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/pkt_sched.h> 40b411b363SPhilipp Reisner #define __KERNEL_SYSCALLS__ 41b411b363SPhilipp Reisner #include <linux/unistd.h> 42b411b363SPhilipp Reisner #include <linux/vmalloc.h> 43b411b363SPhilipp Reisner #include <linux/random.h> 44b411b363SPhilipp Reisner #include <linux/string.h> 45b411b363SPhilipp Reisner #include <linux/scatterlist.h> 46b411b363SPhilipp Reisner #include "drbd_int.h" 47b411b363SPhilipp Reisner #include "drbd_req.h" 48b411b363SPhilipp Reisner 49b411b363SPhilipp Reisner #include "drbd_vli.h" 50b411b363SPhilipp Reisner 51b411b363SPhilipp Reisner enum finish_epoch { 52b411b363SPhilipp Reisner FE_STILL_LIVE, 53b411b363SPhilipp Reisner FE_DESTROYED, 54b411b363SPhilipp Reisner FE_RECYCLED, 55b411b363SPhilipp Reisner }; 56b411b363SPhilipp Reisner 57b411b363SPhilipp Reisner static int drbd_do_handshake(struct drbd_conf *mdev); 58b411b363SPhilipp Reisner static int drbd_do_auth(struct drbd_conf *mdev); 59b411b363SPhilipp Reisner 60b411b363SPhilipp Reisner static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *, struct drbd_epoch *, enum epoch_event); 61b411b363SPhilipp Reisner static int e_end_block(struct drbd_conf *, struct drbd_work *, int); 62b411b363SPhilipp Reisner 63b411b363SPhilipp Reisner 64b411b363SPhilipp Reisner #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) 65b411b363SPhilipp Reisner 6645bb912bSLars Ellenberg /* 6745bb912bSLars Ellenberg * some helper functions to deal with single linked page lists, 6845bb912bSLars Ellenberg * page->private being our "next" pointer. 6945bb912bSLars Ellenberg */ 7045bb912bSLars Ellenberg 7145bb912bSLars Ellenberg /* If at least n pages are linked at head, get n pages off. 7245bb912bSLars Ellenberg * Otherwise, don't modify head, and return NULL. 7345bb912bSLars Ellenberg * Locking is the responsibility of the caller. 7445bb912bSLars Ellenberg */ 7545bb912bSLars Ellenberg static struct page *page_chain_del(struct page **head, int n) 7645bb912bSLars Ellenberg { 7745bb912bSLars Ellenberg struct page *page; 7845bb912bSLars Ellenberg struct page *tmp; 7945bb912bSLars Ellenberg 8045bb912bSLars Ellenberg BUG_ON(!n); 8145bb912bSLars Ellenberg BUG_ON(!head); 8245bb912bSLars Ellenberg 8345bb912bSLars Ellenberg page = *head; 8423ce4227SPhilipp Reisner 8523ce4227SPhilipp Reisner if (!page) 8623ce4227SPhilipp Reisner return NULL; 8723ce4227SPhilipp Reisner 8845bb912bSLars Ellenberg while (page) { 8945bb912bSLars Ellenberg tmp = page_chain_next(page); 9045bb912bSLars Ellenberg if (--n == 0) 9145bb912bSLars Ellenberg break; /* found sufficient pages */ 9245bb912bSLars Ellenberg if (tmp == NULL) 9345bb912bSLars Ellenberg /* insufficient pages, don't use any of them. */ 9445bb912bSLars Ellenberg return NULL; 9545bb912bSLars Ellenberg page = tmp; 9645bb912bSLars Ellenberg } 9745bb912bSLars Ellenberg 9845bb912bSLars Ellenberg /* add end of list marker for the returned list */ 9945bb912bSLars Ellenberg set_page_private(page, 0); 10045bb912bSLars Ellenberg /* actual return value, and adjustment of head */ 10145bb912bSLars Ellenberg page = *head; 10245bb912bSLars Ellenberg *head = tmp; 10345bb912bSLars Ellenberg return page; 10445bb912bSLars Ellenberg } 10545bb912bSLars Ellenberg 10645bb912bSLars Ellenberg /* may be used outside of locks to find the tail of a (usually short) 10745bb912bSLars Ellenberg * "private" page chain, before adding it back to a global chain head 10845bb912bSLars Ellenberg * with page_chain_add() under a spinlock. */ 10945bb912bSLars Ellenberg static struct page *page_chain_tail(struct page *page, int *len) 11045bb912bSLars Ellenberg { 11145bb912bSLars Ellenberg struct page *tmp; 11245bb912bSLars Ellenberg int i = 1; 11345bb912bSLars Ellenberg while ((tmp = page_chain_next(page))) 11445bb912bSLars Ellenberg ++i, page = tmp; 11545bb912bSLars Ellenberg if (len) 11645bb912bSLars Ellenberg *len = i; 11745bb912bSLars Ellenberg return page; 11845bb912bSLars Ellenberg } 11945bb912bSLars Ellenberg 12045bb912bSLars Ellenberg static int page_chain_free(struct page *page) 12145bb912bSLars Ellenberg { 12245bb912bSLars Ellenberg struct page *tmp; 12345bb912bSLars Ellenberg int i = 0; 12445bb912bSLars Ellenberg page_chain_for_each_safe(page, tmp) { 12545bb912bSLars Ellenberg put_page(page); 12645bb912bSLars Ellenberg ++i; 12745bb912bSLars Ellenberg } 12845bb912bSLars Ellenberg return i; 12945bb912bSLars Ellenberg } 13045bb912bSLars Ellenberg 13145bb912bSLars Ellenberg static void page_chain_add(struct page **head, 13245bb912bSLars Ellenberg struct page *chain_first, struct page *chain_last) 13345bb912bSLars Ellenberg { 13445bb912bSLars Ellenberg #if 1 13545bb912bSLars Ellenberg struct page *tmp; 13645bb912bSLars Ellenberg tmp = page_chain_tail(chain_first, NULL); 13745bb912bSLars Ellenberg BUG_ON(tmp != chain_last); 13845bb912bSLars Ellenberg #endif 13945bb912bSLars Ellenberg 14045bb912bSLars Ellenberg /* add chain to head */ 14145bb912bSLars Ellenberg set_page_private(chain_last, (unsigned long)*head); 14245bb912bSLars Ellenberg *head = chain_first; 14345bb912bSLars Ellenberg } 14445bb912bSLars Ellenberg 14545bb912bSLars Ellenberg static struct page *drbd_pp_first_pages_or_try_alloc(struct drbd_conf *mdev, int number) 146b411b363SPhilipp Reisner { 147b411b363SPhilipp Reisner struct page *page = NULL; 14845bb912bSLars Ellenberg struct page *tmp = NULL; 14945bb912bSLars Ellenberg int i = 0; 150b411b363SPhilipp Reisner 151b411b363SPhilipp Reisner /* Yes, testing drbd_pp_vacant outside the lock is racy. 152b411b363SPhilipp Reisner * So what. It saves a spin_lock. */ 15345bb912bSLars Ellenberg if (drbd_pp_vacant >= number) { 154b411b363SPhilipp Reisner spin_lock(&drbd_pp_lock); 15545bb912bSLars Ellenberg page = page_chain_del(&drbd_pp_pool, number); 15645bb912bSLars Ellenberg if (page) 15745bb912bSLars Ellenberg drbd_pp_vacant -= number; 158b411b363SPhilipp Reisner spin_unlock(&drbd_pp_lock); 15945bb912bSLars Ellenberg if (page) 16045bb912bSLars Ellenberg return page; 161b411b363SPhilipp Reisner } 16245bb912bSLars Ellenberg 163b411b363SPhilipp Reisner /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD 164b411b363SPhilipp Reisner * "criss-cross" setup, that might cause write-out on some other DRBD, 165b411b363SPhilipp Reisner * which in turn might block on the other node at this very place. */ 16645bb912bSLars Ellenberg for (i = 0; i < number; i++) { 16745bb912bSLars Ellenberg tmp = alloc_page(GFP_TRY); 16845bb912bSLars Ellenberg if (!tmp) 16945bb912bSLars Ellenberg break; 17045bb912bSLars Ellenberg set_page_private(tmp, (unsigned long)page); 17145bb912bSLars Ellenberg page = tmp; 17245bb912bSLars Ellenberg } 17345bb912bSLars Ellenberg 17445bb912bSLars Ellenberg if (i == number) 175b411b363SPhilipp Reisner return page; 17645bb912bSLars Ellenberg 17745bb912bSLars Ellenberg /* Not enough pages immediately available this time. 17845bb912bSLars Ellenberg * No need to jump around here, drbd_pp_alloc will retry this 17945bb912bSLars Ellenberg * function "soon". */ 18045bb912bSLars Ellenberg if (page) { 18145bb912bSLars Ellenberg tmp = page_chain_tail(page, NULL); 18245bb912bSLars Ellenberg spin_lock(&drbd_pp_lock); 18345bb912bSLars Ellenberg page_chain_add(&drbd_pp_pool, page, tmp); 18445bb912bSLars Ellenberg drbd_pp_vacant += i; 18545bb912bSLars Ellenberg spin_unlock(&drbd_pp_lock); 18645bb912bSLars Ellenberg } 18745bb912bSLars Ellenberg return NULL; 188b411b363SPhilipp Reisner } 189b411b363SPhilipp Reisner 190b411b363SPhilipp Reisner static void reclaim_net_ee(struct drbd_conf *mdev, struct list_head *to_be_freed) 191b411b363SPhilipp Reisner { 192b411b363SPhilipp Reisner struct drbd_epoch_entry *e; 193b411b363SPhilipp Reisner struct list_head *le, *tle; 194b411b363SPhilipp Reisner 195b411b363SPhilipp Reisner /* The EEs are always appended to the end of the list. Since 196b411b363SPhilipp Reisner they are sent in order over the wire, they have to finish 197b411b363SPhilipp Reisner in order. As soon as we see the first not finished we can 198b411b363SPhilipp Reisner stop to examine the list... */ 199b411b363SPhilipp Reisner 200b411b363SPhilipp Reisner list_for_each_safe(le, tle, &mdev->net_ee) { 201b411b363SPhilipp Reisner e = list_entry(le, struct drbd_epoch_entry, w.list); 20245bb912bSLars Ellenberg if (drbd_ee_has_active_page(e)) 203b411b363SPhilipp Reisner break; 204b411b363SPhilipp Reisner list_move(le, to_be_freed); 205b411b363SPhilipp Reisner } 206b411b363SPhilipp Reisner } 207b411b363SPhilipp Reisner 208b411b363SPhilipp Reisner static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev) 209b411b363SPhilipp Reisner { 210b411b363SPhilipp Reisner LIST_HEAD(reclaimed); 211b411b363SPhilipp Reisner struct drbd_epoch_entry *e, *t; 212b411b363SPhilipp Reisner 213b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 214b411b363SPhilipp Reisner reclaim_net_ee(mdev, &reclaimed); 215b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 216b411b363SPhilipp Reisner 217b411b363SPhilipp Reisner list_for_each_entry_safe(e, t, &reclaimed, w.list) 218435f0740SLars Ellenberg drbd_free_net_ee(mdev, e); 219b411b363SPhilipp Reisner } 220b411b363SPhilipp Reisner 221b411b363SPhilipp Reisner /** 22245bb912bSLars Ellenberg * drbd_pp_alloc() - Returns @number pages, retries forever (or until signalled) 223b411b363SPhilipp Reisner * @mdev: DRBD device. 22445bb912bSLars Ellenberg * @number: number of pages requested 22545bb912bSLars Ellenberg * @retry: whether to retry, if not enough pages are available right now 226b411b363SPhilipp Reisner * 22745bb912bSLars Ellenberg * Tries to allocate number pages, first from our own page pool, then from 22845bb912bSLars Ellenberg * the kernel, unless this allocation would exceed the max_buffers setting. 22945bb912bSLars Ellenberg * Possibly retry until DRBD frees sufficient pages somewhere else. 23045bb912bSLars Ellenberg * 23145bb912bSLars Ellenberg * Returns a page chain linked via page->private. 232b411b363SPhilipp Reisner */ 23345bb912bSLars Ellenberg static struct page *drbd_pp_alloc(struct drbd_conf *mdev, unsigned number, bool retry) 234b411b363SPhilipp Reisner { 235b411b363SPhilipp Reisner struct page *page = NULL; 236b411b363SPhilipp Reisner DEFINE_WAIT(wait); 237b411b363SPhilipp Reisner 23845bb912bSLars Ellenberg /* Yes, we may run up to @number over max_buffers. If we 23945bb912bSLars Ellenberg * follow it strictly, the admin will get it wrong anyways. */ 24045bb912bSLars Ellenberg if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) 24145bb912bSLars Ellenberg page = drbd_pp_first_pages_or_try_alloc(mdev, number); 242b411b363SPhilipp Reisner 24345bb912bSLars Ellenberg while (page == NULL) { 244b411b363SPhilipp Reisner prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); 245b411b363SPhilipp Reisner 246b411b363SPhilipp Reisner drbd_kick_lo_and_reclaim_net(mdev); 247b411b363SPhilipp Reisner 248b411b363SPhilipp Reisner if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) { 24945bb912bSLars Ellenberg page = drbd_pp_first_pages_or_try_alloc(mdev, number); 250b411b363SPhilipp Reisner if (page) 251b411b363SPhilipp Reisner break; 252b411b363SPhilipp Reisner } 253b411b363SPhilipp Reisner 254b411b363SPhilipp Reisner if (!retry) 255b411b363SPhilipp Reisner break; 256b411b363SPhilipp Reisner 257b411b363SPhilipp Reisner if (signal_pending(current)) { 258b411b363SPhilipp Reisner dev_warn(DEV, "drbd_pp_alloc interrupted!\n"); 259b411b363SPhilipp Reisner break; 260b411b363SPhilipp Reisner } 261b411b363SPhilipp Reisner 262b411b363SPhilipp Reisner schedule(); 263b411b363SPhilipp Reisner } 264b411b363SPhilipp Reisner finish_wait(&drbd_pp_wait, &wait); 265b411b363SPhilipp Reisner 26645bb912bSLars Ellenberg if (page) 26745bb912bSLars Ellenberg atomic_add(number, &mdev->pp_in_use); 268b411b363SPhilipp Reisner return page; 269b411b363SPhilipp Reisner } 270b411b363SPhilipp Reisner 271b411b363SPhilipp Reisner /* Must not be used from irq, as that may deadlock: see drbd_pp_alloc. 27245bb912bSLars Ellenberg * Is also used from inside an other spin_lock_irq(&mdev->req_lock); 27345bb912bSLars Ellenberg * Either links the page chain back to the global pool, 27445bb912bSLars Ellenberg * or returns all pages to the system. */ 275435f0740SLars Ellenberg static void drbd_pp_free(struct drbd_conf *mdev, struct page *page, int is_net) 276b411b363SPhilipp Reisner { 277435f0740SLars Ellenberg atomic_t *a = is_net ? &mdev->pp_in_use_by_net : &mdev->pp_in_use; 278b411b363SPhilipp Reisner int i; 279435f0740SLars Ellenberg 2801816a2b4SLars Ellenberg if (drbd_pp_vacant > (DRBD_MAX_BIO_SIZE/PAGE_SIZE)*minor_count) 28145bb912bSLars Ellenberg i = page_chain_free(page); 28245bb912bSLars Ellenberg else { 28345bb912bSLars Ellenberg struct page *tmp; 28445bb912bSLars Ellenberg tmp = page_chain_tail(page, &i); 285b411b363SPhilipp Reisner spin_lock(&drbd_pp_lock); 28645bb912bSLars Ellenberg page_chain_add(&drbd_pp_pool, page, tmp); 28745bb912bSLars Ellenberg drbd_pp_vacant += i; 288b411b363SPhilipp Reisner spin_unlock(&drbd_pp_lock); 289b411b363SPhilipp Reisner } 290435f0740SLars Ellenberg i = atomic_sub_return(i, a); 29145bb912bSLars Ellenberg if (i < 0) 292435f0740SLars Ellenberg dev_warn(DEV, "ASSERTION FAILED: %s: %d < 0\n", 293435f0740SLars Ellenberg is_net ? "pp_in_use_by_net" : "pp_in_use", i); 294b411b363SPhilipp Reisner wake_up(&drbd_pp_wait); 295b411b363SPhilipp Reisner } 296b411b363SPhilipp Reisner 297b411b363SPhilipp Reisner /* 298b411b363SPhilipp Reisner You need to hold the req_lock: 299b411b363SPhilipp Reisner _drbd_wait_ee_list_empty() 300b411b363SPhilipp Reisner 301b411b363SPhilipp Reisner You must not have the req_lock: 302b411b363SPhilipp Reisner drbd_free_ee() 303b411b363SPhilipp Reisner drbd_alloc_ee() 304b411b363SPhilipp Reisner drbd_init_ee() 305b411b363SPhilipp Reisner drbd_release_ee() 306b411b363SPhilipp Reisner drbd_ee_fix_bhs() 307b411b363SPhilipp Reisner drbd_process_done_ee() 308b411b363SPhilipp Reisner drbd_clear_done_ee() 309b411b363SPhilipp Reisner drbd_wait_ee_list_empty() 310b411b363SPhilipp Reisner */ 311b411b363SPhilipp Reisner 312b411b363SPhilipp Reisner struct drbd_epoch_entry *drbd_alloc_ee(struct drbd_conf *mdev, 313b411b363SPhilipp Reisner u64 id, 314b411b363SPhilipp Reisner sector_t sector, 315b411b363SPhilipp Reisner unsigned int data_size, 316b411b363SPhilipp Reisner gfp_t gfp_mask) __must_hold(local) 317b411b363SPhilipp Reisner { 318b411b363SPhilipp Reisner struct drbd_epoch_entry *e; 319b411b363SPhilipp Reisner struct page *page; 32045bb912bSLars Ellenberg unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; 321b411b363SPhilipp Reisner 3220cf9d27eSAndreas Gruenbacher if (drbd_insert_fault(mdev, DRBD_FAULT_AL_EE)) 323b411b363SPhilipp Reisner return NULL; 324b411b363SPhilipp Reisner 325b411b363SPhilipp Reisner e = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); 326b411b363SPhilipp Reisner if (!e) { 327b411b363SPhilipp Reisner if (!(gfp_mask & __GFP_NOWARN)) 328b411b363SPhilipp Reisner dev_err(DEV, "alloc_ee: Allocation of an EE failed\n"); 329b411b363SPhilipp Reisner return NULL; 330b411b363SPhilipp Reisner } 331b411b363SPhilipp Reisner 33245bb912bSLars Ellenberg page = drbd_pp_alloc(mdev, nr_pages, (gfp_mask & __GFP_WAIT)); 33345bb912bSLars Ellenberg if (!page) 33445bb912bSLars Ellenberg goto fail; 335b411b363SPhilipp Reisner 33624c4830cSBart Van Assche INIT_HLIST_NODE(&e->collision); 337b411b363SPhilipp Reisner e->epoch = NULL; 33845bb912bSLars Ellenberg e->mdev = mdev; 33945bb912bSLars Ellenberg e->pages = page; 34045bb912bSLars Ellenberg atomic_set(&e->pending_bios, 0); 34145bb912bSLars Ellenberg e->size = data_size; 342b411b363SPhilipp Reisner e->flags = 0; 34345bb912bSLars Ellenberg e->sector = sector; 34445bb912bSLars Ellenberg e->block_id = id; 345b411b363SPhilipp Reisner 346b411b363SPhilipp Reisner return e; 347b411b363SPhilipp Reisner 34845bb912bSLars Ellenberg fail: 349b411b363SPhilipp Reisner mempool_free(e, drbd_ee_mempool); 350b411b363SPhilipp Reisner return NULL; 351b411b363SPhilipp Reisner } 352b411b363SPhilipp Reisner 353435f0740SLars Ellenberg void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, int is_net) 354b411b363SPhilipp Reisner { 355c36c3cedSLars Ellenberg if (e->flags & EE_HAS_DIGEST) 356c36c3cedSLars Ellenberg kfree(e->digest); 357435f0740SLars Ellenberg drbd_pp_free(mdev, e->pages, is_net); 35845bb912bSLars Ellenberg D_ASSERT(atomic_read(&e->pending_bios) == 0); 35924c4830cSBart Van Assche D_ASSERT(hlist_unhashed(&e->collision)); 360b411b363SPhilipp Reisner mempool_free(e, drbd_ee_mempool); 361b411b363SPhilipp Reisner } 362b411b363SPhilipp Reisner 363b411b363SPhilipp Reisner int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list) 364b411b363SPhilipp Reisner { 365b411b363SPhilipp Reisner LIST_HEAD(work_list); 366b411b363SPhilipp Reisner struct drbd_epoch_entry *e, *t; 367b411b363SPhilipp Reisner int count = 0; 368435f0740SLars Ellenberg int is_net = list == &mdev->net_ee; 369b411b363SPhilipp Reisner 370b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 371b411b363SPhilipp Reisner list_splice_init(list, &work_list); 372b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 373b411b363SPhilipp Reisner 374b411b363SPhilipp Reisner list_for_each_entry_safe(e, t, &work_list, w.list) { 375435f0740SLars Ellenberg drbd_free_some_ee(mdev, e, is_net); 376b411b363SPhilipp Reisner count++; 377b411b363SPhilipp Reisner } 378b411b363SPhilipp Reisner return count; 379b411b363SPhilipp Reisner } 380b411b363SPhilipp Reisner 381b411b363SPhilipp Reisner 382b411b363SPhilipp Reisner /* 383b411b363SPhilipp Reisner * This function is called from _asender only_ 384b411b363SPhilipp Reisner * but see also comments in _req_mod(,barrier_acked) 385b411b363SPhilipp Reisner * and receive_Barrier. 386b411b363SPhilipp Reisner * 387b411b363SPhilipp Reisner * Move entries from net_ee to done_ee, if ready. 388b411b363SPhilipp Reisner * Grab done_ee, call all callbacks, free the entries. 389b411b363SPhilipp Reisner * The callbacks typically send out ACKs. 390b411b363SPhilipp Reisner */ 391b411b363SPhilipp Reisner static int drbd_process_done_ee(struct drbd_conf *mdev) 392b411b363SPhilipp Reisner { 393b411b363SPhilipp Reisner LIST_HEAD(work_list); 394b411b363SPhilipp Reisner LIST_HEAD(reclaimed); 395b411b363SPhilipp Reisner struct drbd_epoch_entry *e, *t; 396b411b363SPhilipp Reisner int ok = (mdev->state.conn >= C_WF_REPORT_PARAMS); 397b411b363SPhilipp Reisner 398b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 399b411b363SPhilipp Reisner reclaim_net_ee(mdev, &reclaimed); 400b411b363SPhilipp Reisner list_splice_init(&mdev->done_ee, &work_list); 401b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 402b411b363SPhilipp Reisner 403b411b363SPhilipp Reisner list_for_each_entry_safe(e, t, &reclaimed, w.list) 404435f0740SLars Ellenberg drbd_free_net_ee(mdev, e); 405b411b363SPhilipp Reisner 406b411b363SPhilipp Reisner /* possible callbacks here: 407b411b363SPhilipp Reisner * e_end_block, and e_end_resync_block, e_send_discard_ack. 408b411b363SPhilipp Reisner * all ignore the last argument. 409b411b363SPhilipp Reisner */ 410b411b363SPhilipp Reisner list_for_each_entry_safe(e, t, &work_list, w.list) { 411b411b363SPhilipp Reisner /* list_del not necessary, next/prev members not touched */ 412b411b363SPhilipp Reisner ok = e->w.cb(mdev, &e->w, !ok) && ok; 413b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 414b411b363SPhilipp Reisner } 415b411b363SPhilipp Reisner wake_up(&mdev->ee_wait); 416b411b363SPhilipp Reisner 417b411b363SPhilipp Reisner return ok; 418b411b363SPhilipp Reisner } 419b411b363SPhilipp Reisner 420b411b363SPhilipp Reisner void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) 421b411b363SPhilipp Reisner { 422b411b363SPhilipp Reisner DEFINE_WAIT(wait); 423b411b363SPhilipp Reisner 424b411b363SPhilipp Reisner /* avoids spin_lock/unlock 425b411b363SPhilipp Reisner * and calling prepare_to_wait in the fast path */ 426b411b363SPhilipp Reisner while (!list_empty(head)) { 427b411b363SPhilipp Reisner prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE); 428b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 4297eaceaccSJens Axboe io_schedule(); 430b411b363SPhilipp Reisner finish_wait(&mdev->ee_wait, &wait); 431b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 432b411b363SPhilipp Reisner } 433b411b363SPhilipp Reisner } 434b411b363SPhilipp Reisner 435b411b363SPhilipp Reisner void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) 436b411b363SPhilipp Reisner { 437b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 438b411b363SPhilipp Reisner _drbd_wait_ee_list_empty(mdev, head); 439b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 440b411b363SPhilipp Reisner } 441b411b363SPhilipp Reisner 442b411b363SPhilipp Reisner /* see also kernel_accept; which is only present since 2.6.18. 443b411b363SPhilipp Reisner * also we want to log which part of it failed, exactly */ 444b411b363SPhilipp Reisner static int drbd_accept(struct drbd_conf *mdev, const char **what, 445b411b363SPhilipp Reisner struct socket *sock, struct socket **newsock) 446b411b363SPhilipp Reisner { 447b411b363SPhilipp Reisner struct sock *sk = sock->sk; 448b411b363SPhilipp Reisner int err = 0; 449b411b363SPhilipp Reisner 450b411b363SPhilipp Reisner *what = "listen"; 451b411b363SPhilipp Reisner err = sock->ops->listen(sock, 5); 452b411b363SPhilipp Reisner if (err < 0) 453b411b363SPhilipp Reisner goto out; 454b411b363SPhilipp Reisner 455b411b363SPhilipp Reisner *what = "sock_create_lite"; 456b411b363SPhilipp Reisner err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol, 457b411b363SPhilipp Reisner newsock); 458b411b363SPhilipp Reisner if (err < 0) 459b411b363SPhilipp Reisner goto out; 460b411b363SPhilipp Reisner 461b411b363SPhilipp Reisner *what = "accept"; 462b411b363SPhilipp Reisner err = sock->ops->accept(sock, *newsock, 0); 463b411b363SPhilipp Reisner if (err < 0) { 464b411b363SPhilipp Reisner sock_release(*newsock); 465b411b363SPhilipp Reisner *newsock = NULL; 466b411b363SPhilipp Reisner goto out; 467b411b363SPhilipp Reisner } 468b411b363SPhilipp Reisner (*newsock)->ops = sock->ops; 469b411b363SPhilipp Reisner 470b411b363SPhilipp Reisner out: 471b411b363SPhilipp Reisner return err; 472b411b363SPhilipp Reisner } 473b411b363SPhilipp Reisner 474b411b363SPhilipp Reisner static int drbd_recv_short(struct drbd_conf *mdev, struct socket *sock, 475b411b363SPhilipp Reisner void *buf, size_t size, int flags) 476b411b363SPhilipp Reisner { 477b411b363SPhilipp Reisner mm_segment_t oldfs; 478b411b363SPhilipp Reisner struct kvec iov = { 479b411b363SPhilipp Reisner .iov_base = buf, 480b411b363SPhilipp Reisner .iov_len = size, 481b411b363SPhilipp Reisner }; 482b411b363SPhilipp Reisner struct msghdr msg = { 483b411b363SPhilipp Reisner .msg_iovlen = 1, 484b411b363SPhilipp Reisner .msg_iov = (struct iovec *)&iov, 485b411b363SPhilipp Reisner .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) 486b411b363SPhilipp Reisner }; 487b411b363SPhilipp Reisner int rv; 488b411b363SPhilipp Reisner 489b411b363SPhilipp Reisner oldfs = get_fs(); 490b411b363SPhilipp Reisner set_fs(KERNEL_DS); 491b411b363SPhilipp Reisner rv = sock_recvmsg(sock, &msg, size, msg.msg_flags); 492b411b363SPhilipp Reisner set_fs(oldfs); 493b411b363SPhilipp Reisner 494b411b363SPhilipp Reisner return rv; 495b411b363SPhilipp Reisner } 496b411b363SPhilipp Reisner 497b411b363SPhilipp Reisner static int drbd_recv(struct drbd_conf *mdev, void *buf, size_t size) 498b411b363SPhilipp Reisner { 499b411b363SPhilipp Reisner mm_segment_t oldfs; 500b411b363SPhilipp Reisner struct kvec iov = { 501b411b363SPhilipp Reisner .iov_base = buf, 502b411b363SPhilipp Reisner .iov_len = size, 503b411b363SPhilipp Reisner }; 504b411b363SPhilipp Reisner struct msghdr msg = { 505b411b363SPhilipp Reisner .msg_iovlen = 1, 506b411b363SPhilipp Reisner .msg_iov = (struct iovec *)&iov, 507b411b363SPhilipp Reisner .msg_flags = MSG_WAITALL | MSG_NOSIGNAL 508b411b363SPhilipp Reisner }; 509b411b363SPhilipp Reisner int rv; 510b411b363SPhilipp Reisner 511b411b363SPhilipp Reisner oldfs = get_fs(); 512b411b363SPhilipp Reisner set_fs(KERNEL_DS); 513b411b363SPhilipp Reisner 514b411b363SPhilipp Reisner for (;;) { 515b411b363SPhilipp Reisner rv = sock_recvmsg(mdev->data.socket, &msg, size, msg.msg_flags); 516b411b363SPhilipp Reisner if (rv == size) 517b411b363SPhilipp Reisner break; 518b411b363SPhilipp Reisner 519b411b363SPhilipp Reisner /* Note: 520b411b363SPhilipp Reisner * ECONNRESET other side closed the connection 521b411b363SPhilipp Reisner * ERESTARTSYS (on sock) we got a signal 522b411b363SPhilipp Reisner */ 523b411b363SPhilipp Reisner 524b411b363SPhilipp Reisner if (rv < 0) { 525b411b363SPhilipp Reisner if (rv == -ECONNRESET) 526b411b363SPhilipp Reisner dev_info(DEV, "sock was reset by peer\n"); 527b411b363SPhilipp Reisner else if (rv != -ERESTARTSYS) 528b411b363SPhilipp Reisner dev_err(DEV, "sock_recvmsg returned %d\n", rv); 529b411b363SPhilipp Reisner break; 530b411b363SPhilipp Reisner } else if (rv == 0) { 531b411b363SPhilipp Reisner dev_info(DEV, "sock was shut down by peer\n"); 532b411b363SPhilipp Reisner break; 533b411b363SPhilipp Reisner } else { 534b411b363SPhilipp Reisner /* signal came in, or peer/link went down, 535b411b363SPhilipp Reisner * after we read a partial message 536b411b363SPhilipp Reisner */ 537b411b363SPhilipp Reisner /* D_ASSERT(signal_pending(current)); */ 538b411b363SPhilipp Reisner break; 539b411b363SPhilipp Reisner } 540b411b363SPhilipp Reisner }; 541b411b363SPhilipp Reisner 542b411b363SPhilipp Reisner set_fs(oldfs); 543b411b363SPhilipp Reisner 544b411b363SPhilipp Reisner if (rv != size) 545b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_BROKEN_PIPE)); 546b411b363SPhilipp Reisner 547b411b363SPhilipp Reisner return rv; 548b411b363SPhilipp Reisner } 549b411b363SPhilipp Reisner 5505dbf1673SLars Ellenberg /* quoting tcp(7): 5515dbf1673SLars Ellenberg * On individual connections, the socket buffer size must be set prior to the 5525dbf1673SLars Ellenberg * listen(2) or connect(2) calls in order to have it take effect. 5535dbf1673SLars Ellenberg * This is our wrapper to do so. 5545dbf1673SLars Ellenberg */ 5555dbf1673SLars Ellenberg static void drbd_setbufsize(struct socket *sock, unsigned int snd, 5565dbf1673SLars Ellenberg unsigned int rcv) 5575dbf1673SLars Ellenberg { 5585dbf1673SLars Ellenberg /* open coded SO_SNDBUF, SO_RCVBUF */ 5595dbf1673SLars Ellenberg if (snd) { 5605dbf1673SLars Ellenberg sock->sk->sk_sndbuf = snd; 5615dbf1673SLars Ellenberg sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 5625dbf1673SLars Ellenberg } 5635dbf1673SLars Ellenberg if (rcv) { 5645dbf1673SLars Ellenberg sock->sk->sk_rcvbuf = rcv; 5655dbf1673SLars Ellenberg sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 5665dbf1673SLars Ellenberg } 5675dbf1673SLars Ellenberg } 5685dbf1673SLars Ellenberg 569b411b363SPhilipp Reisner static struct socket *drbd_try_connect(struct drbd_conf *mdev) 570b411b363SPhilipp Reisner { 571b411b363SPhilipp Reisner const char *what; 572b411b363SPhilipp Reisner struct socket *sock; 573b411b363SPhilipp Reisner struct sockaddr_in6 src_in6; 574b411b363SPhilipp Reisner int err; 575b411b363SPhilipp Reisner int disconnect_on_error = 1; 576b411b363SPhilipp Reisner 577b411b363SPhilipp Reisner if (!get_net_conf(mdev)) 578b411b363SPhilipp Reisner return NULL; 579b411b363SPhilipp Reisner 580b411b363SPhilipp Reisner what = "sock_create_kern"; 581b411b363SPhilipp Reisner err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family, 582b411b363SPhilipp Reisner SOCK_STREAM, IPPROTO_TCP, &sock); 583b411b363SPhilipp Reisner if (err < 0) { 584b411b363SPhilipp Reisner sock = NULL; 585b411b363SPhilipp Reisner goto out; 586b411b363SPhilipp Reisner } 587b411b363SPhilipp Reisner 588b411b363SPhilipp Reisner sock->sk->sk_rcvtimeo = 589b411b363SPhilipp Reisner sock->sk->sk_sndtimeo = mdev->net_conf->try_connect_int*HZ; 5905dbf1673SLars Ellenberg drbd_setbufsize(sock, mdev->net_conf->sndbuf_size, 5915dbf1673SLars Ellenberg mdev->net_conf->rcvbuf_size); 592b411b363SPhilipp Reisner 593b411b363SPhilipp Reisner /* explicitly bind to the configured IP as source IP 594b411b363SPhilipp Reisner * for the outgoing connections. 595b411b363SPhilipp Reisner * This is needed for multihomed hosts and to be 596b411b363SPhilipp Reisner * able to use lo: interfaces for drbd. 597b411b363SPhilipp Reisner * Make sure to use 0 as port number, so linux selects 598b411b363SPhilipp Reisner * a free one dynamically. 599b411b363SPhilipp Reisner */ 600b411b363SPhilipp Reisner memcpy(&src_in6, mdev->net_conf->my_addr, 601b411b363SPhilipp Reisner min_t(int, mdev->net_conf->my_addr_len, sizeof(src_in6))); 602b411b363SPhilipp Reisner if (((struct sockaddr *)mdev->net_conf->my_addr)->sa_family == AF_INET6) 603b411b363SPhilipp Reisner src_in6.sin6_port = 0; 604b411b363SPhilipp Reisner else 605b411b363SPhilipp Reisner ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ 606b411b363SPhilipp Reisner 607b411b363SPhilipp Reisner what = "bind before connect"; 608b411b363SPhilipp Reisner err = sock->ops->bind(sock, 609b411b363SPhilipp Reisner (struct sockaddr *) &src_in6, 610b411b363SPhilipp Reisner mdev->net_conf->my_addr_len); 611b411b363SPhilipp Reisner if (err < 0) 612b411b363SPhilipp Reisner goto out; 613b411b363SPhilipp Reisner 614b411b363SPhilipp Reisner /* connect may fail, peer not yet available. 615b411b363SPhilipp Reisner * stay C_WF_CONNECTION, don't go Disconnecting! */ 616b411b363SPhilipp Reisner disconnect_on_error = 0; 617b411b363SPhilipp Reisner what = "connect"; 618b411b363SPhilipp Reisner err = sock->ops->connect(sock, 619b411b363SPhilipp Reisner (struct sockaddr *)mdev->net_conf->peer_addr, 620b411b363SPhilipp Reisner mdev->net_conf->peer_addr_len, 0); 621b411b363SPhilipp Reisner 622b411b363SPhilipp Reisner out: 623b411b363SPhilipp Reisner if (err < 0) { 624b411b363SPhilipp Reisner if (sock) { 625b411b363SPhilipp Reisner sock_release(sock); 626b411b363SPhilipp Reisner sock = NULL; 627b411b363SPhilipp Reisner } 628b411b363SPhilipp Reisner switch (-err) { 629b411b363SPhilipp Reisner /* timeout, busy, signal pending */ 630b411b363SPhilipp Reisner case ETIMEDOUT: case EAGAIN: case EINPROGRESS: 631b411b363SPhilipp Reisner case EINTR: case ERESTARTSYS: 632b411b363SPhilipp Reisner /* peer not (yet) available, network problem */ 633b411b363SPhilipp Reisner case ECONNREFUSED: case ENETUNREACH: 634b411b363SPhilipp Reisner case EHOSTDOWN: case EHOSTUNREACH: 635b411b363SPhilipp Reisner disconnect_on_error = 0; 636b411b363SPhilipp Reisner break; 637b411b363SPhilipp Reisner default: 638b411b363SPhilipp Reisner dev_err(DEV, "%s failed, err = %d\n", what, err); 639b411b363SPhilipp Reisner } 640b411b363SPhilipp Reisner if (disconnect_on_error) 641b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 642b411b363SPhilipp Reisner } 643b411b363SPhilipp Reisner put_net_conf(mdev); 644b411b363SPhilipp Reisner return sock; 645b411b363SPhilipp Reisner } 646b411b363SPhilipp Reisner 647b411b363SPhilipp Reisner static struct socket *drbd_wait_for_connect(struct drbd_conf *mdev) 648b411b363SPhilipp Reisner { 649b411b363SPhilipp Reisner int timeo, err; 650b411b363SPhilipp Reisner struct socket *s_estab = NULL, *s_listen; 651b411b363SPhilipp Reisner const char *what; 652b411b363SPhilipp Reisner 653b411b363SPhilipp Reisner if (!get_net_conf(mdev)) 654b411b363SPhilipp Reisner return NULL; 655b411b363SPhilipp Reisner 656b411b363SPhilipp Reisner what = "sock_create_kern"; 657b411b363SPhilipp Reisner err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family, 658b411b363SPhilipp Reisner SOCK_STREAM, IPPROTO_TCP, &s_listen); 659b411b363SPhilipp Reisner if (err) { 660b411b363SPhilipp Reisner s_listen = NULL; 661b411b363SPhilipp Reisner goto out; 662b411b363SPhilipp Reisner } 663b411b363SPhilipp Reisner 664b411b363SPhilipp Reisner timeo = mdev->net_conf->try_connect_int * HZ; 665b411b363SPhilipp Reisner timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */ 666b411b363SPhilipp Reisner 667b411b363SPhilipp Reisner s_listen->sk->sk_reuse = 1; /* SO_REUSEADDR */ 668b411b363SPhilipp Reisner s_listen->sk->sk_rcvtimeo = timeo; 669b411b363SPhilipp Reisner s_listen->sk->sk_sndtimeo = timeo; 6705dbf1673SLars Ellenberg drbd_setbufsize(s_listen, mdev->net_conf->sndbuf_size, 6715dbf1673SLars Ellenberg mdev->net_conf->rcvbuf_size); 672b411b363SPhilipp Reisner 673b411b363SPhilipp Reisner what = "bind before listen"; 674b411b363SPhilipp Reisner err = s_listen->ops->bind(s_listen, 675b411b363SPhilipp Reisner (struct sockaddr *) mdev->net_conf->my_addr, 676b411b363SPhilipp Reisner mdev->net_conf->my_addr_len); 677b411b363SPhilipp Reisner if (err < 0) 678b411b363SPhilipp Reisner goto out; 679b411b363SPhilipp Reisner 680b411b363SPhilipp Reisner err = drbd_accept(mdev, &what, s_listen, &s_estab); 681b411b363SPhilipp Reisner 682b411b363SPhilipp Reisner out: 683b411b363SPhilipp Reisner if (s_listen) 684b411b363SPhilipp Reisner sock_release(s_listen); 685b411b363SPhilipp Reisner if (err < 0) { 686b411b363SPhilipp Reisner if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { 687b411b363SPhilipp Reisner dev_err(DEV, "%s failed, err = %d\n", what, err); 688b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 689b411b363SPhilipp Reisner } 690b411b363SPhilipp Reisner } 691b411b363SPhilipp Reisner put_net_conf(mdev); 692b411b363SPhilipp Reisner 693b411b363SPhilipp Reisner return s_estab; 694b411b363SPhilipp Reisner } 695b411b363SPhilipp Reisner 696b411b363SPhilipp Reisner static int drbd_send_fp(struct drbd_conf *mdev, 697b411b363SPhilipp Reisner struct socket *sock, enum drbd_packets cmd) 698b411b363SPhilipp Reisner { 69902918be2SPhilipp Reisner struct p_header80 *h = &mdev->data.sbuf.header.h80; 700b411b363SPhilipp Reisner 701b411b363SPhilipp Reisner return _drbd_send_cmd(mdev, sock, cmd, h, sizeof(*h), 0); 702b411b363SPhilipp Reisner } 703b411b363SPhilipp Reisner 704b411b363SPhilipp Reisner static enum drbd_packets drbd_recv_fp(struct drbd_conf *mdev, struct socket *sock) 705b411b363SPhilipp Reisner { 70602918be2SPhilipp Reisner struct p_header80 *h = &mdev->data.rbuf.header.h80; 707b411b363SPhilipp Reisner int rr; 708b411b363SPhilipp Reisner 709b411b363SPhilipp Reisner rr = drbd_recv_short(mdev, sock, h, sizeof(*h), 0); 710b411b363SPhilipp Reisner 711b411b363SPhilipp Reisner if (rr == sizeof(*h) && h->magic == BE_DRBD_MAGIC) 712b411b363SPhilipp Reisner return be16_to_cpu(h->command); 713b411b363SPhilipp Reisner 714b411b363SPhilipp Reisner return 0xffff; 715b411b363SPhilipp Reisner } 716b411b363SPhilipp Reisner 717b411b363SPhilipp Reisner /** 718b411b363SPhilipp Reisner * drbd_socket_okay() - Free the socket if its connection is not okay 719b411b363SPhilipp Reisner * @mdev: DRBD device. 720b411b363SPhilipp Reisner * @sock: pointer to the pointer to the socket. 721b411b363SPhilipp Reisner */ 722b411b363SPhilipp Reisner static int drbd_socket_okay(struct drbd_conf *mdev, struct socket **sock) 723b411b363SPhilipp Reisner { 724b411b363SPhilipp Reisner int rr; 725b411b363SPhilipp Reisner char tb[4]; 726b411b363SPhilipp Reisner 727b411b363SPhilipp Reisner if (!*sock) 72881e84650SAndreas Gruenbacher return false; 729b411b363SPhilipp Reisner 730b411b363SPhilipp Reisner rr = drbd_recv_short(mdev, *sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); 731b411b363SPhilipp Reisner 732b411b363SPhilipp Reisner if (rr > 0 || rr == -EAGAIN) { 73381e84650SAndreas Gruenbacher return true; 734b411b363SPhilipp Reisner } else { 735b411b363SPhilipp Reisner sock_release(*sock); 736b411b363SPhilipp Reisner *sock = NULL; 73781e84650SAndreas Gruenbacher return false; 738b411b363SPhilipp Reisner } 739b411b363SPhilipp Reisner } 740b411b363SPhilipp Reisner 741b411b363SPhilipp Reisner /* 742b411b363SPhilipp Reisner * return values: 743b411b363SPhilipp Reisner * 1 yes, we have a valid connection 744b411b363SPhilipp Reisner * 0 oops, did not work out, please try again 745b411b363SPhilipp Reisner * -1 peer talks different language, 746b411b363SPhilipp Reisner * no point in trying again, please go standalone. 747b411b363SPhilipp Reisner * -2 We do not have a network config... 748b411b363SPhilipp Reisner */ 749b411b363SPhilipp Reisner static int drbd_connect(struct drbd_conf *mdev) 750b411b363SPhilipp Reisner { 751b411b363SPhilipp Reisner struct socket *s, *sock, *msock; 752b411b363SPhilipp Reisner int try, h, ok; 753b411b363SPhilipp Reisner 754b411b363SPhilipp Reisner D_ASSERT(!mdev->data.socket); 755b411b363SPhilipp Reisner 756b411b363SPhilipp Reisner if (drbd_request_state(mdev, NS(conn, C_WF_CONNECTION)) < SS_SUCCESS) 757b411b363SPhilipp Reisner return -2; 758b411b363SPhilipp Reisner 759b411b363SPhilipp Reisner clear_bit(DISCARD_CONCURRENT, &mdev->flags); 760b411b363SPhilipp Reisner 761b411b363SPhilipp Reisner sock = NULL; 762b411b363SPhilipp Reisner msock = NULL; 763b411b363SPhilipp Reisner 764b411b363SPhilipp Reisner do { 765b411b363SPhilipp Reisner for (try = 0;;) { 766b411b363SPhilipp Reisner /* 3 tries, this should take less than a second! */ 767b411b363SPhilipp Reisner s = drbd_try_connect(mdev); 768b411b363SPhilipp Reisner if (s || ++try >= 3) 769b411b363SPhilipp Reisner break; 770b411b363SPhilipp Reisner /* give the other side time to call bind() & listen() */ 77120ee6390SPhilipp Reisner schedule_timeout_interruptible(HZ / 10); 772b411b363SPhilipp Reisner } 773b411b363SPhilipp Reisner 774b411b363SPhilipp Reisner if (s) { 775b411b363SPhilipp Reisner if (!sock) { 776b411b363SPhilipp Reisner drbd_send_fp(mdev, s, P_HAND_SHAKE_S); 777b411b363SPhilipp Reisner sock = s; 778b411b363SPhilipp Reisner s = NULL; 779b411b363SPhilipp Reisner } else if (!msock) { 780b411b363SPhilipp Reisner drbd_send_fp(mdev, s, P_HAND_SHAKE_M); 781b411b363SPhilipp Reisner msock = s; 782b411b363SPhilipp Reisner s = NULL; 783b411b363SPhilipp Reisner } else { 784b411b363SPhilipp Reisner dev_err(DEV, "Logic error in drbd_connect()\n"); 785b411b363SPhilipp Reisner goto out_release_sockets; 786b411b363SPhilipp Reisner } 787b411b363SPhilipp Reisner } 788b411b363SPhilipp Reisner 789b411b363SPhilipp Reisner if (sock && msock) { 790a8e40792SPhilipp Reisner schedule_timeout_interruptible(mdev->net_conf->ping_timeo*HZ/10); 791b411b363SPhilipp Reisner ok = drbd_socket_okay(mdev, &sock); 792b411b363SPhilipp Reisner ok = drbd_socket_okay(mdev, &msock) && ok; 793b411b363SPhilipp Reisner if (ok) 794b411b363SPhilipp Reisner break; 795b411b363SPhilipp Reisner } 796b411b363SPhilipp Reisner 797b411b363SPhilipp Reisner retry: 798b411b363SPhilipp Reisner s = drbd_wait_for_connect(mdev); 799b411b363SPhilipp Reisner if (s) { 800b411b363SPhilipp Reisner try = drbd_recv_fp(mdev, s); 801b411b363SPhilipp Reisner drbd_socket_okay(mdev, &sock); 802b411b363SPhilipp Reisner drbd_socket_okay(mdev, &msock); 803b411b363SPhilipp Reisner switch (try) { 804b411b363SPhilipp Reisner case P_HAND_SHAKE_S: 805b411b363SPhilipp Reisner if (sock) { 806b411b363SPhilipp Reisner dev_warn(DEV, "initial packet S crossed\n"); 807b411b363SPhilipp Reisner sock_release(sock); 808b411b363SPhilipp Reisner } 809b411b363SPhilipp Reisner sock = s; 810b411b363SPhilipp Reisner break; 811b411b363SPhilipp Reisner case P_HAND_SHAKE_M: 812b411b363SPhilipp Reisner if (msock) { 813b411b363SPhilipp Reisner dev_warn(DEV, "initial packet M crossed\n"); 814b411b363SPhilipp Reisner sock_release(msock); 815b411b363SPhilipp Reisner } 816b411b363SPhilipp Reisner msock = s; 817b411b363SPhilipp Reisner set_bit(DISCARD_CONCURRENT, &mdev->flags); 818b411b363SPhilipp Reisner break; 819b411b363SPhilipp Reisner default: 820b411b363SPhilipp Reisner dev_warn(DEV, "Error receiving initial packet\n"); 821b411b363SPhilipp Reisner sock_release(s); 822b411b363SPhilipp Reisner if (random32() & 1) 823b411b363SPhilipp Reisner goto retry; 824b411b363SPhilipp Reisner } 825b411b363SPhilipp Reisner } 826b411b363SPhilipp Reisner 827b411b363SPhilipp Reisner if (mdev->state.conn <= C_DISCONNECTING) 828b411b363SPhilipp Reisner goto out_release_sockets; 829b411b363SPhilipp Reisner if (signal_pending(current)) { 830b411b363SPhilipp Reisner flush_signals(current); 831b411b363SPhilipp Reisner smp_rmb(); 832b411b363SPhilipp Reisner if (get_t_state(&mdev->receiver) == Exiting) 833b411b363SPhilipp Reisner goto out_release_sockets; 834b411b363SPhilipp Reisner } 835b411b363SPhilipp Reisner 836b411b363SPhilipp Reisner if (sock && msock) { 837b411b363SPhilipp Reisner ok = drbd_socket_okay(mdev, &sock); 838b411b363SPhilipp Reisner ok = drbd_socket_okay(mdev, &msock) && ok; 839b411b363SPhilipp Reisner if (ok) 840b411b363SPhilipp Reisner break; 841b411b363SPhilipp Reisner } 842b411b363SPhilipp Reisner } while (1); 843b411b363SPhilipp Reisner 844b411b363SPhilipp Reisner msock->sk->sk_reuse = 1; /* SO_REUSEADDR */ 845b411b363SPhilipp Reisner sock->sk->sk_reuse = 1; /* SO_REUSEADDR */ 846b411b363SPhilipp Reisner 847b411b363SPhilipp Reisner sock->sk->sk_allocation = GFP_NOIO; 848b411b363SPhilipp Reisner msock->sk->sk_allocation = GFP_NOIO; 849b411b363SPhilipp Reisner 850b411b363SPhilipp Reisner sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; 851b411b363SPhilipp Reisner msock->sk->sk_priority = TC_PRIO_INTERACTIVE; 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. 86525985edcSLucas De Marchi * we use TCP_CORK where appropriate, 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 902148efa16SPhilipp Reisner if (drbd_send_protocol(mdev) == -1) 9037e2455c1SPhilipp Reisner return -1; 904b411b363SPhilipp Reisner drbd_send_sync_param(mdev, &mdev->sync_conf); 905e89b591cSPhilipp Reisner drbd_send_sizes(mdev, 0, 0); 906b411b363SPhilipp Reisner drbd_send_uuids(mdev); 907b411b363SPhilipp Reisner drbd_send_state(mdev); 908b411b363SPhilipp Reisner clear_bit(USE_DEGR_WFC_T, &mdev->flags); 909b411b363SPhilipp Reisner clear_bit(RESIZE_PENDING, &mdev->flags); 9107fde2be9SPhilipp Reisner mod_timer(&mdev->request_timer, jiffies + HZ); /* just start it here. */ 911b411b363SPhilipp Reisner 912b411b363SPhilipp Reisner return 1; 913b411b363SPhilipp Reisner 914b411b363SPhilipp Reisner out_release_sockets: 915b411b363SPhilipp Reisner if (sock) 916b411b363SPhilipp Reisner sock_release(sock); 917b411b363SPhilipp Reisner if (msock) 918b411b363SPhilipp Reisner sock_release(msock); 919b411b363SPhilipp Reisner return -1; 920b411b363SPhilipp Reisner } 921b411b363SPhilipp Reisner 92202918be2SPhilipp Reisner static int drbd_recv_header(struct drbd_conf *mdev, enum drbd_packets *cmd, unsigned int *packet_size) 923b411b363SPhilipp Reisner { 92402918be2SPhilipp Reisner union p_header *h = &mdev->data.rbuf.header; 925b411b363SPhilipp Reisner int r; 926b411b363SPhilipp Reisner 927b411b363SPhilipp Reisner r = drbd_recv(mdev, h, sizeof(*h)); 928b411b363SPhilipp Reisner if (unlikely(r != sizeof(*h))) { 9290ddc5549SLars Ellenberg if (!signal_pending(current)) 9300ddc5549SLars Ellenberg dev_warn(DEV, "short read expecting header on sock: r=%d\n", r); 93181e84650SAndreas Gruenbacher return false; 93202918be2SPhilipp Reisner } 93302918be2SPhilipp Reisner 93402918be2SPhilipp Reisner if (likely(h->h80.magic == BE_DRBD_MAGIC)) { 93502918be2SPhilipp Reisner *cmd = be16_to_cpu(h->h80.command); 93602918be2SPhilipp Reisner *packet_size = be16_to_cpu(h->h80.length); 93702918be2SPhilipp Reisner } else if (h->h95.magic == BE_DRBD_MAGIC_BIG) { 93802918be2SPhilipp Reisner *cmd = be16_to_cpu(h->h95.command); 93902918be2SPhilipp Reisner *packet_size = be32_to_cpu(h->h95.length); 94002918be2SPhilipp Reisner } else { 941004352faSLars Ellenberg dev_err(DEV, "magic?? on data m: 0x%08x c: %d l: %d\n", 942004352faSLars Ellenberg be32_to_cpu(h->h80.magic), 943004352faSLars Ellenberg be16_to_cpu(h->h80.command), 944004352faSLars Ellenberg be16_to_cpu(h->h80.length)); 94581e84650SAndreas Gruenbacher return false; 946b411b363SPhilipp Reisner } 947b411b363SPhilipp Reisner mdev->last_received = jiffies; 948b411b363SPhilipp Reisner 94981e84650SAndreas Gruenbacher return true; 950b411b363SPhilipp Reisner } 951b411b363SPhilipp Reisner 9522451fc3bSPhilipp Reisner static void drbd_flush(struct drbd_conf *mdev) 953b411b363SPhilipp Reisner { 954b411b363SPhilipp Reisner int rv; 955b411b363SPhilipp Reisner 956b411b363SPhilipp Reisner if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) { 957fbd9b09aSDmitry Monakhov rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL, 958dd3932edSChristoph Hellwig NULL); 959b411b363SPhilipp Reisner if (rv) { 960b411b363SPhilipp Reisner dev_err(DEV, "local disk flush failed with status %d\n", rv); 961b411b363SPhilipp Reisner /* would rather check on EOPNOTSUPP, but that is not reliable. 962b411b363SPhilipp Reisner * don't try again for ANY return value != 0 963b411b363SPhilipp Reisner * if (rv == -EOPNOTSUPP) */ 964b411b363SPhilipp Reisner drbd_bump_write_ordering(mdev, WO_drain_io); 965b411b363SPhilipp Reisner } 966b411b363SPhilipp Reisner put_ldev(mdev); 967b411b363SPhilipp Reisner } 968b411b363SPhilipp Reisner } 969b411b363SPhilipp Reisner 970b411b363SPhilipp Reisner /** 971b411b363SPhilipp Reisner * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. 972b411b363SPhilipp Reisner * @mdev: DRBD device. 973b411b363SPhilipp Reisner * @epoch: Epoch object. 974b411b363SPhilipp Reisner * @ev: Epoch event. 975b411b363SPhilipp Reisner */ 976b411b363SPhilipp Reisner static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev, 977b411b363SPhilipp Reisner struct drbd_epoch *epoch, 978b411b363SPhilipp Reisner enum epoch_event ev) 979b411b363SPhilipp Reisner { 9802451fc3bSPhilipp Reisner int epoch_size; 981b411b363SPhilipp Reisner struct drbd_epoch *next_epoch; 982b411b363SPhilipp Reisner enum finish_epoch rv = FE_STILL_LIVE; 983b411b363SPhilipp Reisner 984b411b363SPhilipp Reisner spin_lock(&mdev->epoch_lock); 985b411b363SPhilipp Reisner do { 986b411b363SPhilipp Reisner next_epoch = NULL; 987b411b363SPhilipp Reisner 988b411b363SPhilipp Reisner epoch_size = atomic_read(&epoch->epoch_size); 989b411b363SPhilipp Reisner 990b411b363SPhilipp Reisner switch (ev & ~EV_CLEANUP) { 991b411b363SPhilipp Reisner case EV_PUT: 992b411b363SPhilipp Reisner atomic_dec(&epoch->active); 993b411b363SPhilipp Reisner break; 994b411b363SPhilipp Reisner case EV_GOT_BARRIER_NR: 995b411b363SPhilipp Reisner set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); 996b411b363SPhilipp Reisner break; 997b411b363SPhilipp Reisner case EV_BECAME_LAST: 998b411b363SPhilipp Reisner /* nothing to do*/ 999b411b363SPhilipp Reisner break; 1000b411b363SPhilipp Reisner } 1001b411b363SPhilipp Reisner 1002b411b363SPhilipp Reisner if (epoch_size != 0 && 1003b411b363SPhilipp Reisner atomic_read(&epoch->active) == 0 && 10042451fc3bSPhilipp Reisner test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) { 1005b411b363SPhilipp Reisner if (!(ev & EV_CLEANUP)) { 1006b411b363SPhilipp Reisner spin_unlock(&mdev->epoch_lock); 1007b411b363SPhilipp Reisner drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size); 1008b411b363SPhilipp Reisner spin_lock(&mdev->epoch_lock); 1009b411b363SPhilipp Reisner } 1010b411b363SPhilipp Reisner dec_unacked(mdev); 1011b411b363SPhilipp Reisner 1012b411b363SPhilipp Reisner if (mdev->current_epoch != epoch) { 1013b411b363SPhilipp Reisner next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); 1014b411b363SPhilipp Reisner list_del(&epoch->list); 1015b411b363SPhilipp Reisner ev = EV_BECAME_LAST | (ev & EV_CLEANUP); 1016b411b363SPhilipp Reisner mdev->epochs--; 1017b411b363SPhilipp Reisner kfree(epoch); 1018b411b363SPhilipp Reisner 1019b411b363SPhilipp Reisner if (rv == FE_STILL_LIVE) 1020b411b363SPhilipp Reisner rv = FE_DESTROYED; 1021b411b363SPhilipp Reisner } else { 1022b411b363SPhilipp Reisner epoch->flags = 0; 1023b411b363SPhilipp Reisner atomic_set(&epoch->epoch_size, 0); 1024698f9315SUwe Kleine-König /* atomic_set(&epoch->active, 0); is already zero */ 1025b411b363SPhilipp Reisner if (rv == FE_STILL_LIVE) 1026b411b363SPhilipp Reisner rv = FE_RECYCLED; 10272451fc3bSPhilipp Reisner wake_up(&mdev->ee_wait); 1028b411b363SPhilipp Reisner } 1029b411b363SPhilipp Reisner } 1030b411b363SPhilipp Reisner 1031b411b363SPhilipp Reisner if (!next_epoch) 1032b411b363SPhilipp Reisner break; 1033b411b363SPhilipp Reisner 1034b411b363SPhilipp Reisner epoch = next_epoch; 1035b411b363SPhilipp Reisner } while (1); 1036b411b363SPhilipp Reisner 1037b411b363SPhilipp Reisner spin_unlock(&mdev->epoch_lock); 1038b411b363SPhilipp Reisner 1039b411b363SPhilipp Reisner return rv; 1040b411b363SPhilipp Reisner } 1041b411b363SPhilipp Reisner 1042b411b363SPhilipp Reisner /** 1043b411b363SPhilipp Reisner * drbd_bump_write_ordering() - Fall back to an other write ordering method 1044b411b363SPhilipp Reisner * @mdev: DRBD device. 1045b411b363SPhilipp Reisner * @wo: Write ordering method to try. 1046b411b363SPhilipp Reisner */ 1047b411b363SPhilipp Reisner void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local) 1048b411b363SPhilipp Reisner { 1049b411b363SPhilipp Reisner enum write_ordering_e pwo; 1050b411b363SPhilipp Reisner static char *write_ordering_str[] = { 1051b411b363SPhilipp Reisner [WO_none] = "none", 1052b411b363SPhilipp Reisner [WO_drain_io] = "drain", 1053b411b363SPhilipp Reisner [WO_bdev_flush] = "flush", 1054b411b363SPhilipp Reisner }; 1055b411b363SPhilipp Reisner 1056b411b363SPhilipp Reisner pwo = mdev->write_ordering; 1057b411b363SPhilipp Reisner wo = min(pwo, wo); 1058b411b363SPhilipp Reisner if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush) 1059b411b363SPhilipp Reisner wo = WO_drain_io; 1060b411b363SPhilipp Reisner if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain) 1061b411b363SPhilipp Reisner wo = WO_none; 1062b411b363SPhilipp Reisner mdev->write_ordering = wo; 10632451fc3bSPhilipp Reisner if (pwo != mdev->write_ordering || wo == WO_bdev_flush) 1064b411b363SPhilipp Reisner dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]); 1065b411b363SPhilipp Reisner } 1066b411b363SPhilipp Reisner 1067b411b363SPhilipp Reisner /** 106845bb912bSLars Ellenberg * drbd_submit_ee() 106945bb912bSLars Ellenberg * @mdev: DRBD device. 107045bb912bSLars Ellenberg * @e: epoch entry 107145bb912bSLars Ellenberg * @rw: flag field, see bio->bi_rw 107210f6d992SLars Ellenberg * 107310f6d992SLars Ellenberg * May spread the pages to multiple bios, 107410f6d992SLars Ellenberg * depending on bio_add_page restrictions. 107510f6d992SLars Ellenberg * 107610f6d992SLars Ellenberg * Returns 0 if all bios have been submitted, 107710f6d992SLars Ellenberg * -ENOMEM if we could not allocate enough bios, 107810f6d992SLars Ellenberg * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a 107910f6d992SLars Ellenberg * single page to an empty bio (which should never happen and likely indicates 108010f6d992SLars Ellenberg * that the lower level IO stack is in some way broken). This has been observed 108110f6d992SLars Ellenberg * on certain Xen deployments. 108245bb912bSLars Ellenberg */ 108345bb912bSLars Ellenberg /* TODO allocate from our own bio_set. */ 108445bb912bSLars Ellenberg int drbd_submit_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, 108545bb912bSLars Ellenberg const unsigned rw, const int fault_type) 108645bb912bSLars Ellenberg { 108745bb912bSLars Ellenberg struct bio *bios = NULL; 108845bb912bSLars Ellenberg struct bio *bio; 108945bb912bSLars Ellenberg struct page *page = e->pages; 109045bb912bSLars Ellenberg sector_t sector = e->sector; 109145bb912bSLars Ellenberg unsigned ds = e->size; 109245bb912bSLars Ellenberg unsigned n_bios = 0; 109345bb912bSLars Ellenberg unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT; 109410f6d992SLars Ellenberg int err = -ENOMEM; 109545bb912bSLars Ellenberg 109645bb912bSLars Ellenberg /* In most cases, we will only need one bio. But in case the lower 109745bb912bSLars Ellenberg * level restrictions happen to be different at this offset on this 109845bb912bSLars Ellenberg * side than those of the sending peer, we may need to submit the 109945bb912bSLars Ellenberg * request in more than one bio. */ 110045bb912bSLars Ellenberg next_bio: 110145bb912bSLars Ellenberg bio = bio_alloc(GFP_NOIO, nr_pages); 110245bb912bSLars Ellenberg if (!bio) { 110345bb912bSLars Ellenberg dev_err(DEV, "submit_ee: Allocation of a bio failed\n"); 110445bb912bSLars Ellenberg goto fail; 110545bb912bSLars Ellenberg } 110645bb912bSLars Ellenberg /* > e->sector, unless this is the first bio */ 110745bb912bSLars Ellenberg bio->bi_sector = sector; 110845bb912bSLars Ellenberg bio->bi_bdev = mdev->ldev->backing_bdev; 110945bb912bSLars Ellenberg bio->bi_rw = rw; 111045bb912bSLars Ellenberg bio->bi_private = e; 111145bb912bSLars Ellenberg bio->bi_end_io = drbd_endio_sec; 111245bb912bSLars Ellenberg 111345bb912bSLars Ellenberg bio->bi_next = bios; 111445bb912bSLars Ellenberg bios = bio; 111545bb912bSLars Ellenberg ++n_bios; 111645bb912bSLars Ellenberg 111745bb912bSLars Ellenberg page_chain_for_each(page) { 111845bb912bSLars Ellenberg unsigned len = min_t(unsigned, ds, PAGE_SIZE); 111945bb912bSLars Ellenberg if (!bio_add_page(bio, page, len, 0)) { 112010f6d992SLars Ellenberg /* A single page must always be possible! 112110f6d992SLars Ellenberg * But in case it fails anyways, 112210f6d992SLars Ellenberg * we deal with it, and complain (below). */ 112310f6d992SLars Ellenberg if (bio->bi_vcnt == 0) { 112410f6d992SLars Ellenberg dev_err(DEV, 112510f6d992SLars Ellenberg "bio_add_page failed for len=%u, " 112610f6d992SLars Ellenberg "bi_vcnt=0 (bi_sector=%llu)\n", 112710f6d992SLars Ellenberg len, (unsigned long long)bio->bi_sector); 112810f6d992SLars Ellenberg err = -ENOSPC; 112910f6d992SLars Ellenberg goto fail; 113010f6d992SLars Ellenberg } 113145bb912bSLars Ellenberg goto next_bio; 113245bb912bSLars Ellenberg } 113345bb912bSLars Ellenberg ds -= len; 113445bb912bSLars Ellenberg sector += len >> 9; 113545bb912bSLars Ellenberg --nr_pages; 113645bb912bSLars Ellenberg } 113745bb912bSLars Ellenberg D_ASSERT(page == NULL); 113845bb912bSLars Ellenberg D_ASSERT(ds == 0); 113945bb912bSLars Ellenberg 114045bb912bSLars Ellenberg atomic_set(&e->pending_bios, n_bios); 114145bb912bSLars Ellenberg do { 114245bb912bSLars Ellenberg bio = bios; 114345bb912bSLars Ellenberg bios = bios->bi_next; 114445bb912bSLars Ellenberg bio->bi_next = NULL; 114545bb912bSLars Ellenberg 114645bb912bSLars Ellenberg drbd_generic_make_request(mdev, fault_type, bio); 114745bb912bSLars Ellenberg } while (bios); 114845bb912bSLars Ellenberg return 0; 114945bb912bSLars Ellenberg 115045bb912bSLars Ellenberg fail: 115145bb912bSLars Ellenberg while (bios) { 115245bb912bSLars Ellenberg bio = bios; 115345bb912bSLars Ellenberg bios = bios->bi_next; 115445bb912bSLars Ellenberg bio_put(bio); 115545bb912bSLars Ellenberg } 115610f6d992SLars Ellenberg return err; 115745bb912bSLars Ellenberg } 115845bb912bSLars Ellenberg 115902918be2SPhilipp Reisner static int receive_Barrier(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 1160b411b363SPhilipp Reisner { 11612451fc3bSPhilipp Reisner int rv; 116202918be2SPhilipp Reisner struct p_barrier *p = &mdev->data.rbuf.barrier; 1163b411b363SPhilipp Reisner struct drbd_epoch *epoch; 1164b411b363SPhilipp Reisner 1165b411b363SPhilipp Reisner inc_unacked(mdev); 1166b411b363SPhilipp Reisner 1167b411b363SPhilipp Reisner mdev->current_epoch->barrier_nr = p->barrier; 1168b411b363SPhilipp Reisner rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR); 1169b411b363SPhilipp Reisner 1170b411b363SPhilipp Reisner /* P_BARRIER_ACK may imply that the corresponding extent is dropped from 1171b411b363SPhilipp Reisner * the activity log, which means it would not be resynced in case the 1172b411b363SPhilipp Reisner * R_PRIMARY crashes now. 1173b411b363SPhilipp Reisner * Therefore we must send the barrier_ack after the barrier request was 1174b411b363SPhilipp Reisner * completed. */ 1175b411b363SPhilipp Reisner switch (mdev->write_ordering) { 1176b411b363SPhilipp Reisner case WO_none: 1177b411b363SPhilipp Reisner if (rv == FE_RECYCLED) 117881e84650SAndreas Gruenbacher return true; 1179b411b363SPhilipp Reisner 1180b411b363SPhilipp Reisner /* receiver context, in the writeout path of the other node. 1181b411b363SPhilipp Reisner * avoid potential distributed deadlock */ 1182b411b363SPhilipp Reisner epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); 11832451fc3bSPhilipp Reisner if (epoch) 11842451fc3bSPhilipp Reisner break; 11852451fc3bSPhilipp Reisner else 1186b411b363SPhilipp Reisner dev_warn(DEV, "Allocation of an epoch failed, slowing down\n"); 11872451fc3bSPhilipp Reisner /* Fall through */ 11882451fc3bSPhilipp Reisner 11892451fc3bSPhilipp Reisner case WO_bdev_flush: 11902451fc3bSPhilipp Reisner case WO_drain_io: 1191b411b363SPhilipp Reisner drbd_wait_ee_list_empty(mdev, &mdev->active_ee); 11922451fc3bSPhilipp Reisner drbd_flush(mdev); 11932451fc3bSPhilipp Reisner 11942451fc3bSPhilipp Reisner if (atomic_read(&mdev->current_epoch->epoch_size)) { 11952451fc3bSPhilipp Reisner epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); 11962451fc3bSPhilipp Reisner if (epoch) 11972451fc3bSPhilipp Reisner break; 1198b411b363SPhilipp Reisner } 1199b411b363SPhilipp Reisner 12002451fc3bSPhilipp Reisner epoch = mdev->current_epoch; 12012451fc3bSPhilipp Reisner wait_event(mdev->ee_wait, atomic_read(&epoch->epoch_size) == 0); 12022451fc3bSPhilipp Reisner 12032451fc3bSPhilipp Reisner D_ASSERT(atomic_read(&epoch->active) == 0); 12042451fc3bSPhilipp Reisner D_ASSERT(epoch->flags == 0); 1205b411b363SPhilipp Reisner 120681e84650SAndreas Gruenbacher return true; 12072451fc3bSPhilipp Reisner default: 12082451fc3bSPhilipp Reisner dev_err(DEV, "Strangeness in mdev->write_ordering %d\n", mdev->write_ordering); 120981e84650SAndreas Gruenbacher return false; 1210b411b363SPhilipp Reisner } 1211b411b363SPhilipp Reisner 1212b411b363SPhilipp Reisner epoch->flags = 0; 1213b411b363SPhilipp Reisner atomic_set(&epoch->epoch_size, 0); 1214b411b363SPhilipp Reisner atomic_set(&epoch->active, 0); 1215b411b363SPhilipp Reisner 1216b411b363SPhilipp Reisner spin_lock(&mdev->epoch_lock); 1217b411b363SPhilipp Reisner if (atomic_read(&mdev->current_epoch->epoch_size)) { 1218b411b363SPhilipp Reisner list_add(&epoch->list, &mdev->current_epoch->list); 1219b411b363SPhilipp Reisner mdev->current_epoch = epoch; 1220b411b363SPhilipp Reisner mdev->epochs++; 1221b411b363SPhilipp Reisner } else { 1222b411b363SPhilipp Reisner /* The current_epoch got recycled while we allocated this one... */ 1223b411b363SPhilipp Reisner kfree(epoch); 1224b411b363SPhilipp Reisner } 1225b411b363SPhilipp Reisner spin_unlock(&mdev->epoch_lock); 1226b411b363SPhilipp Reisner 122781e84650SAndreas Gruenbacher return true; 1228b411b363SPhilipp Reisner } 1229b411b363SPhilipp Reisner 1230b411b363SPhilipp Reisner /* used from receive_RSDataReply (recv_resync_read) 1231b411b363SPhilipp Reisner * and from receive_Data */ 1232b411b363SPhilipp Reisner static struct drbd_epoch_entry * 1233b411b363SPhilipp Reisner read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, int data_size) __must_hold(local) 1234b411b363SPhilipp Reisner { 12356666032aSLars Ellenberg const sector_t capacity = drbd_get_capacity(mdev->this_bdev); 1236b411b363SPhilipp Reisner struct drbd_epoch_entry *e; 1237b411b363SPhilipp Reisner struct page *page; 123845bb912bSLars Ellenberg int dgs, ds, rr; 1239b411b363SPhilipp Reisner void *dig_in = mdev->int_dig_in; 1240b411b363SPhilipp Reisner void *dig_vv = mdev->int_dig_vv; 12416b4388acSPhilipp Reisner unsigned long *data; 1242b411b363SPhilipp Reisner 1243b411b363SPhilipp Reisner dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ? 1244b411b363SPhilipp Reisner crypto_hash_digestsize(mdev->integrity_r_tfm) : 0; 1245b411b363SPhilipp Reisner 1246b411b363SPhilipp Reisner if (dgs) { 1247b411b363SPhilipp Reisner rr = drbd_recv(mdev, dig_in, dgs); 1248b411b363SPhilipp Reisner if (rr != dgs) { 12490ddc5549SLars Ellenberg if (!signal_pending(current)) 12500ddc5549SLars Ellenberg dev_warn(DEV, 12510ddc5549SLars Ellenberg "short read receiving data digest: read %d expected %d\n", 1252b411b363SPhilipp Reisner rr, dgs); 1253b411b363SPhilipp Reisner return NULL; 1254b411b363SPhilipp Reisner } 1255b411b363SPhilipp Reisner } 1256b411b363SPhilipp Reisner 1257b411b363SPhilipp Reisner data_size -= dgs; 1258b411b363SPhilipp Reisner 1259d07c9c10SPhilipp Reisner ERR_IF(data_size == 0) return NULL; 1260b411b363SPhilipp Reisner ERR_IF(data_size & 0x1ff) return NULL; 12611816a2b4SLars Ellenberg ERR_IF(data_size > DRBD_MAX_BIO_SIZE) return NULL; 1262b411b363SPhilipp Reisner 12636666032aSLars Ellenberg /* even though we trust out peer, 12646666032aSLars Ellenberg * we sometimes have to double check. */ 12656666032aSLars Ellenberg if (sector + (data_size>>9) > capacity) { 1266fdda6544SLars Ellenberg dev_err(DEV, "request from peer beyond end of local disk: " 1267fdda6544SLars Ellenberg "capacity: %llus < sector: %llus + size: %u\n", 12686666032aSLars Ellenberg (unsigned long long)capacity, 12696666032aSLars Ellenberg (unsigned long long)sector, data_size); 12706666032aSLars Ellenberg return NULL; 12716666032aSLars Ellenberg } 12726666032aSLars Ellenberg 1273b411b363SPhilipp Reisner /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD 1274b411b363SPhilipp Reisner * "criss-cross" setup, that might cause write-out on some other DRBD, 1275b411b363SPhilipp Reisner * which in turn might block on the other node at this very place. */ 1276b411b363SPhilipp Reisner e = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO); 1277b411b363SPhilipp Reisner if (!e) 1278b411b363SPhilipp Reisner return NULL; 127945bb912bSLars Ellenberg 1280b411b363SPhilipp Reisner ds = data_size; 128145bb912bSLars Ellenberg page = e->pages; 128245bb912bSLars Ellenberg page_chain_for_each(page) { 128345bb912bSLars Ellenberg unsigned len = min_t(int, ds, PAGE_SIZE); 12846b4388acSPhilipp Reisner data = kmap(page); 128545bb912bSLars Ellenberg rr = drbd_recv(mdev, data, len); 12860cf9d27eSAndreas Gruenbacher if (drbd_insert_fault(mdev, DRBD_FAULT_RECEIVE)) { 12876b4388acSPhilipp Reisner dev_err(DEV, "Fault injection: Corrupting data on receive\n"); 12886b4388acSPhilipp Reisner data[0] = data[0] ^ (unsigned long)-1; 12896b4388acSPhilipp Reisner } 1290b411b363SPhilipp Reisner kunmap(page); 129145bb912bSLars Ellenberg if (rr != len) { 1292b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 12930ddc5549SLars Ellenberg if (!signal_pending(current)) 1294b411b363SPhilipp Reisner dev_warn(DEV, "short read receiving data: read %d expected %d\n", 129545bb912bSLars Ellenberg rr, len); 1296b411b363SPhilipp Reisner return NULL; 1297b411b363SPhilipp Reisner } 1298b411b363SPhilipp Reisner ds -= rr; 1299b411b363SPhilipp Reisner } 1300b411b363SPhilipp Reisner 1301b411b363SPhilipp Reisner if (dgs) { 130245bb912bSLars Ellenberg drbd_csum_ee(mdev, mdev->integrity_r_tfm, e, dig_vv); 1303b411b363SPhilipp Reisner if (memcmp(dig_in, dig_vv, dgs)) { 1304470be44aSLars Ellenberg dev_err(DEV, "Digest integrity check FAILED: %llus +%u\n", 1305470be44aSLars Ellenberg (unsigned long long)sector, data_size); 1306b411b363SPhilipp Reisner drbd_bcast_ee(mdev, "digest failed", 1307b411b363SPhilipp Reisner dgs, dig_in, dig_vv, e); 1308b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 1309b411b363SPhilipp Reisner return NULL; 1310b411b363SPhilipp Reisner } 1311b411b363SPhilipp Reisner } 1312b411b363SPhilipp Reisner mdev->recv_cnt += data_size>>9; 1313b411b363SPhilipp Reisner return e; 1314b411b363SPhilipp Reisner } 1315b411b363SPhilipp Reisner 1316b411b363SPhilipp Reisner /* drbd_drain_block() just takes a data block 1317b411b363SPhilipp Reisner * out of the socket input buffer, and discards it. 1318b411b363SPhilipp Reisner */ 1319b411b363SPhilipp Reisner static int drbd_drain_block(struct drbd_conf *mdev, int data_size) 1320b411b363SPhilipp Reisner { 1321b411b363SPhilipp Reisner struct page *page; 1322b411b363SPhilipp Reisner int rr, rv = 1; 1323b411b363SPhilipp Reisner void *data; 1324b411b363SPhilipp Reisner 1325c3470cdeSLars Ellenberg if (!data_size) 132681e84650SAndreas Gruenbacher return true; 1327c3470cdeSLars Ellenberg 132845bb912bSLars Ellenberg page = drbd_pp_alloc(mdev, 1, 1); 1329b411b363SPhilipp Reisner 1330b411b363SPhilipp Reisner data = kmap(page); 1331b411b363SPhilipp Reisner while (data_size) { 1332b411b363SPhilipp Reisner rr = drbd_recv(mdev, data, min_t(int, data_size, PAGE_SIZE)); 1333b411b363SPhilipp Reisner if (rr != min_t(int, data_size, PAGE_SIZE)) { 1334b411b363SPhilipp Reisner rv = 0; 13350ddc5549SLars Ellenberg if (!signal_pending(current)) 13360ddc5549SLars Ellenberg dev_warn(DEV, 13370ddc5549SLars Ellenberg "short read receiving data: read %d expected %d\n", 1338b411b363SPhilipp Reisner rr, min_t(int, data_size, PAGE_SIZE)); 1339b411b363SPhilipp Reisner break; 1340b411b363SPhilipp Reisner } 1341b411b363SPhilipp Reisner data_size -= rr; 1342b411b363SPhilipp Reisner } 1343b411b363SPhilipp Reisner kunmap(page); 1344435f0740SLars Ellenberg drbd_pp_free(mdev, page, 0); 1345b411b363SPhilipp Reisner return rv; 1346b411b363SPhilipp Reisner } 1347b411b363SPhilipp Reisner 1348b411b363SPhilipp Reisner static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req, 1349b411b363SPhilipp Reisner sector_t sector, int data_size) 1350b411b363SPhilipp Reisner { 1351b411b363SPhilipp Reisner struct bio_vec *bvec; 1352b411b363SPhilipp Reisner struct bio *bio; 1353b411b363SPhilipp Reisner int dgs, rr, i, expect; 1354b411b363SPhilipp Reisner void *dig_in = mdev->int_dig_in; 1355b411b363SPhilipp Reisner void *dig_vv = mdev->int_dig_vv; 1356b411b363SPhilipp Reisner 1357b411b363SPhilipp Reisner dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ? 1358b411b363SPhilipp Reisner crypto_hash_digestsize(mdev->integrity_r_tfm) : 0; 1359b411b363SPhilipp Reisner 1360b411b363SPhilipp Reisner if (dgs) { 1361b411b363SPhilipp Reisner rr = drbd_recv(mdev, dig_in, dgs); 1362b411b363SPhilipp Reisner if (rr != dgs) { 13630ddc5549SLars Ellenberg if (!signal_pending(current)) 13640ddc5549SLars Ellenberg dev_warn(DEV, 13650ddc5549SLars Ellenberg "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) { 13870ddc5549SLars Ellenberg if (!signal_pending(current)) 1388b411b363SPhilipp Reisner dev_warn(DEV, "short read receiving data reply: " 1389b411b363SPhilipp Reisner "read %d expected %d\n", 1390b411b363SPhilipp Reisner rr, expect); 1391b411b363SPhilipp Reisner return 0; 1392b411b363SPhilipp Reisner } 1393b411b363SPhilipp Reisner data_size -= rr; 1394b411b363SPhilipp Reisner } 1395b411b363SPhilipp Reisner 1396b411b363SPhilipp Reisner if (dgs) { 139745bb912bSLars Ellenberg drbd_csum_bio(mdev, mdev->integrity_r_tfm, bio, dig_vv); 1398b411b363SPhilipp Reisner if (memcmp(dig_in, dig_vv, dgs)) { 1399b411b363SPhilipp Reisner dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n"); 1400b411b363SPhilipp Reisner return 0; 1401b411b363SPhilipp Reisner } 1402b411b363SPhilipp Reisner } 1403b411b363SPhilipp Reisner 1404b411b363SPhilipp Reisner D_ASSERT(data_size == 0); 1405b411b363SPhilipp Reisner return 1; 1406b411b363SPhilipp Reisner } 1407b411b363SPhilipp Reisner 1408b411b363SPhilipp Reisner /* e_end_resync_block() is called via 1409b411b363SPhilipp Reisner * drbd_process_done_ee() by asender only */ 1410b411b363SPhilipp Reisner static int e_end_resync_block(struct drbd_conf *mdev, struct drbd_work *w, int unused) 1411b411b363SPhilipp Reisner { 1412b411b363SPhilipp Reisner struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; 1413b411b363SPhilipp Reisner sector_t sector = e->sector; 1414b411b363SPhilipp Reisner int ok; 1415b411b363SPhilipp Reisner 141624c4830cSBart Van Assche D_ASSERT(hlist_unhashed(&e->collision)); 1417b411b363SPhilipp Reisner 141845bb912bSLars Ellenberg if (likely((e->flags & EE_WAS_ERROR) == 0)) { 1419b411b363SPhilipp Reisner drbd_set_in_sync(mdev, sector, e->size); 1420b411b363SPhilipp Reisner ok = drbd_send_ack(mdev, P_RS_WRITE_ACK, e); 1421b411b363SPhilipp Reisner } else { 1422b411b363SPhilipp Reisner /* Record failure to sync */ 1423b411b363SPhilipp Reisner drbd_rs_failed_io(mdev, sector, e->size); 1424b411b363SPhilipp Reisner 1425b411b363SPhilipp Reisner ok = drbd_send_ack(mdev, P_NEG_ACK, e); 1426b411b363SPhilipp Reisner } 1427b411b363SPhilipp Reisner dec_unacked(mdev); 1428b411b363SPhilipp Reisner 1429b411b363SPhilipp Reisner return ok; 1430b411b363SPhilipp Reisner } 1431b411b363SPhilipp Reisner 1432b411b363SPhilipp Reisner static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local) 1433b411b363SPhilipp Reisner { 1434b411b363SPhilipp Reisner struct drbd_epoch_entry *e; 1435b411b363SPhilipp Reisner 1436b411b363SPhilipp Reisner e = read_in_block(mdev, ID_SYNCER, sector, data_size); 143745bb912bSLars Ellenberg if (!e) 143845bb912bSLars Ellenberg goto fail; 1439b411b363SPhilipp Reisner 1440b411b363SPhilipp Reisner dec_rs_pending(mdev); 1441b411b363SPhilipp Reisner 1442b411b363SPhilipp Reisner inc_unacked(mdev); 1443b411b363SPhilipp Reisner /* corresponding dec_unacked() in e_end_resync_block() 1444b411b363SPhilipp Reisner * respective _drbd_clear_done_ee */ 1445b411b363SPhilipp Reisner 144645bb912bSLars Ellenberg e->w.cb = e_end_resync_block; 144745bb912bSLars Ellenberg 1448b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 1449b411b363SPhilipp Reisner list_add(&e->w.list, &mdev->sync_ee); 1450b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1451b411b363SPhilipp Reisner 14520f0601f4SLars Ellenberg atomic_add(data_size >> 9, &mdev->rs_sect_ev); 145345bb912bSLars Ellenberg if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_RS_WR) == 0) 145481e84650SAndreas Gruenbacher return true; 145545bb912bSLars Ellenberg 145610f6d992SLars Ellenberg /* don't care for the reason here */ 145710f6d992SLars Ellenberg dev_err(DEV, "submit failed, triggering re-connect\n"); 145822cc37a9SLars Ellenberg spin_lock_irq(&mdev->req_lock); 145922cc37a9SLars Ellenberg list_del(&e->w.list); 146022cc37a9SLars Ellenberg spin_unlock_irq(&mdev->req_lock); 146122cc37a9SLars Ellenberg 146245bb912bSLars Ellenberg drbd_free_ee(mdev, e); 146345bb912bSLars Ellenberg fail: 146445bb912bSLars Ellenberg put_ldev(mdev); 146581e84650SAndreas Gruenbacher return false; 1466b411b363SPhilipp Reisner } 1467b411b363SPhilipp Reisner 146802918be2SPhilipp Reisner static int receive_DataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 1469b411b363SPhilipp Reisner { 1470b411b363SPhilipp Reisner struct drbd_request *req; 1471b411b363SPhilipp Reisner sector_t sector; 1472b411b363SPhilipp Reisner int ok; 147302918be2SPhilipp Reisner struct p_data *p = &mdev->data.rbuf.data; 1474b411b363SPhilipp Reisner 1475b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 1476b411b363SPhilipp Reisner 1477b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 1478b411b363SPhilipp Reisner req = _ar_id_to_req(mdev, p->block_id, sector); 1479b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1480b411b363SPhilipp Reisner if (unlikely(!req)) { 1481b411b363SPhilipp Reisner dev_err(DEV, "Got a corrupt block_id/sector pair(1).\n"); 148281e84650SAndreas Gruenbacher return false; 1483b411b363SPhilipp Reisner } 1484b411b363SPhilipp Reisner 148524c4830cSBart Van Assche /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid 1486b411b363SPhilipp Reisner * special casing it there for the various failure cases. 1487b411b363SPhilipp Reisner * still no race with drbd_fail_pending_reads */ 1488b411b363SPhilipp Reisner ok = recv_dless_read(mdev, req, sector, data_size); 1489b411b363SPhilipp Reisner 1490b411b363SPhilipp Reisner if (ok) 1491b411b363SPhilipp Reisner req_mod(req, data_received); 1492b411b363SPhilipp Reisner /* else: nothing. handled from drbd_disconnect... 1493b411b363SPhilipp Reisner * I don't think we may complete this just yet 1494b411b363SPhilipp Reisner * in case we are "on-disconnect: freeze" */ 1495b411b363SPhilipp Reisner 1496b411b363SPhilipp Reisner return ok; 1497b411b363SPhilipp Reisner } 1498b411b363SPhilipp Reisner 149902918be2SPhilipp Reisner static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 1500b411b363SPhilipp Reisner { 1501b411b363SPhilipp Reisner sector_t sector; 1502b411b363SPhilipp Reisner int ok; 150302918be2SPhilipp Reisner struct p_data *p = &mdev->data.rbuf.data; 1504b411b363SPhilipp Reisner 1505b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 1506b411b363SPhilipp Reisner D_ASSERT(p->block_id == ID_SYNCER); 1507b411b363SPhilipp Reisner 1508b411b363SPhilipp Reisner if (get_ldev(mdev)) { 1509b411b363SPhilipp Reisner /* data is submitted to disk within recv_resync_read. 1510b411b363SPhilipp Reisner * corresponding put_ldev done below on error, 1511b411b363SPhilipp Reisner * or in drbd_endio_write_sec. */ 1512b411b363SPhilipp Reisner ok = recv_resync_read(mdev, sector, data_size); 1513b411b363SPhilipp Reisner } else { 1514b411b363SPhilipp Reisner if (__ratelimit(&drbd_ratelimit_state)) 1515b411b363SPhilipp Reisner dev_err(DEV, "Can not write resync data to local disk.\n"); 1516b411b363SPhilipp Reisner 1517b411b363SPhilipp Reisner ok = drbd_drain_block(mdev, data_size); 1518b411b363SPhilipp Reisner 15192b2bf214SLars Ellenberg drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); 1520b411b363SPhilipp Reisner } 1521b411b363SPhilipp Reisner 1522778f271dSPhilipp Reisner atomic_add(data_size >> 9, &mdev->rs_sect_in); 1523778f271dSPhilipp Reisner 1524b411b363SPhilipp Reisner return ok; 1525b411b363SPhilipp Reisner } 1526b411b363SPhilipp Reisner 1527b411b363SPhilipp Reisner /* e_end_block() is called via drbd_process_done_ee(). 1528b411b363SPhilipp Reisner * this means this function only runs in the asender thread 1529b411b363SPhilipp Reisner */ 1530b411b363SPhilipp Reisner static int e_end_block(struct drbd_conf *mdev, struct drbd_work *w, int cancel) 1531b411b363SPhilipp Reisner { 1532b411b363SPhilipp Reisner struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; 1533b411b363SPhilipp Reisner sector_t sector = e->sector; 1534b411b363SPhilipp Reisner int ok = 1, pcmd; 1535b411b363SPhilipp Reisner 1536b411b363SPhilipp Reisner if (mdev->net_conf->wire_protocol == DRBD_PROT_C) { 153745bb912bSLars Ellenberg if (likely((e->flags & EE_WAS_ERROR) == 0)) { 1538b411b363SPhilipp Reisner pcmd = (mdev->state.conn >= C_SYNC_SOURCE && 1539b411b363SPhilipp Reisner mdev->state.conn <= C_PAUSED_SYNC_T && 1540b411b363SPhilipp Reisner e->flags & EE_MAY_SET_IN_SYNC) ? 1541b411b363SPhilipp Reisner P_RS_WRITE_ACK : P_WRITE_ACK; 1542b411b363SPhilipp Reisner ok &= drbd_send_ack(mdev, pcmd, e); 1543b411b363SPhilipp Reisner if (pcmd == P_RS_WRITE_ACK) 1544b411b363SPhilipp Reisner drbd_set_in_sync(mdev, sector, e->size); 1545b411b363SPhilipp Reisner } else { 1546b411b363SPhilipp Reisner ok = drbd_send_ack(mdev, P_NEG_ACK, e); 1547b411b363SPhilipp Reisner /* we expect it to be marked out of sync anyways... 1548b411b363SPhilipp Reisner * maybe assert this? */ 1549b411b363SPhilipp Reisner } 1550b411b363SPhilipp Reisner dec_unacked(mdev); 1551b411b363SPhilipp Reisner } 1552b411b363SPhilipp Reisner /* we delete from the conflict detection hash _after_ we sent out the 1553b411b363SPhilipp Reisner * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ 1554b411b363SPhilipp Reisner if (mdev->net_conf->two_primaries) { 1555b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 155624c4830cSBart Van Assche D_ASSERT(!hlist_unhashed(&e->collision)); 155724c4830cSBart Van Assche hlist_del_init(&e->collision); 1558b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1559b411b363SPhilipp Reisner } else { 156024c4830cSBart Van Assche D_ASSERT(hlist_unhashed(&e->collision)); 1561b411b363SPhilipp Reisner } 1562b411b363SPhilipp Reisner 1563b411b363SPhilipp Reisner drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); 1564b411b363SPhilipp Reisner 1565b411b363SPhilipp Reisner return ok; 1566b411b363SPhilipp Reisner } 1567b411b363SPhilipp Reisner 1568b411b363SPhilipp Reisner static int e_send_discard_ack(struct drbd_conf *mdev, struct drbd_work *w, int unused) 1569b411b363SPhilipp Reisner { 1570b411b363SPhilipp Reisner struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w; 1571b411b363SPhilipp Reisner int ok = 1; 1572b411b363SPhilipp Reisner 1573b411b363SPhilipp Reisner D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 1574b411b363SPhilipp Reisner ok = drbd_send_ack(mdev, P_DISCARD_ACK, e); 1575b411b363SPhilipp Reisner 1576b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 157724c4830cSBart Van Assche D_ASSERT(!hlist_unhashed(&e->collision)); 157824c4830cSBart Van Assche hlist_del_init(&e->collision); 1579b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1580b411b363SPhilipp Reisner 1581b411b363SPhilipp Reisner dec_unacked(mdev); 1582b411b363SPhilipp Reisner 1583b411b363SPhilipp Reisner return ok; 1584b411b363SPhilipp Reisner } 1585b411b363SPhilipp Reisner 1586b411b363SPhilipp Reisner /* Called from receive_Data. 1587b411b363SPhilipp Reisner * Synchronize packets on sock with packets on msock. 1588b411b363SPhilipp Reisner * 1589b411b363SPhilipp Reisner * This is here so even when a P_DATA packet traveling via sock overtook an Ack 1590b411b363SPhilipp Reisner * packet traveling on msock, they are still processed in the order they have 1591b411b363SPhilipp Reisner * been sent. 1592b411b363SPhilipp Reisner * 1593b411b363SPhilipp Reisner * Note: we don't care for Ack packets overtaking P_DATA packets. 1594b411b363SPhilipp Reisner * 1595b411b363SPhilipp Reisner * In case packet_seq is larger than mdev->peer_seq number, there are 1596b411b363SPhilipp Reisner * outstanding packets on the msock. We wait for them to arrive. 1597b411b363SPhilipp Reisner * In case we are the logically next packet, we update mdev->peer_seq 1598b411b363SPhilipp Reisner * ourselves. Correctly handles 32bit wrap around. 1599b411b363SPhilipp Reisner * 1600b411b363SPhilipp Reisner * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, 1601b411b363SPhilipp Reisner * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds 1602b411b363SPhilipp Reisner * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have 1603b411b363SPhilipp Reisner * 1<<9 == 512 seconds aka ages for the 32bit wrap around... 1604b411b363SPhilipp Reisner * 1605b411b363SPhilipp Reisner * returns 0 if we may process the packet, 1606b411b363SPhilipp Reisner * -ERESTARTSYS if we were interrupted (by disconnect signal). */ 1607b411b363SPhilipp Reisner static int drbd_wait_peer_seq(struct drbd_conf *mdev, const u32 packet_seq) 1608b411b363SPhilipp Reisner { 1609b411b363SPhilipp Reisner DEFINE_WAIT(wait); 1610b411b363SPhilipp Reisner unsigned int p_seq; 1611b411b363SPhilipp Reisner long timeout; 1612b411b363SPhilipp Reisner int ret = 0; 1613b411b363SPhilipp Reisner spin_lock(&mdev->peer_seq_lock); 1614b411b363SPhilipp Reisner for (;;) { 1615b411b363SPhilipp Reisner prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE); 1616b411b363SPhilipp Reisner if (seq_le(packet_seq, mdev->peer_seq+1)) 1617b411b363SPhilipp Reisner break; 1618b411b363SPhilipp Reisner if (signal_pending(current)) { 1619b411b363SPhilipp Reisner ret = -ERESTARTSYS; 1620b411b363SPhilipp Reisner break; 1621b411b363SPhilipp Reisner } 1622b411b363SPhilipp Reisner p_seq = mdev->peer_seq; 1623b411b363SPhilipp Reisner spin_unlock(&mdev->peer_seq_lock); 1624b411b363SPhilipp Reisner timeout = schedule_timeout(30*HZ); 1625b411b363SPhilipp Reisner spin_lock(&mdev->peer_seq_lock); 1626b411b363SPhilipp Reisner if (timeout == 0 && p_seq == mdev->peer_seq) { 1627b411b363SPhilipp Reisner ret = -ETIMEDOUT; 1628b411b363SPhilipp Reisner dev_err(DEV, "ASSERT FAILED waited 30 seconds for sequence update, forcing reconnect\n"); 1629b411b363SPhilipp Reisner break; 1630b411b363SPhilipp Reisner } 1631b411b363SPhilipp Reisner } 1632b411b363SPhilipp Reisner finish_wait(&mdev->seq_wait, &wait); 1633b411b363SPhilipp Reisner if (mdev->peer_seq+1 == packet_seq) 1634b411b363SPhilipp Reisner mdev->peer_seq++; 1635b411b363SPhilipp Reisner spin_unlock(&mdev->peer_seq_lock); 1636b411b363SPhilipp Reisner return ret; 1637b411b363SPhilipp Reisner } 1638b411b363SPhilipp Reisner 1639688593c5SLars Ellenberg /* see also bio_flags_to_wire() 1640688593c5SLars Ellenberg * DRBD_REQ_*, because we need to semantically map the flags to data packet 1641688593c5SLars Ellenberg * flags and back. We may replicate to other kernel versions. */ 1642688593c5SLars Ellenberg static unsigned long wire_flags_to_bio(struct drbd_conf *mdev, u32 dpf) 164376d2e7ecSPhilipp Reisner { 164476d2e7ecSPhilipp Reisner return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | 164576d2e7ecSPhilipp Reisner (dpf & DP_FUA ? REQ_FUA : 0) | 1646688593c5SLars Ellenberg (dpf & DP_FLUSH ? REQ_FLUSH : 0) | 164776d2e7ecSPhilipp Reisner (dpf & DP_DISCARD ? REQ_DISCARD : 0); 164876d2e7ecSPhilipp Reisner } 164976d2e7ecSPhilipp Reisner 1650b411b363SPhilipp Reisner /* mirrored write */ 165102918be2SPhilipp Reisner static int receive_Data(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 1652b411b363SPhilipp Reisner { 1653b411b363SPhilipp Reisner sector_t sector; 1654b411b363SPhilipp Reisner struct drbd_epoch_entry *e; 165502918be2SPhilipp Reisner struct p_data *p = &mdev->data.rbuf.data; 1656b411b363SPhilipp Reisner int rw = WRITE; 1657b411b363SPhilipp Reisner u32 dp_flags; 1658b411b363SPhilipp Reisner 1659b411b363SPhilipp Reisner if (!get_ldev(mdev)) { 1660b411b363SPhilipp Reisner spin_lock(&mdev->peer_seq_lock); 1661b411b363SPhilipp Reisner if (mdev->peer_seq+1 == be32_to_cpu(p->seq_num)) 1662b411b363SPhilipp Reisner mdev->peer_seq++; 1663b411b363SPhilipp Reisner spin_unlock(&mdev->peer_seq_lock); 1664b411b363SPhilipp Reisner 16652b2bf214SLars Ellenberg drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); 1666b411b363SPhilipp Reisner atomic_inc(&mdev->current_epoch->epoch_size); 1667b411b363SPhilipp Reisner return drbd_drain_block(mdev, data_size); 1668b411b363SPhilipp Reisner } 1669b411b363SPhilipp Reisner 1670b411b363SPhilipp Reisner /* get_ldev(mdev) successful. 1671b411b363SPhilipp Reisner * Corresponding put_ldev done either below (on various errors), 1672b411b363SPhilipp Reisner * or in drbd_endio_write_sec, if we successfully submit the data at 1673b411b363SPhilipp Reisner * the end of this function. */ 1674b411b363SPhilipp Reisner 1675b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 1676b411b363SPhilipp Reisner e = read_in_block(mdev, p->block_id, sector, data_size); 1677b411b363SPhilipp Reisner if (!e) { 1678b411b363SPhilipp Reisner put_ldev(mdev); 167981e84650SAndreas Gruenbacher return false; 1680b411b363SPhilipp Reisner } 1681b411b363SPhilipp Reisner 1682b411b363SPhilipp Reisner e->w.cb = e_end_block; 1683b411b363SPhilipp Reisner 1684688593c5SLars Ellenberg dp_flags = be32_to_cpu(p->dp_flags); 1685688593c5SLars Ellenberg rw |= wire_flags_to_bio(mdev, dp_flags); 1686688593c5SLars Ellenberg 1687688593c5SLars Ellenberg if (dp_flags & DP_MAY_SET_IN_SYNC) 1688688593c5SLars Ellenberg e->flags |= EE_MAY_SET_IN_SYNC; 1689688593c5SLars Ellenberg 1690b411b363SPhilipp Reisner spin_lock(&mdev->epoch_lock); 1691b411b363SPhilipp Reisner e->epoch = mdev->current_epoch; 1692b411b363SPhilipp Reisner atomic_inc(&e->epoch->epoch_size); 1693b411b363SPhilipp Reisner atomic_inc(&e->epoch->active); 1694b411b363SPhilipp Reisner spin_unlock(&mdev->epoch_lock); 1695b411b363SPhilipp Reisner 1696b411b363SPhilipp Reisner /* I'm the receiver, I do hold a net_cnt reference. */ 1697b411b363SPhilipp Reisner if (!mdev->net_conf->two_primaries) { 1698b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 1699b411b363SPhilipp Reisner } else { 1700b411b363SPhilipp Reisner /* don't get the req_lock yet, 1701b411b363SPhilipp Reisner * we may sleep in drbd_wait_peer_seq */ 1702b411b363SPhilipp Reisner const int size = e->size; 1703b411b363SPhilipp Reisner const int discard = test_bit(DISCARD_CONCURRENT, &mdev->flags); 1704b411b363SPhilipp Reisner DEFINE_WAIT(wait); 1705b411b363SPhilipp Reisner struct drbd_request *i; 1706b411b363SPhilipp Reisner struct hlist_node *n; 1707b411b363SPhilipp Reisner struct hlist_head *slot; 1708b411b363SPhilipp Reisner int first; 1709b411b363SPhilipp Reisner 1710b411b363SPhilipp Reisner D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 1711b411b363SPhilipp Reisner BUG_ON(mdev->ee_hash == NULL); 1712b411b363SPhilipp Reisner BUG_ON(mdev->tl_hash == NULL); 1713b411b363SPhilipp Reisner 1714b411b363SPhilipp Reisner /* conflict detection and handling: 1715b411b363SPhilipp Reisner * 1. wait on the sequence number, 1716b411b363SPhilipp Reisner * in case this data packet overtook ACK packets. 1717b411b363SPhilipp Reisner * 2. check our hash tables for conflicting requests. 1718b411b363SPhilipp Reisner * we only need to walk the tl_hash, since an ee can not 1719b411b363SPhilipp Reisner * have a conflict with an other ee: on the submitting 1720b411b363SPhilipp Reisner * node, the corresponding req had already been conflicting, 1721b411b363SPhilipp Reisner * and a conflicting req is never sent. 1722b411b363SPhilipp Reisner * 1723b411b363SPhilipp Reisner * Note: for two_primaries, we are protocol C, 1724b411b363SPhilipp Reisner * so there cannot be any request that is DONE 1725b411b363SPhilipp Reisner * but still on the transfer log. 1726b411b363SPhilipp Reisner * 1727b411b363SPhilipp Reisner * unconditionally add to the ee_hash. 1728b411b363SPhilipp Reisner * 1729b411b363SPhilipp Reisner * if no conflicting request is found: 1730b411b363SPhilipp Reisner * submit. 1731b411b363SPhilipp Reisner * 1732b411b363SPhilipp Reisner * if any conflicting request is found 1733b411b363SPhilipp Reisner * that has not yet been acked, 1734b411b363SPhilipp Reisner * AND I have the "discard concurrent writes" flag: 1735b411b363SPhilipp Reisner * queue (via done_ee) the P_DISCARD_ACK; OUT. 1736b411b363SPhilipp Reisner * 1737b411b363SPhilipp Reisner * if any conflicting request is found: 1738b411b363SPhilipp Reisner * block the receiver, waiting on misc_wait 1739b411b363SPhilipp Reisner * until no more conflicting requests are there, 1740b411b363SPhilipp Reisner * or we get interrupted (disconnect). 1741b411b363SPhilipp Reisner * 1742b411b363SPhilipp Reisner * we do not just write after local io completion of those 1743b411b363SPhilipp Reisner * requests, but only after req is done completely, i.e. 1744b411b363SPhilipp Reisner * we wait for the P_DISCARD_ACK to arrive! 1745b411b363SPhilipp Reisner * 1746b411b363SPhilipp Reisner * then proceed normally, i.e. submit. 1747b411b363SPhilipp Reisner */ 1748b411b363SPhilipp Reisner if (drbd_wait_peer_seq(mdev, be32_to_cpu(p->seq_num))) 1749b411b363SPhilipp Reisner goto out_interrupted; 1750b411b363SPhilipp Reisner 1751b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 1752b411b363SPhilipp Reisner 175324c4830cSBart Van Assche hlist_add_head(&e->collision, ee_hash_slot(mdev, sector)); 1754b411b363SPhilipp Reisner 1755b411b363SPhilipp Reisner #define OVERLAPS overlaps(i->sector, i->size, sector, size) 1756b411b363SPhilipp Reisner slot = tl_hash_slot(mdev, sector); 1757b411b363SPhilipp Reisner first = 1; 1758b411b363SPhilipp Reisner for (;;) { 1759b411b363SPhilipp Reisner int have_unacked = 0; 1760b411b363SPhilipp Reisner int have_conflict = 0; 1761b411b363SPhilipp Reisner prepare_to_wait(&mdev->misc_wait, &wait, 1762b411b363SPhilipp Reisner TASK_INTERRUPTIBLE); 176324c4830cSBart Van Assche hlist_for_each_entry(i, n, slot, collision) { 1764b411b363SPhilipp Reisner if (OVERLAPS) { 1765b411b363SPhilipp Reisner /* only ALERT on first iteration, 1766b411b363SPhilipp Reisner * we may be woken up early... */ 1767b411b363SPhilipp Reisner if (first) 1768b411b363SPhilipp Reisner dev_alert(DEV, "%s[%u] Concurrent local write detected!" 1769b411b363SPhilipp Reisner " new: %llus +%u; pending: %llus +%u\n", 1770b411b363SPhilipp Reisner current->comm, current->pid, 1771b411b363SPhilipp Reisner (unsigned long long)sector, size, 1772b411b363SPhilipp Reisner (unsigned long long)i->sector, i->size); 1773b411b363SPhilipp Reisner if (i->rq_state & RQ_NET_PENDING) 1774b411b363SPhilipp Reisner ++have_unacked; 1775b411b363SPhilipp Reisner ++have_conflict; 1776b411b363SPhilipp Reisner } 1777b411b363SPhilipp Reisner } 1778b411b363SPhilipp Reisner #undef OVERLAPS 1779b411b363SPhilipp Reisner if (!have_conflict) 1780b411b363SPhilipp Reisner break; 1781b411b363SPhilipp Reisner 1782b411b363SPhilipp Reisner /* Discard Ack only for the _first_ iteration */ 1783b411b363SPhilipp Reisner if (first && discard && have_unacked) { 1784b411b363SPhilipp Reisner dev_alert(DEV, "Concurrent write! [DISCARD BY FLAG] sec=%llus\n", 1785b411b363SPhilipp Reisner (unsigned long long)sector); 1786b411b363SPhilipp Reisner inc_unacked(mdev); 1787b411b363SPhilipp Reisner e->w.cb = e_send_discard_ack; 1788b411b363SPhilipp Reisner list_add_tail(&e->w.list, &mdev->done_ee); 1789b411b363SPhilipp Reisner 1790b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1791b411b363SPhilipp Reisner 1792b411b363SPhilipp Reisner /* we could probably send that P_DISCARD_ACK ourselves, 1793b411b363SPhilipp Reisner * but I don't like the receiver using the msock */ 1794b411b363SPhilipp Reisner 1795b411b363SPhilipp Reisner put_ldev(mdev); 1796b411b363SPhilipp Reisner wake_asender(mdev); 1797b411b363SPhilipp Reisner finish_wait(&mdev->misc_wait, &wait); 179881e84650SAndreas Gruenbacher return true; 1799b411b363SPhilipp Reisner } 1800b411b363SPhilipp Reisner 1801b411b363SPhilipp Reisner if (signal_pending(current)) { 180224c4830cSBart Van Assche hlist_del_init(&e->collision); 1803b411b363SPhilipp Reisner 1804b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1805b411b363SPhilipp Reisner 1806b411b363SPhilipp Reisner finish_wait(&mdev->misc_wait, &wait); 1807b411b363SPhilipp Reisner goto out_interrupted; 1808b411b363SPhilipp Reisner } 1809b411b363SPhilipp Reisner 1810b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1811b411b363SPhilipp Reisner if (first) { 1812b411b363SPhilipp Reisner first = 0; 1813b411b363SPhilipp Reisner dev_alert(DEV, "Concurrent write! [W AFTERWARDS] " 1814b411b363SPhilipp Reisner "sec=%llus\n", (unsigned long long)sector); 1815b411b363SPhilipp Reisner } else if (discard) { 1816b411b363SPhilipp Reisner /* we had none on the first iteration. 1817b411b363SPhilipp Reisner * there must be none now. */ 1818b411b363SPhilipp Reisner D_ASSERT(have_unacked == 0); 1819b411b363SPhilipp Reisner } 1820b411b363SPhilipp Reisner schedule(); 1821b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 1822b411b363SPhilipp Reisner } 1823b411b363SPhilipp Reisner finish_wait(&mdev->misc_wait, &wait); 1824b411b363SPhilipp Reisner } 1825b411b363SPhilipp Reisner 1826b411b363SPhilipp Reisner list_add(&e->w.list, &mdev->active_ee); 1827b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 1828b411b363SPhilipp Reisner 1829b411b363SPhilipp Reisner switch (mdev->net_conf->wire_protocol) { 1830b411b363SPhilipp Reisner case DRBD_PROT_C: 1831b411b363SPhilipp Reisner inc_unacked(mdev); 1832b411b363SPhilipp Reisner /* corresponding dec_unacked() in e_end_block() 1833b411b363SPhilipp Reisner * respective _drbd_clear_done_ee */ 1834b411b363SPhilipp Reisner break; 1835b411b363SPhilipp Reisner case DRBD_PROT_B: 1836b411b363SPhilipp Reisner /* I really don't like it that the receiver thread 1837b411b363SPhilipp Reisner * sends on the msock, but anyways */ 1838b411b363SPhilipp Reisner drbd_send_ack(mdev, P_RECV_ACK, e); 1839b411b363SPhilipp Reisner break; 1840b411b363SPhilipp Reisner case DRBD_PROT_A: 1841b411b363SPhilipp Reisner /* nothing to do */ 1842b411b363SPhilipp Reisner break; 1843b411b363SPhilipp Reisner } 1844b411b363SPhilipp Reisner 18456719fb03SLars Ellenberg if (mdev->state.pdsk < D_INCONSISTENT) { 1846b411b363SPhilipp Reisner /* In case we have the only disk of the cluster, */ 1847b411b363SPhilipp Reisner drbd_set_out_of_sync(mdev, e->sector, e->size); 1848b411b363SPhilipp Reisner e->flags |= EE_CALL_AL_COMPLETE_IO; 18496719fb03SLars Ellenberg e->flags &= ~EE_MAY_SET_IN_SYNC; 1850b411b363SPhilipp Reisner drbd_al_begin_io(mdev, e->sector); 1851b411b363SPhilipp Reisner } 1852b411b363SPhilipp Reisner 185345bb912bSLars Ellenberg if (drbd_submit_ee(mdev, e, rw, DRBD_FAULT_DT_WR) == 0) 185481e84650SAndreas Gruenbacher return true; 1855b411b363SPhilipp Reisner 185610f6d992SLars Ellenberg /* don't care for the reason here */ 185710f6d992SLars Ellenberg dev_err(DEV, "submit failed, triggering re-connect\n"); 185822cc37a9SLars Ellenberg spin_lock_irq(&mdev->req_lock); 185922cc37a9SLars Ellenberg list_del(&e->w.list); 186024c4830cSBart Van Assche hlist_del_init(&e->collision); 186122cc37a9SLars Ellenberg spin_unlock_irq(&mdev->req_lock); 186222cc37a9SLars Ellenberg if (e->flags & EE_CALL_AL_COMPLETE_IO) 186322cc37a9SLars Ellenberg drbd_al_complete_io(mdev, e->sector); 186422cc37a9SLars Ellenberg 1865b411b363SPhilipp Reisner out_interrupted: 186610f6d992SLars Ellenberg drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + EV_CLEANUP); 1867b411b363SPhilipp Reisner put_ldev(mdev); 1868b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 186981e84650SAndreas Gruenbacher return false; 1870b411b363SPhilipp Reisner } 1871b411b363SPhilipp Reisner 18720f0601f4SLars Ellenberg /* We may throttle resync, if the lower device seems to be busy, 18730f0601f4SLars Ellenberg * and current sync rate is above c_min_rate. 18740f0601f4SLars Ellenberg * 18750f0601f4SLars Ellenberg * To decide whether or not the lower device is busy, we use a scheme similar 18760f0601f4SLars Ellenberg * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" 18770f0601f4SLars Ellenberg * (more than 64 sectors) of activity we cannot account for with our own resync 18780f0601f4SLars Ellenberg * activity, it obviously is "busy". 18790f0601f4SLars Ellenberg * 18800f0601f4SLars Ellenberg * The current sync rate used here uses only the most recent two step marks, 18810f0601f4SLars Ellenberg * to have a short time average so we can react faster. 18820f0601f4SLars Ellenberg */ 1883e3555d85SPhilipp Reisner int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector) 18840f0601f4SLars Ellenberg { 18850f0601f4SLars Ellenberg struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk; 18860f0601f4SLars Ellenberg unsigned long db, dt, dbdt; 1887e3555d85SPhilipp Reisner struct lc_element *tmp; 18880f0601f4SLars Ellenberg int curr_events; 18890f0601f4SLars Ellenberg int throttle = 0; 18900f0601f4SLars Ellenberg 18910f0601f4SLars Ellenberg /* feature disabled? */ 18920f0601f4SLars Ellenberg if (mdev->sync_conf.c_min_rate == 0) 18930f0601f4SLars Ellenberg return 0; 18940f0601f4SLars Ellenberg 1895e3555d85SPhilipp Reisner spin_lock_irq(&mdev->al_lock); 1896e3555d85SPhilipp Reisner tmp = lc_find(mdev->resync, BM_SECT_TO_EXT(sector)); 1897e3555d85SPhilipp Reisner if (tmp) { 1898e3555d85SPhilipp Reisner struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); 1899e3555d85SPhilipp Reisner if (test_bit(BME_PRIORITY, &bm_ext->flags)) { 1900e3555d85SPhilipp Reisner spin_unlock_irq(&mdev->al_lock); 1901e3555d85SPhilipp Reisner return 0; 1902e3555d85SPhilipp Reisner } 1903e3555d85SPhilipp Reisner /* Do not slow down if app IO is already waiting for this extent */ 1904e3555d85SPhilipp Reisner } 1905e3555d85SPhilipp Reisner spin_unlock_irq(&mdev->al_lock); 1906e3555d85SPhilipp Reisner 19070f0601f4SLars Ellenberg curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + 19080f0601f4SLars Ellenberg (int)part_stat_read(&disk->part0, sectors[1]) - 19090f0601f4SLars Ellenberg atomic_read(&mdev->rs_sect_ev); 1910e3555d85SPhilipp Reisner 19110f0601f4SLars Ellenberg if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) { 19120f0601f4SLars Ellenberg unsigned long rs_left; 19130f0601f4SLars Ellenberg int i; 19140f0601f4SLars Ellenberg 19150f0601f4SLars Ellenberg mdev->rs_last_events = curr_events; 19160f0601f4SLars Ellenberg 19170f0601f4SLars Ellenberg /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, 19180f0601f4SLars Ellenberg * approx. */ 19192649f080SLars Ellenberg i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; 19202649f080SLars Ellenberg 19212649f080SLars Ellenberg if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T) 19222649f080SLars Ellenberg rs_left = mdev->ov_left; 19232649f080SLars Ellenberg else 19240f0601f4SLars Ellenberg rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; 19250f0601f4SLars Ellenberg 19260f0601f4SLars Ellenberg dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ; 19270f0601f4SLars Ellenberg if (!dt) 19280f0601f4SLars Ellenberg dt++; 19290f0601f4SLars Ellenberg db = mdev->rs_mark_left[i] - rs_left; 19300f0601f4SLars Ellenberg dbdt = Bit2KB(db/dt); 19310f0601f4SLars Ellenberg 19320f0601f4SLars Ellenberg if (dbdt > mdev->sync_conf.c_min_rate) 19330f0601f4SLars Ellenberg throttle = 1; 19340f0601f4SLars Ellenberg } 19350f0601f4SLars Ellenberg return throttle; 19360f0601f4SLars Ellenberg } 19370f0601f4SLars Ellenberg 19380f0601f4SLars Ellenberg 193902918be2SPhilipp Reisner static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int digest_size) 1940b411b363SPhilipp Reisner { 1941b411b363SPhilipp Reisner sector_t sector; 1942b411b363SPhilipp Reisner const sector_t capacity = drbd_get_capacity(mdev->this_bdev); 1943b411b363SPhilipp Reisner struct drbd_epoch_entry *e; 1944b411b363SPhilipp Reisner struct digest_info *di = NULL; 1945b18b37beSPhilipp Reisner int size, verb; 1946b411b363SPhilipp Reisner unsigned int fault_type; 194702918be2SPhilipp Reisner struct p_block_req *p = &mdev->data.rbuf.block_req; 1948b411b363SPhilipp Reisner 1949b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 1950b411b363SPhilipp Reisner size = be32_to_cpu(p->blksize); 1951b411b363SPhilipp Reisner 19521816a2b4SLars Ellenberg if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_BIO_SIZE) { 1953b411b363SPhilipp Reisner dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, 1954b411b363SPhilipp Reisner (unsigned long long)sector, size); 195581e84650SAndreas Gruenbacher return false; 1956b411b363SPhilipp Reisner } 1957b411b363SPhilipp Reisner if (sector + (size>>9) > capacity) { 1958b411b363SPhilipp Reisner dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, 1959b411b363SPhilipp Reisner (unsigned long long)sector, size); 196081e84650SAndreas Gruenbacher return false; 1961b411b363SPhilipp Reisner } 1962b411b363SPhilipp Reisner 1963b411b363SPhilipp Reisner if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) { 1964b18b37beSPhilipp Reisner verb = 1; 1965b18b37beSPhilipp Reisner switch (cmd) { 1966b18b37beSPhilipp Reisner case P_DATA_REQUEST: 1967b18b37beSPhilipp Reisner drbd_send_ack_rp(mdev, P_NEG_DREPLY, p); 1968b18b37beSPhilipp Reisner break; 1969b18b37beSPhilipp Reisner case P_RS_DATA_REQUEST: 1970b18b37beSPhilipp Reisner case P_CSUM_RS_REQUEST: 1971b18b37beSPhilipp Reisner case P_OV_REQUEST: 1972b18b37beSPhilipp Reisner drbd_send_ack_rp(mdev, P_NEG_RS_DREPLY , p); 1973b18b37beSPhilipp Reisner break; 1974b18b37beSPhilipp Reisner case P_OV_REPLY: 1975b18b37beSPhilipp Reisner verb = 0; 1976b18b37beSPhilipp Reisner dec_rs_pending(mdev); 1977b18b37beSPhilipp Reisner drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, ID_IN_SYNC); 1978b18b37beSPhilipp Reisner break; 1979b18b37beSPhilipp Reisner default: 1980b18b37beSPhilipp Reisner dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", 1981b18b37beSPhilipp Reisner cmdname(cmd)); 1982b18b37beSPhilipp Reisner } 1983b18b37beSPhilipp Reisner if (verb && __ratelimit(&drbd_ratelimit_state)) 1984b411b363SPhilipp Reisner dev_err(DEV, "Can not satisfy peer's read request, " 1985b411b363SPhilipp Reisner "no local data.\n"); 1986b18b37beSPhilipp Reisner 1987a821cc4aSLars Ellenberg /* drain possibly payload */ 1988a821cc4aSLars Ellenberg return drbd_drain_block(mdev, digest_size); 1989b411b363SPhilipp Reisner } 1990b411b363SPhilipp Reisner 1991b411b363SPhilipp Reisner /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD 1992b411b363SPhilipp Reisner * "criss-cross" setup, that might cause write-out on some other DRBD, 1993b411b363SPhilipp Reisner * which in turn might block on the other node at this very place. */ 1994b411b363SPhilipp Reisner e = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO); 1995b411b363SPhilipp Reisner if (!e) { 1996b411b363SPhilipp Reisner put_ldev(mdev); 199781e84650SAndreas Gruenbacher return false; 1998b411b363SPhilipp Reisner } 1999b411b363SPhilipp Reisner 200002918be2SPhilipp Reisner switch (cmd) { 2001b411b363SPhilipp Reisner case P_DATA_REQUEST: 2002b411b363SPhilipp Reisner e->w.cb = w_e_end_data_req; 2003b411b363SPhilipp Reisner fault_type = DRBD_FAULT_DT_RD; 200480a40e43SLars Ellenberg /* application IO, don't drbd_rs_begin_io */ 200580a40e43SLars Ellenberg goto submit; 200680a40e43SLars Ellenberg 2007b411b363SPhilipp Reisner case P_RS_DATA_REQUEST: 2008b411b363SPhilipp Reisner e->w.cb = w_e_end_rsdata_req; 2009b411b363SPhilipp Reisner fault_type = DRBD_FAULT_RS_RD; 20105f9915bbSLars Ellenberg /* used in the sector offset progress display */ 20115f9915bbSLars Ellenberg mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); 2012b411b363SPhilipp Reisner break; 2013b411b363SPhilipp Reisner 2014b411b363SPhilipp Reisner case P_OV_REPLY: 2015b411b363SPhilipp Reisner case P_CSUM_RS_REQUEST: 2016b411b363SPhilipp Reisner fault_type = DRBD_FAULT_RS_RD; 2017b411b363SPhilipp Reisner di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO); 2018b411b363SPhilipp Reisner if (!di) 2019b411b363SPhilipp Reisner goto out_free_e; 2020b411b363SPhilipp Reisner 2021b411b363SPhilipp Reisner di->digest_size = digest_size; 2022b411b363SPhilipp Reisner di->digest = (((char *)di)+sizeof(struct digest_info)); 2023b411b363SPhilipp Reisner 2024c36c3cedSLars Ellenberg e->digest = di; 2025c36c3cedSLars Ellenberg e->flags |= EE_HAS_DIGEST; 2026c36c3cedSLars Ellenberg 2027b411b363SPhilipp Reisner if (drbd_recv(mdev, di->digest, digest_size) != digest_size) 2028b411b363SPhilipp Reisner goto out_free_e; 2029b411b363SPhilipp Reisner 203002918be2SPhilipp Reisner if (cmd == P_CSUM_RS_REQUEST) { 2031b411b363SPhilipp Reisner D_ASSERT(mdev->agreed_pro_version >= 89); 2032b411b363SPhilipp Reisner e->w.cb = w_e_end_csum_rs_req; 20335f9915bbSLars Ellenberg /* used in the sector offset progress display */ 20345f9915bbSLars Ellenberg mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); 203502918be2SPhilipp Reisner } else if (cmd == P_OV_REPLY) { 20362649f080SLars Ellenberg /* track progress, we may need to throttle */ 20372649f080SLars Ellenberg atomic_add(size >> 9, &mdev->rs_sect_in); 2038b411b363SPhilipp Reisner e->w.cb = w_e_end_ov_reply; 2039b411b363SPhilipp Reisner dec_rs_pending(mdev); 20400f0601f4SLars Ellenberg /* drbd_rs_begin_io done when we sent this request, 20410f0601f4SLars Ellenberg * but accounting still needs to be done. */ 20420f0601f4SLars Ellenberg goto submit_for_resync; 2043b411b363SPhilipp Reisner } 2044b411b363SPhilipp Reisner break; 2045b411b363SPhilipp Reisner 2046b411b363SPhilipp Reisner case P_OV_REQUEST: 2047b411b363SPhilipp Reisner if (mdev->ov_start_sector == ~(sector_t)0 && 2048b411b363SPhilipp Reisner mdev->agreed_pro_version >= 90) { 2049de228bbaSLars Ellenberg unsigned long now = jiffies; 2050de228bbaSLars Ellenberg int i; 2051b411b363SPhilipp Reisner mdev->ov_start_sector = sector; 2052b411b363SPhilipp Reisner mdev->ov_position = sector; 205330b743a2SLars Ellenberg mdev->ov_left = drbd_bm_bits(mdev) - BM_SECT_TO_BIT(sector); 205430b743a2SLars Ellenberg mdev->rs_total = mdev->ov_left; 2055de228bbaSLars Ellenberg for (i = 0; i < DRBD_SYNC_MARKS; i++) { 2056de228bbaSLars Ellenberg mdev->rs_mark_left[i] = mdev->ov_left; 2057de228bbaSLars Ellenberg mdev->rs_mark_time[i] = now; 2058de228bbaSLars Ellenberg } 2059b411b363SPhilipp Reisner dev_info(DEV, "Online Verify start sector: %llu\n", 2060b411b363SPhilipp Reisner (unsigned long long)sector); 2061b411b363SPhilipp Reisner } 2062b411b363SPhilipp Reisner e->w.cb = w_e_end_ov_req; 2063b411b363SPhilipp Reisner fault_type = DRBD_FAULT_RS_RD; 2064b411b363SPhilipp Reisner break; 2065b411b363SPhilipp Reisner 2066b411b363SPhilipp Reisner default: 2067b411b363SPhilipp Reisner dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", 206802918be2SPhilipp Reisner cmdname(cmd)); 2069b411b363SPhilipp Reisner fault_type = DRBD_FAULT_MAX; 207080a40e43SLars Ellenberg goto out_free_e; 2071b411b363SPhilipp Reisner } 2072b411b363SPhilipp Reisner 20730f0601f4SLars Ellenberg /* Throttle, drbd_rs_begin_io and submit should become asynchronous 20740f0601f4SLars Ellenberg * wrt the receiver, but it is not as straightforward as it may seem. 20750f0601f4SLars Ellenberg * Various places in the resync start and stop logic assume resync 20760f0601f4SLars Ellenberg * requests are processed in order, requeuing this on the worker thread 20770f0601f4SLars Ellenberg * introduces a bunch of new code for synchronization between threads. 20780f0601f4SLars Ellenberg * 20790f0601f4SLars Ellenberg * Unlimited throttling before drbd_rs_begin_io may stall the resync 20800f0601f4SLars Ellenberg * "forever", throttling after drbd_rs_begin_io will lock that extent 20810f0601f4SLars Ellenberg * for application writes for the same time. For now, just throttle 20820f0601f4SLars Ellenberg * here, where the rest of the code expects the receiver to sleep for 20830f0601f4SLars Ellenberg * a while, anyways. 20840f0601f4SLars Ellenberg */ 2085b411b363SPhilipp Reisner 20860f0601f4SLars Ellenberg /* Throttle before drbd_rs_begin_io, as that locks out application IO; 20870f0601f4SLars Ellenberg * this defers syncer requests for some time, before letting at least 20880f0601f4SLars Ellenberg * on request through. The resync controller on the receiving side 20890f0601f4SLars Ellenberg * will adapt to the incoming rate accordingly. 20900f0601f4SLars Ellenberg * 20910f0601f4SLars Ellenberg * We cannot throttle here if remote is Primary/SyncTarget: 20920f0601f4SLars Ellenberg * we would also throttle its application reads. 20930f0601f4SLars Ellenberg * In that case, throttling is done on the SyncTarget only. 20940f0601f4SLars Ellenberg */ 2095e3555d85SPhilipp Reisner if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev, sector)) 2096e3555d85SPhilipp Reisner schedule_timeout_uninterruptible(HZ/10); 2097e3555d85SPhilipp Reisner if (drbd_rs_begin_io(mdev, sector)) 209880a40e43SLars Ellenberg goto out_free_e; 2099b411b363SPhilipp Reisner 21000f0601f4SLars Ellenberg submit_for_resync: 21010f0601f4SLars Ellenberg atomic_add(size >> 9, &mdev->rs_sect_ev); 21020f0601f4SLars Ellenberg 210380a40e43SLars Ellenberg submit: 2104b411b363SPhilipp Reisner inc_unacked(mdev); 210580a40e43SLars Ellenberg spin_lock_irq(&mdev->req_lock); 210680a40e43SLars Ellenberg list_add_tail(&e->w.list, &mdev->read_ee); 210780a40e43SLars Ellenberg spin_unlock_irq(&mdev->req_lock); 2108b411b363SPhilipp Reisner 210945bb912bSLars Ellenberg if (drbd_submit_ee(mdev, e, READ, fault_type) == 0) 211081e84650SAndreas Gruenbacher return true; 2111b411b363SPhilipp Reisner 211210f6d992SLars Ellenberg /* don't care for the reason here */ 211310f6d992SLars Ellenberg dev_err(DEV, "submit failed, triggering re-connect\n"); 211422cc37a9SLars Ellenberg spin_lock_irq(&mdev->req_lock); 211522cc37a9SLars Ellenberg list_del(&e->w.list); 211622cc37a9SLars Ellenberg spin_unlock_irq(&mdev->req_lock); 211722cc37a9SLars Ellenberg /* no drbd_rs_complete_io(), we are dropping the connection anyways */ 211822cc37a9SLars Ellenberg 2119b411b363SPhilipp Reisner out_free_e: 2120b411b363SPhilipp Reisner put_ldev(mdev); 2121b411b363SPhilipp Reisner drbd_free_ee(mdev, e); 212281e84650SAndreas Gruenbacher return false; 2123b411b363SPhilipp Reisner } 2124b411b363SPhilipp Reisner 2125b411b363SPhilipp Reisner static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local) 2126b411b363SPhilipp Reisner { 2127b411b363SPhilipp Reisner int self, peer, rv = -100; 2128b411b363SPhilipp Reisner unsigned long ch_self, ch_peer; 2129b411b363SPhilipp Reisner 2130b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_BITMAP] & 1; 2131b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_BITMAP] & 1; 2132b411b363SPhilipp Reisner 2133b411b363SPhilipp Reisner ch_peer = mdev->p_uuid[UI_SIZE]; 2134b411b363SPhilipp Reisner ch_self = mdev->comm_bm_set; 2135b411b363SPhilipp Reisner 2136b411b363SPhilipp Reisner switch (mdev->net_conf->after_sb_0p) { 2137b411b363SPhilipp Reisner case ASB_CONSENSUS: 2138b411b363SPhilipp Reisner case ASB_DISCARD_SECONDARY: 2139b411b363SPhilipp Reisner case ASB_CALL_HELPER: 2140b411b363SPhilipp Reisner dev_err(DEV, "Configuration error.\n"); 2141b411b363SPhilipp Reisner break; 2142b411b363SPhilipp Reisner case ASB_DISCONNECT: 2143b411b363SPhilipp Reisner break; 2144b411b363SPhilipp Reisner case ASB_DISCARD_YOUNGER_PRI: 2145b411b363SPhilipp Reisner if (self == 0 && peer == 1) { 2146b411b363SPhilipp Reisner rv = -1; 2147b411b363SPhilipp Reisner break; 2148b411b363SPhilipp Reisner } 2149b411b363SPhilipp Reisner if (self == 1 && peer == 0) { 2150b411b363SPhilipp Reisner rv = 1; 2151b411b363SPhilipp Reisner break; 2152b411b363SPhilipp Reisner } 2153b411b363SPhilipp Reisner /* Else fall through to one of the other strategies... */ 2154b411b363SPhilipp Reisner case ASB_DISCARD_OLDER_PRI: 2155b411b363SPhilipp Reisner if (self == 0 && peer == 1) { 2156b411b363SPhilipp Reisner rv = 1; 2157b411b363SPhilipp Reisner break; 2158b411b363SPhilipp Reisner } 2159b411b363SPhilipp Reisner if (self == 1 && peer == 0) { 2160b411b363SPhilipp Reisner rv = -1; 2161b411b363SPhilipp Reisner break; 2162b411b363SPhilipp Reisner } 2163b411b363SPhilipp Reisner /* Else fall through to one of the other strategies... */ 2164ad19bf6eSLars Ellenberg dev_warn(DEV, "Discard younger/older primary did not find a decision\n" 2165b411b363SPhilipp Reisner "Using discard-least-changes instead\n"); 2166b411b363SPhilipp Reisner case ASB_DISCARD_ZERO_CHG: 2167b411b363SPhilipp Reisner if (ch_peer == 0 && ch_self == 0) { 2168b411b363SPhilipp Reisner rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) 2169b411b363SPhilipp Reisner ? -1 : 1; 2170b411b363SPhilipp Reisner break; 2171b411b363SPhilipp Reisner } else { 2172b411b363SPhilipp Reisner if (ch_peer == 0) { rv = 1; break; } 2173b411b363SPhilipp Reisner if (ch_self == 0) { rv = -1; break; } 2174b411b363SPhilipp Reisner } 2175b411b363SPhilipp Reisner if (mdev->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG) 2176b411b363SPhilipp Reisner break; 2177b411b363SPhilipp Reisner case ASB_DISCARD_LEAST_CHG: 2178b411b363SPhilipp Reisner if (ch_self < ch_peer) 2179b411b363SPhilipp Reisner rv = -1; 2180b411b363SPhilipp Reisner else if (ch_self > ch_peer) 2181b411b363SPhilipp Reisner rv = 1; 2182b411b363SPhilipp Reisner else /* ( ch_self == ch_peer ) */ 2183b411b363SPhilipp Reisner /* Well, then use something else. */ 2184b411b363SPhilipp Reisner rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) 2185b411b363SPhilipp Reisner ? -1 : 1; 2186b411b363SPhilipp Reisner break; 2187b411b363SPhilipp Reisner case ASB_DISCARD_LOCAL: 2188b411b363SPhilipp Reisner rv = -1; 2189b411b363SPhilipp Reisner break; 2190b411b363SPhilipp Reisner case ASB_DISCARD_REMOTE: 2191b411b363SPhilipp Reisner rv = 1; 2192b411b363SPhilipp Reisner } 2193b411b363SPhilipp Reisner 2194b411b363SPhilipp Reisner return rv; 2195b411b363SPhilipp Reisner } 2196b411b363SPhilipp Reisner 2197b411b363SPhilipp Reisner static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local) 2198b411b363SPhilipp Reisner { 21996184ea21SAndreas Gruenbacher int hg, rv = -100; 2200b411b363SPhilipp Reisner 2201b411b363SPhilipp Reisner switch (mdev->net_conf->after_sb_1p) { 2202b411b363SPhilipp Reisner case ASB_DISCARD_YOUNGER_PRI: 2203b411b363SPhilipp Reisner case ASB_DISCARD_OLDER_PRI: 2204b411b363SPhilipp Reisner case ASB_DISCARD_LEAST_CHG: 2205b411b363SPhilipp Reisner case ASB_DISCARD_LOCAL: 2206b411b363SPhilipp Reisner case ASB_DISCARD_REMOTE: 2207b411b363SPhilipp Reisner dev_err(DEV, "Configuration error.\n"); 2208b411b363SPhilipp Reisner break; 2209b411b363SPhilipp Reisner case ASB_DISCONNECT: 2210b411b363SPhilipp Reisner break; 2211b411b363SPhilipp Reisner case ASB_CONSENSUS: 2212b411b363SPhilipp Reisner hg = drbd_asb_recover_0p(mdev); 2213b411b363SPhilipp Reisner if (hg == -1 && mdev->state.role == R_SECONDARY) 2214b411b363SPhilipp Reisner rv = hg; 2215b411b363SPhilipp Reisner if (hg == 1 && mdev->state.role == R_PRIMARY) 2216b411b363SPhilipp Reisner rv = hg; 2217b411b363SPhilipp Reisner break; 2218b411b363SPhilipp Reisner case ASB_VIOLENTLY: 2219b411b363SPhilipp Reisner rv = drbd_asb_recover_0p(mdev); 2220b411b363SPhilipp Reisner break; 2221b411b363SPhilipp Reisner case ASB_DISCARD_SECONDARY: 2222b411b363SPhilipp Reisner return mdev->state.role == R_PRIMARY ? 1 : -1; 2223b411b363SPhilipp Reisner case ASB_CALL_HELPER: 2224b411b363SPhilipp Reisner hg = drbd_asb_recover_0p(mdev); 2225b411b363SPhilipp Reisner if (hg == -1 && mdev->state.role == R_PRIMARY) { 2226bb437946SAndreas Gruenbacher enum drbd_state_rv rv2; 2227bb437946SAndreas Gruenbacher 2228bb437946SAndreas Gruenbacher drbd_set_role(mdev, R_SECONDARY, 0); 2229b411b363SPhilipp Reisner /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, 2230b411b363SPhilipp Reisner * we might be here in C_WF_REPORT_PARAMS which is transient. 2231b411b363SPhilipp Reisner * we do not need to wait for the after state change work either. */ 2232bb437946SAndreas Gruenbacher rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); 2233bb437946SAndreas Gruenbacher if (rv2 != SS_SUCCESS) { 2234b411b363SPhilipp Reisner drbd_khelper(mdev, "pri-lost-after-sb"); 2235b411b363SPhilipp Reisner } else { 2236b411b363SPhilipp Reisner dev_warn(DEV, "Successfully gave up primary role.\n"); 2237b411b363SPhilipp Reisner rv = hg; 2238b411b363SPhilipp Reisner } 2239b411b363SPhilipp Reisner } else 2240b411b363SPhilipp Reisner rv = hg; 2241b411b363SPhilipp Reisner } 2242b411b363SPhilipp Reisner 2243b411b363SPhilipp Reisner return rv; 2244b411b363SPhilipp Reisner } 2245b411b363SPhilipp Reisner 2246b411b363SPhilipp Reisner static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local) 2247b411b363SPhilipp Reisner { 22486184ea21SAndreas Gruenbacher int hg, rv = -100; 2249b411b363SPhilipp Reisner 2250b411b363SPhilipp Reisner switch (mdev->net_conf->after_sb_2p) { 2251b411b363SPhilipp Reisner case ASB_DISCARD_YOUNGER_PRI: 2252b411b363SPhilipp Reisner case ASB_DISCARD_OLDER_PRI: 2253b411b363SPhilipp Reisner case ASB_DISCARD_LEAST_CHG: 2254b411b363SPhilipp Reisner case ASB_DISCARD_LOCAL: 2255b411b363SPhilipp Reisner case ASB_DISCARD_REMOTE: 2256b411b363SPhilipp Reisner case ASB_CONSENSUS: 2257b411b363SPhilipp Reisner case ASB_DISCARD_SECONDARY: 2258b411b363SPhilipp Reisner dev_err(DEV, "Configuration error.\n"); 2259b411b363SPhilipp Reisner break; 2260b411b363SPhilipp Reisner case ASB_VIOLENTLY: 2261b411b363SPhilipp Reisner rv = drbd_asb_recover_0p(mdev); 2262b411b363SPhilipp Reisner break; 2263b411b363SPhilipp Reisner case ASB_DISCONNECT: 2264b411b363SPhilipp Reisner break; 2265b411b363SPhilipp Reisner case ASB_CALL_HELPER: 2266b411b363SPhilipp Reisner hg = drbd_asb_recover_0p(mdev); 2267b411b363SPhilipp Reisner if (hg == -1) { 2268bb437946SAndreas Gruenbacher enum drbd_state_rv rv2; 2269bb437946SAndreas Gruenbacher 2270b411b363SPhilipp Reisner /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, 2271b411b363SPhilipp Reisner * we might be here in C_WF_REPORT_PARAMS which is transient. 2272b411b363SPhilipp Reisner * we do not need to wait for the after state change work either. */ 2273bb437946SAndreas Gruenbacher rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); 2274bb437946SAndreas Gruenbacher if (rv2 != SS_SUCCESS) { 2275b411b363SPhilipp Reisner drbd_khelper(mdev, "pri-lost-after-sb"); 2276b411b363SPhilipp Reisner } else { 2277b411b363SPhilipp Reisner dev_warn(DEV, "Successfully gave up primary role.\n"); 2278b411b363SPhilipp Reisner rv = hg; 2279b411b363SPhilipp Reisner } 2280b411b363SPhilipp Reisner } else 2281b411b363SPhilipp Reisner rv = hg; 2282b411b363SPhilipp Reisner } 2283b411b363SPhilipp Reisner 2284b411b363SPhilipp Reisner return rv; 2285b411b363SPhilipp Reisner } 2286b411b363SPhilipp Reisner 2287b411b363SPhilipp Reisner static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid, 2288b411b363SPhilipp Reisner u64 bits, u64 flags) 2289b411b363SPhilipp Reisner { 2290b411b363SPhilipp Reisner if (!uuid) { 2291b411b363SPhilipp Reisner dev_info(DEV, "%s uuid info vanished while I was looking!\n", text); 2292b411b363SPhilipp Reisner return; 2293b411b363SPhilipp Reisner } 2294b411b363SPhilipp Reisner dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", 2295b411b363SPhilipp Reisner text, 2296b411b363SPhilipp Reisner (unsigned long long)uuid[UI_CURRENT], 2297b411b363SPhilipp Reisner (unsigned long long)uuid[UI_BITMAP], 2298b411b363SPhilipp Reisner (unsigned long long)uuid[UI_HISTORY_START], 2299b411b363SPhilipp Reisner (unsigned long long)uuid[UI_HISTORY_END], 2300b411b363SPhilipp Reisner (unsigned long long)bits, 2301b411b363SPhilipp Reisner (unsigned long long)flags); 2302b411b363SPhilipp Reisner } 2303b411b363SPhilipp Reisner 2304b411b363SPhilipp Reisner /* 2305b411b363SPhilipp Reisner 100 after split brain try auto recover 2306b411b363SPhilipp Reisner 2 C_SYNC_SOURCE set BitMap 2307b411b363SPhilipp Reisner 1 C_SYNC_SOURCE use BitMap 2308b411b363SPhilipp Reisner 0 no Sync 2309b411b363SPhilipp Reisner -1 C_SYNC_TARGET use BitMap 2310b411b363SPhilipp Reisner -2 C_SYNC_TARGET set BitMap 2311b411b363SPhilipp Reisner -100 after split brain, disconnect 2312b411b363SPhilipp Reisner -1000 unrelated data 23134a23f264SPhilipp Reisner -1091 requires proto 91 23144a23f264SPhilipp Reisner -1096 requires proto 96 2315b411b363SPhilipp Reisner */ 2316b411b363SPhilipp Reisner static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local) 2317b411b363SPhilipp Reisner { 2318b411b363SPhilipp Reisner u64 self, peer; 2319b411b363SPhilipp Reisner int i, j; 2320b411b363SPhilipp Reisner 2321b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); 2322b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); 2323b411b363SPhilipp Reisner 2324b411b363SPhilipp Reisner *rule_nr = 10; 2325b411b363SPhilipp Reisner if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) 2326b411b363SPhilipp Reisner return 0; 2327b411b363SPhilipp Reisner 2328b411b363SPhilipp Reisner *rule_nr = 20; 2329b411b363SPhilipp Reisner if ((self == UUID_JUST_CREATED || self == (u64)0) && 2330b411b363SPhilipp Reisner peer != UUID_JUST_CREATED) 2331b411b363SPhilipp Reisner return -2; 2332b411b363SPhilipp Reisner 2333b411b363SPhilipp Reisner *rule_nr = 30; 2334b411b363SPhilipp Reisner if (self != UUID_JUST_CREATED && 2335b411b363SPhilipp Reisner (peer == UUID_JUST_CREATED || peer == (u64)0)) 2336b411b363SPhilipp Reisner return 2; 2337b411b363SPhilipp Reisner 2338b411b363SPhilipp Reisner if (self == peer) { 2339b411b363SPhilipp Reisner int rct, dc; /* roles at crash time */ 2340b411b363SPhilipp Reisner 2341b411b363SPhilipp Reisner if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) { 2342b411b363SPhilipp Reisner 2343b411b363SPhilipp Reisner if (mdev->agreed_pro_version < 91) 23444a23f264SPhilipp Reisner return -1091; 2345b411b363SPhilipp Reisner 2346b411b363SPhilipp Reisner if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) && 2347b411b363SPhilipp Reisner (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { 2348b411b363SPhilipp Reisner dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n"); 2349b411b363SPhilipp Reisner drbd_uuid_set_bm(mdev, 0UL); 2350b411b363SPhilipp Reisner 2351b411b363SPhilipp Reisner drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, 2352b411b363SPhilipp Reisner mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); 2353b411b363SPhilipp Reisner *rule_nr = 34; 2354b411b363SPhilipp Reisner } else { 2355b411b363SPhilipp Reisner dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n"); 2356b411b363SPhilipp Reisner *rule_nr = 36; 2357b411b363SPhilipp Reisner } 2358b411b363SPhilipp Reisner 2359b411b363SPhilipp Reisner return 1; 2360b411b363SPhilipp Reisner } 2361b411b363SPhilipp Reisner 2362b411b363SPhilipp Reisner if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) { 2363b411b363SPhilipp Reisner 2364b411b363SPhilipp Reisner if (mdev->agreed_pro_version < 91) 23654a23f264SPhilipp Reisner return -1091; 2366b411b363SPhilipp Reisner 2367b411b363SPhilipp Reisner if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) && 2368b411b363SPhilipp Reisner (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) { 2369b411b363SPhilipp Reisner dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); 2370b411b363SPhilipp Reisner 2371b411b363SPhilipp Reisner mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START]; 2372b411b363SPhilipp Reisner mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP]; 2373b411b363SPhilipp Reisner mdev->p_uuid[UI_BITMAP] = 0UL; 2374b411b363SPhilipp Reisner 2375b411b363SPhilipp Reisner drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); 2376b411b363SPhilipp Reisner *rule_nr = 35; 2377b411b363SPhilipp Reisner } else { 2378b411b363SPhilipp Reisner dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n"); 2379b411b363SPhilipp Reisner *rule_nr = 37; 2380b411b363SPhilipp Reisner } 2381b411b363SPhilipp Reisner 2382b411b363SPhilipp Reisner return -1; 2383b411b363SPhilipp Reisner } 2384b411b363SPhilipp Reisner 2385b411b363SPhilipp Reisner /* Common power [off|failure] */ 2386b411b363SPhilipp Reisner rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) + 2387b411b363SPhilipp Reisner (mdev->p_uuid[UI_FLAGS] & 2); 2388b411b363SPhilipp Reisner /* lowest bit is set when we were primary, 2389b411b363SPhilipp Reisner * next bit (weight 2) is set when peer was primary */ 2390b411b363SPhilipp Reisner *rule_nr = 40; 2391b411b363SPhilipp Reisner 2392b411b363SPhilipp Reisner switch (rct) { 2393b411b363SPhilipp Reisner case 0: /* !self_pri && !peer_pri */ return 0; 2394b411b363SPhilipp Reisner case 1: /* self_pri && !peer_pri */ return 1; 2395b411b363SPhilipp Reisner case 2: /* !self_pri && peer_pri */ return -1; 2396b411b363SPhilipp Reisner case 3: /* self_pri && peer_pri */ 2397b411b363SPhilipp Reisner dc = test_bit(DISCARD_CONCURRENT, &mdev->flags); 2398b411b363SPhilipp Reisner return dc ? -1 : 1; 2399b411b363SPhilipp Reisner } 2400b411b363SPhilipp Reisner } 2401b411b363SPhilipp Reisner 2402b411b363SPhilipp Reisner *rule_nr = 50; 2403b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); 2404b411b363SPhilipp Reisner if (self == peer) 2405b411b363SPhilipp Reisner return -1; 2406b411b363SPhilipp Reisner 2407b411b363SPhilipp Reisner *rule_nr = 51; 2408b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); 2409b411b363SPhilipp Reisner if (self == peer) { 24104a23f264SPhilipp Reisner if (mdev->agreed_pro_version < 96 ? 24114a23f264SPhilipp Reisner (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == 24124a23f264SPhilipp Reisner (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : 24134a23f264SPhilipp Reisner peer + UUID_NEW_BM_OFFSET == (mdev->p_uuid[UI_BITMAP] & ~((u64)1))) { 2414b411b363SPhilipp Reisner /* The last P_SYNC_UUID did not get though. Undo the last start of 2415b411b363SPhilipp Reisner resync as sync source modifications of the peer's UUIDs. */ 2416b411b363SPhilipp Reisner 2417b411b363SPhilipp Reisner if (mdev->agreed_pro_version < 91) 24184a23f264SPhilipp Reisner return -1091; 2419b411b363SPhilipp Reisner 2420b411b363SPhilipp Reisner mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START]; 2421b411b363SPhilipp Reisner mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1]; 24224a23f264SPhilipp Reisner 24234a23f264SPhilipp Reisner dev_info(DEV, "Did not got last syncUUID packet, corrected:\n"); 24244a23f264SPhilipp Reisner drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); 24254a23f264SPhilipp Reisner 2426b411b363SPhilipp Reisner return -1; 2427b411b363SPhilipp Reisner } 2428b411b363SPhilipp Reisner } 2429b411b363SPhilipp Reisner 2430b411b363SPhilipp Reisner *rule_nr = 60; 2431b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); 2432b411b363SPhilipp Reisner for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 2433b411b363SPhilipp Reisner peer = mdev->p_uuid[i] & ~((u64)1); 2434b411b363SPhilipp Reisner if (self == peer) 2435b411b363SPhilipp Reisner return -2; 2436b411b363SPhilipp Reisner } 2437b411b363SPhilipp Reisner 2438b411b363SPhilipp Reisner *rule_nr = 70; 2439b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); 2440b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); 2441b411b363SPhilipp Reisner if (self == peer) 2442b411b363SPhilipp Reisner return 1; 2443b411b363SPhilipp Reisner 2444b411b363SPhilipp Reisner *rule_nr = 71; 2445b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); 2446b411b363SPhilipp Reisner if (self == peer) { 24474a23f264SPhilipp Reisner if (mdev->agreed_pro_version < 96 ? 24484a23f264SPhilipp Reisner (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == 24494a23f264SPhilipp Reisner (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) : 24504a23f264SPhilipp Reisner self + UUID_NEW_BM_OFFSET == (mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { 2451b411b363SPhilipp Reisner /* The last P_SYNC_UUID did not get though. Undo the last start of 2452b411b363SPhilipp Reisner resync as sync source modifications of our UUIDs. */ 2453b411b363SPhilipp Reisner 2454b411b363SPhilipp Reisner if (mdev->agreed_pro_version < 91) 24554a23f264SPhilipp Reisner return -1091; 2456b411b363SPhilipp Reisner 2457b411b363SPhilipp Reisner _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]); 2458b411b363SPhilipp Reisner _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]); 2459b411b363SPhilipp Reisner 24604a23f264SPhilipp Reisner dev_info(DEV, "Last syncUUID did not get through, corrected:\n"); 2461b411b363SPhilipp Reisner drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, 2462b411b363SPhilipp Reisner mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); 2463b411b363SPhilipp Reisner 2464b411b363SPhilipp Reisner return 1; 2465b411b363SPhilipp Reisner } 2466b411b363SPhilipp Reisner } 2467b411b363SPhilipp Reisner 2468b411b363SPhilipp Reisner 2469b411b363SPhilipp Reisner *rule_nr = 80; 2470d8c2a36bSPhilipp Reisner peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); 2471b411b363SPhilipp Reisner for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 2472b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[i] & ~((u64)1); 2473b411b363SPhilipp Reisner if (self == peer) 2474b411b363SPhilipp Reisner return 2; 2475b411b363SPhilipp Reisner } 2476b411b363SPhilipp Reisner 2477b411b363SPhilipp Reisner *rule_nr = 90; 2478b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); 2479b411b363SPhilipp Reisner peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); 2480b411b363SPhilipp Reisner if (self == peer && self != ((u64)0)) 2481b411b363SPhilipp Reisner return 100; 2482b411b363SPhilipp Reisner 2483b411b363SPhilipp Reisner *rule_nr = 100; 2484b411b363SPhilipp Reisner for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 2485b411b363SPhilipp Reisner self = mdev->ldev->md.uuid[i] & ~((u64)1); 2486b411b363SPhilipp Reisner for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { 2487b411b363SPhilipp Reisner peer = mdev->p_uuid[j] & ~((u64)1); 2488b411b363SPhilipp Reisner if (self == peer) 2489b411b363SPhilipp Reisner return -100; 2490b411b363SPhilipp Reisner } 2491b411b363SPhilipp Reisner } 2492b411b363SPhilipp Reisner 2493b411b363SPhilipp Reisner return -1000; 2494b411b363SPhilipp Reisner } 2495b411b363SPhilipp Reisner 2496b411b363SPhilipp Reisner /* drbd_sync_handshake() returns the new conn state on success, or 2497b411b363SPhilipp Reisner CONN_MASK (-1) on failure. 2498b411b363SPhilipp Reisner */ 2499b411b363SPhilipp Reisner static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role, 2500b411b363SPhilipp Reisner enum drbd_disk_state peer_disk) __must_hold(local) 2501b411b363SPhilipp Reisner { 2502b411b363SPhilipp Reisner int hg, rule_nr; 2503b411b363SPhilipp Reisner enum drbd_conns rv = C_MASK; 2504b411b363SPhilipp Reisner enum drbd_disk_state mydisk; 2505b411b363SPhilipp Reisner 2506b411b363SPhilipp Reisner mydisk = mdev->state.disk; 2507b411b363SPhilipp Reisner if (mydisk == D_NEGOTIATING) 2508b411b363SPhilipp Reisner mydisk = mdev->new_state_tmp.disk; 2509b411b363SPhilipp Reisner 2510b411b363SPhilipp Reisner dev_info(DEV, "drbd_sync_handshake:\n"); 2511b411b363SPhilipp Reisner drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0); 2512b411b363SPhilipp Reisner drbd_uuid_dump(mdev, "peer", mdev->p_uuid, 2513b411b363SPhilipp Reisner mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); 2514b411b363SPhilipp Reisner 2515b411b363SPhilipp Reisner hg = drbd_uuid_compare(mdev, &rule_nr); 2516b411b363SPhilipp Reisner 2517b411b363SPhilipp Reisner dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr); 2518b411b363SPhilipp Reisner 2519b411b363SPhilipp Reisner if (hg == -1000) { 2520b411b363SPhilipp Reisner dev_alert(DEV, "Unrelated data, aborting!\n"); 2521b411b363SPhilipp Reisner return C_MASK; 2522b411b363SPhilipp Reisner } 25234a23f264SPhilipp Reisner if (hg < -1000) { 25244a23f264SPhilipp Reisner dev_alert(DEV, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); 2525b411b363SPhilipp Reisner return C_MASK; 2526b411b363SPhilipp Reisner } 2527b411b363SPhilipp Reisner 2528b411b363SPhilipp Reisner if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || 2529b411b363SPhilipp Reisner (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { 2530b411b363SPhilipp Reisner int f = (hg == -100) || abs(hg) == 2; 2531b411b363SPhilipp Reisner hg = mydisk > D_INCONSISTENT ? 1 : -1; 2532b411b363SPhilipp Reisner if (f) 2533b411b363SPhilipp Reisner hg = hg*2; 2534b411b363SPhilipp Reisner dev_info(DEV, "Becoming sync %s due to disk states.\n", 2535b411b363SPhilipp Reisner hg > 0 ? "source" : "target"); 2536b411b363SPhilipp Reisner } 2537b411b363SPhilipp Reisner 25383a11a487SAdam Gandelman if (abs(hg) == 100) 25393a11a487SAdam Gandelman drbd_khelper(mdev, "initial-split-brain"); 25403a11a487SAdam Gandelman 2541b411b363SPhilipp Reisner if (hg == 100 || (hg == -100 && mdev->net_conf->always_asbp)) { 2542b411b363SPhilipp Reisner int pcount = (mdev->state.role == R_PRIMARY) 2543b411b363SPhilipp Reisner + (peer_role == R_PRIMARY); 2544b411b363SPhilipp Reisner int forced = (hg == -100); 2545b411b363SPhilipp Reisner 2546b411b363SPhilipp Reisner switch (pcount) { 2547b411b363SPhilipp Reisner case 0: 2548b411b363SPhilipp Reisner hg = drbd_asb_recover_0p(mdev); 2549b411b363SPhilipp Reisner break; 2550b411b363SPhilipp Reisner case 1: 2551b411b363SPhilipp Reisner hg = drbd_asb_recover_1p(mdev); 2552b411b363SPhilipp Reisner break; 2553b411b363SPhilipp Reisner case 2: 2554b411b363SPhilipp Reisner hg = drbd_asb_recover_2p(mdev); 2555b411b363SPhilipp Reisner break; 2556b411b363SPhilipp Reisner } 2557b411b363SPhilipp Reisner if (abs(hg) < 100) { 2558b411b363SPhilipp Reisner dev_warn(DEV, "Split-Brain detected, %d primaries, " 2559b411b363SPhilipp Reisner "automatically solved. Sync from %s node\n", 2560b411b363SPhilipp Reisner pcount, (hg < 0) ? "peer" : "this"); 2561b411b363SPhilipp Reisner if (forced) { 2562b411b363SPhilipp Reisner dev_warn(DEV, "Doing a full sync, since" 2563b411b363SPhilipp Reisner " UUIDs where ambiguous.\n"); 2564b411b363SPhilipp Reisner hg = hg*2; 2565b411b363SPhilipp Reisner } 2566b411b363SPhilipp Reisner } 2567b411b363SPhilipp Reisner } 2568b411b363SPhilipp Reisner 2569b411b363SPhilipp Reisner if (hg == -100) { 2570b411b363SPhilipp Reisner if (mdev->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1)) 2571b411b363SPhilipp Reisner hg = -1; 2572b411b363SPhilipp Reisner if (!mdev->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1)) 2573b411b363SPhilipp Reisner hg = 1; 2574b411b363SPhilipp Reisner 2575b411b363SPhilipp Reisner if (abs(hg) < 100) 2576b411b363SPhilipp Reisner dev_warn(DEV, "Split-Brain detected, manually solved. " 2577b411b363SPhilipp Reisner "Sync from %s node\n", 2578b411b363SPhilipp Reisner (hg < 0) ? "peer" : "this"); 2579b411b363SPhilipp Reisner } 2580b411b363SPhilipp Reisner 2581b411b363SPhilipp Reisner if (hg == -100) { 2582580b9767SLars Ellenberg /* FIXME this log message is not correct if we end up here 2583580b9767SLars Ellenberg * after an attempted attach on a diskless node. 2584580b9767SLars Ellenberg * We just refuse to attach -- well, we drop the "connection" 2585580b9767SLars Ellenberg * to that disk, in a way... */ 25863a11a487SAdam Gandelman dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n"); 2587b411b363SPhilipp Reisner drbd_khelper(mdev, "split-brain"); 2588b411b363SPhilipp Reisner return C_MASK; 2589b411b363SPhilipp Reisner } 2590b411b363SPhilipp Reisner 2591b411b363SPhilipp Reisner if (hg > 0 && mydisk <= D_INCONSISTENT) { 2592b411b363SPhilipp Reisner dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n"); 2593b411b363SPhilipp Reisner return C_MASK; 2594b411b363SPhilipp Reisner } 2595b411b363SPhilipp Reisner 2596b411b363SPhilipp Reisner if (hg < 0 && /* by intention we do not use mydisk here. */ 2597b411b363SPhilipp Reisner mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) { 2598b411b363SPhilipp Reisner switch (mdev->net_conf->rr_conflict) { 2599b411b363SPhilipp Reisner case ASB_CALL_HELPER: 2600b411b363SPhilipp Reisner drbd_khelper(mdev, "pri-lost"); 2601b411b363SPhilipp Reisner /* fall through */ 2602b411b363SPhilipp Reisner case ASB_DISCONNECT: 2603b411b363SPhilipp Reisner dev_err(DEV, "I shall become SyncTarget, but I am primary!\n"); 2604b411b363SPhilipp Reisner return C_MASK; 2605b411b363SPhilipp Reisner case ASB_VIOLENTLY: 2606b411b363SPhilipp Reisner dev_warn(DEV, "Becoming SyncTarget, violating the stable-data" 2607b411b363SPhilipp Reisner "assumption\n"); 2608b411b363SPhilipp Reisner } 2609b411b363SPhilipp Reisner } 2610b411b363SPhilipp Reisner 2611cf14c2e9SPhilipp Reisner if (mdev->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->flags)) { 2612cf14c2e9SPhilipp Reisner if (hg == 0) 2613cf14c2e9SPhilipp Reisner dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n"); 2614cf14c2e9SPhilipp Reisner else 2615cf14c2e9SPhilipp Reisner dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.", 2616cf14c2e9SPhilipp Reisner drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), 2617cf14c2e9SPhilipp Reisner abs(hg) >= 2 ? "full" : "bit-map based"); 2618cf14c2e9SPhilipp Reisner return C_MASK; 2619cf14c2e9SPhilipp Reisner } 2620cf14c2e9SPhilipp Reisner 2621b411b363SPhilipp Reisner if (abs(hg) >= 2) { 2622b411b363SPhilipp Reisner dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); 262320ceb2b2SLars Ellenberg if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", 262420ceb2b2SLars Ellenberg BM_LOCKED_SET_ALLOWED)) 2625b411b363SPhilipp Reisner return C_MASK; 2626b411b363SPhilipp Reisner } 2627b411b363SPhilipp Reisner 2628b411b363SPhilipp Reisner if (hg > 0) { /* become sync source. */ 2629b411b363SPhilipp Reisner rv = C_WF_BITMAP_S; 2630b411b363SPhilipp Reisner } else if (hg < 0) { /* become sync target */ 2631b411b363SPhilipp Reisner rv = C_WF_BITMAP_T; 2632b411b363SPhilipp Reisner } else { 2633b411b363SPhilipp Reisner rv = C_CONNECTED; 2634b411b363SPhilipp Reisner if (drbd_bm_total_weight(mdev)) { 2635b411b363SPhilipp Reisner dev_info(DEV, "No resync, but %lu bits in bitmap!\n", 2636b411b363SPhilipp Reisner drbd_bm_total_weight(mdev)); 2637b411b363SPhilipp Reisner } 2638b411b363SPhilipp Reisner } 2639b411b363SPhilipp Reisner 2640b411b363SPhilipp Reisner return rv; 2641b411b363SPhilipp Reisner } 2642b411b363SPhilipp Reisner 2643b411b363SPhilipp Reisner /* returns 1 if invalid */ 2644b411b363SPhilipp Reisner static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self) 2645b411b363SPhilipp Reisner { 2646b411b363SPhilipp Reisner /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ 2647b411b363SPhilipp Reisner if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) || 2648b411b363SPhilipp Reisner (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL)) 2649b411b363SPhilipp Reisner return 0; 2650b411b363SPhilipp Reisner 2651b411b363SPhilipp Reisner /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ 2652b411b363SPhilipp Reisner if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL || 2653b411b363SPhilipp Reisner self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL) 2654b411b363SPhilipp Reisner return 1; 2655b411b363SPhilipp Reisner 2656b411b363SPhilipp Reisner /* everything else is valid if they are equal on both sides. */ 2657b411b363SPhilipp Reisner if (peer == self) 2658b411b363SPhilipp Reisner return 0; 2659b411b363SPhilipp Reisner 2660b411b363SPhilipp Reisner /* everything es is invalid. */ 2661b411b363SPhilipp Reisner return 1; 2662b411b363SPhilipp Reisner } 2663b411b363SPhilipp Reisner 266402918be2SPhilipp Reisner static int receive_protocol(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 2665b411b363SPhilipp Reisner { 266602918be2SPhilipp Reisner struct p_protocol *p = &mdev->data.rbuf.protocol; 2667b411b363SPhilipp Reisner int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; 2668cf14c2e9SPhilipp Reisner int p_want_lose, p_two_primaries, cf; 2669b411b363SPhilipp Reisner char p_integrity_alg[SHARED_SECRET_MAX] = ""; 2670b411b363SPhilipp Reisner 2671b411b363SPhilipp Reisner p_proto = be32_to_cpu(p->protocol); 2672b411b363SPhilipp Reisner p_after_sb_0p = be32_to_cpu(p->after_sb_0p); 2673b411b363SPhilipp Reisner p_after_sb_1p = be32_to_cpu(p->after_sb_1p); 2674b411b363SPhilipp Reisner p_after_sb_2p = be32_to_cpu(p->after_sb_2p); 2675b411b363SPhilipp Reisner p_two_primaries = be32_to_cpu(p->two_primaries); 2676cf14c2e9SPhilipp Reisner cf = be32_to_cpu(p->conn_flags); 2677cf14c2e9SPhilipp Reisner p_want_lose = cf & CF_WANT_LOSE; 2678cf14c2e9SPhilipp Reisner 2679cf14c2e9SPhilipp Reisner clear_bit(CONN_DRY_RUN, &mdev->flags); 2680cf14c2e9SPhilipp Reisner 2681cf14c2e9SPhilipp Reisner if (cf & CF_DRY_RUN) 2682cf14c2e9SPhilipp Reisner set_bit(CONN_DRY_RUN, &mdev->flags); 2683b411b363SPhilipp Reisner 2684b411b363SPhilipp Reisner if (p_proto != mdev->net_conf->wire_protocol) { 2685b411b363SPhilipp Reisner dev_err(DEV, "incompatible communication protocols\n"); 2686b411b363SPhilipp Reisner goto disconnect; 2687b411b363SPhilipp Reisner } 2688b411b363SPhilipp Reisner 2689b411b363SPhilipp Reisner if (cmp_after_sb(p_after_sb_0p, mdev->net_conf->after_sb_0p)) { 2690b411b363SPhilipp Reisner dev_err(DEV, "incompatible after-sb-0pri settings\n"); 2691b411b363SPhilipp Reisner goto disconnect; 2692b411b363SPhilipp Reisner } 2693b411b363SPhilipp Reisner 2694b411b363SPhilipp Reisner if (cmp_after_sb(p_after_sb_1p, mdev->net_conf->after_sb_1p)) { 2695b411b363SPhilipp Reisner dev_err(DEV, "incompatible after-sb-1pri settings\n"); 2696b411b363SPhilipp Reisner goto disconnect; 2697b411b363SPhilipp Reisner } 2698b411b363SPhilipp Reisner 2699b411b363SPhilipp Reisner if (cmp_after_sb(p_after_sb_2p, mdev->net_conf->after_sb_2p)) { 2700b411b363SPhilipp Reisner dev_err(DEV, "incompatible after-sb-2pri settings\n"); 2701b411b363SPhilipp Reisner goto disconnect; 2702b411b363SPhilipp Reisner } 2703b411b363SPhilipp Reisner 2704b411b363SPhilipp Reisner if (p_want_lose && mdev->net_conf->want_lose) { 2705b411b363SPhilipp Reisner dev_err(DEV, "both sides have the 'want_lose' flag set\n"); 2706b411b363SPhilipp Reisner goto disconnect; 2707b411b363SPhilipp Reisner } 2708b411b363SPhilipp Reisner 2709b411b363SPhilipp Reisner if (p_two_primaries != mdev->net_conf->two_primaries) { 2710b411b363SPhilipp Reisner dev_err(DEV, "incompatible setting of the two-primaries options\n"); 2711b411b363SPhilipp Reisner goto disconnect; 2712b411b363SPhilipp Reisner } 2713b411b363SPhilipp Reisner 2714b411b363SPhilipp Reisner if (mdev->agreed_pro_version >= 87) { 2715b411b363SPhilipp Reisner unsigned char *my_alg = mdev->net_conf->integrity_alg; 2716b411b363SPhilipp Reisner 2717b411b363SPhilipp Reisner if (drbd_recv(mdev, p_integrity_alg, data_size) != data_size) 271881e84650SAndreas Gruenbacher return false; 2719b411b363SPhilipp Reisner 2720b411b363SPhilipp Reisner p_integrity_alg[SHARED_SECRET_MAX-1] = 0; 2721b411b363SPhilipp Reisner if (strcmp(p_integrity_alg, my_alg)) { 2722b411b363SPhilipp Reisner dev_err(DEV, "incompatible setting of the data-integrity-alg\n"); 2723b411b363SPhilipp Reisner goto disconnect; 2724b411b363SPhilipp Reisner } 2725b411b363SPhilipp Reisner dev_info(DEV, "data-integrity-alg: %s\n", 2726b411b363SPhilipp Reisner my_alg[0] ? my_alg : (unsigned char *)"<not-used>"); 2727b411b363SPhilipp Reisner } 2728b411b363SPhilipp Reisner 272981e84650SAndreas Gruenbacher return true; 2730b411b363SPhilipp Reisner 2731b411b363SPhilipp Reisner disconnect: 2732b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 273381e84650SAndreas Gruenbacher return false; 2734b411b363SPhilipp Reisner } 2735b411b363SPhilipp Reisner 2736b411b363SPhilipp Reisner /* helper function 2737b411b363SPhilipp Reisner * input: alg name, feature name 2738b411b363SPhilipp Reisner * return: NULL (alg name was "") 2739b411b363SPhilipp Reisner * ERR_PTR(error) if something goes wrong 2740b411b363SPhilipp Reisner * or the crypto hash ptr, if it worked out ok. */ 2741b411b363SPhilipp Reisner struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev, 2742b411b363SPhilipp Reisner const char *alg, const char *name) 2743b411b363SPhilipp Reisner { 2744b411b363SPhilipp Reisner struct crypto_hash *tfm; 2745b411b363SPhilipp Reisner 2746b411b363SPhilipp Reisner if (!alg[0]) 2747b411b363SPhilipp Reisner return NULL; 2748b411b363SPhilipp Reisner 2749b411b363SPhilipp Reisner tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); 2750b411b363SPhilipp Reisner if (IS_ERR(tfm)) { 2751b411b363SPhilipp Reisner dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n", 2752b411b363SPhilipp Reisner alg, name, PTR_ERR(tfm)); 2753b411b363SPhilipp Reisner return tfm; 2754b411b363SPhilipp Reisner } 2755b411b363SPhilipp Reisner if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) { 2756b411b363SPhilipp Reisner crypto_free_hash(tfm); 2757b411b363SPhilipp Reisner dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name); 2758b411b363SPhilipp Reisner return ERR_PTR(-EINVAL); 2759b411b363SPhilipp Reisner } 2760b411b363SPhilipp Reisner return tfm; 2761b411b363SPhilipp Reisner } 2762b411b363SPhilipp Reisner 276302918be2SPhilipp Reisner static int receive_SyncParam(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int packet_size) 2764b411b363SPhilipp Reisner { 276581e84650SAndreas Gruenbacher int ok = true; 276602918be2SPhilipp Reisner struct p_rs_param_95 *p = &mdev->data.rbuf.rs_param_95; 2767b411b363SPhilipp Reisner unsigned int header_size, data_size, exp_max_sz; 2768b411b363SPhilipp Reisner struct crypto_hash *verify_tfm = NULL; 2769b411b363SPhilipp Reisner struct crypto_hash *csums_tfm = NULL; 2770b411b363SPhilipp Reisner const int apv = mdev->agreed_pro_version; 2771778f271dSPhilipp Reisner int *rs_plan_s = NULL; 2772778f271dSPhilipp Reisner int fifo_size = 0; 2773b411b363SPhilipp Reisner 2774b411b363SPhilipp Reisner exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) 2775b411b363SPhilipp Reisner : apv == 88 ? sizeof(struct p_rs_param) 2776b411b363SPhilipp Reisner + SHARED_SECRET_MAX 27778e26f9ccSPhilipp Reisner : apv <= 94 ? sizeof(struct p_rs_param_89) 27788e26f9ccSPhilipp Reisner : /* apv >= 95 */ sizeof(struct p_rs_param_95); 2779b411b363SPhilipp Reisner 278002918be2SPhilipp Reisner if (packet_size > exp_max_sz) { 2781b411b363SPhilipp Reisner dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n", 278202918be2SPhilipp Reisner packet_size, exp_max_sz); 278381e84650SAndreas Gruenbacher return false; 2784b411b363SPhilipp Reisner } 2785b411b363SPhilipp Reisner 2786b411b363SPhilipp Reisner if (apv <= 88) { 278702918be2SPhilipp Reisner header_size = sizeof(struct p_rs_param) - sizeof(struct p_header80); 278802918be2SPhilipp Reisner data_size = packet_size - header_size; 27898e26f9ccSPhilipp Reisner } else if (apv <= 94) { 279002918be2SPhilipp Reisner header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header80); 279102918be2SPhilipp Reisner data_size = packet_size - header_size; 2792b411b363SPhilipp Reisner D_ASSERT(data_size == 0); 27938e26f9ccSPhilipp Reisner } else { 279402918be2SPhilipp Reisner header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header80); 279502918be2SPhilipp Reisner data_size = packet_size - header_size; 2796b411b363SPhilipp Reisner D_ASSERT(data_size == 0); 2797b411b363SPhilipp Reisner } 2798b411b363SPhilipp Reisner 2799b411b363SPhilipp Reisner /* initialize verify_alg and csums_alg */ 2800b411b363SPhilipp Reisner memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); 2801b411b363SPhilipp Reisner 280202918be2SPhilipp Reisner if (drbd_recv(mdev, &p->head.payload, header_size) != header_size) 280381e84650SAndreas Gruenbacher return false; 2804b411b363SPhilipp Reisner 2805b411b363SPhilipp Reisner mdev->sync_conf.rate = be32_to_cpu(p->rate); 2806b411b363SPhilipp Reisner 2807b411b363SPhilipp Reisner if (apv >= 88) { 2808b411b363SPhilipp Reisner if (apv == 88) { 2809b411b363SPhilipp Reisner if (data_size > SHARED_SECRET_MAX) { 2810b411b363SPhilipp Reisner dev_err(DEV, "verify-alg too long, " 2811b411b363SPhilipp Reisner "peer wants %u, accepting only %u byte\n", 2812b411b363SPhilipp Reisner data_size, SHARED_SECRET_MAX); 281381e84650SAndreas Gruenbacher return false; 2814b411b363SPhilipp Reisner } 2815b411b363SPhilipp Reisner 2816b411b363SPhilipp Reisner if (drbd_recv(mdev, p->verify_alg, data_size) != data_size) 281781e84650SAndreas Gruenbacher return false; 2818b411b363SPhilipp Reisner 2819b411b363SPhilipp Reisner /* we expect NUL terminated string */ 2820b411b363SPhilipp Reisner /* but just in case someone tries to be evil */ 2821b411b363SPhilipp Reisner D_ASSERT(p->verify_alg[data_size-1] == 0); 2822b411b363SPhilipp Reisner p->verify_alg[data_size-1] = 0; 2823b411b363SPhilipp Reisner 2824b411b363SPhilipp Reisner } else /* apv >= 89 */ { 2825b411b363SPhilipp Reisner /* we still expect NUL terminated strings */ 2826b411b363SPhilipp Reisner /* but just in case someone tries to be evil */ 2827b411b363SPhilipp Reisner D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0); 2828b411b363SPhilipp Reisner D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0); 2829b411b363SPhilipp Reisner p->verify_alg[SHARED_SECRET_MAX-1] = 0; 2830b411b363SPhilipp Reisner p->csums_alg[SHARED_SECRET_MAX-1] = 0; 2831b411b363SPhilipp Reisner } 2832b411b363SPhilipp Reisner 2833b411b363SPhilipp Reisner if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) { 2834b411b363SPhilipp Reisner if (mdev->state.conn == C_WF_REPORT_PARAMS) { 2835b411b363SPhilipp Reisner dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", 2836b411b363SPhilipp Reisner mdev->sync_conf.verify_alg, p->verify_alg); 2837b411b363SPhilipp Reisner goto disconnect; 2838b411b363SPhilipp Reisner } 2839b411b363SPhilipp Reisner verify_tfm = drbd_crypto_alloc_digest_safe(mdev, 2840b411b363SPhilipp Reisner p->verify_alg, "verify-alg"); 2841b411b363SPhilipp Reisner if (IS_ERR(verify_tfm)) { 2842b411b363SPhilipp Reisner verify_tfm = NULL; 2843b411b363SPhilipp Reisner goto disconnect; 2844b411b363SPhilipp Reisner } 2845b411b363SPhilipp Reisner } 2846b411b363SPhilipp Reisner 2847b411b363SPhilipp Reisner if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) { 2848b411b363SPhilipp Reisner if (mdev->state.conn == C_WF_REPORT_PARAMS) { 2849b411b363SPhilipp Reisner dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", 2850b411b363SPhilipp Reisner mdev->sync_conf.csums_alg, p->csums_alg); 2851b411b363SPhilipp Reisner goto disconnect; 2852b411b363SPhilipp Reisner } 2853b411b363SPhilipp Reisner csums_tfm = drbd_crypto_alloc_digest_safe(mdev, 2854b411b363SPhilipp Reisner p->csums_alg, "csums-alg"); 2855b411b363SPhilipp Reisner if (IS_ERR(csums_tfm)) { 2856b411b363SPhilipp Reisner csums_tfm = NULL; 2857b411b363SPhilipp Reisner goto disconnect; 2858b411b363SPhilipp Reisner } 2859b411b363SPhilipp Reisner } 2860b411b363SPhilipp Reisner 28618e26f9ccSPhilipp Reisner if (apv > 94) { 28628e26f9ccSPhilipp Reisner mdev->sync_conf.rate = be32_to_cpu(p->rate); 28638e26f9ccSPhilipp Reisner mdev->sync_conf.c_plan_ahead = be32_to_cpu(p->c_plan_ahead); 28648e26f9ccSPhilipp Reisner mdev->sync_conf.c_delay_target = be32_to_cpu(p->c_delay_target); 28658e26f9ccSPhilipp Reisner mdev->sync_conf.c_fill_target = be32_to_cpu(p->c_fill_target); 28668e26f9ccSPhilipp Reisner mdev->sync_conf.c_max_rate = be32_to_cpu(p->c_max_rate); 2867778f271dSPhilipp Reisner 2868778f271dSPhilipp Reisner fifo_size = (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ; 2869778f271dSPhilipp Reisner if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) { 2870778f271dSPhilipp Reisner rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL); 2871778f271dSPhilipp Reisner if (!rs_plan_s) { 2872778f271dSPhilipp Reisner dev_err(DEV, "kmalloc of fifo_buffer failed"); 2873778f271dSPhilipp Reisner goto disconnect; 2874778f271dSPhilipp Reisner } 2875778f271dSPhilipp Reisner } 28768e26f9ccSPhilipp Reisner } 2877b411b363SPhilipp Reisner 2878b411b363SPhilipp Reisner spin_lock(&mdev->peer_seq_lock); 2879b411b363SPhilipp Reisner /* lock against drbd_nl_syncer_conf() */ 2880b411b363SPhilipp Reisner if (verify_tfm) { 2881b411b363SPhilipp Reisner strcpy(mdev->sync_conf.verify_alg, p->verify_alg); 2882b411b363SPhilipp Reisner mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1; 2883b411b363SPhilipp Reisner crypto_free_hash(mdev->verify_tfm); 2884b411b363SPhilipp Reisner mdev->verify_tfm = verify_tfm; 2885b411b363SPhilipp Reisner dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg); 2886b411b363SPhilipp Reisner } 2887b411b363SPhilipp Reisner if (csums_tfm) { 2888b411b363SPhilipp Reisner strcpy(mdev->sync_conf.csums_alg, p->csums_alg); 2889b411b363SPhilipp Reisner mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1; 2890b411b363SPhilipp Reisner crypto_free_hash(mdev->csums_tfm); 2891b411b363SPhilipp Reisner mdev->csums_tfm = csums_tfm; 2892b411b363SPhilipp Reisner dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg); 2893b411b363SPhilipp Reisner } 2894778f271dSPhilipp Reisner if (fifo_size != mdev->rs_plan_s.size) { 2895778f271dSPhilipp Reisner kfree(mdev->rs_plan_s.values); 2896778f271dSPhilipp Reisner mdev->rs_plan_s.values = rs_plan_s; 2897778f271dSPhilipp Reisner mdev->rs_plan_s.size = fifo_size; 2898778f271dSPhilipp Reisner mdev->rs_planed = 0; 2899778f271dSPhilipp Reisner } 2900b411b363SPhilipp Reisner spin_unlock(&mdev->peer_seq_lock); 2901b411b363SPhilipp Reisner } 2902b411b363SPhilipp Reisner 2903b411b363SPhilipp Reisner return ok; 2904b411b363SPhilipp Reisner disconnect: 2905b411b363SPhilipp Reisner /* just for completeness: actually not needed, 2906b411b363SPhilipp Reisner * as this is not reached if csums_tfm was ok. */ 2907b411b363SPhilipp Reisner crypto_free_hash(csums_tfm); 2908b411b363SPhilipp Reisner /* but free the verify_tfm again, if csums_tfm did not work out */ 2909b411b363SPhilipp Reisner crypto_free_hash(verify_tfm); 2910b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 291181e84650SAndreas Gruenbacher return false; 2912b411b363SPhilipp Reisner } 2913b411b363SPhilipp Reisner 2914b411b363SPhilipp Reisner /* warn if the arguments differ by more than 12.5% */ 2915b411b363SPhilipp Reisner static void warn_if_differ_considerably(struct drbd_conf *mdev, 2916b411b363SPhilipp Reisner const char *s, sector_t a, sector_t b) 2917b411b363SPhilipp Reisner { 2918b411b363SPhilipp Reisner sector_t d; 2919b411b363SPhilipp Reisner if (a == 0 || b == 0) 2920b411b363SPhilipp Reisner return; 2921b411b363SPhilipp Reisner d = (a > b) ? (a - b) : (b - a); 2922b411b363SPhilipp Reisner if (d > (a>>3) || d > (b>>3)) 2923b411b363SPhilipp Reisner dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s, 2924b411b363SPhilipp Reisner (unsigned long long)a, (unsigned long long)b); 2925b411b363SPhilipp Reisner } 2926b411b363SPhilipp Reisner 292702918be2SPhilipp Reisner static int receive_sizes(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 2928b411b363SPhilipp Reisner { 292902918be2SPhilipp Reisner struct p_sizes *p = &mdev->data.rbuf.sizes; 2930b411b363SPhilipp Reisner enum determine_dev_size dd = unchanged; 2931b411b363SPhilipp Reisner sector_t p_size, p_usize, my_usize; 2932b411b363SPhilipp Reisner int ldsc = 0; /* local disk size changed */ 2933e89b591cSPhilipp Reisner enum dds_flags ddsf; 2934b411b363SPhilipp Reisner 2935b411b363SPhilipp Reisner p_size = be64_to_cpu(p->d_size); 2936b411b363SPhilipp Reisner p_usize = be64_to_cpu(p->u_size); 2937b411b363SPhilipp Reisner 2938b411b363SPhilipp Reisner if (p_size == 0 && mdev->state.disk == D_DISKLESS) { 2939b411b363SPhilipp Reisner dev_err(DEV, "some backing storage is needed\n"); 2940b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 294181e84650SAndreas Gruenbacher return false; 2942b411b363SPhilipp Reisner } 2943b411b363SPhilipp Reisner 2944b411b363SPhilipp Reisner /* just store the peer's disk size for now. 2945b411b363SPhilipp Reisner * we still need to figure out whether we accept that. */ 2946b411b363SPhilipp Reisner mdev->p_size = p_size; 2947b411b363SPhilipp Reisner 2948b411b363SPhilipp Reisner if (get_ldev(mdev)) { 2949b411b363SPhilipp Reisner warn_if_differ_considerably(mdev, "lower level device sizes", 2950b411b363SPhilipp Reisner p_size, drbd_get_max_capacity(mdev->ldev)); 2951b411b363SPhilipp Reisner warn_if_differ_considerably(mdev, "user requested size", 2952b411b363SPhilipp Reisner p_usize, mdev->ldev->dc.disk_size); 2953b411b363SPhilipp Reisner 2954b411b363SPhilipp Reisner /* if this is the first connect, or an otherwise expected 2955b411b363SPhilipp Reisner * param exchange, choose the minimum */ 2956b411b363SPhilipp Reisner if (mdev->state.conn == C_WF_REPORT_PARAMS) 2957b411b363SPhilipp Reisner p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size, 2958b411b363SPhilipp Reisner p_usize); 2959b411b363SPhilipp Reisner 2960b411b363SPhilipp Reisner my_usize = mdev->ldev->dc.disk_size; 2961b411b363SPhilipp Reisner 2962b411b363SPhilipp Reisner if (mdev->ldev->dc.disk_size != p_usize) { 2963b411b363SPhilipp Reisner mdev->ldev->dc.disk_size = p_usize; 2964b411b363SPhilipp Reisner dev_info(DEV, "Peer sets u_size to %lu sectors\n", 2965b411b363SPhilipp Reisner (unsigned long)mdev->ldev->dc.disk_size); 2966b411b363SPhilipp Reisner } 2967b411b363SPhilipp Reisner 2968b411b363SPhilipp Reisner /* Never shrink a device with usable data during connect. 2969b411b363SPhilipp Reisner But allow online shrinking if we are connected. */ 2970a393db6fSPhilipp Reisner if (drbd_new_dev_size(mdev, mdev->ldev, 0) < 2971b411b363SPhilipp Reisner drbd_get_capacity(mdev->this_bdev) && 2972b411b363SPhilipp Reisner mdev->state.disk >= D_OUTDATED && 2973b411b363SPhilipp Reisner mdev->state.conn < C_CONNECTED) { 2974b411b363SPhilipp Reisner dev_err(DEV, "The peer's disk size is too small!\n"); 2975b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 2976b411b363SPhilipp Reisner mdev->ldev->dc.disk_size = my_usize; 2977b411b363SPhilipp Reisner put_ldev(mdev); 297881e84650SAndreas Gruenbacher return false; 2979b411b363SPhilipp Reisner } 2980b411b363SPhilipp Reisner put_ldev(mdev); 2981b411b363SPhilipp Reisner } 2982b411b363SPhilipp Reisner 2983e89b591cSPhilipp Reisner ddsf = be16_to_cpu(p->dds_flags); 2984b411b363SPhilipp Reisner if (get_ldev(mdev)) { 298524c4830cSBart Van Assche dd = drbd_determine_dev_size(mdev, ddsf); 2986b411b363SPhilipp Reisner put_ldev(mdev); 2987b411b363SPhilipp Reisner if (dd == dev_size_error) 298881e84650SAndreas Gruenbacher return false; 2989b411b363SPhilipp Reisner drbd_md_sync(mdev); 2990b411b363SPhilipp Reisner } else { 2991b411b363SPhilipp Reisner /* I am diskless, need to accept the peer's size. */ 2992b411b363SPhilipp Reisner drbd_set_my_capacity(mdev, p_size); 2993b411b363SPhilipp Reisner } 2994b411b363SPhilipp Reisner 299599432fccSPhilipp Reisner mdev->peer_max_bio_size = be32_to_cpu(p->max_bio_size); 299699432fccSPhilipp Reisner drbd_reconsider_max_bio_size(mdev); 299799432fccSPhilipp Reisner 2998b411b363SPhilipp Reisner if (get_ldev(mdev)) { 2999b411b363SPhilipp Reisner if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) { 3000b411b363SPhilipp Reisner mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); 3001b411b363SPhilipp Reisner ldsc = 1; 3002b411b363SPhilipp Reisner } 3003b411b363SPhilipp Reisner 3004b411b363SPhilipp Reisner put_ldev(mdev); 3005b411b363SPhilipp Reisner } 3006b411b363SPhilipp Reisner 3007b411b363SPhilipp Reisner if (mdev->state.conn > C_WF_REPORT_PARAMS) { 3008b411b363SPhilipp Reisner if (be64_to_cpu(p->c_size) != 3009b411b363SPhilipp Reisner drbd_get_capacity(mdev->this_bdev) || ldsc) { 3010b411b363SPhilipp Reisner /* we have different sizes, probably peer 3011b411b363SPhilipp Reisner * needs to know my new size... */ 3012e89b591cSPhilipp Reisner drbd_send_sizes(mdev, 0, ddsf); 3013b411b363SPhilipp Reisner } 3014b411b363SPhilipp Reisner if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) || 3015b411b363SPhilipp Reisner (dd == grew && mdev->state.conn == C_CONNECTED)) { 3016b411b363SPhilipp Reisner if (mdev->state.pdsk >= D_INCONSISTENT && 3017e89b591cSPhilipp Reisner mdev->state.disk >= D_INCONSISTENT) { 3018e89b591cSPhilipp Reisner if (ddsf & DDSF_NO_RESYNC) 3019e89b591cSPhilipp Reisner dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n"); 3020b411b363SPhilipp Reisner else 3021e89b591cSPhilipp Reisner resync_after_online_grow(mdev); 3022e89b591cSPhilipp Reisner } else 3023b411b363SPhilipp Reisner set_bit(RESYNC_AFTER_NEG, &mdev->flags); 3024b411b363SPhilipp Reisner } 3025b411b363SPhilipp Reisner } 3026b411b363SPhilipp Reisner 302781e84650SAndreas Gruenbacher return true; 3028b411b363SPhilipp Reisner } 3029b411b363SPhilipp Reisner 303002918be2SPhilipp Reisner static int receive_uuids(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3031b411b363SPhilipp Reisner { 303202918be2SPhilipp Reisner struct p_uuids *p = &mdev->data.rbuf.uuids; 3033b411b363SPhilipp Reisner u64 *p_uuid; 303462b0da3aSLars Ellenberg int i, updated_uuids = 0; 3035b411b363SPhilipp Reisner 3036b411b363SPhilipp Reisner p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); 3037b411b363SPhilipp Reisner 3038b411b363SPhilipp Reisner for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) 3039b411b363SPhilipp Reisner p_uuid[i] = be64_to_cpu(p->uuid[i]); 3040b411b363SPhilipp Reisner 3041b411b363SPhilipp Reisner kfree(mdev->p_uuid); 3042b411b363SPhilipp Reisner mdev->p_uuid = p_uuid; 3043b411b363SPhilipp Reisner 3044b411b363SPhilipp Reisner if (mdev->state.conn < C_CONNECTED && 3045b411b363SPhilipp Reisner mdev->state.disk < D_INCONSISTENT && 3046b411b363SPhilipp Reisner mdev->state.role == R_PRIMARY && 3047b411b363SPhilipp Reisner (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { 3048b411b363SPhilipp Reisner dev_err(DEV, "Can only connect to data with current UUID=%016llX\n", 3049b411b363SPhilipp Reisner (unsigned long long)mdev->ed_uuid); 3050b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 305181e84650SAndreas Gruenbacher return false; 3052b411b363SPhilipp Reisner } 3053b411b363SPhilipp Reisner 3054b411b363SPhilipp Reisner if (get_ldev(mdev)) { 3055b411b363SPhilipp Reisner int skip_initial_sync = 3056b411b363SPhilipp Reisner mdev->state.conn == C_CONNECTED && 3057b411b363SPhilipp Reisner mdev->agreed_pro_version >= 90 && 3058b411b363SPhilipp Reisner mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && 3059b411b363SPhilipp Reisner (p_uuid[UI_FLAGS] & 8); 3060b411b363SPhilipp Reisner if (skip_initial_sync) { 3061b411b363SPhilipp Reisner dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n"); 3062b411b363SPhilipp Reisner drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, 306320ceb2b2SLars Ellenberg "clear_n_write from receive_uuids", 306420ceb2b2SLars Ellenberg BM_LOCKED_TEST_ALLOWED); 3065b411b363SPhilipp Reisner _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]); 3066b411b363SPhilipp Reisner _drbd_uuid_set(mdev, UI_BITMAP, 0); 3067b411b363SPhilipp Reisner _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), 3068b411b363SPhilipp Reisner CS_VERBOSE, NULL); 3069b411b363SPhilipp Reisner drbd_md_sync(mdev); 307062b0da3aSLars Ellenberg updated_uuids = 1; 3071b411b363SPhilipp Reisner } 3072b411b363SPhilipp Reisner put_ldev(mdev); 307318a50fa2SPhilipp Reisner } else if (mdev->state.disk < D_INCONSISTENT && 307418a50fa2SPhilipp Reisner mdev->state.role == R_PRIMARY) { 307518a50fa2SPhilipp Reisner /* I am a diskless primary, the peer just created a new current UUID 307618a50fa2SPhilipp Reisner for me. */ 307762b0da3aSLars Ellenberg updated_uuids = drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); 3078b411b363SPhilipp Reisner } 3079b411b363SPhilipp Reisner 3080b411b363SPhilipp Reisner /* Before we test for the disk state, we should wait until an eventually 3081b411b363SPhilipp Reisner ongoing cluster wide state change is finished. That is important if 3082b411b363SPhilipp Reisner we are primary and are detaching from our disk. We need to see the 3083b411b363SPhilipp Reisner new disk state... */ 3084b411b363SPhilipp Reisner wait_event(mdev->misc_wait, !test_bit(CLUSTER_ST_CHANGE, &mdev->flags)); 3085b411b363SPhilipp Reisner if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT) 308662b0da3aSLars Ellenberg updated_uuids |= drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); 308762b0da3aSLars Ellenberg 308862b0da3aSLars Ellenberg if (updated_uuids) 308962b0da3aSLars Ellenberg drbd_print_uuids(mdev, "receiver updated UUIDs to"); 3090b411b363SPhilipp Reisner 309181e84650SAndreas Gruenbacher return true; 3092b411b363SPhilipp Reisner } 3093b411b363SPhilipp Reisner 3094b411b363SPhilipp Reisner /** 3095b411b363SPhilipp Reisner * convert_state() - Converts the peer's view of the cluster state to our point of view 3096b411b363SPhilipp Reisner * @ps: The state as seen by the peer. 3097b411b363SPhilipp Reisner */ 3098b411b363SPhilipp Reisner static union drbd_state convert_state(union drbd_state ps) 3099b411b363SPhilipp Reisner { 3100b411b363SPhilipp Reisner union drbd_state ms; 3101b411b363SPhilipp Reisner 3102b411b363SPhilipp Reisner static enum drbd_conns c_tab[] = { 3103b411b363SPhilipp Reisner [C_CONNECTED] = C_CONNECTED, 3104b411b363SPhilipp Reisner 3105b411b363SPhilipp Reisner [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, 3106b411b363SPhilipp Reisner [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, 3107b411b363SPhilipp Reisner [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ 3108b411b363SPhilipp Reisner [C_VERIFY_S] = C_VERIFY_T, 3109b411b363SPhilipp Reisner [C_MASK] = C_MASK, 3110b411b363SPhilipp Reisner }; 3111b411b363SPhilipp Reisner 3112b411b363SPhilipp Reisner ms.i = ps.i; 3113b411b363SPhilipp Reisner 3114b411b363SPhilipp Reisner ms.conn = c_tab[ps.conn]; 3115b411b363SPhilipp Reisner ms.peer = ps.role; 3116b411b363SPhilipp Reisner ms.role = ps.peer; 3117b411b363SPhilipp Reisner ms.pdsk = ps.disk; 3118b411b363SPhilipp Reisner ms.disk = ps.pdsk; 3119b411b363SPhilipp Reisner ms.peer_isp = (ps.aftr_isp | ps.user_isp); 3120b411b363SPhilipp Reisner 3121b411b363SPhilipp Reisner return ms; 3122b411b363SPhilipp Reisner } 3123b411b363SPhilipp Reisner 312402918be2SPhilipp Reisner static int receive_req_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3125b411b363SPhilipp Reisner { 312602918be2SPhilipp Reisner struct p_req_state *p = &mdev->data.rbuf.req_state; 3127b411b363SPhilipp Reisner union drbd_state mask, val; 3128bf885f8aSAndreas Gruenbacher enum drbd_state_rv rv; 3129b411b363SPhilipp Reisner 3130b411b363SPhilipp Reisner mask.i = be32_to_cpu(p->mask); 3131b411b363SPhilipp Reisner val.i = be32_to_cpu(p->val); 3132b411b363SPhilipp Reisner 3133b411b363SPhilipp Reisner if (test_bit(DISCARD_CONCURRENT, &mdev->flags) && 3134b411b363SPhilipp Reisner test_bit(CLUSTER_ST_CHANGE, &mdev->flags)) { 3135b411b363SPhilipp Reisner drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG); 313681e84650SAndreas Gruenbacher return true; 3137b411b363SPhilipp Reisner } 3138b411b363SPhilipp Reisner 3139b411b363SPhilipp Reisner mask = convert_state(mask); 3140b411b363SPhilipp Reisner val = convert_state(val); 3141b411b363SPhilipp Reisner 3142b411b363SPhilipp Reisner rv = drbd_change_state(mdev, CS_VERBOSE, mask, val); 3143b411b363SPhilipp Reisner 3144b411b363SPhilipp Reisner drbd_send_sr_reply(mdev, rv); 3145b411b363SPhilipp Reisner drbd_md_sync(mdev); 3146b411b363SPhilipp Reisner 314781e84650SAndreas Gruenbacher return true; 3148b411b363SPhilipp Reisner } 3149b411b363SPhilipp Reisner 315002918be2SPhilipp Reisner static int receive_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3151b411b363SPhilipp Reisner { 315202918be2SPhilipp Reisner struct p_state *p = &mdev->data.rbuf.state; 31534ac4aadaSLars Ellenberg union drbd_state os, ns, peer_state; 3154b411b363SPhilipp Reisner enum drbd_disk_state real_peer_disk; 315565d922c3SPhilipp Reisner enum chg_state_flags cs_flags; 3156b411b363SPhilipp Reisner int rv; 3157b411b363SPhilipp Reisner 3158b411b363SPhilipp Reisner peer_state.i = be32_to_cpu(p->state); 3159b411b363SPhilipp Reisner 3160b411b363SPhilipp Reisner real_peer_disk = peer_state.disk; 3161b411b363SPhilipp Reisner if (peer_state.disk == D_NEGOTIATING) { 3162b411b363SPhilipp Reisner real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; 3163b411b363SPhilipp Reisner dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); 3164b411b363SPhilipp Reisner } 3165b411b363SPhilipp Reisner 3166b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 3167b411b363SPhilipp Reisner retry: 31684ac4aadaSLars Ellenberg os = ns = mdev->state; 3169b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 3170b411b363SPhilipp Reisner 3171e9ef7bb6SLars Ellenberg /* peer says his disk is uptodate, while we think it is inconsistent, 3172e9ef7bb6SLars Ellenberg * and this happens while we think we have a sync going on. */ 3173e9ef7bb6SLars Ellenberg if (os.pdsk == D_INCONSISTENT && real_peer_disk == D_UP_TO_DATE && 3174e9ef7bb6SLars Ellenberg os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { 3175e9ef7bb6SLars Ellenberg /* If we are (becoming) SyncSource, but peer is still in sync 3176e9ef7bb6SLars Ellenberg * preparation, ignore its uptodate-ness to avoid flapping, it 3177e9ef7bb6SLars Ellenberg * will change to inconsistent once the peer reaches active 3178e9ef7bb6SLars Ellenberg * syncing states. 3179e9ef7bb6SLars Ellenberg * It may have changed syncer-paused flags, however, so we 3180e9ef7bb6SLars Ellenberg * cannot ignore this completely. */ 3181e9ef7bb6SLars Ellenberg if (peer_state.conn > C_CONNECTED && 3182e9ef7bb6SLars Ellenberg peer_state.conn < C_SYNC_SOURCE) 3183e9ef7bb6SLars Ellenberg real_peer_disk = D_INCONSISTENT; 3184e9ef7bb6SLars Ellenberg 3185e9ef7bb6SLars Ellenberg /* if peer_state changes to connected at the same time, 3186e9ef7bb6SLars Ellenberg * it explicitly notifies us that it finished resync. 3187e9ef7bb6SLars Ellenberg * Maybe we should finish it up, too? */ 3188e9ef7bb6SLars Ellenberg else if (os.conn >= C_SYNC_SOURCE && 3189e9ef7bb6SLars Ellenberg peer_state.conn == C_CONNECTED) { 3190e9ef7bb6SLars Ellenberg if (drbd_bm_total_weight(mdev) <= mdev->rs_failed) 3191e9ef7bb6SLars Ellenberg drbd_resync_finished(mdev); 319281e84650SAndreas Gruenbacher return true; 3193e9ef7bb6SLars Ellenberg } 3194e9ef7bb6SLars Ellenberg } 3195e9ef7bb6SLars Ellenberg 3196e9ef7bb6SLars Ellenberg /* peer says his disk is inconsistent, while we think it is uptodate, 3197e9ef7bb6SLars Ellenberg * and this happens while the peer still thinks we have a sync going on, 3198e9ef7bb6SLars Ellenberg * but we think we are already done with the sync. 3199e9ef7bb6SLars Ellenberg * We ignore this to avoid flapping pdsk. 3200e9ef7bb6SLars Ellenberg * This should not happen, if the peer is a recent version of drbd. */ 3201e9ef7bb6SLars Ellenberg if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && 3202e9ef7bb6SLars Ellenberg os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) 3203e9ef7bb6SLars Ellenberg real_peer_disk = D_UP_TO_DATE; 3204e9ef7bb6SLars Ellenberg 32054ac4aadaSLars Ellenberg if (ns.conn == C_WF_REPORT_PARAMS) 32064ac4aadaSLars Ellenberg ns.conn = C_CONNECTED; 3207b411b363SPhilipp Reisner 320867531718SPhilipp Reisner if (peer_state.conn == C_AHEAD) 320967531718SPhilipp Reisner ns.conn = C_BEHIND; 321067531718SPhilipp Reisner 3211b411b363SPhilipp Reisner if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING && 3212b411b363SPhilipp Reisner get_ldev_if_state(mdev, D_NEGOTIATING)) { 3213b411b363SPhilipp Reisner int cr; /* consider resync */ 3214b411b363SPhilipp Reisner 3215b411b363SPhilipp Reisner /* if we established a new connection */ 32164ac4aadaSLars Ellenberg cr = (os.conn < C_CONNECTED); 3217b411b363SPhilipp Reisner /* if we had an established connection 3218b411b363SPhilipp Reisner * and one of the nodes newly attaches a disk */ 32194ac4aadaSLars Ellenberg cr |= (os.conn == C_CONNECTED && 3220b411b363SPhilipp Reisner (peer_state.disk == D_NEGOTIATING || 32214ac4aadaSLars Ellenberg os.disk == D_NEGOTIATING)); 3222b411b363SPhilipp Reisner /* if we have both been inconsistent, and the peer has been 3223b411b363SPhilipp Reisner * forced to be UpToDate with --overwrite-data */ 3224b411b363SPhilipp Reisner cr |= test_bit(CONSIDER_RESYNC, &mdev->flags); 3225b411b363SPhilipp Reisner /* if we had been plain connected, and the admin requested to 3226b411b363SPhilipp Reisner * start a sync by "invalidate" or "invalidate-remote" */ 32274ac4aadaSLars Ellenberg cr |= (os.conn == C_CONNECTED && 3228b411b363SPhilipp Reisner (peer_state.conn >= C_STARTING_SYNC_S && 3229b411b363SPhilipp Reisner peer_state.conn <= C_WF_BITMAP_T)); 3230b411b363SPhilipp Reisner 3231b411b363SPhilipp Reisner if (cr) 32324ac4aadaSLars Ellenberg ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk); 3233b411b363SPhilipp Reisner 3234b411b363SPhilipp Reisner put_ldev(mdev); 32354ac4aadaSLars Ellenberg if (ns.conn == C_MASK) { 32364ac4aadaSLars Ellenberg ns.conn = C_CONNECTED; 3237b411b363SPhilipp Reisner if (mdev->state.disk == D_NEGOTIATING) { 323882f59cc6SLars Ellenberg drbd_force_state(mdev, NS(disk, D_FAILED)); 3239b411b363SPhilipp Reisner } else if (peer_state.disk == D_NEGOTIATING) { 3240b411b363SPhilipp Reisner dev_err(DEV, "Disk attach process on the peer node was aborted.\n"); 3241b411b363SPhilipp Reisner peer_state.disk = D_DISKLESS; 3242580b9767SLars Ellenberg real_peer_disk = D_DISKLESS; 3243b411b363SPhilipp Reisner } else { 3244cf14c2e9SPhilipp Reisner if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags)) 324581e84650SAndreas Gruenbacher return false; 32464ac4aadaSLars Ellenberg D_ASSERT(os.conn == C_WF_REPORT_PARAMS); 3247b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 324881e84650SAndreas Gruenbacher return false; 3249b411b363SPhilipp Reisner } 3250b411b363SPhilipp Reisner } 3251b411b363SPhilipp Reisner } 3252b411b363SPhilipp Reisner 3253b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 32544ac4aadaSLars Ellenberg if (mdev->state.i != os.i) 3255b411b363SPhilipp Reisner goto retry; 3256b411b363SPhilipp Reisner clear_bit(CONSIDER_RESYNC, &mdev->flags); 3257b411b363SPhilipp Reisner ns.peer = peer_state.role; 3258b411b363SPhilipp Reisner ns.pdsk = real_peer_disk; 3259b411b363SPhilipp Reisner ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); 32604ac4aadaSLars Ellenberg if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) 3261b411b363SPhilipp Reisner ns.disk = mdev->new_state_tmp.disk; 32624ac4aadaSLars Ellenberg cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); 32634ac4aadaSLars Ellenberg if (ns.pdsk == D_CONSISTENT && is_susp(ns) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && 3264481c6f50SPhilipp Reisner test_bit(NEW_CUR_UUID, &mdev->flags)) { 3265481c6f50SPhilipp Reisner /* Do not allow tl_restart(resend) for a rebooted peer. We can only allow this 3266481c6f50SPhilipp Reisner for temporal network outages! */ 3267481c6f50SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 3268481c6f50SPhilipp Reisner dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); 3269481c6f50SPhilipp Reisner tl_clear(mdev); 3270481c6f50SPhilipp Reisner drbd_uuid_new_current(mdev); 3271481c6f50SPhilipp Reisner clear_bit(NEW_CUR_UUID, &mdev->flags); 3272481c6f50SPhilipp Reisner drbd_force_state(mdev, NS2(conn, C_PROTOCOL_ERROR, susp, 0)); 327381e84650SAndreas Gruenbacher return false; 3274481c6f50SPhilipp Reisner } 327565d922c3SPhilipp Reisner rv = _drbd_set_state(mdev, ns, cs_flags, NULL); 3276b411b363SPhilipp Reisner ns = mdev->state; 3277b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 3278b411b363SPhilipp Reisner 3279b411b363SPhilipp Reisner if (rv < SS_SUCCESS) { 3280b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 328181e84650SAndreas Gruenbacher return false; 3282b411b363SPhilipp Reisner } 3283b411b363SPhilipp Reisner 32844ac4aadaSLars Ellenberg if (os.conn > C_WF_REPORT_PARAMS) { 32854ac4aadaSLars Ellenberg if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && 3286b411b363SPhilipp Reisner peer_state.disk != D_NEGOTIATING ) { 3287b411b363SPhilipp Reisner /* we want resync, peer has not yet decided to sync... */ 3288b411b363SPhilipp Reisner /* Nowadays only used when forcing a node into primary role and 3289b411b363SPhilipp Reisner setting its disk to UpToDate with that */ 3290b411b363SPhilipp Reisner drbd_send_uuids(mdev); 3291b411b363SPhilipp Reisner drbd_send_state(mdev); 3292b411b363SPhilipp Reisner } 3293b411b363SPhilipp Reisner } 3294b411b363SPhilipp Reisner 3295b411b363SPhilipp Reisner mdev->net_conf->want_lose = 0; 3296b411b363SPhilipp Reisner 3297b411b363SPhilipp Reisner drbd_md_sync(mdev); /* update connected indicator, la_size, ... */ 3298b411b363SPhilipp Reisner 329981e84650SAndreas Gruenbacher return true; 3300b411b363SPhilipp Reisner } 3301b411b363SPhilipp Reisner 330202918be2SPhilipp Reisner static int receive_sync_uuid(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3303b411b363SPhilipp Reisner { 330402918be2SPhilipp Reisner struct p_rs_uuid *p = &mdev->data.rbuf.rs_uuid; 3305b411b363SPhilipp Reisner 3306b411b363SPhilipp Reisner wait_event(mdev->misc_wait, 3307b411b363SPhilipp Reisner mdev->state.conn == C_WF_SYNC_UUID || 3308c4752ef1SPhilipp Reisner mdev->state.conn == C_BEHIND || 3309b411b363SPhilipp Reisner mdev->state.conn < C_CONNECTED || 3310b411b363SPhilipp Reisner mdev->state.disk < D_NEGOTIATING); 3311b411b363SPhilipp Reisner 3312b411b363SPhilipp Reisner /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */ 3313b411b363SPhilipp Reisner 3314b411b363SPhilipp Reisner /* Here the _drbd_uuid_ functions are right, current should 3315b411b363SPhilipp Reisner _not_ be rotated into the history */ 3316b411b363SPhilipp Reisner if (get_ldev_if_state(mdev, D_NEGOTIATING)) { 3317b411b363SPhilipp Reisner _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid)); 3318b411b363SPhilipp Reisner _drbd_uuid_set(mdev, UI_BITMAP, 0UL); 3319b411b363SPhilipp Reisner 332062b0da3aSLars Ellenberg drbd_print_uuids(mdev, "updated sync uuid"); 3321b411b363SPhilipp Reisner drbd_start_resync(mdev, C_SYNC_TARGET); 3322b411b363SPhilipp Reisner 3323b411b363SPhilipp Reisner put_ldev(mdev); 3324b411b363SPhilipp Reisner } else 3325b411b363SPhilipp Reisner dev_err(DEV, "Ignoring SyncUUID packet!\n"); 3326b411b363SPhilipp Reisner 332781e84650SAndreas Gruenbacher return true; 3328b411b363SPhilipp Reisner } 3329b411b363SPhilipp Reisner 33302c46407dSAndreas Gruenbacher /** 33312c46407dSAndreas Gruenbacher * receive_bitmap_plain 33322c46407dSAndreas Gruenbacher * 33332c46407dSAndreas Gruenbacher * Return 0 when done, 1 when another iteration is needed, and a negative error 33342c46407dSAndreas Gruenbacher * code upon failure. 33352c46407dSAndreas Gruenbacher */ 33362c46407dSAndreas Gruenbacher static int 333702918be2SPhilipp Reisner receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size, 3338b411b363SPhilipp Reisner unsigned long *buffer, struct bm_xfer_ctx *c) 3339b411b363SPhilipp Reisner { 3340b411b363SPhilipp Reisner unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset); 3341b411b363SPhilipp Reisner unsigned want = num_words * sizeof(long); 33422c46407dSAndreas Gruenbacher int err; 3343b411b363SPhilipp Reisner 334402918be2SPhilipp Reisner if (want != data_size) { 334502918be2SPhilipp Reisner dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size); 33462c46407dSAndreas Gruenbacher return -EIO; 3347b411b363SPhilipp Reisner } 3348b411b363SPhilipp Reisner if (want == 0) 33492c46407dSAndreas Gruenbacher return 0; 33502c46407dSAndreas Gruenbacher err = drbd_recv(mdev, buffer, want); 33512c46407dSAndreas Gruenbacher if (err != want) { 33522c46407dSAndreas Gruenbacher if (err >= 0) 33532c46407dSAndreas Gruenbacher err = -EIO; 33542c46407dSAndreas Gruenbacher return err; 33552c46407dSAndreas Gruenbacher } 3356b411b363SPhilipp Reisner 3357b411b363SPhilipp Reisner drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer); 3358b411b363SPhilipp Reisner 3359b411b363SPhilipp Reisner c->word_offset += num_words; 3360b411b363SPhilipp Reisner c->bit_offset = c->word_offset * BITS_PER_LONG; 3361b411b363SPhilipp Reisner if (c->bit_offset > c->bm_bits) 3362b411b363SPhilipp Reisner c->bit_offset = c->bm_bits; 3363b411b363SPhilipp Reisner 33642c46407dSAndreas Gruenbacher return 1; 3365b411b363SPhilipp Reisner } 3366b411b363SPhilipp Reisner 33672c46407dSAndreas Gruenbacher /** 33682c46407dSAndreas Gruenbacher * recv_bm_rle_bits 33692c46407dSAndreas Gruenbacher * 33702c46407dSAndreas Gruenbacher * Return 0 when done, 1 when another iteration is needed, and a negative error 33712c46407dSAndreas Gruenbacher * code upon failure. 33722c46407dSAndreas Gruenbacher */ 33732c46407dSAndreas Gruenbacher static int 3374b411b363SPhilipp Reisner recv_bm_rle_bits(struct drbd_conf *mdev, 3375b411b363SPhilipp Reisner struct p_compressed_bm *p, 3376b411b363SPhilipp Reisner struct bm_xfer_ctx *c) 3377b411b363SPhilipp Reisner { 3378b411b363SPhilipp Reisner struct bitstream bs; 3379b411b363SPhilipp Reisner u64 look_ahead; 3380b411b363SPhilipp Reisner u64 rl; 3381b411b363SPhilipp Reisner u64 tmp; 3382b411b363SPhilipp Reisner unsigned long s = c->bit_offset; 3383b411b363SPhilipp Reisner unsigned long e; 3384004352faSLars Ellenberg int len = be16_to_cpu(p->head.length) - (sizeof(*p) - sizeof(p->head)); 3385b411b363SPhilipp Reisner int toggle = DCBP_get_start(p); 3386b411b363SPhilipp Reisner int have; 3387b411b363SPhilipp Reisner int bits; 3388b411b363SPhilipp Reisner 3389b411b363SPhilipp Reisner bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p)); 3390b411b363SPhilipp Reisner 3391b411b363SPhilipp Reisner bits = bitstream_get_bits(&bs, &look_ahead, 64); 3392b411b363SPhilipp Reisner if (bits < 0) 33932c46407dSAndreas Gruenbacher return -EIO; 3394b411b363SPhilipp Reisner 3395b411b363SPhilipp Reisner for (have = bits; have > 0; s += rl, toggle = !toggle) { 3396b411b363SPhilipp Reisner bits = vli_decode_bits(&rl, look_ahead); 3397b411b363SPhilipp Reisner if (bits <= 0) 33982c46407dSAndreas Gruenbacher return -EIO; 3399b411b363SPhilipp Reisner 3400b411b363SPhilipp Reisner if (toggle) { 3401b411b363SPhilipp Reisner e = s + rl -1; 3402b411b363SPhilipp Reisner if (e >= c->bm_bits) { 3403b411b363SPhilipp Reisner dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); 34042c46407dSAndreas Gruenbacher return -EIO; 3405b411b363SPhilipp Reisner } 3406b411b363SPhilipp Reisner _drbd_bm_set_bits(mdev, s, e); 3407b411b363SPhilipp Reisner } 3408b411b363SPhilipp Reisner 3409b411b363SPhilipp Reisner if (have < bits) { 3410b411b363SPhilipp Reisner dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", 3411b411b363SPhilipp Reisner have, bits, look_ahead, 3412b411b363SPhilipp Reisner (unsigned int)(bs.cur.b - p->code), 3413b411b363SPhilipp Reisner (unsigned int)bs.buf_len); 34142c46407dSAndreas Gruenbacher return -EIO; 3415b411b363SPhilipp Reisner } 3416b411b363SPhilipp Reisner look_ahead >>= bits; 3417b411b363SPhilipp Reisner have -= bits; 3418b411b363SPhilipp Reisner 3419b411b363SPhilipp Reisner bits = bitstream_get_bits(&bs, &tmp, 64 - have); 3420b411b363SPhilipp Reisner if (bits < 0) 34212c46407dSAndreas Gruenbacher return -EIO; 3422b411b363SPhilipp Reisner look_ahead |= tmp << have; 3423b411b363SPhilipp Reisner have += bits; 3424b411b363SPhilipp Reisner } 3425b411b363SPhilipp Reisner 3426b411b363SPhilipp Reisner c->bit_offset = s; 3427b411b363SPhilipp Reisner bm_xfer_ctx_bit_to_word_offset(c); 3428b411b363SPhilipp Reisner 34292c46407dSAndreas Gruenbacher return (s != c->bm_bits); 3430b411b363SPhilipp Reisner } 3431b411b363SPhilipp Reisner 34322c46407dSAndreas Gruenbacher /** 34332c46407dSAndreas Gruenbacher * decode_bitmap_c 34342c46407dSAndreas Gruenbacher * 34352c46407dSAndreas Gruenbacher * Return 0 when done, 1 when another iteration is needed, and a negative error 34362c46407dSAndreas Gruenbacher * code upon failure. 34372c46407dSAndreas Gruenbacher */ 34382c46407dSAndreas Gruenbacher static int 3439b411b363SPhilipp Reisner decode_bitmap_c(struct drbd_conf *mdev, 3440b411b363SPhilipp Reisner struct p_compressed_bm *p, 3441b411b363SPhilipp Reisner struct bm_xfer_ctx *c) 3442b411b363SPhilipp Reisner { 3443b411b363SPhilipp Reisner if (DCBP_get_code(p) == RLE_VLI_Bits) 3444b411b363SPhilipp Reisner return recv_bm_rle_bits(mdev, p, c); 3445b411b363SPhilipp Reisner 3446b411b363SPhilipp Reisner /* other variants had been implemented for evaluation, 3447b411b363SPhilipp Reisner * but have been dropped as this one turned out to be "best" 3448b411b363SPhilipp Reisner * during all our tests. */ 3449b411b363SPhilipp Reisner 3450b411b363SPhilipp Reisner dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding); 3451b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); 34522c46407dSAndreas Gruenbacher return -EIO; 3453b411b363SPhilipp Reisner } 3454b411b363SPhilipp Reisner 3455b411b363SPhilipp Reisner void INFO_bm_xfer_stats(struct drbd_conf *mdev, 3456b411b363SPhilipp Reisner const char *direction, struct bm_xfer_ctx *c) 3457b411b363SPhilipp Reisner { 3458b411b363SPhilipp Reisner /* what would it take to transfer it "plaintext" */ 34590b70a13dSPhilipp Reisner unsigned plain = sizeof(struct p_header80) * 3460b411b363SPhilipp Reisner ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1) 3461b411b363SPhilipp Reisner + c->bm_words * sizeof(long); 3462b411b363SPhilipp Reisner unsigned total = c->bytes[0] + c->bytes[1]; 3463b411b363SPhilipp Reisner unsigned r; 3464b411b363SPhilipp Reisner 3465b411b363SPhilipp Reisner /* total can not be zero. but just in case: */ 3466b411b363SPhilipp Reisner if (total == 0) 3467b411b363SPhilipp Reisner return; 3468b411b363SPhilipp Reisner 3469b411b363SPhilipp Reisner /* don't report if not compressed */ 3470b411b363SPhilipp Reisner if (total >= plain) 3471b411b363SPhilipp Reisner return; 3472b411b363SPhilipp Reisner 3473b411b363SPhilipp Reisner /* total < plain. check for overflow, still */ 3474b411b363SPhilipp Reisner r = (total > UINT_MAX/1000) ? (total / (plain/1000)) 3475b411b363SPhilipp Reisner : (1000 * total / plain); 3476b411b363SPhilipp Reisner 3477b411b363SPhilipp Reisner if (r > 1000) 3478b411b363SPhilipp Reisner r = 1000; 3479b411b363SPhilipp Reisner 3480b411b363SPhilipp Reisner r = 1000 - r; 3481b411b363SPhilipp Reisner dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " 3482b411b363SPhilipp Reisner "total %u; compression: %u.%u%%\n", 3483b411b363SPhilipp Reisner direction, 3484b411b363SPhilipp Reisner c->bytes[1], c->packets[1], 3485b411b363SPhilipp Reisner c->bytes[0], c->packets[0], 3486b411b363SPhilipp Reisner total, r/10, r % 10); 3487b411b363SPhilipp Reisner } 3488b411b363SPhilipp Reisner 3489b411b363SPhilipp Reisner /* Since we are processing the bitfield from lower addresses to higher, 3490b411b363SPhilipp Reisner it does not matter if the process it in 32 bit chunks or 64 bit 3491b411b363SPhilipp Reisner chunks as long as it is little endian. (Understand it as byte stream, 3492b411b363SPhilipp Reisner beginning with the lowest byte...) If we would use big endian 3493b411b363SPhilipp Reisner we would need to process it from the highest address to the lowest, 3494b411b363SPhilipp Reisner in order to be agnostic to the 32 vs 64 bits issue. 3495b411b363SPhilipp Reisner 3496b411b363SPhilipp Reisner returns 0 on failure, 1 if we successfully received it. */ 349702918be2SPhilipp Reisner static int receive_bitmap(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3498b411b363SPhilipp Reisner { 3499b411b363SPhilipp Reisner struct bm_xfer_ctx c; 3500b411b363SPhilipp Reisner void *buffer; 35012c46407dSAndreas Gruenbacher int err; 350281e84650SAndreas Gruenbacher int ok = false; 350302918be2SPhilipp Reisner struct p_header80 *h = &mdev->data.rbuf.header.h80; 3504b411b363SPhilipp Reisner 350520ceb2b2SLars Ellenberg drbd_bm_lock(mdev, "receive bitmap", BM_LOCKED_SET_ALLOWED); 350620ceb2b2SLars Ellenberg /* you are supposed to send additional out-of-sync information 350720ceb2b2SLars Ellenberg * if you actually set bits during this phase */ 3508b411b363SPhilipp Reisner 3509b411b363SPhilipp Reisner /* maybe we should use some per thread scratch page, 3510b411b363SPhilipp Reisner * and allocate that during initial device creation? */ 3511b411b363SPhilipp Reisner buffer = (unsigned long *) __get_free_page(GFP_NOIO); 3512b411b363SPhilipp Reisner if (!buffer) { 3513b411b363SPhilipp Reisner dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__); 3514b411b363SPhilipp Reisner goto out; 3515b411b363SPhilipp Reisner } 3516b411b363SPhilipp Reisner 3517b411b363SPhilipp Reisner c = (struct bm_xfer_ctx) { 3518b411b363SPhilipp Reisner .bm_bits = drbd_bm_bits(mdev), 3519b411b363SPhilipp Reisner .bm_words = drbd_bm_words(mdev), 3520b411b363SPhilipp Reisner }; 3521b411b363SPhilipp Reisner 35222c46407dSAndreas Gruenbacher for(;;) { 352302918be2SPhilipp Reisner if (cmd == P_BITMAP) { 35242c46407dSAndreas Gruenbacher err = receive_bitmap_plain(mdev, data_size, buffer, &c); 352502918be2SPhilipp Reisner } else if (cmd == P_COMPRESSED_BITMAP) { 3526b411b363SPhilipp Reisner /* MAYBE: sanity check that we speak proto >= 90, 3527b411b363SPhilipp Reisner * and the feature is enabled! */ 3528b411b363SPhilipp Reisner struct p_compressed_bm *p; 3529b411b363SPhilipp Reisner 353002918be2SPhilipp Reisner if (data_size > BM_PACKET_PAYLOAD_BYTES) { 3531b411b363SPhilipp Reisner dev_err(DEV, "ReportCBitmap packet too large\n"); 3532b411b363SPhilipp Reisner goto out; 3533b411b363SPhilipp Reisner } 3534b411b363SPhilipp Reisner /* use the page buff */ 3535b411b363SPhilipp Reisner p = buffer; 3536b411b363SPhilipp Reisner memcpy(p, h, sizeof(*h)); 353702918be2SPhilipp Reisner if (drbd_recv(mdev, p->head.payload, data_size) != data_size) 3538b411b363SPhilipp Reisner goto out; 3539004352faSLars Ellenberg if (data_size <= (sizeof(*p) - sizeof(p->head))) { 3540004352faSLars Ellenberg dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", data_size); 354178fcbdaeSAndreas Gruenbacher goto out; 3542b411b363SPhilipp Reisner } 35432c46407dSAndreas Gruenbacher err = decode_bitmap_c(mdev, p, &c); 3544b411b363SPhilipp Reisner } else { 354502918be2SPhilipp Reisner dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", cmd); 3546b411b363SPhilipp Reisner goto out; 3547b411b363SPhilipp Reisner } 3548b411b363SPhilipp Reisner 354902918be2SPhilipp Reisner c.packets[cmd == P_BITMAP]++; 355002918be2SPhilipp Reisner c.bytes[cmd == P_BITMAP] += sizeof(struct p_header80) + data_size; 3551b411b363SPhilipp Reisner 35522c46407dSAndreas Gruenbacher if (err <= 0) { 35532c46407dSAndreas Gruenbacher if (err < 0) 35542c46407dSAndreas Gruenbacher goto out; 3555b411b363SPhilipp Reisner break; 35562c46407dSAndreas Gruenbacher } 355702918be2SPhilipp Reisner if (!drbd_recv_header(mdev, &cmd, &data_size)) 3558b411b363SPhilipp Reisner goto out; 35592c46407dSAndreas Gruenbacher } 3560b411b363SPhilipp Reisner 3561b411b363SPhilipp Reisner INFO_bm_xfer_stats(mdev, "receive", &c); 3562b411b363SPhilipp Reisner 3563b411b363SPhilipp Reisner if (mdev->state.conn == C_WF_BITMAP_T) { 3564de1f8e4aSAndreas Gruenbacher enum drbd_state_rv rv; 3565de1f8e4aSAndreas Gruenbacher 3566b411b363SPhilipp Reisner ok = !drbd_send_bitmap(mdev); 3567b411b363SPhilipp Reisner if (!ok) 3568b411b363SPhilipp Reisner goto out; 3569b411b363SPhilipp Reisner /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ 3570de1f8e4aSAndreas Gruenbacher rv = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); 3571de1f8e4aSAndreas Gruenbacher D_ASSERT(rv == SS_SUCCESS); 3572b411b363SPhilipp Reisner } else if (mdev->state.conn != C_WF_BITMAP_S) { 3573b411b363SPhilipp Reisner /* admin may have requested C_DISCONNECTING, 3574b411b363SPhilipp Reisner * other threads may have noticed network errors */ 3575b411b363SPhilipp Reisner dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n", 3576b411b363SPhilipp Reisner drbd_conn_str(mdev->state.conn)); 3577b411b363SPhilipp Reisner } 3578b411b363SPhilipp Reisner 357981e84650SAndreas Gruenbacher ok = true; 3580b411b363SPhilipp Reisner out: 358120ceb2b2SLars Ellenberg drbd_bm_unlock(mdev); 3582b411b363SPhilipp Reisner if (ok && mdev->state.conn == C_WF_BITMAP_S) 3583b411b363SPhilipp Reisner drbd_start_resync(mdev, C_SYNC_SOURCE); 3584b411b363SPhilipp Reisner free_page((unsigned long) buffer); 3585b411b363SPhilipp Reisner return ok; 3586b411b363SPhilipp Reisner } 3587b411b363SPhilipp Reisner 358802918be2SPhilipp Reisner static int receive_skip(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3589b411b363SPhilipp Reisner { 3590b411b363SPhilipp Reisner /* TODO zero copy sink :) */ 3591b411b363SPhilipp Reisner static char sink[128]; 3592b411b363SPhilipp Reisner int size, want, r; 3593b411b363SPhilipp Reisner 3594b411b363SPhilipp Reisner dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n", 359502918be2SPhilipp Reisner cmd, data_size); 3596b411b363SPhilipp Reisner 359702918be2SPhilipp Reisner size = data_size; 3598b411b363SPhilipp Reisner while (size > 0) { 3599b411b363SPhilipp Reisner want = min_t(int, size, sizeof(sink)); 3600b411b363SPhilipp Reisner r = drbd_recv(mdev, sink, want); 3601b411b363SPhilipp Reisner ERR_IF(r <= 0) break; 3602b411b363SPhilipp Reisner size -= r; 3603b411b363SPhilipp Reisner } 3604b411b363SPhilipp Reisner return size == 0; 3605b411b363SPhilipp Reisner } 3606b411b363SPhilipp Reisner 360702918be2SPhilipp Reisner static int receive_UnplugRemote(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 3608b411b363SPhilipp Reisner { 3609b411b363SPhilipp Reisner /* Make sure we've acked all the TCP data associated 3610b411b363SPhilipp Reisner * with the data requests being unplugged */ 3611b411b363SPhilipp Reisner drbd_tcp_quickack(mdev->data.socket); 3612b411b363SPhilipp Reisner 361381e84650SAndreas Gruenbacher return true; 3614b411b363SPhilipp Reisner } 3615b411b363SPhilipp Reisner 361673a01a18SPhilipp Reisner static int receive_out_of_sync(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size) 361773a01a18SPhilipp Reisner { 361873a01a18SPhilipp Reisner struct p_block_desc *p = &mdev->data.rbuf.block_desc; 361973a01a18SPhilipp Reisner 3620f735e363SLars Ellenberg switch (mdev->state.conn) { 3621f735e363SLars Ellenberg case C_WF_SYNC_UUID: 3622f735e363SLars Ellenberg case C_WF_BITMAP_T: 3623f735e363SLars Ellenberg case C_BEHIND: 3624f735e363SLars Ellenberg break; 3625f735e363SLars Ellenberg default: 3626f735e363SLars Ellenberg dev_err(DEV, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", 3627f735e363SLars Ellenberg drbd_conn_str(mdev->state.conn)); 3628f735e363SLars Ellenberg } 3629f735e363SLars Ellenberg 363073a01a18SPhilipp Reisner drbd_set_out_of_sync(mdev, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); 363173a01a18SPhilipp Reisner 363281e84650SAndreas Gruenbacher return true; 363373a01a18SPhilipp Reisner } 363473a01a18SPhilipp Reisner 363502918be2SPhilipp Reisner typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, enum drbd_packets cmd, unsigned int to_receive); 3636b411b363SPhilipp Reisner 363702918be2SPhilipp Reisner struct data_cmd { 363802918be2SPhilipp Reisner int expect_payload; 363902918be2SPhilipp Reisner size_t pkt_size; 364002918be2SPhilipp Reisner drbd_cmd_handler_f function; 3641b411b363SPhilipp Reisner }; 3642b411b363SPhilipp Reisner 364302918be2SPhilipp Reisner static struct data_cmd drbd_cmd_handler[] = { 364402918be2SPhilipp Reisner [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, 364502918be2SPhilipp Reisner [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, 364602918be2SPhilipp Reisner [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , 364702918be2SPhilipp Reisner [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , 364802918be2SPhilipp Reisner [P_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } , 364902918be2SPhilipp Reisner [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } , 365002918be2SPhilipp Reisner [P_UNPLUG_REMOTE] = { 0, sizeof(struct p_header80), receive_UnplugRemote }, 365102918be2SPhilipp Reisner [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, 365202918be2SPhilipp Reisner [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, 365302918be2SPhilipp Reisner [P_SYNC_PARAM] = { 1, sizeof(struct p_header80), receive_SyncParam }, 365402918be2SPhilipp Reisner [P_SYNC_PARAM89] = { 1, sizeof(struct p_header80), receive_SyncParam }, 365502918be2SPhilipp Reisner [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, 365602918be2SPhilipp Reisner [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, 365702918be2SPhilipp Reisner [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, 365802918be2SPhilipp Reisner [P_STATE] = { 0, sizeof(struct p_state), receive_state }, 365902918be2SPhilipp Reisner [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, 366002918be2SPhilipp Reisner [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, 366102918be2SPhilipp Reisner [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, 366202918be2SPhilipp Reisner [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, 366302918be2SPhilipp Reisner [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, 366402918be2SPhilipp Reisner [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, 366573a01a18SPhilipp Reisner [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync }, 366602918be2SPhilipp Reisner /* anything missing from this table is in 366702918be2SPhilipp Reisner * the asender_tbl, see get_asender_cmd */ 366802918be2SPhilipp Reisner [P_MAX_CMD] = { 0, 0, NULL }, 366902918be2SPhilipp Reisner }; 367002918be2SPhilipp Reisner 367102918be2SPhilipp Reisner /* All handler functions that expect a sub-header get that sub-heder in 367202918be2SPhilipp Reisner mdev->data.rbuf.header.head.payload. 367302918be2SPhilipp Reisner 367402918be2SPhilipp Reisner Usually in mdev->data.rbuf.header.head the callback can find the usual 367502918be2SPhilipp Reisner p_header, but they may not rely on that. Since there is also p_header95 ! 367602918be2SPhilipp Reisner */ 3677b411b363SPhilipp Reisner 3678b411b363SPhilipp Reisner static void drbdd(struct drbd_conf *mdev) 3679b411b363SPhilipp Reisner { 368002918be2SPhilipp Reisner union p_header *header = &mdev->data.rbuf.header; 368102918be2SPhilipp Reisner unsigned int packet_size; 368202918be2SPhilipp Reisner enum drbd_packets cmd; 368302918be2SPhilipp Reisner size_t shs; /* sub header size */ 368402918be2SPhilipp Reisner int rv; 3685b411b363SPhilipp Reisner 3686b411b363SPhilipp Reisner while (get_t_state(&mdev->receiver) == Running) { 3687b411b363SPhilipp Reisner drbd_thread_current_set_cpu(mdev); 368802918be2SPhilipp Reisner if (!drbd_recv_header(mdev, &cmd, &packet_size)) 368902918be2SPhilipp Reisner goto err_out; 369002918be2SPhilipp Reisner 369102918be2SPhilipp Reisner if (unlikely(cmd >= P_MAX_CMD || !drbd_cmd_handler[cmd].function)) { 369202918be2SPhilipp Reisner dev_err(DEV, "unknown packet type %d, l: %d!\n", cmd, packet_size); 369302918be2SPhilipp Reisner goto err_out; 36940b33a916SLars Ellenberg } 3695b411b363SPhilipp Reisner 369602918be2SPhilipp Reisner shs = drbd_cmd_handler[cmd].pkt_size - sizeof(union p_header); 3697c13f7e1aSLars Ellenberg if (packet_size - shs > 0 && !drbd_cmd_handler[cmd].expect_payload) { 3698c13f7e1aSLars Ellenberg dev_err(DEV, "No payload expected %s l:%d\n", cmdname(cmd), packet_size); 3699c13f7e1aSLars Ellenberg goto err_out; 3700c13f7e1aSLars Ellenberg } 3701c13f7e1aSLars Ellenberg 3702c13f7e1aSLars Ellenberg if (shs) { 370302918be2SPhilipp Reisner rv = drbd_recv(mdev, &header->h80.payload, shs); 370402918be2SPhilipp Reisner if (unlikely(rv != shs)) { 37050ddc5549SLars Ellenberg if (!signal_pending(current)) 37060ddc5549SLars Ellenberg dev_warn(DEV, "short read while reading sub header: rv=%d\n", rv); 370702918be2SPhilipp Reisner goto err_out; 3708b411b363SPhilipp Reisner } 370902918be2SPhilipp Reisner } 371002918be2SPhilipp Reisner 371102918be2SPhilipp Reisner rv = drbd_cmd_handler[cmd].function(mdev, cmd, packet_size - shs); 371202918be2SPhilipp Reisner 371302918be2SPhilipp Reisner if (unlikely(!rv)) { 3714b411b363SPhilipp Reisner dev_err(DEV, "error receiving %s, l: %d!\n", 371502918be2SPhilipp Reisner cmdname(cmd), packet_size); 371602918be2SPhilipp Reisner goto err_out; 3717b411b363SPhilipp Reisner } 3718b411b363SPhilipp Reisner } 371902918be2SPhilipp Reisner 372002918be2SPhilipp Reisner if (0) { 372102918be2SPhilipp Reisner err_out: 3722b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); 3723b411b363SPhilipp Reisner } 3724856c50c7SLars Ellenberg /* If we leave here, we probably want to update at least the 3725856c50c7SLars Ellenberg * "Connected" indicator on stable storage. Do so explicitly here. */ 3726856c50c7SLars Ellenberg drbd_md_sync(mdev); 3727b411b363SPhilipp Reisner } 3728b411b363SPhilipp Reisner 3729b411b363SPhilipp Reisner void drbd_flush_workqueue(struct drbd_conf *mdev) 3730b411b363SPhilipp Reisner { 3731b411b363SPhilipp Reisner struct drbd_wq_barrier barr; 3732b411b363SPhilipp Reisner 3733b411b363SPhilipp Reisner barr.w.cb = w_prev_work_done; 3734b411b363SPhilipp Reisner init_completion(&barr.done); 3735b411b363SPhilipp Reisner drbd_queue_work(&mdev->data.work, &barr.w); 3736b411b363SPhilipp Reisner wait_for_completion(&barr.done); 3737b411b363SPhilipp Reisner } 3738b411b363SPhilipp Reisner 3739f70b3511SPhilipp Reisner void drbd_free_tl_hash(struct drbd_conf *mdev) 3740f70b3511SPhilipp Reisner { 3741f70b3511SPhilipp Reisner struct hlist_head *h; 3742f70b3511SPhilipp Reisner 3743f70b3511SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 3744f70b3511SPhilipp Reisner 3745f70b3511SPhilipp Reisner if (!mdev->tl_hash || mdev->state.conn != C_STANDALONE) { 3746f70b3511SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 3747f70b3511SPhilipp Reisner return; 3748f70b3511SPhilipp Reisner } 3749f70b3511SPhilipp Reisner /* paranoia code */ 3750f70b3511SPhilipp Reisner for (h = mdev->ee_hash; h < mdev->ee_hash + mdev->ee_hash_s; h++) 3751f70b3511SPhilipp Reisner if (h->first) 3752f70b3511SPhilipp Reisner dev_err(DEV, "ASSERT FAILED ee_hash[%u].first == %p, expected NULL\n", 3753f70b3511SPhilipp Reisner (int)(h - mdev->ee_hash), h->first); 3754f70b3511SPhilipp Reisner kfree(mdev->ee_hash); 3755f70b3511SPhilipp Reisner mdev->ee_hash = NULL; 3756f70b3511SPhilipp Reisner mdev->ee_hash_s = 0; 3757f70b3511SPhilipp Reisner 3758f70b3511SPhilipp Reisner /* paranoia code */ 3759f70b3511SPhilipp Reisner for (h = mdev->tl_hash; h < mdev->tl_hash + mdev->tl_hash_s; h++) 3760f70b3511SPhilipp Reisner if (h->first) 3761f70b3511SPhilipp Reisner dev_err(DEV, "ASSERT FAILED tl_hash[%u] == %p, expected NULL\n", 3762f70b3511SPhilipp Reisner (int)(h - mdev->tl_hash), h->first); 3763f70b3511SPhilipp Reisner kfree(mdev->tl_hash); 3764f70b3511SPhilipp Reisner mdev->tl_hash = NULL; 3765f70b3511SPhilipp Reisner mdev->tl_hash_s = 0; 3766f70b3511SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 3767f70b3511SPhilipp Reisner } 3768f70b3511SPhilipp Reisner 3769b411b363SPhilipp Reisner static void drbd_disconnect(struct drbd_conf *mdev) 3770b411b363SPhilipp Reisner { 3771b411b363SPhilipp Reisner enum drbd_fencing_p fp; 3772b411b363SPhilipp Reisner union drbd_state os, ns; 3773b411b363SPhilipp Reisner int rv = SS_UNKNOWN_ERROR; 3774b411b363SPhilipp Reisner unsigned int i; 3775b411b363SPhilipp Reisner 3776b411b363SPhilipp Reisner if (mdev->state.conn == C_STANDALONE) 3777b411b363SPhilipp Reisner return; 3778b411b363SPhilipp Reisner 3779b411b363SPhilipp Reisner /* asender does not clean up anything. it must not interfere, either */ 3780b411b363SPhilipp Reisner drbd_thread_stop(&mdev->asender); 3781b411b363SPhilipp Reisner drbd_free_sock(mdev); 3782b411b363SPhilipp Reisner 378385719573SPhilipp Reisner /* wait for current activity to cease. */ 3784b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 3785b411b363SPhilipp Reisner _drbd_wait_ee_list_empty(mdev, &mdev->active_ee); 3786b411b363SPhilipp Reisner _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee); 3787b411b363SPhilipp Reisner _drbd_wait_ee_list_empty(mdev, &mdev->read_ee); 3788b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 3789b411b363SPhilipp Reisner 3790b411b363SPhilipp Reisner /* We do not have data structures that would allow us to 3791b411b363SPhilipp Reisner * get the rs_pending_cnt down to 0 again. 3792b411b363SPhilipp Reisner * * On C_SYNC_TARGET we do not have any data structures describing 3793b411b363SPhilipp Reisner * the pending RSDataRequest's we have sent. 3794b411b363SPhilipp Reisner * * On C_SYNC_SOURCE there is no data structure that tracks 3795b411b363SPhilipp Reisner * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. 3796b411b363SPhilipp Reisner * And no, it is not the sum of the reference counts in the 3797b411b363SPhilipp Reisner * resync_LRU. The resync_LRU tracks the whole operation including 3798b411b363SPhilipp Reisner * the disk-IO, while the rs_pending_cnt only tracks the blocks 3799b411b363SPhilipp Reisner * on the fly. */ 3800b411b363SPhilipp Reisner drbd_rs_cancel_all(mdev); 3801b411b363SPhilipp Reisner mdev->rs_total = 0; 3802b411b363SPhilipp Reisner mdev->rs_failed = 0; 3803b411b363SPhilipp Reisner atomic_set(&mdev->rs_pending_cnt, 0); 3804b411b363SPhilipp Reisner wake_up(&mdev->misc_wait); 3805b411b363SPhilipp Reisner 38067fde2be9SPhilipp Reisner del_timer(&mdev->request_timer); 38077fde2be9SPhilipp Reisner 3808b411b363SPhilipp Reisner /* make sure syncer is stopped and w_resume_next_sg queued */ 3809b411b363SPhilipp Reisner del_timer_sync(&mdev->resync_timer); 3810b411b363SPhilipp Reisner resync_timer_fn((unsigned long)mdev); 3811b411b363SPhilipp Reisner 3812b411b363SPhilipp Reisner /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, 3813b411b363SPhilipp Reisner * w_make_resync_request etc. which may still be on the worker queue 3814b411b363SPhilipp Reisner * to be "canceled" */ 3815b411b363SPhilipp Reisner drbd_flush_workqueue(mdev); 3816b411b363SPhilipp Reisner 3817b411b363SPhilipp Reisner /* This also does reclaim_net_ee(). If we do this too early, we might 3818b411b363SPhilipp Reisner * miss some resync ee and pages.*/ 3819b411b363SPhilipp Reisner drbd_process_done_ee(mdev); 3820b411b363SPhilipp Reisner 3821b411b363SPhilipp Reisner kfree(mdev->p_uuid); 3822b411b363SPhilipp Reisner mdev->p_uuid = NULL; 3823b411b363SPhilipp Reisner 3824fb22c402SPhilipp Reisner if (!is_susp(mdev->state)) 3825b411b363SPhilipp Reisner tl_clear(mdev); 3826b411b363SPhilipp Reisner 3827b411b363SPhilipp Reisner dev_info(DEV, "Connection closed\n"); 3828b411b363SPhilipp Reisner 3829b411b363SPhilipp Reisner drbd_md_sync(mdev); 3830b411b363SPhilipp Reisner 3831b411b363SPhilipp Reisner fp = FP_DONT_CARE; 3832b411b363SPhilipp Reisner if (get_ldev(mdev)) { 3833b411b363SPhilipp Reisner fp = mdev->ldev->dc.fencing; 3834b411b363SPhilipp Reisner put_ldev(mdev); 3835b411b363SPhilipp Reisner } 3836b411b363SPhilipp Reisner 383787f7be4cSPhilipp Reisner if (mdev->state.role == R_PRIMARY && fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN) 383887f7be4cSPhilipp Reisner drbd_try_outdate_peer_async(mdev); 3839b411b363SPhilipp Reisner 3840b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 3841b411b363SPhilipp Reisner os = mdev->state; 3842b411b363SPhilipp Reisner if (os.conn >= C_UNCONNECTED) { 3843b411b363SPhilipp Reisner /* Do not restart in case we are C_DISCONNECTING */ 3844b411b363SPhilipp Reisner ns = os; 3845b411b363SPhilipp Reisner ns.conn = C_UNCONNECTED; 3846b411b363SPhilipp Reisner rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); 3847b411b363SPhilipp Reisner } 3848b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 3849b411b363SPhilipp Reisner 3850b411b363SPhilipp Reisner if (os.conn == C_DISCONNECTING) { 385184dfb9f5SPhilipp Reisner wait_event(mdev->net_cnt_wait, atomic_read(&mdev->net_cnt) == 0); 3852b411b363SPhilipp Reisner 3853b411b363SPhilipp Reisner crypto_free_hash(mdev->cram_hmac_tfm); 3854b411b363SPhilipp Reisner mdev->cram_hmac_tfm = NULL; 3855b411b363SPhilipp Reisner 3856b411b363SPhilipp Reisner kfree(mdev->net_conf); 3857b411b363SPhilipp Reisner mdev->net_conf = NULL; 3858b411b363SPhilipp Reisner drbd_request_state(mdev, NS(conn, C_STANDALONE)); 3859b411b363SPhilipp Reisner } 3860b411b363SPhilipp Reisner 386120ceb2b2SLars Ellenberg /* serialize with bitmap writeout triggered by the state change, 386220ceb2b2SLars Ellenberg * if any. */ 386320ceb2b2SLars Ellenberg wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags)); 386420ceb2b2SLars Ellenberg 3865b411b363SPhilipp Reisner /* tcp_close and release of sendpage pages can be deferred. I don't 3866b411b363SPhilipp Reisner * want to use SO_LINGER, because apparently it can be deferred for 3867b411b363SPhilipp Reisner * more than 20 seconds (longest time I checked). 3868b411b363SPhilipp Reisner * 3869b411b363SPhilipp Reisner * Actually we don't care for exactly when the network stack does its 3870b411b363SPhilipp Reisner * put_page(), but release our reference on these pages right here. 3871b411b363SPhilipp Reisner */ 3872b411b363SPhilipp Reisner i = drbd_release_ee(mdev, &mdev->net_ee); 3873b411b363SPhilipp Reisner if (i) 3874b411b363SPhilipp Reisner dev_info(DEV, "net_ee not empty, killed %u entries\n", i); 3875435f0740SLars Ellenberg i = atomic_read(&mdev->pp_in_use_by_net); 3876435f0740SLars Ellenberg if (i) 3877435f0740SLars Ellenberg dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i); 3878b411b363SPhilipp Reisner i = atomic_read(&mdev->pp_in_use); 3879b411b363SPhilipp Reisner if (i) 388045bb912bSLars Ellenberg dev_info(DEV, "pp_in_use = %d, expected 0\n", i); 3881b411b363SPhilipp Reisner 3882b411b363SPhilipp Reisner D_ASSERT(list_empty(&mdev->read_ee)); 3883b411b363SPhilipp Reisner D_ASSERT(list_empty(&mdev->active_ee)); 3884b411b363SPhilipp Reisner D_ASSERT(list_empty(&mdev->sync_ee)); 3885b411b363SPhilipp Reisner D_ASSERT(list_empty(&mdev->done_ee)); 3886b411b363SPhilipp Reisner 3887b411b363SPhilipp Reisner /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ 3888b411b363SPhilipp Reisner atomic_set(&mdev->current_epoch->epoch_size, 0); 3889b411b363SPhilipp Reisner D_ASSERT(list_empty(&mdev->current_epoch->list)); 3890b411b363SPhilipp Reisner } 3891b411b363SPhilipp Reisner 3892b411b363SPhilipp Reisner /* 3893b411b363SPhilipp Reisner * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version 3894b411b363SPhilipp Reisner * we can agree on is stored in agreed_pro_version. 3895b411b363SPhilipp Reisner * 3896b411b363SPhilipp Reisner * feature flags and the reserved array should be enough room for future 3897b411b363SPhilipp Reisner * enhancements of the handshake protocol, and possible plugins... 3898b411b363SPhilipp Reisner * 3899b411b363SPhilipp Reisner * for now, they are expected to be zero, but ignored. 3900b411b363SPhilipp Reisner */ 3901b411b363SPhilipp Reisner static int drbd_send_handshake(struct drbd_conf *mdev) 3902b411b363SPhilipp Reisner { 3903b411b363SPhilipp Reisner /* ASSERT current == mdev->receiver ... */ 3904b411b363SPhilipp Reisner struct p_handshake *p = &mdev->data.sbuf.handshake; 3905b411b363SPhilipp Reisner int ok; 3906b411b363SPhilipp Reisner 3907b411b363SPhilipp Reisner if (mutex_lock_interruptible(&mdev->data.mutex)) { 3908b411b363SPhilipp Reisner dev_err(DEV, "interrupted during initial handshake\n"); 3909b411b363SPhilipp Reisner return 0; /* interrupted. not ok. */ 3910b411b363SPhilipp Reisner } 3911b411b363SPhilipp Reisner 3912b411b363SPhilipp Reisner if (mdev->data.socket == NULL) { 3913b411b363SPhilipp Reisner mutex_unlock(&mdev->data.mutex); 3914b411b363SPhilipp Reisner return 0; 3915b411b363SPhilipp Reisner } 3916b411b363SPhilipp Reisner 3917b411b363SPhilipp Reisner memset(p, 0, sizeof(*p)); 3918b411b363SPhilipp Reisner p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); 3919b411b363SPhilipp Reisner p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); 3920b411b363SPhilipp Reisner ok = _drbd_send_cmd( mdev, mdev->data.socket, P_HAND_SHAKE, 39210b70a13dSPhilipp Reisner (struct p_header80 *)p, sizeof(*p), 0 ); 3922b411b363SPhilipp Reisner mutex_unlock(&mdev->data.mutex); 3923b411b363SPhilipp Reisner return ok; 3924b411b363SPhilipp Reisner } 3925b411b363SPhilipp Reisner 3926b411b363SPhilipp Reisner /* 3927b411b363SPhilipp Reisner * return values: 3928b411b363SPhilipp Reisner * 1 yes, we have a valid connection 3929b411b363SPhilipp Reisner * 0 oops, did not work out, please try again 3930b411b363SPhilipp Reisner * -1 peer talks different language, 3931b411b363SPhilipp Reisner * no point in trying again, please go standalone. 3932b411b363SPhilipp Reisner */ 3933b411b363SPhilipp Reisner static int drbd_do_handshake(struct drbd_conf *mdev) 3934b411b363SPhilipp Reisner { 3935b411b363SPhilipp Reisner /* ASSERT current == mdev->receiver ... */ 3936b411b363SPhilipp Reisner struct p_handshake *p = &mdev->data.rbuf.handshake; 393702918be2SPhilipp Reisner const int expect = sizeof(struct p_handshake) - sizeof(struct p_header80); 393802918be2SPhilipp Reisner unsigned int length; 393902918be2SPhilipp Reisner enum drbd_packets cmd; 3940b411b363SPhilipp Reisner int rv; 3941b411b363SPhilipp Reisner 3942b411b363SPhilipp Reisner rv = drbd_send_handshake(mdev); 3943b411b363SPhilipp Reisner if (!rv) 3944b411b363SPhilipp Reisner return 0; 3945b411b363SPhilipp Reisner 394602918be2SPhilipp Reisner rv = drbd_recv_header(mdev, &cmd, &length); 3947b411b363SPhilipp Reisner if (!rv) 3948b411b363SPhilipp Reisner return 0; 3949b411b363SPhilipp Reisner 395002918be2SPhilipp Reisner if (cmd != P_HAND_SHAKE) { 3951b411b363SPhilipp Reisner dev_err(DEV, "expected HandShake packet, received: %s (0x%04x)\n", 395202918be2SPhilipp Reisner cmdname(cmd), cmd); 3953b411b363SPhilipp Reisner return -1; 3954b411b363SPhilipp Reisner } 3955b411b363SPhilipp Reisner 395602918be2SPhilipp Reisner if (length != expect) { 3957b411b363SPhilipp Reisner dev_err(DEV, "expected HandShake length: %u, received: %u\n", 395802918be2SPhilipp Reisner expect, length); 3959b411b363SPhilipp Reisner return -1; 3960b411b363SPhilipp Reisner } 3961b411b363SPhilipp Reisner 3962b411b363SPhilipp Reisner rv = drbd_recv(mdev, &p->head.payload, expect); 3963b411b363SPhilipp Reisner 3964b411b363SPhilipp Reisner if (rv != expect) { 39650ddc5549SLars Ellenberg if (!signal_pending(current)) 39660ddc5549SLars Ellenberg dev_warn(DEV, "short read receiving handshake packet: l=%u\n", rv); 3967b411b363SPhilipp Reisner return 0; 3968b411b363SPhilipp Reisner } 3969b411b363SPhilipp Reisner 3970b411b363SPhilipp Reisner p->protocol_min = be32_to_cpu(p->protocol_min); 3971b411b363SPhilipp Reisner p->protocol_max = be32_to_cpu(p->protocol_max); 3972b411b363SPhilipp Reisner if (p->protocol_max == 0) 3973b411b363SPhilipp Reisner p->protocol_max = p->protocol_min; 3974b411b363SPhilipp Reisner 3975b411b363SPhilipp Reisner if (PRO_VERSION_MAX < p->protocol_min || 3976b411b363SPhilipp Reisner PRO_VERSION_MIN > p->protocol_max) 3977b411b363SPhilipp Reisner goto incompat; 3978b411b363SPhilipp Reisner 3979b411b363SPhilipp Reisner mdev->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); 3980b411b363SPhilipp Reisner 3981b411b363SPhilipp Reisner dev_info(DEV, "Handshake successful: " 3982b411b363SPhilipp Reisner "Agreed network protocol version %d\n", mdev->agreed_pro_version); 3983b411b363SPhilipp Reisner 3984b411b363SPhilipp Reisner return 1; 3985b411b363SPhilipp Reisner 3986b411b363SPhilipp Reisner incompat: 3987b411b363SPhilipp Reisner dev_err(DEV, "incompatible DRBD dialects: " 3988b411b363SPhilipp Reisner "I support %d-%d, peer supports %d-%d\n", 3989b411b363SPhilipp Reisner PRO_VERSION_MIN, PRO_VERSION_MAX, 3990b411b363SPhilipp Reisner p->protocol_min, p->protocol_max); 3991b411b363SPhilipp Reisner return -1; 3992b411b363SPhilipp Reisner } 3993b411b363SPhilipp Reisner 3994b411b363SPhilipp Reisner #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) 3995b411b363SPhilipp Reisner static int drbd_do_auth(struct drbd_conf *mdev) 3996b411b363SPhilipp Reisner { 3997b411b363SPhilipp Reisner dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); 3998b411b363SPhilipp Reisner dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); 3999b10d96cbSJohannes Thoma return -1; 4000b411b363SPhilipp Reisner } 4001b411b363SPhilipp Reisner #else 4002b411b363SPhilipp Reisner #define CHALLENGE_LEN 64 4003b10d96cbSJohannes Thoma 4004b10d96cbSJohannes Thoma /* Return value: 4005b10d96cbSJohannes Thoma 1 - auth succeeded, 4006b10d96cbSJohannes Thoma 0 - failed, try again (network error), 4007b10d96cbSJohannes Thoma -1 - auth failed, don't try again. 4008b10d96cbSJohannes Thoma */ 4009b10d96cbSJohannes Thoma 4010b411b363SPhilipp Reisner static int drbd_do_auth(struct drbd_conf *mdev) 4011b411b363SPhilipp Reisner { 4012b411b363SPhilipp Reisner char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ 4013b411b363SPhilipp Reisner struct scatterlist sg; 4014b411b363SPhilipp Reisner char *response = NULL; 4015b411b363SPhilipp Reisner char *right_response = NULL; 4016b411b363SPhilipp Reisner char *peers_ch = NULL; 4017b411b363SPhilipp Reisner unsigned int key_len = strlen(mdev->net_conf->shared_secret); 4018b411b363SPhilipp Reisner unsigned int resp_size; 4019b411b363SPhilipp Reisner struct hash_desc desc; 402002918be2SPhilipp Reisner enum drbd_packets cmd; 402102918be2SPhilipp Reisner unsigned int length; 4022b411b363SPhilipp Reisner int rv; 4023b411b363SPhilipp Reisner 4024b411b363SPhilipp Reisner desc.tfm = mdev->cram_hmac_tfm; 4025b411b363SPhilipp Reisner desc.flags = 0; 4026b411b363SPhilipp Reisner 4027b411b363SPhilipp Reisner rv = crypto_hash_setkey(mdev->cram_hmac_tfm, 4028b411b363SPhilipp Reisner (u8 *)mdev->net_conf->shared_secret, key_len); 4029b411b363SPhilipp Reisner if (rv) { 4030b411b363SPhilipp Reisner dev_err(DEV, "crypto_hash_setkey() failed with %d\n", rv); 4031b10d96cbSJohannes Thoma rv = -1; 4032b411b363SPhilipp Reisner goto fail; 4033b411b363SPhilipp Reisner } 4034b411b363SPhilipp Reisner 4035b411b363SPhilipp Reisner get_random_bytes(my_challenge, CHALLENGE_LEN); 4036b411b363SPhilipp Reisner 4037b411b363SPhilipp Reisner rv = drbd_send_cmd2(mdev, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN); 4038b411b363SPhilipp Reisner if (!rv) 4039b411b363SPhilipp Reisner goto fail; 4040b411b363SPhilipp Reisner 404102918be2SPhilipp Reisner rv = drbd_recv_header(mdev, &cmd, &length); 4042b411b363SPhilipp Reisner if (!rv) 4043b411b363SPhilipp Reisner goto fail; 4044b411b363SPhilipp Reisner 404502918be2SPhilipp Reisner if (cmd != P_AUTH_CHALLENGE) { 4046b411b363SPhilipp Reisner dev_err(DEV, "expected AuthChallenge packet, received: %s (0x%04x)\n", 404702918be2SPhilipp Reisner cmdname(cmd), cmd); 4048b411b363SPhilipp Reisner rv = 0; 4049b411b363SPhilipp Reisner goto fail; 4050b411b363SPhilipp Reisner } 4051b411b363SPhilipp Reisner 405202918be2SPhilipp Reisner if (length > CHALLENGE_LEN * 2) { 4053b411b363SPhilipp Reisner dev_err(DEV, "expected AuthChallenge payload too big.\n"); 4054b10d96cbSJohannes Thoma rv = -1; 4055b411b363SPhilipp Reisner goto fail; 4056b411b363SPhilipp Reisner } 4057b411b363SPhilipp Reisner 405802918be2SPhilipp Reisner peers_ch = kmalloc(length, GFP_NOIO); 4059b411b363SPhilipp Reisner if (peers_ch == NULL) { 4060b411b363SPhilipp Reisner dev_err(DEV, "kmalloc of peers_ch failed\n"); 4061b10d96cbSJohannes Thoma rv = -1; 4062b411b363SPhilipp Reisner goto fail; 4063b411b363SPhilipp Reisner } 4064b411b363SPhilipp Reisner 406502918be2SPhilipp Reisner rv = drbd_recv(mdev, peers_ch, length); 4066b411b363SPhilipp Reisner 406702918be2SPhilipp Reisner if (rv != length) { 40680ddc5549SLars Ellenberg if (!signal_pending(current)) 40690ddc5549SLars Ellenberg dev_warn(DEV, "short read AuthChallenge: l=%u\n", rv); 4070b411b363SPhilipp Reisner rv = 0; 4071b411b363SPhilipp Reisner goto fail; 4072b411b363SPhilipp Reisner } 4073b411b363SPhilipp Reisner 4074b411b363SPhilipp Reisner resp_size = crypto_hash_digestsize(mdev->cram_hmac_tfm); 4075b411b363SPhilipp Reisner response = kmalloc(resp_size, GFP_NOIO); 4076b411b363SPhilipp Reisner if (response == NULL) { 4077b411b363SPhilipp Reisner dev_err(DEV, "kmalloc of response failed\n"); 4078b10d96cbSJohannes Thoma rv = -1; 4079b411b363SPhilipp Reisner goto fail; 4080b411b363SPhilipp Reisner } 4081b411b363SPhilipp Reisner 4082b411b363SPhilipp Reisner sg_init_table(&sg, 1); 408302918be2SPhilipp Reisner sg_set_buf(&sg, peers_ch, length); 4084b411b363SPhilipp Reisner 4085b411b363SPhilipp Reisner rv = crypto_hash_digest(&desc, &sg, sg.length, response); 4086b411b363SPhilipp Reisner if (rv) { 4087b411b363SPhilipp Reisner dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); 4088b10d96cbSJohannes Thoma rv = -1; 4089b411b363SPhilipp Reisner goto fail; 4090b411b363SPhilipp Reisner } 4091b411b363SPhilipp Reisner 4092b411b363SPhilipp Reisner rv = drbd_send_cmd2(mdev, P_AUTH_RESPONSE, response, resp_size); 4093b411b363SPhilipp Reisner if (!rv) 4094b411b363SPhilipp Reisner goto fail; 4095b411b363SPhilipp Reisner 409602918be2SPhilipp Reisner rv = drbd_recv_header(mdev, &cmd, &length); 4097b411b363SPhilipp Reisner if (!rv) 4098b411b363SPhilipp Reisner goto fail; 4099b411b363SPhilipp Reisner 410002918be2SPhilipp Reisner if (cmd != P_AUTH_RESPONSE) { 4101b411b363SPhilipp Reisner dev_err(DEV, "expected AuthResponse packet, received: %s (0x%04x)\n", 410202918be2SPhilipp Reisner cmdname(cmd), cmd); 4103b411b363SPhilipp Reisner rv = 0; 4104b411b363SPhilipp Reisner goto fail; 4105b411b363SPhilipp Reisner } 4106b411b363SPhilipp Reisner 410702918be2SPhilipp Reisner if (length != resp_size) { 4108b411b363SPhilipp Reisner dev_err(DEV, "expected AuthResponse payload of wrong size\n"); 4109b411b363SPhilipp Reisner rv = 0; 4110b411b363SPhilipp Reisner goto fail; 4111b411b363SPhilipp Reisner } 4112b411b363SPhilipp Reisner 4113b411b363SPhilipp Reisner rv = drbd_recv(mdev, response , resp_size); 4114b411b363SPhilipp Reisner 4115b411b363SPhilipp Reisner if (rv != resp_size) { 41160ddc5549SLars Ellenberg if (!signal_pending(current)) 41170ddc5549SLars Ellenberg dev_warn(DEV, "short read receiving AuthResponse: l=%u\n", rv); 4118b411b363SPhilipp Reisner rv = 0; 4119b411b363SPhilipp Reisner goto fail; 4120b411b363SPhilipp Reisner } 4121b411b363SPhilipp Reisner 4122b411b363SPhilipp Reisner right_response = kmalloc(resp_size, GFP_NOIO); 41232d1ee87dSJulia Lawall if (right_response == NULL) { 4124b411b363SPhilipp Reisner dev_err(DEV, "kmalloc of right_response failed\n"); 4125b10d96cbSJohannes Thoma rv = -1; 4126b411b363SPhilipp Reisner goto fail; 4127b411b363SPhilipp Reisner } 4128b411b363SPhilipp Reisner 4129b411b363SPhilipp Reisner sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); 4130b411b363SPhilipp Reisner 4131b411b363SPhilipp Reisner rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); 4132b411b363SPhilipp Reisner if (rv) { 4133b411b363SPhilipp Reisner dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); 4134b10d96cbSJohannes Thoma rv = -1; 4135b411b363SPhilipp Reisner goto fail; 4136b411b363SPhilipp Reisner } 4137b411b363SPhilipp Reisner 4138b411b363SPhilipp Reisner rv = !memcmp(response, right_response, resp_size); 4139b411b363SPhilipp Reisner 4140b411b363SPhilipp Reisner if (rv) 4141b411b363SPhilipp Reisner dev_info(DEV, "Peer authenticated using %d bytes of '%s' HMAC\n", 4142b411b363SPhilipp Reisner resp_size, mdev->net_conf->cram_hmac_alg); 4143b10d96cbSJohannes Thoma else 4144b10d96cbSJohannes Thoma rv = -1; 4145b411b363SPhilipp Reisner 4146b411b363SPhilipp Reisner fail: 4147b411b363SPhilipp Reisner kfree(peers_ch); 4148b411b363SPhilipp Reisner kfree(response); 4149b411b363SPhilipp Reisner kfree(right_response); 4150b411b363SPhilipp Reisner 4151b411b363SPhilipp Reisner return rv; 4152b411b363SPhilipp Reisner } 4153b411b363SPhilipp Reisner #endif 4154b411b363SPhilipp Reisner 4155b411b363SPhilipp Reisner int drbdd_init(struct drbd_thread *thi) 4156b411b363SPhilipp Reisner { 4157b411b363SPhilipp Reisner struct drbd_conf *mdev = thi->mdev; 4158b411b363SPhilipp Reisner unsigned int minor = mdev_to_minor(mdev); 4159b411b363SPhilipp Reisner int h; 4160b411b363SPhilipp Reisner 4161b411b363SPhilipp Reisner sprintf(current->comm, "drbd%d_receiver", minor); 4162b411b363SPhilipp Reisner 4163b411b363SPhilipp Reisner dev_info(DEV, "receiver (re)started\n"); 4164b411b363SPhilipp Reisner 4165b411b363SPhilipp Reisner do { 4166b411b363SPhilipp Reisner h = drbd_connect(mdev); 4167b411b363SPhilipp Reisner if (h == 0) { 4168b411b363SPhilipp Reisner drbd_disconnect(mdev); 416920ee6390SPhilipp Reisner schedule_timeout_interruptible(HZ); 4170b411b363SPhilipp Reisner } 4171b411b363SPhilipp Reisner if (h == -1) { 4172b411b363SPhilipp Reisner dev_warn(DEV, "Discarding network configuration.\n"); 4173b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 4174b411b363SPhilipp Reisner } 4175b411b363SPhilipp Reisner } while (h == 0); 4176b411b363SPhilipp Reisner 4177b411b363SPhilipp Reisner if (h > 0) { 4178b411b363SPhilipp Reisner if (get_net_conf(mdev)) { 4179b411b363SPhilipp Reisner drbdd(mdev); 4180b411b363SPhilipp Reisner put_net_conf(mdev); 4181b411b363SPhilipp Reisner } 4182b411b363SPhilipp Reisner } 4183b411b363SPhilipp Reisner 4184b411b363SPhilipp Reisner drbd_disconnect(mdev); 4185b411b363SPhilipp Reisner 4186b411b363SPhilipp Reisner dev_info(DEV, "receiver terminated\n"); 4187b411b363SPhilipp Reisner return 0; 4188b411b363SPhilipp Reisner } 4189b411b363SPhilipp Reisner 4190b411b363SPhilipp Reisner /* ********* acknowledge sender ******** */ 4191b411b363SPhilipp Reisner 41920b70a13dSPhilipp Reisner static int got_RqSReply(struct drbd_conf *mdev, struct p_header80 *h) 4193b411b363SPhilipp Reisner { 4194b411b363SPhilipp Reisner struct p_req_state_reply *p = (struct p_req_state_reply *)h; 4195b411b363SPhilipp Reisner 4196b411b363SPhilipp Reisner int retcode = be32_to_cpu(p->retcode); 4197b411b363SPhilipp Reisner 4198b411b363SPhilipp Reisner if (retcode >= SS_SUCCESS) { 4199b411b363SPhilipp Reisner set_bit(CL_ST_CHG_SUCCESS, &mdev->flags); 4200b411b363SPhilipp Reisner } else { 4201b411b363SPhilipp Reisner set_bit(CL_ST_CHG_FAIL, &mdev->flags); 4202b411b363SPhilipp Reisner dev_err(DEV, "Requested state change failed by peer: %s (%d)\n", 4203b411b363SPhilipp Reisner drbd_set_st_err_str(retcode), retcode); 4204b411b363SPhilipp Reisner } 4205b411b363SPhilipp Reisner wake_up(&mdev->state_wait); 4206b411b363SPhilipp Reisner 420781e84650SAndreas Gruenbacher return true; 4208b411b363SPhilipp Reisner } 4209b411b363SPhilipp Reisner 42100b70a13dSPhilipp Reisner static int got_Ping(struct drbd_conf *mdev, struct p_header80 *h) 4211b411b363SPhilipp Reisner { 4212b411b363SPhilipp Reisner return drbd_send_ping_ack(mdev); 4213b411b363SPhilipp Reisner 4214b411b363SPhilipp Reisner } 4215b411b363SPhilipp Reisner 42160b70a13dSPhilipp Reisner static int got_PingAck(struct drbd_conf *mdev, struct p_header80 *h) 4217b411b363SPhilipp Reisner { 4218b411b363SPhilipp Reisner /* restore idle timeout */ 4219b411b363SPhilipp Reisner mdev->meta.socket->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ; 4220309d1608SPhilipp Reisner if (!test_and_set_bit(GOT_PING_ACK, &mdev->flags)) 4221309d1608SPhilipp Reisner wake_up(&mdev->misc_wait); 4222b411b363SPhilipp Reisner 422381e84650SAndreas Gruenbacher return true; 4224b411b363SPhilipp Reisner } 4225b411b363SPhilipp Reisner 42260b70a13dSPhilipp Reisner static int got_IsInSync(struct drbd_conf *mdev, struct p_header80 *h) 4227b411b363SPhilipp Reisner { 4228b411b363SPhilipp Reisner struct p_block_ack *p = (struct p_block_ack *)h; 4229b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 4230b411b363SPhilipp Reisner int blksize = be32_to_cpu(p->blksize); 4231b411b363SPhilipp Reisner 4232b411b363SPhilipp Reisner D_ASSERT(mdev->agreed_pro_version >= 89); 4233b411b363SPhilipp Reisner 4234b411b363SPhilipp Reisner update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4235b411b363SPhilipp Reisner 42361d53f09eSLars Ellenberg if (get_ldev(mdev)) { 4237b411b363SPhilipp Reisner drbd_rs_complete_io(mdev, sector); 4238b411b363SPhilipp Reisner drbd_set_in_sync(mdev, sector, blksize); 4239b411b363SPhilipp Reisner /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ 4240b411b363SPhilipp Reisner mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); 42411d53f09eSLars Ellenberg put_ldev(mdev); 42421d53f09eSLars Ellenberg } 4243b411b363SPhilipp Reisner dec_rs_pending(mdev); 4244778f271dSPhilipp Reisner atomic_add(blksize >> 9, &mdev->rs_sect_in); 4245b411b363SPhilipp Reisner 424681e84650SAndreas Gruenbacher return true; 4247b411b363SPhilipp Reisner } 4248b411b363SPhilipp Reisner 4249b411b363SPhilipp Reisner /* when we receive the ACK for a write request, 4250b411b363SPhilipp Reisner * verify that we actually know about it */ 4251b411b363SPhilipp Reisner static struct drbd_request *_ack_id_to_req(struct drbd_conf *mdev, 4252b411b363SPhilipp Reisner u64 id, sector_t sector) 4253b411b363SPhilipp Reisner { 4254b411b363SPhilipp Reisner struct hlist_head *slot = tl_hash_slot(mdev, sector); 4255b411b363SPhilipp Reisner struct hlist_node *n; 4256b411b363SPhilipp Reisner struct drbd_request *req; 4257b411b363SPhilipp Reisner 425824c4830cSBart Van Assche hlist_for_each_entry(req, n, slot, collision) { 4259b411b363SPhilipp Reisner if ((unsigned long)req == (unsigned long)id) { 4260b411b363SPhilipp Reisner if (req->sector != sector) { 4261b411b363SPhilipp Reisner dev_err(DEV, "_ack_id_to_req: found req %p but it has " 4262b411b363SPhilipp Reisner "wrong sector (%llus versus %llus)\n", req, 4263b411b363SPhilipp Reisner (unsigned long long)req->sector, 4264b411b363SPhilipp Reisner (unsigned long long)sector); 4265b411b363SPhilipp Reisner break; 4266b411b363SPhilipp Reisner } 4267b411b363SPhilipp Reisner return req; 4268b411b363SPhilipp Reisner } 4269b411b363SPhilipp Reisner } 4270b411b363SPhilipp Reisner return NULL; 4271b411b363SPhilipp Reisner } 4272b411b363SPhilipp Reisner 4273b411b363SPhilipp Reisner typedef struct drbd_request *(req_validator_fn) 4274b411b363SPhilipp Reisner (struct drbd_conf *mdev, u64 id, sector_t sector); 4275b411b363SPhilipp Reisner 4276b411b363SPhilipp Reisner static int validate_req_change_req_state(struct drbd_conf *mdev, 4277b411b363SPhilipp Reisner u64 id, sector_t sector, req_validator_fn validator, 4278b411b363SPhilipp Reisner const char *func, enum drbd_req_event what) 4279b411b363SPhilipp Reisner { 4280b411b363SPhilipp Reisner struct drbd_request *req; 4281b411b363SPhilipp Reisner struct bio_and_error m; 4282b411b363SPhilipp Reisner 4283b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 4284b411b363SPhilipp Reisner req = validator(mdev, id, sector); 4285b411b363SPhilipp Reisner if (unlikely(!req)) { 4286b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 42872deb8336SPhilipp Reisner 42882deb8336SPhilipp Reisner dev_err(DEV, "%s: failed to find req %p, sector %llus\n", func, 42892deb8336SPhilipp Reisner (void *)(unsigned long)id, (unsigned long long)sector); 429081e84650SAndreas Gruenbacher return false; 4291b411b363SPhilipp Reisner } 4292b411b363SPhilipp Reisner __req_mod(req, what, &m); 4293b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 4294b411b363SPhilipp Reisner 4295b411b363SPhilipp Reisner if (m.bio) 4296b411b363SPhilipp Reisner complete_master_bio(mdev, &m); 429781e84650SAndreas Gruenbacher return true; 4298b411b363SPhilipp Reisner } 4299b411b363SPhilipp Reisner 43000b70a13dSPhilipp Reisner static int got_BlockAck(struct drbd_conf *mdev, struct p_header80 *h) 4301b411b363SPhilipp Reisner { 4302b411b363SPhilipp Reisner struct p_block_ack *p = (struct p_block_ack *)h; 4303b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 4304b411b363SPhilipp Reisner int blksize = be32_to_cpu(p->blksize); 4305b411b363SPhilipp Reisner enum drbd_req_event what; 4306b411b363SPhilipp Reisner 4307b411b363SPhilipp Reisner update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4308b411b363SPhilipp Reisner 4309b411b363SPhilipp Reisner if (is_syncer_block_id(p->block_id)) { 4310b411b363SPhilipp Reisner drbd_set_in_sync(mdev, sector, blksize); 4311b411b363SPhilipp Reisner dec_rs_pending(mdev); 431281e84650SAndreas Gruenbacher return true; 4313b411b363SPhilipp Reisner } 4314b411b363SPhilipp Reisner switch (be16_to_cpu(h->command)) { 4315b411b363SPhilipp Reisner case P_RS_WRITE_ACK: 4316b411b363SPhilipp Reisner D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 4317b411b363SPhilipp Reisner what = write_acked_by_peer_and_sis; 4318b411b363SPhilipp Reisner break; 4319b411b363SPhilipp Reisner case P_WRITE_ACK: 4320b411b363SPhilipp Reisner D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 4321b411b363SPhilipp Reisner what = write_acked_by_peer; 4322b411b363SPhilipp Reisner break; 4323b411b363SPhilipp Reisner case P_RECV_ACK: 4324b411b363SPhilipp Reisner D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_B); 4325b411b363SPhilipp Reisner what = recv_acked_by_peer; 4326b411b363SPhilipp Reisner break; 4327b411b363SPhilipp Reisner case P_DISCARD_ACK: 4328b411b363SPhilipp Reisner D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C); 4329b411b363SPhilipp Reisner what = conflict_discarded_by_peer; 4330b411b363SPhilipp Reisner break; 4331b411b363SPhilipp Reisner default: 4332b411b363SPhilipp Reisner D_ASSERT(0); 433381e84650SAndreas Gruenbacher return false; 4334b411b363SPhilipp Reisner } 4335b411b363SPhilipp Reisner 4336b411b363SPhilipp Reisner return validate_req_change_req_state(mdev, p->block_id, sector, 4337b411b363SPhilipp Reisner _ack_id_to_req, __func__ , what); 4338b411b363SPhilipp Reisner } 4339b411b363SPhilipp Reisner 43400b70a13dSPhilipp Reisner static int got_NegAck(struct drbd_conf *mdev, struct p_header80 *h) 4341b411b363SPhilipp Reisner { 4342b411b363SPhilipp Reisner struct p_block_ack *p = (struct p_block_ack *)h; 4343b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 43442deb8336SPhilipp Reisner int size = be32_to_cpu(p->blksize); 43452deb8336SPhilipp Reisner struct drbd_request *req; 43462deb8336SPhilipp Reisner struct bio_and_error m; 4347b411b363SPhilipp Reisner 4348b411b363SPhilipp Reisner update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4349b411b363SPhilipp Reisner 4350b411b363SPhilipp Reisner if (is_syncer_block_id(p->block_id)) { 4351b411b363SPhilipp Reisner dec_rs_pending(mdev); 4352b411b363SPhilipp Reisner drbd_rs_failed_io(mdev, sector, size); 435381e84650SAndreas Gruenbacher return true; 4354b411b363SPhilipp Reisner } 43552deb8336SPhilipp Reisner 43562deb8336SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 43572deb8336SPhilipp Reisner req = _ack_id_to_req(mdev, p->block_id, sector); 43582deb8336SPhilipp Reisner if (!req) { 43592deb8336SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 43602deb8336SPhilipp Reisner if (mdev->net_conf->wire_protocol == DRBD_PROT_A || 43612deb8336SPhilipp Reisner mdev->net_conf->wire_protocol == DRBD_PROT_B) { 43622deb8336SPhilipp Reisner /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. 43632deb8336SPhilipp Reisner The master bio might already be completed, therefore the 43642deb8336SPhilipp Reisner request is no longer in the collision hash. 43652deb8336SPhilipp Reisner => Do not try to validate block_id as request. */ 43662deb8336SPhilipp Reisner /* In Protocol B we might already have got a P_RECV_ACK 43672deb8336SPhilipp Reisner but then get a P_NEG_ACK after wards. */ 43682deb8336SPhilipp Reisner drbd_set_out_of_sync(mdev, sector, size); 43692deb8336SPhilipp Reisner return true; 43702deb8336SPhilipp Reisner } else { 43712deb8336SPhilipp Reisner dev_err(DEV, "%s: failed to find req %p, sector %llus\n", __func__, 43722deb8336SPhilipp Reisner (void *)(unsigned long)p->block_id, (unsigned long long)sector); 43732deb8336SPhilipp Reisner return false; 43742deb8336SPhilipp Reisner } 43752deb8336SPhilipp Reisner } 43762deb8336SPhilipp Reisner __req_mod(req, neg_acked, &m); 43772deb8336SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 43782deb8336SPhilipp Reisner 43792deb8336SPhilipp Reisner if (m.bio) 43802deb8336SPhilipp Reisner complete_master_bio(mdev, &m); 43812deb8336SPhilipp Reisner return true; 4382b411b363SPhilipp Reisner } 4383b411b363SPhilipp Reisner 43840b70a13dSPhilipp Reisner static int got_NegDReply(struct drbd_conf *mdev, struct p_header80 *h) 4385b411b363SPhilipp Reisner { 4386b411b363SPhilipp Reisner struct p_block_ack *p = (struct p_block_ack *)h; 4387b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 4388b411b363SPhilipp Reisner 4389b411b363SPhilipp Reisner update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4390b411b363SPhilipp Reisner dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n", 4391b411b363SPhilipp Reisner (unsigned long long)sector, be32_to_cpu(p->blksize)); 4392b411b363SPhilipp Reisner 4393b411b363SPhilipp Reisner return validate_req_change_req_state(mdev, p->block_id, sector, 4394b411b363SPhilipp Reisner _ar_id_to_req, __func__ , neg_acked); 4395b411b363SPhilipp Reisner } 4396b411b363SPhilipp Reisner 43970b70a13dSPhilipp Reisner static int got_NegRSDReply(struct drbd_conf *mdev, struct p_header80 *h) 4398b411b363SPhilipp Reisner { 4399b411b363SPhilipp Reisner sector_t sector; 4400b411b363SPhilipp Reisner int size; 4401b411b363SPhilipp Reisner struct p_block_ack *p = (struct p_block_ack *)h; 4402b411b363SPhilipp Reisner 4403b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 4404b411b363SPhilipp Reisner size = be32_to_cpu(p->blksize); 4405b411b363SPhilipp Reisner 4406b411b363SPhilipp Reisner update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4407b411b363SPhilipp Reisner 4408b411b363SPhilipp Reisner dec_rs_pending(mdev); 4409b411b363SPhilipp Reisner 4410b411b363SPhilipp Reisner if (get_ldev_if_state(mdev, D_FAILED)) { 4411b411b363SPhilipp Reisner drbd_rs_complete_io(mdev, sector); 4412d612d309SPhilipp Reisner switch (be16_to_cpu(h->command)) { 4413d612d309SPhilipp Reisner case P_NEG_RS_DREPLY: 4414b411b363SPhilipp Reisner drbd_rs_failed_io(mdev, sector, size); 4415d612d309SPhilipp Reisner case P_RS_CANCEL: 4416d612d309SPhilipp Reisner break; 4417d612d309SPhilipp Reisner default: 4418d612d309SPhilipp Reisner D_ASSERT(0); 4419d612d309SPhilipp Reisner put_ldev(mdev); 4420d612d309SPhilipp Reisner return false; 4421d612d309SPhilipp Reisner } 4422b411b363SPhilipp Reisner put_ldev(mdev); 4423b411b363SPhilipp Reisner } 4424b411b363SPhilipp Reisner 442581e84650SAndreas Gruenbacher return true; 4426b411b363SPhilipp Reisner } 4427b411b363SPhilipp Reisner 44280b70a13dSPhilipp Reisner static int got_BarrierAck(struct drbd_conf *mdev, struct p_header80 *h) 4429b411b363SPhilipp Reisner { 4430b411b363SPhilipp Reisner struct p_barrier_ack *p = (struct p_barrier_ack *)h; 4431b411b363SPhilipp Reisner 4432b411b363SPhilipp Reisner tl_release(mdev, p->barrier, be32_to_cpu(p->set_size)); 4433b411b363SPhilipp Reisner 4434c4752ef1SPhilipp Reisner if (mdev->state.conn == C_AHEAD && 4435c4752ef1SPhilipp Reisner atomic_read(&mdev->ap_in_flight) == 0 && 4436370a43e7SPhilipp Reisner !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &mdev->current_epoch->flags)) { 4437370a43e7SPhilipp Reisner mdev->start_resync_timer.expires = jiffies + HZ; 4438370a43e7SPhilipp Reisner add_timer(&mdev->start_resync_timer); 4439c4752ef1SPhilipp Reisner } 4440c4752ef1SPhilipp Reisner 444181e84650SAndreas Gruenbacher return true; 4442b411b363SPhilipp Reisner } 4443b411b363SPhilipp Reisner 44440b70a13dSPhilipp Reisner static int got_OVResult(struct drbd_conf *mdev, struct p_header80 *h) 4445b411b363SPhilipp Reisner { 4446b411b363SPhilipp Reisner struct p_block_ack *p = (struct p_block_ack *)h; 4447b411b363SPhilipp Reisner struct drbd_work *w; 4448b411b363SPhilipp Reisner sector_t sector; 4449b411b363SPhilipp Reisner int size; 4450b411b363SPhilipp Reisner 4451b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 4452b411b363SPhilipp Reisner size = be32_to_cpu(p->blksize); 4453b411b363SPhilipp Reisner 4454b411b363SPhilipp Reisner update_peer_seq(mdev, be32_to_cpu(p->seq_num)); 4455b411b363SPhilipp Reisner 4456b411b363SPhilipp Reisner if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) 4457b411b363SPhilipp Reisner drbd_ov_oos_found(mdev, sector, size); 4458b411b363SPhilipp Reisner else 4459b411b363SPhilipp Reisner ov_oos_print(mdev); 4460b411b363SPhilipp Reisner 44611d53f09eSLars Ellenberg if (!get_ldev(mdev)) 446281e84650SAndreas Gruenbacher return true; 44631d53f09eSLars Ellenberg 4464b411b363SPhilipp Reisner drbd_rs_complete_io(mdev, sector); 4465b411b363SPhilipp Reisner dec_rs_pending(mdev); 4466b411b363SPhilipp Reisner 4467ea5442afSLars Ellenberg --mdev->ov_left; 4468ea5442afSLars Ellenberg 4469ea5442afSLars Ellenberg /* let's advance progress step marks only for every other megabyte */ 4470ea5442afSLars Ellenberg if ((mdev->ov_left & 0x200) == 0x200) 4471ea5442afSLars Ellenberg drbd_advance_rs_marks(mdev, mdev->ov_left); 4472ea5442afSLars Ellenberg 4473ea5442afSLars Ellenberg if (mdev->ov_left == 0) { 4474b411b363SPhilipp Reisner w = kmalloc(sizeof(*w), GFP_NOIO); 4475b411b363SPhilipp Reisner if (w) { 4476b411b363SPhilipp Reisner w->cb = w_ov_finished; 4477b411b363SPhilipp Reisner drbd_queue_work_front(&mdev->data.work, w); 4478b411b363SPhilipp Reisner } else { 4479b411b363SPhilipp Reisner dev_err(DEV, "kmalloc(w) failed."); 4480b411b363SPhilipp Reisner ov_oos_print(mdev); 4481b411b363SPhilipp Reisner drbd_resync_finished(mdev); 4482b411b363SPhilipp Reisner } 4483b411b363SPhilipp Reisner } 44841d53f09eSLars Ellenberg put_ldev(mdev); 448581e84650SAndreas Gruenbacher return true; 4486b411b363SPhilipp Reisner } 4487b411b363SPhilipp Reisner 448802918be2SPhilipp Reisner static int got_skip(struct drbd_conf *mdev, struct p_header80 *h) 44890ced55a3SPhilipp Reisner { 449081e84650SAndreas Gruenbacher return true; 44910ced55a3SPhilipp Reisner } 44920ced55a3SPhilipp Reisner 4493b411b363SPhilipp Reisner struct asender_cmd { 4494b411b363SPhilipp Reisner size_t pkt_size; 44950b70a13dSPhilipp Reisner int (*process)(struct drbd_conf *mdev, struct p_header80 *h); 4496b411b363SPhilipp Reisner }; 4497b411b363SPhilipp Reisner 4498b411b363SPhilipp Reisner static struct asender_cmd *get_asender_cmd(int cmd) 4499b411b363SPhilipp Reisner { 4500b411b363SPhilipp Reisner static struct asender_cmd asender_tbl[] = { 4501b411b363SPhilipp Reisner /* anything missing from this table is in 4502b411b363SPhilipp Reisner * the drbd_cmd_handler (drbd_default_handler) table, 4503b411b363SPhilipp Reisner * see the beginning of drbdd() */ 45040b70a13dSPhilipp Reisner [P_PING] = { sizeof(struct p_header80), got_Ping }, 45050b70a13dSPhilipp Reisner [P_PING_ACK] = { sizeof(struct p_header80), got_PingAck }, 4506b411b363SPhilipp Reisner [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 4507b411b363SPhilipp Reisner [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 4508b411b363SPhilipp Reisner [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 4509b411b363SPhilipp Reisner [P_DISCARD_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 4510b411b363SPhilipp Reisner [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, 4511b411b363SPhilipp Reisner [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, 4512b411b363SPhilipp Reisner [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply}, 4513b411b363SPhilipp Reisner [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, 4514b411b363SPhilipp Reisner [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, 4515b411b363SPhilipp Reisner [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, 4516b411b363SPhilipp Reisner [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, 451702918be2SPhilipp Reisner [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, 4518d612d309SPhilipp Reisner [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply}, 4519b411b363SPhilipp Reisner [P_MAX_CMD] = { 0, NULL }, 4520b411b363SPhilipp Reisner }; 4521b411b363SPhilipp Reisner if (cmd > P_MAX_CMD || asender_tbl[cmd].process == NULL) 4522b411b363SPhilipp Reisner return NULL; 4523b411b363SPhilipp Reisner return &asender_tbl[cmd]; 4524b411b363SPhilipp Reisner } 4525b411b363SPhilipp Reisner 4526b411b363SPhilipp Reisner int drbd_asender(struct drbd_thread *thi) 4527b411b363SPhilipp Reisner { 4528b411b363SPhilipp Reisner struct drbd_conf *mdev = thi->mdev; 452902918be2SPhilipp Reisner struct p_header80 *h = &mdev->meta.rbuf.header.h80; 4530b411b363SPhilipp Reisner struct asender_cmd *cmd = NULL; 4531b411b363SPhilipp Reisner 4532b411b363SPhilipp Reisner int rv, len; 4533b411b363SPhilipp Reisner void *buf = h; 4534b411b363SPhilipp Reisner int received = 0; 45350b70a13dSPhilipp Reisner int expect = sizeof(struct p_header80); 4536b411b363SPhilipp Reisner int empty; 4537f36af18cSLars Ellenberg int ping_timeout_active = 0; 4538b411b363SPhilipp Reisner 4539b411b363SPhilipp Reisner sprintf(current->comm, "drbd%d_asender", mdev_to_minor(mdev)); 4540b411b363SPhilipp Reisner 4541b411b363SPhilipp Reisner current->policy = SCHED_RR; /* Make this a realtime task! */ 4542b411b363SPhilipp Reisner current->rt_priority = 2; /* more important than all other tasks */ 4543b411b363SPhilipp Reisner 4544b411b363SPhilipp Reisner while (get_t_state(thi) == Running) { 4545b411b363SPhilipp Reisner drbd_thread_current_set_cpu(mdev); 4546b411b363SPhilipp Reisner if (test_and_clear_bit(SEND_PING, &mdev->flags)) { 4547b411b363SPhilipp Reisner ERR_IF(!drbd_send_ping(mdev)) goto reconnect; 4548b411b363SPhilipp Reisner mdev->meta.socket->sk->sk_rcvtimeo = 4549b411b363SPhilipp Reisner mdev->net_conf->ping_timeo*HZ/10; 4550f36af18cSLars Ellenberg ping_timeout_active = 1; 4551b411b363SPhilipp Reisner } 4552b411b363SPhilipp Reisner 4553b411b363SPhilipp Reisner /* conditionally cork; 4554b411b363SPhilipp Reisner * it may hurt latency if we cork without much to send */ 4555b411b363SPhilipp Reisner if (!mdev->net_conf->no_cork && 4556b411b363SPhilipp Reisner 3 < atomic_read(&mdev->unacked_cnt)) 4557b411b363SPhilipp Reisner drbd_tcp_cork(mdev->meta.socket); 4558b411b363SPhilipp Reisner while (1) { 4559b411b363SPhilipp Reisner clear_bit(SIGNAL_ASENDER, &mdev->flags); 4560b411b363SPhilipp Reisner flush_signals(current); 45610f8488e1SLars Ellenberg if (!drbd_process_done_ee(mdev)) 4562b411b363SPhilipp Reisner goto reconnect; 4563b411b363SPhilipp Reisner /* to avoid race with newly queued ACKs */ 4564b411b363SPhilipp Reisner set_bit(SIGNAL_ASENDER, &mdev->flags); 4565b411b363SPhilipp Reisner spin_lock_irq(&mdev->req_lock); 4566b411b363SPhilipp Reisner empty = list_empty(&mdev->done_ee); 4567b411b363SPhilipp Reisner spin_unlock_irq(&mdev->req_lock); 4568b411b363SPhilipp Reisner /* new ack may have been queued right here, 4569b411b363SPhilipp Reisner * but then there is also a signal pending, 4570b411b363SPhilipp Reisner * and we start over... */ 4571b411b363SPhilipp Reisner if (empty) 4572b411b363SPhilipp Reisner break; 4573b411b363SPhilipp Reisner } 4574b411b363SPhilipp Reisner /* but unconditionally uncork unless disabled */ 4575b411b363SPhilipp Reisner if (!mdev->net_conf->no_cork) 4576b411b363SPhilipp Reisner drbd_tcp_uncork(mdev->meta.socket); 4577b411b363SPhilipp Reisner 4578b411b363SPhilipp Reisner /* short circuit, recv_msg would return EINTR anyways. */ 4579b411b363SPhilipp Reisner if (signal_pending(current)) 4580b411b363SPhilipp Reisner continue; 4581b411b363SPhilipp Reisner 4582b411b363SPhilipp Reisner rv = drbd_recv_short(mdev, mdev->meta.socket, 4583b411b363SPhilipp Reisner buf, expect-received, 0); 4584b411b363SPhilipp Reisner clear_bit(SIGNAL_ASENDER, &mdev->flags); 4585b411b363SPhilipp Reisner 4586b411b363SPhilipp Reisner flush_signals(current); 4587b411b363SPhilipp Reisner 4588b411b363SPhilipp Reisner /* Note: 4589b411b363SPhilipp Reisner * -EINTR (on meta) we got a signal 4590b411b363SPhilipp Reisner * -EAGAIN (on meta) rcvtimeo expired 4591b411b363SPhilipp Reisner * -ECONNRESET other side closed the connection 4592b411b363SPhilipp Reisner * -ERESTARTSYS (on data) we got a signal 4593b411b363SPhilipp Reisner * rv < 0 other than above: unexpected error! 4594b411b363SPhilipp Reisner * rv == expected: full header or command 4595b411b363SPhilipp Reisner * rv < expected: "woken" by signal during receive 4596b411b363SPhilipp Reisner * rv == 0 : "connection shut down by peer" 4597b411b363SPhilipp Reisner */ 4598b411b363SPhilipp Reisner if (likely(rv > 0)) { 4599b411b363SPhilipp Reisner received += rv; 4600b411b363SPhilipp Reisner buf += rv; 4601b411b363SPhilipp Reisner } else if (rv == 0) { 4602b411b363SPhilipp Reisner dev_err(DEV, "meta connection shut down by peer.\n"); 4603b411b363SPhilipp Reisner goto reconnect; 4604b411b363SPhilipp Reisner } else if (rv == -EAGAIN) { 4605cb6518cbSLars Ellenberg /* If the data socket received something meanwhile, 4606cb6518cbSLars Ellenberg * that is good enough: peer is still alive. */ 4607cb6518cbSLars Ellenberg if (time_after(mdev->last_received, 4608cb6518cbSLars Ellenberg jiffies - mdev->meta.socket->sk->sk_rcvtimeo)) 4609cb6518cbSLars Ellenberg continue; 4610f36af18cSLars Ellenberg if (ping_timeout_active) { 4611b411b363SPhilipp Reisner dev_err(DEV, "PingAck did not arrive in time.\n"); 4612b411b363SPhilipp Reisner goto reconnect; 4613b411b363SPhilipp Reisner } 4614b411b363SPhilipp Reisner set_bit(SEND_PING, &mdev->flags); 4615b411b363SPhilipp Reisner continue; 4616b411b363SPhilipp Reisner } else if (rv == -EINTR) { 4617b411b363SPhilipp Reisner continue; 4618b411b363SPhilipp Reisner } else { 4619b411b363SPhilipp Reisner dev_err(DEV, "sock_recvmsg returned %d\n", rv); 4620b411b363SPhilipp Reisner goto reconnect; 4621b411b363SPhilipp Reisner } 4622b411b363SPhilipp Reisner 4623b411b363SPhilipp Reisner if (received == expect && cmd == NULL) { 4624b411b363SPhilipp Reisner if (unlikely(h->magic != BE_DRBD_MAGIC)) { 4625004352faSLars Ellenberg dev_err(DEV, "magic?? on meta m: 0x%08x c: %d l: %d\n", 4626004352faSLars Ellenberg be32_to_cpu(h->magic), 4627004352faSLars Ellenberg be16_to_cpu(h->command), 4628004352faSLars Ellenberg be16_to_cpu(h->length)); 4629b411b363SPhilipp Reisner goto reconnect; 4630b411b363SPhilipp Reisner } 4631b411b363SPhilipp Reisner cmd = get_asender_cmd(be16_to_cpu(h->command)); 4632b411b363SPhilipp Reisner len = be16_to_cpu(h->length); 4633b411b363SPhilipp Reisner if (unlikely(cmd == NULL)) { 4634004352faSLars Ellenberg dev_err(DEV, "unknown command?? on meta m: 0x%08x c: %d l: %d\n", 4635004352faSLars Ellenberg be32_to_cpu(h->magic), 4636004352faSLars Ellenberg be16_to_cpu(h->command), 4637004352faSLars Ellenberg be16_to_cpu(h->length)); 4638b411b363SPhilipp Reisner goto disconnect; 4639b411b363SPhilipp Reisner } 4640b411b363SPhilipp Reisner expect = cmd->pkt_size; 46410b70a13dSPhilipp Reisner ERR_IF(len != expect-sizeof(struct p_header80)) 4642b411b363SPhilipp Reisner goto reconnect; 4643b411b363SPhilipp Reisner } 4644b411b363SPhilipp Reisner if (received == expect) { 4645cb6518cbSLars Ellenberg mdev->last_received = jiffies; 4646b411b363SPhilipp Reisner D_ASSERT(cmd != NULL); 4647b411b363SPhilipp Reisner if (!cmd->process(mdev, h)) 4648b411b363SPhilipp Reisner goto reconnect; 4649b411b363SPhilipp Reisner 4650f36af18cSLars Ellenberg /* the idle_timeout (ping-int) 4651f36af18cSLars Ellenberg * has been restored in got_PingAck() */ 4652f36af18cSLars Ellenberg if (cmd == get_asender_cmd(P_PING_ACK)) 4653f36af18cSLars Ellenberg ping_timeout_active = 0; 4654f36af18cSLars Ellenberg 4655b411b363SPhilipp Reisner buf = h; 4656b411b363SPhilipp Reisner received = 0; 46570b70a13dSPhilipp Reisner expect = sizeof(struct p_header80); 4658b411b363SPhilipp Reisner cmd = NULL; 4659b411b363SPhilipp Reisner } 4660b411b363SPhilipp Reisner } 4661b411b363SPhilipp Reisner 4662b411b363SPhilipp Reisner if (0) { 4663b411b363SPhilipp Reisner reconnect: 4664b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE)); 4665856c50c7SLars Ellenberg drbd_md_sync(mdev); 4666b411b363SPhilipp Reisner } 4667b411b363SPhilipp Reisner if (0) { 4668b411b363SPhilipp Reisner disconnect: 4669b411b363SPhilipp Reisner drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 4670856c50c7SLars Ellenberg drbd_md_sync(mdev); 4671b411b363SPhilipp Reisner } 4672b411b363SPhilipp Reisner clear_bit(SIGNAL_ASENDER, &mdev->flags); 4673b411b363SPhilipp Reisner 4674b411b363SPhilipp Reisner D_ASSERT(mdev->state.conn < C_CONNECTED); 4675b411b363SPhilipp Reisner dev_info(DEV, "asender terminated\n"); 4676b411b363SPhilipp Reisner 4677b411b363SPhilipp Reisner return 0; 4678b411b363SPhilipp Reisner } 4679