_size, PAGE_SIZE)); 1341b411b363SPhilipp Reisner break; 1342b411b363SPhilipp Reisner } 1343b411b363SPhilipp Reisner data_size -= rr; 1344b411b363SPhilipp Reisner } 1345b411b363SPhilipp Reisner kunmap(page); 1346b411b363SPhilipp Reisner drbd_pp_free(mdev, page); 1347b411b363SPhilipp Reisner return rv; 1348b411b363SPhilipp Reisner } 1349b411b363SPhilipp Reisner 1350b411b363SPhilipp Reisner static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req, 1351b411b363SPhilipp Reisner sector_t sector, int data_size) 1352b411b363SPhilipp Reisner { 1353b411b363SPhilipp Reisner struct bio_vec *bvec; 1354b411b363SPhilipp Reisner struct bio *bio; 1355b411b363SPhilipp Reisner int dgs, rr, i, expect; 1356b411b363SPhilipp Reisner void *dig_in = mdev->int_dig_in; 1357b411b363SPhilipp Reisner void *dig_vv = mdev->int_dig_vv; 1358b411b363SPhilipp Reisner 1359b411b363SPhilipp Reisner dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ? 1360b411b363SPhilipp Reisner crypto_hash_digestsize(mdev->integrity_r_tfm) : 0; 1361b411b363SPhilipp Reisner 1362b411b363SPhilipp Reisner if (dgs) { 1363b411b363SPhilipp Reisner rr = drbd_recv(mdev, dig_in, dgs); 1364b411b363SPhilipp Reisner if (rr != dgs) { 1365b411b363SPhilipp Reisner dev_warn(DEV, "short read receiving data reply digest: read %d expected %d\n", 1366b411b363SPhilipp Reisner rr, dgs); 1367b411b363SPhilipp Reisner return 0; 1368b411b363SPhilipp Reisner } 1369b411b363SPhilipp Reisner } 1370b411b363SPhilipp Reisner 1371b411b363SPhilipp Reisner data_size -= dgs; 1372b411b363SPhilipp Reisner 1373b411b363SPhilipp Reisner /* optimistically update recv_cnt. if receiving fails below, 1374b411b363SPhilipp Reisner * we disconnect anyways, and counters will be reset. */ 1375b411b363SPhilipp Reisner mdev->recv_cnt += data_size>>9; 1376b411b363SPhilipp Reisner 1377b411b363SPhilipp Reisner bio = req->master_bio; 1378b411b363SPhilipp Reisner D_ASSERT(sector == bio->bi_sector); 1379b411b363SPhilipp Reisner 1380b411b363SPhilipp Reisner bio_for_each_segment(bvec, bio, i) { 1381b411b363SPhilipp Reisner expect = min_t(int, data_size, bvec->bv_len); 1382b411b363SPhilipp Reisner rr = drbd_recv(mdev, 1383b411b363SPhilipp Reisner kmap(bvec->bv_page)+bvec->bv_offset, 1384b411b363SPhilipp Reisner expect); 1385b411b363SPhilipp Reisner kunmap(bvec->bv_page); 1386b411b363SPhilipp Reisner if (rr != expect) { 1387b411b363SPhilipp Reisner dev_warn(DEV, "short read receiving data reply: " 1388b411b363SPhilipp Reisner "read %d expected %d\n", 1389b411b363SPhilipp Reisner rr, expect); 1390b411b363SPhilipp Reisner return 0; 1391b411b363SPhilipp Reisner } 1392b411b363SPhilipp Reisner data_size -= rr; 1393b411b363SPhilipp Reisner } 1394b411b363SPhilipp Reisner 1395b411b363SPhilipp Reisner if (dgs) { 1396b411b363SPhilipp Reisner drbd_csum(mdev, mdev->integrity_r_tfm, bio, dig_vv); 1397b411b363SPhilipp Reisner if (memcmp(dig_in, dig_vv, dgs)) { 1398b411b363SPhilipp Reisner dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n"); 1399b411b363SPhilipp Reisner return 0; 1400b411b363SPhilipp Reisner } 1401b411b363SPhilipp Reisner } 1402b411b363SPhilipp Reisner 1403b411b363SPhilipp Reisner D_ASSERT(data_size == 0); 1404b411b363SPhilipp Reisner return 1; 1405b411b363SPhilipp Reisner } 1406b411b363SPhilipp Reisner 1407b411b363SPhilipp Reisner /* e_end_resync_block() is called via 1408b411b363SPhilipp Reisner * drbd_process_done_ee() by asender only */ 1409b411b363SPhilipp Reisner static int e_end_resync_block(struct drbd_conf *mdev, struct drbd_work *w, int unused) 1410b411b363SPhilipp Reisner { 1411b411b363SPhilipp Reisner struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; 1412b411b363SPhilipp Reisner sector_t sector = e->sector; 1413b411b363SPhilipp Reisner int ok; 1414b411b363SPhilipp Reisner 1415b411b363SPhilipp Reisner D_ASSERT(hlist_unhashed(&e->colision)); 1416b411b363SPhilipp Reisner 1417b411b363SPhilipp Reisner if (likely(drbd_bio_uptodate(e->private_bio))) { 1418b411b363SPhilipp Reisner drbd_set_in_sync(mdev, sector, e->size); 1419b411b363SPhilipp Reisner ok = drbd_send_ack(mdev, P_RS_WRITE_ACK, e); 1420b411b363SPhilipp Reisner } else { 1421b411b363SPhilipp Reisner /* Record failure to sync */ 1422b411b363SPhilipp Reisner drbd_rs_failed_io(mdev, sector, e->size); 1423b411b363SPhilipp Reisner 1424b411b363SPhilipp Reisner ok = drbd_send_ack(mdev, P_NEG_ACK, e); 1425b411b363SPhilipp Reisner } 1426b411b363SPhilipp Reisner dec_unacked(mdev); 1427b411b363SPhilipp Reisner 1428b411b363SPhilipp Reisner return ok; 1429b411b363SPhilipp Reisner } 1430b411b363SPhilipp Reisner 1431b411b363SPhilipp Reisner static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local) 1432b411b363SPhilipp Reisner { 1433b411b363SPhilipp Reisner struct drbd_epoch_entry *e; 1434b411b363SPhilipp Reisner 1435b411b363SPhilipp Reisner e = read_in_block(mdev, ID_SYNCER, sector, data_size); 1436b411b363SPhilipp Reisner if (!e) { 1437b411b363SPhilipp Reisner put_ldev(mdev); 1438b411b363SPhilipp Reisner return FALSE; 1439b411b363SPhilipp Reisner } 1440b411b363SPhilipp Reisner 1441b411b363SPhilipp Reisner dec_rs_pending(mdev); 1442b411b363SPhilipp Reisner 1443b411b363SPhilipp Reisner e->private_bio->bi_end_io = drbd_endio_write_sec; 1444b411b363SPhilipp Reisner e->private_bio->bi_rw = WRITE; 1445b411b363SPhilipp Reisner e->w.cb = e_end_resync_block; 1446b411b363SPhilipp Reisner 1447b411b363SPhilipp Reisner inc_unacked(mdev); 1448b411b363SPhilipp Reisner /* corresponding dec_unacked() in e_end_resync_block() 1449b411b363SPhilipp Reisner * respective _drbd_clear_done_ee */ 1450b411b363SPhilipp Reisner 1451b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 1452b411b363SPhilipp Reisner list_add(&e->w.list, &mdev->sync_ee); 1453b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1454b411b363SPhilipp Reisner 1455b411b363SPhilipp Reisner drbd_generic_make_request(mdev, DRBD_FAULT_RS_WR, e->private_bio); 1456b411b363SPhilipp Reisner /* accounting done in endio */ 1457b411b363SPhilipp Reisner 1458b411b363SPhilipp Reisner maybe_kick_lo(mdev); 1459b411b363SPhilipp Reisner return TRUE; 1460b411b363SPhilipp Reisner } 1461b411b363SPhilipp Reisner 1462b411b363SPhilipp Reisner static int receive_DataReply(struct drbd_conf *mdev, struct p_header *h) 1463b411b363SPhilipp Reisner { 1464b411b363SPhilipp Reisner struct drbd_request *req; 1465b411b363SPhilipp Reisner sector_t sector; 1466b411b363SPhilipp Reisner unsigned int header_size, data_size; 1467b411b363SPhilipp Reisner int ok; 1468b411b363SPhilipp Reisner struct p_data *p = (struct p_data *)h; 1469b411b363SPhilipp Reisner 1470b411b363SPhilipp Reisner header_size = sizeof(*p) - sizeof(*h); 1471b411b363SPhilipp Reisner data_size = h->length - header_size; 1472b411b363SPhilipp Reisner 1473b411b363SPhilipp Reisner ERR_IF(data_size == 0) return FALSE; 1474b411b363SPhilipp Reisner 1475b411b363SPhilipp Reisner if (drbd_recv(mdev, h->payload, header_size) != header_size) 1476b411b363SPhilipp Reisner return FALSE; 1477b411b363SPhilipp Reisner 1478b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 1479b411b363SPhilipp Reisner 1480b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 1481b411b363SPhilipp Reisner req = _ar_id_to_req(mdev, p->block_id, sector); 1482b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1483b411b363SPhilipp Reisner if (unlikely(!req)) { 1484b411b363SPhilipp Reisner dev_err(DEV, "Got a corrupt block_id/sector pair(1).\n"); 1485b411b363SPhilipp Reisner return FALSE; 1486b411b363SPhilipp Reisner } 1487b411b363SPhilipp Reisner 1488b411b363SPhilipp Reisner /* hlist_del(&req->colision) is done in _req_may_be_done, to avoid 1489b411b363SPhilipp Reisner * special casing it there for the various failure cases. 1490b411b363SPhilipp Reisner * still no race with drbd_fail_pending_reads */ 1491b411b363SPhilipp Reisner ok = recv_dless_read(mdev, req, sector, data_size); 1492b411b363SPhilipp Reisner 1493b411b363SPhilipp Reisner if (ok) 1494b411b363SPhilipp Reisner req_mod(req, data_received); 1495b411b363SPhilipp Reisner /* else: nothing. handled from drbd_disconnect... 1496b411b363SPhilipp Reisner * I don't think we may complete this just yet 1497b411b363SPhilipp Reisner * in case we are "on-disconnect: freeze" */ 1498b411b363SPhilipp Reisner 1499b411b363SPhilipp Reisner return ok; 1500b411b363SPhilipp Reisner } 1501b411b363SPhilipp Reisner 1502b411b363SPhilipp Reisner static int receive_RSDataReply(struct drbd_conf *mdev, struct p_header *h) 1503b411b363SPhilipp Reisner { 1504b411b363SPhilipp Reisner sector_t sector; 1505b411b363SPhilipp Reisner unsigned int header_size, data_size; 1506b411b363SPhilipp Reisner int ok; 1507b411b363SPhilipp Reisner struct p_data *p = (struct p_data *)h; 1508b411b363SPhilipp Reisner 1509b411b363SPhilipp Reisner header_size = sizeof(*p) - sizeof(*h); 1510b411b363SPhilipp Reisner data_size = h->length - header_size; 1511b411b363SPhilipp Reisner 1512b411b363SPhilipp Reisner ERR_IF(data_size == 0) return FALSE; 1513b411b363SPhilipp Reisner 1514b411b363SPhilipp Reisner if (drbd_recv(mdev, h->payload, header_size) != header_size) 1515b411b363SPhilipp Reisner return FALSE; 1516b411b363SPhilipp Reisner 1517b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 1518b411b363SPhilipp Reisner D_ASSERT(p->block_id == ID_SYNCER); 1519b411b363SPhilipp Reisner 1520b411b363SPhilipp Reisner if (get_ldev(mdev)) { 1521b411b363SPhilipp Reisner /* data is submitted to disk within recv_resync_read. 1522b411b363SPhilipp Reisner * corresponding put_ldev done below on error, 1523b411b363SPhilipp Reisner * or in drbd_endio_write_sec. */ 1524b411b363SPhilipp Reisner ok = recv_resync_read(mdev, sector, data_size); 1525b411b363SPhilipp Reisner } else { 1526b411b363SPhilipp Reisner if (__ratelimit(&drbd_ratelimit_state)) 1527b411b363SPhilipp Reisner dev_err(DEV, "Can not write resync data to local disk.\n"); 1528b411b363SPhilipp Reisner 1529b411b363SPhilipp Reisner ok = drbd_drain_block(mdev, data_size); 1530b411b363SPhilipp Reisner 1531b411b363SPhilipp Reisner drbd_send_ack_dp(mdev, P_NEG_ACK, p); 1532b411b363SPhilipp Reisner } 1533b411b363SPhilipp Reisner 1534b411b363SPhilipp Reisner return ok; 1535b411b363SPhilipp Reisner } 1536b411b363SPhilipp Reisner 1537b411b363SPhilipp Reisner /* e_end_block() is called via drbd_process_done_ee(). 1538b411b363SPhilipp Reisner * this means this function only runs in the asender thread 1539b411b363SPhilipp Reisner */ 1540b411b363SPhilipp Reisner static int e_end_block(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1541b411b363SPhilipp Reisner { 1542b411b363SPhilipp Reisner struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; 1543b411b363SPhilipp Reisner sector_t sector = e->sector; 1544b411b363SPhilipp Reisner struct drbd_epoch *epoch; 1545b411b363SPhilipp Reisner int ok = 1, pcmd; 1546b411b363SPhilipp Reisner 1547b411b363SPhilipp Reisner if (e->flags & EE_IS_BARRIER) { 1548b411b363SPhilipp Reisner epoch = previous_epoch(mdev, e->epoch); 1549b411b363SPhilipp Reisner if (epoch) 1550b411b363SPhilipp Reisner drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE + (cancel ? EV_CLEANUP : 0)); 1551b411b363SPhilipp Reisner } 1552b411b363SPhilipp Reisner 1553b411b363SPhilipp Reisner if (mdev->net_conf->wire_protocol == DRBD_PROT_C) { 1554b411b363SPhilipp Reisner if (likely(drbd_bio_uptodate(e->private_bio))) { 1555b411b363SPhilipp Reisner pcmd = (mdev->state.conn >= C_SYNC_SOURCE && 1556b411b363SPhilipp Reisner mdev->state.conn <= C_PAUSED_SYNC_T && 1557b411b363SPhilipp Reisner e->flags & EE_MAY_SET_IN_SYNC) ? 1558b411b363SPhilipp Reisner P_RS_WRITE_ACK : P_WRITE_ACK; 1559b411b363SPhilipp Reisner ok &= drbd_send_ack(mdev, pcmd, e); 1560b411b363SPhilipp Reisner if (pcmd == P_RS_WRITE_ACK) 1561b411b363SPhilipp Reisner drbd_set_in_sync(mdev, sector, e->size); 1562b411b363SPhilipp Reisner } else { 1563b411b363SPhilipp Reisner ok = drbd_send_ack(mdev, P_NEG_ACK, e); 1564b411b363SPhilipp Reisner /* we expect it to be marked out of sync anyways... 1565b411b363SPhilipp Reisner * maybe assert this? */ 1566b411b363SPhilipp Reisner } 1567b411b363SPhilipp Reisner dec_unacked(mdev); 1568b411b363SPhilipp Reisner } 1569b411b363SPhilipp Reisner /* we delete from the conflict detection hash _after_ we sent out the 1570b411b363SPhilipp Reisner * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ 1571b411b363SPhilipp Reisner if (mdev->net_conf->two_primaries) { 1572b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 1573b411b363SPhilipp Reisner D_ASSERT(!hlist_unhashed(&e->colision)); 1574b411b363SPhilipp Reisner hlist_del_init(&e->colision); 1575b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1576b411b363SPhilipp Reisner } else { 1577b411b363SPhilipp Reisner D_ASSERT(hlist_unhashed(&e->colision)); 1578b411b363SPhilipp Reisner } 1579b411b363SPhilipp Reisner 1580b411b363SPhilipp Reisner drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); 1581b411b363SPhilipp Reisner 1582b411b363SPhilipp Reisner return ok; 1583b411b363SPhilipp Reisner } 1584b411b363SPhilipp Reisner 1585b411b363SPhilipp Reisner static int e_send_discard_ack(struct drbd_conf *mdev, struct drbd_work *w, int unused) 1586b411b363SPhilipp Reisner { 1587b411b363SPhilipp Reisner struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; 1588b411b363SPhilipp Reisner int ok = 1; 1589b411b363SPhilipp Reisner 1590b411b363SPhilipp Reisner D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 1591b411b363SPhilipp Reisner ok = drbd_send_ack(mdev, P_DISCARD_ACK, e); 1592b411b363SPhilipp Reisner 1593b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 1594b411b363SPhilipp Reisner D_ASSERT(!hlist_unhashed(&e->colision)); 1595b411b363SPhilipp Reisner hlist_del_init(&e->colision); 1596b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1597b411b363SPhilipp Reisner 1598b411b363SPhilipp Reisner dec_unacked(mdev); 1599b411b363SPhilipp Reisner 1600b411b363SPhilipp Reisner return ok; 1601b411b363SPhilipp Reisner } 1602b411b363SPhilipp Reisner 1603b411b363SPhilipp Reisner /* Called from receive_Data. 1604b411b363SPhilipp Reisner * Synchronize packets on sock with packets on msock. 1605b411b363SPhilipp Reisner * 1606b411b363SPhilipp Reisner * This is here so even when a P_DATA packet traveling via sock overtook an Ack 1607b411b363SPhilipp Reisner * packet traveling on msock, they are still processed in the order they have 1608b411b363SPhilipp Reisner * been sent. 1609b411b363SPhilipp Reisner * 1610b411b363SPhilipp Reisner * Note: we don't care for Ack packets overtaking P_DATA packets. 1611b411b363SPhilipp Reisner * 1612b411b363SPhilipp Reisner * In case packet_seq is larger than mdev->peer_seq number, there are 1613b411b363SPhilipp Reisner * outstanding packets on the msock. We wait for them to arrive. 1614b411b363SPhilipp Reisner * In case we are the logically next packet, we update mdev->peer_seq 1615b411b363SPhilipp Reisner * ourselves. Correctly handles 32bit wrap around. 1616b411b363SPhilipp Reisner * 1617b411b363SPhilipp Reisner * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, 1618b411b363SPhilipp Reisner * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds 1619b411b363SPhilipp Reisner * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have 1620b411b363SPhilipp Reisner * 1<<9 == 512 seconds aka ages for the 32bit wrap around... 1621b411b363SPhilipp Reisner * 1622b411b363SPhilipp Reisner * returns 0 if we may process the packet, 1623b411b363SPhilipp Reisner * -ERESTARTSYS if we were interrupted (by disconnect signal). */ 1624b411b363SPhilipp Reisner static int drbd_wait_peer_seq(struct drbd_conf *mdev, const u32 packet_seq) 1625b411b363SPhilipp Reisner { 1626b411b363SPhilipp Reisner DEFINE_WAIT(wait); 1627b411b363SPhilipp Reisner unsigned int p_seq; 1628b411b363SPhilipp Reisner long timeout; 1629b411b363SPhilipp Reisner int ret = 0; 1630b411b363SPhilipp Reisner spin_lock(&mdev->peer_seq_lock); 1631b411b363SPhilipp Reisner for (;;) { 1632b411b363SPhilipp Reisner prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE); 1633b411b363SPhilipp Reisner if (seq_le(packet_seq, mdev->peer_seq+1)) 1634b411b363SPhilipp Reisner break; 1635b411b363SPhilipp Reisner if (signal_pending(current)) { 1636b411b363SPhilipp Reisner ret = -ERESTARTSYS; 1637b411b363SPhilipp Reisner break; 1638b411b363SPhilipp Reisner } 1639b411b363SPhilipp Reisner p_seq = mdev->peer_seq; 1640b411b363SPhilipp Reisner spin_unlock(&mdev->peer_seq_lock); 1641b411b363SPhilipp Reisner timeout = schedule_timeout(30*HZ); 1642b411b363SPhilipp Reisner spin_lock(&mdev->peer_seq_lock); 1643b411b363SPhilipp Reisner if (timeout == 0 && p_seq == mdev->peer_seq) { 1644b411b363SPhilipp Reisner ret = -ETIMEDOUT; 1645b411b363SPhilipp Reisner dev_err(DEV, "ASSERT FAILED waited 30 seconds for sequence update, forcing reconnect\n"); 1646b411b363SPhilipp Reisner break; 1647b411b363SPhilipp Reisner } 1648b411b363SPhilipp Reisner } 1649b411b363SPhilipp Reisner finish_wait(&mdev->seq_wait, &wait); 1650b411b363SPhilipp Reisner if (mdev->peer_seq+1 == packet_seq) 1651b411b363SPhilipp Reisner mdev->peer_seq++; 1652b411b363SPhilipp Reisner spin_unlock(&mdev->peer_seq_lock); 1653b411b363SPhilipp Reisner return ret; 1654b411b363SPhilipp Reisner } 1655b411b363SPhilipp Reisner 1656b411b363SPhilipp Reisner /* mirrored write */ 1657b411b363SPhilipp Reisner static int receive_Data(struct drbd_conf *mdev, struct p_header *h) 1658b411b363SPhilipp Reisner { 1659b411b363SPhilipp Reisner sector_t sector; 1660b411b363SPhilipp Reisner struct drbd_epoch_entry *e; 1661b411b363SPhilipp Reisner struct p_data *p = (struct p_data *)h; 1662b411b363SPhilipp Reisner int header_size, data_size; 1663b411b363SPhilipp Reisner int rw = WRITE; 1664b411b363SPhilipp Reisner u32 dp_flags; 1665b411b363SPhilipp Reisner 1666b411b363SPhilipp Reisner header_size = sizeof(*p) - sizeof(*h); 1667b411b363SPhilipp Reisner data_size = h->length - header_size; 1668b411b363SPhilipp Reisner 1669b411b363SPhilipp Reisner ERR_IF(data_size == 0) return FALSE; 1670b411b363SPhilipp Reisner 1671b411b363SPhilipp Reisner if (drbd_recv(mdev, h->payload, header_size) != header_size) 1672b411b363SPhilipp Reisner return FALSE; 1673b411b363SPhilipp Reisner 1674b411b363SPhilipp Reisner if (!get_ldev(mdev)) { 1675b411b363SPhilipp Reisner if (__ratelimit(&drbd_ratelimit_state)) 1676b411b363SPhilipp Reisner dev_err(DEV, "Can not write mirrored data block " 1677b411b363SPhilipp Reisner "to local disk.\n"); 1678b411b363SPhilipp Reisner spin_lock(&mdev->peer_seq_lock); 1679b411b363SPhilipp Reisner if (mdev->peer_seq+1 == be32_to_cpu(p->seq_num)) 1680b411b363SPhilipp Reisner mdev->peer_seq++; 1681b411b363SPhilipp Reisner spin_unlock(&mdev->peer_seq_lock); 1682b411b363SPhilipp Reisner 1683b411b363SPhilipp Reisner drbd_send_ack_dp(mdev, P_NEG_ACK, p); 1684b411b363SPhilipp Reisner atomic_inc(&mdev->current_epoch->epoch_size); 1685b411b363SPhilipp Reisner return drbd_drain_block(mdev, data_size); 1686b411b363SPhilipp Reisner } 1687b411b363SPhilipp Reisner 1688b411b363SPhilipp Reisner /* get_ldev(mdev) successful. 1689b411b363SPhilipp Reisner * Corresponding put_ldev done either below (on various errors), 1690b411b363SPhilipp Reisner * or in drbd_endio_write_sec, if we successfully submit the data at 1691b411b363SPhilipp Reisner * the end of this function. */ 1692b411b363SPhilipp Reisner 1693b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 1694b411b363SPhilipp Reisner e = read_in_block(mdev, p->block_id, sector, data_size); 1695b411b363SPhilipp Reisner if (!e) { 1696b411b363SPhilipp Reisner put_ldev(mdev); 1697b411b363SPhilipp Reisner return FALSE; 1698b411b363SPhilipp Reisner } 1699b411b363SPhilipp Reisner 1700b411b363SPhilipp Reisner e->private_bio->bi_end_io = drbd_endio_write_sec; 1701b411b363SPhilipp Reisner e->w.cb = e_end_block; 1702b411b363SPhilipp Reisner 1703b411b363SPhilipp Reisner spin_lock(&mdev->epoch_lock); 1704b411b363SPhilipp Reisner e->epoch = mdev->current_epoch; 1705b411b363SPhilipp Reisner atomic_inc(&e->epoch->epoch_size); 1706b411b363SPhilipp Reisner atomic_inc(&e->epoch->active); 1707b411b363SPhilipp Reisner 1708b411b363SPhilipp Reisner if (mdev->write_ordering == WO_bio_barrier && atomic_read(&e->epoch->epoch_size) == 1) { 1709b411b363SPhilipp Reisner struct drbd_epoch *epoch; 1710b411b363SPhilipp Reisner /* Issue a barrier if we start a new epoch, and the previous epoch 1711b411b363SPhilipp Reisner was not a epoch containing a single request which already was 1712b411b363SPhilipp Reisner a Barrier. */ 1713b411b363SPhilipp Reisner epoch = list_entry(e->epoch->list.prev, struct drbd_epoch, list); 1714b411b363SPhilipp Reisner if (epoch == e->epoch) { 1715b411b363SPhilipp Reisner set_bit(DE_CONTAINS_A_BARRIER, &e->epoch->flags); 1716b411b363SPhilipp Reisner rw |= (1<<BIO_RW_BARRIER); 1717b411b363SPhilipp Reisner e->flags |= EE_IS_BARRIER; 1718b411b363SPhilipp Reisner } else { 1719b411b363SPhilipp Reisner if (atomic_read(&epoch->epoch_size) > 1 || 1720b411b363SPhilipp Reisner !test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags)) { 1721b411b363SPhilipp Reisner set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags); 1722b411b363SPhilipp Reisner set_bit(DE_CONTAINS_A_BARRIER, &e->epoch->flags); 1723b411b363SPhilipp Reisner rw |= (1<<BIO_RW_BARRIER); 1724b411b363SPhilipp Reisner e->flags |= EE_IS_BARRIER; 1725b411b363SPhilipp Reisner } 1726b411b363SPhilipp Reisner } 1727b411b363SPhilipp Reisner } 1728b411b363SPhilipp Reisner spin_unlock(&mdev->epoch_lock); 1729b411b363SPhilipp Reisner 1730b411b363SPhilipp Reisner dp_flags = be32_to_cpu(p->dp_flags); 1731b411b363SPhilipp Reisner if (dp_flags & DP_HARDBARRIER) { 1732b411b363SPhilipp Reisner dev_err(DEV, "ASSERT FAILED would have submitted barrier request\n"); 1733b411b363SPhilipp Reisner /* rw |= (1<<BIO_RW_BARRIER); */ 1734b411b363SPhilipp Reisner } 1735b411b363SPhilipp Reisner if (dp_flags & DP_RW_SYNC) 1736b411b363SPhilipp Reisner rw |= (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG); 1737b411b363SPhilipp Reisner if (dp_flags & DP_MAY_SET_IN_SYNC) 1738b411b363SPhilipp Reisner e->flags |= EE_MAY_SET_IN_SYNC; 1739b411b363SPhilipp Reisner 1740b411b363SPhilipp Reisner /* I'm the receiver, I do hold a net_cnt reference. */ 1741b411b363SPhilipp Reisner if (!mdev->net_conf->two_primaries) { 1742b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 1743b411b363SPhilipp Reisner } else { 1744b411b363SPhilipp Reisner /* don't get the req_lock yet, 1745b411b363SPhilipp Reisner * we may sleep in drbd_wait_peer_seq */ 1746b411b363SPhilipp Reisner const int size = e->size; 1747b411b363SPhilipp Reisner const int discard = test_bit(DISCARD_CONCURRENT, &mdev->flags); 1748b411b363SPhilipp Reisner DEFINE_WAIT(wait); 1749b411b363SPhilipp Reisner struct drbd_request *i; 1750b411b363SPhilipp Reisner struct hlist_node *n; 1751b411b363SPhilipp Reisner struct hlist_head *slot; 1752b411b363SPhilipp Reisner int first; 1753b411b363SPhilipp Reisner 1754b411b363SPhilipp Reisner D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 1755b411b363SPhilipp Reisner BUG_ON(mdev->ee_hash == NULL); 1756b411b363SPhilipp Reisner BUG_ON(mdev->tl_hash == NULL); 1757b411b363SPhilipp Reisner 1758b411b363SPhilipp Reisner /* conflict detection and handling: 1759b411b363SPhilipp Reisner * 1. wait on the sequence number, 1760b411b363SPhilipp Reisner * in case this data packet overtook ACK packets. 1761b411b363SPhilipp Reisner * 2. check our hash tables for conflicting requests. 1762b411b363SPhilipp Reisner * we only need to walk the tl_hash, since an ee can not 1763b411b363SPhilipp Reisner * have a conflict with an other ee: on the submitting 1764b411b363SPhilipp Reisner * node, the corresponding req had already been conflicting, 1765b411b363SPhilipp Reisner * and a conflicting req is never sent. 1766b411b363SPhilipp Reisner * 1767b411b363SPhilipp Reisner * Note: for two_primaries, we are protocol C, 1768b411b363SPhilipp Reisner * so there cannot be any request that is DONE 1769b411b363SPhilipp Reisner * but still on the transfer log. 1770b411b363SPhilipp Reisner * 1771b411b363SPhilipp Reisner * unconditionally add to the ee_hash. 1772b411b363SPhilipp Reisner * 1773b411b363SPhilipp Reisner * if no conflicting request is found: 1774b411b363SPhilipp Reisner * submit. 1775b411b363SPhilipp Reisner * 1776b411b363SPhilipp Reisner * if any conflicting request is found 1777b411b363SPhilipp Reisner * that has not yet been acked, 1778b411b363SPhilipp Reisner * AND I have the "discard concurrent writes" flag: 1779b411b363SPhilipp Reisner * queue (via done_ee) the P_DISCARD_ACK; OUT. 1780b411b363SPhilipp Reisner * 1781b411b363SPhilipp Reisner * if any conflicting request is found: 1782b411b363SPhilipp Reisner * block the receiver, waiting on misc_wait 1783b411b363SPhilipp Reisner * until no more conflicting requests are there, 1784b411b363SPhilipp Reisner * or we get interrupted (disconnect). 1785b411b363SPhilipp Reisner * 1786b411b363SPhilipp Reisner * we do not just write after local io completion of those 1787b411b363SPhilipp Reisner * requests, but only after req is done completely, i.e. 1788b411b363SPhilipp Reisner * we wait for the P_DISCARD_ACK to arrive! 1789b411b363SPhilipp Reisner * 1790b411b363SPhilipp Reisner * then proceed normally, i.e. submit. 1791b411b363SPhilipp Reisner */ 1792b411b363SPhilipp Reisner if (drbd_wait_peer_seq(mdev, be32_to_cpu(p->seq_num))) 1793b411b363SPhilipp Reisner goto out_interrupted; 1794b411b363SPhilipp Reisner 1795b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 1796b411b363SPhilipp Reisner 1797b411b363SPhilipp Reisner hlist_add_head(&e->colision, ee_hash_slot(mdev, sector)); 1798b411b363SPhilipp Reisner 1799b411b363SPhilipp Reisner #define OVERLAPS overlaps(i->sector, i->size, sector, size) 1800b411b363SPhilipp Reisner slot = tl_hash_slot(mdev, sector); 1801b411b363SPhilipp Reisner first = 1; 1802b411b363SPhilipp Reisner for (;;) { 1803b411b363SPhilipp Reisner int have_unacked = 0; 1804b411b363SPhilipp Reisner int have_conflict = 0; 1805b411b363SPhilipp Reisner prepare_to_wait(&mdev->misc_wait, &wait, 1806b411b363SPhilipp Reisner TASK_INTERRUPTIBLE); 1807b411b363SPhilipp Reisner hlist_for_each_entry(i, n, slot, colision) { 1808b411b363SPhilipp Reisner if (OVERLAPS) { 1809b411b363SPhilipp Reisner /* only ALERT on first iteration, 1810b411b363SPhilipp Reisner * we may be woken up early... */ 1811b411b363SPhilipp Reisner if (first) 1812b411b363SPhilipp Reisner dev_alert(DEV, "%s[%u] Concurrent local write detected!" 1813b411b363SPhilipp Reisner " new: %llus +%u; pending: %llus +%u\n", 1814b411b363SPhilipp Reisner current->comm, current->pid, 1815b411b363SPhilipp Reisner (unsigned long long)sector, size, 1816b411b363SPhilipp Reisner (unsigned long long)i->sector, i->size); 1817b411b363SPhilipp Reisner if (i->rq_state & RQ_NET_PENDING) 1818b411b363SPhilipp Reisner ++have_unacked; 1819b411b363SPhilipp Reisner ++have_conflict; 1820b411b363SPhilipp Reisner } 1821b411b363SPhilipp Reisner } 1822b411b363SPhilipp Reisner #undef OVERLAPS 1823b411b363SPhilipp Reisner if (!have_conflict) 1824b411b363SPhilipp Reisner break; 1825b411b363SPhilipp Reisner 1826b411b363SPhilipp Reisner /* Discard Ack only for the _first_ iteration */ 1827b411b363SPhilipp Reisner if (first && discard && have_unacked) { 1828b411b363SPhilipp Reisner dev_alert(DEV, "Concurrent write! [DISCARD BY FLAG] sec=%llus\n", 1829b411b363SPhilipp Reisner (unsigned long long)sector); 1830b411b363SPhilipp Reisner inc_unacked(mdev); 1831b411b363SPhilipp Reisner e->w.cb = e_send_discard_ack; 1832b411b363SPhilipp Reisner list_add_tail(&e->w.list, &mdev->done_ee); 1833b411b363SPhilipp Reisner 1834b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1835b411b363SPhilipp Reisner 1836b411b363SPhilipp Reisner /* we could probably send that P_DISCARD_ACK ourselves, 1837b411b363SPhilipp Reisner * but I don't like the receiver using the msock */ 1838b411b363SPhilipp Reisner 1839b411b363SPhilipp Reisner put_ldev(mdev); 1840b411b363SPhilipp Reisner wake_asender(mdev); 1841b411b363SPhilipp Reisner finish_wait(&mdev->misc_wait, &wait); 1842b411b363SPhilipp Reisner return TRUE; 1843b411b363SPhilipp Reisner } 1844b411b363SPhilipp Reisner 1845b411b363SPhilipp Reisner if (signal_pending(current)) { 1846b411b363SPhilipp Reisner hlist_del_init(&e->colision); 1847b411b363SPhilipp Reisner 1848b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1849b411b363SPhilipp Reisner 1850b411b363SPhilipp Reisner finish_wait(&mdev->misc_wait, &wait); 1851b411b363SPhilipp Reisner goto out_interrupted; 1852b411b363SPhilipp Reisner } 1853b411b363SPhilipp Reisner 1854b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1855b411b363SPhilipp Reisner if (first) { 1856b411b363SPhilipp Reisner first = 0; 1857b411b363SPhilipp Reisner dev_alert(DEV, "Concurrent write! [W AFTERWARDS] " 1858b411b363SPhilipp Reisner "sec=%llus\n", (unsigned long long)sector); 1859b411b363SPhilipp Reisner } else if (discard) { 1860b411b363SPhilipp Reisner /* we had none on the first iteration. 1861b411b363SPhilipp Reisner * there must be none now. */ 1862b411b363SPhilipp Reisner D_ASSERT(have_unacked == 0); 1863b411b363SPhilipp Reisner } 1864b411b363SPhilipp Reisner schedule(); 1865b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 1866b411b363SPhilipp Reisner } 1867b411b363SPhilipp Reisner finish_wait(&mdev->misc_wait, &wait); 1868b411b363SPhilipp Reisner } 1869b411b363SPhilipp Reisner 1870b411b363SPhilipp Reisner list_add(&e->w.list, &mdev->active_ee); 1871b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1872b411b363SPhilipp Reisner 1873b411b363SPhilipp Reisner switch (mdev->net_conf->wire_protocol) { 1874b411b363SPhilipp Reisner case DRBD_PROT_C: 1875b411b363SPhilipp Reisner inc_unacked(mdev); 1876b411b363SPhilipp Reisner /* corresponding dec_unacked() in e_end_block() 1877b411b363SPhilipp Reisner * respective _drbd_clear_done_ee */ 1878b411b363SPhilipp Reisner break; 1879b411b363SPhilipp Reisner case DRBD_PROT_B: 1880b411b363SPhilipp Reisner /* I really don't like it that the receiver thread 1881b411b363SPhilipp Reisner * sends on the msock, but anyways */ 1882b411b363SPhilipp Reisner drbd_send_ack(mdev, P_RECV_ACK, e); 1883b411b363SPhilipp Reisner break; 1884b411b363SPhilipp Reisner case DRBD_PROT_A: 1885b411b363SPhilipp Reisner /* nothing to do */ 1886b411b363SPhilipp Reisner break; 1887b411b363SPhilipp Reisner } 1888b411b363SPhilipp Reisner 1889b411b363SPhilipp Reisner if (mdev->state.pdsk == D_DISKLESS) { 1890b411b363SPhilipp Reisner /* In case we have the only disk of the cluster, */ 1891b411b363SPhilipp Reisner drbd_set_out_of_sync(mdev, e->sector, e->size); 1892b411b363SPhilipp Reisner e->flags |= EE_CALL_AL_COMPLETE_IO; 1893b411b363SPhilipp Reisner drbd_al_begin_io(mdev, e->sector); 1894b411b363SPhilipp Reisner } 1895b411b363SPhilipp Reisner 1896b411b363SPhilipp Reisner e->private_bio->bi_rw = rw; 1897b411b363SPhilipp Reisner drbd_generic_make_request(mdev, DRBD_FAULT_DT_WR, e->private_bio); 1898b411b363SPhilipp Reisner /* accounting done in endio */ 1899b411b363SPhilipp Reisner 1900b411b363SPhilipp Reisner maybe_kick_lo(mdev); 1901b411b363SPhilipp Reisner return TRUE; 1902b411b363SPhilipp Reisner 1903b411b363SPhilipp Reisner out_interrupted: 1904b411b363SPhilipp Reisner /* yes, the epoch_size now is imbalanced. 1905b411b363SPhilipp Reisner * but we drop the connection anyways, so we don't have a chance to 1906b411b363SPhilipp Reisner * receive a barrier... atomic_inc(&mdev->epoch_size); */ 1907b411b363SPhilipp Reisner put_ldev(mdev); 1908b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 1909b411b363SPhilipp Reisner return FALSE; 1910b411b363SPhilipp Reisner } 1911b411b363SPhilipp Reisner 1912b411b363SPhilipp Reisner static int receive_DataRequest(struct drbd_conf *mdev, struct p_header *h) 1913b411b363SPhilipp Reisner { 1914b411b363SPhilipp Reisner sector_t sector; 1915b411b363SPhilipp Reisner const sector_t capacity = drbd_get_capacity(mdev->this_bdev); 1916b411b363SPhilipp Reisner struct drbd_epoch_entry *e; 1917b411b363SPhilipp Reisner struct digest_info *di = NULL; 1918b411b363SPhilipp Reisner int size, digest_size; 1919b411b363SPhilipp Reisner unsigned int fault_type; 1920b411b363SPhilipp Reisner struct p_block_req *p = 1921b411b363SPhilipp Reisner (struct p_block_req *)h; 1922b411b363SPhilipp Reisner const int brps = sizeof(*p)-sizeof(*h); 1923b411b363SPhilipp Reisner 1924b411b363SPhilipp Reisner if (drbd_recv(mdev, h->payload, brps) != brps) 1925b411b363SPhilipp Reisner return FALSE; 1926b411b363SPhilipp Reisner 1927b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 1928b411b363SPhilipp Reisner size = be32_to_cpu(p->blksize); 1929b411b363SPhilipp Reisner 1930b411b363SPhilipp Reisner if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_SEGMENT_SIZE) { 1931b411b363SPhilipp Reisner dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, 1932b411b363SPhilipp Reisner (unsigned long long)sector, size); 1933b411b363SPhilipp Reisner return FALSE; 1934b411b363SPhilipp Reisner } 1935b411b363SPhilipp Reisner if (sector + (size>>9) > capacity) { 1936b411b363SPhilipp Reisner dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, 1937b411b363SPhilipp Reisner (unsigned long long)sector, size); 1938b411b363SPhilipp Reisner return FALSE; 1939b411b363SPhilipp Reisner } 1940b411b363SPhilipp Reisner 1941b411b363SPhilipp Reisner if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) { 1942b411b363SPhilipp Reisner if (__ratelimit(&drbd_ratelimit_state)) 1943b411b363SPhilipp Reisner dev_err(DEV, "Can not satisfy peer's read request, " 1944b411b363SPhilipp Reisner "no local data.\n"); 1945b411b363SPhilipp Reisner drbd_send_ack_rp(mdev, h->command == P_DATA_REQUEST ? P_NEG_DREPLY : 1946b411b363SPhilipp Reisner P_NEG_RS_DREPLY , p); 1947b411b363SPhilipp Reisner return TRUE; 1948b411b363SPhilipp Reisner } 1949b411b363SPhilipp Reisner 1950b411b363SPhilipp Reisner /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD 1951b411b363SPhilipp Reisner * "criss-cross" setup, that might cause write-out on some other DRBD, 1952b411b363SPhilipp Reisner * which in turn might block on the other node at this very place. */ 1953b411b363SPhilipp Reisner e = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO); 1954b411b363SPhilipp Reisner if (!e) { 1955b411b363SPhilipp Reisner put_ldev(mdev); 1956b411b363SPhilipp Reisner return FALSE; 1957b411b363SPhilipp Reisner } 1958b411b363SPhilipp Reisner 1959b411b363SPhilipp Reisner e->private_bio->bi_rw = READ; 1960b411b363SPhilipp Reisner e->private_bio->bi_end_io = drbd_endio_read_sec; 1961b411b363SPhilipp Reisner 1962b411b363SPhilipp Reisner switch (h->command) { 1963b411b363SPhilipp Reisner case P_DATA_REQUEST: 1964b411b363SPhilipp Reisner e->w.cb = w_e_end_data_req; 1965b411b363SPhilipp Reisner fault_type = DRBD_FAULT_DT_RD; 1966b411b363SPhilipp Reisner break; 1967b411b363SPhilipp Reisner case P_RS_DATA_REQUEST: 1968b411b363SPhilipp Reisner e->w.cb = w_e_end_rsdata_req; 1969b411b363SPhilipp Reisner fault_type = DRBD_FAULT_RS_RD; 1970b411b363SPhilipp Reisner /* Eventually this should become asynchronously. Currently it 1971b411b363SPhilipp Reisner * blocks the whole receiver just to delay the reading of a 1972b411b363SPhilipp Reisner * resync data block. 1973b411b363SPhilipp Reisner * the drbd_work_queue mechanism is made for this... 1974b411b363SPhilipp Reisner */ 1975b411b363SPhilipp Reisner if (!drbd_rs_begin_io(mdev, sector)) { 1976b411b363SPhilipp Reisner /* we have been interrupted, 1977b411b363SPhilipp Reisner * probably connection lost! */ 1978b411b363SPhilipp Reisner D_ASSERT(signal_pending(current)); 1979b411b363SPhilipp Reisner goto out_free_e; 1980b411b363SPhilipp Reisner } 1981b411b363SPhilipp Reisner break; 1982b411b363SPhilipp Reisner 1983b411b363SPhilipp Reisner case P_OV_REPLY: 1984b411b363SPhilipp Reisner case P_CSUM_RS_REQUEST: 1985b411b363SPhilipp Reisner fault_type = DRBD_FAULT_RS_RD; 1986b411b363SPhilipp Reisner digest_size = h->length - brps ; 1987b411b363SPhilipp Reisner di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO); 1988b411b363SPhilipp Reisner if (!di) 1989b411b363SPhilipp Reisner goto out_free_e; 1990b411b363SPhilipp Reisner 1991b411b363SPhilipp Reisner di->digest_size = digest_size; 1992b411b363SPhilipp Reisner di->digest = (((char *)di)+sizeof(struct digest_info)); 1993b411b363SPhilipp Reisner 1994b411b363SPhilipp Reisner if (drbd_recv(mdev, di->digest, digest_size) != digest_size) 1995b411b363SPhilipp Reisner goto out_free_e; 1996b411b363SPhilipp Reisner 1997b411b363SPhilipp Reisner e->block_id = (u64)(unsigned long)di; 1998b411b363SPhilipp Reisner if (h->command == P_CSUM_RS_REQUEST) { 1999b411b363SPhilipp Reisner D_ASSERT(mdev->agreed_pro_version >= 89); 2000b411b363SPhilipp Reisner e->w.cb = w_e_end_csum_rs_req; 2001b411b363SPhilipp Reisner } else if (h->command == P_OV_REPLY) { 2002b411b363SPhilipp Reisner e->w.cb = w_e_end_ov_reply; 2003b411b363SPhilipp Reisner dec_rs_pending(mdev); 2004b411b363SPhilipp Reisner break; 2005b411b363SPhilipp Reisner } 2006b411b363SPhilipp Reisner 2007b411b363SPhilipp Reisner if (!drbd_rs_begin_io(mdev, sector)) { 2008b411b363SPhilipp Reisner /* we have been interrupted, probably connection lost! */ 2009b411b363SPhilipp Reisner D_ASSERT(signal_pending(current)); 2010b411b363SPhilipp Reisner goto out_free_e; 2011b411b363SPhilipp Reisner } 2012b411b363SPhilipp Reisner break; 2013b411b363SPhilipp Reisner 2014b411b363SPhilipp Reisner case P_OV_REQUEST: 2015b411b363SPhilipp Reisner if (mdev->state.conn >= C_CONNECTED && 2016b411b363SPhilipp Reisner mdev->state.conn != C_VERIFY_T) 2017b411b363SPhilipp Reisner dev_warn(DEV, "ASSERT FAILED: got P_OV_REQUEST while being %s\n", 2018b411b363SPhilipp Reisner drbd_conn_str(mdev->state.conn)); 2019b411b363SPhilipp Reisner if (mdev->ov_start_sector == ~(sector_t)0 && 2020b411b363SPhilipp Reisner mdev->agreed_pro_version >= 90) { 2021b411b363SPhilipp Reisner mdev->ov_start_sector = sector; 2022b411b363SPhilipp Reisner mdev->ov_position = sector; 2023b411b363SPhilipp Reisner mdev->ov_left = mdev->rs_total - BM_SECT_TO_BIT(sector); 2024b411b363SPhilipp Reisner dev_info(DEV, "Online Verify start sector: %llu\n", 2025b411b363SPhilipp Reisner (unsigned long long)sector); 2026b411b363SPhilipp Reisner } 2027b411b363SPhilipp Reisner e->w.cb = w_e_end_ov_req; 2028b411b363SPhilipp Reisner fault_type = DRBD_FAULT_RS_RD; 2029b411b363SPhilipp Reisner /* Eventually this should become asynchronous. Currently it 2030b411b363SPhilipp Reisner * blocks the whole receiver just to delay the reading of a 2031b411b363SPhilipp Reisner * resync data block. 2032b411b363SPhilipp Reisner * the drbd_work_queue mechanism is made for this... 2033b411b363SPhilipp Reisner */ 2034b411b363SPhilipp Reisner if (!drbd_rs_begin_io(mdev, sector)) { 2035b411b363SPhilipp Reisner /* we have been interrupted, 2036b411b363SPhilipp Reisner * probably connection lost! */ 2037b411b363SPhilipp Reisner D_ASSERT(signal_pending(current)); 2038b411b363SPhilipp Reisner goto out_free_e; 2039b411b363SPhilipp Reisner } 2040b411b363SPhilipp Reisner break; 2041b411b363SPhilipp Reisner 2042b411b363SPhilipp Reisner 2043b411b363SPhilipp Reisner default: 2044b411b363SPhilipp Reisner dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", 2045b411b363SPhilipp Reisner cmdname(h->command)); 2046b411b363SPhilipp Reisner fault_type = DRBD_FAULT_MAX; 2047b411b363SPhilipp Reisner } 2048b411b363SPhilipp Reisner 2049b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 2050b411b363SPhilipp Reisner list_add(&e->w.list, &mdev->read_ee); 2051b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 2052b411b363SPhilipp Reisner 2053b411b363SPhilipp Reisner inc_unacked(mdev); 2054b411b363SPhilipp Reisner 2055b411b363SPhilipp Reisner drbd_generic_make_request(mdev, fault_type, e->private_bio); 2056b411b363SPhilipp Reisner maybe_kick_lo(mdev); 2057b411b363SPhilipp Reisner 2058b411b363SPhilipp Reisner return TRUE; 2059b411b363SPhilipp Reisner 2060b411b363SPhilipp Reisner out_free_e: 2061b411b363SPhilipp Reisner kfree(di); 2062b411b363SPhilipp Reisner put_ldev(mdev); 2063b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 2064b411b363SPhilipp Reisner return FALSE; 2065b411b363SPhilipp Reisner } 2066b411b363SPhilipp Reisner 2067b411b363SPhilipp Reisner static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local) 2068b411b363SPhilipp Reisner { 2069b411b363SPhilipp Reisner int self, peer, rv = -100; 2070b411b363SPhilipp Reisner unsigned long ch_self, ch_peer; 2071b411b363SPhilipp Reisner 2072b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_BITMAP] & 1; 2073b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_BITMAP] & 1; 2074b411b363SPhilipp Reisner 2075b411b363SPhilipp Reisner ch_peer = mdev->p_uuid[UI_SIZE]; 2076b411b363SPhilipp Reisner ch_self = mdev->comm_bm_set; 2077b411b363SPhilipp Reisner 2078b411b363SPhilipp Reisner switch (mdev->net_conf->after_sb_0p) { 2079b411b363SPhilipp Reisner case ASB_CONSENSUS: 2080b411b363SPhilipp Reisner case ASB_DISCARD_SECONDARY: 2081b411b363SPhilipp Reisner case ASB_CALL_HELPER: 2082b411b363SPhilipp Reisner dev_err(DEV, "Configuration error.\n"); 2083b411b363SPhilipp Reisner break; 2084b411b363SPhilipp Reisner case ASB_DISCONNECT: 2085b411b363SPhilipp Reisner break; 2086b411b363SPhilipp Reisner case ASB_DISCARD_YOUNGER_PRI: 2087b411b363SPhilipp Reisner if (self == 0 && peer == 1) { 2088b411b363SPhilipp Reisner rv = -1; 2089b411b363SPhilipp Reisner break; 2090b411b363SPhilipp Reisner } 2091b411b363SPhilipp Reisner if (self == 1 && peer == 0) { 2092b411b363SPhilipp Reisner rv = 1; 2093b411b363SPhilipp Reisner break; 2094b411b363SPhilipp Reisner } 2095b411b363SPhilipp Reisner /* Else fall through to one of the other strategies... */ 2096b411b363SPhilipp Reisner case ASB_DISCARD_OLDER_PRI: 2097b411b363SPhilipp Reisner if (self == 0 && peer == 1) { 2098b411b363SPhilipp Reisner rv = 1; 2099b411b363SPhilipp Reisner break; 2100b411b363SPhilipp Reisner } 2101b411b363SPhilipp Reisner if (self == 1 && peer == 0) { 2102b411b363SPhilipp Reisner rv = -1; 2103b411b363SPhilipp Reisner break; 2104b411b363SPhilipp Reisner } 2105b411b363SPhilipp Reisner /* Else fall through to one of the other strategies... */ 2106ad19bf6eSLars Ellenberg dev_warn(DEV, "Discard younger/older primary did not find a decision\n" 2107b411b363SPhilipp Reisner "Using discard-least-changes instead\n"); 2108b411b363SPhilipp Reisner case ASB_DISCARD_ZERO_CHG: 2109b411b363SPhilipp Reisner if (ch_peer == 0 && ch_self == 0) { 2110b411b363SPhilipp Reisner rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) 2111b411b363SPhilipp Reisner ? -1 : 1; 2112b411b363SPhilipp Reisner break; 2113b411b363SPhilipp Reisner } else { 2114b411b363SPhilipp Reisner if (ch_peer == 0) { rv = 1; break; } 2115b411b363SPhilipp Reisner if (ch_self == 0) { rv = -1; break; } 2116b411b363SPhilipp Reisner } 2117b411b363SPhilipp Reisner if (mdev->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG) 2118b411b363SPhilipp Reisner break; 2119b411b363SPhilipp Reisner case ASB_DISCARD_LEAST_CHG: 2120b411b363SPhilipp Reisner if (ch_self < ch_peer) 2121b411b363SPhilipp Reisner rv = -1; 2122b411b363SPhilipp Reisner else if (ch_self > ch_peer) 2123b411b363SPhilipp Reisner rv = 1; 2124b411b363SPhilipp Reisner else /* ( ch_self == ch_peer ) */ 2125b411b363SPhilipp Reisner /* Well, then use something else. */ 2126b411b363SPhilipp Reisner rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) 2127b411b363SPhilipp Reisner ? -1 : 1; 2128b411b363SPhilipp Reisner break; 2129b411b363SPhilipp Reisner case ASB_DISCARD_LOCAL: 2130b411b363SPhilipp Reisner rv = -1; 2131b411b363SPhilipp Reisner break; 2132b411b363SPhilipp Reisner case ASB_DISCARD_REMOTE: 2133b411b363SPhilipp Reisner rv = 1; 2134b411b363SPhilipp Reisner } 2135b411b363SPhilipp Reisner 2136b411b363SPhilipp Reisner return rv; 2137b411b363SPhilipp Reisner } 2138b411b363SPhilipp Reisner 2139b411b363SPhilipp Reisner static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local) 2140b411b363SPhilipp Reisner { 2141b411b363SPhilipp Reisner int self, peer, hg, rv = -100; 2142b411b363SPhilipp Reisner 2143b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_BITMAP] & 1; 2144b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_BITMAP] & 1; 2145b411b363SPhilipp Reisner 2146b411b363SPhilipp Reisner switch (mdev->net_conf->after_sb_1p) { 2147b411b363SPhilipp Reisner case ASB_DISCARD_YOUNGER_PRI: 2148b411b363SPhilipp Reisner case ASB_DISCARD_OLDER_PRI: 2149b411b363SPhilipp Reisner case ASB_DISCARD_LEAST_CHG: 2150b411b363SPhilipp Reisner case ASB_DISCARD_LOCAL: 2151b411b363SPhilipp Reisner case ASB_DISCARD_REMOTE: 2152b411b363SPhilipp Reisner dev_err(DEV, "Configuration error.\n"); 2153b411b363SPhilipp Reisner break; 2154b411b363SPhilipp Reisner case ASB_DISCONNECT: 2155b411b363SPhilipp Reisner break; 2156b411b363SPhilipp Reisner case ASB_CONSENSUS: 2157b411b363SPhilipp Reisner hg = drbd_asb_recover_0p(mdev); 2158b411b363SPhilipp Reisner if (hg == -1 && mdev->state.role == R_SECONDARY) 2159b411b363SPhilipp Reisner rv = hg; 2160b411b363SPhilipp Reisner if (hg == 1 && mdev->state.role == R_PRIMARY) 2161b411b363SPhilipp Reisner rv = hg; 2162b411b363SPhilipp Reisner break; 2163b411b363SPhilipp Reisner case ASB_VIOLENTLY: 2164b411b363SPhilipp Reisner rv = drbd_asb_recover_0p(mdev); 2165b411b363SPhilipp Reisner break; 2166b411b363SPhilipp Reisner case ASB_DISCARD_SECONDARY: 2167b411b363SPhilipp Reisner return mdev->state.role == R_PRIMARY ? 1 : -1; 2168b411b363SPhilipp Reisner case ASB_CALL_HELPER: 2169b411b363SPhilipp Reisner hg = drbd_asb_recover_0p(mdev); 2170b411b363SPhilipp Reisner if (hg == -1 && mdev->state.role == R_PRIMARY) { 2171b411b363SPhilipp Reisner self = drbd_set_role(mdev, R_SECONDARY, 0); 2172b411b363SPhilipp Reisner /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, 2173b411b363SPhilipp Reisner * we might be here in C_WF_REPORT_PARAMS which is transient. 2174b411b363SPhilipp Reisner * we do not need to wait for the after state change work either. */ 2175b411b363SPhilipp Reisner self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); 2176b411b363SPhilipp Reisner if (self != SS_SUCCESS) { 2177b411b363SPhilipp Reisner drbd_khelper(mdev, "pri-lost-after-sb"); 2178b411b363SPhilipp Reisner } else { 2179b411b363SPhilipp Reisner dev_warn(DEV, "Successfully gave up primary role.\n"); 2180b411b363SPhilipp Reisner rv = hg; 2181b411b363SPhilipp Reisner } 2182b411b363SPhilipp Reisner } else 2183b411b363SPhilipp Reisner rv = hg; 2184b411b363SPhilipp Reisner } 2185b411b363SPhilipp Reisner 2186b411b363SPhilipp Reisner return rv; 2187b411b363SPhilipp Reisner } 2188b411b363SPhilipp Reisner 2189b411b363SPhilipp Reisner static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local) 2190b411b363SPhilipp Reisner { 2191b411b363SPhilipp Reisner int self, peer, hg, rv = -100; 2192b411b363SPhilipp Reisner 2193b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_BITMAP] & 1; 2194b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_BITMAP] & 1; 2195b411b363SPhilipp Reisner 2196b411b363SPhilipp Reisner switch (mdev->net_conf->after_sb_2p) { 2197b411b363SPhilipp Reisner case ASB_DISCARD_YOUNGER_PRI: 2198b411b363SPhilipp Reisner case ASB_DISCARD_OLDER_PRI: 2199b411b363SPhilipp Reisner case ASB_DISCARD_LEAST_CHG: 2200b411b363SPhilipp Reisner case ASB_DISCARD_LOCAL: 2201b411b363SPhilipp Reisner case ASB_DISCARD_REMOTE: 2202b411b363SPhilipp Reisner case ASB_CONSENSUS: 2203b411b363SPhilipp Reisner case ASB_DISCARD_SECONDARY: 2204b411b363SPhilipp Reisner dev_err(DEV, "Configuration error.\n"); 2205b411b363SPhilipp Reisner break; 2206b411b363SPhilipp Reisner case ASB_VIOLENTLY: 2207b411b363SPhilipp Reisner rv = drbd_asb_recover_0p(mdev); 2208b411b363SPhilipp Reisner break; 2209b411b363SPhilipp Reisner case ASB_DISCONNECT: 2210b411b363SPhilipp Reisner break; 2211b411b363SPhilipp Reisner case ASB_CALL_HELPER: 2212b411b363SPhilipp Reisner hg = drbd_asb_recover_0p(mdev); 2213b411b363SPhilipp Reisner if (hg == -1) { 2214b411b363SPhilipp Reisner /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, 2215b411b363SPhilipp Reisner * we might be here in C_WF_REPORT_PARAMS which is transient. 2216b411b363SPhilipp Reisner * we do not need to wait for the after state change work either. */ 2217b411b363SPhilipp Reisner self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); 2218b411b363SPhilipp Reisner if (self != SS_SUCCESS) { 2219b411b363SPhilipp Reisner drbd_khelper(mdev, "pri-lost-after-sb"); 2220b411b363SPhilipp Reisner } else { 2221b411b363SPhilipp Reisner dev_warn(DEV, "Successfully gave up primary role.\n"); 2222b411b363SPhilipp Reisner rv = hg; 2223b411b363SPhilipp Reisner } 2224b411b363SPhilipp Reisner } else 2225b411b363SPhilipp Reisner rv = hg; 2226b411b363SPhilipp Reisner } 2227b411b363SPhilipp Reisner 2228b411b363SPhilipp Reisner return rv; 2229b411b363SPhilipp Reisner } 2230b411b363SPhilipp Reisner 2231b411b363SPhilipp Reisner static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid, 2232b411b363SPhilipp Reisner u64 bits, u64 flags) 2233b411b363SPhilipp Reisner { 2234b411b363SPhilipp Reisner if (!uuid) { 2235b411b363SPhilipp Reisner dev_info(DEV, "%s uuid info vanished while I was looking!\n", text); 2236b411b363SPhilipp Reisner return; 2237b411b363SPhilipp Reisner } 2238b411b363SPhilipp Reisner dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", 2239b411b363SPhilipp Reisner text, 2240b411b363SPhilipp Reisner (unsigned long long)uuid[UI_CURRENT], 2241b411b363SPhilipp Reisner (unsigned long long)uuid[UI_BITMAP], 2242b411b363SPhilipp Reisner (unsigned long long)uuid[UI_HISTORY_START], 2243b411b363SPhilipp Reisner (unsigned long long)uuid[UI_HISTORY_END], 2244b411b363SPhilipp Reisner (unsigned long long)bits, 2245b411b363SPhilipp Reisner (unsigned long long)flags); 2246b411b363SPhilipp Reisner } 2247b411b363SPhilipp Reisner 2248b411b363SPhilipp Reisner /* 2249b411b363SPhilipp Reisner 100 after split brain try auto recover 2250b411b363SPhilipp Reisner 2 C_SYNC_SOURCE set BitMap 2251b411b363SPhilipp Reisner 1 C_SYNC_SOURCE use BitMap 2252b411b363SPhilipp Reisner 0 no Sync 2253b411b363SPhilipp Reisner -1 C_SYNC_TARGET use BitMap 2254b411b363SPhilipp Reisner -2 C_SYNC_TARGET set BitMap 2255b411b363SPhilipp Reisner -100 after split brain, disconnect 2256b411b363SPhilipp Reisner -1000 unrelated data 2257b411b363SPhilipp Reisner */ 2258b411b363SPhilipp Reisner static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local) 2259b411b363SPhilipp Reisner { 2260b411b363SPhilipp Reisner u64 self, peer; 2261b411b363SPhilipp Reisner int i, j; 2262b411b363SPhilipp Reisner 2263b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); 2264b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); 2265b411b363SPhilipp Reisner 2266b411b363SPhilipp Reisner *rule_nr = 10; 2267b411b363SPhilipp Reisner if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) 2268b411b363SPhilipp Reisner return 0; 2269b411b363SPhilipp Reisner 2270b411b363SPhilipp Reisner *rule_nr = 20; 2271b411b363SPhilipp Reisner if ((self == UUID_JUST_CREATED || self == (u64)0) && 2272b411b363SPhilipp Reisner peer != UUID_JUST_CREATED) 2273b411b363SPhilipp Reisner return -2; 2274b411b363SPhilipp Reisner 2275b411b363SPhilipp Reisner *rule_nr = 30; 2276b411b363SPhilipp Reisner if (self != UUID_JUST_CREATED && 2277b411b363SPhilipp Reisner (peer == UUID_JUST_CREATED || peer == (u64)0)) 2278b411b363SPhilipp Reisner return 2; 2279b411b363SPhilipp Reisner 2280b411b363SPhilipp Reisner if (self == peer) { 2281b411b363SPhilipp Reisner int rct, dc; /* roles at crash time */ 2282b411b363SPhilipp Reisner 2283b411b363SPhilipp Reisner if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) { 2284b411b363SPhilipp Reisner 2285b411b363SPhilipp Reisner if (mdev->agreed_pro_version < 91) 2286b411b363SPhilipp Reisner return -1001; 2287b411b363SPhilipp Reisner 2288b411b363SPhilipp Reisner if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) && 2289b411b363SPhilipp Reisner (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { 2290b411b363SPhilipp Reisner dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n"); 2291b411b363SPhilipp Reisner drbd_uuid_set_bm(mdev, 0UL); 2292b411b363SPhilipp Reisner 2293b411b363SPhilipp Reisner drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, 2294b411b363SPhilipp Reisner mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); 2295b411b363SPhilipp Reisner *rule_nr = 34; 2296b411b363SPhilipp Reisner } else { 2297b411b363SPhilipp Reisner dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n"); 2298b411b363SPhilipp Reisner *rule_nr = 36; 2299b411b363SPhilipp Reisner } 2300b411b363SPhilipp Reisner 2301b411b363SPhilipp Reisner return 1; 2302b411b363SPhilipp Reisner } 2303b411b363SPhilipp Reisner 2304b411b363SPhilipp Reisner if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) { 2305b411b363SPhilipp Reisner 2306b411b363SPhilipp Reisner if (mdev->agreed_pro_version < 91) 2307b411b363SPhilipp Reisner return -1001; 2308b411b363SPhilipp Reisner 2309b411b363SPhilipp Reisner if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) && 2310b411b363SPhilipp Reisner (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) { 2311b411b363SPhilipp Reisner dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); 2312b411b363SPhilipp Reisner 2313b411b363SPhilipp Reisner mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START]; 2314b411b363SPhilipp Reisner mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP]; 2315b411b363SPhilipp Reisner mdev->p_uuid[UI_BITMAP] = 0UL; 2316b411b363SPhilipp Reisner 2317b411b363SPhilipp Reisner drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); 2318b411b363SPhilipp Reisner *rule_nr = 35; 2319b411b363SPhilipp Reisner } else { 2320b411b363SPhilipp Reisner dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n"); 2321b411b363SPhilipp Reisner *rule_nr = 37; 2322b411b363SPhilipp Reisner } 2323b411b363SPhilipp Reisner 2324b411b363SPhilipp Reisner return -1; 2325b411b363SPhilipp Reisner } 2326b411b363SPhilipp Reisner 2327b411b363SPhilipp Reisner /* Common power [off|failure] */ 2328b411b363SPhilipp Reisner rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) + 2329b411b363SPhilipp Reisner (mdev->p_uuid[UI_FLAGS] & 2); 2330b411b363SPhilipp Reisner /* lowest bit is set when we were primary, 2331b411b363SPhilipp Reisner * next bit (weight 2) is set when peer was primary */ 2332b411b363SPhilipp Reisner *rule_nr = 40; 2333b411b363SPhilipp Reisner 2334b411b363SPhilipp Reisner switch (rct) { 2335b411b363SPhilipp Reisner case 0: /* !self_pri && !peer_pri */ return 0; 2336b411b363SPhilipp Reisner case 1: /* self_pri && !peer_pri */ return 1; 2337b411b363SPhilipp Reisner case 2: /* !self_pri && peer_pri */ return -1; 2338b411b363SPhilipp Reisner case 3: /* self_pri && peer_pri */ 2339b411b363SPhilipp Reisner dc = test_bit(DISCARD_CONCURRENT, &mdev->flags); 2340b411b363SPhilipp Reisner return dc ? -1 : 1; 2341b411b363SPhilipp Reisner } 2342b411b363SPhilipp Reisner } 2343b411b363SPhilipp Reisner 2344b411b363SPhilipp Reisner *rule_nr = 50; 2345b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); 2346b411b363SPhilipp Reisner if (self == peer) 2347b411b363SPhilipp Reisner return -1; 2348b411b363SPhilipp Reisner 2349b411b363SPhilipp Reisner *rule_nr = 51; 2350b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); 2351b411b363SPhilipp Reisner if (self == peer) { 2352b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); 2353b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1); 2354b411b363SPhilipp Reisner if (self == peer) { 2355b411b363SPhilipp Reisner /* The last P_SYNC_UUID did not get though. Undo the last start of 2356b411b363SPhilipp Reisner resync as sync source modifications of the peer's UUIDs. */ 2357b411b363SPhilipp Reisner 2358b411b363SPhilipp Reisner if (mdev->agreed_pro_version < 91) 2359b411b363SPhilipp Reisner return -1001; 2360b411b363SPhilipp Reisner 2361b411b363SPhilipp Reisner mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START]; 2362b411b363SPhilipp Reisner mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1]; 2363b411b363SPhilipp Reisner return -1; 2364b411b363SPhilipp Reisner } 2365b411b363SPhilipp Reisner } 2366b411b363SPhilipp Reisner 2367b411b363SPhilipp Reisner *rule_nr = 60; 2368b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); 2369b411b363SPhilipp Reisner for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 2370b411b363SPhilipp Reisner peer = mdev->p_uuid[i] & ~((u64)1); 2371b411b363SPhilipp Reisner if (self == peer) 2372b411b363SPhilipp Reisner return -2; 2373b411b363SPhilipp Reisner } 2374b411b363SPhilipp Reisner 2375b411b363SPhilipp Reisner *rule_nr = 70; 2376b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); 2377b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); 2378b411b363SPhilipp Reisner if (self == peer) 2379b411b363SPhilipp Reisner return 1; 2380b411b363SPhilipp Reisner 2381b411b363SPhilipp Reisner *rule_nr = 71; 2382b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); 2383b411b363SPhilipp Reisner if (self == peer) { 2384b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1); 2385b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); 2386b411b363SPhilipp Reisner if (self == peer) { 2387b411b363SPhilipp Reisner /* The last P_SYNC_UUID did not get though. Undo the last start of 2388b411b363SPhilipp Reisner resync as sync source modifications of our UUIDs. */ 2389b411b363SPhilipp Reisner 2390b411b363SPhilipp Reisner if (mdev->agreed_pro_version < 91) 2391b411b363SPhilipp Reisner return -1001; 2392b411b363SPhilipp Reisner 2393b411b363SPhilipp Reisner _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]); 2394b411b363SPhilipp Reisner _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]); 2395b411b363SPhilipp Reisner 2396b411b363SPhilipp Reisner dev_info(DEV, "Undid last start of resync:\n"); 2397b411b363SPhilipp Reisner 2398b411b363SPhilipp Reisner drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, 2399b411b363SPhilipp Reisner mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); 2400b411b363SPhilipp Reisner 2401b411b363SPhilipp Reisner return 1; 2402b411b363SPhilipp Reisner } 2403b411b363SPhilipp Reisner } 2404b411b363SPhilipp Reisner 2405b411b363SPhilipp Reisner 2406b411b363SPhilipp Reisner *rule_nr = 80; 2407d8c2a36bSPhilipp Reisner peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); 2408b411b363SPhilipp Reisner for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 2409b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[i] & ~((u64)1); 2410b411b363SPhilipp Reisner if (self == peer) 2411b411b363SPhilipp Reisner return 2; 2412b411b363SPhilipp Reisner } 2413b411b363SPhilipp Reisner 2414b411b363SPhilipp Reisner *rule_nr = 90; 2415b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); 2416b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); 2417b411b363SPhilipp Reisner if (self == peer && self != ((u64)0)) 2418b411b363SPhilipp Reisner return 100; 2419b411b363SPhilipp Reisner 2420b411b363SPhilipp Reisner *rule_nr = 100; 2421b411b363SPhilipp Reisner for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 2422b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[i] & ~((u64)1); 2423b411b363SPhilipp Reisner for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { 2424b411b363SPhilipp Reisner peer = mdev->p_uuid[j] & ~((u64)1); 2425b411b363SPhilipp Reisner if (self == peer) 2426b411b363SPhilipp Reisner return -100; 2427b411b363SPhilipp Reisner } 2428b411b363SPhilipp Reisner } 2429b411b363SPhilipp Reisner 2430b411b363SPhilipp Reisner return -1000; 2431b411b363SPhilipp Reisner } 2432b411b363SPhilipp Reisner 2433b411b363SPhilipp Reisner /* drbd_sync_handshake() returns the new conn state on success, or 2434b411b363SPhilipp Reisner CONN_MASK (-1) on failure. 2435b411b363SPhilipp Reisner */ 2436b411b363SPhilipp Reisner static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role, 2437b411b363SPhilipp Reisner enum drbd_disk_state peer_disk) __must_hold(local) 2438b411b363SPhilipp Reisner { 2439b411b363SPhilipp Reisner int hg, rule_nr; 2440b411b363SPhilipp Reisner enum drbd_conns rv = C_MASK; 2441b411b363SPhilipp Reisner enum drbd_disk_state mydisk; 2442b411b363SPhilipp Reisner 2443b411b363SPhilipp Reisner mydisk = mdev->state.disk; 2444b411b363SPhilipp Reisner if (mydisk == D_NEGOTIATING) 2445b411b363SPhilipp Reisner mydisk = mdev->new_state_tmp.disk; 2446b411b363SPhilipp Reisner 2447b411b363SPhilipp Reisner dev_info(DEV, "drbd_sync_handshake:\n"); 2448b411b363SPhilipp Reisner drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0); 2449b411b363SPhilipp Reisner drbd_uuid_dump(mdev, "peer", mdev->p_uuid, 2450b411b363SPhilipp Reisner mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); 2451b411b363SPhilipp Reisner 2452b411b363SPhilipp Reisner hg = drbd_uuid_compare(mdev, &rule_nr); 2453b411b363SPhilipp Reisner 2454b411b363SPhilipp Reisner dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr); 2455b411b363SPhilipp Reisner 2456b411b363SPhilipp Reisner if (hg == -1000) { 2457b411b363SPhilipp Reisner dev_alert(DEV, "Unrelated data, aborting!\n"); 2458b411b363SPhilipp Reisner return C_MASK; 2459b411b363SPhilipp Reisner } 2460b411b363SPhilipp Reisner if (hg == -1001) { 2461b411b363SPhilipp Reisner dev_alert(DEV, "To resolve this both sides have to support at least protocol\n"); 2462b411b363SPhilipp Reisner return C_MASK; 2463b411b363SPhilipp Reisner } 2464b411b363SPhilipp Reisner 2465b411b363SPhilipp Reisner if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || 2466b411b363SPhilipp Reisner (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { 2467b411b363SPhilipp Reisner int f = (hg == -100) || abs(hg) == 2; 2468b411b363SPhilipp Reisner hg = mydisk > D_INCONSISTENT ? 1 : -1; 2469b411b363SPhilipp Reisner if (f) 2470b411b363SPhilipp Reisner hg = hg*2; 2471b411b363SPhilipp Reisner dev_info(DEV, "Becoming sync %s due to disk states.\n", 2472b411b363SPhilipp Reisner hg > 0 ? "source" : "target"); 2473b411b363SPhilipp Reisner } 2474b411b363SPhilipp Reisner 2475b411b363SPhilipp Reisner if (hg == 100 || (hg == -100 && mdev->net_conf->always_asbp)) { 2476b411b363SPhilipp Reisner int pcount = (mdev->state.role == R_PRIMARY) 2477b411b363SPhilipp Reisner + (peer_role == R_PRIMARY); 2478b411b363SPhilipp Reisner int forced = (hg == -100); 2479b411b363SPhilipp Reisner 2480b411b363SPhilipp Reisner switch (pcount) { 2481b411b363SPhilipp Reisner case 0: 2482b411b363SPhilipp Reisner hg = drbd_asb_recover_0p(mdev); 2483b411b363SPhilipp Reisner break; 2484b411b363SPhilipp Reisner case 1: 2485b411b363SPhilipp Reisner hg = drbd_asb_recover_1p(mdev); 2486b411b363SPhilipp Reisner break; 2487b411b363SPhilipp Reisner case 2: 2488b411b363SPhilipp Reisner hg = drbd_asb_recover_2p(mdev); 2489b411b363SPhilipp Reisner break; 2490b411b363SPhilipp Reisner } 2491b411b363SPhilipp Reisner if (abs(hg) < 100) { 2492b411b363SPhilipp Reisner dev_warn(DEV, "Split-Brain detected, %d primaries, " 2493b411b363SPhilipp Reisner "automatically solved. Sync from %s node\n", 2494b411b363SPhilipp Reisner pcount, (hg < 0) ? "peer" : "this"); 2495b411b363SPhilipp Reisner if (forced) { 2496b411b363SPhilipp Reisner dev_warn(DEV, "Doing a full sync, since" 2497b411b363SPhilipp Reisner " UUIDs where ambiguous.\n"); 2498b411b363SPhilipp Reisner hg = hg*2; 2499b411b363SPhilipp Reisner } 2500b411b363SPhilipp Reisner } 2501b411b363SPhilipp Reisner } 2502b411b363SPhilipp Reisner 2503b411b363SPhilipp Reisner if (hg == -100) { 2504b411b363SPhilipp Reisner if (mdev->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1)) 2505b411b363SPhilipp Reisner hg = -1; 2506b411b363SPhilipp Reisner if (!mdev->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1)) 2507b411b363SPhilipp Reisner hg = 1; 2508b411b363SPhilipp Reisner 2509b411b363SPhilipp Reisner if (abs(hg) < 100) 2510b411b363SPhilipp Reisner dev_warn(DEV, "Split-Brain detected, manually solved. " 2511b411b363SPhilipp Reisner "Sync from %s node\n", 2512b411b363SPhilipp Reisner (hg < 0) ? "peer" : "this"); 2513b411b363SPhilipp Reisner } 2514b411b363SPhilipp Reisner 2515b411b363SPhilipp Reisner if (hg == -100) { 2516b411b363SPhilipp Reisner dev_alert(DEV, "Split-Brain detected, dropping connection!\n"); 2517b411b363SPhilipp Reisner drbd_khelper(mdev, "split-brain"); 2518b411b363SPhilipp Reisner return C_MASK; 2519b411b363SPhilipp Reisner } 2520b411b363SPhilipp Reisner 2521b411b363SPhilipp Reisner if (hg > 0 && mydisk <= D_INCONSISTENT) { 2522b411b363SPhilipp Reisner dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n"); 2523b411b363SPhilipp Reisner return C_MASK; 2524b411b363SPhilipp Reisner } 2525b411b363SPhilipp Reisner 2526b411b363SPhilipp Reisner if (hg < 0 && /* by intention we do not use mydisk here. */ 2527b411b363SPhilipp Reisner mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) { 2528b411b363SPhilipp Reisner switch (mdev->net_conf->rr_conflict) { 2529b411b363SPhilipp Reisner case ASB_CALL_HELPER: 2530b411b363SPhilipp Reisner drbd_khelper(mdev, "pri-lost"); 2531b411b363SPhilipp Reisner /* fall through */ 2532b411b363SPhilipp Reisner case ASB_DISCONNECT: 2533b411b363SPhilipp Reisner dev_err(DEV, "I shall become SyncTarget, but I am primary!\n"); 2534b411b363SPhilipp Reisner return C_MASK; 2535b411b363SPhilipp Reisner case ASB_VIOLENTLY: 2536b411b363SPhilipp Reisner dev_warn(DEV, "Becoming SyncTarget, violating the stable-data" 2537b411b363SPhilipp Reisner "assumption\n"); 2538b411b363SPhilipp Reisner } 2539b411b363SPhilipp Reisner } 2540b411b363SPhilipp Reisner 2541b411b363SPhilipp Reisner if (abs(hg) >= 2) { 2542b411b363SPhilipp Reisner dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); 2543b411b363SPhilipp Reisner if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake")) 2544b411b363SPhilipp Reisner return C_MASK; 2545b411b363SPhilipp Reisner } 2546b411b363SPhilipp Reisner 2547b411b363SPhilipp Reisner if (hg > 0) { /* become sync source. */ 2548b411b363SPhilipp Reisner rv = C_WF_BITMAP_S; 2549b411b363SPhilipp Reisner } else if (hg < 0) { /* become sync target */ 2550b411b363SPhilipp Reisner rv = C_WF_BITMAP_T; 2551b411b363SPhilipp Reisner } else { 2552b411b363SPhilipp Reisner rv = C_CONNECTED; 2553b411b363SPhilipp Reisner if (drbd_bm_total_weight(mdev)) { 2554b411b363SPhilipp Reisner dev_info(DEV, "No resync, but %lu bits in bitmap!\n", 2555b411b363SPhilipp Reisner drbd_bm_total_weight(mdev)); 2556b411b363SPhilipp Reisner } 2557b411b363SPhilipp Reisner } 2558b411b363SPhilipp Reisner 2559b411b363SPhilipp Reisner return rv; 2560b411b363SPhilipp Reisner } 2561b411b363SPhilipp Reisner 2562b411b363SPhilipp Reisner /* returns 1 if invalid */ 2563b411b363SPhilipp Reisner static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self) 2564b411b363SPhilipp Reisner { 2565b411b363SPhilipp Reisner /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ 2566b411b363SPhilipp Reisner if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) || 2567b411b363SPhilipp Reisner (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL)) 2568b411b363SPhilipp Reisner return 0; 2569b411b363SPhilipp Reisner 2570b411b363SPhilipp Reisner /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ 2571b411b363SPhilipp Reisner if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL || 2572b411b363SPhilipp Reisner self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL) 2573b411b363SPhilipp Reisner return 1; 2574b411b363SPhilipp Reisner 2575b411b363SPhilipp Reisner /* everything else is valid if they are equal on both sides. */ 2576b411b363SPhilipp Reisner if (peer == self) 2577b411b363SPhilipp Reisner return 0; 2578b411b363SPhilipp Reisner 2579b411b363SPhilipp Reisner /* everything es is invalid. */ 2580b411b363SPhilipp Reisner return 1; 2581b411b363SPhilipp Reisner } 2582b411b363SPhilipp Reisner 2583b411b363SPhilipp Reisner static int receive_protocol(struct drbd_conf *mdev, struct p_header *h) 2584b411b363SPhilipp Reisner { 2585b411b363SPhilipp Reisner struct p_protocol *p = (struct p_protocol *)h; 2586b411b363SPhilipp Reisner int header_size, data_size; 2587b411b363SPhilipp Reisner int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; 2588b411b363SPhilipp Reisner int p_want_lose, p_two_primaries; 2589b411b363SPhilipp Reisner char p_integrity_alg[SHARED_SECRET_MAX] = ""; 2590b411b363SPhilipp Reisner 2591b411b363SPhilipp Reisner header_size = sizeof(*p) - sizeof(*h); 2592b411b363SPhilipp Reisner data_size = h->length - header_size; 2593b411b363SPhilipp Reisner 2594b411b363SPhilipp Reisner if (drbd_recv(mdev, h->payload, header_size) != header_size) 2595b411b363SPhilipp Reisner return FALSE; 2596b411b363SPhilipp Reisner 2597b411b363SPhilipp Reisner p_proto = be32_to_cpu(p->protocol); 2598b411b363SPhilipp Reisner p_after_sb_0p = be32_to_cpu(p->after_sb_0p); 2599b411b363SPhilipp Reisner p_after_sb_1p = be32_to_cpu(p->after_sb_1p); 2600b411b363SPhilipp Reisner p_after_sb_2p = be32_to_cpu(p->after_sb_2p); 2601b411b363SPhilipp Reisner p_want_lose = be32_to_cpu(p->want_lose); 2602b411b363SPhilipp Reisner p_two_primaries = be32_to_cpu(p->two_primaries); 2603b411b363SPhilipp Reisner 2604b411b363SPhilipp Reisner if (p_proto != mdev->net_conf->wire_protocol) { 2605b411b363SPhilipp Reisner dev_err(DEV, "incompatible communication protocols\n"); 2606b411b363SPhilipp Reisner goto disconnect; 2607b411b363SPhilipp Reisner } 2608b411b363SPhilipp Reisner 2609b411b363SPhilipp Reisner if (cmp_after_sb(p_after_sb_0p, mdev->net_conf->after_sb_0p)) { 2610b411b363SPhilipp Reisner dev_err(DEV, "incompatible after-sb-0pri settings\n"); 2611b411b363SPhilipp Reisner goto disconnect; 2612b411b363SPhilipp Reisner } 2613b411b363SPhilipp Reisner 2614b411b363SPhilipp Reisner if (cmp_after_sb(p_after_sb_1p, mdev->net_conf->after_sb_1p)) { 2615b411b363SPhilipp Reisner dev_err(DEV, "incompatible after-sb-1pri settings\n"); 2616b411b363SPhilipp Reisner goto disconnect; 2617b411b363SPhilipp Reisner } 2618b411b363SPhilipp Reisner 2619b411b363SPhilipp Reisner if (cmp_after_sb(p_after_sb_2p, mdev->net_conf->after_sb_2p)) { 2620b411b363SPhilipp Reisner dev_err(DEV, "incompatible after-sb-2pri settings\n"); 2621b411b363SPhilipp Reisner goto disconnect; 2622b411b363SPhilipp Reisner } 2623b411b363SPhilipp Reisner 2624b411b363SPhilipp Reisner if (p_want_lose && mdev->net_conf->want_lose) { 2625b411b363SPhilipp Reisner dev_err(DEV, "both sides have the 'want_lose' flag set\n"); 2626b411b363SPhilipp Reisner goto disconnect; 2627b411b363SPhilipp Reisner } 2628b411b363SPhilipp Reisner 2629b411b363SPhilipp Reisner if (p_two_primaries != mdev->net_conf->two_primaries) { 2630b411b363SPhilipp Reisner dev_err(DEV, "incompatible setting of the two-primaries options\n"); 2631b411b363SPhilipp Reisner goto disconnect; 2632b411b363SPhilipp Reisner } 2633b411b363SPhilipp Reisner 2634b411b363SPhilipp Reisner if (mdev->agreed_pro_version >= 87) { 2635b411b363SPhilipp Reisner unsigned char *my_alg = mdev->net_conf->integrity_alg; 2636b411b363SPhilipp Reisner 2637b411b363SPhilipp Reisner if (drbd_recv(mdev, p_integrity_alg, data_size) != data_size) 2638b411b363SPhilipp Reisner return FALSE; 2639b411b363SPhilipp Reisner 2640b411b363SPhilipp Reisner p_integrity_alg[SHARED_SECRET_MAX-1] = 0; 2641b411b363SPhilipp Reisner if (strcmp(p_integrity_alg, my_alg)) { 2642b411b363SPhilipp Reisner dev_err(DEV, "incompatible setting of the data-integrity-alg\n"); 2643b411b363SPhilipp Reisner goto disconnect; 2644b411b363SPhilipp Reisner } 2645b411b363SPhilipp Reisner dev_info(DEV, "data-integrity-alg: %s\n", 2646b411b363SPhilipp Reisner my_alg[0] ? my_alg : (unsigned char *)"<not-used>"); 2647b411b363SPhilipp Reisner } 2648b411b363SPhilipp Reisner 2649b411b363SPhilipp Reisner return TRUE; 2650b411b363SPhilipp Reisner 2651b411b363SPhilipp Reisner disconnect: 2652b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 2653b411b363SPhilipp Reisner return FALSE; 2654b411b363SPhilipp Reisner } 2655b411b363SPhilipp Reisner 2656b411b363SPhilipp Reisner /* helper function 2657b411b363SPhilipp Reisner * input: alg name, feature name 2658b411b363SPhilipp Reisner * return: NULL (alg name was "") 2659b411b363SPhilipp Reisner * ERR_PTR(error) if something goes wrong 2660b411b363SPhilipp Reisner * or the crypto hash ptr, if it worked out ok. */ 2661b411b363SPhilipp Reisner struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev, 2662b411b363SPhilipp Reisner const char *alg, const char *name) 2663b411b363SPhilipp Reisner { 2664b411b363SPhilipp Reisner struct crypto_hash *tfm; 2665b411b363SPhilipp Reisner 2666b411b363SPhilipp Reisner if (!alg[0]) 2667b411b363SPhilipp Reisner return NULL; 2668b411b363SPhilipp Reisner 2669b411b363SPhilipp Reisner tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); 2670b411b363SPhilipp Reisner if (IS_ERR(tfm)) { 2671b411b363SPhilipp Reisner dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n", 2672b411b363SPhilipp Reisner alg, name, PTR_ERR(tfm)); 2673b411b363SPhilipp Reisner return tfm; 2674b411b363SPhilipp Reisner } 2675b411b363SPhilipp Reisner if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) { 2676b411b363SPhilipp Reisner crypto_free_hash(tfm); 2677b411b363SPhilipp Reisner dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name); 2678b411b363SPhilipp Reisner return ERR_PTR(-EINVAL); 2679b411b363SPhilipp Reisner } 2680b411b363SPhilipp Reisner return tfm; 2681b411b363SPhilipp Reisner } 2682b411b363SPhilipp Reisner 2683b411b363SPhilipp Reisner static int receive_SyncParam(struct drbd_conf *mdev, struct p_header *h) 2684b411b363SPhilipp Reisner { 2685b411b363SPhilipp Reisner int ok = TRUE; 2686b411b363SPhilipp Reisner struct p_rs_param_89 *p = (struct p_rs_param_89 *)h; 2687b411b363SPhilipp Reisner unsigned int header_size, data_size, exp_max_sz; 2688b411b363SPhilipp Reisner struct crypto_hash *verify_tfm = NULL; 2689b411b363SPhilipp Reisner struct crypto_hash *csums_tfm = NULL; 2690b411b363SPhilipp Reisner const int apv = mdev->agreed_pro_version; 2691b411b363SPhilipp Reisner 2692b411b363SPhilipp Reisner exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) 2693b411b363SPhilipp Reisner : apv == 88 ? sizeof(struct p_rs_param) 2694b411b363SPhilipp Reisner + SHARED_SECRET_MAX 2695b411b363SPhilipp Reisner : /* 89 */ sizeof(struct p_rs_param_89); 2696b411b363SPhilipp Reisner 2697b411b363SPhilipp Reisner if (h->length > exp_max_sz) { 2698b411b363SPhilipp Reisner dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n", 2699b411b363SPhilipp Reisner h->length, exp_max_sz); 2700b411b363SPhilipp Reisner return FALSE; 2701b411b363SPhilipp Reisner } 2702b411b363SPhilipp Reisner 2703b411b363SPhilipp Reisner if (apv <= 88) { 2704b411b363SPhilipp Reisner header_size = sizeof(struct p_rs_param) - sizeof(*h); 2705b411b363SPhilipp Reisner data_size = h->length - header_size; 2706b411b363SPhilipp Reisner } else /* apv >= 89 */ { 2707b411b363SPhilipp Reisner header_size = sizeof(struct p_rs_param_89) - sizeof(*h); 2708b411b363SPhilipp Reisner data_size = h->length - header_size; 2709b411b363SPhilipp Reisner D_ASSERT(data_size == 0); 2710b411b363SPhilipp Reisner } 2711b411b363SPhilipp Reisner 2712b411b363SPhilipp Reisner /* initialize verify_alg and csums_alg */ 2713b411b363SPhilipp Reisner memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); 2714b411b363SPhilipp Reisner 2715b411b363SPhilipp Reisner if (drbd_recv(mdev, h->payload, header_size) != header_size) 2716b411b363SPhilipp Reisner return FALSE; 2717b411b363SPhilipp Reisner 2718b411b363SPhilipp Reisner mdev->sync_conf.rate = be32_to_cpu(p->rate); 2719b411b363SPhilipp Reisner 2720b411b363SPhilipp Reisner if (apv >= 88) { 2721b411b363SPhilipp Reisner if (apv == 88) { 2722b411b363SPhilipp Reisner if (data_size > SHARED_SECRET_MAX) { 2723b411b363SPhilipp Reisner dev_err(DEV, "verify-alg too long, " 2724b411b363SPhilipp Reisner "peer wants %u, accepting only %u byte\n", 2725b411b363SPhilipp Reisner data_size, SHARED_SECRET_MAX); 2726b411b363SPhilipp Reisner return FALSE; 2727b411b363SPhilipp Reisner } 2728b411b363SPhilipp Reisner 2729b411b363SPhilipp Reisner if (drbd_recv(mdev, p->verify_alg, data_size) != data_size) 2730b411b363SPhilipp Reisner return FALSE; 2731b411b363SPhilipp Reisner 2732b411b363SPhilipp Reisner /* we expect NUL terminated string */ 2733b411b363SPhilipp Reisner /* but just in case someone tries to be evil */ 2734b411b363SPhilipp Reisner D_ASSERT(p->verify_alg[data_size-1] == 0); 2735b411b363SPhilipp Reisner p->verify_alg[data_size-1] = 0; 2736b411b363SPhilipp Reisner 2737b411b363SPhilipp Reisner } else /* apv >= 89 */ { 2738b411b363SPhilipp Reisner /* we still expect NUL terminated strings */ 2739b411b363SPhilipp Reisner /* but just in case someone tries to be evil */ 2740b411b363SPhilipp Reisner D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0); 2741b411b363SPhilipp Reisner D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0); 2742b411b363SPhilipp Reisner p->verify_alg[SHARED_SECRET_MAX-1] = 0; 2743b411b363SPhilipp Reisner p->csums_alg[SHARED_SECRET_MAX-1] = 0; 2744b411b363SPhilipp Reisner } 2745b411b363SPhilipp Reisner 2746b411b363SPhilipp Reisner if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) { 2747b411b363SPhilipp Reisner if (mdev->state.conn == C_WF_REPORT_PARAMS) { 2748b411b363SPhilipp Reisner dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", 2749b411b363SPhilipp Reisner mdev->sync_conf.verify_alg, p->verify_alg); 2750b411b363SPhilipp Reisner goto disconnect; 2751b411b363SPhilipp Reisner } 2752b411b363SPhilipp Reisner verify_tfm = drbd_crypto_alloc_digest_safe(mdev, 2753b411b363SPhilipp Reisner p->verify_alg, "verify-alg"); 2754b411b363SPhilipp Reisner if (IS_ERR(verify_tfm)) { 2755b411b363SPhilipp Reisner verify_tfm = NULL; 2756b411b363SPhilipp Reisner goto disconnect; 2757b411b363SPhilipp Reisner } 2758b411b363SPhilipp Reisner } 2759b411b363SPhilipp Reisner 2760b411b363SPhilipp Reisner if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) { 2761b411b363SPhilipp Reisner if (mdev->state.conn == C_WF_REPORT_PARAMS) { 2762b411b363SPhilipp Reisner dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", 2763b411b363SPhilipp Reisner mdev->sync_conf.csums_alg, p->csums_alg); 2764b411b363SPhilipp Reisner goto disconnect; 2765b411b363SPhilipp Reisner } 2766b411b363SPhilipp Reisner csums_tfm = drbd_crypto_alloc_digest_safe(mdev, 2767b411b363SPhilipp Reisner p->csums_alg, "csums-alg"); 2768b411b363SPhilipp Reisner if (IS_ERR(csums_tfm)) { 2769b411b363SPhilipp Reisner csums_tfm = NULL; 2770b411b363SPhilipp Reisner goto disconnect; 2771b411b363SPhilipp Reisner } 2772b411b363SPhilipp Reisner } 2773b411b363SPhilipp Reisner 2774b411b363SPhilipp Reisner 2775b411b363SPhilipp Reisner spin_lock(&mdev->peer_seq_lock); 2776b411b363SPhilipp Reisner /* lock against drbd_nl_syncer_conf() */ 2777b411b363SPhilipp Reisner if (verify_tfm) { 2778b411b363SPhilipp Reisner strcpy(mdev->sync_conf.verify_alg, p->verify_alg); 2779b411b363SPhilipp Reisner mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1; 2780b411b363SPhilipp Reisner crypto_free_hash(mdev->verify_tfm); 2781b411b363SPhilipp Reisner mdev->verify_tfm = verify_tfm; 2782b411b363SPhilipp Reisner dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg); 2783b411b363SPhilipp Reisner } 2784b411b363SPhilipp Reisner if (csums_tfm) { 2785b411b363SPhilipp Reisner strcpy(mdev->sync_conf.csums_alg, p->csums_alg); 2786b411b363SPhilipp Reisner mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1; 2787b411b363SPhilipp Reisner crypto_free_hash(mdev->csums_tfm); 2788b411b363SPhilipp Reisner mdev->csums_tfm = csums_tfm; 2789b411b363SPhilipp Reisner dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg); 2790b411b363SPhilipp Reisner } 2791b411b363SPhilipp Reisner spin_unlock(&mdev->peer_seq_lock); 2792b411b363SPhilipp Reisner } 2793b411b363SPhilipp Reisner 2794b411b363SPhilipp Reisner return ok; 2795b411b363SPhilipp Reisner disconnect: 2796b411b363SPhilipp Reisner /* just for completeness: actually not needed, 2797b411b363SPhilipp Reisner * as this is not reached if csums_tfm was ok. */ 2798b411b363SPhilipp Reisner crypto_free_hash(csums_tfm); 2799b411b363SPhilipp Reisner /* but free the verify_tfm again, if csums_tfm did not work out */ 2800b411b363SPhilipp Reisner crypto_free_hash(verify_tfm); 2801b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 2802b411b363SPhilipp Reisner return FALSE; 2803b411b363SPhilipp Reisner } 2804b411b363SPhilipp Reisner 2805b411b363SPhilipp Reisner static void drbd_setup_order_type(struct drbd_conf *mdev, int peer) 2806b411b363SPhilipp Reisner { 2807b411b363SPhilipp Reisner /* sorry, we currently have no working implementation 2808b411b363SPhilipp Reisner * of distributed TCQ */ 2809b411b363SPhilipp Reisner } 2810b411b363SPhilipp Reisner 2811b411b363SPhilipp Reisner /* warn if the arguments differ by more than 12.5% */ 2812b411b363SPhilipp Reisner static void warn_if_differ_considerably(struct drbd_conf *mdev, 2813b411b363SPhilipp Reisner const char *s, sector_t a, sector_t b) 2814b411b363SPhilipp Reisner { 2815b411b363SPhilipp Reisner sector_t d; 2816b411b363SPhilipp Reisner if (a == 0 || b == 0) 2817b411b363SPhilipp Reisner return; 2818b411b363SPhilipp Reisner d = (a > b) ? (a - b) : (b - a); 2819b411b363SPhilipp Reisner if (d > (a>>3) || d > (b>>3)) 2820b411b363SPhilipp Reisner dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s, 2821b411b363SPhilipp Reisner (unsigned long long)a, (unsigned long long)b); 2822b411b363SPhilipp Reisner } 2823b411b363SPhilipp Reisner 2824b411b363SPhilipp Reisner static int receive_sizes(struct drbd_conf *mdev, struct p_header *h) 2825b411b363SPhilipp Reisner { 2826b411b363SPhilipp Reisner struct p_sizes *p = (struct p_sizes *)h; 2827b411b363SPhilipp Reisner enum determine_dev_size dd = unchanged; 2828b411b363SPhilipp Reisner unsigned int max_seg_s; 2829b411b363SPhilipp Reisner sector_t p_size, p_usize, my_usize; 2830b411b363SPhilipp Reisner int ldsc = 0; /* local disk size changed */ 2831b411b363SPhilipp Reisner enum drbd_conns nconn; 2832b411b363SPhilipp Reisner 2833b411b363SPhilipp Reisner ERR_IF(h->length != (sizeof(*p)-sizeof(*h))) return FALSE; 2834b411b363SPhilipp Reisner if (drbd_recv(mdev, h->payload, h->length) != h->length) 2835b411b363SPhilipp Reisner return FALSE; 2836b411b363SPhilipp Reisner 2837b411b363SPhilipp Reisner p_size = be64_to_cpu(p->d_size); 2838b411b363SPhilipp Reisner p_usize = be64_to_cpu(p->u_size); 2839b411b363SPhilipp Reisner 2840b411b363SPhilipp Reisner if (p_size == 0 && mdev->state.disk == D_DISKLESS) { 2841b411b363SPhilipp Reisner dev_err(DEV, "some backing storage is needed\n"); 2842b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 2843b411b363SPhilipp Reisner return FALSE; 2844b411b363SPhilipp Reisner } 2845b411b363SPhilipp Reisner 2846b411b363SPhilipp Reisner /* just store the peer's disk size for now. 2847b411b363SPhilipp Reisner * we still need to figure out whether we accept that. */ 2848b411b363SPhilipp Reisner mdev->p_size = p_size; 2849b411b363SPhilipp Reisner 2850b411b363SPhilipp Reisner #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) 2851b411b363SPhilipp Reisner if (get_ldev(mdev)) { 2852b411b363SPhilipp Reisner warn_if_differ_considerably(mdev, "lower level device sizes", 2853b411b363SPhilipp Reisner p_size, drbd_get_max_capacity(mdev->ldev)); 2854b411b363SPhilipp Reisner warn_if_differ_considerably(mdev, "user requested size", 2855b411b363SPhilipp Reisner p_usize, mdev->ldev->dc.disk_size); 2856b411b363SPhilipp Reisner 2857b411b363SPhilipp Reisner /* if this is the first connect, or an otherwise expected 2858b411b363SPhilipp Reisner * param exchange, choose the minimum */ 2859b411b363SPhilipp Reisner if (mdev->state.conn == C_WF_REPORT_PARAMS) 2860b411b363SPhilipp Reisner p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size, 2861b411b363SPhilipp Reisner p_usize); 2862b411b363SPhilipp Reisner 2863b411b363SPhilipp Reisner my_usize = mdev->ldev->dc.disk_size; 2864b411b363SPhilipp Reisner 2865b411b363SPhilipp Reisner if (mdev->ldev->dc.disk_size != p_usize) { 2866b411b363SPhilipp Reisner mdev->ldev->dc.disk_size = p_usize; 2867b411b363SPhilipp Reisner dev_info(DEV, "Peer sets u_size to %lu sectors\n", 2868b411b363SPhilipp Reisner (unsigned long)mdev->ldev->dc.disk_size); 2869b411b363SPhilipp Reisner } 2870b411b363SPhilipp Reisner 2871b411b363SPhilipp Reisner /* Never shrink a device with usable data during connect. 2872b411b363SPhilipp Reisner But allow online shrinking if we are connected. */ 2873a393db6fSPhilipp Reisner if (drbd_new_dev_size(mdev, mdev->ldev, 0) < 2874b411b363SPhilipp Reisner drbd_get_capacity(mdev->this_bdev) && 2875b411b363SPhilipp Reisner mdev->state.disk >= D_OUTDATED && 2876b411b363SPhilipp Reisner mdev->state.conn < C_CONNECTED) { 2877b411b363SPhilipp Reisner dev_err(DEV, "The peer's disk size is too small!\n"); 2878b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 2879b411b363SPhilipp Reisner mdev->ldev->dc.disk_size = my_usize; 2880b411b363SPhilipp Reisner put_ldev(mdev); 2881b411b363SPhilipp Reisner return FALSE; 2882b411b363SPhilipp Reisner } 2883b411b363SPhilipp Reisner put_ldev(mdev); 2884b411b363SPhilipp Reisner } 2885b411b363SPhilipp Reisner #undef min_not_zero 2886b411b363SPhilipp Reisner 2887b411b363SPhilipp Reisner if (get_ldev(mdev)) { 2888a393db6fSPhilipp Reisner dd = drbd_determin_dev_size(mdev, 0); 2889b411b363SPhilipp Reisner put_ldev(mdev); 2890b411b363SPhilipp Reisner if (dd == dev_size_error) 2891b411b363SPhilipp Reisner return FALSE; 2892b411b363SPhilipp Reisner drbd_md_sync(mdev); 2893b411b363SPhilipp Reisner } else { 2894b411b363SPhilipp Reisner /* I am diskless, need to accept the peer's size. */ 2895b411b363SPhilipp Reisner drbd_set_my_capacity(mdev, p_size); 2896b411b363SPhilipp Reisner } 2897b411b363SPhilipp Reisner 2898b411b363SPhilipp Reisner if (mdev->p_uuid && mdev->state.conn <= C_CONNECTED && get_ldev(mdev)) { 2899b411b363SPhilipp Reisner nconn = drbd_sync_handshake(mdev, 2900b411b363SPhilipp Reisner mdev->state.peer, mdev->state.pdsk); 2901b411b363SPhilipp Reisner put_ldev(mdev); 2902b411b363SPhilipp Reisner 2903b411b363SPhilipp Reisner if (nconn == C_MASK) { 2904b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 2905b411b363SPhilipp Reisner return FALSE; 2906b411b363SPhilipp Reisner } 2907b411b363SPhilipp Reisner 2908b411b363SPhilipp Reisner if (drbd_request_state(mdev, NS(conn, nconn)) < SS_SUCCESS) { 2909b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 2910b411b363SPhilipp Reisner return FALSE; 2911b411b363SPhilipp Reisner } 2912b411b363SPhilipp Reisner } 2913b411b363SPhilipp Reisner 2914b411b363SPhilipp Reisner if (get_ldev(mdev)) { 2915b411b363SPhilipp Reisner if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) { 2916b411b363SPhilipp Reisner mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); 2917b411b363SPhilipp Reisner ldsc = 1; 2918b411b363SPhilipp Reisner } 2919b411b363SPhilipp Reisner 2920b411b363SPhilipp Reisner max_seg_s = be32_to_cpu(p->max_segment_size); 2921b411b363SPhilipp Reisner if (max_seg_s != queue_max_segment_size(mdev->rq_queue)) 2922b411b363SPhilipp Reisner drbd_setup_queue_param(mdev, max_seg_s); 2923b411b363SPhilipp Reisner 2924b411b363SPhilipp Reisner drbd_setup_order_type(mdev, be32_to_cpu(p->queue_order_type)); 2925b411b363SPhilipp Reisner put_ldev(mdev); 2926b411b363SPhilipp Reisner } 2927b411b363SPhilipp Reisner 2928b411b363SPhilipp Reisner if (mdev->state.conn > C_WF_REPORT_PARAMS) { 2929b411b363SPhilipp Reisner if (be64_to_cpu(p->c_size) != 2930b411b363SPhilipp Reisner drbd_get_capacity(mdev->this_bdev) || ldsc) { 2931b411b363SPhilipp Reisner /* we have different sizes, probably peer 2932b411b363SPhilipp Reisner * needs to know my new size... */ 2933b411b363SPhilipp Reisner drbd_send_sizes(mdev, 0); 2934b411b363SPhilipp Reisner } 2935b411b363SPhilipp Reisner if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) || 2936b411b363SPhilipp Reisner (dd == grew && mdev->state.conn == C_CONNECTED)) { 2937b411b363SPhilipp Reisner if (mdev->state.pdsk >= D_INCONSISTENT && 2938b411b363SPhilipp Reisner mdev->state.disk >= D_INCONSISTENT) 2939b411b363SPhilipp Reisner resync_after_online_grow(mdev); 2940b411b363SPhilipp Reisner else 2941b411b363SPhilipp Reisner set_bit(RESYNC_AFTER_NEG, &mdev->flags); 2942b411b363SPhilipp Reisner } 2943b411b363SPhilipp Reisner } 2944b411b363SPhilipp Reisner 2945b411b363SPhilipp Reisner return TRUE; 2946b411b363SPhilipp Reisner } 2947b411b363SPhilipp Reisner 2948b411b363SPhilipp Reisner static int receive_uuids(struct drbd_conf *mdev, struct p_header *h) 2949b411b363SPhilipp Reisner { 2950b411b363SPhilipp Reisner struct p_uuids *p = (struct p_uuids *)h; 2951b411b363SPhilipp Reisner u64 *p_uuid; 2952b411b363SPhilipp Reisner int i; 2953b411b363SPhilipp Reisner 2954b411b363SPhilipp Reisner ERR_IF(h->length != (sizeof(*p)-sizeof(*h))) return FALSE; 2955b411b363SPhilipp Reisner if (drbd_recv(mdev, h->payload, h->length) != h->length) 2956b411b363SPhilipp Reisner return FALSE; 2957b411b363SPhilipp Reisner 2958b411b363SPhilipp Reisner p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); 2959b411b363SPhilipp Reisner 2960b411b363SPhilipp Reisner for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) 2961b411b363SPhilipp Reisner p_uuid[i] = be64_to_cpu(p->uuid[i]); 2962b411b363SPhilipp Reisner 2963b411b363SPhilipp Reisner kfree(mdev->p_uuid); 2964b411b363SPhilipp Reisner mdev->p_uuid = p_uuid; 2965b411b363SPhilipp Reisner 2966b411b363SPhilipp Reisner if (mdev->state.conn < C_CONNECTED && 2967b411b363SPhilipp Reisner mdev->state.disk < D_INCONSISTENT && 2968b411b363SPhilipp Reisner mdev->state.role == R_PRIMARY && 2969b411b363SPhilipp Reisner (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { 2970b411b363SPhilipp Reisner dev_err(DEV, "Can only connect to data with current UUID=%016llX\n", 2971b411b363SPhilipp Reisner (unsigned long long)mdev->ed_uuid); 2972b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 2973b411b363SPhilipp Reisner return FALSE; 2974b411b363SPhilipp Reisner } 2975b411b363SPhilipp Reisner 2976b411b363SPhilipp Reisner if (get_ldev(mdev)) { 2977b411b363SPhilipp Reisner int skip_initial_sync = 2978b411b363SPhilipp Reisner mdev->state.conn == C_CONNECTED && 2979b411b363SPhilipp Reisner mdev->agreed_pro_version >= 90 && 2980b411b363SPhilipp Reisner mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && 2981b411b363SPhilipp Reisner (p_uuid[UI_FLAGS] & 8); 2982b411b363SPhilipp Reisner if (skip_initial_sync) { 2983b411b363SPhilipp Reisner dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n"); 2984b411b363SPhilipp Reisner drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, 2985b411b363SPhilipp Reisner "clear_n_write from receive_uuids"); 2986b411b363SPhilipp Reisner _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]); 2987b411b363SPhilipp Reisner _drbd_uuid_set(mdev, UI_BITMAP, 0); 2988b411b363SPhilipp Reisner _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), 2989b411b363SPhilipp Reisner CS_VERBOSE, NULL); 2990b411b363SPhilipp Reisner drbd_md_sync(mdev); 2991b411b363SPhilipp Reisner } 2992b411b363SPhilipp Reisner put_ldev(mdev); 2993b411b363SPhilipp Reisner } 2994b411b363SPhilipp Reisner 2995b411b363SPhilipp Reisner /* Before we test for the disk state, we should wait until an eventually 2996b411b363SPhilipp Reisner ongoing cluster wide state change is finished. That is important if 2997b411b363SPhilipp Reisner we are primary and are detaching from our disk. We need to see the 2998b411b363SPhilipp Reisner new disk state... */ 2999b411b363SPhilipp Reisner wait_event(mdev->misc_wait, !test_bit(CLUSTER_ST_CHANGE, &mdev->flags)); 3000b411b363SPhilipp Reisner if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT) 3001b411b363SPhilipp Reisner drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); 3002b411b363SPhilipp Reisner 3003b411b363SPhilipp Reisner return TRUE; 3004b411b363SPhilipp Reisner } 3005b411b363SPhilipp Reisner 3006b411b363SPhilipp Reisner /** 3007b411b363SPhilipp Reisner * convert_state() - Converts the peer's view of the cluster state to our point of view 3008b411b363SPhilipp Reisner * @ps: The state as seen by the peer. 3009b411b363SPhilipp Reisner */ 3010b411b363SPhilipp Reisner static union drbd_state convert_state(union drbd_state ps) 3011b411b363SPhilipp Reisner { 3012b411b363SPhilipp Reisner union drbd_state ms; 3013b411b363SPhilipp Reisner 3014b411b363SPhilipp Reisner static enum drbd_conns c_tab[] = { 3015b411b363SPhilipp Reisner [C_CONNECTED] = C_CONNECTED, 3016b411b363SPhilipp Reisner 3017b411b363SPhilipp Reisner [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, 3018b411b363SPhilipp Reisner [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, 3019b411b363SPhilipp Reisner [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ 3020b411b363SPhilipp Reisner [C_VERIFY_S] = C_VERIFY_T, 3021b411b363SPhilipp Reisner [C_MASK] = C_MASK, 3022b411b363SPhilipp Reisner }; 3023b411b363SPhilipp Reisner 3024b411b363SPhilipp Reisner ms.i = ps.i; 3025b411b363SPhilipp Reisner 3026b411b363SPhilipp Reisner ms.conn = c_tab[ps.conn]; 3027b411b363SPhilipp Reisner ms.peer = ps.role; 3028b411b363SPhilipp Reisner ms.role = ps.peer; 3029b411b363SPhilipp Reisner ms.pdsk = ps.disk; 3030b411b363SPhilipp Reisner ms.disk = ps.pdsk; 3031b411b363SPhilipp Reisner ms.peer_isp = (ps.aftr_isp | ps.user_isp); 3032b411b363SPhilipp Reisner 3033b411b363SPhilipp Reisner return ms; 3034b411b363SPhilipp Reisner } 3035b411b363SPhilipp Reisner 3036b411b363SPhilipp Reisner static int receive_req_state(struct drbd_conf *mdev, struct p_header *h) 3037b411b363SPhilipp Reisner { 3038b411b363SPhilipp Reisner struct p_req_state *p = (struct p_req_state *)h; 3039b411b363SPhilipp Reisner union drbd_state mask, val; 3040b411b363SPhilipp Reisner int rv; 3041b411b363SPhilipp Reisner 3042b411b363SPhilipp Reisner ERR_IF(h->length != (sizeof(*p)-sizeof(*h))) return FALSE; 3043b411b363SPhilipp Reisner if (drbd_recv(mdev, h->payload, h->length) != h->length) 3044b411b363SPhilipp Reisner return FALSE; 3045b411b363SPhilipp Reisner 3046b411b363SPhilipp Reisner mask.i = be32_to_cpu(p->mask); 3047b411b363SPhilipp Reisner val.i = be32_to_cpu(p->val); 3048b411b363SPhilipp Reisner 3049b411b363SPhilipp Reisner if (test_bit(DISCARD_CONCURRENT, &mdev->flags) && 3050b411b363SPhilipp Reisner test_bit(CLUSTER_ST_CHANGE, &mdev->flags)) { 3051b411b363SPhilipp Reisner drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG); 3052b411b363SPhilipp Reisner return TRUE; 3053b411b363SPhilipp Reisner } 3054b411b363SPhilipp Reisner 3055b411b363SPhilipp Reisner mask = convert_state(mask); 3056b411b363SPhilipp Reisner val = convert_state(val); 3057b411b363SPhilipp Reisner 3058b411b363SPhilipp Reisner rv = drbd_change_state(mdev, CS_VERBOSE, mask, val); 3059b411b363SPhilipp Reisner 3060b411b363SPhilipp Reisner drbd_send_sr_reply(mdev, rv); 3061b411b363SPhilipp Reisner drbd_md_sync(mdev); 3062b411b363SPhilipp Reisner 3063b411b363SPhilipp Reisner return TRUE; 3064b411b363SPhilipp Reisner } 3065b411b363SPhilipp Reisner 3066b411b363SPhilipp Reisner static int receive_state(struct drbd_conf *mdev, struct p_header *h) 3067b411b363SPhilipp Reisner { 3068b411b363SPhilipp Reisner struct p_state *p = (struct p_state *)h; 3069b411b363SPhilipp Reisner enum drbd_conns nconn, oconn; 3070b411b363SPhilipp Reisner union drbd_state ns, peer_state; 3071b411b363SPhilipp Reisner enum drbd_disk_state real_peer_disk; 3072b411b363SPhilipp Reisner int rv; 3073b411b363SPhilipp Reisner 3074b411b363SPhilipp Reisner ERR_IF(h->length != (sizeof(*p)-sizeof(*h))) 3075b411b363SPhilipp Reisner return FALSE; 3076b411b363SPhilipp Reisner 3077b411b363SPhilipp Reisner if (drbd_recv(mdev, h->payload, h->length) != h->length) 3078b411b363SPhilipp Reisner return FALSE; 3079b411b363SPhilipp Reisner 3080b411b363SPhilipp Reisner peer_state.i = be32_to_cpu(p->state); 3081b411b363SPhilipp Reisner 3082b411b363SPhilipp Reisner real_peer_disk = peer_state.disk; 3083b411b363SPhilipp Reisner if (peer_state.disk == D_NEGOTIATING) { 3084b411b363SPhilipp Reisner real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; 3085b411b363SPhilipp Reisner dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); 3086b411b363SPhilipp Reisner } 3087b411b363SPhilipp Reisner 3088b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 3089b411b363SPhilipp Reisner retry: 3090b411b363SPhilipp Reisner oconn = nconn = mdev->state.conn; 3091b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 3092b411b363SPhilipp Reisner 3093b411b363SPhilipp Reisner if (nconn == C_WF_REPORT_PARAMS) 3094b411b363SPhilipp Reisner nconn = C_CONNECTED; 3095b411b363SPhilipp Reisner 3096b411b363SPhilipp Reisner if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING && 3097b411b363SPhilipp Reisner get_ldev_if_state(mdev, D_NEGOTIATING)) { 3098b411b363SPhilipp Reisner int cr; /* consider resync */ 3099b411b363SPhilipp Reisner 3100b411b363SPhilipp Reisner /* if we established a new connection */ 3101b411b363SPhilipp Reisner cr = (oconn < C_CONNECTED); 3102b411b363SPhilipp Reisner /* if we had an established connection 3103b411b363SPhilipp Reisner * and one of the nodes newly attaches a disk */ 3104b411b363SPhilipp Reisner cr |= (oconn == C_CONNECTED && 3105b411b363SPhilipp Reisner (peer_state.disk == D_NEGOTIATING || 3106b411b363SPhilipp Reisner mdev->state.disk == D_NEGOTIATING)); 3107b411b363SPhilipp Reisner /* if we have both been inconsistent, and the peer has been 3108b411b363SPhilipp Reisner * forced to be UpToDate with --overwrite-data */ 3109b411b363SPhilipp Reisner cr |= test_bit(CONSIDER_RESYNC, &mdev->flags); 3110b411b363SPhilipp Reisner /* if we had been plain connected, and the admin requested to 3111b411b363SPhilipp Reisner * start a sync by "invalidate" or "invalidate-remote" */ 3112b411b363SPhilipp Reisner cr |= (oconn == C_CONNECTED && 3113b411b363SPhilipp Reisner (peer_state.conn >= C_STARTING_SYNC_S && 3114b411b363SPhilipp Reisner peer_state.conn <= C_WF_BITMAP_T)); 3115b411b363SPhilipp Reisner 3116b411b363SPhilipp Reisner if (cr) 3117b411b363SPhilipp Reisner nconn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk); 3118b411b363SPhilipp Reisner 3119b411b363SPhilipp Reisner put_ldev(mdev); 3120b411b363SPhilipp Reisner if (nconn == C_MASK) { 3121b411b363SPhilipp Reisner if (mdev->state.disk == D_NEGOTIATING) { 3122b411b363SPhilipp Reisner drbd_force_state(mdev, NS(disk, D_DISKLESS)); 3123b411b363SPhilipp Reisner nconn = C_CONNECTED; 3124b411b363SPhilipp Reisner } else if (peer_state.disk == D_NEGOTIATING) { 3125b411b363SPhilipp Reisner dev_err(DEV, "Disk attach process on the peer node was aborted.\n"); 3126b411b363SPhilipp Reisner peer_state.disk = D_DISKLESS; 3127b411b363SPhilipp Reisner } else { 3128b411b363SPhilipp Reisner D_ASSERT(oconn == C_WF_REPORT_PARAMS); 3129b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 3130b411b363SPhilipp Reisner return FALSE; 3131b411b363SPhilipp Reisner } 3132b411b363SPhilipp Reisner } 3133b411b363SPhilipp Reisner } 3134b411b363SPhilipp Reisner 3135b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 3136b411b363SPhilipp Reisner if (mdev->state.conn != oconn) 3137b411b363SPhilipp Reisner goto retry; 3138b411b363SPhilipp Reisner clear_bit(CONSIDER_RESYNC, &mdev->flags); 3139b411b363SPhilipp Reisner ns.i = mdev->state.i; 3140b411b363SPhilipp Reisner ns.conn = nconn; 3141b411b363SPhilipp Reisner ns.peer = peer_state.role; 3142b411b363SPhilipp Reisner ns.pdsk = real_peer_disk; 3143b411b363SPhilipp Reisner ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); 3144b411b363SPhilipp Reisner if ((nconn == C_CONNECTED || nconn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) 3145b411b363SPhilipp Reisner ns.disk = mdev->new_state_tmp.disk; 3146b411b363SPhilipp Reisner 3147b411b363SPhilipp Reisner rv = _drbd_set_state(mdev, ns, CS_VERBOSE | CS_HARD, NULL); 3148b411b363SPhilipp Reisner ns = mdev->state; 3149b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 3150b411b363SPhilipp Reisner 3151b411b363SPhilipp Reisner if (rv < SS_SUCCESS) { 3152b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 3153b411b363SPhilipp Reisner return FALSE; 3154b411b363SPhilipp Reisner } 3155b411b363SPhilipp Reisner 3156b411b363SPhilipp Reisner if (oconn > C_WF_REPORT_PARAMS) { 3157b411b363SPhilipp Reisner if (nconn > C_CONNECTED && peer_state.conn <= C_CONNECTED && 3158b411b363SPhilipp Reisner peer_state.disk != D_NEGOTIATING ) { 3159b411b363SPhilipp Reisner /* we want resync, peer has not yet decided to sync... */ 3160b411b363SPhilipp Reisner /* Nowadays only used when forcing a node into primary role and 3161b411b363SPhilipp Reisner setting its disk to UpToDate with that */ 3162b411b363SPhilipp Reisner drbd_send_uuids(mdev); 3163b411b363SPhilipp Reisner drbd_send_state(mdev); 3164b411b363SPhilipp Reisner } 3165b411b363SPhilipp Reisner } 3166b411b363SPhilipp Reisner 3167b411b363SPhilipp Reisner mdev->net_conf->want_lose = 0; 3168b411b363SPhilipp Reisner 3169b411b363SPhilipp Reisner drbd_md_sync(mdev); /* update connected indicator, la_size, ... */ 3170b411b363SPhilipp Reisner 3171b411b363SPhilipp Reisner return TRUE; 3172b411b363SPhilipp Reisner } 3173b411b363SPhilipp Reisner 3174b411b363SPhilipp Reisner static int receive_sync_uuid(struct drbd_conf *mdev, struct p_header *h) 3175b411b363SPhilipp Reisner { 3176b411b363SPhilipp Reisner struct p_rs_uuid *p = (struct p_rs_uuid *)h; 3177b411b363SPhilipp Reisner 3178b411b363SPhilipp Reisner wait_event(mdev->misc_wait, 3179b411b363SPhilipp Reisner mdev->state.conn == C_WF_SYNC_UUID || 3180b411b363SPhilipp Reisner mdev->state.conn < C_CONNECTED || 3181b411b363SPhilipp Reisner mdev->state.disk < D_NEGOTIATING); 3182b411b363SPhilipp Reisner 3183b411b363SPhilipp Reisner /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */ 3184b411b363SPhilipp Reisner 3185b411b363SPhilipp Reisner ERR_IF(h->length != (sizeof(*p)-sizeof(*h))) return FALSE; 3186b411b363SPhilipp Reisner if (drbd_recv(mdev, h->payload, h->length) != h->length) 3187b411b363SPhilipp Reisner return FALSE; 3188b411b363SPhilipp Reisner 3189b411b363SPhilipp Reisner /* Here the _drbd_uuid_ functions are right, current should 3190b411b363SPhilipp Reisner _not_ be rotated into the history */ 3191b411b363SPhilipp Reisner if (get_ldev_if_state(mdev, D_NEGOTIATING)) { 3192b411b363SPhilipp Reisner _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid)); 3193b411b363SPhilipp Reisner _drbd_uuid_set(mdev, UI_BITMAP, 0UL); 3194b411b363SPhilipp Reisner 3195b411b363SPhilipp Reisner drbd_start_resync(mdev, C_SYNC_TARGET); 3196b411b363SPhilipp Reisner 3197b411b363SPhilipp Reisner put_ldev(mdev); 3198b411b363SPhilipp Reisner } else 3199b411b363SPhilipp Reisner dev_err(DEV, "Ignoring SyncUUID packet!\n"); 3200b411b363SPhilipp Reisner 3201b411b363SPhilipp Reisner return TRUE; 3202b411b363SPhilipp Reisner } 3203b411b363SPhilipp Reisner 3204b411b363SPhilipp Reisner enum receive_bitmap_ret { OK, DONE, FAILED }; 3205b411b363SPhilipp Reisner 3206b411b363SPhilipp Reisner static enum receive_bitmap_ret 3207b411b363SPhilipp Reisner receive_bitmap_plain(struct drbd_conf *mdev, struct p_header *h, 3208b411b363SPhilipp Reisner unsigned long *buffer, struct bm_xfer_ctx *c) 3209b411b363SPhilipp Reisner { 3210b411b363SPhilipp Reisner unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset); 3211b411b363SPhilipp Reisner unsigned want = num_words * sizeof(long); 3212b411b363SPhilipp Reisner 3213b411b363SPhilipp Reisner if (want != h->length) { 3214b411b363SPhilipp Reisner dev_err(DEV, "%s:want (%u) != h->length (%u)\n", __func__, want, h->length); 3215b411b363SPhilipp Reisner return FAILED; 3216b411b363SPhilipp Reisner } 3217b411b363SPhilipp Reisner if (want == 0) 3218b411b363SPhilipp Reisner return DONE; 3219b411b363SPhilipp Reisner if (drbd_recv(mdev, buffer, want) != want) 3220b411b363SPhilipp Reisner return FAILED; 3221b411b363SPhilipp Reisner 3222b411b363SPhilipp Reisner drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer); 3223b411b363SPhilipp Reisner 3224b411b363SPhilipp Reisner c->word_offset += num_words; 3225b411b363SPhilipp Reisner c->bit_offset = c->word_offset * BITS_PER_LONG; 3226b411b363SPhilipp Reisner if (c->bit_offset > c->bm_bits) 3227b411b363SPhilipp Reisner c->bit_offset = c->bm_bits; 3228b411b363SPhilipp Reisner 3229b411b363SPhilipp Reisner return OK; 3230b411b363SPhilipp Reisner } 3231b411b363SPhilipp Reisner 3232b411b363SPhilipp Reisner static enum receive_bitmap_ret 3233b411b363SPhilipp Reisner recv_bm_rle_bits(struct drbd_conf *mdev, 3234b411b363SPhilipp Reisner struct p_compressed_bm *p, 3235b411b363SPhilipp Reisner struct bm_xfer_ctx *c) 3236b411b363SPhilipp Reisner { 3237b411b363SPhilipp Reisner struct bitstream bs; 3238b411b363SPhilipp Reisner u64 look_ahead; 3239b411b363SPhilipp Reisner u64 rl; 3240b411b363SPhilipp Reisner u64 tmp; 3241b411b363SPhilipp Reisner unsigned long s = c->bit_offset; 3242b411b363SPhilipp Reisner unsigned long e; 3243b411b363SPhilipp Reisner int len = p->head.length - (sizeof(*p) - sizeof(p->head)); 3244b411b363SPhilipp Reisner int toggle = DCBP_get_start(p); 3245b411b363SPhilipp Reisner int have; 3246b411b363SPhilipp Reisner int bits; 3247b411b363SPhilipp Reisner 3248b411b363SPhilipp Reisner bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p)); 3249b411b363SPhilipp Reisner 3250b411b363SPhilipp Reisner bits = bitstream_get_bits(&bs, &look_ahead, 64); 3251b411b363SPhilipp Reisner if (bits < 0) 3252b411b363SPhilipp Reisner return FAILED; 3253b411b363SPhilipp Reisner 3254b411b363SPhilipp Reisner for (have = bits; have > 0; s += rl, toggle = !toggle) { 3255b411b363SPhilipp Reisner bits = vli_decode_bits(&rl, look_ahead); 3256b411b363SPhilipp Reisner if (bits <= 0) 3257b411b363SPhilipp Reisner return FAILED; 3258b411b363SPhilipp Reisner 3259b411b363SPhilipp Reisner if (toggle) { 3260b411b363SPhilipp Reisner e = s + rl -1; 3261b411b363SPhilipp Reisner if (e >= c->bm_bits) { 3262b411b363SPhilipp Reisner dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); 3263b411b363SPhilipp Reisner return FAILED; 3264b411b363SPhilipp Reisner } 3265b411b363SPhilipp Reisner _drbd_bm_set_bits(mdev, s, e); 3266b411b363SPhilipp Reisner } 3267b411b363SPhilipp Reisner 3268b411b363SPhilipp Reisner if (have < bits) { 3269b411b363SPhilipp Reisner dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", 3270b411b363SPhilipp Reisner have, bits, look_ahead, 3271b411b363SPhilipp Reisner (unsigned int)(bs.cur.b - p->code), 3272b411b363SPhilipp Reisner (unsigned int)bs.buf_len); 3273b411b363SPhilipp Reisner return FAILED; 3274b411b363SPhilipp Reisner } 3275b411b363SPhilipp Reisner look_ahead >>= bits; 3276b411b363SPhilipp Reisner have -= bits; 3277b411b363SPhilipp Reisner 3278b411b363SPhilipp Reisner bits = bitstream_get_bits(&bs, &tmp, 64 - have); 3279b411b363SPhilipp Reisner if (bits < 0) 3280b411b363SPhilipp Reisner return FAILED; 3281b411b363SPhilipp Reisner look_ahead |= tmp << have; 3282b411b363SPhilipp Reisner have += bits; 3283b411b363SPhilipp Reisner } 3284b411b363SPhilipp Reisner 3285b411b363SPhilipp Reisner c->bit_offset = s; 3286b411b363SPhilipp Reisner bm_xfer_ctx_bit_to_word_offset(c); 3287b411b363SPhilipp Reisner 3288b411b363SPhilipp Reisner return (s == c->bm_bits) ? DONE : OK; 3289b411b363SPhilipp Reisner } 3290b411b363SPhilipp Reisner 3291b411b363SPhilipp Reisner static enum receive_bitmap_ret 3292b411b363SPhilipp Reisner decode_bitmap_c(struct drbd_conf *mdev, 3293b411b363SPhilipp Reisner struct p_compressed_bm *p, 3294b411b363SPhilipp Reisner struct bm_xfer_ctx *c) 3295b411b363SPhilipp Reisner { 3296b411b363SPhilipp Reisner if (DCBP_get_code(p) == RLE_VLI_Bits) 3297b411b363SPhilipp Reisner return recv_bm_rle_bits(mdev, p, c); 3298b411b363SPhilipp Reisner 3299b411b363SPhilipp Reisner /* other variants had been implemented for evaluation, 3300b411b363SPhilipp Reisner * but have been dropped as this one turned out to be "best" 3301b411b363SPhilipp Reisner * during all our tests. */ 3302b411b363SPhilipp Reisner 3303b411b363SPhilipp Reisner dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding); 3304b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); 3305b411b363SPhilipp Reisner return FAILED; 3306b411b363SPhilipp Reisner } 3307b411b363SPhilipp Reisner 3308b411b363SPhilipp Reisner void INFO_bm_xfer_stats(struct drbd_conf *mdev, 3309b411b363SPhilipp Reisner const char *direction, struct bm_xfer_ctx *c) 3310b411b363SPhilipp Reisner { 3311b411b363SPhilipp Reisner /* what would it take to transfer it "plaintext" */ 3312b411b363SPhilipp Reisner unsigned plain = sizeof(struct p_header) * 3313b411b363SPhilipp Reisner ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1) 3314b411b363SPhilipp Reisner + c->bm_words * sizeof(long); 3315b411b363SPhilipp Reisner unsigned total = c->bytes[0] + c->bytes[1]; 3316b411b363SPhilipp Reisner unsigned r; 3317b411b363SPhilipp Reisner 3318b411b363SPhilipp Reisner /* total can not be zero. but just in case: */ 3319b411b363SPhilipp Reisner if (total == 0) 3320b411b363SPhilipp Reisner return; 3321b411b363SPhilipp Reisner 3322b411b363SPhilipp Reisner /* don't report if not compressed */ 3323b411b363SPhilipp Reisner if (total >= plain) 3324b411b363SPhilipp Reisner return; 3325b411b363SPhilipp Reisner 3326b411b363SPhilipp Reisner /* total < plain. check for overflow, still */ 3327b411b363SPhilipp Reisner r = (total > UINT_MAX/1000) ? (total / (plain/1000)) 3328b411b363SPhilipp Reisner : (1000 * total / plain); 3329b411b363SPhilipp Reisner 3330b411b363SPhilipp Reisner if (r > 1000) 3331b411b363SPhilipp Reisner r = 1000; 3332b411b363SPhilipp Reisner 3333b411b363SPhilipp Reisner r = 1000 - r; 3334b411b363SPhilipp Reisner dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " 3335b411b363SPhilipp Reisner "total %u; compression: %u.%u%%\n", 3336b411b363SPhilipp Reisner direction, 3337b411b363SPhilipp Reisner c->bytes[1], c->packets[1], 3338b411b363SPhilipp Reisner c->bytes[0], c->packets[0], 3339b411b363SPhilipp Reisner total, r/10, r % 10); 3340b411b363SPhilipp Reisner } 3341b411b363SPhilipp Reisner 3342b411b363SPhilipp Reisner /* Since we are processing the bitfield from lower addresses to higher, 3343b411b363SPhilipp Reisner it does not matter if the process it in 32 bit chunks or 64 bit 3344b411b363SPhilipp Reisner chunks as long as it is little endian. (Understand it as byte stream, 3345b411b363SPhilipp Reisner beginning with the lowest byte...) If we would use big endian 3346b411b363SPhilipp Reisner we would need to process it from the highest address to the lowest, 3347b411b363SPhilipp Reisner in order to be agnostic to the 32 vs 64 bits issue. 3348b411b363SPhilipp Reisner 3349b411b363SPhilipp Reisner returns 0 on failure, 1 if we successfully received it. */ 3350b411b363SPhilipp Reisner static int receive_bitmap(struct drbd_conf *mdev, struct p_header *h) 3351b411b363SPhilipp Reisner { 3352b411b363SPhilipp Reisner struct bm_xfer_ctx c; 3353b411b363SPhilipp Reisner void *buffer; 3354b411b363SPhilipp Reisner enum receive_bitmap_ret ret; 3355b411b363SPhilipp Reisner int ok = FALSE; 3356b411b363SPhilipp Reisner 3357b411b363SPhilipp Reisner wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt)); 3358b411b363SPhilipp Reisner 3359b411b363SPhilipp Reisner drbd_bm_lock(mdev, "receive bitmap"); 3360b411b363SPhilipp Reisner 3361b411b363SPhilipp Reisner /* maybe we should use some per thread scratch page, 3362b411b363SPhilipp Reisner * and allocate that during initial device creation? */ 3363b411b363SPhilipp Reisner buffer = (unsigned long *) __get_free_page(GFP_NOIO); 3364b411b363SPhilipp Reisner if (!buffer) { 3365b411b363SPhilipp Reisner dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__); 3366b411b363SPhilipp Reisner goto out; 3367b411b363SPhilipp Reisner } 3368b411b363SPhilipp Reisner 3369b411b363SPhilipp Reisner c = (struct bm_xfer_ctx) { 3370b411b363SPhilipp Reisner .bm_bits = drbd_bm_bits(mdev), 3371b411b363SPhilipp Reisner .bm_words = drbd_bm_words(mdev), 3372b411b363SPhilipp Reisner }; 3373b411b363SPhilipp Reisner 3374b411b363SPhilipp Reisner do { 3375b411b363SPhilipp Reisner if (h->command == P_BITMAP) { 3376b411b363SPhilipp Reisner ret = receive_bitmap_plain(mdev, h, buffer, &c); 3377b411b363SPhilipp Reisner } else if (h->command == P_COMPRESSED_BITMAP) { 3378b411b363SPhilipp Reisner /* MAYBE: sanity check that we speak proto >= 90, 3379b411b363SPhilipp Reisner * and the feature is enabled! */ 3380b411b363SPhilipp Reisner struct p_compressed_bm *p; 3381b411b363SPhilipp Reisner 3382b411b363SPhilipp Reisner if (h->length > BM_PACKET_PAYLOAD_BYTES) { 3383b411b363SPhilipp Reisner dev_err(DEV, "ReportCBitmap packet too large\n"); 3384b411b363SPhilipp Reisner goto out; 3385b411b363SPhilipp Reisner } 3386b411b363SPhilipp Reisner /* use the page buff */ 3387b411b363SPhilipp Reisner p = buffer; 3388b411b363SPhilipp Reisner memcpy(p, h, sizeof(*h)); 3389b411b363SPhilipp Reisner if (drbd_recv(mdev, p->head.payload, h->length) != h->length) 3390b411b363SPhilipp Reisner goto out; 3391b411b363SPhilipp Reisner if (p->head.length <= (sizeof(*p) - sizeof(p->head))) { 3392b411b363SPhilipp Reisner dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", p->head.length); 3393b411b363SPhilipp Reisner return FAILED; 3394b411b363SPhilipp Reisner } 3395b411b363SPhilipp Reisner ret = decode_bitmap_c(mdev, p, &c); 3396b411b363SPhilipp Reisner } else { 3397b411b363SPhilipp Reisner dev_warn(DEV, "receive_bitmap: h->command neither ReportBitMap nor ReportCBitMap (is 0x%x)", h->command); 3398b411b363SPhilipp Reisner goto out; 3399b411b363SPhilipp Reisner } 3400b411b363SPhilipp Reisner 3401b411b363SPhilipp Reisner c.packets[h->command == P_BITMAP]++; 3402b411b363SPhilipp Reisner c.bytes[h->command == P_BITMAP] += sizeof(struct p_header) + h->length; 3403b411b363SPhilipp Reisner 3404b411b363SPhilipp Reisner if (ret != OK) 3405b411b363SPhilipp Reisner break; 3406b411b363SPhilipp Reisner 3407b411b363SPhilipp Reisner if (!drbd_recv_header(mdev, h)) 3408b411b363SPhilipp Reisner goto out; 3409b411b363SPhilipp Reisner } while (ret == OK); 3410b411b363SPhilipp Reisner if (ret == FAILED) 3411b411b363SPhilipp Reisner goto out; 3412b411b363SPhilipp Reisner 3413b411b363SPhilipp Reisner INFO_bm_xfer_stats(mdev, "receive", &c); 3414b411b363SPhilipp Reisner 3415b411b363SPhilipp Reisner if (mdev->state.conn == C_WF_BITMAP_T) { 3416b411b363SPhilipp Reisner ok = !drbd_send_bitmap(mdev); 3417b411b363SPhilipp Reisner if (!ok) 3418b411b363SPhilipp Reisner goto out; 3419b411b363SPhilipp Reisner /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ 3420b411b363SPhilipp Reisner ok = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); 3421b411b363SPhilipp Reisner D_ASSERT(ok == SS_SUCCESS); 3422b411b363SPhilipp Reisner } else if (mdev->state.conn != C_WF_BITMAP_S) { 3423b411b363SPhilipp Reisner /* admin may have requested C_DISCONNECTING, 3424b411b363SPhilipp Reisner * other threads may have noticed network errors */ 3425b411b363SPhilipp Reisner dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n", 3426b411b363SPhilipp Reisner drbd_conn_str(mdev->state.conn)); 3427b411b363SPhilipp Reisner } 3428b411b363SPhilipp Reisner 3429b411b363SPhilipp Reisner ok = TRUE; 3430b411b363SPhilipp Reisner out: 3431b411b363SPhilipp Reisner drbd_bm_unlock(mdev); 3432b411b363SPhilipp Reisner if (ok && mdev->state.conn == C_WF_BITMAP_S) 3433b411b363SPhilipp Reisner drbd_start_resync(mdev, C_SYNC_SOURCE); 3434b411b363SPhilipp Reisner free_page((unsigned long) buffer); 3435b411b363SPhilipp Reisner return ok; 3436b411b363SPhilipp Reisner } 3437b411b363SPhilipp Reisner 3438b411b363SPhilipp Reisner static int receive_skip(struct drbd_conf *mdev, struct p_header *h) 3439b411b363SPhilipp Reisner { 3440b411b363SPhilipp Reisner /* TODO zero copy sink :) */ 3441b411b363SPhilipp Reisner static char sink[128]; 3442b411b363SPhilipp Reisner int size, want, r; 3443b411b363SPhilipp Reisner 3444b411b363SPhilipp Reisner dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n", 3445b411b363SPhilipp Reisner h->command, h->length); 3446b411b363SPhilipp Reisner 3447b411b363SPhilipp Reisner size = h->length; 3448b411b363SPhilipp Reisner while (size > 0) { 3449b411b363SPhilipp Reisner want = min_t(int, size, sizeof(sink)); 3450b411b363SPhilipp Reisner r = drbd_recv(mdev, sink, want); 3451b411b363SPhilipp Reisner ERR_IF(r <= 0) break; 3452b411b363SPhilipp Reisner size -= r; 3453b411b363SPhilipp Reisner } 3454b411b363SPhilipp Reisner return size == 0; 3455b411b363SPhilipp Reisner } 3456b411b363SPhilipp Reisner 3457b411b363SPhilipp Reisner static int receive_UnplugRemote(struct drbd_conf *mdev, struct p_header *h) 3458b411b363SPhilipp Reisner { 3459b411b363SPhilipp Reisner if (mdev->state.disk >= D_INCONSISTENT) 3460b411b363SPhilipp Reisner drbd_kick_lo(mdev); 3461b411b363SPhilipp Reisner 3462b411b363SPhilipp Reisner /* Make sure we've acked all the TCP data associated 3463b411b363SPhilipp Reisner * with the data requests being unplugged */ 3464b411b363SPhilipp Reisner drbd_tcp_quickack(mdev->data.socket); 3465b411b363SPhilipp Reisner 3466b411b363SPhilipp Reisner return TRUE; 3467b411b363SPhilipp Reisner } 3468b411b363SPhilipp Reisner 3469b411b363SPhilipp Reisner typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, struct p_header *); 3470b411b363SPhilipp Reisner 3471b411b363SPhilipp Reisner static drbd_cmd_handler_f drbd_default_handler[] = { 3472b411b363SPhilipp Reisner [P_DATA] = receive_Data, 3473b411b363SPhilipp Reisner [P_DATA_REPLY] = receive_DataReply, 3474b411b363SPhilipp Reisner [P_RS_DATA_REPLY] = receive_RSDataReply, 3475b411b363SPhilipp Reisner [P_BARRIER] = receive_Barrier, 3476b411b363SPhilipp Reisner [P_BITMAP] = receive_bitmap, 3477b411b363SPhilipp Reisner [P_COMPRESSED_BITMAP] = receive_bitmap, 3478b411b363SPhilipp Reisner [P_UNPLUG_REMOTE] = receive_UnplugRemote, 3479b411b363SPhilipp Reisner [P_DATA_REQUEST] = receive_DataRequest, 3480b411b363SPhilipp Reisner [P_RS_DATA_REQUEST] = receive_DataRequest, 3481b411b363SPhilipp Reisner [P_SYNC_PARAM] = receive_SyncParam, 3482b411b363SPhilipp Reisner [P_SYNC_PARAM89] = receive_SyncParam, 3483b411b363SPhilipp Reisner [P_PROTOCOL] = receive_protocol, 3484b411b363SPhilipp Reisner [P_UUIDS] = receive_uuids, 3485b411b363SPhilipp Reisner [P_SIZES] = receive_sizes, 3486b411b363SPhilipp Reisner [P_STATE] = receive_state, 3487b411b363SPhilipp Reisner [P_STATE_CHG_REQ] = receive_req_state, 3488b411b363SPhilipp Reisner [P_SYNC_UUID] = receive_sync_uuid, 3489b411b363SPhilipp Reisner [P_OV_REQUEST] = receive_DataRequest, 3490b411b363SPhilipp Reisner [P_OV_REPLY] = receive_DataRequest, 3491b411b363SPhilipp Reisner [P_CSUM_RS_REQUEST] = receive_DataRequest, 3492b411b363SPhilipp Reisner /* anything missing from this table is in 3493b411b363SPhilipp Reisner * the asender_tbl, see get_asender_cmd */ 3494b411b363SPhilipp Reisner [P_MAX_CMD] = NULL, 3495b411b363SPhilipp Reisner }; 3496b411b363SPhilipp Reisner 3497b411b363SPhilipp Reisner static drbd_cmd_handler_f *drbd_cmd_handler = drbd_default_handler; 3498b411b363SPhilipp Reisner static drbd_cmd_handler_f *drbd_opt_cmd_handler; 3499b411b363SPhilipp Reisner 3500b411b363SPhilipp Reisner static void drbdd(struct drbd_conf *mdev) 3501b411b363SPhilipp Reisner { 3502b411b363SPhilipp Reisner drbd_cmd_handler_f handler; 3503b411b363SPhilipp Reisner struct p_header *header = &mdev->data.rbuf.header; 3504b411b363SPhilipp Reisner 3505b411b363SPhilipp Reisner while (get_t_state(&mdev->receiver) == Running) { 3506b411b363SPhilipp Reisner drbd_thread_current_set_cpu(mdev); 35070b33a916SLars Ellenberg if (!drbd_recv_header(mdev, header)) { 35080b33a916SLars Ellenberg drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); 3509b411b363SPhilipp Reisner break; 35100b33a916SLars Ellenberg } 3511b411b363SPhilipp Reisner 3512b411b363SPhilipp Reisner if (header->command < P_MAX_CMD) 3513b411b363SPhilipp Reisner handler = drbd_cmd_handler[header->command]; 3514b411b363SPhilipp Reisner else if (P_MAY_IGNORE < header->command 3515b411b363SPhilipp Reisner && header->command < P_MAX_OPT_CMD) 3516b411b363SPhilipp Reisner handler = drbd_opt_cmd_handler[header->command-P_MAY_IGNORE]; 3517b411b363SPhilipp Reisner else if (header->command > P_MAX_OPT_CMD) 3518b411b363SPhilipp Reisner handler = receive_skip; 3519b411b363SPhilipp Reisner else 3520b411b363SPhilipp Reisner handler = NULL; 3521b411b363SPhilipp Reisner 3522b411b363SPhilipp Reisner if (unlikely(!handler)) { 3523b411b363SPhilipp Reisner dev_err(DEV, "unknown packet type %d, l: %d!\n", 3524b411b363SPhilipp Reisner header->command, header->length); 3525b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); 3526b411b363SPhilipp Reisner break; 3527b411b363SPhilipp Reisner } 3528b411b363SPhilipp Reisner if (unlikely(!handler(mdev, header))) { 3529b411b363SPhilipp Reisner dev_err(DEV, "error receiving %s, l: %d!\n", 3530b411b363SPhilipp Reisner cmdname(header->command), header->length); 3531b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); 3532b411b363SPhilipp Reisner break; 3533b411b363SPhilipp Reisner } 3534b411b363SPhilipp Reisner } 3535b411b363SPhilipp Reisner } 3536b411b363SPhilipp Reisner 3537b411b363SPhilipp Reisner static void drbd_fail_pending_reads(struct drbd_conf *mdev) 3538b411b363SPhilipp Reisner { 3539b411b363SPhilipp Reisner struct hlist_head *slot; 3540b411b363SPhilipp Reisner struct hlist_node *pos; 3541b411b363SPhilipp Reisner struct hlist_node *tmp; 3542b411b363SPhilipp Reisner struct drbd_request *req; 3543b411b363SPhilipp Reisner int i; 3544b411b363SPhilipp Reisner 3545b411b363SPhilipp Reisner /* 3546b411b363SPhilipp Reisner * Application READ requests 3547b411b363SPhilipp Reisner */ 3548b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 3549b411b363SPhilipp Reisner for (i = 0; i < APP_R_HSIZE; i++) { 3550b411b363SPhilipp Reisner slot = mdev->app_reads_hash+i; 3551b411b363SPhilipp Reisner hlist_for_each_entry_safe(req, pos, tmp, slot, colision) { 3552b411b363SPhilipp Reisner /* it may (but should not any longer!) 3553b411b363SPhilipp Reisner * be on the work queue; if that assert triggers, 3554b411b363SPhilipp Reisner * we need to also grab the 3555b411b363SPhilipp Reisner * spin_lock_irq(&mdev->data.work.q_lock); 3556b411b363SPhilipp Reisner * and list_del_init here. */ 3557b411b363SPhilipp Reisner D_ASSERT(list_empty(&req->w.list)); 3558b411b363SPhilipp Reisner /* It would be nice to complete outside of spinlock. 3559b411b363SPhilipp Reisner * But this is easier for now. */ 3560b411b363SPhilipp Reisner _req_mod(req, connection_lost_while_pending); 3561b411b363SPhilipp Reisner } 3562b411b363SPhilipp Reisner } 3563b411b363SPhilipp Reisner for (i = 0; i < APP_R_HSIZE; i++) 3564b411b363SPhilipp Reisner if (!hlist_empty(mdev->app_reads_hash+i)) 3565b411b363SPhilipp Reisner dev_warn(DEV, "ASSERT FAILED: app_reads_hash[%d].first: " 3566b411b363SPhilipp Reisner "%p, should be NULL\n", i, mdev->app_reads_hash[i].first); 3567b411b363SPhilipp Reisner 3568b411b363SPhilipp Reisner memset(mdev->app_reads_hash, 0, APP_R_HSIZE*sizeof(void *)); 3569b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 3570b411b363SPhilipp Reisner } 3571b411b363SPhilipp Reisner 3572b411b363SPhilipp Reisner void drbd_flush_workqueue(struct drbd_conf *mdev) 3573b411b363SPhilipp Reisner { 3574b411b363SPhilipp Reisner struct drbd_wq_barrier barr; 3575b411b363SPhilipp Reisner 3576b411b363SPhilipp Reisner barr.w.cb = w_prev_work_done; 3577b411b363SPhilipp Reisner init_completion(&barr.done); 3578b411b363SPhilipp Reisner drbd_queue_work(&mdev->data.work, &barr.w); 3579b411b363SPhilipp Reisner wait_for_completion(&barr.done); 3580b411b363SPhilipp Reisner } 3581b411b363SPhilipp Reisner 3582b411b363SPhilipp Reisner static void drbd_disconnect(struct drbd_conf *mdev) 3583b411b363SPhilipp Reisner { 3584b411b363SPhilipp Reisner enum drbd_fencing_p fp; 3585b411b363SPhilipp Reisner union drbd_state os, ns; 3586b411b363SPhilipp Reisner int rv = SS_UNKNOWN_ERROR; 3587b411b363SPhilipp Reisner unsigned int i; 3588b411b363SPhilipp Reisner 3589b411b363SPhilipp Reisner if (mdev->state.conn == C_STANDALONE) 3590b411b363SPhilipp Reisner return; 3591b411b363SPhilipp Reisner if (mdev->state.conn >= C_WF_CONNECTION) 3592b411b363SPhilipp Reisner dev_err(DEV, "ASSERT FAILED cstate = %s, expected < WFConnection\n", 3593b411b363SPhilipp Reisner drbd_conn_str(mdev->state.conn)); 3594b411b363SPhilipp Reisner 3595b411b363SPhilipp Reisner /* asender does not clean up anything. it must not interfere, either */ 3596b411b363SPhilipp Reisner drbd_thread_stop(&mdev->asender); 3597b411b363SPhilipp Reisner 3598b411b363SPhilipp Reisner mutex_lock(&mdev->data.mutex); 3599b411b363SPhilipp Reisner drbd_free_sock(mdev); 3600b411b363SPhilipp Reisner mutex_unlock(&mdev->data.mutex); 3601b411b363SPhilipp Reisner 3602b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 3603b411b363SPhilipp Reisner _drbd_wait_ee_list_empty(mdev, &mdev->active_ee); 3604b411b363SPhilipp Reisner _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee); 3605b411b363SPhilipp Reisner _drbd_wait_ee_list_empty(mdev, &mdev->read_ee); 3606b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 3607b411b363SPhilipp Reisner 3608b411b363SPhilipp Reisner /* We do not have data structures that would allow us to 3609b411b363SPhilipp Reisner * get the rs_pending_cnt down to 0 again. 3610b411b363SPhilipp Reisner * * On C_SYNC_TARGET we do not have any data structures describing 3611b411b363SPhilipp Reisner * the pending RSDataRequest's we have sent. 3612b411b363SPhilipp Reisner * * On C_SYNC_SOURCE there is no data structure that tracks 3613b411b363SPhilipp Reisner * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. 3614b411b363SPhilipp Reisner * And no, it is not the sum of the reference counts in the 3615b411b363SPhilipp Reisner * resync_LRU. The resync_LRU tracks the whole operation including 3616b411b363SPhilipp Reisner * the disk-IO, while the rs_pending_cnt only tracks the blocks 3617b411b363SPhilipp Reisner * on the fly. */ 3618b411b363SPhilipp Reisner drbd_rs_cancel_all(mdev); 3619b411b363SPhilipp Reisner mdev->rs_total = 0; 3620b411b363SPhilipp Reisner mdev->rs_failed = 0; 3621b411b363SPhilipp Reisner atomic_set(&mdev->rs_pending_cnt, 0); 3622b411b363SPhilipp Reisner wake_up(&mdev->misc_wait); 3623b411b363SPhilipp Reisner 3624b411b363SPhilipp Reisner /* make sure syncer is stopped and w_resume_next_sg queued */ 3625b411b363SPhilipp Reisner del_timer_sync(&mdev->resync_timer); 3626b411b363SPhilipp Reisner set_bit(STOP_SYNC_TIMER, &mdev->flags); 3627b411b363SPhilipp Reisner resync_timer_fn((unsigned long)mdev); 3628b411b363SPhilipp Reisner 3629b411b363SPhilipp Reisner /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, 3630b411b363SPhilipp Reisner * w_make_resync_request etc. which may still be on the worker queue 3631b411b363SPhilipp Reisner * to be "canceled" */ 3632b411b363SPhilipp Reisner drbd_flush_workqueue(mdev); 3633b411b363SPhilipp Reisner 3634b411b363SPhilipp Reisner /* This also does reclaim_net_ee(). If we do this too early, we might 3635b411b363SPhilipp Reisner * miss some resync ee and pages.*/ 3636b411b363SPhilipp Reisner drbd_process_done_ee(mdev); 3637b411b363SPhilipp Reisner 3638b411b363SPhilipp Reisner kfree(mdev->p_uuid); 3639b411b363SPhilipp Reisner mdev->p_uuid = NULL; 3640b411b363SPhilipp Reisner 3641b411b363SPhilipp Reisner if (!mdev->state.susp) 3642b411b363SPhilipp Reisner tl_clear(mdev); 3643b411b363SPhilipp Reisner 3644b411b363SPhilipp Reisner drbd_fail_pending_reads(mdev); 3645b411b363SPhilipp Reisner 3646b411b363SPhilipp Reisner dev_info(DEV, "Connection closed\n"); 3647b411b363SPhilipp Reisner 3648b411b363SPhilipp Reisner drbd_md_sync(mdev); 3649b411b363SPhilipp Reisner 3650b411b363SPhilipp Reisner fp = FP_DONT_CARE; 3651b411b363SPhilipp Reisner if (get_ldev(mdev)) { 3652b411b363SPhilipp Reisner fp = mdev->ldev->dc.fencing; 3653b411b363SPhilipp Reisner put_ldev(mdev); 3654b411b363SPhilipp Reisner } 3655b411b363SPhilipp Reisner 3656b411b363SPhilipp Reisner if (mdev->state.role == R_PRIMARY) { 3657b411b363SPhilipp Reisner if (fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN) { 3658b411b363SPhilipp Reisner enum drbd_disk_state nps = drbd_try_outdate_peer(mdev); 3659b411b363SPhilipp Reisner drbd_request_state(mdev, NS(pdsk, nps)); 3660b411b363SPhilipp Reisner } 3661b411b363SPhilipp Reisner } 3662b411b363SPhilipp Reisner 3663b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 3664b411b363SPhilipp Reisner os = mdev->state; 3665b411b363SPhilipp Reisner if (os.conn >= C_UNCONNECTED) { 3666b411b363SPhilipp Reisner /* Do not restart in case we are C_DISCONNECTING */ 3667b411b363SPhilipp Reisner ns = os; 3668b411b363SPhilipp Reisner ns.conn = C_UNCONNECTED; 3669b411b363SPhilipp Reisner rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); 3670b411b363SPhilipp Reisner } 3671b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 3672b411b363SPhilipp Reisner 3673b411b363SPhilipp Reisner if (os.conn == C_DISCONNECTING) { 3674b411b363SPhilipp Reisner struct hlist_head *h; 3675b411b363SPhilipp Reisner wait_event(mdev->misc_wait, atomic_read(&mdev->net_cnt) == 0); 3676b411b363SPhilipp Reisner 3677b411b363SPhilipp Reisner /* we must not free the tl_hash 3678b411b363SPhilipp Reisner * while application io is still on the fly */ 3679b411b363SPhilipp Reisner wait_event(mdev->misc_wait, atomic_read(&mdev->ap_bio_cnt) == 0); 3680b411b363SPhilipp Reisner 3681b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 3682b411b363SPhilipp Reisner /* paranoia code */ 3683b411b363SPhilipp Reisner for (h = mdev->ee_hash; h < mdev->ee_hash + mdev->ee_hash_s; h++) 3684b411b363SPhilipp Reisner if (h->first) 3685b411b363SPhilipp Reisner dev_err(DEV, "ASSERT FAILED ee_hash[%u].first == %p, expected NULL\n", 3686b411b363SPhilipp Reisner (int)(h - mdev->ee_hash), h->first); 3687b411b363SPhilipp Reisner kfree(mdev->ee_hash); 3688b411b363SPhilipp Reisner mdev->ee_hash = NULL; 3689b411b363SPhilipp Reisner mdev->ee_hash_s = 0; 3690b411b363SPhilipp Reisner 3691b411b363SPhilipp Reisner /* paranoia code */ 3692b411b363SPhilipp Reisner for (h = mdev->tl_hash; h < mdev->tl_hash + mdev->tl_hash_s; h++) 3693b411b363SPhilipp Reisner if (h->first) 3694b411b363SPhilipp Reisner dev_err(DEV, "ASSERT FAILED tl_hash[%u] == %p, expected NULL\n", 3695b411b363SPhilipp Reisner (int)(h - mdev->tl_hash), h->first); 3696b411b363SPhilipp Reisner kfree(mdev->tl_hash); 3697b411b363SPhilipp Reisner mdev->tl_hash = NULL; 3698b411b363SPhilipp Reisner mdev->tl_hash_s = 0; 3699b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 3700b411b363SPhilipp Reisner 3701b411b363SPhilipp Reisner crypto_free_hash(mdev->cram_hmac_tfm); 3702b411b363SPhilipp Reisner mdev->cram_hmac_tfm = NULL; 3703b411b363SPhilipp Reisner 3704b411b363SPhilipp Reisner kfree(mdev->net_conf); 3705b411b363SPhilipp Reisner mdev->net_conf = NULL; 3706b411b363SPhilipp Reisner drbd_request_state(mdev, NS(conn, C_STANDALONE)); 3707b411b363SPhilipp Reisner } 3708b411b363SPhilipp Reisner 3709b411b363SPhilipp Reisner /* tcp_close and release of sendpage pages can be deferred. I don't 3710b411b363SPhilipp Reisner * want to use SO_LINGER, because apparently it can be deferred for 3711b411b363SPhilipp Reisner * more than 20 seconds (longest time I checked). 3712b411b363SPhilipp Reisner * 3713b411b363SPhilipp Reisner * Actually we don't care for exactly when the network stack does its 3714b411b363SPhilipp Reisner * put_page(), but release our reference on these pages right here. 3715b411b363SPhilipp Reisner */ 3716b411b363SPhilipp Reisner i = drbd_release_ee(mdev, &mdev->net_ee); 3717b411b363SPhilipp Reisner if (i) 3718b411b363SPhilipp Reisner dev_info(DEV, "net_ee not empty, killed %u entries\n", i); 3719b411b363SPhilipp Reisner i = atomic_read(&mdev->pp_in_use); 3720b411b363SPhilipp Reisner if (i) 3721b411b363SPhilipp Reisner dev_info(DEV, "pp_in_use = %u, expected 0\n", i); 3722b411b363SPhilipp Reisner 3723b411b363SPhilipp Reisner D_ASSERT(list_empty(&mdev->read_ee)); 3724b411b363SPhilipp Reisner D_ASSERT(list_empty(&mdev->active_ee)); 3725b411b363SPhilipp Reisner D_ASSERT(list_empty(&mdev->sync_ee)); 3726b411b363SPhilipp Reisner D_ASSERT(list_empty(&mdev->done_ee)); 3727b411b363SPhilipp Reisner 3728b411b363SPhilipp Reisner /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ 3729b411b363SPhilipp Reisner atomic_set(&mdev->current_epoch->epoch_size, 0); 3730b411b363SPhilipp Reisner D_ASSERT(list_empty(&mdev->current_epoch->list)); 3731b411b363SPhilipp Reisner } 3732b411b363SPhilipp Reisner 3733b411b363SPhilipp Reisner /* 3734b411b363SPhilipp Reisner * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version 3735b411b363SPhilipp Reisner * we can agree on is stored in agreed_pro_version. 3736b411b363SPhilipp Reisner * 3737b411b363SPhilipp Reisner * feature flags and the reserved array should be enough room for future 3738b411b363SPhilipp Reisner * enhancements of the handshake protocol, and possible plugins... 3739b411b363SPhilipp Reisner * 3740b411b363SPhilipp Reisner * for now, they are expected to be zero, but ignored. 3741b411b363SPhilipp Reisner */ 3742b411b363SPhilipp Reisner static int drbd_send_handshake(struct drbd_conf *mdev) 3743b411b363SPhilipp Reisner { 3744b411b363SPhilipp Reisner /* ASSERT current == mdev->receiver ... */ 3745b411b363SPhilipp Reisner struct p_handshake *p = &mdev->data.sbuf.handshake; 3746b411b363SPhilipp Reisner int ok; 3747b411b363SPhilipp Reisner 3748b411b363SPhilipp Reisner if (mutex_lock_interruptible(&mdev->data.mutex)) { 3749b411b363SPhilipp Reisner dev_err(DEV, "interrupted during initial handshake\n"); 3750b411b363SPhilipp Reisner return 0; /* interrupted. not ok. */ 3751b411b363SPhilipp Reisner } 3752b411b363SPhilipp Reisner 3753b411b363SPhilipp Reisner if (mdev->data.socket == NULL) { 3754b411b363SPhilipp Reisner mutex_unlock(&mdev->data.mutex); 3755b411b363SPhilipp Reisner return 0; 3756b411b363SPhilipp Reisner } 3757b411b363SPhilipp Reisner 3758b411b363SPhilipp Reisner memset(p, 0, sizeof(*p)); 3759b411b363SPhilipp Reisner p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); 3760b411b363SPhilipp Reisner p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); 3761b411b363SPhilipp Reisner ok = _drbd_send_cmd( mdev, mdev->data.socket, P_HAND_SHAKE, 3762b411b363SPhilipp Reisner (struct p_header *)p, sizeof(*p), 0 ); 3763b411b363SPhilipp Reisner mutex_unlock(&mdev->data.mutex); 3764b411b363SPhilipp Reisner return ok; 3765b411b363SPhilipp Reisner } 3766b411b363SPhilipp Reisner 3767b411b363SPhilipp Reisner /* 3768b411b363SPhilipp Reisner * return values: 3769b411b363SPhilipp Reisner * 1 yes, we have a valid connection 3770b411b363SPhilipp Reisner * 0 oops, did not work out, please try again 3771b411b363SPhilipp Reisner * -1 peer talks different language, 3772b411b363SPhilipp Reisner * no point in trying again, please go standalone. 3773b411b363SPhilipp Reisner */ 3774b411b363SPhilipp Reisner static int drbd_do_handshake(struct drbd_conf *mdev) 3775b411b363SPhilipp Reisner { 3776b411b363SPhilipp Reisner /* ASSERT current == mdev->receiver ... */ 3777b411b363SPhilipp Reisner struct p_handshake *p = &mdev->data.rbuf.handshake; 3778b411b363SPhilipp Reisner const int expect = sizeof(struct p_handshake) 3779b411b363SPhilipp Reisner -sizeof(struct p_header); 3780b411b363SPhilipp Reisner int rv; 3781b411b363SPhilipp Reisner 3782b411b363SPhilipp Reisner rv = drbd_send_handshake(mdev); 3783b411b363SPhilipp Reisner if (!rv) 3784b411b363SPhilipp Reisner return 0; 3785b411b363SPhilipp Reisner 3786b411b363SPhilipp Reisner rv = drbd_recv_header(mdev, &p->head); 3787b411b363SPhilipp Reisner if (!rv) 3788b411b363SPhilipp Reisner return 0; 3789b411b363SPhilipp Reisner 3790b411b363SPhilipp Reisner if (p->head.command != P_HAND_SHAKE) { 3791b411b363SPhilipp Reisner dev_err(DEV, "expected HandShake packet, received: %s (0x%04x)\n", 3792b411b363SPhilipp Reisner cmdname(p->head.command), p->head.command); 3793b411b363SPhilipp Reisner return -1; 3794b411b363SPhilipp Reisner } 3795b411b363SPhilipp Reisner 3796b411b363SPhilipp Reisner if (p->head.length != expect) { 3797b411b363SPhilipp Reisner dev_err(DEV, "expected HandShake length: %u, received: %u\n", 3798b411b363SPhilipp Reisner expect, p->head.length); 3799b411b363SPhilipp Reisner return -1; 3800b411b363SPhilipp Reisner } 3801b411b363SPhilipp Reisner 3802b411b363SPhilipp Reisner rv = drbd_recv(mdev, &p->head.payload, expect); 3803b411b363SPhilipp Reisner 3804b411b363SPhilipp Reisner if (rv != expect) { 3805b411b363SPhilipp Reisner dev_err(DEV, "short read receiving handshake packet: l=%u\n", rv); 3806b411b363SPhilipp Reisner return 0; 3807b411b363SPhilipp Reisner } 3808b411b363SPhilipp Reisner 3809b411b363SPhilipp Reisner p->protocol_min = be32_to_cpu(p->protocol_min); 3810b411b363SPhilipp Reisner p->protocol_max = be32_to_cpu(p->protocol_max); 3811b411b363SPhilipp Reisner if (p->protocol_max == 0) 3812b411b363SPhilipp Reisner p->protocol_max = p->protocol_min; 3813b411b363SPhilipp Reisner 3814b411b363SPhilipp Reisner if (PRO_VERSION_MAX < p->protocol_min || 3815b411b363SPhilipp Reisner PRO_VERSION_MIN > p->protocol_max) 3816b411b363SPhilipp Reisner goto incompat; 3817b411b363SPhilipp Reisner 3818b411b363SPhilipp Reisner mdev->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); 3819b411b363SPhilipp Reisner 3820b411b363SPhilipp Reisner dev_info(DEV, "Handshake successful: " 3821b411b363SPhilipp Reisner "Agreed network protocol version %d\n", mdev->agreed_pro_version); 3822b411b363SPhilipp Reisner 3823b411b363SPhilipp Reisner return 1; 3824b411b363SPhilipp Reisner 3825b411b363SPhilipp Reisner incompat: 3826b411b363SPhilipp Reisner dev_err(DEV, "incompatible DRBD dialects: " 3827b411b363SPhilipp Reisner "I support %d-%d, peer supports %d-%d\n", 3828b411b363SPhilipp Reisner PRO_VERSION_MIN, PRO_VERSION_MAX, 3829b411b363SPhilipp Reisner p->protocol_min, p->protocol_max); 3830b411b363SPhilipp Reisner return -1; 3831b411b363SPhilipp Reisner } 3832b411b363SPhilipp Reisner 3833b411b363SPhilipp Reisner #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) 3834b411b363SPhilipp Reisner static int drbd_do_auth(struct drbd_conf *mdev) 3835b411b363SPhilipp Reisner { 3836b411b363SPhilipp Reisner dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); 3837b411b363SPhilipp Reisner dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); 3838b10d96cbSJohannes Thoma return -1; 3839b411b363SPhilipp Reisner } 3840b411b363SPhilipp Reisner #else 3841b411b363SPhilipp Reisner #define CHALLENGE_LEN 64 3842b10d96cbSJohannes Thoma 3843b10d96cbSJohannes Thoma /* Return value: 3844b10d96cbSJohannes Thoma 1 - auth succeeded, 3845b10d96cbSJohannes Thoma 0 - failed, try again (network error), 3846b10d96cbSJohannes Thoma -1 - auth failed, don't try again. 3847b10d96cbSJohannes Thoma */ 3848b10d96cbSJohannes Thoma 3849b411b363SPhilipp Reisner static int drbd_do_auth(struct drbd_conf *mdev) 3850b411b363SPhilipp Reisner { 3851b411b363SPhilipp Reisner char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ 3852b411b363SPhilipp Reisner struct scatterlist sg; 3853b411b363SPhilipp Reisner char *response = NULL; 3854b411b363SPhilipp Reisner char *right_response = NULL; 3855b411b363SPhilipp Reisner char *peers_ch = NULL; 3856b411b363SPhilipp Reisner struct p_header p; 3857b411b363SPhilipp Reisner unsigned int key_len = strlen(mdev->net_conf->shared_secret); 3858b411b363SPhilipp Reisner unsigned int resp_size; 3859b411b363SPhilipp Reisner struct hash_desc desc; 3860b411b363SPhilipp Reisner int rv; 3861b411b363SPhilipp Reisner 3862b411b363SPhilipp Reisner desc.tfm = mdev->cram_hmac_tfm; 3863b411b363SPhilipp Reisner desc.flags = 0; 3864b411b363SPhilipp Reisner 3865b411b363SPhilipp Reisner rv = crypto_hash_setkey(mdev->cram_hmac_tfm, 3866b411b363SPhilipp Reisner (u8 *)mdev->net_conf->shared_secret, key_len); 3867b411b363SPhilipp Reisner if (rv) { 3868b411b363SPhilipp Reisner dev_err(DEV, "crypto_hash_setkey() failed with %d\n", rv); 3869b10d96cbSJohannes Thoma rv = -1; 3870b411b363SPhilipp Reisner goto fail; 3871b411b363SPhilipp Reisner } 3872b411b363SPhilipp Reisner 3873b411b363SPhilipp Reisner get_random_bytes(my_challenge, CHALLENGE_LEN); 3874b411b363SPhilipp Reisner 3875b411b363SPhilipp Reisner rv = drbd_send_cmd2(mdev, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN); 3876b411b363SPhilipp Reisner if (!rv) 3877b411b363SPhilipp Reisner goto fail; 3878b411b363SPhilipp Reisner 3879b411b363SPhilipp Reisner rv = drbd_recv_header(mdev, &p); 3880b411b363SPhilipp Reisner if (!rv) 3881b411b363SPhilipp Reisner goto fail; 3882b411b363SPhilipp Reisner 3883b411b363SPhilipp Reisner if (p.command != P_AUTH_CHALLENGE) { 3884b411b363SPhilipp Reisner dev_err(DEV, "expected AuthChallenge packet, received: %s (0x%04x)\n", 3885b411b363SPhilipp Reisner cmdname(p.command), p.command); 3886b411b363SPhilipp Reisner rv = 0; 3887b411b363SPhilipp Reisner goto fail; 3888b411b363SPhilipp Reisner } 3889b411b363SPhilipp Reisner 3890b411b363SPhilipp Reisner if (p.length > CHALLENGE_LEN*2) { 3891b411b363SPhilipp Reisner dev_err(DEV, "expected AuthChallenge payload too big.\n"); 3892b10d96cbSJohannes Thoma rv = -1; 3893b411b363SPhilipp Reisner goto fail; 3894b411b363SPhilipp Reisner } 3895b411b363SPhilipp Reisner 3896b411b363SPhilipp Reisner peers_ch = kmalloc(p.length, GFP_NOIO); 3897b411b363SPhilipp Reisner if (peers_ch == NULL) { 3898b411b363SPhilipp Reisner dev_err(DEV, "kmalloc of peers_ch failed\n"); 3899b10d96cbSJohannes Thoma rv = -1; 3900b411b363SPhilipp Reisner goto fail; 3901b411b363SPhilipp Reisner } 3902b411b363SPhilipp Reisner 3903b411b363SPhilipp Reisner rv = drbd_recv(mdev, peers_ch, p.length); 3904b411b363SPhilipp Reisner 3905b411b363SPhilipp Reisner if (rv != p.length) { 3906b411b363SPhilipp Reisner dev_err(DEV, "short read AuthChallenge: l=%u\n", rv); 3907b411b363SPhilipp Reisner rv = 0; 3908b411b363SPhilipp Reisner goto fail; 3909b411b363SPhilipp Reisner } 3910b411b363SPhilipp Reisner 3911b411b363SPhilipp Reisner resp_size = crypto_hash_digestsize(mdev->cram_hmac_tfm); 3912b411b363SPhilipp Reisner response = kmalloc(resp_size, GFP_NOIO); 3913b411b363SPhilipp Reisner if (response == NULL) { 3914b411b363SPhilipp Reisner dev_err(DEV, "kmalloc of response failed\n"); 3915b10d96cbSJohannes Thoma rv = -1; 3916b411b363SPhilipp Reisner goto fail; 3917b411b363SPhilipp Reisner } 3918b411b363SPhilipp Reisner 3919b411b363SPhilipp Reisner sg_init_table(&sg, 1); 3920b411b363SPhilipp Reisner sg_set_buf(&sg, peers_ch, p.length); 3921b411b363SPhilipp Reisner 3922b411b363SPhilipp Reisner rv = crypto_hash_digest(&desc, &sg, sg.length, response); 3923b411b363SPhilipp Reisner if (rv) { 3924b411b363SPhilipp Reisner dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); 3925b10d96cbSJohannes Thoma rv = -1; 3926b411b363SPhilipp Reisner goto fail; 3927b411b363SPhilipp Reisner } 3928b411b363SPhilipp Reisner 3929b411b363SPhilipp Reisner rv = drbd_send_cmd2(mdev, P_AUTH_RESPONSE, response, resp_size); 3930b411b363SPhilipp Reisner if (!rv) 3931b411b363SPhilipp Reisner goto fail; 3932b411b363SPhilipp Reisner 3933b411b363SPhilipp Reisner rv = drbd_recv_header(mdev, &p); 3934b411b363SPhilipp Reisner if (!rv) 3935b411b363SPhilipp Reisner goto fail; 3936b411b363SPhilipp Reisner 3937b411b363SPhilipp Reisner if (p.command != P_AUTH_RESPONSE) { 3938b411b363SPhilipp Reisner dev_err(DEV, "expected AuthResponse packet, received: %s (0x%04x)\n", 3939b411b363SPhilipp Reisner cmdname(p.command), p.command); 3940b411b363SPhilipp Reisner rv = 0; 3941b411b363SPhilipp Reisner goto fail; 3942b411b363SPhilipp Reisner } 3943b411b363SPhilipp Reisner 3944b411b363SPhilipp Reisner if (p.length != resp_size) { 3945b411b363SPhilipp Reisner dev_err(DEV, "expected AuthResponse payload of wrong size\n"); 3946b411b363SPhilipp Reisner rv = 0; 3947b411b363SPhilipp Reisner goto fail; 3948b411b363SPhilipp Reisner } 3949b411b363SPhilipp Reisner 3950b411b363SPhilipp Reisner rv = drbd_recv(mdev, response , resp_size); 3951b411b363SPhilipp Reisner 3952b411b363SPhilipp Reisner if (rv != resp_size) { 3953b411b363SPhilipp Reisner dev_err(DEV, "short read receiving AuthResponse: l=%u\n", rv); 3954b411b363SPhilipp Reisner rv = 0; 3955b411b363SPhilipp Reisner goto fail; 3956b411b363SPhilipp Reisner } 3957b411b363SPhilipp Reisner 3958b411b363SPhilipp Reisner right_response = kmalloc(resp_size, GFP_NOIO); 39592d1ee87dSJulia Lawall if (right_response == NULL) { 3960b411b363SPhilipp Reisner dev_err(DEV, "kmalloc of right_response failed\n"); 3961b10d96cbSJohannes Thoma rv = -1; 3962b411b363SPhilipp Reisner goto fail; 3963b411b363SPhilipp Reisner } 3964b411b363SPhilipp Reisner 3965b411b363SPhilipp Reisner sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); 3966b411b363SPhilipp Reisner 3967b411b363SPhilipp Reisner rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); 3968b411b363SPhilipp Reisner if (rv) { 3969b411b363SPhilipp Reisner dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); 3970b10d96cbSJohannes Thoma rv = -1; 3971b411b363SPhilipp Reisner goto fail; 3972b411b363SPhilipp Reisner } 3973b411b363SPhilipp Reisner 3974b411b363SPhilipp Reisner rv = !memcmp(response, right_response, resp_size); 3975b411b363SPhilipp Reisner 3976b411b363SPhilipp Reisner if (rv) 3977b411b363SPhilipp Reisner dev_info(DEV, "Peer authenticated using %d bytes of '%s' HMAC\n", 3978b411b363SPhilipp Reisner resp_size, mdev->net_conf->cram_hmac_alg); 3979b10d96cbSJohannes Thoma else 3980b10d96cbSJohannes Thoma rv = -1; 3981b411b363SPhilipp Reisner 3982b411b363SPhilipp Reisner fail: 3983b411b363SPhilipp Reisner kfree(peers_ch); 3984b411b363SPhilipp Reisner kfree(response); 3985b411b363SPhilipp Reisner kfree(right_response); 3986b411b363SPhilipp Reisner 3987b411b363SPhilipp Reisner return rv; 3988b411b363SPhilipp Reisner } 3989b411b363SPhilipp Reisner #endif 3990b411b363SPhilipp Reisner 3991b411b363SPhilipp Reisner int drbdd_init(struct drbd_thread *thi) 3992b411b363SPhilipp Reisner { 3993b411b363SPhilipp Reisner struct drbd_conf *mdev = thi->mdev; 3994b411b363SPhilipp Reisner unsigned int minor = mdev_to_minor(mdev); 3995b411b363SPhilipp Reisner int h; 3996b411b363SPhilipp Reisner 3997b411b363SPhilipp Reisner sprintf(current->comm, "drbd%d_receiver", minor); 3998b411b363SPhilipp Reisner 3999b411b363SPhilipp Reisner dev_info(DEV, "receiver (re)started\n"); 4000b411b363SPhilipp Reisner 4001b411b363SPhilipp Reisner do { 4002b411b363SPhilipp Reisner h = drbd_connect(mdev); 4003b411b363SPhilipp Reisner if (h == 0) { 4004b411b363SPhilipp Reisner drbd_disconnect(mdev); 4005b411b363SPhilipp Reisner __set_current_state(TASK_INTERRUPTIBLE); 4006b411b363SPhilipp Reisner schedule_timeout(HZ); 4007b411b363SPhilipp Reisner } 4008b411b363SPhilipp Reisner if (h == -1) { 4009b411b363SPhilipp Reisner dev_warn(DEV, "Discarding network configuration.\n"); 4010b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 4011b411b363SPhilipp Reisner } 4012b411b363SPhilipp Reisner } while (h == 0); 4013b411b363SPhilipp Reisner 4014b411b363SPhilipp Reisner if (h > 0) { 4015b411b363SPhilipp Reisner if (get_net_conf(mdev)) { 4016b411b363SPhilipp Reisner drbdd(mdev); 4017b411b363SPhilipp Reisner put_net_conf(mdev); 4018b411b363SPhilipp Reisner } 4019b411b363SPhilipp Reisner } 4020b411b363SPhilipp Reisner 4021b411b363SPhilipp Reisner drbd_disconnect(mdev); 4022b411b363SPhilipp Reisner 4023b411b363SPhilipp Reisner dev_info(DEV, "receiver terminated\n"); 4024b411b363SPhilipp Reisner return 0; 4025b411b363SPhilipp Reisner } 4026b411b363SPhilipp Reisner 4027b411b363SPhilipp Reisner /* ********* acknowledge sender ******** */ 4028b411b363SPhilipp Reisner 4029b411b363SPhilipp Reisner static int got_RqSReply(struct drbd_conf *mdev, struct p_header *h) 4030b411b363SPhilipp Reisner { 4031b411b363SPhilipp Reisner struct p_req_state_reply *p = (struct p_req_state_reply *)h; 4032b411b363SPhilipp Reisner 4033b411b363SPhilipp Reisner int retcode = be32_to_cpu(p->retcode); 4034b411b363SPhilipp Reisner 4035b411b363SPhilipp Reisner if (retcode >= SS_SUCCESS) { 4036b411b363SPhilipp Reisner set_bit(CL_ST_CHG_SUCCESS, &mdev->flags); 4037b411b363SPhilipp Reisner } else { 4038b411b363SPhilipp Reisner set_bit(CL_ST_CHG_FAIL, &mdev->flags); 4039b411b363SPhilipp Reisner dev_err(DEV, "Requested state change failed by peer: %s (%d)\n", 4040b411b363SPhilipp Reisner drbd_set_st_err_str(retcode), retcode); 4041b411b363SPhilipp Reisner } 4042b411b363SPhilipp Reisner wake_up(&mdev->state_wait); 4043b411b363SPhilipp Reisner 4044b411b363SPhilipp Reisner return TRUE; 4045b411b363SPhilipp Reisner } 4046b411b363SPhilipp Reisner 4047b411b363SPhilipp Reisner static int got_Ping(struct drbd_conf *mdev, struct p_header *h) 4048b411b363SPhilipp Reisner { 4049b411b363SPhilipp Reisner return drbd_send_ping_ack(mdev); 4050b411b363SPhilipp Reisner 4051b411b363SPhilipp Reisner } 4052b411b363SPhilipp Reisner 4053b411b363SPhilipp Reisner static int got_PingAck(struct drbd_conf *mdev, struct p_header *h) 4054b411b363SPhilipp Reisner { 4055b411b363SPhilipp Reisner /* restore idle timeout */ 4056b411b363SPhilipp Reisner mdev->meta.socket->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ; 4057b411b363SPhilipp Reisner 4058b411b363SPhilipp Reisner return TRUE; 4059b411b363SPhilipp Reisner } 4060b411b363SPhilipp Reisner 4061b411b363SPhilipp Reisner static int got_IsInSync(struct drbd_conf *mdev, struct p_header *h) 4062b411b363SPhilipp Reisner { 4063b411b363SPhilipp Reisner struct p_block_ack *p = (struct p_block_ack *)h; 4064b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 4065b411b363SPhilipp Reisner int blksize = be32_to_cpu(p->blksize); 4066b411b363SPhilipp Reisner 4067b411b363SPhilipp Reisner D_ASSERT(mdev->agreed_pro_version >= 89); 4068b411b363SPhilipp Reisner 4069b411b363SPhilipp Reisner update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4070b411b363SPhilipp Reisner 4071b411b363SPhilipp Reisner drbd_rs_complete_io(mdev, sector); 4072b411b363SPhilipp Reisner drbd_set_in_sync(mdev, sector, blksize); 4073b411b363SPhilipp Reisner /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ 4074b411b363SPhilipp Reisner mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); 4075b411b363SPhilipp Reisner dec_rs_pending(mdev); 4076b411b363SPhilipp Reisner 4077b411b363SPhilipp Reisner return TRUE; 4078b411b363SPhilipp Reisner } 4079b411b363SPhilipp Reisner 4080b411b363SPhilipp Reisner /* when we receive the ACK for a write request, 4081b411b363SPhilipp Reisner * verify that we actually know about it */ 4082b411b363SPhilipp Reisner static struct drbd_request *_ack_id_to_req(struct drbd_conf *mdev, 4083b411b363SPhilipp Reisner u64 id, sector_t sector) 4084b411b363SPhilipp Reisner { 4085b411b363SPhilipp Reisner struct hlist_head *slot = tl_hash_slot(mdev, sector); 4086b411b363SPhilipp Reisner struct hlist_node *n; 4087b411b363SPhilipp Reisner struct drbd_request *req; 4088b411b363SPhilipp Reisner 4089b411b363SPhilipp Reisner hlist_for_each_entry(req, n, slot, colision) { 4090b411b363SPhilipp Reisner if ((unsigned long)req == (unsigned long)id) { 4091b411b363SPhilipp Reisner if (req->sector != sector) { 4092b411b363SPhilipp Reisner dev_err(DEV, "_ack_id_to_req: found req %p but it has " 4093b411b363SPhilipp Reisner "wrong sector (%llus versus %llus)\n", req, 4094b411b363SPhilipp Reisner (unsigned long long)req->sector, 4095b411b363SPhilipp Reisner (unsigned long long)sector); 4096b411b363SPhilipp Reisner break; 4097b411b363SPhilipp Reisner } 4098b411b363SPhilipp Reisner return req; 4099b411b363SPhilipp Reisner } 4100b411b363SPhilipp Reisner } 4101b411b363SPhilipp Reisner dev_err(DEV, "_ack_id_to_req: failed to find req %p, sector %llus in list\n", 4102b411b363SPhilipp Reisner (void *)(unsigned long)id, (unsigned long long)sector); 4103b411b363SPhilipp Reisner return NULL; 4104b411b363SPhilipp Reisner } 4105b411b363SPhilipp Reisner 4106b411b363SPhilipp Reisner typedef struct drbd_request *(req_validator_fn) 4107b411b363SPhilipp Reisner (struct drbd_conf *mdev, u64 id, sector_t sector); 4108b411b363SPhilipp Reisner 4109b411b363SPhilipp Reisner static int validate_req_change_req_state(struct drbd_conf *mdev, 4110b411b363SPhilipp Reisner u64 id, sector_t sector, req_validator_fn validator, 4111b411b363SPhilipp Reisner const char *func, enum drbd_req_event what) 4112b411b363SPhilipp Reisner { 4113b411b363SPhilipp Reisner struct drbd_request *req; 4114b411b363SPhilipp Reisner struct bio_and_error m; 4115b411b363SPhilipp Reisner 4116b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 4117b411b363SPhilipp Reisner req = validator(mdev, id, sector); 4118b411b363SPhilipp Reisner if (unlikely(!req)) { 4119b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 4120b411b363SPhilipp Reisner dev_err(DEV, "%s: got a corrupt block_id/sector pair\n", func); 4121b411b363SPhilipp Reisner return FALSE; 4122b411b363SPhilipp Reisner } 4123b411b363SPhilipp Reisner __req_mod(req, what, &m); 4124b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 4125b411b363SPhilipp Reisner 4126b411b363SPhilipp Reisner if (m.bio) 4127b411b363SPhilipp Reisner complete_master_bio(mdev, &m); 4128b411b363SPhilipp Reisner return TRUE; 4129b411b363SPhilipp Reisner } 4130b411b363SPhilipp Reisner 4131b411b363SPhilipp Reisner static int got_BlockAck(struct drbd_conf *mdev, struct p_header *h) 4132b411b363SPhilipp Reisner { 4133b411b363SPhilipp Reisner struct p_block_ack *p = (struct p_block_ack *)h; 4134b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 4135b411b363SPhilipp Reisner int blksize = be32_to_cpu(p->blksize); 4136b411b363SPhilipp Reisner enum drbd_req_event what; 4137b411b363SPhilipp Reisner 4138b411b363SPhilipp Reisner update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4139b411b363SPhilipp Reisner 4140b411b363SPhilipp Reisner if (is_syncer_block_id(p->block_id)) { 4141b411b363SPhilipp Reisner drbd_set_in_sync(mdev, sector, blksize); 4142b411b363SPhilipp Reisner dec_rs_pending(mdev); 4143b411b363SPhilipp Reisner return TRUE; 4144b411b363SPhilipp Reisner } 4145b411b363SPhilipp Reisner switch (be16_to_cpu(h->command)) { 4146b411b363SPhilipp Reisner case P_RS_WRITE_ACK: 4147b411b363SPhilipp Reisner D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 4148b411b363SPhilipp Reisner what = write_acked_by_peer_and_sis; 4149b411b363SPhilipp Reisner break; 4150b411b363SPhilipp Reisner case P_WRITE_ACK: 4151b411b363SPhilipp Reisner D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 4152b411b363SPhilipp Reisner what = write_acked_by_peer; 4153b411b363SPhilipp Reisner break; 4154b411b363SPhilipp Reisner case P_RECV_ACK: 4155b411b363SPhilipp Reisner D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_B); 4156b411b363SPhilipp Reisner what = recv_acked_by_peer; 4157b411b363SPhilipp Reisner break; 4158b411b363SPhilipp Reisner case P_DISCARD_ACK: 4159b411b363SPhilipp Reisner D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 4160b411b363SPhilipp Reisner what = conflict_discarded_by_peer; 4161b411b363SPhilipp Reisner break; 4162b411b363SPhilipp Reisner default: 4163b411b363SPhilipp Reisner D_ASSERT(0); 4164b411b363SPhilipp Reisner return FALSE; 4165b411b363SPhilipp Reisner } 4166b411b363SPhilipp Reisner 4167b411b363SPhilipp Reisner return validate_req_change_req_state(mdev, p->block_id, sector, 4168b411b363SPhilipp Reisner _ack_id_to_req, __func__ , what); 4169b411b363SPhilipp Reisner } 4170b411b363SPhilipp Reisner 4171b411b363SPhilipp Reisner static int got_NegAck(struct drbd_conf *mdev, struct p_header *h) 4172b411b363SPhilipp Reisner { 4173b411b363SPhilipp Reisner struct p_block_ack *p = (struct p_block_ack *)h; 4174b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 4175b411b363SPhilipp Reisner 4176b411b363SPhilipp Reisner if (__ratelimit(&drbd_ratelimit_state)) 4177b411b363SPhilipp Reisner dev_warn(DEV, "Got NegAck packet. Peer is in troubles?\n"); 4178b411b363SPhilipp Reisner 4179b411b363SPhilipp Reisner update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4180b411b363SPhilipp Reisner 4181b411b363SPhilipp Reisner if (is_syncer_block_id(p->block_id)) { 4182b411b363SPhilipp Reisner int size = be32_to_cpu(p->blksize); 4183b411b363SPhilipp Reisner dec_rs_pending(mdev); 4184b411b363SPhilipp Reisner drbd_rs_failed_io(mdev, sector, size); 4185b411b363SPhilipp Reisner return TRUE; 4186b411b363SPhilipp Reisner } 4187b411b363SPhilipp Reisner return validate_req_change_req_state(mdev, p->block_id, sector, 4188b411b363SPhilipp Reisner _ack_id_to_req, __func__ , neg_acked); 4189b411b363SPhilipp Reisner } 4190b411b363SPhilipp Reisner 4191b411b363SPhilipp Reisner static int got_NegDReply(struct drbd_conf *mdev, struct p_header *h) 4192b411b363SPhilipp Reisner { 4193b411b363SPhilipp Reisner struct p_block_ack *p = (struct p_block_ack *)h; 4194b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 4195b411b363SPhilipp Reisner 4196b411b363SPhilipp Reisner update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4197b411b363SPhilipp Reisner dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n", 4198b411b363SPhilipp Reisner (unsigned long long)sector, be32_to_cpu(p->blksize)); 4199b411b363SPhilipp Reisner 4200b411b363SPhilipp Reisner return validate_req_change_req_state(mdev, p->block_id, sector, 4201b411b363SPhilipp Reisner _ar_id_to_req, __func__ , neg_acked); 4202b411b363SPhilipp Reisner } 4203b411b363SPhilipp Reisner 4204b411b363SPhilipp Reisner static int got_NegRSDReply(struct drbd_conf *mdev, struct p_header *h) 4205b411b363SPhilipp Reisner { 4206b411b363SPhilipp Reisner sector_t sector; 4207b411b363SPhilipp Reisner int size; 4208b411b363SPhilipp Reisner struct p_block_ack *p = (struct p_block_ack *)h; 4209b411b363SPhilipp Reisner 4210b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 4211b411b363SPhilipp Reisner size = be32_to_cpu(p->blksize); 4212b411b363SPhilipp Reisner D_ASSERT(p->block_id == ID_SYNCER); 4213b411b363SPhilipp Reisner 4214b411b363SPhilipp Reisner update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4215b411b363SPhilipp Reisner 4216b411b363SPhilipp Reisner dec_rs_pending(mdev); 4217b411b363SPhilipp Reisner 4218b411b363SPhilipp Reisner if (get_ldev_if_state(mdev, D_FAILED)) { 4219b411b363SPhilipp Reisner drbd_rs_complete_io(mdev, sector); 4220b411b363SPhilipp Reisner drbd_rs_failed_io(mdev, sector, size); 4221b411b363SPhilipp Reisner put_ldev(mdev); 4222b411b363SPhilipp Reisner } 4223b411b363SPhilipp Reisner 4224b411b363SPhilipp Reisner return TRUE; 4225b411b363SPhilipp Reisner } 4226b411b363SPhilipp Reisner 4227b411b363SPhilipp Reisner static int got_BarrierAck(struct drbd_conf *mdev, struct p_header *h) 4228b411b363SPhilipp Reisner { 4229b411b363SPhilipp Reisner struct p_barrier_ack *p = (struct p_barrier_ack *)h; 4230b411b363SPhilipp Reisner 4231b411b363SPhilipp Reisner tl_release(mdev, p->barrier, be32_to_cpu(p->set_size)); 4232b411b363SPhilipp Reisner 4233b411b363SPhilipp Reisner return TRUE; 4234b411b363SPhilipp Reisner } 4235b411b363SPhilipp Reisner 4236b411b363SPhilipp Reisner static int got_OVResult(struct drbd_conf *mdev, struct p_header *h) 4237b411b363SPhilipp Reisner { 4238b411b363SPhilipp Reisner struct p_block_ack *p = (struct p_block_ack *)h; 4239b411b363SPhilipp Reisner struct drbd_work *w; 4240b411b363SPhilipp Reisner sector_t sector; 4241b411b363SPhilipp Reisner int size; 4242b411b363SPhilipp Reisner 4243b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 4244b411b363SPhilipp Reisner size = be32_to_cpu(p->blksize); 4245b411b363SPhilipp Reisner 4246b411b363SPhilipp Reisner update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4247b411b363SPhilipp Reisner 4248b411b363SPhilipp Reisner if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) 4249b411b363SPhilipp Reisner drbd_ov_oos_found(mdev, sector, size); 4250b411b363SPhilipp Reisner else 4251b411b363SPhilipp Reisner ov_oos_print(mdev); 4252b411b363SPhilipp Reisner 4253b411b363SPhilipp Reisner drbd_rs_complete_io(mdev, sector); 4254b411b363SPhilipp Reisner dec_rs_pending(mdev); 4255b411b363SPhilipp Reisner 4256b411b363SPhilipp Reisner if (--mdev->ov_left == 0) { 4257b411b363SPhilipp Reisner w = kmalloc(sizeof(*w), GFP_NOIO); 4258b411b363SPhilipp Reisner if (w) { 4259b411b363SPhilipp Reisner w->cb = w_ov_finished; 4260b411b363SPhilipp Reisner drbd_queue_work_front(&mdev->data.work, w); 4261b411b363SPhilipp Reisner } else { 4262b411b363SPhilipp Reisner dev_err(DEV, "kmalloc(w) failed."); 4263b411b363SPhilipp Reisner ov_oos_print(mdev); 4264b411b363SPhilipp Reisner drbd_resync_finished(mdev); 4265b411b363SPhilipp Reisner } 4266b411b363SPhilipp Reisner } 4267b411b363SPhilipp Reisner return TRUE; 4268b411b363SPhilipp Reisner } 4269b411b363SPhilipp Reisner 4270b411b363SPhilipp Reisner struct asender_cmd { 4271b411b363SPhilipp Reisner size_t pkt_size; 4272b411b363SPhilipp Reisner int (*process)(struct drbd_conf *mdev, struct p_header *h); 4273b411b363SPhilipp Reisner }; 4274b411b363SPhilipp Reisner 4275b411b363SPhilipp Reisner static struct asender_cmd *get_asender_cmd(int cmd) 4276b411b363SPhilipp Reisner { 4277b411b363SPhilipp Reisner static struct asender_cmd asender_tbl[] = { 4278b411b363SPhilipp Reisner /* anything missing from this table is in 4279b411b363SPhilipp Reisner * the drbd_cmd_handler (drbd_default_handler) table, 4280b411b363SPhilipp Reisner * see the beginning of drbdd() */ 4281b411b363SPhilipp Reisner [P_PING] = { sizeof(struct p_header), got_Ping }, 4282b411b363SPhilipp Reisner [P_PING_ACK] = { sizeof(struct p_header), got_PingAck }, 4283b411b363SPhilipp Reisner [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 4284b411b363SPhilipp Reisner [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 4285b411b363SPhilipp Reisner [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 4286b411b363SPhilipp Reisner [P_DISCARD_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 4287b411b363SPhilipp Reisner [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, 4288b411b363SPhilipp Reisner [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, 4289b411b363SPhilipp Reisner [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply}, 4290b411b363SPhilipp Reisner [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, 4291b411b363SPhilipp Reisner [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, 4292b411b363SPhilipp Reisner [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, 4293b411b363SPhilipp Reisner [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, 4294b411b363SPhilipp Reisner [P_MAX_CMD] = { 0, NULL }, 4295b411b363SPhilipp Reisner }; 4296b411b363SPhilipp Reisner if (cmd > P_MAX_CMD || asender_tbl[cmd].process == NULL) 4297b411b363SPhilipp Reisner return NULL; 4298b411b363SPhilipp Reisner return &asender_tbl[cmd]; 4299b411b363SPhilipp Reisner } 4300b411b363SPhilipp Reisner 4301b411b363SPhilipp Reisner int drbd_asender(struct drbd_thread *thi) 4302b411b363SPhilipp Reisner { 4303b411b363SPhilipp Reisner struct drbd_conf *mdev = thi->mdev; 4304b411b363SPhilipp Reisner struct p_header *h = &mdev->meta.rbuf.header; 4305b411b363SPhilipp Reisner struct asender_cmd *cmd = NULL; 4306b411b363SPhilipp Reisner 4307b411b363SPhilipp Reisner int rv, len; 4308b411b363SPhilipp Reisner void *buf = h; 4309b411b363SPhilipp Reisner int received = 0; 4310b411b363SPhilipp Reisner int expect = sizeof(struct p_header); 4311b411b363SPhilipp Reisner int empty; 4312b411b363SPhilipp Reisner 4313b411b363SPhilipp Reisner sprintf(current->comm, "drbd%d_asender", mdev_to_minor(mdev)); 4314b411b363SPhilipp Reisner 4315b411b363SPhilipp Reisner current->policy = SCHED_RR; /* Make this a realtime task! */ 4316b411b363SPhilipp Reisner current->rt_priority = 2; /* more important than all other tasks */ 4317b411b363SPhilipp Reisner 4318b411b363SPhilipp Reisner while (get_t_state(thi) == Running) { 4319b411b363SPhilipp Reisner drbd_thread_current_set_cpu(mdev); 4320b411b363SPhilipp Reisner if (test_and_clear_bit(SEND_PING, &mdev->flags)) { 4321b411b363SPhilipp Reisner ERR_IF(!drbd_send_ping(mdev)) goto reconnect; 4322b411b363SPhilipp Reisner mdev->meta.socket->sk->sk_rcvtimeo = 4323b411b363SPhilipp Reisner mdev->net_conf->ping_timeo*HZ/10; 4324b411b363SPhilipp Reisner } 4325b411b363SPhilipp Reisner 4326b411b363SPhilipp Reisner /* conditionally cork; 4327b411b363SPhilipp Reisner * it may hurt latency if we cork without much to send */ 4328b411b363SPhilipp Reisner if (!mdev->net_conf->no_cork && 4329b411b363SPhilipp Reisner 3 < atomic_read(&mdev->unacked_cnt)) 4330b411b363SPhilipp Reisner drbd_tcp_cork(mdev->meta.socket); 4331b411b363SPhilipp Reisner while (1) { 4332b411b363SPhilipp Reisner clear_bit(SIGNAL_ASENDER, &mdev->flags); 4333b411b363SPhilipp Reisner flush_signals(current); 4334b411b363SPhilipp Reisner if (!drbd_process_done_ee(mdev)) { 4335b411b363SPhilipp Reisner dev_err(DEV, "process_done_ee() = NOT_OK\n"); 4336b411b363SPhilipp Reisner goto reconnect; 4337b411b363SPhilipp Reisner } 4338b411b363SPhilipp Reisner /* to avoid race with newly queued ACKs */ 4339b411b363SPhilipp Reisner set_bit(SIGNAL_ASENDER, &mdev->flags); 4340b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 4341b411b363SPhilipp Reisner empty = list_empty(&mdev->done_ee); 4342b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 4343b411b363SPhilipp Reisner /* new ack may have been queued right here, 4344b411b363SPhilipp Reisner * but then there is also a signal pending, 4345b411b363SPhilipp Reisner * and we start over... */ 4346b411b363SPhilipp Reisner if (empty) 4347b411b363SPhilipp Reisner break; 4348b411b363SPhilipp Reisner } 4349b411b363SPhilipp Reisner /* but unconditionally uncork unless disabled */ 4350b411b363SPhilipp Reisner if (!mdev->net_conf->no_cork) 4351b411b363SPhilipp Reisner drbd_tcp_uncork(mdev->meta.socket); 4352b411b363SPhilipp Reisner 4353b411b363SPhilipp Reisner /* short circuit, recv_msg would return EINTR anyways. */ 4354b411b363SPhilipp Reisner if (signal_pending(current)) 4355b411b363SPhilipp Reisner continue; 4356b411b363SPhilipp Reisner 4357b411b363SPhilipp Reisner rv = drbd_recv_short(mdev, mdev->meta.socket, 4358b411b363SPhilipp Reisner buf, expect-received, 0); 4359b411b363SPhilipp Reisner clear_bit(SIGNAL_ASENDER, &mdev->flags); 4360b411b363SPhilipp Reisner 4361b411b363SPhilipp Reisner flush_signals(current); 4362b411b363SPhilipp Reisner 4363b411b363SPhilipp Reisner /* Note: 4364b411b363SPhilipp Reisner * -EINTR (on meta) we got a signal 4365b411b363SPhilipp Reisner * -EAGAIN (on meta) rcvtimeo expired 4366b411b363SPhilipp Reisner * -ECONNRESET other side closed the connection 4367b411b363SPhilipp Reisner * -ERESTARTSYS (on data) we got a signal 4368b411b363SPhilipp Reisner * rv < 0 other than above: unexpected error! 4369b411b363SPhilipp Reisner * rv == expected: full header or command 4370b411b363SPhilipp Reisner * rv < expected: "woken" by signal during receive 4371b411b363SPhilipp Reisner * rv == 0 : "connection shut down by peer" 4372b411b363SPhilipp Reisner */ 4373b411b363SPhilipp Reisner if (likely(rv > 0)) { 4374b411b363SPhilipp Reisner received += rv; 4375b411b363SPhilipp Reisner buf += rv; 4376b411b363SPhilipp Reisner } else if (rv == 0) { 4377b411b363SPhilipp Reisner dev_err(DEV, "meta connection shut down by peer.\n"); 4378b411b363SPhilipp Reisner goto reconnect; 4379b411b363SPhilipp Reisner } else if (rv == -EAGAIN) { 4380b411b363SPhilipp Reisner if (mdev->meta.socket->sk->sk_rcvtimeo == 4381b411b363SPhilipp Reisner mdev->net_conf->ping_timeo*HZ/10) { 4382b411b363SPhilipp Reisner dev_err(DEV, "PingAck did not arrive in time.\n"); 4383b411b363SPhilipp Reisner goto reconnect; 4384b411b363SPhilipp Reisner } 4385b411b363SPhilipp Reisner set_bit(SEND_PING, &mdev->flags); 4386b411b363SPhilipp Reisner continue; 4387b411b363SPhilipp Reisner } else if (rv == -EINTR) { 4388b411b363SPhilipp Reisner continue; 4389b411b363SPhilipp Reisner } else { 4390b411b363SPhilipp Reisner dev_err(DEV, "sock_recvmsg returned %d\n", rv); 4391b411b363SPhilipp Reisner goto reconnect; 4392b411b363SPhilipp Reisner } 4393b411b363SPhilipp Reisner 4394b411b363SPhilipp Reisner if (received == expect && cmd == NULL) { 4395b411b363SPhilipp Reisner if (unlikely(h->magic != BE_DRBD_MAGIC)) { 4396b411b363SPhilipp Reisner dev_err(DEV, "magic?? on meta m: 0x%lx c: %d l: %d\n", 4397b411b363SPhilipp Reisner (long)be32_to_cpu(h->magic), 4398b411b363SPhilipp Reisner h->command, h->length); 4399b411b363SPhilipp Reisner goto reconnect; 4400b411b363SPhilipp Reisner } 4401b411b363SPhilipp Reisner cmd = get_asender_cmd(be16_to_cpu(h->command)); 4402b411b363SPhilipp Reisner len = be16_to_cpu(h->length); 4403b411b363SPhilipp Reisner if (unlikely(cmd == NULL)) { 4404b411b363SPhilipp Reisner dev_err(DEV, "unknown command?? on meta m: 0x%lx c: %d l: %d\n", 4405b411b363SPhilipp Reisner (long)be32_to_cpu(h->magic), 4406b411b363SPhilipp Reisner h->command, h->length); 4407b411b363SPhilipp Reisner goto disconnect; 4408b411b363SPhilipp Reisner } 4409b411b363SPhilipp Reisner expect = cmd->pkt_size; 44106a0afdf5SJens Axboe ERR_IF(len != expect-sizeof(struct p_header)) 4411b411b363SPhilipp Reisner goto reconnect; 4412b411b363SPhilipp Reisner } 4413b411b363SPhilipp Reisner if (received == expect) { 4414b411b363SPhilipp Reisner D_ASSERT(cmd != NULL); 4415b411b363SPhilipp Reisner if (!cmd->process(mdev, h)) 4416b411b363SPhilipp Reisner goto reconnect; 4417b411b363SPhilipp Reisner 4418b411b363SPhilipp Reisner buf = h; 4419b411b363SPhilipp Reisner received = 0; 4420b411b363SPhilipp Reisner expect = sizeof(struct p_header); 4421b411b363SPhilipp Reisner cmd = NULL; 4422b411b363SPhilipp Reisner } 4423b411b363SPhilipp Reisner } 4424b411b363SPhilipp Reisner 4425b411b363SPhilipp Reisner if (0) { 4426b411b363SPhilipp Reisner reconnect: 4427b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE)); 4428b411b363SPhilipp Reisner } 4429b411b363SPhilipp Reisner if (0) { 4430b411b363SPhilipp Reisner disconnect: 4431b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 4432b411b363SPhilipp Reisner } 4433b411b363SPhilipp Reisner clear_bit(SIGNAL_ASENDER, &mdev->flags); 4434b411b363SPhilipp Reisner 4435b411b363SPhilipp Reisner D_ASSERT(mdev->state.conn < C_CONNECTED); 4436b411b363SPhilipp Reisner dev_info(DEV, "asender terminated\n"); 4437b411b363SPhilipp Reisner 4438b411b363SPhilipp Reisner return 0; 4439b411b363SPhilipp Reisner } 4440