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 287e5fec31SFabian Frederick #include <linux/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> 39ae7e81c0SIngo Molnar #include <uapi/linux/sched/types.h> 40174cd4b1SIngo Molnar #include <linux/sched/signal.h> 41b411b363SPhilipp Reisner #include <linux/pkt_sched.h> 42b411b363SPhilipp Reisner #define __KERNEL_SYSCALLS__ 43b411b363SPhilipp Reisner #include <linux/unistd.h> 44b411b363SPhilipp Reisner #include <linux/vmalloc.h> 45b411b363SPhilipp Reisner #include <linux/random.h> 46b411b363SPhilipp Reisner #include <linux/string.h> 47b411b363SPhilipp Reisner #include <linux/scatterlist.h> 48b411b363SPhilipp Reisner #include "drbd_int.h" 49a3603a6eSAndreas Gruenbacher #include "drbd_protocol.h" 50b411b363SPhilipp Reisner #include "drbd_req.h" 51b411b363SPhilipp Reisner #include "drbd_vli.h" 52b411b363SPhilipp Reisner 539104d31aSLars Ellenberg #define PRO_FEATURES (DRBD_FF_TRIM|DRBD_FF_THIN_RESYNC|DRBD_FF_WSAME) 5420c68fdeSLars Ellenberg 5577351055SPhilipp Reisner struct packet_info { 5677351055SPhilipp Reisner enum drbd_packet cmd; 57e2857216SAndreas Gruenbacher unsigned int size; 58e2857216SAndreas Gruenbacher unsigned int vnr; 59e658983aSAndreas Gruenbacher void *data; 6077351055SPhilipp Reisner }; 6177351055SPhilipp Reisner 62b411b363SPhilipp Reisner enum finish_epoch { 63b411b363SPhilipp Reisner FE_STILL_LIVE, 64b411b363SPhilipp Reisner FE_DESTROYED, 65b411b363SPhilipp Reisner FE_RECYCLED, 66b411b363SPhilipp Reisner }; 67b411b363SPhilipp Reisner 68bde89a9eSAndreas Gruenbacher static int drbd_do_features(struct drbd_connection *connection); 69bde89a9eSAndreas Gruenbacher static int drbd_do_auth(struct drbd_connection *connection); 7069a22773SAndreas Gruenbacher static int drbd_disconnected(struct drbd_peer_device *); 71a0fb3c47SLars Ellenberg static void conn_wait_active_ee_empty(struct drbd_connection *connection); 72bde89a9eSAndreas Gruenbacher static enum finish_epoch drbd_may_finish_epoch(struct drbd_connection *, struct drbd_epoch *, enum epoch_event); 7399920dc5SAndreas Gruenbacher static int e_end_block(struct drbd_work *, int); 74b411b363SPhilipp Reisner 75b411b363SPhilipp Reisner 76b411b363SPhilipp Reisner #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) 77b411b363SPhilipp Reisner 7845bb912bSLars Ellenberg /* 7945bb912bSLars Ellenberg * some helper functions to deal with single linked page lists, 8045bb912bSLars Ellenberg * page->private being our "next" pointer. 8145bb912bSLars Ellenberg */ 8245bb912bSLars Ellenberg 8345bb912bSLars Ellenberg /* If at least n pages are linked at head, get n pages off. 8445bb912bSLars Ellenberg * Otherwise, don't modify head, and return NULL. 8545bb912bSLars Ellenberg * Locking is the responsibility of the caller. 8645bb912bSLars Ellenberg */ 8745bb912bSLars Ellenberg static struct page *page_chain_del(struct page **head, int n) 8845bb912bSLars Ellenberg { 8945bb912bSLars Ellenberg struct page *page; 9045bb912bSLars Ellenberg struct page *tmp; 9145bb912bSLars Ellenberg 9245bb912bSLars Ellenberg BUG_ON(!n); 9345bb912bSLars Ellenberg BUG_ON(!head); 9445bb912bSLars Ellenberg 9545bb912bSLars Ellenberg page = *head; 9623ce4227SPhilipp Reisner 9723ce4227SPhilipp Reisner if (!page) 9823ce4227SPhilipp Reisner return NULL; 9923ce4227SPhilipp Reisner 10045bb912bSLars Ellenberg while (page) { 10145bb912bSLars Ellenberg tmp = page_chain_next(page); 10245bb912bSLars Ellenberg if (--n == 0) 10345bb912bSLars Ellenberg break; /* found sufficient pages */ 10445bb912bSLars Ellenberg if (tmp == NULL) 10545bb912bSLars Ellenberg /* insufficient pages, don't use any of them. */ 10645bb912bSLars Ellenberg return NULL; 10745bb912bSLars Ellenberg page = tmp; 10845bb912bSLars Ellenberg } 10945bb912bSLars Ellenberg 11045bb912bSLars Ellenberg /* add end of list marker for the returned list */ 11145bb912bSLars Ellenberg set_page_private(page, 0); 11245bb912bSLars Ellenberg /* actual return value, and adjustment of head */ 11345bb912bSLars Ellenberg page = *head; 11445bb912bSLars Ellenberg *head = tmp; 11545bb912bSLars Ellenberg return page; 11645bb912bSLars Ellenberg } 11745bb912bSLars Ellenberg 11845bb912bSLars Ellenberg /* may be used outside of locks to find the tail of a (usually short) 11945bb912bSLars Ellenberg * "private" page chain, before adding it back to a global chain head 12045bb912bSLars Ellenberg * with page_chain_add() under a spinlock. */ 12145bb912bSLars Ellenberg static struct page *page_chain_tail(struct page *page, int *len) 12245bb912bSLars Ellenberg { 12345bb912bSLars Ellenberg struct page *tmp; 12445bb912bSLars Ellenberg int i = 1; 12545bb912bSLars Ellenberg while ((tmp = page_chain_next(page))) 12645bb912bSLars Ellenberg ++i, page = tmp; 12745bb912bSLars Ellenberg if (len) 12845bb912bSLars Ellenberg *len = i; 12945bb912bSLars Ellenberg return page; 13045bb912bSLars Ellenberg } 13145bb912bSLars Ellenberg 13245bb912bSLars Ellenberg static int page_chain_free(struct page *page) 13345bb912bSLars Ellenberg { 13445bb912bSLars Ellenberg struct page *tmp; 13545bb912bSLars Ellenberg int i = 0; 13645bb912bSLars Ellenberg page_chain_for_each_safe(page, tmp) { 13745bb912bSLars Ellenberg put_page(page); 13845bb912bSLars Ellenberg ++i; 13945bb912bSLars Ellenberg } 14045bb912bSLars Ellenberg return i; 14145bb912bSLars Ellenberg } 14245bb912bSLars Ellenberg 14345bb912bSLars Ellenberg static void page_chain_add(struct page **head, 14445bb912bSLars Ellenberg struct page *chain_first, struct page *chain_last) 14545bb912bSLars Ellenberg { 14645bb912bSLars Ellenberg #if 1 14745bb912bSLars Ellenberg struct page *tmp; 14845bb912bSLars Ellenberg tmp = page_chain_tail(chain_first, NULL); 14945bb912bSLars Ellenberg BUG_ON(tmp != chain_last); 15045bb912bSLars Ellenberg #endif 15145bb912bSLars Ellenberg 15245bb912bSLars Ellenberg /* add chain to head */ 15345bb912bSLars Ellenberg set_page_private(chain_last, (unsigned long)*head); 15445bb912bSLars Ellenberg *head = chain_first; 15545bb912bSLars Ellenberg } 15645bb912bSLars Ellenberg 157b30ab791SAndreas Gruenbacher static struct page *__drbd_alloc_pages(struct drbd_device *device, 15818c2d522SAndreas Gruenbacher unsigned int number) 159b411b363SPhilipp Reisner { 160b411b363SPhilipp Reisner struct page *page = NULL; 16145bb912bSLars Ellenberg struct page *tmp = NULL; 16218c2d522SAndreas Gruenbacher unsigned int i = 0; 163b411b363SPhilipp Reisner 164b411b363SPhilipp Reisner /* Yes, testing drbd_pp_vacant outside the lock is racy. 165b411b363SPhilipp Reisner * So what. It saves a spin_lock. */ 16645bb912bSLars Ellenberg if (drbd_pp_vacant >= number) { 167b411b363SPhilipp Reisner spin_lock(&drbd_pp_lock); 16845bb912bSLars Ellenberg page = page_chain_del(&drbd_pp_pool, number); 16945bb912bSLars Ellenberg if (page) 17045bb912bSLars Ellenberg drbd_pp_vacant -= number; 171b411b363SPhilipp Reisner spin_unlock(&drbd_pp_lock); 17245bb912bSLars Ellenberg if (page) 17345bb912bSLars Ellenberg return page; 174b411b363SPhilipp Reisner } 17545bb912bSLars Ellenberg 176b411b363SPhilipp Reisner /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD 177b411b363SPhilipp Reisner * "criss-cross" setup, that might cause write-out on some other DRBD, 178b411b363SPhilipp Reisner * which in turn might block on the other node at this very place. */ 17945bb912bSLars Ellenberg for (i = 0; i < number; i++) { 18045bb912bSLars Ellenberg tmp = alloc_page(GFP_TRY); 18145bb912bSLars Ellenberg if (!tmp) 18245bb912bSLars Ellenberg break; 18345bb912bSLars Ellenberg set_page_private(tmp, (unsigned long)page); 18445bb912bSLars Ellenberg page = tmp; 18545bb912bSLars Ellenberg } 18645bb912bSLars Ellenberg 18745bb912bSLars Ellenberg if (i == number) 188b411b363SPhilipp Reisner return page; 18945bb912bSLars Ellenberg 19045bb912bSLars Ellenberg /* Not enough pages immediately available this time. 191c37c8ecfSAndreas Gruenbacher * No need to jump around here, drbd_alloc_pages will retry this 19245bb912bSLars Ellenberg * function "soon". */ 19345bb912bSLars Ellenberg if (page) { 19445bb912bSLars Ellenberg tmp = page_chain_tail(page, NULL); 19545bb912bSLars Ellenberg spin_lock(&drbd_pp_lock); 19645bb912bSLars Ellenberg page_chain_add(&drbd_pp_pool, page, tmp); 19745bb912bSLars Ellenberg drbd_pp_vacant += i; 19845bb912bSLars Ellenberg spin_unlock(&drbd_pp_lock); 19945bb912bSLars Ellenberg } 20045bb912bSLars Ellenberg return NULL; 201b411b363SPhilipp Reisner } 202b411b363SPhilipp Reisner 203b30ab791SAndreas Gruenbacher static void reclaim_finished_net_peer_reqs(struct drbd_device *device, 204a990be46SAndreas Gruenbacher struct list_head *to_be_freed) 205b411b363SPhilipp Reisner { 206a8cd15baSAndreas Gruenbacher struct drbd_peer_request *peer_req, *tmp; 207b411b363SPhilipp Reisner 208b411b363SPhilipp Reisner /* The EEs are always appended to the end of the list. Since 209b411b363SPhilipp Reisner they are sent in order over the wire, they have to finish 210b411b363SPhilipp Reisner in order. As soon as we see the first not finished we can 211b411b363SPhilipp Reisner stop to examine the list... */ 212b411b363SPhilipp Reisner 213a8cd15baSAndreas Gruenbacher list_for_each_entry_safe(peer_req, tmp, &device->net_ee, w.list) { 214045417f7SAndreas Gruenbacher if (drbd_peer_req_has_active_page(peer_req)) 215b411b363SPhilipp Reisner break; 216a8cd15baSAndreas Gruenbacher list_move(&peer_req->w.list, to_be_freed); 217b411b363SPhilipp Reisner } 218b411b363SPhilipp Reisner } 219b411b363SPhilipp Reisner 220668700b4SPhilipp Reisner static void drbd_reclaim_net_peer_reqs(struct drbd_device *device) 221b411b363SPhilipp Reisner { 222b411b363SPhilipp Reisner LIST_HEAD(reclaimed); 223db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req, *t; 224b411b363SPhilipp Reisner 2250500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 226b30ab791SAndreas Gruenbacher reclaim_finished_net_peer_reqs(device, &reclaimed); 2270500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 228a8cd15baSAndreas Gruenbacher list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) 229b30ab791SAndreas Gruenbacher drbd_free_net_peer_req(device, peer_req); 230b411b363SPhilipp Reisner } 231b411b363SPhilipp Reisner 232668700b4SPhilipp Reisner static void conn_reclaim_net_peer_reqs(struct drbd_connection *connection) 233668700b4SPhilipp Reisner { 234668700b4SPhilipp Reisner struct drbd_peer_device *peer_device; 235668700b4SPhilipp Reisner int vnr; 236668700b4SPhilipp Reisner 237668700b4SPhilipp Reisner rcu_read_lock(); 238668700b4SPhilipp Reisner idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 239668700b4SPhilipp Reisner struct drbd_device *device = peer_device->device; 240668700b4SPhilipp Reisner if (!atomic_read(&device->pp_in_use_by_net)) 241668700b4SPhilipp Reisner continue; 242668700b4SPhilipp Reisner 243668700b4SPhilipp Reisner kref_get(&device->kref); 244668700b4SPhilipp Reisner rcu_read_unlock(); 245668700b4SPhilipp Reisner drbd_reclaim_net_peer_reqs(device); 246668700b4SPhilipp Reisner kref_put(&device->kref, drbd_destroy_device); 247668700b4SPhilipp Reisner rcu_read_lock(); 248668700b4SPhilipp Reisner } 249668700b4SPhilipp Reisner rcu_read_unlock(); 250668700b4SPhilipp Reisner } 251668700b4SPhilipp Reisner 252b411b363SPhilipp Reisner /** 253c37c8ecfSAndreas Gruenbacher * drbd_alloc_pages() - Returns @number pages, retries forever (or until signalled) 254b30ab791SAndreas Gruenbacher * @device: DRBD device. 25545bb912bSLars Ellenberg * @number: number of pages requested 25645bb912bSLars Ellenberg * @retry: whether to retry, if not enough pages are available right now 257b411b363SPhilipp Reisner * 25845bb912bSLars Ellenberg * Tries to allocate number pages, first from our own page pool, then from 2590e49d7b0SLars Ellenberg * the kernel. 26045bb912bSLars Ellenberg * Possibly retry until DRBD frees sufficient pages somewhere else. 26145bb912bSLars Ellenberg * 2620e49d7b0SLars Ellenberg * If this allocation would exceed the max_buffers setting, we throttle 2630e49d7b0SLars Ellenberg * allocation (schedule_timeout) to give the system some room to breathe. 2640e49d7b0SLars Ellenberg * 2650e49d7b0SLars Ellenberg * We do not use max-buffers as hard limit, because it could lead to 2660e49d7b0SLars Ellenberg * congestion and further to a distributed deadlock during online-verify or 2670e49d7b0SLars Ellenberg * (checksum based) resync, if the max-buffers, socket buffer sizes and 2680e49d7b0SLars Ellenberg * resync-rate settings are mis-configured. 2690e49d7b0SLars Ellenberg * 27045bb912bSLars Ellenberg * Returns a page chain linked via page->private. 271b411b363SPhilipp Reisner */ 27269a22773SAndreas Gruenbacher struct page *drbd_alloc_pages(struct drbd_peer_device *peer_device, unsigned int number, 273c37c8ecfSAndreas Gruenbacher bool retry) 274b411b363SPhilipp Reisner { 27569a22773SAndreas Gruenbacher struct drbd_device *device = peer_device->device; 276b411b363SPhilipp Reisner struct page *page = NULL; 27744ed167dSPhilipp Reisner struct net_conf *nc; 278b411b363SPhilipp Reisner DEFINE_WAIT(wait); 2790e49d7b0SLars Ellenberg unsigned int mxb; 280b411b363SPhilipp Reisner 28144ed167dSPhilipp Reisner rcu_read_lock(); 28269a22773SAndreas Gruenbacher nc = rcu_dereference(peer_device->connection->net_conf); 28344ed167dSPhilipp Reisner mxb = nc ? nc->max_buffers : 1000000; 28444ed167dSPhilipp Reisner rcu_read_unlock(); 28544ed167dSPhilipp Reisner 286b30ab791SAndreas Gruenbacher if (atomic_read(&device->pp_in_use) < mxb) 287b30ab791SAndreas Gruenbacher page = __drbd_alloc_pages(device, number); 288b411b363SPhilipp Reisner 289668700b4SPhilipp Reisner /* Try to keep the fast path fast, but occasionally we need 290668700b4SPhilipp Reisner * to reclaim the pages we lended to the network stack. */ 291668700b4SPhilipp Reisner if (page && atomic_read(&device->pp_in_use_by_net) > 512) 292668700b4SPhilipp Reisner drbd_reclaim_net_peer_reqs(device); 293668700b4SPhilipp Reisner 29445bb912bSLars Ellenberg while (page == NULL) { 295b411b363SPhilipp Reisner prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); 296b411b363SPhilipp Reisner 297668700b4SPhilipp Reisner drbd_reclaim_net_peer_reqs(device); 298b411b363SPhilipp Reisner 299b30ab791SAndreas Gruenbacher if (atomic_read(&device->pp_in_use) < mxb) { 300b30ab791SAndreas Gruenbacher page = __drbd_alloc_pages(device, number); 301b411b363SPhilipp Reisner if (page) 302b411b363SPhilipp Reisner break; 303b411b363SPhilipp Reisner } 304b411b363SPhilipp Reisner 305b411b363SPhilipp Reisner if (!retry) 306b411b363SPhilipp Reisner break; 307b411b363SPhilipp Reisner 308b411b363SPhilipp Reisner if (signal_pending(current)) { 309d0180171SAndreas Gruenbacher drbd_warn(device, "drbd_alloc_pages interrupted!\n"); 310b411b363SPhilipp Reisner break; 311b411b363SPhilipp Reisner } 312b411b363SPhilipp Reisner 3130e49d7b0SLars Ellenberg if (schedule_timeout(HZ/10) == 0) 3140e49d7b0SLars Ellenberg mxb = UINT_MAX; 315b411b363SPhilipp Reisner } 316b411b363SPhilipp Reisner finish_wait(&drbd_pp_wait, &wait); 317b411b363SPhilipp Reisner 31845bb912bSLars Ellenberg if (page) 319b30ab791SAndreas Gruenbacher atomic_add(number, &device->pp_in_use); 320b411b363SPhilipp Reisner return page; 321b411b363SPhilipp Reisner } 322b411b363SPhilipp Reisner 323c37c8ecfSAndreas Gruenbacher /* Must not be used from irq, as that may deadlock: see drbd_alloc_pages. 3240500813fSAndreas Gruenbacher * Is also used from inside an other spin_lock_irq(&resource->req_lock); 32545bb912bSLars Ellenberg * Either links the page chain back to the global pool, 32645bb912bSLars Ellenberg * or returns all pages to the system. */ 327b30ab791SAndreas Gruenbacher static void drbd_free_pages(struct drbd_device *device, struct page *page, int is_net) 328b411b363SPhilipp Reisner { 329b30ab791SAndreas Gruenbacher atomic_t *a = is_net ? &device->pp_in_use_by_net : &device->pp_in_use; 330b411b363SPhilipp Reisner int i; 331435f0740SLars Ellenberg 332a73ff323SLars Ellenberg if (page == NULL) 333a73ff323SLars Ellenberg return; 334a73ff323SLars Ellenberg 335183ece30SRoland Kammerer if (drbd_pp_vacant > (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count) 33645bb912bSLars Ellenberg i = page_chain_free(page); 33745bb912bSLars Ellenberg else { 33845bb912bSLars Ellenberg struct page *tmp; 33945bb912bSLars Ellenberg tmp = page_chain_tail(page, &i); 340b411b363SPhilipp Reisner spin_lock(&drbd_pp_lock); 34145bb912bSLars Ellenberg page_chain_add(&drbd_pp_pool, page, tmp); 34245bb912bSLars Ellenberg drbd_pp_vacant += i; 343b411b363SPhilipp Reisner spin_unlock(&drbd_pp_lock); 344b411b363SPhilipp Reisner } 345435f0740SLars Ellenberg i = atomic_sub_return(i, a); 34645bb912bSLars Ellenberg if (i < 0) 347d0180171SAndreas Gruenbacher drbd_warn(device, "ASSERTION FAILED: %s: %d < 0\n", 348435f0740SLars Ellenberg is_net ? "pp_in_use_by_net" : "pp_in_use", i); 349b411b363SPhilipp Reisner wake_up(&drbd_pp_wait); 350b411b363SPhilipp Reisner } 351b411b363SPhilipp Reisner 352b411b363SPhilipp Reisner /* 353b411b363SPhilipp Reisner You need to hold the req_lock: 354b411b363SPhilipp Reisner _drbd_wait_ee_list_empty() 355b411b363SPhilipp Reisner 356b411b363SPhilipp Reisner You must not have the req_lock: 3573967deb1SAndreas Gruenbacher drbd_free_peer_req() 3580db55363SAndreas Gruenbacher drbd_alloc_peer_req() 3597721f567SAndreas Gruenbacher drbd_free_peer_reqs() 360b411b363SPhilipp Reisner drbd_ee_fix_bhs() 361a990be46SAndreas Gruenbacher drbd_finish_peer_reqs() 362b411b363SPhilipp Reisner drbd_clear_done_ee() 363b411b363SPhilipp Reisner drbd_wait_ee_list_empty() 364b411b363SPhilipp Reisner */ 365b411b363SPhilipp Reisner 3669104d31aSLars Ellenberg /* normal: payload_size == request size (bi_size) 3679104d31aSLars Ellenberg * w_same: payload_size == logical_block_size 3689104d31aSLars Ellenberg * trim: payload_size == 0 */ 369f6ffca9fSAndreas Gruenbacher struct drbd_peer_request * 37069a22773SAndreas Gruenbacher drbd_alloc_peer_req(struct drbd_peer_device *peer_device, u64 id, sector_t sector, 3719104d31aSLars Ellenberg unsigned int request_size, unsigned int payload_size, gfp_t gfp_mask) __must_hold(local) 372b411b363SPhilipp Reisner { 37369a22773SAndreas Gruenbacher struct drbd_device *device = peer_device->device; 374db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req; 375a73ff323SLars Ellenberg struct page *page = NULL; 3769104d31aSLars Ellenberg unsigned nr_pages = (payload_size + PAGE_SIZE -1) >> PAGE_SHIFT; 377b411b363SPhilipp Reisner 378b30ab791SAndreas Gruenbacher if (drbd_insert_fault(device, DRBD_FAULT_AL_EE)) 379b411b363SPhilipp Reisner return NULL; 380b411b363SPhilipp Reisner 381db830c46SAndreas Gruenbacher peer_req = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); 382db830c46SAndreas Gruenbacher if (!peer_req) { 383b411b363SPhilipp Reisner if (!(gfp_mask & __GFP_NOWARN)) 384d0180171SAndreas Gruenbacher drbd_err(device, "%s: allocation failed\n", __func__); 385b411b363SPhilipp Reisner return NULL; 386b411b363SPhilipp Reisner } 387b411b363SPhilipp Reisner 3889104d31aSLars Ellenberg if (nr_pages) { 389d0164adcSMel Gorman page = drbd_alloc_pages(peer_device, nr_pages, 390d0164adcSMel Gorman gfpflags_allow_blocking(gfp_mask)); 39145bb912bSLars Ellenberg if (!page) 39245bb912bSLars Ellenberg goto fail; 393a73ff323SLars Ellenberg } 394b411b363SPhilipp Reisner 395c5a2c150SLars Ellenberg memset(peer_req, 0, sizeof(*peer_req)); 396c5a2c150SLars Ellenberg INIT_LIST_HEAD(&peer_req->w.list); 397db830c46SAndreas Gruenbacher drbd_clear_interval(&peer_req->i); 3989104d31aSLars Ellenberg peer_req->i.size = request_size; 399db830c46SAndreas Gruenbacher peer_req->i.sector = sector; 400c5a2c150SLars Ellenberg peer_req->submit_jif = jiffies; 401a8cd15baSAndreas Gruenbacher peer_req->peer_device = peer_device; 402db830c46SAndreas Gruenbacher peer_req->pages = page; 4039a8e7753SAndreas Gruenbacher /* 4049a8e7753SAndreas Gruenbacher * The block_id is opaque to the receiver. It is not endianness 4059a8e7753SAndreas Gruenbacher * converted, and sent back to the sender unchanged. 4069a8e7753SAndreas Gruenbacher */ 407db830c46SAndreas Gruenbacher peer_req->block_id = id; 408b411b363SPhilipp Reisner 409db830c46SAndreas Gruenbacher return peer_req; 410b411b363SPhilipp Reisner 41145bb912bSLars Ellenberg fail: 412db830c46SAndreas Gruenbacher mempool_free(peer_req, drbd_ee_mempool); 413b411b363SPhilipp Reisner return NULL; 414b411b363SPhilipp Reisner } 415b411b363SPhilipp Reisner 416b30ab791SAndreas Gruenbacher void __drbd_free_peer_req(struct drbd_device *device, struct drbd_peer_request *peer_req, 417f6ffca9fSAndreas Gruenbacher int is_net) 418b411b363SPhilipp Reisner { 41921ae5d7fSLars Ellenberg might_sleep(); 420db830c46SAndreas Gruenbacher if (peer_req->flags & EE_HAS_DIGEST) 421db830c46SAndreas Gruenbacher kfree(peer_req->digest); 422b30ab791SAndreas Gruenbacher drbd_free_pages(device, peer_req->pages, is_net); 4230b0ba1efSAndreas Gruenbacher D_ASSERT(device, atomic_read(&peer_req->pending_bios) == 0); 4240b0ba1efSAndreas Gruenbacher D_ASSERT(device, drbd_interval_empty(&peer_req->i)); 42521ae5d7fSLars Ellenberg if (!expect(!(peer_req->flags & EE_CALL_AL_COMPLETE_IO))) { 42621ae5d7fSLars Ellenberg peer_req->flags &= ~EE_CALL_AL_COMPLETE_IO; 42721ae5d7fSLars Ellenberg drbd_al_complete_io(device, &peer_req->i); 42821ae5d7fSLars Ellenberg } 429db830c46SAndreas Gruenbacher mempool_free(peer_req, drbd_ee_mempool); 430b411b363SPhilipp Reisner } 431b411b363SPhilipp Reisner 432b30ab791SAndreas Gruenbacher int drbd_free_peer_reqs(struct drbd_device *device, struct list_head *list) 433b411b363SPhilipp Reisner { 434b411b363SPhilipp Reisner LIST_HEAD(work_list); 435db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req, *t; 436b411b363SPhilipp Reisner int count = 0; 437b30ab791SAndreas Gruenbacher int is_net = list == &device->net_ee; 438b411b363SPhilipp Reisner 4390500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 440b411b363SPhilipp Reisner list_splice_init(list, &work_list); 4410500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 442b411b363SPhilipp Reisner 443a8cd15baSAndreas Gruenbacher list_for_each_entry_safe(peer_req, t, &work_list, w.list) { 444b30ab791SAndreas Gruenbacher __drbd_free_peer_req(device, peer_req, is_net); 445b411b363SPhilipp Reisner count++; 446b411b363SPhilipp Reisner } 447b411b363SPhilipp Reisner return count; 448b411b363SPhilipp Reisner } 449b411b363SPhilipp Reisner 450b411b363SPhilipp Reisner /* 451a990be46SAndreas Gruenbacher * See also comments in _req_mod(,BARRIER_ACKED) and receive_Barrier. 452b411b363SPhilipp Reisner */ 453b30ab791SAndreas Gruenbacher static int drbd_finish_peer_reqs(struct drbd_device *device) 454b411b363SPhilipp Reisner { 455b411b363SPhilipp Reisner LIST_HEAD(work_list); 456b411b363SPhilipp Reisner LIST_HEAD(reclaimed); 457db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req, *t; 458e2b3032bSAndreas Gruenbacher int err = 0; 459b411b363SPhilipp Reisner 4600500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 461b30ab791SAndreas Gruenbacher reclaim_finished_net_peer_reqs(device, &reclaimed); 462b30ab791SAndreas Gruenbacher list_splice_init(&device->done_ee, &work_list); 4630500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 464b411b363SPhilipp Reisner 465a8cd15baSAndreas Gruenbacher list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) 466b30ab791SAndreas Gruenbacher drbd_free_net_peer_req(device, peer_req); 467b411b363SPhilipp Reisner 468b411b363SPhilipp Reisner /* possible callbacks here: 469d4dabbe2SLars Ellenberg * e_end_block, and e_end_resync_block, e_send_superseded. 470b411b363SPhilipp Reisner * all ignore the last argument. 471b411b363SPhilipp Reisner */ 472a8cd15baSAndreas Gruenbacher list_for_each_entry_safe(peer_req, t, &work_list, w.list) { 473e2b3032bSAndreas Gruenbacher int err2; 474e2b3032bSAndreas Gruenbacher 475b411b363SPhilipp Reisner /* list_del not necessary, next/prev members not touched */ 476a8cd15baSAndreas Gruenbacher err2 = peer_req->w.cb(&peer_req->w, !!err); 477e2b3032bSAndreas Gruenbacher if (!err) 478e2b3032bSAndreas Gruenbacher err = err2; 479b30ab791SAndreas Gruenbacher drbd_free_peer_req(device, peer_req); 480b411b363SPhilipp Reisner } 481b30ab791SAndreas Gruenbacher wake_up(&device->ee_wait); 482b411b363SPhilipp Reisner 483e2b3032bSAndreas Gruenbacher return err; 484b411b363SPhilipp Reisner } 485b411b363SPhilipp Reisner 486b30ab791SAndreas Gruenbacher static void _drbd_wait_ee_list_empty(struct drbd_device *device, 487d4da1537SAndreas Gruenbacher struct list_head *head) 488b411b363SPhilipp Reisner { 489b411b363SPhilipp Reisner DEFINE_WAIT(wait); 490b411b363SPhilipp Reisner 491b411b363SPhilipp Reisner /* avoids spin_lock/unlock 492b411b363SPhilipp Reisner * and calling prepare_to_wait in the fast path */ 493b411b363SPhilipp Reisner while (!list_empty(head)) { 494b30ab791SAndreas Gruenbacher prepare_to_wait(&device->ee_wait, &wait, TASK_UNINTERRUPTIBLE); 4950500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 4967eaceaccSJens Axboe io_schedule(); 497b30ab791SAndreas Gruenbacher finish_wait(&device->ee_wait, &wait); 4980500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 499b411b363SPhilipp Reisner } 500b411b363SPhilipp Reisner } 501b411b363SPhilipp Reisner 502b30ab791SAndreas Gruenbacher static void drbd_wait_ee_list_empty(struct drbd_device *device, 503d4da1537SAndreas Gruenbacher struct list_head *head) 504b411b363SPhilipp Reisner { 5050500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 506b30ab791SAndreas Gruenbacher _drbd_wait_ee_list_empty(device, head); 5070500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 508b411b363SPhilipp Reisner } 509b411b363SPhilipp Reisner 510dbd9eea0SPhilipp Reisner static int drbd_recv_short(struct socket *sock, void *buf, size_t size, int flags) 511b411b363SPhilipp Reisner { 512b411b363SPhilipp Reisner struct kvec iov = { 513b411b363SPhilipp Reisner .iov_base = buf, 514b411b363SPhilipp Reisner .iov_len = size, 515b411b363SPhilipp Reisner }; 516b411b363SPhilipp Reisner struct msghdr msg = { 517b411b363SPhilipp Reisner .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) 518b411b363SPhilipp Reisner }; 519f730c848SAl Viro return kernel_recvmsg(sock, &msg, &iov, 1, size, msg.msg_flags); 520b411b363SPhilipp Reisner } 521b411b363SPhilipp Reisner 522bde89a9eSAndreas Gruenbacher static int drbd_recv(struct drbd_connection *connection, void *buf, size_t size) 523b411b363SPhilipp Reisner { 524b411b363SPhilipp Reisner int rv; 525b411b363SPhilipp Reisner 526bde89a9eSAndreas Gruenbacher rv = drbd_recv_short(connection->data.socket, buf, size, 0); 527b411b363SPhilipp Reisner 528b411b363SPhilipp Reisner if (rv < 0) { 529b411b363SPhilipp Reisner if (rv == -ECONNRESET) 5301ec861ebSAndreas Gruenbacher drbd_info(connection, "sock was reset by peer\n"); 531b411b363SPhilipp Reisner else if (rv != -ERESTARTSYS) 5321ec861ebSAndreas Gruenbacher drbd_err(connection, "sock_recvmsg returned %d\n", rv); 533b411b363SPhilipp Reisner } else if (rv == 0) { 534bde89a9eSAndreas Gruenbacher if (test_bit(DISCONNECT_SENT, &connection->flags)) { 535b66623e3SPhilipp Reisner long t; 536b66623e3SPhilipp Reisner rcu_read_lock(); 537bde89a9eSAndreas Gruenbacher t = rcu_dereference(connection->net_conf)->ping_timeo * HZ/10; 538b66623e3SPhilipp Reisner rcu_read_unlock(); 539b66623e3SPhilipp Reisner 540bde89a9eSAndreas Gruenbacher t = wait_event_timeout(connection->ping_wait, connection->cstate < C_WF_REPORT_PARAMS, t); 541b66623e3SPhilipp Reisner 542599377acSPhilipp Reisner if (t) 543599377acSPhilipp Reisner goto out; 544599377acSPhilipp Reisner } 5451ec861ebSAndreas Gruenbacher drbd_info(connection, "sock was shut down by peer\n"); 546599377acSPhilipp Reisner } 547599377acSPhilipp Reisner 548b411b363SPhilipp Reisner if (rv != size) 549bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD); 550b411b363SPhilipp Reisner 551599377acSPhilipp Reisner out: 552b411b363SPhilipp Reisner return rv; 553b411b363SPhilipp Reisner } 554b411b363SPhilipp Reisner 555bde89a9eSAndreas Gruenbacher static int drbd_recv_all(struct drbd_connection *connection, void *buf, size_t size) 556c6967746SAndreas Gruenbacher { 557c6967746SAndreas Gruenbacher int err; 558c6967746SAndreas Gruenbacher 559bde89a9eSAndreas Gruenbacher err = drbd_recv(connection, buf, size); 560c6967746SAndreas Gruenbacher if (err != size) { 561c6967746SAndreas Gruenbacher if (err >= 0) 562c6967746SAndreas Gruenbacher err = -EIO; 563c6967746SAndreas Gruenbacher } else 564c6967746SAndreas Gruenbacher err = 0; 565c6967746SAndreas Gruenbacher return err; 566c6967746SAndreas Gruenbacher } 567c6967746SAndreas Gruenbacher 568bde89a9eSAndreas Gruenbacher static int drbd_recv_all_warn(struct drbd_connection *connection, void *buf, size_t size) 569a5c31904SAndreas Gruenbacher { 570a5c31904SAndreas Gruenbacher int err; 571a5c31904SAndreas Gruenbacher 572bde89a9eSAndreas Gruenbacher err = drbd_recv_all(connection, buf, size); 573a5c31904SAndreas Gruenbacher if (err && !signal_pending(current)) 5741ec861ebSAndreas Gruenbacher drbd_warn(connection, "short read (expected size %d)\n", (int)size); 575a5c31904SAndreas Gruenbacher return err; 576a5c31904SAndreas Gruenbacher } 577a5c31904SAndreas Gruenbacher 5785dbf1673SLars Ellenberg /* quoting tcp(7): 5795dbf1673SLars Ellenberg * On individual connections, the socket buffer size must be set prior to the 5805dbf1673SLars Ellenberg * listen(2) or connect(2) calls in order to have it take effect. 5815dbf1673SLars Ellenberg * This is our wrapper to do so. 5825dbf1673SLars Ellenberg */ 5835dbf1673SLars Ellenberg static void drbd_setbufsize(struct socket *sock, unsigned int snd, 5845dbf1673SLars Ellenberg unsigned int rcv) 5855dbf1673SLars Ellenberg { 5865dbf1673SLars Ellenberg /* open coded SO_SNDBUF, SO_RCVBUF */ 5875dbf1673SLars Ellenberg if (snd) { 5885dbf1673SLars Ellenberg sock->sk->sk_sndbuf = snd; 5895dbf1673SLars Ellenberg sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 5905dbf1673SLars Ellenberg } 5915dbf1673SLars Ellenberg if (rcv) { 5925dbf1673SLars Ellenberg sock->sk->sk_rcvbuf = rcv; 5935dbf1673SLars Ellenberg sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 5945dbf1673SLars Ellenberg } 5955dbf1673SLars Ellenberg } 5965dbf1673SLars Ellenberg 597bde89a9eSAndreas Gruenbacher static struct socket *drbd_try_connect(struct drbd_connection *connection) 598b411b363SPhilipp Reisner { 599b411b363SPhilipp Reisner const char *what; 600b411b363SPhilipp Reisner struct socket *sock; 601b411b363SPhilipp Reisner struct sockaddr_in6 src_in6; 60244ed167dSPhilipp Reisner struct sockaddr_in6 peer_in6; 60344ed167dSPhilipp Reisner struct net_conf *nc; 60444ed167dSPhilipp Reisner int err, peer_addr_len, my_addr_len; 60569ef82deSAndreas Gruenbacher int sndbuf_size, rcvbuf_size, connect_int; 606b411b363SPhilipp Reisner int disconnect_on_error = 1; 607b411b363SPhilipp Reisner 60844ed167dSPhilipp Reisner rcu_read_lock(); 609bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 61044ed167dSPhilipp Reisner if (!nc) { 61144ed167dSPhilipp Reisner rcu_read_unlock(); 612b411b363SPhilipp Reisner return NULL; 61344ed167dSPhilipp Reisner } 61444ed167dSPhilipp Reisner sndbuf_size = nc->sndbuf_size; 61544ed167dSPhilipp Reisner rcvbuf_size = nc->rcvbuf_size; 61669ef82deSAndreas Gruenbacher connect_int = nc->connect_int; 617089c075dSAndreas Gruenbacher rcu_read_unlock(); 61844ed167dSPhilipp Reisner 619bde89a9eSAndreas Gruenbacher my_addr_len = min_t(int, connection->my_addr_len, sizeof(src_in6)); 620bde89a9eSAndreas Gruenbacher memcpy(&src_in6, &connection->my_addr, my_addr_len); 62144ed167dSPhilipp Reisner 622bde89a9eSAndreas Gruenbacher if (((struct sockaddr *)&connection->my_addr)->sa_family == AF_INET6) 62344ed167dSPhilipp Reisner src_in6.sin6_port = 0; 62444ed167dSPhilipp Reisner else 62544ed167dSPhilipp Reisner ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ 62644ed167dSPhilipp Reisner 627bde89a9eSAndreas Gruenbacher peer_addr_len = min_t(int, connection->peer_addr_len, sizeof(src_in6)); 628bde89a9eSAndreas Gruenbacher memcpy(&peer_in6, &connection->peer_addr, peer_addr_len); 629b411b363SPhilipp Reisner 630b411b363SPhilipp Reisner what = "sock_create_kern"; 631eeb1bd5cSEric W. Biederman err = sock_create_kern(&init_net, ((struct sockaddr *)&src_in6)->sa_family, 632b411b363SPhilipp Reisner SOCK_STREAM, IPPROTO_TCP, &sock); 633b411b363SPhilipp Reisner if (err < 0) { 634b411b363SPhilipp Reisner sock = NULL; 635b411b363SPhilipp Reisner goto out; 636b411b363SPhilipp Reisner } 637b411b363SPhilipp Reisner 638b411b363SPhilipp Reisner sock->sk->sk_rcvtimeo = 63969ef82deSAndreas Gruenbacher sock->sk->sk_sndtimeo = connect_int * HZ; 64044ed167dSPhilipp Reisner drbd_setbufsize(sock, sndbuf_size, rcvbuf_size); 641b411b363SPhilipp Reisner 642b411b363SPhilipp Reisner /* explicitly bind to the configured IP as source IP 643b411b363SPhilipp Reisner * for the outgoing connections. 644b411b363SPhilipp Reisner * This is needed for multihomed hosts and to be 645b411b363SPhilipp Reisner * able to use lo: interfaces for drbd. 646b411b363SPhilipp Reisner * Make sure to use 0 as port number, so linux selects 647b411b363SPhilipp Reisner * a free one dynamically. 648b411b363SPhilipp Reisner */ 649b411b363SPhilipp Reisner what = "bind before connect"; 65044ed167dSPhilipp Reisner err = sock->ops->bind(sock, (struct sockaddr *) &src_in6, my_addr_len); 651b411b363SPhilipp Reisner if (err < 0) 652b411b363SPhilipp Reisner goto out; 653b411b363SPhilipp Reisner 654b411b363SPhilipp Reisner /* connect may fail, peer not yet available. 655b411b363SPhilipp Reisner * stay C_WF_CONNECTION, don't go Disconnecting! */ 656b411b363SPhilipp Reisner disconnect_on_error = 0; 657b411b363SPhilipp Reisner what = "connect"; 65844ed167dSPhilipp Reisner err = sock->ops->connect(sock, (struct sockaddr *) &peer_in6, peer_addr_len, 0); 659b411b363SPhilipp Reisner 660b411b363SPhilipp Reisner out: 661b411b363SPhilipp Reisner if (err < 0) { 662b411b363SPhilipp Reisner if (sock) { 663b411b363SPhilipp Reisner sock_release(sock); 664b411b363SPhilipp Reisner sock = NULL; 665b411b363SPhilipp Reisner } 666b411b363SPhilipp Reisner switch (-err) { 667b411b363SPhilipp Reisner /* timeout, busy, signal pending */ 668b411b363SPhilipp Reisner case ETIMEDOUT: case EAGAIN: case EINPROGRESS: 669b411b363SPhilipp Reisner case EINTR: case ERESTARTSYS: 670b411b363SPhilipp Reisner /* peer not (yet) available, network problem */ 671b411b363SPhilipp Reisner case ECONNREFUSED: case ENETUNREACH: 672b411b363SPhilipp Reisner case EHOSTDOWN: case EHOSTUNREACH: 673b411b363SPhilipp Reisner disconnect_on_error = 0; 674b411b363SPhilipp Reisner break; 675b411b363SPhilipp Reisner default: 6761ec861ebSAndreas Gruenbacher drbd_err(connection, "%s failed, err = %d\n", what, err); 677b411b363SPhilipp Reisner } 678b411b363SPhilipp Reisner if (disconnect_on_error) 679bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); 680b411b363SPhilipp Reisner } 68144ed167dSPhilipp Reisner 682b411b363SPhilipp Reisner return sock; 683b411b363SPhilipp Reisner } 684b411b363SPhilipp Reisner 6857a426fd8SPhilipp Reisner struct accept_wait_data { 686bde89a9eSAndreas Gruenbacher struct drbd_connection *connection; 6877a426fd8SPhilipp Reisner struct socket *s_listen; 6887a426fd8SPhilipp Reisner struct completion door_bell; 6897a426fd8SPhilipp Reisner void (*original_sk_state_change)(struct sock *sk); 6907a426fd8SPhilipp Reisner 6917a426fd8SPhilipp Reisner }; 6927a426fd8SPhilipp Reisner 693715306f6SAndreas Gruenbacher static void drbd_incoming_connection(struct sock *sk) 694b411b363SPhilipp Reisner { 6957a426fd8SPhilipp Reisner struct accept_wait_data *ad = sk->sk_user_data; 696715306f6SAndreas Gruenbacher void (*state_change)(struct sock *sk); 6977a426fd8SPhilipp Reisner 698715306f6SAndreas Gruenbacher state_change = ad->original_sk_state_change; 699715306f6SAndreas Gruenbacher if (sk->sk_state == TCP_ESTABLISHED) 7007a426fd8SPhilipp Reisner complete(&ad->door_bell); 701715306f6SAndreas Gruenbacher state_change(sk); 7027a426fd8SPhilipp Reisner } 7037a426fd8SPhilipp Reisner 704bde89a9eSAndreas Gruenbacher static int prepare_listen_socket(struct drbd_connection *connection, struct accept_wait_data *ad) 705b411b363SPhilipp Reisner { 7061f3e509bSPhilipp Reisner int err, sndbuf_size, rcvbuf_size, my_addr_len; 70744ed167dSPhilipp Reisner struct sockaddr_in6 my_addr; 7081f3e509bSPhilipp Reisner struct socket *s_listen; 70944ed167dSPhilipp Reisner struct net_conf *nc; 710b411b363SPhilipp Reisner const char *what; 711b411b363SPhilipp Reisner 71244ed167dSPhilipp Reisner rcu_read_lock(); 713bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 71444ed167dSPhilipp Reisner if (!nc) { 71544ed167dSPhilipp Reisner rcu_read_unlock(); 7167a426fd8SPhilipp Reisner return -EIO; 71744ed167dSPhilipp Reisner } 71844ed167dSPhilipp Reisner sndbuf_size = nc->sndbuf_size; 71944ed167dSPhilipp Reisner rcvbuf_size = nc->rcvbuf_size; 72044ed167dSPhilipp Reisner rcu_read_unlock(); 721b411b363SPhilipp Reisner 722bde89a9eSAndreas Gruenbacher my_addr_len = min_t(int, connection->my_addr_len, sizeof(struct sockaddr_in6)); 723bde89a9eSAndreas Gruenbacher memcpy(&my_addr, &connection->my_addr, my_addr_len); 724b411b363SPhilipp Reisner 725b411b363SPhilipp Reisner what = "sock_create_kern"; 726eeb1bd5cSEric W. Biederman err = sock_create_kern(&init_net, ((struct sockaddr *)&my_addr)->sa_family, 727b411b363SPhilipp Reisner SOCK_STREAM, IPPROTO_TCP, &s_listen); 728b411b363SPhilipp Reisner if (err) { 729b411b363SPhilipp Reisner s_listen = NULL; 730b411b363SPhilipp Reisner goto out; 731b411b363SPhilipp Reisner } 732b411b363SPhilipp Reisner 7334a17fd52SPavel Emelyanov s_listen->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ 73444ed167dSPhilipp Reisner drbd_setbufsize(s_listen, sndbuf_size, rcvbuf_size); 735b411b363SPhilipp Reisner 736b411b363SPhilipp Reisner what = "bind before listen"; 73744ed167dSPhilipp Reisner err = s_listen->ops->bind(s_listen, (struct sockaddr *)&my_addr, my_addr_len); 738b411b363SPhilipp Reisner if (err < 0) 739b411b363SPhilipp Reisner goto out; 740b411b363SPhilipp Reisner 7417a426fd8SPhilipp Reisner ad->s_listen = s_listen; 7427a426fd8SPhilipp Reisner write_lock_bh(&s_listen->sk->sk_callback_lock); 7437a426fd8SPhilipp Reisner ad->original_sk_state_change = s_listen->sk->sk_state_change; 744715306f6SAndreas Gruenbacher s_listen->sk->sk_state_change = drbd_incoming_connection; 7457a426fd8SPhilipp Reisner s_listen->sk->sk_user_data = ad; 7467a426fd8SPhilipp Reisner write_unlock_bh(&s_listen->sk->sk_callback_lock); 747b411b363SPhilipp Reisner 7482820fd39SPhilipp Reisner what = "listen"; 7492820fd39SPhilipp Reisner err = s_listen->ops->listen(s_listen, 5); 7502820fd39SPhilipp Reisner if (err < 0) 7512820fd39SPhilipp Reisner goto out; 7522820fd39SPhilipp Reisner 7537a426fd8SPhilipp Reisner return 0; 754b411b363SPhilipp Reisner out: 755b411b363SPhilipp Reisner if (s_listen) 756b411b363SPhilipp Reisner sock_release(s_listen); 757b411b363SPhilipp Reisner if (err < 0) { 758b411b363SPhilipp Reisner if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { 7591ec861ebSAndreas Gruenbacher drbd_err(connection, "%s failed, err = %d\n", what, err); 760bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); 761b411b363SPhilipp Reisner } 762b411b363SPhilipp Reisner } 7631f3e509bSPhilipp Reisner 7647a426fd8SPhilipp Reisner return -EIO; 7651f3e509bSPhilipp Reisner } 7661f3e509bSPhilipp Reisner 767715306f6SAndreas Gruenbacher static void unregister_state_change(struct sock *sk, struct accept_wait_data *ad) 768715306f6SAndreas Gruenbacher { 769715306f6SAndreas Gruenbacher write_lock_bh(&sk->sk_callback_lock); 770715306f6SAndreas Gruenbacher sk->sk_state_change = ad->original_sk_state_change; 771715306f6SAndreas Gruenbacher sk->sk_user_data = NULL; 772715306f6SAndreas Gruenbacher write_unlock_bh(&sk->sk_callback_lock); 773715306f6SAndreas Gruenbacher } 774715306f6SAndreas Gruenbacher 775bde89a9eSAndreas Gruenbacher static struct socket *drbd_wait_for_connect(struct drbd_connection *connection, struct accept_wait_data *ad) 7761f3e509bSPhilipp Reisner { 7771f3e509bSPhilipp Reisner int timeo, connect_int, err = 0; 7781f3e509bSPhilipp Reisner struct socket *s_estab = NULL; 7791f3e509bSPhilipp Reisner struct net_conf *nc; 7801f3e509bSPhilipp Reisner 7811f3e509bSPhilipp Reisner rcu_read_lock(); 782bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 7831f3e509bSPhilipp Reisner if (!nc) { 7841f3e509bSPhilipp Reisner rcu_read_unlock(); 7851f3e509bSPhilipp Reisner return NULL; 7861f3e509bSPhilipp Reisner } 7871f3e509bSPhilipp Reisner connect_int = nc->connect_int; 7881f3e509bSPhilipp Reisner rcu_read_unlock(); 7891f3e509bSPhilipp Reisner 7901f3e509bSPhilipp Reisner timeo = connect_int * HZ; 79138b682b2SAkinobu Mita /* 28.5% random jitter */ 79238b682b2SAkinobu Mita timeo += (prandom_u32() & 1) ? timeo / 7 : -timeo / 7; 7931f3e509bSPhilipp Reisner 7947a426fd8SPhilipp Reisner err = wait_for_completion_interruptible_timeout(&ad->door_bell, timeo); 7957a426fd8SPhilipp Reisner if (err <= 0) 7967a426fd8SPhilipp Reisner return NULL; 7971f3e509bSPhilipp Reisner 7987a426fd8SPhilipp Reisner err = kernel_accept(ad->s_listen, &s_estab, 0); 799b411b363SPhilipp Reisner if (err < 0) { 800b411b363SPhilipp Reisner if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { 8011ec861ebSAndreas Gruenbacher drbd_err(connection, "accept failed, err = %d\n", err); 802bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); 803b411b363SPhilipp Reisner } 804b411b363SPhilipp Reisner } 805b411b363SPhilipp Reisner 806715306f6SAndreas Gruenbacher if (s_estab) 807715306f6SAndreas Gruenbacher unregister_state_change(s_estab->sk, ad); 808b411b363SPhilipp Reisner 809b411b363SPhilipp Reisner return s_estab; 810b411b363SPhilipp Reisner } 811b411b363SPhilipp Reisner 812bde89a9eSAndreas Gruenbacher static int decode_header(struct drbd_connection *, void *, struct packet_info *); 813b411b363SPhilipp Reisner 814bde89a9eSAndreas Gruenbacher static int send_first_packet(struct drbd_connection *connection, struct drbd_socket *sock, 8159f5bdc33SAndreas Gruenbacher enum drbd_packet cmd) 8169f5bdc33SAndreas Gruenbacher { 817bde89a9eSAndreas Gruenbacher if (!conn_prepare_command(connection, sock)) 8189f5bdc33SAndreas Gruenbacher return -EIO; 819bde89a9eSAndreas Gruenbacher return conn_send_command(connection, sock, cmd, 0, NULL, 0); 820b411b363SPhilipp Reisner } 821b411b363SPhilipp Reisner 822bde89a9eSAndreas Gruenbacher static int receive_first_packet(struct drbd_connection *connection, struct socket *sock) 823b411b363SPhilipp Reisner { 824bde89a9eSAndreas Gruenbacher unsigned int header_size = drbd_header_size(connection); 8259f5bdc33SAndreas Gruenbacher struct packet_info pi; 8264920e37aSPhilipp Reisner struct net_conf *nc; 8279f5bdc33SAndreas Gruenbacher int err; 828b411b363SPhilipp Reisner 8294920e37aSPhilipp Reisner rcu_read_lock(); 8304920e37aSPhilipp Reisner nc = rcu_dereference(connection->net_conf); 8314920e37aSPhilipp Reisner if (!nc) { 8324920e37aSPhilipp Reisner rcu_read_unlock(); 8334920e37aSPhilipp Reisner return -EIO; 8344920e37aSPhilipp Reisner } 8354920e37aSPhilipp Reisner sock->sk->sk_rcvtimeo = nc->ping_timeo * 4 * HZ / 10; 8364920e37aSPhilipp Reisner rcu_read_unlock(); 8374920e37aSPhilipp Reisner 838bde89a9eSAndreas Gruenbacher err = drbd_recv_short(sock, connection->data.rbuf, header_size, 0); 8399f5bdc33SAndreas Gruenbacher if (err != header_size) { 8409f5bdc33SAndreas Gruenbacher if (err >= 0) 8419f5bdc33SAndreas Gruenbacher err = -EIO; 8429f5bdc33SAndreas Gruenbacher return err; 8439f5bdc33SAndreas Gruenbacher } 844bde89a9eSAndreas Gruenbacher err = decode_header(connection, connection->data.rbuf, &pi); 8459f5bdc33SAndreas Gruenbacher if (err) 8469f5bdc33SAndreas Gruenbacher return err; 8479f5bdc33SAndreas Gruenbacher return pi.cmd; 848b411b363SPhilipp Reisner } 849b411b363SPhilipp Reisner 850b411b363SPhilipp Reisner /** 851b411b363SPhilipp Reisner * drbd_socket_okay() - Free the socket if its connection is not okay 852b411b363SPhilipp Reisner * @sock: pointer to the pointer to the socket. 853b411b363SPhilipp Reisner */ 8545d0b17f1SPhilipp Reisner static bool drbd_socket_okay(struct socket **sock) 855b411b363SPhilipp Reisner { 856b411b363SPhilipp Reisner int rr; 857b411b363SPhilipp Reisner char tb[4]; 858b411b363SPhilipp Reisner 859b411b363SPhilipp Reisner if (!*sock) 86081e84650SAndreas Gruenbacher return false; 861b411b363SPhilipp Reisner 862dbd9eea0SPhilipp Reisner rr = drbd_recv_short(*sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); 863b411b363SPhilipp Reisner 864b411b363SPhilipp Reisner if (rr > 0 || rr == -EAGAIN) { 86581e84650SAndreas Gruenbacher return true; 866b411b363SPhilipp Reisner } else { 867b411b363SPhilipp Reisner sock_release(*sock); 868b411b363SPhilipp Reisner *sock = NULL; 86981e84650SAndreas Gruenbacher return false; 870b411b363SPhilipp Reisner } 871b411b363SPhilipp Reisner } 8725d0b17f1SPhilipp Reisner 8735d0b17f1SPhilipp Reisner static bool connection_established(struct drbd_connection *connection, 8745d0b17f1SPhilipp Reisner struct socket **sock1, 8755d0b17f1SPhilipp Reisner struct socket **sock2) 8765d0b17f1SPhilipp Reisner { 8775d0b17f1SPhilipp Reisner struct net_conf *nc; 8785d0b17f1SPhilipp Reisner int timeout; 8795d0b17f1SPhilipp Reisner bool ok; 8805d0b17f1SPhilipp Reisner 8815d0b17f1SPhilipp Reisner if (!*sock1 || !*sock2) 8825d0b17f1SPhilipp Reisner return false; 8835d0b17f1SPhilipp Reisner 8845d0b17f1SPhilipp Reisner rcu_read_lock(); 8855d0b17f1SPhilipp Reisner nc = rcu_dereference(connection->net_conf); 8865d0b17f1SPhilipp Reisner timeout = (nc->sock_check_timeo ?: nc->ping_timeo) * HZ / 10; 8875d0b17f1SPhilipp Reisner rcu_read_unlock(); 8885d0b17f1SPhilipp Reisner schedule_timeout_interruptible(timeout); 8895d0b17f1SPhilipp Reisner 8905d0b17f1SPhilipp Reisner ok = drbd_socket_okay(sock1); 8915d0b17f1SPhilipp Reisner ok = drbd_socket_okay(sock2) && ok; 8925d0b17f1SPhilipp Reisner 8935d0b17f1SPhilipp Reisner return ok; 8945d0b17f1SPhilipp Reisner } 8955d0b17f1SPhilipp Reisner 8962325eb66SPhilipp Reisner /* Gets called if a connection is established, or if a new minor gets created 8972325eb66SPhilipp Reisner in a connection */ 89869a22773SAndreas Gruenbacher int drbd_connected(struct drbd_peer_device *peer_device) 899907599e0SPhilipp Reisner { 90069a22773SAndreas Gruenbacher struct drbd_device *device = peer_device->device; 9010829f5edSAndreas Gruenbacher int err; 902907599e0SPhilipp Reisner 903b30ab791SAndreas Gruenbacher atomic_set(&device->packet_seq, 0); 904b30ab791SAndreas Gruenbacher device->peer_seq = 0; 905907599e0SPhilipp Reisner 90669a22773SAndreas Gruenbacher device->state_mutex = peer_device->connection->agreed_pro_version < 100 ? 90769a22773SAndreas Gruenbacher &peer_device->connection->cstate_mutex : 908b30ab791SAndreas Gruenbacher &device->own_state_mutex; 9098410da8fSPhilipp Reisner 91069a22773SAndreas Gruenbacher err = drbd_send_sync_param(peer_device); 9110829f5edSAndreas Gruenbacher if (!err) 91269a22773SAndreas Gruenbacher err = drbd_send_sizes(peer_device, 0, 0); 9130829f5edSAndreas Gruenbacher if (!err) 91469a22773SAndreas Gruenbacher err = drbd_send_uuids(peer_device); 9150829f5edSAndreas Gruenbacher if (!err) 91669a22773SAndreas Gruenbacher err = drbd_send_current_state(peer_device); 917b30ab791SAndreas Gruenbacher clear_bit(USE_DEGR_WFC_T, &device->flags); 918b30ab791SAndreas Gruenbacher clear_bit(RESIZE_PENDING, &device->flags); 919b30ab791SAndreas Gruenbacher atomic_set(&device->ap_in_flight, 0); 920b30ab791SAndreas Gruenbacher mod_timer(&device->request_timer, jiffies + HZ); /* just start it here. */ 9210829f5edSAndreas Gruenbacher return err; 922907599e0SPhilipp Reisner } 923b411b363SPhilipp Reisner 924b411b363SPhilipp Reisner /* 925b411b363SPhilipp Reisner * return values: 926b411b363SPhilipp Reisner * 1 yes, we have a valid connection 927b411b363SPhilipp Reisner * 0 oops, did not work out, please try again 928b411b363SPhilipp Reisner * -1 peer talks different language, 929b411b363SPhilipp Reisner * no point in trying again, please go standalone. 930b411b363SPhilipp Reisner * -2 We do not have a network config... 931b411b363SPhilipp Reisner */ 932bde89a9eSAndreas Gruenbacher static int conn_connect(struct drbd_connection *connection) 933b411b363SPhilipp Reisner { 9347da35862SPhilipp Reisner struct drbd_socket sock, msock; 935c06ece6bSAndreas Gruenbacher struct drbd_peer_device *peer_device; 93644ed167dSPhilipp Reisner struct net_conf *nc; 9375d0b17f1SPhilipp Reisner int vnr, timeout, h; 9385d0b17f1SPhilipp Reisner bool discard_my_data, ok; 939197296ffSPhilipp Reisner enum drbd_state_rv rv; 9407a426fd8SPhilipp Reisner struct accept_wait_data ad = { 941bde89a9eSAndreas Gruenbacher .connection = connection, 9427a426fd8SPhilipp Reisner .door_bell = COMPLETION_INITIALIZER_ONSTACK(ad.door_bell), 9437a426fd8SPhilipp Reisner }; 944b411b363SPhilipp Reisner 945bde89a9eSAndreas Gruenbacher clear_bit(DISCONNECT_SENT, &connection->flags); 946bde89a9eSAndreas Gruenbacher if (conn_request_state(connection, NS(conn, C_WF_CONNECTION), CS_VERBOSE) < SS_SUCCESS) 947b411b363SPhilipp Reisner return -2; 948b411b363SPhilipp Reisner 9497da35862SPhilipp Reisner mutex_init(&sock.mutex); 950bde89a9eSAndreas Gruenbacher sock.sbuf = connection->data.sbuf; 951bde89a9eSAndreas Gruenbacher sock.rbuf = connection->data.rbuf; 9527da35862SPhilipp Reisner sock.socket = NULL; 9537da35862SPhilipp Reisner mutex_init(&msock.mutex); 954bde89a9eSAndreas Gruenbacher msock.sbuf = connection->meta.sbuf; 955bde89a9eSAndreas Gruenbacher msock.rbuf = connection->meta.rbuf; 9567da35862SPhilipp Reisner msock.socket = NULL; 9577da35862SPhilipp Reisner 9580916e0e3SAndreas Gruenbacher /* Assume that the peer only understands protocol 80 until we know better. */ 959bde89a9eSAndreas Gruenbacher connection->agreed_pro_version = 80; 960b411b363SPhilipp Reisner 961bde89a9eSAndreas Gruenbacher if (prepare_listen_socket(connection, &ad)) 9627a426fd8SPhilipp Reisner return 0; 963b411b363SPhilipp Reisner 964b411b363SPhilipp Reisner do { 9652bf89621SAndreas Gruenbacher struct socket *s; 966b411b363SPhilipp Reisner 967bde89a9eSAndreas Gruenbacher s = drbd_try_connect(connection); 968b411b363SPhilipp Reisner if (s) { 9697da35862SPhilipp Reisner if (!sock.socket) { 9707da35862SPhilipp Reisner sock.socket = s; 971bde89a9eSAndreas Gruenbacher send_first_packet(connection, &sock, P_INITIAL_DATA); 9727da35862SPhilipp Reisner } else if (!msock.socket) { 973bde89a9eSAndreas Gruenbacher clear_bit(RESOLVE_CONFLICTS, &connection->flags); 9747da35862SPhilipp Reisner msock.socket = s; 975bde89a9eSAndreas Gruenbacher send_first_packet(connection, &msock, P_INITIAL_META); 976b411b363SPhilipp Reisner } else { 9771ec861ebSAndreas Gruenbacher drbd_err(connection, "Logic error in conn_connect()\n"); 978b411b363SPhilipp Reisner goto out_release_sockets; 979b411b363SPhilipp Reisner } 980b411b363SPhilipp Reisner } 981b411b363SPhilipp Reisner 9825d0b17f1SPhilipp Reisner if (connection_established(connection, &sock.socket, &msock.socket)) 983b411b363SPhilipp Reisner break; 984b411b363SPhilipp Reisner 985b411b363SPhilipp Reisner retry: 986bde89a9eSAndreas Gruenbacher s = drbd_wait_for_connect(connection, &ad); 987b411b363SPhilipp Reisner if (s) { 988bde89a9eSAndreas Gruenbacher int fp = receive_first_packet(connection, s); 9897da35862SPhilipp Reisner drbd_socket_okay(&sock.socket); 9907da35862SPhilipp Reisner drbd_socket_okay(&msock.socket); 99192f14951SPhilipp Reisner switch (fp) { 992e5d6f33aSAndreas Gruenbacher case P_INITIAL_DATA: 9937da35862SPhilipp Reisner if (sock.socket) { 9941ec861ebSAndreas Gruenbacher drbd_warn(connection, "initial packet S crossed\n"); 9957da35862SPhilipp Reisner sock_release(sock.socket); 99680c6eed4SPhilipp Reisner sock.socket = s; 99780c6eed4SPhilipp Reisner goto randomize; 998b411b363SPhilipp Reisner } 9997da35862SPhilipp Reisner sock.socket = s; 1000b411b363SPhilipp Reisner break; 1001e5d6f33aSAndreas Gruenbacher case P_INITIAL_META: 1002bde89a9eSAndreas Gruenbacher set_bit(RESOLVE_CONFLICTS, &connection->flags); 10037da35862SPhilipp Reisner if (msock.socket) { 10041ec861ebSAndreas Gruenbacher drbd_warn(connection, "initial packet M crossed\n"); 10057da35862SPhilipp Reisner sock_release(msock.socket); 100680c6eed4SPhilipp Reisner msock.socket = s; 100780c6eed4SPhilipp Reisner goto randomize; 1008b411b363SPhilipp Reisner } 10097da35862SPhilipp Reisner msock.socket = s; 1010b411b363SPhilipp Reisner break; 1011b411b363SPhilipp Reisner default: 10121ec861ebSAndreas Gruenbacher drbd_warn(connection, "Error receiving initial packet\n"); 1013b411b363SPhilipp Reisner sock_release(s); 101480c6eed4SPhilipp Reisner randomize: 101538b682b2SAkinobu Mita if (prandom_u32() & 1) 1016b411b363SPhilipp Reisner goto retry; 1017b411b363SPhilipp Reisner } 1018b411b363SPhilipp Reisner } 1019b411b363SPhilipp Reisner 1020bde89a9eSAndreas Gruenbacher if (connection->cstate <= C_DISCONNECTING) 1021b411b363SPhilipp Reisner goto out_release_sockets; 1022b411b363SPhilipp Reisner if (signal_pending(current)) { 1023b411b363SPhilipp Reisner flush_signals(current); 1024b411b363SPhilipp Reisner smp_rmb(); 1025bde89a9eSAndreas Gruenbacher if (get_t_state(&connection->receiver) == EXITING) 1026b411b363SPhilipp Reisner goto out_release_sockets; 1027b411b363SPhilipp Reisner } 1028b411b363SPhilipp Reisner 10295d0b17f1SPhilipp Reisner ok = connection_established(connection, &sock.socket, &msock.socket); 1030b666dbf8SPhilipp Reisner } while (!ok); 1031b411b363SPhilipp Reisner 10327a426fd8SPhilipp Reisner if (ad.s_listen) 10337a426fd8SPhilipp Reisner sock_release(ad.s_listen); 1034b411b363SPhilipp Reisner 103598683650SPhilipp Reisner sock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ 103698683650SPhilipp Reisner msock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ 1037b411b363SPhilipp Reisner 10387da35862SPhilipp Reisner sock.socket->sk->sk_allocation = GFP_NOIO; 10397da35862SPhilipp Reisner msock.socket->sk->sk_allocation = GFP_NOIO; 1040b411b363SPhilipp Reisner 10417da35862SPhilipp Reisner sock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; 10427da35862SPhilipp Reisner msock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE; 1043b411b363SPhilipp Reisner 1044b411b363SPhilipp Reisner /* NOT YET ... 1045bde89a9eSAndreas Gruenbacher * sock.socket->sk->sk_sndtimeo = connection->net_conf->timeout*HZ/10; 10467da35862SPhilipp Reisner * sock.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 10476038178eSAndreas Gruenbacher * first set it to the P_CONNECTION_FEATURES timeout, 1048b411b363SPhilipp Reisner * which we set to 4x the configured ping_timeout. */ 104944ed167dSPhilipp Reisner rcu_read_lock(); 1050bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 1051b411b363SPhilipp Reisner 10527da35862SPhilipp Reisner sock.socket->sk->sk_sndtimeo = 10537da35862SPhilipp Reisner sock.socket->sk->sk_rcvtimeo = nc->ping_timeo*4*HZ/10; 105444ed167dSPhilipp Reisner 10557da35862SPhilipp Reisner msock.socket->sk->sk_rcvtimeo = nc->ping_int*HZ; 105644ed167dSPhilipp Reisner timeout = nc->timeout * HZ / 10; 105708b165baSPhilipp Reisner discard_my_data = nc->discard_my_data; 105844ed167dSPhilipp Reisner rcu_read_unlock(); 105944ed167dSPhilipp Reisner 10607da35862SPhilipp Reisner msock.socket->sk->sk_sndtimeo = timeout; 1061b411b363SPhilipp Reisner 1062b411b363SPhilipp Reisner /* we don't want delays. 106325985edcSLucas De Marchi * we use TCP_CORK where appropriate, though */ 10647da35862SPhilipp Reisner drbd_tcp_nodelay(sock.socket); 10657da35862SPhilipp Reisner drbd_tcp_nodelay(msock.socket); 1066b411b363SPhilipp Reisner 1067bde89a9eSAndreas Gruenbacher connection->data.socket = sock.socket; 1068bde89a9eSAndreas Gruenbacher connection->meta.socket = msock.socket; 1069bde89a9eSAndreas Gruenbacher connection->last_received = jiffies; 1070b411b363SPhilipp Reisner 1071bde89a9eSAndreas Gruenbacher h = drbd_do_features(connection); 1072b411b363SPhilipp Reisner if (h <= 0) 1073b411b363SPhilipp Reisner return h; 1074b411b363SPhilipp Reisner 1075bde89a9eSAndreas Gruenbacher if (connection->cram_hmac_tfm) { 1076b30ab791SAndreas Gruenbacher /* drbd_request_state(device, NS(conn, WFAuth)); */ 1077bde89a9eSAndreas Gruenbacher switch (drbd_do_auth(connection)) { 1078b10d96cbSJohannes Thoma case -1: 10791ec861ebSAndreas Gruenbacher drbd_err(connection, "Authentication of peer failed\n"); 1080b411b363SPhilipp Reisner return -1; 1081b10d96cbSJohannes Thoma case 0: 10821ec861ebSAndreas Gruenbacher drbd_err(connection, "Authentication of peer failed, trying again.\n"); 1083b10d96cbSJohannes Thoma return 0; 1084b411b363SPhilipp Reisner } 1085b411b363SPhilipp Reisner } 1086b411b363SPhilipp Reisner 1087bde89a9eSAndreas Gruenbacher connection->data.socket->sk->sk_sndtimeo = timeout; 1088bde89a9eSAndreas Gruenbacher connection->data.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 1089b411b363SPhilipp Reisner 1090bde89a9eSAndreas Gruenbacher if (drbd_send_protocol(connection) == -EOPNOTSUPP) 10917e2455c1SPhilipp Reisner return -1; 10921e86ac48SPhilipp Reisner 109313c76abaSPhilipp Reisner /* Prevent a race between resync-handshake and 109413c76abaSPhilipp Reisner * being promoted to Primary. 109513c76abaSPhilipp Reisner * 109613c76abaSPhilipp Reisner * Grab and release the state mutex, so we know that any current 109713c76abaSPhilipp Reisner * drbd_set_role() is finished, and any incoming drbd_set_role 109813c76abaSPhilipp Reisner * will see the STATE_SENT flag, and wait for it to be cleared. 109913c76abaSPhilipp Reisner */ 110031007745SPhilipp Reisner idr_for_each_entry(&connection->peer_devices, peer_device, vnr) 110131007745SPhilipp Reisner mutex_lock(peer_device->device->state_mutex); 110231007745SPhilipp Reisner 1103cde81d99SLars Ellenberg /* avoid a race with conn_request_state( C_DISCONNECTING ) */ 1104cde81d99SLars Ellenberg spin_lock_irq(&connection->resource->req_lock); 110531007745SPhilipp Reisner set_bit(STATE_SENT, &connection->flags); 1106cde81d99SLars Ellenberg spin_unlock_irq(&connection->resource->req_lock); 110731007745SPhilipp Reisner 110831007745SPhilipp Reisner idr_for_each_entry(&connection->peer_devices, peer_device, vnr) 110931007745SPhilipp Reisner mutex_unlock(peer_device->device->state_mutex); 111031007745SPhilipp Reisner 111131007745SPhilipp Reisner rcu_read_lock(); 111231007745SPhilipp Reisner idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 111331007745SPhilipp Reisner struct drbd_device *device = peer_device->device; 111431007745SPhilipp Reisner kref_get(&device->kref); 111531007745SPhilipp Reisner rcu_read_unlock(); 111613c76abaSPhilipp Reisner 111708b165baSPhilipp Reisner if (discard_my_data) 1118b30ab791SAndreas Gruenbacher set_bit(DISCARD_MY_DATA, &device->flags); 111908b165baSPhilipp Reisner else 1120b30ab791SAndreas Gruenbacher clear_bit(DISCARD_MY_DATA, &device->flags); 112108b165baSPhilipp Reisner 112269a22773SAndreas Gruenbacher drbd_connected(peer_device); 112305a10ec7SAndreas Gruenbacher kref_put(&device->kref, drbd_destroy_device); 1124c141ebdaSPhilipp Reisner rcu_read_lock(); 1125c141ebdaSPhilipp Reisner } 1126c141ebdaSPhilipp Reisner rcu_read_unlock(); 1127c141ebdaSPhilipp Reisner 1128bde89a9eSAndreas Gruenbacher rv = conn_request_state(connection, NS(conn, C_WF_REPORT_PARAMS), CS_VERBOSE); 1129bde89a9eSAndreas Gruenbacher if (rv < SS_SUCCESS || connection->cstate != C_WF_REPORT_PARAMS) { 1130bde89a9eSAndreas Gruenbacher clear_bit(STATE_SENT, &connection->flags); 11311e86ac48SPhilipp Reisner return 0; 1132a1096a6eSPhilipp Reisner } 11331e86ac48SPhilipp Reisner 11341c03e520SPhilipp Reisner drbd_thread_start(&connection->ack_receiver); 113539e91a60SLars Ellenberg /* opencoded create_singlethread_workqueue(), 113639e91a60SLars Ellenberg * to be able to use format string arguments */ 113739e91a60SLars Ellenberg connection->ack_sender = 113839e91a60SLars Ellenberg alloc_ordered_workqueue("drbd_as_%s", WQ_MEM_RECLAIM, connection->resource->name); 1139668700b4SPhilipp Reisner if (!connection->ack_sender) { 1140668700b4SPhilipp Reisner drbd_err(connection, "Failed to create workqueue ack_sender\n"); 1141668700b4SPhilipp Reisner return 0; 1142668700b4SPhilipp Reisner } 1143b411b363SPhilipp Reisner 11440500813fSAndreas Gruenbacher mutex_lock(&connection->resource->conf_update); 114508b165baSPhilipp Reisner /* The discard_my_data flag is a single-shot modifier to the next 114608b165baSPhilipp Reisner * connection attempt, the handshake of which is now well underway. 114708b165baSPhilipp Reisner * No need for rcu style copying of the whole struct 114808b165baSPhilipp Reisner * just to clear a single value. */ 1149bde89a9eSAndreas Gruenbacher connection->net_conf->discard_my_data = 0; 11500500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 115108b165baSPhilipp Reisner 1152d3fcb490SPhilipp Reisner return h; 1153b411b363SPhilipp Reisner 1154b411b363SPhilipp Reisner out_release_sockets: 11557a426fd8SPhilipp Reisner if (ad.s_listen) 11567a426fd8SPhilipp Reisner sock_release(ad.s_listen); 11577da35862SPhilipp Reisner if (sock.socket) 11587da35862SPhilipp Reisner sock_release(sock.socket); 11597da35862SPhilipp Reisner if (msock.socket) 11607da35862SPhilipp Reisner sock_release(msock.socket); 1161b411b363SPhilipp Reisner return -1; 1162b411b363SPhilipp Reisner } 1163b411b363SPhilipp Reisner 1164bde89a9eSAndreas Gruenbacher static int decode_header(struct drbd_connection *connection, void *header, struct packet_info *pi) 1165b411b363SPhilipp Reisner { 1166bde89a9eSAndreas Gruenbacher unsigned int header_size = drbd_header_size(connection); 1167b411b363SPhilipp Reisner 11680c8e36d9SAndreas Gruenbacher if (header_size == sizeof(struct p_header100) && 11690c8e36d9SAndreas Gruenbacher *(__be32 *)header == cpu_to_be32(DRBD_MAGIC_100)) { 11700c8e36d9SAndreas Gruenbacher struct p_header100 *h = header; 11710c8e36d9SAndreas Gruenbacher if (h->pad != 0) { 11721ec861ebSAndreas Gruenbacher drbd_err(connection, "Header padding is not zero\n"); 11730c8e36d9SAndreas Gruenbacher return -EINVAL; 117402918be2SPhilipp Reisner } 11750c8e36d9SAndreas Gruenbacher pi->vnr = be16_to_cpu(h->volume); 11760c8e36d9SAndreas Gruenbacher pi->cmd = be16_to_cpu(h->command); 11770c8e36d9SAndreas Gruenbacher pi->size = be32_to_cpu(h->length); 11780c8e36d9SAndreas Gruenbacher } else if (header_size == sizeof(struct p_header95) && 1179e658983aSAndreas Gruenbacher *(__be16 *)header == cpu_to_be16(DRBD_MAGIC_BIG)) { 1180e658983aSAndreas Gruenbacher struct p_header95 *h = header; 1181e658983aSAndreas Gruenbacher pi->cmd = be16_to_cpu(h->command); 1182b55d84baSAndreas Gruenbacher pi->size = be32_to_cpu(h->length); 1183eefc2f7dSPhilipp Reisner pi->vnr = 0; 1184e658983aSAndreas Gruenbacher } else if (header_size == sizeof(struct p_header80) && 1185e658983aSAndreas Gruenbacher *(__be32 *)header == cpu_to_be32(DRBD_MAGIC)) { 1186e658983aSAndreas Gruenbacher struct p_header80 *h = header; 1187e658983aSAndreas Gruenbacher pi->cmd = be16_to_cpu(h->command); 1188e658983aSAndreas Gruenbacher pi->size = be16_to_cpu(h->length); 118977351055SPhilipp Reisner pi->vnr = 0; 119002918be2SPhilipp Reisner } else { 11911ec861ebSAndreas Gruenbacher drbd_err(connection, "Wrong magic value 0x%08x in protocol version %d\n", 1192e658983aSAndreas Gruenbacher be32_to_cpu(*(__be32 *)header), 1193bde89a9eSAndreas Gruenbacher connection->agreed_pro_version); 11948172f3e9SAndreas Gruenbacher return -EINVAL; 1195b411b363SPhilipp Reisner } 1196e658983aSAndreas Gruenbacher pi->data = header + header_size; 11978172f3e9SAndreas Gruenbacher return 0; 1198b411b363SPhilipp Reisner } 1199b411b363SPhilipp Reisner 1200c51a0ef3SLars Ellenberg static void drbd_unplug_all_devices(struct drbd_connection *connection) 1201c51a0ef3SLars Ellenberg { 1202c51a0ef3SLars Ellenberg if (current->plug == &connection->receiver_plug) { 1203c51a0ef3SLars Ellenberg blk_finish_plug(&connection->receiver_plug); 1204c51a0ef3SLars Ellenberg blk_start_plug(&connection->receiver_plug); 1205c51a0ef3SLars Ellenberg } /* else: maybe just schedule() ?? */ 1206c51a0ef3SLars Ellenberg } 1207c51a0ef3SLars Ellenberg 1208bde89a9eSAndreas Gruenbacher static int drbd_recv_header(struct drbd_connection *connection, struct packet_info *pi) 1209257d0af6SPhilipp Reisner { 1210bde89a9eSAndreas Gruenbacher void *buffer = connection->data.rbuf; 121169bc7bc3SAndreas Gruenbacher int err; 1212257d0af6SPhilipp Reisner 1213bde89a9eSAndreas Gruenbacher err = drbd_recv_all_warn(connection, buffer, drbd_header_size(connection)); 1214a5c31904SAndreas Gruenbacher if (err) 121569bc7bc3SAndreas Gruenbacher return err; 1216257d0af6SPhilipp Reisner 1217bde89a9eSAndreas Gruenbacher err = decode_header(connection, buffer, pi); 1218bde89a9eSAndreas Gruenbacher connection->last_received = jiffies; 1219b411b363SPhilipp Reisner 122069bc7bc3SAndreas Gruenbacher return err; 1221b411b363SPhilipp Reisner } 1222b411b363SPhilipp Reisner 1223c51a0ef3SLars Ellenberg static int drbd_recv_header_maybe_unplug(struct drbd_connection *connection, struct packet_info *pi) 1224c51a0ef3SLars Ellenberg { 1225c51a0ef3SLars Ellenberg void *buffer = connection->data.rbuf; 1226c51a0ef3SLars Ellenberg unsigned int size = drbd_header_size(connection); 1227c51a0ef3SLars Ellenberg int err; 1228c51a0ef3SLars Ellenberg 1229c51a0ef3SLars Ellenberg err = drbd_recv_short(connection->data.socket, buffer, size, MSG_NOSIGNAL|MSG_DONTWAIT); 1230c51a0ef3SLars Ellenberg if (err != size) { 1231c51a0ef3SLars Ellenberg /* If we have nothing in the receive buffer now, to reduce 1232c51a0ef3SLars Ellenberg * application latency, try to drain the backend queues as 1233c51a0ef3SLars Ellenberg * quickly as possible, and let remote TCP know what we have 1234c51a0ef3SLars Ellenberg * received so far. */ 1235c51a0ef3SLars Ellenberg if (err == -EAGAIN) { 1236c51a0ef3SLars Ellenberg drbd_tcp_quickack(connection->data.socket); 1237c51a0ef3SLars Ellenberg drbd_unplug_all_devices(connection); 1238c51a0ef3SLars Ellenberg } 1239c51a0ef3SLars Ellenberg if (err > 0) { 1240c51a0ef3SLars Ellenberg buffer += err; 1241c51a0ef3SLars Ellenberg size -= err; 1242c51a0ef3SLars Ellenberg } 1243c51a0ef3SLars Ellenberg err = drbd_recv_all_warn(connection, buffer, size); 1244c51a0ef3SLars Ellenberg if (err) 1245c51a0ef3SLars Ellenberg return err; 1246c51a0ef3SLars Ellenberg } 1247c51a0ef3SLars Ellenberg 1248c51a0ef3SLars Ellenberg err = decode_header(connection, connection->data.rbuf, pi); 1249c51a0ef3SLars Ellenberg connection->last_received = jiffies; 1250c51a0ef3SLars Ellenberg 1251c51a0ef3SLars Ellenberg return err; 1252c51a0ef3SLars Ellenberg } 1253f9ff0da5SLars Ellenberg /* This is blkdev_issue_flush, but asynchronous. 1254f9ff0da5SLars Ellenberg * We want to submit to all component volumes in parallel, 1255f9ff0da5SLars Ellenberg * then wait for all completions. 1256f9ff0da5SLars Ellenberg */ 1257f9ff0da5SLars Ellenberg struct issue_flush_context { 1258f9ff0da5SLars Ellenberg atomic_t pending; 1259f9ff0da5SLars Ellenberg int error; 1260f9ff0da5SLars Ellenberg struct completion done; 1261f9ff0da5SLars Ellenberg }; 1262f9ff0da5SLars Ellenberg struct one_flush_context { 1263f9ff0da5SLars Ellenberg struct drbd_device *device; 1264f9ff0da5SLars Ellenberg struct issue_flush_context *ctx; 1265f9ff0da5SLars Ellenberg }; 1266f9ff0da5SLars Ellenberg 12671ffa7bfaSBaoyou Xie static void one_flush_endio(struct bio *bio) 1268f9ff0da5SLars Ellenberg { 1269f9ff0da5SLars Ellenberg struct one_flush_context *octx = bio->bi_private; 1270f9ff0da5SLars Ellenberg struct drbd_device *device = octx->device; 1271f9ff0da5SLars Ellenberg struct issue_flush_context *ctx = octx->ctx; 1272f9ff0da5SLars Ellenberg 12734e4cbee9SChristoph Hellwig if (bio->bi_status) { 12744e4cbee9SChristoph Hellwig ctx->error = blk_status_to_errno(bio->bi_status); 12754e4cbee9SChristoph Hellwig drbd_info(device, "local disk FLUSH FAILED with status %d\n", bio->bi_status); 1276f9ff0da5SLars Ellenberg } 1277f9ff0da5SLars Ellenberg kfree(octx); 1278f9ff0da5SLars Ellenberg bio_put(bio); 1279f9ff0da5SLars Ellenberg 1280f9ff0da5SLars Ellenberg clear_bit(FLUSH_PENDING, &device->flags); 1281f9ff0da5SLars Ellenberg put_ldev(device); 1282f9ff0da5SLars Ellenberg kref_put(&device->kref, drbd_destroy_device); 1283f9ff0da5SLars Ellenberg 1284f9ff0da5SLars Ellenberg if (atomic_dec_and_test(&ctx->pending)) 1285f9ff0da5SLars Ellenberg complete(&ctx->done); 1286f9ff0da5SLars Ellenberg } 1287f9ff0da5SLars Ellenberg 1288f9ff0da5SLars Ellenberg static void submit_one_flush(struct drbd_device *device, struct issue_flush_context *ctx) 1289f9ff0da5SLars Ellenberg { 1290f9ff0da5SLars Ellenberg struct bio *bio = bio_alloc(GFP_NOIO, 0); 1291f9ff0da5SLars Ellenberg struct one_flush_context *octx = kmalloc(sizeof(*octx), GFP_NOIO); 1292f9ff0da5SLars Ellenberg if (!bio || !octx) { 1293f9ff0da5SLars Ellenberg drbd_warn(device, "Could not allocate a bio, CANNOT ISSUE FLUSH\n"); 1294f9ff0da5SLars Ellenberg /* FIXME: what else can I do now? disconnecting or detaching 1295f9ff0da5SLars Ellenberg * really does not help to improve the state of the world, either. 1296f9ff0da5SLars Ellenberg */ 1297f9ff0da5SLars Ellenberg kfree(octx); 1298f9ff0da5SLars Ellenberg if (bio) 1299f9ff0da5SLars Ellenberg bio_put(bio); 1300f9ff0da5SLars Ellenberg 1301f9ff0da5SLars Ellenberg ctx->error = -ENOMEM; 1302f9ff0da5SLars Ellenberg put_ldev(device); 1303f9ff0da5SLars Ellenberg kref_put(&device->kref, drbd_destroy_device); 1304f9ff0da5SLars Ellenberg return; 1305f9ff0da5SLars Ellenberg } 1306f9ff0da5SLars Ellenberg 1307f9ff0da5SLars Ellenberg octx->device = device; 1308f9ff0da5SLars Ellenberg octx->ctx = ctx; 130974d46992SChristoph Hellwig bio_set_dev(bio, device->ldev->backing_bdev); 1310f9ff0da5SLars Ellenberg bio->bi_private = octx; 1311f9ff0da5SLars Ellenberg bio->bi_end_io = one_flush_endio; 131270fd7614SChristoph Hellwig bio->bi_opf = REQ_OP_FLUSH | REQ_PREFLUSH; 1313f9ff0da5SLars Ellenberg 1314f9ff0da5SLars Ellenberg device->flush_jif = jiffies; 1315f9ff0da5SLars Ellenberg set_bit(FLUSH_PENDING, &device->flags); 1316f9ff0da5SLars Ellenberg atomic_inc(&ctx->pending); 1317f9ff0da5SLars Ellenberg submit_bio(bio); 1318f9ff0da5SLars Ellenberg } 1319f9ff0da5SLars Ellenberg 1320bde89a9eSAndreas Gruenbacher static void drbd_flush(struct drbd_connection *connection) 1321b411b363SPhilipp Reisner { 1322f9ff0da5SLars Ellenberg if (connection->resource->write_ordering >= WO_BDEV_FLUSH) { 1323c06ece6bSAndreas Gruenbacher struct drbd_peer_device *peer_device; 1324f9ff0da5SLars Ellenberg struct issue_flush_context ctx; 13254b0007c0SPhilipp Reisner int vnr; 1326b411b363SPhilipp Reisner 1327f9ff0da5SLars Ellenberg atomic_set(&ctx.pending, 1); 1328f9ff0da5SLars Ellenberg ctx.error = 0; 1329f9ff0da5SLars Ellenberg init_completion(&ctx.done); 1330f9ff0da5SLars Ellenberg 1331615e087fSLars Ellenberg rcu_read_lock(); 1332c06ece6bSAndreas Gruenbacher idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 1333c06ece6bSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 1334c06ece6bSAndreas Gruenbacher 1335b30ab791SAndreas Gruenbacher if (!get_ldev(device)) 1336615e087fSLars Ellenberg continue; 1337b30ab791SAndreas Gruenbacher kref_get(&device->kref); 1338615e087fSLars Ellenberg rcu_read_unlock(); 13394b0007c0SPhilipp Reisner 1340f9ff0da5SLars Ellenberg submit_one_flush(device, &ctx); 1341f9ff0da5SLars Ellenberg 1342f9ff0da5SLars Ellenberg rcu_read_lock(); 1343f9ff0da5SLars Ellenberg } 1344f9ff0da5SLars Ellenberg rcu_read_unlock(); 1345f9ff0da5SLars Ellenberg 1346f9ff0da5SLars Ellenberg /* Do we want to add a timeout, 1347f9ff0da5SLars Ellenberg * if disk-timeout is set? */ 1348f9ff0da5SLars Ellenberg if (!atomic_dec_and_test(&ctx.pending)) 1349f9ff0da5SLars Ellenberg wait_for_completion(&ctx.done); 1350f9ff0da5SLars Ellenberg 1351f9ff0da5SLars Ellenberg if (ctx.error) { 1352b411b363SPhilipp Reisner /* would rather check on EOPNOTSUPP, but that is not reliable. 1353b411b363SPhilipp Reisner * don't try again for ANY return value != 0 1354b411b363SPhilipp Reisner * if (rv == -EOPNOTSUPP) */ 1355f9ff0da5SLars Ellenberg /* Any error is already reported by bio_endio callback. */ 1356f6ba8636SAndreas Gruenbacher drbd_bump_write_ordering(connection->resource, NULL, WO_DRAIN_IO); 1357b411b363SPhilipp Reisner } 1358b411b363SPhilipp Reisner } 1359b411b363SPhilipp Reisner } 1360b411b363SPhilipp Reisner 1361b411b363SPhilipp Reisner /** 1362b411b363SPhilipp Reisner * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. 1363b30ab791SAndreas Gruenbacher * @device: DRBD device. 1364b411b363SPhilipp Reisner * @epoch: Epoch object. 1365b411b363SPhilipp Reisner * @ev: Epoch event. 1366b411b363SPhilipp Reisner */ 1367bde89a9eSAndreas Gruenbacher static enum finish_epoch drbd_may_finish_epoch(struct drbd_connection *connection, 1368b411b363SPhilipp Reisner struct drbd_epoch *epoch, 1369b411b363SPhilipp Reisner enum epoch_event ev) 1370b411b363SPhilipp Reisner { 13712451fc3bSPhilipp Reisner int epoch_size; 1372b411b363SPhilipp Reisner struct drbd_epoch *next_epoch; 1373b411b363SPhilipp Reisner enum finish_epoch rv = FE_STILL_LIVE; 1374b411b363SPhilipp Reisner 1375bde89a9eSAndreas Gruenbacher spin_lock(&connection->epoch_lock); 1376b411b363SPhilipp Reisner do { 1377b411b363SPhilipp Reisner next_epoch = NULL; 1378b411b363SPhilipp Reisner 1379b411b363SPhilipp Reisner epoch_size = atomic_read(&epoch->epoch_size); 1380b411b363SPhilipp Reisner 1381b411b363SPhilipp Reisner switch (ev & ~EV_CLEANUP) { 1382b411b363SPhilipp Reisner case EV_PUT: 1383b411b363SPhilipp Reisner atomic_dec(&epoch->active); 1384b411b363SPhilipp Reisner break; 1385b411b363SPhilipp Reisner case EV_GOT_BARRIER_NR: 1386b411b363SPhilipp Reisner set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); 1387b411b363SPhilipp Reisner break; 1388b411b363SPhilipp Reisner case EV_BECAME_LAST: 1389b411b363SPhilipp Reisner /* nothing to do*/ 1390b411b363SPhilipp Reisner break; 1391b411b363SPhilipp Reisner } 1392b411b363SPhilipp Reisner 1393b411b363SPhilipp Reisner if (epoch_size != 0 && 1394b411b363SPhilipp Reisner atomic_read(&epoch->active) == 0 && 139580f9fd55SPhilipp Reisner (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) || ev & EV_CLEANUP)) { 1396b411b363SPhilipp Reisner if (!(ev & EV_CLEANUP)) { 1397bde89a9eSAndreas Gruenbacher spin_unlock(&connection->epoch_lock); 1398bde89a9eSAndreas Gruenbacher drbd_send_b_ack(epoch->connection, epoch->barrier_nr, epoch_size); 1399bde89a9eSAndreas Gruenbacher spin_lock(&connection->epoch_lock); 1400b411b363SPhilipp Reisner } 14019ed57dcbSLars Ellenberg #if 0 14029ed57dcbSLars Ellenberg /* FIXME: dec unacked on connection, once we have 14039ed57dcbSLars Ellenberg * something to count pending connection packets in. */ 140480f9fd55SPhilipp Reisner if (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) 1405bde89a9eSAndreas Gruenbacher dec_unacked(epoch->connection); 14069ed57dcbSLars Ellenberg #endif 1407b411b363SPhilipp Reisner 1408bde89a9eSAndreas Gruenbacher if (connection->current_epoch != epoch) { 1409b411b363SPhilipp Reisner next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); 1410b411b363SPhilipp Reisner list_del(&epoch->list); 1411b411b363SPhilipp Reisner ev = EV_BECAME_LAST | (ev & EV_CLEANUP); 1412bde89a9eSAndreas Gruenbacher connection->epochs--; 1413b411b363SPhilipp Reisner kfree(epoch); 1414b411b363SPhilipp Reisner 1415b411b363SPhilipp Reisner if (rv == FE_STILL_LIVE) 1416b411b363SPhilipp Reisner rv = FE_DESTROYED; 1417b411b363SPhilipp Reisner } else { 1418b411b363SPhilipp Reisner epoch->flags = 0; 1419b411b363SPhilipp Reisner atomic_set(&epoch->epoch_size, 0); 1420698f9315SUwe Kleine-König /* atomic_set(&epoch->active, 0); is already zero */ 1421b411b363SPhilipp Reisner if (rv == FE_STILL_LIVE) 1422b411b363SPhilipp Reisner rv = FE_RECYCLED; 1423b411b363SPhilipp Reisner } 1424b411b363SPhilipp Reisner } 1425b411b363SPhilipp Reisner 1426b411b363SPhilipp Reisner if (!next_epoch) 1427b411b363SPhilipp Reisner break; 1428b411b363SPhilipp Reisner 1429b411b363SPhilipp Reisner epoch = next_epoch; 1430b411b363SPhilipp Reisner } while (1); 1431b411b363SPhilipp Reisner 1432bde89a9eSAndreas Gruenbacher spin_unlock(&connection->epoch_lock); 1433b411b363SPhilipp Reisner 1434b411b363SPhilipp Reisner return rv; 1435b411b363SPhilipp Reisner } 1436b411b363SPhilipp Reisner 14378fe39aacSPhilipp Reisner static enum write_ordering_e 14388fe39aacSPhilipp Reisner max_allowed_wo(struct drbd_backing_dev *bdev, enum write_ordering_e wo) 14398fe39aacSPhilipp Reisner { 14408fe39aacSPhilipp Reisner struct disk_conf *dc; 14418fe39aacSPhilipp Reisner 14428fe39aacSPhilipp Reisner dc = rcu_dereference(bdev->disk_conf); 14438fe39aacSPhilipp Reisner 1444f6ba8636SAndreas Gruenbacher if (wo == WO_BDEV_FLUSH && !dc->disk_flushes) 1445f6ba8636SAndreas Gruenbacher wo = WO_DRAIN_IO; 1446f6ba8636SAndreas Gruenbacher if (wo == WO_DRAIN_IO && !dc->disk_drain) 1447f6ba8636SAndreas Gruenbacher wo = WO_NONE; 14488fe39aacSPhilipp Reisner 14498fe39aacSPhilipp Reisner return wo; 14508fe39aacSPhilipp Reisner } 14518fe39aacSPhilipp Reisner 1452b411b363SPhilipp Reisner /** 1453b411b363SPhilipp Reisner * drbd_bump_write_ordering() - Fall back to an other write ordering method 1454bde89a9eSAndreas Gruenbacher * @connection: DRBD connection. 1455b411b363SPhilipp Reisner * @wo: Write ordering method to try. 1456b411b363SPhilipp Reisner */ 14578fe39aacSPhilipp Reisner void drbd_bump_write_ordering(struct drbd_resource *resource, struct drbd_backing_dev *bdev, 14588fe39aacSPhilipp Reisner enum write_ordering_e wo) 1459b411b363SPhilipp Reisner { 1460e9526580SPhilipp Reisner struct drbd_device *device; 1461b411b363SPhilipp Reisner enum write_ordering_e pwo; 14624b0007c0SPhilipp Reisner int vnr; 1463b411b363SPhilipp Reisner static char *write_ordering_str[] = { 1464f6ba8636SAndreas Gruenbacher [WO_NONE] = "none", 1465f6ba8636SAndreas Gruenbacher [WO_DRAIN_IO] = "drain", 1466f6ba8636SAndreas Gruenbacher [WO_BDEV_FLUSH] = "flush", 1467b411b363SPhilipp Reisner }; 1468b411b363SPhilipp Reisner 1469e9526580SPhilipp Reisner pwo = resource->write_ordering; 1470f6ba8636SAndreas Gruenbacher if (wo != WO_BDEV_FLUSH) 1471b411b363SPhilipp Reisner wo = min(pwo, wo); 1472daeda1ccSPhilipp Reisner rcu_read_lock(); 1473e9526580SPhilipp Reisner idr_for_each_entry(&resource->devices, device, vnr) { 14748fe39aacSPhilipp Reisner if (get_ldev(device)) { 14758fe39aacSPhilipp Reisner wo = max_allowed_wo(device->ldev, wo); 14768fe39aacSPhilipp Reisner if (device->ldev == bdev) 14778fe39aacSPhilipp Reisner bdev = NULL; 1478b30ab791SAndreas Gruenbacher put_ldev(device); 14794b0007c0SPhilipp Reisner } 14808fe39aacSPhilipp Reisner } 14818fe39aacSPhilipp Reisner 14828fe39aacSPhilipp Reisner if (bdev) 14838fe39aacSPhilipp Reisner wo = max_allowed_wo(bdev, wo); 14848fe39aacSPhilipp Reisner 148570df7092SLars Ellenberg rcu_read_unlock(); 148670df7092SLars Ellenberg 1487e9526580SPhilipp Reisner resource->write_ordering = wo; 1488f6ba8636SAndreas Gruenbacher if (pwo != resource->write_ordering || wo == WO_BDEV_FLUSH) 1489e9526580SPhilipp Reisner drbd_info(resource, "Method to ensure write ordering: %s\n", write_ordering_str[resource->write_ordering]); 1490b411b363SPhilipp Reisner } 1491b411b363SPhilipp Reisner 14929104d31aSLars Ellenberg static void drbd_issue_peer_discard(struct drbd_device *device, struct drbd_peer_request *peer_req) 1493dd4f699dSLars Ellenberg { 14940dbed96aSChristoph Hellwig struct block_device *bdev = device->ldev->backing_bdev; 1495dd4f699dSLars Ellenberg 14960dbed96aSChristoph Hellwig if (blkdev_issue_zeroout(bdev, peer_req->i.sector, peer_req->i.size >> 9, 14970dbed96aSChristoph Hellwig GFP_NOIO, 0)) 1498dd4f699dSLars Ellenberg peer_req->flags |= EE_WAS_ERROR; 14990dbed96aSChristoph Hellwig 1500dd4f699dSLars Ellenberg drbd_endio_write_sec_final(peer_req); 1501dd4f699dSLars Ellenberg } 1502dd4f699dSLars Ellenberg 15039104d31aSLars Ellenberg static void drbd_issue_peer_wsame(struct drbd_device *device, 15049104d31aSLars Ellenberg struct drbd_peer_request *peer_req) 15059104d31aSLars Ellenberg { 15069104d31aSLars Ellenberg struct block_device *bdev = device->ldev->backing_bdev; 15079104d31aSLars Ellenberg sector_t s = peer_req->i.sector; 15089104d31aSLars Ellenberg sector_t nr = peer_req->i.size >> 9; 15099104d31aSLars Ellenberg if (blkdev_issue_write_same(bdev, s, nr, GFP_NOIO, peer_req->pages)) 15109104d31aSLars Ellenberg peer_req->flags |= EE_WAS_ERROR; 15119104d31aSLars Ellenberg drbd_endio_write_sec_final(peer_req); 15129104d31aSLars Ellenberg } 15139104d31aSLars Ellenberg 15149104d31aSLars Ellenberg 1515b411b363SPhilipp Reisner /** 1516fbe29decSAndreas Gruenbacher * drbd_submit_peer_request() 1517b30ab791SAndreas Gruenbacher * @device: DRBD device. 1518db830c46SAndreas Gruenbacher * @peer_req: peer request 15191eff9d32SJens Axboe * @rw: flag field, see bio->bi_opf 152010f6d992SLars Ellenberg * 152110f6d992SLars Ellenberg * May spread the pages to multiple bios, 152210f6d992SLars Ellenberg * depending on bio_add_page restrictions. 152310f6d992SLars Ellenberg * 152410f6d992SLars Ellenberg * Returns 0 if all bios have been submitted, 152510f6d992SLars Ellenberg * -ENOMEM if we could not allocate enough bios, 152610f6d992SLars Ellenberg * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a 152710f6d992SLars Ellenberg * single page to an empty bio (which should never happen and likely indicates 152810f6d992SLars Ellenberg * that the lower level IO stack is in some way broken). This has been observed 152910f6d992SLars Ellenberg * on certain Xen deployments. 153045bb912bSLars Ellenberg */ 153145bb912bSLars Ellenberg /* TODO allocate from our own bio_set. */ 1532b30ab791SAndreas Gruenbacher int drbd_submit_peer_request(struct drbd_device *device, 1533fbe29decSAndreas Gruenbacher struct drbd_peer_request *peer_req, 1534bb3cc85eSMike Christie const unsigned op, const unsigned op_flags, 1535bb3cc85eSMike Christie const int fault_type) 153645bb912bSLars Ellenberg { 153745bb912bSLars Ellenberg struct bio *bios = NULL; 153845bb912bSLars Ellenberg struct bio *bio; 1539db830c46SAndreas Gruenbacher struct page *page = peer_req->pages; 1540db830c46SAndreas Gruenbacher sector_t sector = peer_req->i.sector; 154111f8b2b6SAndreas Gruenbacher unsigned data_size = peer_req->i.size; 154245bb912bSLars Ellenberg unsigned n_bios = 0; 154311f8b2b6SAndreas Gruenbacher unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; 154410f6d992SLars Ellenberg int err = -ENOMEM; 154545bb912bSLars Ellenberg 1546dd4f699dSLars Ellenberg /* TRIM/DISCARD: for now, always use the helper function 1547dd4f699dSLars Ellenberg * blkdev_issue_zeroout(..., discard=true). 1548dd4f699dSLars Ellenberg * It's synchronous, but it does the right thing wrt. bio splitting. 1549dd4f699dSLars Ellenberg * Correctness first, performance later. Next step is to code an 1550dd4f699dSLars Ellenberg * asynchronous variant of the same. 1551dd4f699dSLars Ellenberg */ 15529104d31aSLars Ellenberg if (peer_req->flags & (EE_IS_TRIM|EE_WRITE_SAME)) { 1553a0fb3c47SLars Ellenberg /* wait for all pending IO completions, before we start 1554a0fb3c47SLars Ellenberg * zeroing things out. */ 15555dd2ca19SAndreas Gruenbacher conn_wait_active_ee_empty(peer_req->peer_device->connection); 155645d2933cSLars Ellenberg /* add it to the active list now, 155745d2933cSLars Ellenberg * so we can find it to present it in debugfs */ 155821ae5d7fSLars Ellenberg peer_req->submit_jif = jiffies; 155921ae5d7fSLars Ellenberg peer_req->flags |= EE_SUBMITTED; 1560700ca8c0SPhilipp Reisner 1561700ca8c0SPhilipp Reisner /* If this was a resync request from receive_rs_deallocated(), 1562700ca8c0SPhilipp Reisner * it is already on the sync_ee list */ 1563700ca8c0SPhilipp Reisner if (list_empty(&peer_req->w.list)) { 156445d2933cSLars Ellenberg spin_lock_irq(&device->resource->req_lock); 156545d2933cSLars Ellenberg list_add_tail(&peer_req->w.list, &device->active_ee); 156645d2933cSLars Ellenberg spin_unlock_irq(&device->resource->req_lock); 1567700ca8c0SPhilipp Reisner } 1568700ca8c0SPhilipp Reisner 15699104d31aSLars Ellenberg if (peer_req->flags & EE_IS_TRIM) 1570dd4f699dSLars Ellenberg drbd_issue_peer_discard(device, peer_req); 15719104d31aSLars Ellenberg else /* EE_WRITE_SAME */ 15729104d31aSLars Ellenberg drbd_issue_peer_wsame(device, peer_req); 1573a0fb3c47SLars Ellenberg return 0; 1574a0fb3c47SLars Ellenberg } 1575a0fb3c47SLars Ellenberg 157645bb912bSLars Ellenberg /* In most cases, we will only need one bio. But in case the lower 157745bb912bSLars Ellenberg * level restrictions happen to be different at this offset on this 157845bb912bSLars Ellenberg * side than those of the sending peer, we may need to submit the 15799476f39dSLars Ellenberg * request in more than one bio. 15809476f39dSLars Ellenberg * 15819476f39dSLars Ellenberg * Plain bio_alloc is good enough here, this is no DRBD internally 15829476f39dSLars Ellenberg * generated bio, but a bio allocated on behalf of the peer. 15839476f39dSLars Ellenberg */ 158445bb912bSLars Ellenberg next_bio: 158545bb912bSLars Ellenberg bio = bio_alloc(GFP_NOIO, nr_pages); 158645bb912bSLars Ellenberg if (!bio) { 1587a0fb3c47SLars Ellenberg drbd_err(device, "submit_ee: Allocation of a bio failed (nr_pages=%u)\n", nr_pages); 158845bb912bSLars Ellenberg goto fail; 158945bb912bSLars Ellenberg } 1590db830c46SAndreas Gruenbacher /* > peer_req->i.sector, unless this is the first bio */ 15914f024f37SKent Overstreet bio->bi_iter.bi_sector = sector; 159274d46992SChristoph Hellwig bio_set_dev(bio, device->ldev->backing_bdev); 1593bb3cc85eSMike Christie bio_set_op_attrs(bio, op, op_flags); 1594db830c46SAndreas Gruenbacher bio->bi_private = peer_req; 1595fcefa62eSAndreas Gruenbacher bio->bi_end_io = drbd_peer_request_endio; 159645bb912bSLars Ellenberg 159745bb912bSLars Ellenberg bio->bi_next = bios; 159845bb912bSLars Ellenberg bios = bio; 159945bb912bSLars Ellenberg ++n_bios; 160045bb912bSLars Ellenberg 160145bb912bSLars Ellenberg page_chain_for_each(page) { 160211f8b2b6SAndreas Gruenbacher unsigned len = min_t(unsigned, data_size, PAGE_SIZE); 160306efffdaSMing Lei if (!bio_add_page(bio, page, len, 0)) 160445bb912bSLars Ellenberg goto next_bio; 160511f8b2b6SAndreas Gruenbacher data_size -= len; 160645bb912bSLars Ellenberg sector += len >> 9; 160745bb912bSLars Ellenberg --nr_pages; 160845bb912bSLars Ellenberg } 160911f8b2b6SAndreas Gruenbacher D_ASSERT(device, data_size == 0); 1610a0fb3c47SLars Ellenberg D_ASSERT(device, page == NULL); 161145bb912bSLars Ellenberg 1612db830c46SAndreas Gruenbacher atomic_set(&peer_req->pending_bios, n_bios); 161321ae5d7fSLars Ellenberg /* for debugfs: update timestamp, mark as submitted */ 161421ae5d7fSLars Ellenberg peer_req->submit_jif = jiffies; 161521ae5d7fSLars Ellenberg peer_req->flags |= EE_SUBMITTED; 161645bb912bSLars Ellenberg do { 161745bb912bSLars Ellenberg bio = bios; 161845bb912bSLars Ellenberg bios = bios->bi_next; 161945bb912bSLars Ellenberg bio->bi_next = NULL; 162045bb912bSLars Ellenberg 1621b30ab791SAndreas Gruenbacher drbd_generic_make_request(device, fault_type, bio); 162245bb912bSLars Ellenberg } while (bios); 162345bb912bSLars Ellenberg return 0; 162445bb912bSLars Ellenberg 162545bb912bSLars Ellenberg fail: 162645bb912bSLars Ellenberg while (bios) { 162745bb912bSLars Ellenberg bio = bios; 162845bb912bSLars Ellenberg bios = bios->bi_next; 162945bb912bSLars Ellenberg bio_put(bio); 163045bb912bSLars Ellenberg } 163110f6d992SLars Ellenberg return err; 163245bb912bSLars Ellenberg } 163345bb912bSLars Ellenberg 1634b30ab791SAndreas Gruenbacher static void drbd_remove_epoch_entry_interval(struct drbd_device *device, 1635db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req) 163653840641SAndreas Gruenbacher { 1637db830c46SAndreas Gruenbacher struct drbd_interval *i = &peer_req->i; 163853840641SAndreas Gruenbacher 1639b30ab791SAndreas Gruenbacher drbd_remove_interval(&device->write_requests, i); 164053840641SAndreas Gruenbacher drbd_clear_interval(i); 164153840641SAndreas Gruenbacher 16426c852becSAndreas Gruenbacher /* Wake up any processes waiting for this peer request to complete. */ 164353840641SAndreas Gruenbacher if (i->waiting) 1644b30ab791SAndreas Gruenbacher wake_up(&device->misc_wait); 164553840641SAndreas Gruenbacher } 164653840641SAndreas Gruenbacher 1647bde89a9eSAndreas Gruenbacher static void conn_wait_active_ee_empty(struct drbd_connection *connection) 164877fede51SPhilipp Reisner { 1649c06ece6bSAndreas Gruenbacher struct drbd_peer_device *peer_device; 165077fede51SPhilipp Reisner int vnr; 165177fede51SPhilipp Reisner 165277fede51SPhilipp Reisner rcu_read_lock(); 1653c06ece6bSAndreas Gruenbacher idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 1654c06ece6bSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 1655c06ece6bSAndreas Gruenbacher 1656b30ab791SAndreas Gruenbacher kref_get(&device->kref); 165777fede51SPhilipp Reisner rcu_read_unlock(); 1658b30ab791SAndreas Gruenbacher drbd_wait_ee_list_empty(device, &device->active_ee); 165905a10ec7SAndreas Gruenbacher kref_put(&device->kref, drbd_destroy_device); 166077fede51SPhilipp Reisner rcu_read_lock(); 166177fede51SPhilipp Reisner } 166277fede51SPhilipp Reisner rcu_read_unlock(); 166377fede51SPhilipp Reisner } 166477fede51SPhilipp Reisner 1665bde89a9eSAndreas Gruenbacher static int receive_Barrier(struct drbd_connection *connection, struct packet_info *pi) 1666b411b363SPhilipp Reisner { 16672451fc3bSPhilipp Reisner int rv; 1668e658983aSAndreas Gruenbacher struct p_barrier *p = pi->data; 1669b411b363SPhilipp Reisner struct drbd_epoch *epoch; 1670b411b363SPhilipp Reisner 16719ed57dcbSLars Ellenberg /* FIXME these are unacked on connection, 16729ed57dcbSLars Ellenberg * not a specific (peer)device. 16739ed57dcbSLars Ellenberg */ 1674bde89a9eSAndreas Gruenbacher connection->current_epoch->barrier_nr = p->barrier; 1675bde89a9eSAndreas Gruenbacher connection->current_epoch->connection = connection; 1676bde89a9eSAndreas Gruenbacher rv = drbd_may_finish_epoch(connection, connection->current_epoch, EV_GOT_BARRIER_NR); 1677b411b363SPhilipp Reisner 1678b411b363SPhilipp Reisner /* P_BARRIER_ACK may imply that the corresponding extent is dropped from 1679b411b363SPhilipp Reisner * the activity log, which means it would not be resynced in case the 1680b411b363SPhilipp Reisner * R_PRIMARY crashes now. 1681b411b363SPhilipp Reisner * Therefore we must send the barrier_ack after the barrier request was 1682b411b363SPhilipp Reisner * completed. */ 1683e9526580SPhilipp Reisner switch (connection->resource->write_ordering) { 1684f6ba8636SAndreas Gruenbacher case WO_NONE: 1685b411b363SPhilipp Reisner if (rv == FE_RECYCLED) 168682bc0194SAndreas Gruenbacher return 0; 1687b411b363SPhilipp Reisner 1688b411b363SPhilipp Reisner /* receiver context, in the writeout path of the other node. 1689b411b363SPhilipp Reisner * avoid potential distributed deadlock */ 1690b411b363SPhilipp Reisner epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); 16912451fc3bSPhilipp Reisner if (epoch) 16922451fc3bSPhilipp Reisner break; 16932451fc3bSPhilipp Reisner else 16941ec861ebSAndreas Gruenbacher drbd_warn(connection, "Allocation of an epoch failed, slowing down\n"); 16952451fc3bSPhilipp Reisner /* Fall through */ 16962451fc3bSPhilipp Reisner 1697f6ba8636SAndreas Gruenbacher case WO_BDEV_FLUSH: 1698f6ba8636SAndreas Gruenbacher case WO_DRAIN_IO: 1699bde89a9eSAndreas Gruenbacher conn_wait_active_ee_empty(connection); 1700bde89a9eSAndreas Gruenbacher drbd_flush(connection); 17012451fc3bSPhilipp Reisner 1702bde89a9eSAndreas Gruenbacher if (atomic_read(&connection->current_epoch->epoch_size)) { 17032451fc3bSPhilipp Reisner epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); 17042451fc3bSPhilipp Reisner if (epoch) 17052451fc3bSPhilipp Reisner break; 1706b411b363SPhilipp Reisner } 1707b411b363SPhilipp Reisner 170882bc0194SAndreas Gruenbacher return 0; 17092451fc3bSPhilipp Reisner default: 1710e9526580SPhilipp Reisner drbd_err(connection, "Strangeness in connection->write_ordering %d\n", 1711e9526580SPhilipp Reisner connection->resource->write_ordering); 171282bc0194SAndreas Gruenbacher return -EIO; 1713b411b363SPhilipp Reisner } 1714b411b363SPhilipp Reisner 1715b411b363SPhilipp Reisner epoch->flags = 0; 1716b411b363SPhilipp Reisner atomic_set(&epoch->epoch_size, 0); 1717b411b363SPhilipp Reisner atomic_set(&epoch->active, 0); 1718b411b363SPhilipp Reisner 1719bde89a9eSAndreas Gruenbacher spin_lock(&connection->epoch_lock); 1720bde89a9eSAndreas Gruenbacher if (atomic_read(&connection->current_epoch->epoch_size)) { 1721bde89a9eSAndreas Gruenbacher list_add(&epoch->list, &connection->current_epoch->list); 1722bde89a9eSAndreas Gruenbacher connection->current_epoch = epoch; 1723bde89a9eSAndreas Gruenbacher connection->epochs++; 1724b411b363SPhilipp Reisner } else { 1725b411b363SPhilipp Reisner /* The current_epoch got recycled while we allocated this one... */ 1726b411b363SPhilipp Reisner kfree(epoch); 1727b411b363SPhilipp Reisner } 1728bde89a9eSAndreas Gruenbacher spin_unlock(&connection->epoch_lock); 1729b411b363SPhilipp Reisner 173082bc0194SAndreas Gruenbacher return 0; 1731b411b363SPhilipp Reisner } 1732b411b363SPhilipp Reisner 17339104d31aSLars Ellenberg /* quick wrapper in case payload size != request_size (write same) */ 17349104d31aSLars Ellenberg static void drbd_csum_ee_size(struct crypto_ahash *h, 17359104d31aSLars Ellenberg struct drbd_peer_request *r, void *d, 17369104d31aSLars Ellenberg unsigned int payload_size) 17379104d31aSLars Ellenberg { 17389104d31aSLars Ellenberg unsigned int tmp = r->i.size; 17399104d31aSLars Ellenberg r->i.size = payload_size; 17409104d31aSLars Ellenberg drbd_csum_ee(h, r, d); 17419104d31aSLars Ellenberg r->i.size = tmp; 17429104d31aSLars Ellenberg } 17439104d31aSLars Ellenberg 1744b411b363SPhilipp Reisner /* used from receive_RSDataReply (recv_resync_read) 17459104d31aSLars Ellenberg * and from receive_Data. 17469104d31aSLars Ellenberg * data_size: actual payload ("data in") 17479104d31aSLars Ellenberg * for normal writes that is bi_size. 17489104d31aSLars Ellenberg * for discards, that is zero. 17499104d31aSLars Ellenberg * for write same, it is logical_block_size. 17509104d31aSLars Ellenberg * both trim and write same have the bi_size ("data len to be affected") 17519104d31aSLars Ellenberg * as extra argument in the packet header. 17529104d31aSLars Ellenberg */ 1753f6ffca9fSAndreas Gruenbacher static struct drbd_peer_request * 175469a22773SAndreas Gruenbacher read_in_block(struct drbd_peer_device *peer_device, u64 id, sector_t sector, 1755a0fb3c47SLars Ellenberg struct packet_info *pi) __must_hold(local) 1756b411b363SPhilipp Reisner { 175769a22773SAndreas Gruenbacher struct drbd_device *device = peer_device->device; 1758b30ab791SAndreas Gruenbacher const sector_t capacity = drbd_get_capacity(device->this_bdev); 1759db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req; 1760b411b363SPhilipp Reisner struct page *page; 176111f8b2b6SAndreas Gruenbacher int digest_size, err; 176211f8b2b6SAndreas Gruenbacher unsigned int data_size = pi->size, ds; 176369a22773SAndreas Gruenbacher void *dig_in = peer_device->connection->int_dig_in; 176469a22773SAndreas Gruenbacher void *dig_vv = peer_device->connection->int_dig_vv; 17656b4388acSPhilipp Reisner unsigned long *data; 1766a0fb3c47SLars Ellenberg struct p_trim *trim = (pi->cmd == P_TRIM) ? pi->data : NULL; 17679104d31aSLars Ellenberg struct p_trim *wsame = (pi->cmd == P_WSAME) ? pi->data : NULL; 1768b411b363SPhilipp Reisner 176911f8b2b6SAndreas Gruenbacher digest_size = 0; 1770a0fb3c47SLars Ellenberg if (!trim && peer_device->connection->peer_integrity_tfm) { 17719534d671SHerbert Xu digest_size = crypto_ahash_digestsize(peer_device->connection->peer_integrity_tfm); 17729f5bdc33SAndreas Gruenbacher /* 17739f5bdc33SAndreas Gruenbacher * FIXME: Receive the incoming digest into the receive buffer 17749f5bdc33SAndreas Gruenbacher * here, together with its struct p_data? 17759f5bdc33SAndreas Gruenbacher */ 177611f8b2b6SAndreas Gruenbacher err = drbd_recv_all_warn(peer_device->connection, dig_in, digest_size); 1777a5c31904SAndreas Gruenbacher if (err) 1778b411b363SPhilipp Reisner return NULL; 177911f8b2b6SAndreas Gruenbacher data_size -= digest_size; 178088104ca4SAndreas Gruenbacher } 1781b411b363SPhilipp Reisner 17829104d31aSLars Ellenberg /* assume request_size == data_size, but special case trim and wsame. */ 17839104d31aSLars Ellenberg ds = data_size; 1784a0fb3c47SLars Ellenberg if (trim) { 17859104d31aSLars Ellenberg if (!expect(data_size == 0)) 17869104d31aSLars Ellenberg return NULL; 17879104d31aSLars Ellenberg ds = be32_to_cpu(trim->size); 17889104d31aSLars Ellenberg } else if (wsame) { 17899104d31aSLars Ellenberg if (data_size != queue_logical_block_size(device->rq_queue)) { 17909104d31aSLars Ellenberg drbd_err(peer_device, "data size (%u) != drbd logical block size (%u)\n", 17919104d31aSLars Ellenberg data_size, queue_logical_block_size(device->rq_queue)); 17929104d31aSLars Ellenberg return NULL; 17939104d31aSLars Ellenberg } 17949104d31aSLars Ellenberg if (data_size != bdev_logical_block_size(device->ldev->backing_bdev)) { 17959104d31aSLars Ellenberg drbd_err(peer_device, "data size (%u) != backend logical block size (%u)\n", 17969104d31aSLars Ellenberg data_size, bdev_logical_block_size(device->ldev->backing_bdev)); 17979104d31aSLars Ellenberg return NULL; 17989104d31aSLars Ellenberg } 17999104d31aSLars Ellenberg ds = be32_to_cpu(wsame->size); 1800a0fb3c47SLars Ellenberg } 1801a0fb3c47SLars Ellenberg 18029104d31aSLars Ellenberg if (!expect(IS_ALIGNED(ds, 512))) 1803841ce241SAndreas Gruenbacher return NULL; 18049104d31aSLars Ellenberg if (trim || wsame) { 18059104d31aSLars Ellenberg if (!expect(ds <= (DRBD_MAX_BBIO_SECTORS << 9))) 18069104d31aSLars Ellenberg return NULL; 18079104d31aSLars Ellenberg } else if (!expect(ds <= DRBD_MAX_BIO_SIZE)) 1808841ce241SAndreas Gruenbacher return NULL; 1809b411b363SPhilipp Reisner 18106666032aSLars Ellenberg /* even though we trust out peer, 18116666032aSLars Ellenberg * we sometimes have to double check. */ 18129104d31aSLars Ellenberg if (sector + (ds>>9) > capacity) { 1813d0180171SAndreas Gruenbacher drbd_err(device, "request from peer beyond end of local disk: " 1814fdda6544SLars Ellenberg "capacity: %llus < sector: %llus + size: %u\n", 18156666032aSLars Ellenberg (unsigned long long)capacity, 18169104d31aSLars Ellenberg (unsigned long long)sector, ds); 18176666032aSLars Ellenberg return NULL; 18186666032aSLars Ellenberg } 18196666032aSLars Ellenberg 1820b411b363SPhilipp Reisner /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD 1821b411b363SPhilipp Reisner * "criss-cross" setup, that might cause write-out on some other DRBD, 1822b411b363SPhilipp Reisner * which in turn might block on the other node at this very place. */ 18239104d31aSLars Ellenberg peer_req = drbd_alloc_peer_req(peer_device, id, sector, ds, data_size, GFP_NOIO); 1824db830c46SAndreas Gruenbacher if (!peer_req) 1825b411b363SPhilipp Reisner return NULL; 182645bb912bSLars Ellenberg 182721ae5d7fSLars Ellenberg peer_req->flags |= EE_WRITE; 18289104d31aSLars Ellenberg if (trim) { 18299104d31aSLars Ellenberg peer_req->flags |= EE_IS_TRIM; 183081a3537aSLars Ellenberg return peer_req; 18319104d31aSLars Ellenberg } 18329104d31aSLars Ellenberg if (wsame) 18339104d31aSLars Ellenberg peer_req->flags |= EE_WRITE_SAME; 1834a73ff323SLars Ellenberg 18359104d31aSLars Ellenberg /* receive payload size bytes into page chain */ 1836b411b363SPhilipp Reisner ds = data_size; 1837db830c46SAndreas Gruenbacher page = peer_req->pages; 183845bb912bSLars Ellenberg page_chain_for_each(page) { 183945bb912bSLars Ellenberg unsigned len = min_t(int, ds, PAGE_SIZE); 18406b4388acSPhilipp Reisner data = kmap(page); 184169a22773SAndreas Gruenbacher err = drbd_recv_all_warn(peer_device->connection, data, len); 1842b30ab791SAndreas Gruenbacher if (drbd_insert_fault(device, DRBD_FAULT_RECEIVE)) { 1843d0180171SAndreas Gruenbacher drbd_err(device, "Fault injection: Corrupting data on receive\n"); 18446b4388acSPhilipp Reisner data[0] = data[0] ^ (unsigned long)-1; 18456b4388acSPhilipp Reisner } 1846b411b363SPhilipp Reisner kunmap(page); 1847a5c31904SAndreas Gruenbacher if (err) { 1848b30ab791SAndreas Gruenbacher drbd_free_peer_req(device, peer_req); 1849b411b363SPhilipp Reisner return NULL; 1850b411b363SPhilipp Reisner } 1851a5c31904SAndreas Gruenbacher ds -= len; 1852b411b363SPhilipp Reisner } 1853b411b363SPhilipp Reisner 185411f8b2b6SAndreas Gruenbacher if (digest_size) { 18559104d31aSLars Ellenberg drbd_csum_ee_size(peer_device->connection->peer_integrity_tfm, peer_req, dig_vv, data_size); 185611f8b2b6SAndreas Gruenbacher if (memcmp(dig_in, dig_vv, digest_size)) { 1857d0180171SAndreas Gruenbacher drbd_err(device, "Digest integrity check FAILED: %llus +%u\n", 1858470be44aSLars Ellenberg (unsigned long long)sector, data_size); 1859b30ab791SAndreas Gruenbacher drbd_free_peer_req(device, peer_req); 1860b411b363SPhilipp Reisner return NULL; 1861b411b363SPhilipp Reisner } 1862b411b363SPhilipp Reisner } 1863b30ab791SAndreas Gruenbacher device->recv_cnt += data_size >> 9; 1864db830c46SAndreas Gruenbacher return peer_req; 1865b411b363SPhilipp Reisner } 1866b411b363SPhilipp Reisner 1867b411b363SPhilipp Reisner /* drbd_drain_block() just takes a data block 1868b411b363SPhilipp Reisner * out of the socket input buffer, and discards it. 1869b411b363SPhilipp Reisner */ 187069a22773SAndreas Gruenbacher static int drbd_drain_block(struct drbd_peer_device *peer_device, int data_size) 1871b411b363SPhilipp Reisner { 1872b411b363SPhilipp Reisner struct page *page; 1873a5c31904SAndreas Gruenbacher int err = 0; 1874b411b363SPhilipp Reisner void *data; 1875b411b363SPhilipp Reisner 1876c3470cdeSLars Ellenberg if (!data_size) 1877fc5be839SAndreas Gruenbacher return 0; 1878c3470cdeSLars Ellenberg 187969a22773SAndreas Gruenbacher page = drbd_alloc_pages(peer_device, 1, 1); 1880b411b363SPhilipp Reisner 1881b411b363SPhilipp Reisner data = kmap(page); 1882b411b363SPhilipp Reisner while (data_size) { 1883fc5be839SAndreas Gruenbacher unsigned int len = min_t(int, data_size, PAGE_SIZE); 1884fc5be839SAndreas Gruenbacher 188569a22773SAndreas Gruenbacher err = drbd_recv_all_warn(peer_device->connection, data, len); 1886a5c31904SAndreas Gruenbacher if (err) 1887b411b363SPhilipp Reisner break; 1888a5c31904SAndreas Gruenbacher data_size -= len; 1889b411b363SPhilipp Reisner } 1890b411b363SPhilipp Reisner kunmap(page); 189169a22773SAndreas Gruenbacher drbd_free_pages(peer_device->device, page, 0); 1892fc5be839SAndreas Gruenbacher return err; 1893b411b363SPhilipp Reisner } 1894b411b363SPhilipp Reisner 189569a22773SAndreas Gruenbacher static int recv_dless_read(struct drbd_peer_device *peer_device, struct drbd_request *req, 1896b411b363SPhilipp Reisner sector_t sector, int data_size) 1897b411b363SPhilipp Reisner { 18987988613bSKent Overstreet struct bio_vec bvec; 18997988613bSKent Overstreet struct bvec_iter iter; 1900b411b363SPhilipp Reisner struct bio *bio; 190111f8b2b6SAndreas Gruenbacher int digest_size, err, expect; 190269a22773SAndreas Gruenbacher void *dig_in = peer_device->connection->int_dig_in; 190369a22773SAndreas Gruenbacher void *dig_vv = peer_device->connection->int_dig_vv; 1904b411b363SPhilipp Reisner 190511f8b2b6SAndreas Gruenbacher digest_size = 0; 190669a22773SAndreas Gruenbacher if (peer_device->connection->peer_integrity_tfm) { 19079534d671SHerbert Xu digest_size = crypto_ahash_digestsize(peer_device->connection->peer_integrity_tfm); 190811f8b2b6SAndreas Gruenbacher err = drbd_recv_all_warn(peer_device->connection, dig_in, digest_size); 1909a5c31904SAndreas Gruenbacher if (err) 1910a5c31904SAndreas Gruenbacher return err; 191111f8b2b6SAndreas Gruenbacher data_size -= digest_size; 191288104ca4SAndreas Gruenbacher } 1913b411b363SPhilipp Reisner 1914b411b363SPhilipp Reisner /* optimistically update recv_cnt. if receiving fails below, 1915b411b363SPhilipp Reisner * we disconnect anyways, and counters will be reset. */ 191669a22773SAndreas Gruenbacher peer_device->device->recv_cnt += data_size>>9; 1917b411b363SPhilipp Reisner 1918b411b363SPhilipp Reisner bio = req->master_bio; 191969a22773SAndreas Gruenbacher D_ASSERT(peer_device->device, sector == bio->bi_iter.bi_sector); 1920b411b363SPhilipp Reisner 19217988613bSKent Overstreet bio_for_each_segment(bvec, bio, iter) { 19227988613bSKent Overstreet void *mapped = kmap(bvec.bv_page) + bvec.bv_offset; 19237988613bSKent Overstreet expect = min_t(int, data_size, bvec.bv_len); 192469a22773SAndreas Gruenbacher err = drbd_recv_all_warn(peer_device->connection, mapped, expect); 19257988613bSKent Overstreet kunmap(bvec.bv_page); 1926a5c31904SAndreas Gruenbacher if (err) 1927a5c31904SAndreas Gruenbacher return err; 1928a5c31904SAndreas Gruenbacher data_size -= expect; 1929b411b363SPhilipp Reisner } 1930b411b363SPhilipp Reisner 193111f8b2b6SAndreas Gruenbacher if (digest_size) { 193269a22773SAndreas Gruenbacher drbd_csum_bio(peer_device->connection->peer_integrity_tfm, bio, dig_vv); 193311f8b2b6SAndreas Gruenbacher if (memcmp(dig_in, dig_vv, digest_size)) { 193469a22773SAndreas Gruenbacher drbd_err(peer_device, "Digest integrity check FAILED. Broken NICs?\n"); 193528284cefSAndreas Gruenbacher return -EINVAL; 1936b411b363SPhilipp Reisner } 1937b411b363SPhilipp Reisner } 1938b411b363SPhilipp Reisner 193969a22773SAndreas Gruenbacher D_ASSERT(peer_device->device, data_size == 0); 194028284cefSAndreas Gruenbacher return 0; 1941b411b363SPhilipp Reisner } 1942b411b363SPhilipp Reisner 1943a990be46SAndreas Gruenbacher /* 1944668700b4SPhilipp Reisner * e_end_resync_block() is called in ack_sender context via 1945a990be46SAndreas Gruenbacher * drbd_finish_peer_reqs(). 1946a990be46SAndreas Gruenbacher */ 194799920dc5SAndreas Gruenbacher static int e_end_resync_block(struct drbd_work *w, int unused) 1948b411b363SPhilipp Reisner { 19498050e6d0SAndreas Gruenbacher struct drbd_peer_request *peer_req = 1950a8cd15baSAndreas Gruenbacher container_of(w, struct drbd_peer_request, w); 1951a8cd15baSAndreas Gruenbacher struct drbd_peer_device *peer_device = peer_req->peer_device; 1952a8cd15baSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 1953db830c46SAndreas Gruenbacher sector_t sector = peer_req->i.sector; 195499920dc5SAndreas Gruenbacher int err; 1955b411b363SPhilipp Reisner 19560b0ba1efSAndreas Gruenbacher D_ASSERT(device, drbd_interval_empty(&peer_req->i)); 1957b411b363SPhilipp Reisner 1958db830c46SAndreas Gruenbacher if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { 1959b30ab791SAndreas Gruenbacher drbd_set_in_sync(device, sector, peer_req->i.size); 1960a8cd15baSAndreas Gruenbacher err = drbd_send_ack(peer_device, P_RS_WRITE_ACK, peer_req); 1961b411b363SPhilipp Reisner } else { 1962b411b363SPhilipp Reisner /* Record failure to sync */ 1963b30ab791SAndreas Gruenbacher drbd_rs_failed_io(device, sector, peer_req->i.size); 1964b411b363SPhilipp Reisner 1965a8cd15baSAndreas Gruenbacher err = drbd_send_ack(peer_device, P_NEG_ACK, peer_req); 1966b411b363SPhilipp Reisner } 1967b30ab791SAndreas Gruenbacher dec_unacked(device); 1968b411b363SPhilipp Reisner 196999920dc5SAndreas Gruenbacher return err; 1970b411b363SPhilipp Reisner } 1971b411b363SPhilipp Reisner 197269a22773SAndreas Gruenbacher static int recv_resync_read(struct drbd_peer_device *peer_device, sector_t sector, 1973a0fb3c47SLars Ellenberg struct packet_info *pi) __releases(local) 1974b411b363SPhilipp Reisner { 197569a22773SAndreas Gruenbacher struct drbd_device *device = peer_device->device; 1976db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req; 1977b411b363SPhilipp Reisner 1978a0fb3c47SLars Ellenberg peer_req = read_in_block(peer_device, ID_SYNCER, sector, pi); 1979db830c46SAndreas Gruenbacher if (!peer_req) 198045bb912bSLars Ellenberg goto fail; 1981b411b363SPhilipp Reisner 1982b30ab791SAndreas Gruenbacher dec_rs_pending(device); 1983b411b363SPhilipp Reisner 1984b30ab791SAndreas Gruenbacher inc_unacked(device); 1985b411b363SPhilipp Reisner /* corresponding dec_unacked() in e_end_resync_block() 1986b411b363SPhilipp Reisner * respective _drbd_clear_done_ee */ 1987b411b363SPhilipp Reisner 1988a8cd15baSAndreas Gruenbacher peer_req->w.cb = e_end_resync_block; 198921ae5d7fSLars Ellenberg peer_req->submit_jif = jiffies; 199045bb912bSLars Ellenberg 19910500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 1992b9ed7080SLars Ellenberg list_add_tail(&peer_req->w.list, &device->sync_ee); 19930500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 1994b411b363SPhilipp Reisner 1995a0fb3c47SLars Ellenberg atomic_add(pi->size >> 9, &device->rs_sect_ev); 1996bb3cc85eSMike Christie if (drbd_submit_peer_request(device, peer_req, REQ_OP_WRITE, 0, 1997bb3cc85eSMike Christie DRBD_FAULT_RS_WR) == 0) 1998e1c1b0fcSAndreas Gruenbacher return 0; 199945bb912bSLars Ellenberg 200010f6d992SLars Ellenberg /* don't care for the reason here */ 2001d0180171SAndreas Gruenbacher drbd_err(device, "submit failed, triggering re-connect\n"); 20020500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 2003a8cd15baSAndreas Gruenbacher list_del(&peer_req->w.list); 20040500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 200522cc37a9SLars Ellenberg 2006b30ab791SAndreas Gruenbacher drbd_free_peer_req(device, peer_req); 200745bb912bSLars Ellenberg fail: 2008b30ab791SAndreas Gruenbacher put_ldev(device); 2009e1c1b0fcSAndreas Gruenbacher return -EIO; 2010b411b363SPhilipp Reisner } 2011b411b363SPhilipp Reisner 2012668eebc6SAndreas Gruenbacher static struct drbd_request * 2013b30ab791SAndreas Gruenbacher find_request(struct drbd_device *device, struct rb_root *root, u64 id, 2014bc9c5c41SAndreas Gruenbacher sector_t sector, bool missing_ok, const char *func) 2015b411b363SPhilipp Reisner { 2016b411b363SPhilipp Reisner struct drbd_request *req; 2017668eebc6SAndreas Gruenbacher 2018bc9c5c41SAndreas Gruenbacher /* Request object according to our peer */ 2019bc9c5c41SAndreas Gruenbacher req = (struct drbd_request *)(unsigned long)id; 20205e472264SAndreas Gruenbacher if (drbd_contains_interval(root, sector, &req->i) && req->i.local) 2021668eebc6SAndreas Gruenbacher return req; 2022c3afd8f5SAndreas Gruenbacher if (!missing_ok) { 2023d0180171SAndreas Gruenbacher drbd_err(device, "%s: failed to find request 0x%lx, sector %llus\n", func, 2024c3afd8f5SAndreas Gruenbacher (unsigned long)id, (unsigned long long)sector); 2025c3afd8f5SAndreas Gruenbacher } 2026668eebc6SAndreas Gruenbacher return NULL; 2027668eebc6SAndreas Gruenbacher } 2028668eebc6SAndreas Gruenbacher 2029bde89a9eSAndreas Gruenbacher static int receive_DataReply(struct drbd_connection *connection, struct packet_info *pi) 2030b411b363SPhilipp Reisner { 20319f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 2032b30ab791SAndreas Gruenbacher struct drbd_device *device; 2033b411b363SPhilipp Reisner struct drbd_request *req; 2034b411b363SPhilipp Reisner sector_t sector; 203582bc0194SAndreas Gruenbacher int err; 2036e658983aSAndreas Gruenbacher struct p_data *p = pi->data; 20374a76b161SAndreas Gruenbacher 20389f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 20399f4fe9adSAndreas Gruenbacher if (!peer_device) 20404a76b161SAndreas Gruenbacher return -EIO; 20419f4fe9adSAndreas Gruenbacher device = peer_device->device; 2042b411b363SPhilipp Reisner 2043b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 2044b411b363SPhilipp Reisner 20450500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 2046b30ab791SAndreas Gruenbacher req = find_request(device, &device->read_requests, p->block_id, sector, false, __func__); 20470500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 2048c3afd8f5SAndreas Gruenbacher if (unlikely(!req)) 204982bc0194SAndreas Gruenbacher return -EIO; 2050b411b363SPhilipp Reisner 205124c4830cSBart Van Assche /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid 2052b411b363SPhilipp Reisner * special casing it there for the various failure cases. 2053b411b363SPhilipp Reisner * still no race with drbd_fail_pending_reads */ 205469a22773SAndreas Gruenbacher err = recv_dless_read(peer_device, req, sector, pi->size); 205582bc0194SAndreas Gruenbacher if (!err) 20568554df1cSAndreas Gruenbacher req_mod(req, DATA_RECEIVED); 2057b411b363SPhilipp Reisner /* else: nothing. handled from drbd_disconnect... 2058b411b363SPhilipp Reisner * I don't think we may complete this just yet 2059b411b363SPhilipp Reisner * in case we are "on-disconnect: freeze" */ 2060b411b363SPhilipp Reisner 206182bc0194SAndreas Gruenbacher return err; 2062b411b363SPhilipp Reisner } 2063b411b363SPhilipp Reisner 2064bde89a9eSAndreas Gruenbacher static int receive_RSDataReply(struct drbd_connection *connection, struct packet_info *pi) 2065b411b363SPhilipp Reisner { 20669f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 2067b30ab791SAndreas Gruenbacher struct drbd_device *device; 2068b411b363SPhilipp Reisner sector_t sector; 206982bc0194SAndreas Gruenbacher int err; 2070e658983aSAndreas Gruenbacher struct p_data *p = pi->data; 20714a76b161SAndreas Gruenbacher 20729f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 20739f4fe9adSAndreas Gruenbacher if (!peer_device) 20744a76b161SAndreas Gruenbacher return -EIO; 20759f4fe9adSAndreas Gruenbacher device = peer_device->device; 2076b411b363SPhilipp Reisner 2077b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 20780b0ba1efSAndreas Gruenbacher D_ASSERT(device, p->block_id == ID_SYNCER); 2079b411b363SPhilipp Reisner 2080b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 2081b411b363SPhilipp Reisner /* data is submitted to disk within recv_resync_read. 2082b411b363SPhilipp Reisner * corresponding put_ldev done below on error, 2083fcefa62eSAndreas Gruenbacher * or in drbd_peer_request_endio. */ 2084a0fb3c47SLars Ellenberg err = recv_resync_read(peer_device, sector, pi); 2085b411b363SPhilipp Reisner } else { 2086b411b363SPhilipp Reisner if (__ratelimit(&drbd_ratelimit_state)) 2087d0180171SAndreas Gruenbacher drbd_err(device, "Can not write resync data to local disk.\n"); 2088b411b363SPhilipp Reisner 208969a22773SAndreas Gruenbacher err = drbd_drain_block(peer_device, pi->size); 2090b411b363SPhilipp Reisner 209169a22773SAndreas Gruenbacher drbd_send_ack_dp(peer_device, P_NEG_ACK, p, pi->size); 2092b411b363SPhilipp Reisner } 2093b411b363SPhilipp Reisner 2094b30ab791SAndreas Gruenbacher atomic_add(pi->size >> 9, &device->rs_sect_in); 2095778f271dSPhilipp Reisner 209682bc0194SAndreas Gruenbacher return err; 2097b411b363SPhilipp Reisner } 2098b411b363SPhilipp Reisner 2099b30ab791SAndreas Gruenbacher static void restart_conflicting_writes(struct drbd_device *device, 21007be8da07SAndreas Gruenbacher sector_t sector, int size) 2101b411b363SPhilipp Reisner { 21027be8da07SAndreas Gruenbacher struct drbd_interval *i; 21037be8da07SAndreas Gruenbacher struct drbd_request *req; 2104b411b363SPhilipp Reisner 2105b30ab791SAndreas Gruenbacher drbd_for_each_overlap(i, &device->write_requests, sector, size) { 21067be8da07SAndreas Gruenbacher if (!i->local) 21077be8da07SAndreas Gruenbacher continue; 21087be8da07SAndreas Gruenbacher req = container_of(i, struct drbd_request, i); 21097be8da07SAndreas Gruenbacher if (req->rq_state & RQ_LOCAL_PENDING || 21107be8da07SAndreas Gruenbacher !(req->rq_state & RQ_POSTPONED)) 21117be8da07SAndreas Gruenbacher continue; 21122312f0b3SLars Ellenberg /* as it is RQ_POSTPONED, this will cause it to 21132312f0b3SLars Ellenberg * be queued on the retry workqueue. */ 2114d4dabbe2SLars Ellenberg __req_mod(req, CONFLICT_RESOLVED, NULL); 21157be8da07SAndreas Gruenbacher } 21167be8da07SAndreas Gruenbacher } 21177be8da07SAndreas Gruenbacher 2118a990be46SAndreas Gruenbacher /* 2119668700b4SPhilipp Reisner * e_end_block() is called in ack_sender context via drbd_finish_peer_reqs(). 2120b411b363SPhilipp Reisner */ 212199920dc5SAndreas Gruenbacher static int e_end_block(struct drbd_work *w, int cancel) 2122b411b363SPhilipp Reisner { 21238050e6d0SAndreas Gruenbacher struct drbd_peer_request *peer_req = 2124a8cd15baSAndreas Gruenbacher container_of(w, struct drbd_peer_request, w); 2125a8cd15baSAndreas Gruenbacher struct drbd_peer_device *peer_device = peer_req->peer_device; 2126a8cd15baSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 2127db830c46SAndreas Gruenbacher sector_t sector = peer_req->i.sector; 212899920dc5SAndreas Gruenbacher int err = 0, pcmd; 2129b411b363SPhilipp Reisner 2130303d1448SPhilipp Reisner if (peer_req->flags & EE_SEND_WRITE_ACK) { 2131db830c46SAndreas Gruenbacher if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { 2132b30ab791SAndreas Gruenbacher pcmd = (device->state.conn >= C_SYNC_SOURCE && 2133b30ab791SAndreas Gruenbacher device->state.conn <= C_PAUSED_SYNC_T && 2134db830c46SAndreas Gruenbacher peer_req->flags & EE_MAY_SET_IN_SYNC) ? 2135b411b363SPhilipp Reisner P_RS_WRITE_ACK : P_WRITE_ACK; 2136a8cd15baSAndreas Gruenbacher err = drbd_send_ack(peer_device, pcmd, peer_req); 2137b411b363SPhilipp Reisner if (pcmd == P_RS_WRITE_ACK) 2138b30ab791SAndreas Gruenbacher drbd_set_in_sync(device, sector, peer_req->i.size); 2139b411b363SPhilipp Reisner } else { 2140a8cd15baSAndreas Gruenbacher err = drbd_send_ack(peer_device, P_NEG_ACK, peer_req); 2141b411b363SPhilipp Reisner /* we expect it to be marked out of sync anyways... 2142b411b363SPhilipp Reisner * maybe assert this? */ 2143b411b363SPhilipp Reisner } 2144b30ab791SAndreas Gruenbacher dec_unacked(device); 2145b411b363SPhilipp Reisner } 214608d0dabfSLars Ellenberg 2147b411b363SPhilipp Reisner /* we delete from the conflict detection hash _after_ we sent out the 2148b411b363SPhilipp Reisner * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ 2149302bdeaeSPhilipp Reisner if (peer_req->flags & EE_IN_INTERVAL_TREE) { 21500500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 21510b0ba1efSAndreas Gruenbacher D_ASSERT(device, !drbd_interval_empty(&peer_req->i)); 2152b30ab791SAndreas Gruenbacher drbd_remove_epoch_entry_interval(device, peer_req); 21537be8da07SAndreas Gruenbacher if (peer_req->flags & EE_RESTART_REQUESTS) 2154b30ab791SAndreas Gruenbacher restart_conflicting_writes(device, sector, peer_req->i.size); 21550500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 2156bb3bfe96SAndreas Gruenbacher } else 21570b0ba1efSAndreas Gruenbacher D_ASSERT(device, drbd_interval_empty(&peer_req->i)); 2158b411b363SPhilipp Reisner 21595dd2ca19SAndreas Gruenbacher drbd_may_finish_epoch(peer_device->connection, peer_req->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); 2160b411b363SPhilipp Reisner 216199920dc5SAndreas Gruenbacher return err; 2162b411b363SPhilipp Reisner } 2163b411b363SPhilipp Reisner 2164a8cd15baSAndreas Gruenbacher static int e_send_ack(struct drbd_work *w, enum drbd_packet ack) 2165b411b363SPhilipp Reisner { 21668050e6d0SAndreas Gruenbacher struct drbd_peer_request *peer_req = 2167a8cd15baSAndreas Gruenbacher container_of(w, struct drbd_peer_request, w); 2168a8cd15baSAndreas Gruenbacher struct drbd_peer_device *peer_device = peer_req->peer_device; 216999920dc5SAndreas Gruenbacher int err; 2170b411b363SPhilipp Reisner 2171a8cd15baSAndreas Gruenbacher err = drbd_send_ack(peer_device, ack, peer_req); 2172a8cd15baSAndreas Gruenbacher dec_unacked(peer_device->device); 2173b411b363SPhilipp Reisner 217499920dc5SAndreas Gruenbacher return err; 2175b411b363SPhilipp Reisner } 2176b411b363SPhilipp Reisner 2177d4dabbe2SLars Ellenberg static int e_send_superseded(struct drbd_work *w, int unused) 2178b6a370baSPhilipp Reisner { 2179a8cd15baSAndreas Gruenbacher return e_send_ack(w, P_SUPERSEDED); 21807be8da07SAndreas Gruenbacher } 2181b6a370baSPhilipp Reisner 218299920dc5SAndreas Gruenbacher static int e_send_retry_write(struct drbd_work *w, int unused) 21837be8da07SAndreas Gruenbacher { 2184a8cd15baSAndreas Gruenbacher struct drbd_peer_request *peer_req = 2185a8cd15baSAndreas Gruenbacher container_of(w, struct drbd_peer_request, w); 2186a8cd15baSAndreas Gruenbacher struct drbd_connection *connection = peer_req->peer_device->connection; 21877be8da07SAndreas Gruenbacher 2188a8cd15baSAndreas Gruenbacher return e_send_ack(w, connection->agreed_pro_version >= 100 ? 2189d4dabbe2SLars Ellenberg P_RETRY_WRITE : P_SUPERSEDED); 21907be8da07SAndreas Gruenbacher } 21917be8da07SAndreas Gruenbacher 21923e394da1SAndreas Gruenbacher static bool seq_greater(u32 a, u32 b) 21933e394da1SAndreas Gruenbacher { 21943e394da1SAndreas Gruenbacher /* 21953e394da1SAndreas Gruenbacher * We assume 32-bit wrap-around here. 21963e394da1SAndreas Gruenbacher * For 24-bit wrap-around, we would have to shift: 21973e394da1SAndreas Gruenbacher * a <<= 8; b <<= 8; 21983e394da1SAndreas Gruenbacher */ 21993e394da1SAndreas Gruenbacher return (s32)a - (s32)b > 0; 22003e394da1SAndreas Gruenbacher } 22013e394da1SAndreas Gruenbacher 22023e394da1SAndreas Gruenbacher static u32 seq_max(u32 a, u32 b) 22033e394da1SAndreas Gruenbacher { 22043e394da1SAndreas Gruenbacher return seq_greater(a, b) ? a : b; 22053e394da1SAndreas Gruenbacher } 22063e394da1SAndreas Gruenbacher 220769a22773SAndreas Gruenbacher static void update_peer_seq(struct drbd_peer_device *peer_device, unsigned int peer_seq) 22083e394da1SAndreas Gruenbacher { 220969a22773SAndreas Gruenbacher struct drbd_device *device = peer_device->device; 22103c13b680SLars Ellenberg unsigned int newest_peer_seq; 22113e394da1SAndreas Gruenbacher 221269a22773SAndreas Gruenbacher if (test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags)) { 2213b30ab791SAndreas Gruenbacher spin_lock(&device->peer_seq_lock); 2214b30ab791SAndreas Gruenbacher newest_peer_seq = seq_max(device->peer_seq, peer_seq); 2215b30ab791SAndreas Gruenbacher device->peer_seq = newest_peer_seq; 2216b30ab791SAndreas Gruenbacher spin_unlock(&device->peer_seq_lock); 2217b30ab791SAndreas Gruenbacher /* wake up only if we actually changed device->peer_seq */ 22183c13b680SLars Ellenberg if (peer_seq == newest_peer_seq) 2219b30ab791SAndreas Gruenbacher wake_up(&device->seq_wait); 22203e394da1SAndreas Gruenbacher } 22217be8da07SAndreas Gruenbacher } 22223e394da1SAndreas Gruenbacher 2223d93f6302SLars Ellenberg static inline int overlaps(sector_t s1, int l1, sector_t s2, int l2) 2224d93f6302SLars Ellenberg { 2225d93f6302SLars Ellenberg return !((s1 + (l1>>9) <= s2) || (s1 >= s2 + (l2>>9))); 2226d93f6302SLars Ellenberg } 2227d93f6302SLars Ellenberg 2228d93f6302SLars Ellenberg /* maybe change sync_ee into interval trees as well? */ 2229b30ab791SAndreas Gruenbacher static bool overlapping_resync_write(struct drbd_device *device, struct drbd_peer_request *peer_req) 2230d93f6302SLars Ellenberg { 2231d93f6302SLars Ellenberg struct drbd_peer_request *rs_req; 22327e5fec31SFabian Frederick bool rv = false; 2233b6a370baSPhilipp Reisner 22340500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 2235a8cd15baSAndreas Gruenbacher list_for_each_entry(rs_req, &device->sync_ee, w.list) { 2236d93f6302SLars Ellenberg if (overlaps(peer_req->i.sector, peer_req->i.size, 2237d93f6302SLars Ellenberg rs_req->i.sector, rs_req->i.size)) { 22387e5fec31SFabian Frederick rv = true; 2239b6a370baSPhilipp Reisner break; 2240b6a370baSPhilipp Reisner } 2241b6a370baSPhilipp Reisner } 22420500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 2243b6a370baSPhilipp Reisner 2244b6a370baSPhilipp Reisner return rv; 2245b6a370baSPhilipp Reisner } 2246b6a370baSPhilipp Reisner 2247b411b363SPhilipp Reisner /* Called from receive_Data. 2248b411b363SPhilipp Reisner * Synchronize packets on sock with packets on msock. 2249b411b363SPhilipp Reisner * 2250b411b363SPhilipp Reisner * This is here so even when a P_DATA packet traveling via sock overtook an Ack 2251b411b363SPhilipp Reisner * packet traveling on msock, they are still processed in the order they have 2252b411b363SPhilipp Reisner * been sent. 2253b411b363SPhilipp Reisner * 2254b411b363SPhilipp Reisner * Note: we don't care for Ack packets overtaking P_DATA packets. 2255b411b363SPhilipp Reisner * 2256b30ab791SAndreas Gruenbacher * In case packet_seq is larger than device->peer_seq number, there are 2257b411b363SPhilipp Reisner * outstanding packets on the msock. We wait for them to arrive. 2258b30ab791SAndreas Gruenbacher * In case we are the logically next packet, we update device->peer_seq 2259b411b363SPhilipp Reisner * ourselves. Correctly handles 32bit wrap around. 2260b411b363SPhilipp Reisner * 2261b411b363SPhilipp Reisner * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, 2262b411b363SPhilipp Reisner * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds 2263b411b363SPhilipp Reisner * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have 2264b411b363SPhilipp Reisner * 1<<9 == 512 seconds aka ages for the 32bit wrap around... 2265b411b363SPhilipp Reisner * 2266b411b363SPhilipp Reisner * returns 0 if we may process the packet, 2267b411b363SPhilipp Reisner * -ERESTARTSYS if we were interrupted (by disconnect signal). */ 226869a22773SAndreas Gruenbacher static int wait_for_and_update_peer_seq(struct drbd_peer_device *peer_device, const u32 peer_seq) 2269b411b363SPhilipp Reisner { 227069a22773SAndreas Gruenbacher struct drbd_device *device = peer_device->device; 2271b411b363SPhilipp Reisner DEFINE_WAIT(wait); 2272b411b363SPhilipp Reisner long timeout; 2273b874d231SPhilipp Reisner int ret = 0, tp; 22747be8da07SAndreas Gruenbacher 227569a22773SAndreas Gruenbacher if (!test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags)) 22767be8da07SAndreas Gruenbacher return 0; 22777be8da07SAndreas Gruenbacher 2278b30ab791SAndreas Gruenbacher spin_lock(&device->peer_seq_lock); 2279b411b363SPhilipp Reisner for (;;) { 2280b30ab791SAndreas Gruenbacher if (!seq_greater(peer_seq - 1, device->peer_seq)) { 2281b30ab791SAndreas Gruenbacher device->peer_seq = seq_max(device->peer_seq, peer_seq); 2282b411b363SPhilipp Reisner break; 22837be8da07SAndreas Gruenbacher } 2284b874d231SPhilipp Reisner 2285b411b363SPhilipp Reisner if (signal_pending(current)) { 2286b411b363SPhilipp Reisner ret = -ERESTARTSYS; 2287b411b363SPhilipp Reisner break; 2288b411b363SPhilipp Reisner } 2289b874d231SPhilipp Reisner 2290b874d231SPhilipp Reisner rcu_read_lock(); 22915dd2ca19SAndreas Gruenbacher tp = rcu_dereference(peer_device->connection->net_conf)->two_primaries; 2292b874d231SPhilipp Reisner rcu_read_unlock(); 2293b874d231SPhilipp Reisner 2294b874d231SPhilipp Reisner if (!tp) 2295b874d231SPhilipp Reisner break; 2296b874d231SPhilipp Reisner 2297b874d231SPhilipp Reisner /* Only need to wait if two_primaries is enabled */ 2298b30ab791SAndreas Gruenbacher prepare_to_wait(&device->seq_wait, &wait, TASK_INTERRUPTIBLE); 2299b30ab791SAndreas Gruenbacher spin_unlock(&device->peer_seq_lock); 230044ed167dSPhilipp Reisner rcu_read_lock(); 230169a22773SAndreas Gruenbacher timeout = rcu_dereference(peer_device->connection->net_conf)->ping_timeo*HZ/10; 230244ed167dSPhilipp Reisner rcu_read_unlock(); 230371b1c1ebSAndreas Gruenbacher timeout = schedule_timeout(timeout); 2304b30ab791SAndreas Gruenbacher spin_lock(&device->peer_seq_lock); 23057be8da07SAndreas Gruenbacher if (!timeout) { 2306b411b363SPhilipp Reisner ret = -ETIMEDOUT; 2307d0180171SAndreas Gruenbacher drbd_err(device, "Timed out waiting for missing ack packets; disconnecting\n"); 2308b411b363SPhilipp Reisner break; 2309b411b363SPhilipp Reisner } 2310b411b363SPhilipp Reisner } 2311b30ab791SAndreas Gruenbacher spin_unlock(&device->peer_seq_lock); 2312b30ab791SAndreas Gruenbacher finish_wait(&device->seq_wait, &wait); 2313b411b363SPhilipp Reisner return ret; 2314b411b363SPhilipp Reisner } 2315b411b363SPhilipp Reisner 2316688593c5SLars Ellenberg /* see also bio_flags_to_wire() 2317688593c5SLars Ellenberg * DRBD_REQ_*, because we need to semantically map the flags to data packet 2318688593c5SLars Ellenberg * flags and back. We may replicate to other kernel versions. */ 2319bb3cc85eSMike Christie static unsigned long wire_flags_to_bio_flags(u32 dpf) 232076d2e7ecSPhilipp Reisner { 232176d2e7ecSPhilipp Reisner return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | 232276d2e7ecSPhilipp Reisner (dpf & DP_FUA ? REQ_FUA : 0) | 232328a8f0d3SMike Christie (dpf & DP_FLUSH ? REQ_PREFLUSH : 0); 2324bb3cc85eSMike Christie } 2325bb3cc85eSMike Christie 2326bb3cc85eSMike Christie static unsigned long wire_flags_to_bio_op(u32 dpf) 2327bb3cc85eSMike Christie { 2328bb3cc85eSMike Christie if (dpf & DP_DISCARD) 232945c21793SChristoph Hellwig return REQ_OP_WRITE_ZEROES; 2330bb3cc85eSMike Christie else 2331bb3cc85eSMike Christie return REQ_OP_WRITE; 233276d2e7ecSPhilipp Reisner } 233376d2e7ecSPhilipp Reisner 2334b30ab791SAndreas Gruenbacher static void fail_postponed_requests(struct drbd_device *device, sector_t sector, 23357be8da07SAndreas Gruenbacher unsigned int size) 2336b411b363SPhilipp Reisner { 23377be8da07SAndreas Gruenbacher struct drbd_interval *i; 23387be8da07SAndreas Gruenbacher 23397be8da07SAndreas Gruenbacher repeat: 2340b30ab791SAndreas Gruenbacher drbd_for_each_overlap(i, &device->write_requests, sector, size) { 23417be8da07SAndreas Gruenbacher struct drbd_request *req; 23427be8da07SAndreas Gruenbacher struct bio_and_error m; 23437be8da07SAndreas Gruenbacher 23447be8da07SAndreas Gruenbacher if (!i->local) 23457be8da07SAndreas Gruenbacher continue; 23467be8da07SAndreas Gruenbacher req = container_of(i, struct drbd_request, i); 23477be8da07SAndreas Gruenbacher if (!(req->rq_state & RQ_POSTPONED)) 23487be8da07SAndreas Gruenbacher continue; 23497be8da07SAndreas Gruenbacher req->rq_state &= ~RQ_POSTPONED; 23507be8da07SAndreas Gruenbacher __req_mod(req, NEG_ACKED, &m); 23510500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 23527be8da07SAndreas Gruenbacher if (m.bio) 2353b30ab791SAndreas Gruenbacher complete_master_bio(device, &m); 23540500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 23557be8da07SAndreas Gruenbacher goto repeat; 23567be8da07SAndreas Gruenbacher } 23577be8da07SAndreas Gruenbacher } 23587be8da07SAndreas Gruenbacher 2359b30ab791SAndreas Gruenbacher static int handle_write_conflicts(struct drbd_device *device, 23607be8da07SAndreas Gruenbacher struct drbd_peer_request *peer_req) 23617be8da07SAndreas Gruenbacher { 2362e33b32deSAndreas Gruenbacher struct drbd_connection *connection = peer_req->peer_device->connection; 2363bde89a9eSAndreas Gruenbacher bool resolve_conflicts = test_bit(RESOLVE_CONFLICTS, &connection->flags); 23647be8da07SAndreas Gruenbacher sector_t sector = peer_req->i.sector; 23657be8da07SAndreas Gruenbacher const unsigned int size = peer_req->i.size; 23667be8da07SAndreas Gruenbacher struct drbd_interval *i; 23677be8da07SAndreas Gruenbacher bool equal; 23687be8da07SAndreas Gruenbacher int err; 23697be8da07SAndreas Gruenbacher 23707be8da07SAndreas Gruenbacher /* 23717be8da07SAndreas Gruenbacher * Inserting the peer request into the write_requests tree will prevent 23727be8da07SAndreas Gruenbacher * new conflicting local requests from being added. 23737be8da07SAndreas Gruenbacher */ 2374b30ab791SAndreas Gruenbacher drbd_insert_interval(&device->write_requests, &peer_req->i); 23757be8da07SAndreas Gruenbacher 23767be8da07SAndreas Gruenbacher repeat: 2377b30ab791SAndreas Gruenbacher drbd_for_each_overlap(i, &device->write_requests, sector, size) { 23787be8da07SAndreas Gruenbacher if (i == &peer_req->i) 23797be8da07SAndreas Gruenbacher continue; 238008d0dabfSLars Ellenberg if (i->completed) 238108d0dabfSLars Ellenberg continue; 23827be8da07SAndreas Gruenbacher 23837be8da07SAndreas Gruenbacher if (!i->local) { 23847be8da07SAndreas Gruenbacher /* 23857be8da07SAndreas Gruenbacher * Our peer has sent a conflicting remote request; this 23867be8da07SAndreas Gruenbacher * should not happen in a two-node setup. Wait for the 23877be8da07SAndreas Gruenbacher * earlier peer request to complete. 23887be8da07SAndreas Gruenbacher */ 2389b30ab791SAndreas Gruenbacher err = drbd_wait_misc(device, i); 23907be8da07SAndreas Gruenbacher if (err) 23917be8da07SAndreas Gruenbacher goto out; 23927be8da07SAndreas Gruenbacher goto repeat; 23937be8da07SAndreas Gruenbacher } 23947be8da07SAndreas Gruenbacher 23957be8da07SAndreas Gruenbacher equal = i->sector == sector && i->size == size; 23967be8da07SAndreas Gruenbacher if (resolve_conflicts) { 23977be8da07SAndreas Gruenbacher /* 23987be8da07SAndreas Gruenbacher * If the peer request is fully contained within the 2399d4dabbe2SLars Ellenberg * overlapping request, it can be considered overwritten 2400d4dabbe2SLars Ellenberg * and thus superseded; otherwise, it will be retried 2401d4dabbe2SLars Ellenberg * once all overlapping requests have completed. 24027be8da07SAndreas Gruenbacher */ 2403d4dabbe2SLars Ellenberg bool superseded = i->sector <= sector && i->sector + 24047be8da07SAndreas Gruenbacher (i->size >> 9) >= sector + (size >> 9); 24057be8da07SAndreas Gruenbacher 24067be8da07SAndreas Gruenbacher if (!equal) 2407d0180171SAndreas Gruenbacher drbd_alert(device, "Concurrent writes detected: " 24087be8da07SAndreas Gruenbacher "local=%llus +%u, remote=%llus +%u, " 24097be8da07SAndreas Gruenbacher "assuming %s came first\n", 24107be8da07SAndreas Gruenbacher (unsigned long long)i->sector, i->size, 24117be8da07SAndreas Gruenbacher (unsigned long long)sector, size, 2412d4dabbe2SLars Ellenberg superseded ? "local" : "remote"); 24137be8da07SAndreas Gruenbacher 2414a8cd15baSAndreas Gruenbacher peer_req->w.cb = superseded ? e_send_superseded : 24157be8da07SAndreas Gruenbacher e_send_retry_write; 2416a8cd15baSAndreas Gruenbacher list_add_tail(&peer_req->w.list, &device->done_ee); 2417668700b4SPhilipp Reisner queue_work(connection->ack_sender, &peer_req->peer_device->send_acks_work); 24187be8da07SAndreas Gruenbacher 24197be8da07SAndreas Gruenbacher err = -ENOENT; 24207be8da07SAndreas Gruenbacher goto out; 24217be8da07SAndreas Gruenbacher } else { 24227be8da07SAndreas Gruenbacher struct drbd_request *req = 24237be8da07SAndreas Gruenbacher container_of(i, struct drbd_request, i); 24247be8da07SAndreas Gruenbacher 24257be8da07SAndreas Gruenbacher if (!equal) 2426d0180171SAndreas Gruenbacher drbd_alert(device, "Concurrent writes detected: " 24277be8da07SAndreas Gruenbacher "local=%llus +%u, remote=%llus +%u\n", 24287be8da07SAndreas Gruenbacher (unsigned long long)i->sector, i->size, 24297be8da07SAndreas Gruenbacher (unsigned long long)sector, size); 24307be8da07SAndreas Gruenbacher 24317be8da07SAndreas Gruenbacher if (req->rq_state & RQ_LOCAL_PENDING || 24327be8da07SAndreas Gruenbacher !(req->rq_state & RQ_POSTPONED)) { 24337be8da07SAndreas Gruenbacher /* 24347be8da07SAndreas Gruenbacher * Wait for the node with the discard flag to 2435d4dabbe2SLars Ellenberg * decide if this request has been superseded 2436d4dabbe2SLars Ellenberg * or needs to be retried. 2437d4dabbe2SLars Ellenberg * Requests that have been superseded will 24387be8da07SAndreas Gruenbacher * disappear from the write_requests tree. 24397be8da07SAndreas Gruenbacher * 24407be8da07SAndreas Gruenbacher * In addition, wait for the conflicting 24417be8da07SAndreas Gruenbacher * request to finish locally before submitting 24427be8da07SAndreas Gruenbacher * the conflicting peer request. 24437be8da07SAndreas Gruenbacher */ 2444b30ab791SAndreas Gruenbacher err = drbd_wait_misc(device, &req->i); 24457be8da07SAndreas Gruenbacher if (err) { 2446e33b32deSAndreas Gruenbacher _conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD); 2447b30ab791SAndreas Gruenbacher fail_postponed_requests(device, sector, size); 24487be8da07SAndreas Gruenbacher goto out; 24497be8da07SAndreas Gruenbacher } 24507be8da07SAndreas Gruenbacher goto repeat; 24517be8da07SAndreas Gruenbacher } 24527be8da07SAndreas Gruenbacher /* 24537be8da07SAndreas Gruenbacher * Remember to restart the conflicting requests after 24547be8da07SAndreas Gruenbacher * the new peer request has completed. 24557be8da07SAndreas Gruenbacher */ 24567be8da07SAndreas Gruenbacher peer_req->flags |= EE_RESTART_REQUESTS; 24577be8da07SAndreas Gruenbacher } 24587be8da07SAndreas Gruenbacher } 24597be8da07SAndreas Gruenbacher err = 0; 24607be8da07SAndreas Gruenbacher 24617be8da07SAndreas Gruenbacher out: 24627be8da07SAndreas Gruenbacher if (err) 2463b30ab791SAndreas Gruenbacher drbd_remove_epoch_entry_interval(device, peer_req); 24647be8da07SAndreas Gruenbacher return err; 24657be8da07SAndreas Gruenbacher } 24667be8da07SAndreas Gruenbacher 2467b411b363SPhilipp Reisner /* mirrored write */ 2468bde89a9eSAndreas Gruenbacher static int receive_Data(struct drbd_connection *connection, struct packet_info *pi) 2469b411b363SPhilipp Reisner { 24709f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 2471b30ab791SAndreas Gruenbacher struct drbd_device *device; 247221ae5d7fSLars Ellenberg struct net_conf *nc; 2473b411b363SPhilipp Reisner sector_t sector; 2474db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req; 2475e658983aSAndreas Gruenbacher struct p_data *p = pi->data; 24767be8da07SAndreas Gruenbacher u32 peer_seq = be32_to_cpu(p->seq_num); 2477bb3cc85eSMike Christie int op, op_flags; 2478b411b363SPhilipp Reisner u32 dp_flags; 2479302bdeaeSPhilipp Reisner int err, tp; 24807be8da07SAndreas Gruenbacher 24819f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 24829f4fe9adSAndreas Gruenbacher if (!peer_device) 24834a76b161SAndreas Gruenbacher return -EIO; 24849f4fe9adSAndreas Gruenbacher device = peer_device->device; 2485b411b363SPhilipp Reisner 2486b30ab791SAndreas Gruenbacher if (!get_ldev(device)) { 248782bc0194SAndreas Gruenbacher int err2; 2488b411b363SPhilipp Reisner 248969a22773SAndreas Gruenbacher err = wait_for_and_update_peer_seq(peer_device, peer_seq); 249069a22773SAndreas Gruenbacher drbd_send_ack_dp(peer_device, P_NEG_ACK, p, pi->size); 2491bde89a9eSAndreas Gruenbacher atomic_inc(&connection->current_epoch->epoch_size); 249269a22773SAndreas Gruenbacher err2 = drbd_drain_block(peer_device, pi->size); 249382bc0194SAndreas Gruenbacher if (!err) 249482bc0194SAndreas Gruenbacher err = err2; 249582bc0194SAndreas Gruenbacher return err; 2496b411b363SPhilipp Reisner } 2497b411b363SPhilipp Reisner 2498fcefa62eSAndreas Gruenbacher /* 2499fcefa62eSAndreas Gruenbacher * Corresponding put_ldev done either below (on various errors), or in 2500fcefa62eSAndreas Gruenbacher * drbd_peer_request_endio, if we successfully submit the data at the 2501fcefa62eSAndreas Gruenbacher * end of this function. 2502fcefa62eSAndreas Gruenbacher */ 2503b411b363SPhilipp Reisner 2504b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 2505a0fb3c47SLars Ellenberg peer_req = read_in_block(peer_device, p->block_id, sector, pi); 2506db830c46SAndreas Gruenbacher if (!peer_req) { 2507b30ab791SAndreas Gruenbacher put_ldev(device); 250882bc0194SAndreas Gruenbacher return -EIO; 2509b411b363SPhilipp Reisner } 2510b411b363SPhilipp Reisner 2511a8cd15baSAndreas Gruenbacher peer_req->w.cb = e_end_block; 251221ae5d7fSLars Ellenberg peer_req->submit_jif = jiffies; 251321ae5d7fSLars Ellenberg peer_req->flags |= EE_APPLICATION; 2514b411b363SPhilipp Reisner 2515688593c5SLars Ellenberg dp_flags = be32_to_cpu(p->dp_flags); 2516bb3cc85eSMike Christie op = wire_flags_to_bio_op(dp_flags); 2517bb3cc85eSMike Christie op_flags = wire_flags_to_bio_flags(dp_flags); 2518a0fb3c47SLars Ellenberg if (pi->cmd == P_TRIM) { 2519a0fb3c47SLars Ellenberg D_ASSERT(peer_device, peer_req->i.size > 0); 252045c21793SChristoph Hellwig D_ASSERT(peer_device, op == REQ_OP_WRITE_ZEROES); 2521a0fb3c47SLars Ellenberg D_ASSERT(peer_device, peer_req->pages == NULL); 2522a0fb3c47SLars Ellenberg } else if (peer_req->pages == NULL) { 25230b0ba1efSAndreas Gruenbacher D_ASSERT(device, peer_req->i.size == 0); 25240b0ba1efSAndreas Gruenbacher D_ASSERT(device, dp_flags & DP_FLUSH); 2525a73ff323SLars Ellenberg } 2526688593c5SLars Ellenberg 2527688593c5SLars Ellenberg if (dp_flags & DP_MAY_SET_IN_SYNC) 2528db830c46SAndreas Gruenbacher peer_req->flags |= EE_MAY_SET_IN_SYNC; 2529688593c5SLars Ellenberg 2530bde89a9eSAndreas Gruenbacher spin_lock(&connection->epoch_lock); 2531bde89a9eSAndreas Gruenbacher peer_req->epoch = connection->current_epoch; 2532db830c46SAndreas Gruenbacher atomic_inc(&peer_req->epoch->epoch_size); 2533db830c46SAndreas Gruenbacher atomic_inc(&peer_req->epoch->active); 2534bde89a9eSAndreas Gruenbacher spin_unlock(&connection->epoch_lock); 2535b411b363SPhilipp Reisner 2536302bdeaeSPhilipp Reisner rcu_read_lock(); 253721ae5d7fSLars Ellenberg nc = rcu_dereference(peer_device->connection->net_conf); 253821ae5d7fSLars Ellenberg tp = nc->two_primaries; 253921ae5d7fSLars Ellenberg if (peer_device->connection->agreed_pro_version < 100) { 254021ae5d7fSLars Ellenberg switch (nc->wire_protocol) { 254121ae5d7fSLars Ellenberg case DRBD_PROT_C: 254221ae5d7fSLars Ellenberg dp_flags |= DP_SEND_WRITE_ACK; 254321ae5d7fSLars Ellenberg break; 254421ae5d7fSLars Ellenberg case DRBD_PROT_B: 254521ae5d7fSLars Ellenberg dp_flags |= DP_SEND_RECEIVE_ACK; 254621ae5d7fSLars Ellenberg break; 254721ae5d7fSLars Ellenberg } 254821ae5d7fSLars Ellenberg } 2549302bdeaeSPhilipp Reisner rcu_read_unlock(); 255021ae5d7fSLars Ellenberg 255121ae5d7fSLars Ellenberg if (dp_flags & DP_SEND_WRITE_ACK) { 255221ae5d7fSLars Ellenberg peer_req->flags |= EE_SEND_WRITE_ACK; 255321ae5d7fSLars Ellenberg inc_unacked(device); 255421ae5d7fSLars Ellenberg /* corresponding dec_unacked() in e_end_block() 255521ae5d7fSLars Ellenberg * respective _drbd_clear_done_ee */ 255621ae5d7fSLars Ellenberg } 255721ae5d7fSLars Ellenberg 255821ae5d7fSLars Ellenberg if (dp_flags & DP_SEND_RECEIVE_ACK) { 255921ae5d7fSLars Ellenberg /* I really don't like it that the receiver thread 256021ae5d7fSLars Ellenberg * sends on the msock, but anyways */ 25615dd2ca19SAndreas Gruenbacher drbd_send_ack(peer_device, P_RECV_ACK, peer_req); 256221ae5d7fSLars Ellenberg } 256321ae5d7fSLars Ellenberg 2564302bdeaeSPhilipp Reisner if (tp) { 256521ae5d7fSLars Ellenberg /* two primaries implies protocol C */ 256621ae5d7fSLars Ellenberg D_ASSERT(device, dp_flags & DP_SEND_WRITE_ACK); 2567302bdeaeSPhilipp Reisner peer_req->flags |= EE_IN_INTERVAL_TREE; 256869a22773SAndreas Gruenbacher err = wait_for_and_update_peer_seq(peer_device, peer_seq); 25697be8da07SAndreas Gruenbacher if (err) 2570b411b363SPhilipp Reisner goto out_interrupted; 25710500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 2572b30ab791SAndreas Gruenbacher err = handle_write_conflicts(device, peer_req); 25737be8da07SAndreas Gruenbacher if (err) { 25740500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 25757be8da07SAndreas Gruenbacher if (err == -ENOENT) { 2576b30ab791SAndreas Gruenbacher put_ldev(device); 257782bc0194SAndreas Gruenbacher return 0; 2578b411b363SPhilipp Reisner } 2579b411b363SPhilipp Reisner goto out_interrupted; 2580b411b363SPhilipp Reisner } 2581b874d231SPhilipp Reisner } else { 258269a22773SAndreas Gruenbacher update_peer_seq(peer_device, peer_seq); 25830500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 2584b874d231SPhilipp Reisner } 25859104d31aSLars Ellenberg /* TRIM and WRITE_SAME are processed synchronously, 25869104d31aSLars Ellenberg * we wait for all pending requests, respectively wait for 2587a0fb3c47SLars Ellenberg * active_ee to become empty in drbd_submit_peer_request(); 2588a0fb3c47SLars Ellenberg * better not add ourselves here. */ 25899104d31aSLars Ellenberg if ((peer_req->flags & (EE_IS_TRIM|EE_WRITE_SAME)) == 0) 2590b9ed7080SLars Ellenberg list_add_tail(&peer_req->w.list, &device->active_ee); 25910500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 2592b411b363SPhilipp Reisner 2593b30ab791SAndreas Gruenbacher if (device->state.conn == C_SYNC_TARGET) 2594b30ab791SAndreas Gruenbacher wait_event(device->ee_wait, !overlapping_resync_write(device, peer_req)); 2595b6a370baSPhilipp Reisner 2596b30ab791SAndreas Gruenbacher if (device->state.pdsk < D_INCONSISTENT) { 2597b411b363SPhilipp Reisner /* In case we have the only disk of the cluster, */ 2598b30ab791SAndreas Gruenbacher drbd_set_out_of_sync(device, peer_req->i.sector, peer_req->i.size); 2599db830c46SAndreas Gruenbacher peer_req->flags &= ~EE_MAY_SET_IN_SYNC; 26004dd726f0SLars Ellenberg drbd_al_begin_io(device, &peer_req->i); 260121ae5d7fSLars Ellenberg peer_req->flags |= EE_CALL_AL_COMPLETE_IO; 2602b411b363SPhilipp Reisner } 2603b411b363SPhilipp Reisner 2604bb3cc85eSMike Christie err = drbd_submit_peer_request(device, peer_req, op, op_flags, 2605bb3cc85eSMike Christie DRBD_FAULT_DT_WR); 260682bc0194SAndreas Gruenbacher if (!err) 260782bc0194SAndreas Gruenbacher return 0; 2608b411b363SPhilipp Reisner 260910f6d992SLars Ellenberg /* don't care for the reason here */ 2610d0180171SAndreas Gruenbacher drbd_err(device, "submit failed, triggering re-connect\n"); 26110500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 2612a8cd15baSAndreas Gruenbacher list_del(&peer_req->w.list); 2613b30ab791SAndreas Gruenbacher drbd_remove_epoch_entry_interval(device, peer_req); 26140500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 261521ae5d7fSLars Ellenberg if (peer_req->flags & EE_CALL_AL_COMPLETE_IO) { 261621ae5d7fSLars Ellenberg peer_req->flags &= ~EE_CALL_AL_COMPLETE_IO; 2617b30ab791SAndreas Gruenbacher drbd_al_complete_io(device, &peer_req->i); 261821ae5d7fSLars Ellenberg } 261922cc37a9SLars Ellenberg 2620b411b363SPhilipp Reisner out_interrupted: 26217e5fec31SFabian Frederick drbd_may_finish_epoch(connection, peer_req->epoch, EV_PUT | EV_CLEANUP); 2622b30ab791SAndreas Gruenbacher put_ldev(device); 2623b30ab791SAndreas Gruenbacher drbd_free_peer_req(device, peer_req); 262482bc0194SAndreas Gruenbacher return err; 2625b411b363SPhilipp Reisner } 2626b411b363SPhilipp Reisner 26270f0601f4SLars Ellenberg /* We may throttle resync, if the lower device seems to be busy, 26280f0601f4SLars Ellenberg * and current sync rate is above c_min_rate. 26290f0601f4SLars Ellenberg * 26300f0601f4SLars Ellenberg * To decide whether or not the lower device is busy, we use a scheme similar 26310f0601f4SLars Ellenberg * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" 26320f0601f4SLars Ellenberg * (more than 64 sectors) of activity we cannot account for with our own resync 26330f0601f4SLars Ellenberg * activity, it obviously is "busy". 26340f0601f4SLars Ellenberg * 26350f0601f4SLars Ellenberg * The current sync rate used here uses only the most recent two step marks, 26360f0601f4SLars Ellenberg * to have a short time average so we can react faster. 26370f0601f4SLars Ellenberg */ 2638ad3fee79SLars Ellenberg bool drbd_rs_should_slow_down(struct drbd_device *device, sector_t sector, 2639ad3fee79SLars Ellenberg bool throttle_if_app_is_waiting) 2640e8299874SLars Ellenberg { 2641e8299874SLars Ellenberg struct lc_element *tmp; 2642ad3fee79SLars Ellenberg bool throttle = drbd_rs_c_min_rate_throttle(device); 2643e8299874SLars Ellenberg 2644ad3fee79SLars Ellenberg if (!throttle || throttle_if_app_is_waiting) 2645ad3fee79SLars Ellenberg return throttle; 2646e8299874SLars Ellenberg 2647e8299874SLars Ellenberg spin_lock_irq(&device->al_lock); 2648e8299874SLars Ellenberg tmp = lc_find(device->resync, BM_SECT_TO_EXT(sector)); 2649e8299874SLars Ellenberg if (tmp) { 2650e8299874SLars Ellenberg struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); 2651e8299874SLars Ellenberg if (test_bit(BME_PRIORITY, &bm_ext->flags)) 2652e8299874SLars Ellenberg throttle = false; 2653ad3fee79SLars Ellenberg /* Do not slow down if app IO is already waiting for this extent, 2654ad3fee79SLars Ellenberg * and our progress is necessary for application IO to complete. */ 2655e8299874SLars Ellenberg } 2656e8299874SLars Ellenberg spin_unlock_irq(&device->al_lock); 2657e8299874SLars Ellenberg 2658e8299874SLars Ellenberg return throttle; 2659e8299874SLars Ellenberg } 2660e8299874SLars Ellenberg 2661e8299874SLars Ellenberg bool drbd_rs_c_min_rate_throttle(struct drbd_device *device) 26620f0601f4SLars Ellenberg { 2663b30ab791SAndreas Gruenbacher struct gendisk *disk = device->ldev->backing_bdev->bd_contains->bd_disk; 26640f0601f4SLars Ellenberg unsigned long db, dt, dbdt; 2665daeda1ccSPhilipp Reisner unsigned int c_min_rate; 2666e8299874SLars Ellenberg int curr_events; 2667daeda1ccSPhilipp Reisner 2668daeda1ccSPhilipp Reisner rcu_read_lock(); 2669b30ab791SAndreas Gruenbacher c_min_rate = rcu_dereference(device->ldev->disk_conf)->c_min_rate; 2670daeda1ccSPhilipp Reisner rcu_read_unlock(); 26710f0601f4SLars Ellenberg 26720f0601f4SLars Ellenberg /* feature disabled? */ 2673daeda1ccSPhilipp Reisner if (c_min_rate == 0) 2674e8299874SLars Ellenberg return false; 2675e3555d85SPhilipp Reisner 26760f0601f4SLars Ellenberg curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + 26770f0601f4SLars Ellenberg (int)part_stat_read(&disk->part0, sectors[1]) - 2678b30ab791SAndreas Gruenbacher atomic_read(&device->rs_sect_ev); 2679ad3fee79SLars Ellenberg 2680ad3fee79SLars Ellenberg if (atomic_read(&device->ap_actlog_cnt) 2681ff8bd88bSLars Ellenberg || curr_events - device->rs_last_events > 64) { 26820f0601f4SLars Ellenberg unsigned long rs_left; 26830f0601f4SLars Ellenberg int i; 26840f0601f4SLars Ellenberg 2685b30ab791SAndreas Gruenbacher device->rs_last_events = curr_events; 26860f0601f4SLars Ellenberg 26870f0601f4SLars Ellenberg /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, 26880f0601f4SLars Ellenberg * approx. */ 2689b30ab791SAndreas Gruenbacher i = (device->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; 26902649f080SLars Ellenberg 2691b30ab791SAndreas Gruenbacher if (device->state.conn == C_VERIFY_S || device->state.conn == C_VERIFY_T) 2692b30ab791SAndreas Gruenbacher rs_left = device->ov_left; 26932649f080SLars Ellenberg else 2694b30ab791SAndreas Gruenbacher rs_left = drbd_bm_total_weight(device) - device->rs_failed; 26950f0601f4SLars Ellenberg 2696b30ab791SAndreas Gruenbacher dt = ((long)jiffies - (long)device->rs_mark_time[i]) / HZ; 26970f0601f4SLars Ellenberg if (!dt) 26980f0601f4SLars Ellenberg dt++; 2699b30ab791SAndreas Gruenbacher db = device->rs_mark_left[i] - rs_left; 27000f0601f4SLars Ellenberg dbdt = Bit2KB(db/dt); 27010f0601f4SLars Ellenberg 2702daeda1ccSPhilipp Reisner if (dbdt > c_min_rate) 2703e8299874SLars Ellenberg return true; 27040f0601f4SLars Ellenberg } 2705e8299874SLars Ellenberg return false; 27060f0601f4SLars Ellenberg } 27070f0601f4SLars Ellenberg 2708bde89a9eSAndreas Gruenbacher static int receive_DataRequest(struct drbd_connection *connection, struct packet_info *pi) 2709b411b363SPhilipp Reisner { 27109f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 2711b30ab791SAndreas Gruenbacher struct drbd_device *device; 2712b411b363SPhilipp Reisner sector_t sector; 27134a76b161SAndreas Gruenbacher sector_t capacity; 2714db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req; 2715b411b363SPhilipp Reisner struct digest_info *di = NULL; 2716b18b37beSPhilipp Reisner int size, verb; 2717b411b363SPhilipp Reisner unsigned int fault_type; 2718e658983aSAndreas Gruenbacher struct p_block_req *p = pi->data; 27194a76b161SAndreas Gruenbacher 27209f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 27219f4fe9adSAndreas Gruenbacher if (!peer_device) 27224a76b161SAndreas Gruenbacher return -EIO; 27239f4fe9adSAndreas Gruenbacher device = peer_device->device; 2724b30ab791SAndreas Gruenbacher capacity = drbd_get_capacity(device->this_bdev); 2725b411b363SPhilipp Reisner 2726b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 2727b411b363SPhilipp Reisner size = be32_to_cpu(p->blksize); 2728b411b363SPhilipp Reisner 2729c670a398SAndreas Gruenbacher if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { 2730d0180171SAndreas Gruenbacher drbd_err(device, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, 2731b411b363SPhilipp Reisner (unsigned long long)sector, size); 273282bc0194SAndreas Gruenbacher return -EINVAL; 2733b411b363SPhilipp Reisner } 2734b411b363SPhilipp Reisner if (sector + (size>>9) > capacity) { 2735d0180171SAndreas Gruenbacher drbd_err(device, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, 2736b411b363SPhilipp Reisner (unsigned long long)sector, size); 273782bc0194SAndreas Gruenbacher return -EINVAL; 2738b411b363SPhilipp Reisner } 2739b411b363SPhilipp Reisner 2740b30ab791SAndreas Gruenbacher if (!get_ldev_if_state(device, D_UP_TO_DATE)) { 2741b18b37beSPhilipp Reisner verb = 1; 2742e2857216SAndreas Gruenbacher switch (pi->cmd) { 2743b18b37beSPhilipp Reisner case P_DATA_REQUEST: 274469a22773SAndreas Gruenbacher drbd_send_ack_rp(peer_device, P_NEG_DREPLY, p); 2745b18b37beSPhilipp Reisner break; 2746700ca8c0SPhilipp Reisner case P_RS_THIN_REQ: 2747b18b37beSPhilipp Reisner case P_RS_DATA_REQUEST: 2748b18b37beSPhilipp Reisner case P_CSUM_RS_REQUEST: 2749b18b37beSPhilipp Reisner case P_OV_REQUEST: 275069a22773SAndreas Gruenbacher drbd_send_ack_rp(peer_device, P_NEG_RS_DREPLY , p); 2751b18b37beSPhilipp Reisner break; 2752b18b37beSPhilipp Reisner case P_OV_REPLY: 2753b18b37beSPhilipp Reisner verb = 0; 2754b30ab791SAndreas Gruenbacher dec_rs_pending(device); 275569a22773SAndreas Gruenbacher drbd_send_ack_ex(peer_device, P_OV_RESULT, sector, size, ID_IN_SYNC); 2756b18b37beSPhilipp Reisner break; 2757b18b37beSPhilipp Reisner default: 275849ba9b1bSAndreas Gruenbacher BUG(); 2759b18b37beSPhilipp Reisner } 2760b18b37beSPhilipp Reisner if (verb && __ratelimit(&drbd_ratelimit_state)) 2761d0180171SAndreas Gruenbacher drbd_err(device, "Can not satisfy peer's read request, " 2762b411b363SPhilipp Reisner "no local data.\n"); 2763b18b37beSPhilipp Reisner 2764a821cc4aSLars Ellenberg /* drain possibly payload */ 276569a22773SAndreas Gruenbacher return drbd_drain_block(peer_device, pi->size); 2766b411b363SPhilipp Reisner } 2767b411b363SPhilipp Reisner 2768b411b363SPhilipp Reisner /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD 2769b411b363SPhilipp Reisner * "criss-cross" setup, that might cause write-out on some other DRBD, 2770b411b363SPhilipp Reisner * which in turn might block on the other node at this very place. */ 2771a0fb3c47SLars Ellenberg peer_req = drbd_alloc_peer_req(peer_device, p->block_id, sector, size, 27729104d31aSLars Ellenberg size, GFP_NOIO); 2773db830c46SAndreas Gruenbacher if (!peer_req) { 2774b30ab791SAndreas Gruenbacher put_ldev(device); 277582bc0194SAndreas Gruenbacher return -ENOMEM; 2776b411b363SPhilipp Reisner } 2777b411b363SPhilipp Reisner 2778e2857216SAndreas Gruenbacher switch (pi->cmd) { 2779b411b363SPhilipp Reisner case P_DATA_REQUEST: 2780a8cd15baSAndreas Gruenbacher peer_req->w.cb = w_e_end_data_req; 2781b411b363SPhilipp Reisner fault_type = DRBD_FAULT_DT_RD; 278280a40e43SLars Ellenberg /* application IO, don't drbd_rs_begin_io */ 278321ae5d7fSLars Ellenberg peer_req->flags |= EE_APPLICATION; 278480a40e43SLars Ellenberg goto submit; 278580a40e43SLars Ellenberg 2786700ca8c0SPhilipp Reisner case P_RS_THIN_REQ: 2787700ca8c0SPhilipp Reisner /* If at some point in the future we have a smart way to 2788700ca8c0SPhilipp Reisner find out if this data block is completely deallocated, 2789700ca8c0SPhilipp Reisner then we would do something smarter here than reading 2790700ca8c0SPhilipp Reisner the block... */ 2791700ca8c0SPhilipp Reisner peer_req->flags |= EE_RS_THIN_REQ; 2792b411b363SPhilipp Reisner case P_RS_DATA_REQUEST: 2793a8cd15baSAndreas Gruenbacher peer_req->w.cb = w_e_end_rsdata_req; 2794b411b363SPhilipp Reisner fault_type = DRBD_FAULT_RS_RD; 27955f9915bbSLars Ellenberg /* used in the sector offset progress display */ 2796b30ab791SAndreas Gruenbacher device->bm_resync_fo = BM_SECT_TO_BIT(sector); 2797b411b363SPhilipp Reisner break; 2798b411b363SPhilipp Reisner 2799b411b363SPhilipp Reisner case P_OV_REPLY: 2800b411b363SPhilipp Reisner case P_CSUM_RS_REQUEST: 2801b411b363SPhilipp Reisner fault_type = DRBD_FAULT_RS_RD; 2802e2857216SAndreas Gruenbacher di = kmalloc(sizeof(*di) + pi->size, GFP_NOIO); 2803b411b363SPhilipp Reisner if (!di) 2804b411b363SPhilipp Reisner goto out_free_e; 2805b411b363SPhilipp Reisner 2806e2857216SAndreas Gruenbacher di->digest_size = pi->size; 2807b411b363SPhilipp Reisner di->digest = (((char *)di)+sizeof(struct digest_info)); 2808b411b363SPhilipp Reisner 2809db830c46SAndreas Gruenbacher peer_req->digest = di; 2810db830c46SAndreas Gruenbacher peer_req->flags |= EE_HAS_DIGEST; 2811c36c3cedSLars Ellenberg 28129f4fe9adSAndreas Gruenbacher if (drbd_recv_all(peer_device->connection, di->digest, pi->size)) 2813b411b363SPhilipp Reisner goto out_free_e; 2814b411b363SPhilipp Reisner 2815e2857216SAndreas Gruenbacher if (pi->cmd == P_CSUM_RS_REQUEST) { 28169f4fe9adSAndreas Gruenbacher D_ASSERT(device, peer_device->connection->agreed_pro_version >= 89); 2817a8cd15baSAndreas Gruenbacher peer_req->w.cb = w_e_end_csum_rs_req; 28185f9915bbSLars Ellenberg /* used in the sector offset progress display */ 2819b30ab791SAndreas Gruenbacher device->bm_resync_fo = BM_SECT_TO_BIT(sector); 2820aaaba345SLars Ellenberg /* remember to report stats in drbd_resync_finished */ 2821aaaba345SLars Ellenberg device->use_csums = true; 2822e2857216SAndreas Gruenbacher } else if (pi->cmd == P_OV_REPLY) { 28232649f080SLars Ellenberg /* track progress, we may need to throttle */ 2824b30ab791SAndreas Gruenbacher atomic_add(size >> 9, &device->rs_sect_in); 2825a8cd15baSAndreas Gruenbacher peer_req->w.cb = w_e_end_ov_reply; 2826b30ab791SAndreas Gruenbacher dec_rs_pending(device); 28270f0601f4SLars Ellenberg /* drbd_rs_begin_io done when we sent this request, 28280f0601f4SLars Ellenberg * but accounting still needs to be done. */ 28290f0601f4SLars Ellenberg goto submit_for_resync; 2830b411b363SPhilipp Reisner } 2831b411b363SPhilipp Reisner break; 2832b411b363SPhilipp Reisner 2833b411b363SPhilipp Reisner case P_OV_REQUEST: 2834b30ab791SAndreas Gruenbacher if (device->ov_start_sector == ~(sector_t)0 && 28359f4fe9adSAndreas Gruenbacher peer_device->connection->agreed_pro_version >= 90) { 2836de228bbaSLars Ellenberg unsigned long now = jiffies; 2837de228bbaSLars Ellenberg int i; 2838b30ab791SAndreas Gruenbacher device->ov_start_sector = sector; 2839b30ab791SAndreas Gruenbacher device->ov_position = sector; 2840b30ab791SAndreas Gruenbacher device->ov_left = drbd_bm_bits(device) - BM_SECT_TO_BIT(sector); 2841b30ab791SAndreas Gruenbacher device->rs_total = device->ov_left; 2842de228bbaSLars Ellenberg for (i = 0; i < DRBD_SYNC_MARKS; i++) { 2843b30ab791SAndreas Gruenbacher device->rs_mark_left[i] = device->ov_left; 2844b30ab791SAndreas Gruenbacher device->rs_mark_time[i] = now; 2845de228bbaSLars Ellenberg } 2846d0180171SAndreas Gruenbacher drbd_info(device, "Online Verify start sector: %llu\n", 2847b411b363SPhilipp Reisner (unsigned long long)sector); 2848b411b363SPhilipp Reisner } 2849a8cd15baSAndreas Gruenbacher peer_req->w.cb = w_e_end_ov_req; 2850b411b363SPhilipp Reisner fault_type = DRBD_FAULT_RS_RD; 2851b411b363SPhilipp Reisner break; 2852b411b363SPhilipp Reisner 2853b411b363SPhilipp Reisner default: 285449ba9b1bSAndreas Gruenbacher BUG(); 2855b411b363SPhilipp Reisner } 2856b411b363SPhilipp Reisner 28570f0601f4SLars Ellenberg /* Throttle, drbd_rs_begin_io and submit should become asynchronous 28580f0601f4SLars Ellenberg * wrt the receiver, but it is not as straightforward as it may seem. 28590f0601f4SLars Ellenberg * Various places in the resync start and stop logic assume resync 28600f0601f4SLars Ellenberg * requests are processed in order, requeuing this on the worker thread 28610f0601f4SLars Ellenberg * introduces a bunch of new code for synchronization between threads. 28620f0601f4SLars Ellenberg * 28630f0601f4SLars Ellenberg * Unlimited throttling before drbd_rs_begin_io may stall the resync 28640f0601f4SLars Ellenberg * "forever", throttling after drbd_rs_begin_io will lock that extent 28650f0601f4SLars Ellenberg * for application writes for the same time. For now, just throttle 28660f0601f4SLars Ellenberg * here, where the rest of the code expects the receiver to sleep for 28670f0601f4SLars Ellenberg * a while, anyways. 28680f0601f4SLars Ellenberg */ 2869b411b363SPhilipp Reisner 28700f0601f4SLars Ellenberg /* Throttle before drbd_rs_begin_io, as that locks out application IO; 28710f0601f4SLars Ellenberg * this defers syncer requests for some time, before letting at least 28720f0601f4SLars Ellenberg * on request through. The resync controller on the receiving side 28730f0601f4SLars Ellenberg * will adapt to the incoming rate accordingly. 28740f0601f4SLars Ellenberg * 28750f0601f4SLars Ellenberg * We cannot throttle here if remote is Primary/SyncTarget: 28760f0601f4SLars Ellenberg * we would also throttle its application reads. 28770f0601f4SLars Ellenberg * In that case, throttling is done on the SyncTarget only. 28780f0601f4SLars Ellenberg */ 2879c5a2c150SLars Ellenberg 2880c5a2c150SLars Ellenberg /* Even though this may be a resync request, we do add to "read_ee"; 2881c5a2c150SLars Ellenberg * "sync_ee" is only used for resync WRITEs. 2882c5a2c150SLars Ellenberg * Add to list early, so debugfs can find this request 2883c5a2c150SLars Ellenberg * even if we have to sleep below. */ 2884c5a2c150SLars Ellenberg spin_lock_irq(&device->resource->req_lock); 2885c5a2c150SLars Ellenberg list_add_tail(&peer_req->w.list, &device->read_ee); 2886c5a2c150SLars Ellenberg spin_unlock_irq(&device->resource->req_lock); 2887c5a2c150SLars Ellenberg 2888944410e9SLars Ellenberg update_receiver_timing_details(connection, drbd_rs_should_slow_down); 2889ad3fee79SLars Ellenberg if (device->state.peer != R_PRIMARY 2890ad3fee79SLars Ellenberg && drbd_rs_should_slow_down(device, sector, false)) 2891e3555d85SPhilipp Reisner schedule_timeout_uninterruptible(HZ/10); 2892944410e9SLars Ellenberg update_receiver_timing_details(connection, drbd_rs_begin_io); 2893b30ab791SAndreas Gruenbacher if (drbd_rs_begin_io(device, sector)) 289480a40e43SLars Ellenberg goto out_free_e; 2895b411b363SPhilipp Reisner 28960f0601f4SLars Ellenberg submit_for_resync: 2897b30ab791SAndreas Gruenbacher atomic_add(size >> 9, &device->rs_sect_ev); 28980f0601f4SLars Ellenberg 289980a40e43SLars Ellenberg submit: 2900944410e9SLars Ellenberg update_receiver_timing_details(connection, drbd_submit_peer_request); 2901b30ab791SAndreas Gruenbacher inc_unacked(device); 2902bb3cc85eSMike Christie if (drbd_submit_peer_request(device, peer_req, REQ_OP_READ, 0, 2903bb3cc85eSMike Christie fault_type) == 0) 290482bc0194SAndreas Gruenbacher return 0; 2905b411b363SPhilipp Reisner 290610f6d992SLars Ellenberg /* don't care for the reason here */ 2907d0180171SAndreas Gruenbacher drbd_err(device, "submit failed, triggering re-connect\n"); 2908c5a2c150SLars Ellenberg 2909c5a2c150SLars Ellenberg out_free_e: 29100500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 2911a8cd15baSAndreas Gruenbacher list_del(&peer_req->w.list); 29120500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 291322cc37a9SLars Ellenberg /* no drbd_rs_complete_io(), we are dropping the connection anyways */ 291422cc37a9SLars Ellenberg 2915b30ab791SAndreas Gruenbacher put_ldev(device); 2916b30ab791SAndreas Gruenbacher drbd_free_peer_req(device, peer_req); 291782bc0194SAndreas Gruenbacher return -EIO; 2918b411b363SPhilipp Reisner } 2919b411b363SPhilipp Reisner 292069a22773SAndreas Gruenbacher /** 292169a22773SAndreas Gruenbacher * drbd_asb_recover_0p - Recover after split-brain with no remaining primaries 292269a22773SAndreas Gruenbacher */ 292369a22773SAndreas Gruenbacher static int drbd_asb_recover_0p(struct drbd_peer_device *peer_device) __must_hold(local) 2924b411b363SPhilipp Reisner { 292569a22773SAndreas Gruenbacher struct drbd_device *device = peer_device->device; 2926b411b363SPhilipp Reisner int self, peer, rv = -100; 2927b411b363SPhilipp Reisner unsigned long ch_self, ch_peer; 292844ed167dSPhilipp Reisner enum drbd_after_sb_p after_sb_0p; 2929b411b363SPhilipp Reisner 2930b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[UI_BITMAP] & 1; 2931b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_BITMAP] & 1; 2932b411b363SPhilipp Reisner 2933b30ab791SAndreas Gruenbacher ch_peer = device->p_uuid[UI_SIZE]; 2934b30ab791SAndreas Gruenbacher ch_self = device->comm_bm_set; 2935b411b363SPhilipp Reisner 293644ed167dSPhilipp Reisner rcu_read_lock(); 293769a22773SAndreas Gruenbacher after_sb_0p = rcu_dereference(peer_device->connection->net_conf)->after_sb_0p; 293844ed167dSPhilipp Reisner rcu_read_unlock(); 293944ed167dSPhilipp Reisner switch (after_sb_0p) { 2940b411b363SPhilipp Reisner case ASB_CONSENSUS: 2941b411b363SPhilipp Reisner case ASB_DISCARD_SECONDARY: 2942b411b363SPhilipp Reisner case ASB_CALL_HELPER: 294344ed167dSPhilipp Reisner case ASB_VIOLENTLY: 2944d0180171SAndreas Gruenbacher drbd_err(device, "Configuration error.\n"); 2945b411b363SPhilipp Reisner break; 2946b411b363SPhilipp Reisner case ASB_DISCONNECT: 2947b411b363SPhilipp Reisner break; 2948b411b363SPhilipp Reisner case ASB_DISCARD_YOUNGER_PRI: 2949b411b363SPhilipp Reisner if (self == 0 && peer == 1) { 2950b411b363SPhilipp Reisner rv = -1; 2951b411b363SPhilipp Reisner break; 2952b411b363SPhilipp Reisner } 2953b411b363SPhilipp Reisner if (self == 1 && peer == 0) { 2954b411b363SPhilipp Reisner rv = 1; 2955b411b363SPhilipp Reisner break; 2956b411b363SPhilipp Reisner } 2957b411b363SPhilipp Reisner /* Else fall through to one of the other strategies... */ 2958b411b363SPhilipp Reisner case ASB_DISCARD_OLDER_PRI: 2959b411b363SPhilipp Reisner if (self == 0 && peer == 1) { 2960b411b363SPhilipp Reisner rv = 1; 2961b411b363SPhilipp Reisner break; 2962b411b363SPhilipp Reisner } 2963b411b363SPhilipp Reisner if (self == 1 && peer == 0) { 2964b411b363SPhilipp Reisner rv = -1; 2965b411b363SPhilipp Reisner break; 2966b411b363SPhilipp Reisner } 2967b411b363SPhilipp Reisner /* Else fall through to one of the other strategies... */ 2968d0180171SAndreas Gruenbacher drbd_warn(device, "Discard younger/older primary did not find a decision\n" 2969b411b363SPhilipp Reisner "Using discard-least-changes instead\n"); 2970b411b363SPhilipp Reisner case ASB_DISCARD_ZERO_CHG: 2971b411b363SPhilipp Reisner if (ch_peer == 0 && ch_self == 0) { 297269a22773SAndreas Gruenbacher rv = test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags) 2973b411b363SPhilipp Reisner ? -1 : 1; 2974b411b363SPhilipp Reisner break; 2975b411b363SPhilipp Reisner } else { 2976b411b363SPhilipp Reisner if (ch_peer == 0) { rv = 1; break; } 2977b411b363SPhilipp Reisner if (ch_self == 0) { rv = -1; break; } 2978b411b363SPhilipp Reisner } 297944ed167dSPhilipp Reisner if (after_sb_0p == ASB_DISCARD_ZERO_CHG) 2980b411b363SPhilipp Reisner break; 2981b411b363SPhilipp Reisner case ASB_DISCARD_LEAST_CHG: 2982b411b363SPhilipp Reisner if (ch_self < ch_peer) 2983b411b363SPhilipp Reisner rv = -1; 2984b411b363SPhilipp Reisner else if (ch_self > ch_peer) 2985b411b363SPhilipp Reisner rv = 1; 2986b411b363SPhilipp Reisner else /* ( ch_self == ch_peer ) */ 2987b411b363SPhilipp Reisner /* Well, then use something else. */ 298869a22773SAndreas Gruenbacher rv = test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags) 2989b411b363SPhilipp Reisner ? -1 : 1; 2990b411b363SPhilipp Reisner break; 2991b411b363SPhilipp Reisner case ASB_DISCARD_LOCAL: 2992b411b363SPhilipp Reisner rv = -1; 2993b411b363SPhilipp Reisner break; 2994b411b363SPhilipp Reisner case ASB_DISCARD_REMOTE: 2995b411b363SPhilipp Reisner rv = 1; 2996b411b363SPhilipp Reisner } 2997b411b363SPhilipp Reisner 2998b411b363SPhilipp Reisner return rv; 2999b411b363SPhilipp Reisner } 3000b411b363SPhilipp Reisner 300169a22773SAndreas Gruenbacher /** 300269a22773SAndreas Gruenbacher * drbd_asb_recover_1p - Recover after split-brain with one remaining primary 300369a22773SAndreas Gruenbacher */ 300469a22773SAndreas Gruenbacher static int drbd_asb_recover_1p(struct drbd_peer_device *peer_device) __must_hold(local) 3005b411b363SPhilipp Reisner { 300669a22773SAndreas Gruenbacher struct drbd_device *device = peer_device->device; 30076184ea21SAndreas Gruenbacher int hg, rv = -100; 300844ed167dSPhilipp Reisner enum drbd_after_sb_p after_sb_1p; 3009b411b363SPhilipp Reisner 301044ed167dSPhilipp Reisner rcu_read_lock(); 301169a22773SAndreas Gruenbacher after_sb_1p = rcu_dereference(peer_device->connection->net_conf)->after_sb_1p; 301244ed167dSPhilipp Reisner rcu_read_unlock(); 301344ed167dSPhilipp Reisner switch (after_sb_1p) { 3014b411b363SPhilipp Reisner case ASB_DISCARD_YOUNGER_PRI: 3015b411b363SPhilipp Reisner case ASB_DISCARD_OLDER_PRI: 3016b411b363SPhilipp Reisner case ASB_DISCARD_LEAST_CHG: 3017b411b363SPhilipp Reisner case ASB_DISCARD_LOCAL: 3018b411b363SPhilipp Reisner case ASB_DISCARD_REMOTE: 301944ed167dSPhilipp Reisner case ASB_DISCARD_ZERO_CHG: 3020d0180171SAndreas Gruenbacher drbd_err(device, "Configuration error.\n"); 3021b411b363SPhilipp Reisner break; 3022b411b363SPhilipp Reisner case ASB_DISCONNECT: 3023b411b363SPhilipp Reisner break; 3024b411b363SPhilipp Reisner case ASB_CONSENSUS: 302569a22773SAndreas Gruenbacher hg = drbd_asb_recover_0p(peer_device); 3026b30ab791SAndreas Gruenbacher if (hg == -1 && device->state.role == R_SECONDARY) 3027b411b363SPhilipp Reisner rv = hg; 3028b30ab791SAndreas Gruenbacher if (hg == 1 && device->state.role == R_PRIMARY) 3029b411b363SPhilipp Reisner rv = hg; 3030b411b363SPhilipp Reisner break; 3031b411b363SPhilipp Reisner case ASB_VIOLENTLY: 303269a22773SAndreas Gruenbacher rv = drbd_asb_recover_0p(peer_device); 3033b411b363SPhilipp Reisner break; 3034b411b363SPhilipp Reisner case ASB_DISCARD_SECONDARY: 3035b30ab791SAndreas Gruenbacher return device->state.role == R_PRIMARY ? 1 : -1; 3036b411b363SPhilipp Reisner case ASB_CALL_HELPER: 303769a22773SAndreas Gruenbacher hg = drbd_asb_recover_0p(peer_device); 3038b30ab791SAndreas Gruenbacher if (hg == -1 && device->state.role == R_PRIMARY) { 3039bb437946SAndreas Gruenbacher enum drbd_state_rv rv2; 3040bb437946SAndreas Gruenbacher 3041b411b363SPhilipp Reisner /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, 3042b411b363SPhilipp Reisner * we might be here in C_WF_REPORT_PARAMS which is transient. 3043b411b363SPhilipp Reisner * we do not need to wait for the after state change work either. */ 3044b30ab791SAndreas Gruenbacher rv2 = drbd_change_state(device, CS_VERBOSE, NS(role, R_SECONDARY)); 3045bb437946SAndreas Gruenbacher if (rv2 != SS_SUCCESS) { 3046b30ab791SAndreas Gruenbacher drbd_khelper(device, "pri-lost-after-sb"); 3047b411b363SPhilipp Reisner } else { 3048d0180171SAndreas Gruenbacher drbd_warn(device, "Successfully gave up primary role.\n"); 3049b411b363SPhilipp Reisner rv = hg; 3050b411b363SPhilipp Reisner } 3051b411b363SPhilipp Reisner } else 3052b411b363SPhilipp Reisner rv = hg; 3053b411b363SPhilipp Reisner } 3054b411b363SPhilipp Reisner 3055b411b363SPhilipp Reisner return rv; 3056b411b363SPhilipp Reisner } 3057b411b363SPhilipp Reisner 305869a22773SAndreas Gruenbacher /** 305969a22773SAndreas Gruenbacher * drbd_asb_recover_2p - Recover after split-brain with two remaining primaries 306069a22773SAndreas Gruenbacher */ 306169a22773SAndreas Gruenbacher static int drbd_asb_recover_2p(struct drbd_peer_device *peer_device) __must_hold(local) 3062b411b363SPhilipp Reisner { 306369a22773SAndreas Gruenbacher struct drbd_device *device = peer_device->device; 30646184ea21SAndreas Gruenbacher int hg, rv = -100; 306544ed167dSPhilipp Reisner enum drbd_after_sb_p after_sb_2p; 3066b411b363SPhilipp Reisner 306744ed167dSPhilipp Reisner rcu_read_lock(); 306869a22773SAndreas Gruenbacher after_sb_2p = rcu_dereference(peer_device->connection->net_conf)->after_sb_2p; 306944ed167dSPhilipp Reisner rcu_read_unlock(); 307044ed167dSPhilipp Reisner switch (after_sb_2p) { 3071b411b363SPhilipp Reisner case ASB_DISCARD_YOUNGER_PRI: 3072b411b363SPhilipp Reisner case ASB_DISCARD_OLDER_PRI: 3073b411b363SPhilipp Reisner case ASB_DISCARD_LEAST_CHG: 3074b411b363SPhilipp Reisner case ASB_DISCARD_LOCAL: 3075b411b363SPhilipp Reisner case ASB_DISCARD_REMOTE: 3076b411b363SPhilipp Reisner case ASB_CONSENSUS: 3077b411b363SPhilipp Reisner case ASB_DISCARD_SECONDARY: 307844ed167dSPhilipp Reisner case ASB_DISCARD_ZERO_CHG: 3079d0180171SAndreas Gruenbacher drbd_err(device, "Configuration error.\n"); 3080b411b363SPhilipp Reisner break; 3081b411b363SPhilipp Reisner case ASB_VIOLENTLY: 308269a22773SAndreas Gruenbacher rv = drbd_asb_recover_0p(peer_device); 3083b411b363SPhilipp Reisner break; 3084b411b363SPhilipp Reisner case ASB_DISCONNECT: 3085b411b363SPhilipp Reisner break; 3086b411b363SPhilipp Reisner case ASB_CALL_HELPER: 308769a22773SAndreas Gruenbacher hg = drbd_asb_recover_0p(peer_device); 3088b411b363SPhilipp Reisner if (hg == -1) { 3089bb437946SAndreas Gruenbacher enum drbd_state_rv rv2; 3090bb437946SAndreas Gruenbacher 3091b411b363SPhilipp Reisner /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, 3092b411b363SPhilipp Reisner * we might be here in C_WF_REPORT_PARAMS which is transient. 3093b411b363SPhilipp Reisner * we do not need to wait for the after state change work either. */ 3094b30ab791SAndreas Gruenbacher rv2 = drbd_change_state(device, CS_VERBOSE, NS(role, R_SECONDARY)); 3095bb437946SAndreas Gruenbacher if (rv2 != SS_SUCCESS) { 3096b30ab791SAndreas Gruenbacher drbd_khelper(device, "pri-lost-after-sb"); 3097b411b363SPhilipp Reisner } else { 3098d0180171SAndreas Gruenbacher drbd_warn(device, "Successfully gave up primary role.\n"); 3099b411b363SPhilipp Reisner rv = hg; 3100b411b363SPhilipp Reisner } 3101b411b363SPhilipp Reisner } else 3102b411b363SPhilipp Reisner rv = hg; 3103b411b363SPhilipp Reisner } 3104b411b363SPhilipp Reisner 3105b411b363SPhilipp Reisner return rv; 3106b411b363SPhilipp Reisner } 3107b411b363SPhilipp Reisner 3108b30ab791SAndreas Gruenbacher static void drbd_uuid_dump(struct drbd_device *device, char *text, u64 *uuid, 3109b411b363SPhilipp Reisner u64 bits, u64 flags) 3110b411b363SPhilipp Reisner { 3111b411b363SPhilipp Reisner if (!uuid) { 3112d0180171SAndreas Gruenbacher drbd_info(device, "%s uuid info vanished while I was looking!\n", text); 3113b411b363SPhilipp Reisner return; 3114b411b363SPhilipp Reisner } 3115d0180171SAndreas Gruenbacher drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", 3116b411b363SPhilipp Reisner text, 3117b411b363SPhilipp Reisner (unsigned long long)uuid[UI_CURRENT], 3118b411b363SPhilipp Reisner (unsigned long long)uuid[UI_BITMAP], 3119b411b363SPhilipp Reisner (unsigned long long)uuid[UI_HISTORY_START], 3120b411b363SPhilipp Reisner (unsigned long long)uuid[UI_HISTORY_END], 3121b411b363SPhilipp Reisner (unsigned long long)bits, 3122b411b363SPhilipp Reisner (unsigned long long)flags); 3123b411b363SPhilipp Reisner } 3124b411b363SPhilipp Reisner 3125b411b363SPhilipp Reisner /* 3126b411b363SPhilipp Reisner 100 after split brain try auto recover 3127b411b363SPhilipp Reisner 2 C_SYNC_SOURCE set BitMap 3128b411b363SPhilipp Reisner 1 C_SYNC_SOURCE use BitMap 3129b411b363SPhilipp Reisner 0 no Sync 3130b411b363SPhilipp Reisner -1 C_SYNC_TARGET use BitMap 3131b411b363SPhilipp Reisner -2 C_SYNC_TARGET set BitMap 3132b411b363SPhilipp Reisner -100 after split brain, disconnect 3133b411b363SPhilipp Reisner -1000 unrelated data 31344a23f264SPhilipp Reisner -1091 requires proto 91 31354a23f264SPhilipp Reisner -1096 requires proto 96 3136b411b363SPhilipp Reisner */ 3137f2d3d75bSLars Ellenberg 3138f2d3d75bSLars Ellenberg static int drbd_uuid_compare(struct drbd_device *const device, enum drbd_role const peer_role, int *rule_nr) __must_hold(local) 3139b411b363SPhilipp Reisner { 314044a4d551SLars Ellenberg struct drbd_peer_device *const peer_device = first_peer_device(device); 314144a4d551SLars Ellenberg struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL; 3142b411b363SPhilipp Reisner u64 self, peer; 3143b411b363SPhilipp Reisner int i, j; 3144b411b363SPhilipp Reisner 3145b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[UI_CURRENT] & ~((u64)1); 3146b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_CURRENT] & ~((u64)1); 3147b411b363SPhilipp Reisner 3148b411b363SPhilipp Reisner *rule_nr = 10; 3149b411b363SPhilipp Reisner if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) 3150b411b363SPhilipp Reisner return 0; 3151b411b363SPhilipp Reisner 3152b411b363SPhilipp Reisner *rule_nr = 20; 3153b411b363SPhilipp Reisner if ((self == UUID_JUST_CREATED || self == (u64)0) && 3154b411b363SPhilipp Reisner peer != UUID_JUST_CREATED) 3155b411b363SPhilipp Reisner return -2; 3156b411b363SPhilipp Reisner 3157b411b363SPhilipp Reisner *rule_nr = 30; 3158b411b363SPhilipp Reisner if (self != UUID_JUST_CREATED && 3159b411b363SPhilipp Reisner (peer == UUID_JUST_CREATED || peer == (u64)0)) 3160b411b363SPhilipp Reisner return 2; 3161b411b363SPhilipp Reisner 3162b411b363SPhilipp Reisner if (self == peer) { 3163b411b363SPhilipp Reisner int rct, dc; /* roles at crash time */ 3164b411b363SPhilipp Reisner 3165b30ab791SAndreas Gruenbacher if (device->p_uuid[UI_BITMAP] == (u64)0 && device->ldev->md.uuid[UI_BITMAP] != (u64)0) { 3166b411b363SPhilipp Reisner 316744a4d551SLars Ellenberg if (connection->agreed_pro_version < 91) 31684a23f264SPhilipp Reisner return -1091; 3169b411b363SPhilipp Reisner 3170b30ab791SAndreas Gruenbacher if ((device->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START] & ~((u64)1)) && 3171b30ab791SAndreas Gruenbacher (device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { 3172d0180171SAndreas Gruenbacher drbd_info(device, "was SyncSource, missed the resync finished event, corrected myself:\n"); 3173b30ab791SAndreas Gruenbacher drbd_uuid_move_history(device); 3174b30ab791SAndreas Gruenbacher device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP]; 3175b30ab791SAndreas Gruenbacher device->ldev->md.uuid[UI_BITMAP] = 0; 3176b411b363SPhilipp Reisner 3177b30ab791SAndreas Gruenbacher drbd_uuid_dump(device, "self", device->ldev->md.uuid, 3178b30ab791SAndreas Gruenbacher device->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(device) : 0, 0); 3179b411b363SPhilipp Reisner *rule_nr = 34; 3180b411b363SPhilipp Reisner } else { 3181d0180171SAndreas Gruenbacher drbd_info(device, "was SyncSource (peer failed to write sync_uuid)\n"); 3182b411b363SPhilipp Reisner *rule_nr = 36; 3183b411b363SPhilipp Reisner } 3184b411b363SPhilipp Reisner 3185b411b363SPhilipp Reisner return 1; 3186b411b363SPhilipp Reisner } 3187b411b363SPhilipp Reisner 3188b30ab791SAndreas Gruenbacher if (device->ldev->md.uuid[UI_BITMAP] == (u64)0 && device->p_uuid[UI_BITMAP] != (u64)0) { 3189b411b363SPhilipp Reisner 319044a4d551SLars Ellenberg if (connection->agreed_pro_version < 91) 31914a23f264SPhilipp Reisner return -1091; 3192b411b363SPhilipp Reisner 3193b30ab791SAndreas Gruenbacher if ((device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (device->p_uuid[UI_BITMAP] & ~((u64)1)) && 3194b30ab791SAndreas Gruenbacher (device->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START] & ~((u64)1))) { 3195d0180171SAndreas Gruenbacher drbd_info(device, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); 3196b411b363SPhilipp Reisner 3197b30ab791SAndreas Gruenbacher device->p_uuid[UI_HISTORY_START + 1] = device->p_uuid[UI_HISTORY_START]; 3198b30ab791SAndreas Gruenbacher device->p_uuid[UI_HISTORY_START] = device->p_uuid[UI_BITMAP]; 3199b30ab791SAndreas Gruenbacher device->p_uuid[UI_BITMAP] = 0UL; 3200b411b363SPhilipp Reisner 3201b30ab791SAndreas Gruenbacher drbd_uuid_dump(device, "peer", device->p_uuid, device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); 3202b411b363SPhilipp Reisner *rule_nr = 35; 3203b411b363SPhilipp Reisner } else { 3204d0180171SAndreas Gruenbacher drbd_info(device, "was SyncTarget (failed to write sync_uuid)\n"); 3205b411b363SPhilipp Reisner *rule_nr = 37; 3206b411b363SPhilipp Reisner } 3207b411b363SPhilipp Reisner 3208b411b363SPhilipp Reisner return -1; 3209b411b363SPhilipp Reisner } 3210b411b363SPhilipp Reisner 3211b411b363SPhilipp Reisner /* Common power [off|failure] */ 3212b30ab791SAndreas Gruenbacher rct = (test_bit(CRASHED_PRIMARY, &device->flags) ? 1 : 0) + 3213b30ab791SAndreas Gruenbacher (device->p_uuid[UI_FLAGS] & 2); 3214b411b363SPhilipp Reisner /* lowest bit is set when we were primary, 3215b411b363SPhilipp Reisner * next bit (weight 2) is set when peer was primary */ 3216b411b363SPhilipp Reisner *rule_nr = 40; 3217b411b363SPhilipp Reisner 3218f2d3d75bSLars Ellenberg /* Neither has the "crashed primary" flag set, 3219f2d3d75bSLars Ellenberg * only a replication link hickup. */ 3220f2d3d75bSLars Ellenberg if (rct == 0) 3221f2d3d75bSLars Ellenberg return 0; 3222f2d3d75bSLars Ellenberg 3223f2d3d75bSLars Ellenberg /* Current UUID equal and no bitmap uuid; does not necessarily 3224f2d3d75bSLars Ellenberg * mean this was a "simultaneous hard crash", maybe IO was 3225f2d3d75bSLars Ellenberg * frozen, so no UUID-bump happened. 3226f2d3d75bSLars Ellenberg * This is a protocol change, overload DRBD_FF_WSAME as flag 3227f2d3d75bSLars Ellenberg * for "new-enough" peer DRBD version. */ 3228f2d3d75bSLars Ellenberg if (device->state.role == R_PRIMARY || peer_role == R_PRIMARY) { 3229f2d3d75bSLars Ellenberg *rule_nr = 41; 3230f2d3d75bSLars Ellenberg if (!(connection->agreed_features & DRBD_FF_WSAME)) { 3231f2d3d75bSLars Ellenberg drbd_warn(peer_device, "Equivalent unrotated UUIDs, but current primary present.\n"); 3232f2d3d75bSLars Ellenberg return -(0x10000 | PRO_VERSION_MAX | (DRBD_FF_WSAME << 8)); 3233f2d3d75bSLars Ellenberg } 3234f2d3d75bSLars Ellenberg if (device->state.role == R_PRIMARY && peer_role == R_PRIMARY) { 3235f2d3d75bSLars Ellenberg /* At least one has the "crashed primary" bit set, 3236f2d3d75bSLars Ellenberg * both are primary now, but neither has rotated its UUIDs? 3237f2d3d75bSLars Ellenberg * "Can not happen." */ 3238f2d3d75bSLars Ellenberg drbd_err(peer_device, "Equivalent unrotated UUIDs, but both are primary. Can not resolve this.\n"); 3239f2d3d75bSLars Ellenberg return -100; 3240f2d3d75bSLars Ellenberg } 3241f2d3d75bSLars Ellenberg if (device->state.role == R_PRIMARY) 3242f2d3d75bSLars Ellenberg return 1; 3243f2d3d75bSLars Ellenberg return -1; 3244f2d3d75bSLars Ellenberg } 3245f2d3d75bSLars Ellenberg 3246f2d3d75bSLars Ellenberg /* Both are secondary. 3247f2d3d75bSLars Ellenberg * Really looks like recovery from simultaneous hard crash. 3248f2d3d75bSLars Ellenberg * Check which had been primary before, and arbitrate. */ 3249b411b363SPhilipp Reisner switch (rct) { 3250f2d3d75bSLars Ellenberg case 0: /* !self_pri && !peer_pri */ return 0; /* already handled */ 3251b411b363SPhilipp Reisner case 1: /* self_pri && !peer_pri */ return 1; 3252b411b363SPhilipp Reisner case 2: /* !self_pri && peer_pri */ return -1; 3253b411b363SPhilipp Reisner case 3: /* self_pri && peer_pri */ 325444a4d551SLars Ellenberg dc = test_bit(RESOLVE_CONFLICTS, &connection->flags); 3255b411b363SPhilipp Reisner return dc ? -1 : 1; 3256b411b363SPhilipp Reisner } 3257b411b363SPhilipp Reisner } 3258b411b363SPhilipp Reisner 3259b411b363SPhilipp Reisner *rule_nr = 50; 3260b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_BITMAP] & ~((u64)1); 3261b411b363SPhilipp Reisner if (self == peer) 3262b411b363SPhilipp Reisner return -1; 3263b411b363SPhilipp Reisner 3264b411b363SPhilipp Reisner *rule_nr = 51; 3265b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_HISTORY_START] & ~((u64)1); 3266b411b363SPhilipp Reisner if (self == peer) { 326744a4d551SLars Ellenberg if (connection->agreed_pro_version < 96 ? 3268b30ab791SAndreas Gruenbacher (device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == 3269b30ab791SAndreas Gruenbacher (device->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : 3270b30ab791SAndreas Gruenbacher peer + UUID_NEW_BM_OFFSET == (device->p_uuid[UI_BITMAP] & ~((u64)1))) { 3271b411b363SPhilipp Reisner /* The last P_SYNC_UUID did not get though. Undo the last start of 3272b411b363SPhilipp Reisner resync as sync source modifications of the peer's UUIDs. */ 3273b411b363SPhilipp Reisner 327444a4d551SLars Ellenberg if (connection->agreed_pro_version < 91) 32754a23f264SPhilipp Reisner return -1091; 3276b411b363SPhilipp Reisner 3277b30ab791SAndreas Gruenbacher device->p_uuid[UI_BITMAP] = device->p_uuid[UI_HISTORY_START]; 3278b30ab791SAndreas Gruenbacher device->p_uuid[UI_HISTORY_START] = device->p_uuid[UI_HISTORY_START + 1]; 32794a23f264SPhilipp Reisner 3280d0180171SAndreas Gruenbacher drbd_info(device, "Lost last syncUUID packet, corrected:\n"); 3281b30ab791SAndreas Gruenbacher drbd_uuid_dump(device, "peer", device->p_uuid, device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); 32824a23f264SPhilipp Reisner 3283b411b363SPhilipp Reisner return -1; 3284b411b363SPhilipp Reisner } 3285b411b363SPhilipp Reisner } 3286b411b363SPhilipp Reisner 3287b411b363SPhilipp Reisner *rule_nr = 60; 3288b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[UI_CURRENT] & ~((u64)1); 3289b411b363SPhilipp Reisner for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 3290b30ab791SAndreas Gruenbacher peer = device->p_uuid[i] & ~((u64)1); 3291b411b363SPhilipp Reisner if (self == peer) 3292b411b363SPhilipp Reisner return -2; 3293b411b363SPhilipp Reisner } 3294b411b363SPhilipp Reisner 3295b411b363SPhilipp Reisner *rule_nr = 70; 3296b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[UI_BITMAP] & ~((u64)1); 3297b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_CURRENT] & ~((u64)1); 3298b411b363SPhilipp Reisner if (self == peer) 3299b411b363SPhilipp Reisner return 1; 3300b411b363SPhilipp Reisner 3301b411b363SPhilipp Reisner *rule_nr = 71; 3302b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); 3303b411b363SPhilipp Reisner if (self == peer) { 330444a4d551SLars Ellenberg if (connection->agreed_pro_version < 96 ? 3305b30ab791SAndreas Gruenbacher (device->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == 3306b30ab791SAndreas Gruenbacher (device->p_uuid[UI_HISTORY_START] & ~((u64)1)) : 3307b30ab791SAndreas Gruenbacher self + UUID_NEW_BM_OFFSET == (device->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { 3308b411b363SPhilipp Reisner /* The last P_SYNC_UUID did not get though. Undo the last start of 3309b411b363SPhilipp Reisner resync as sync source modifications of our UUIDs. */ 3310b411b363SPhilipp Reisner 331144a4d551SLars Ellenberg if (connection->agreed_pro_version < 91) 33124a23f264SPhilipp Reisner return -1091; 3313b411b363SPhilipp Reisner 3314b30ab791SAndreas Gruenbacher __drbd_uuid_set(device, UI_BITMAP, device->ldev->md.uuid[UI_HISTORY_START]); 3315b30ab791SAndreas Gruenbacher __drbd_uuid_set(device, UI_HISTORY_START, device->ldev->md.uuid[UI_HISTORY_START + 1]); 3316b411b363SPhilipp Reisner 3317d0180171SAndreas Gruenbacher drbd_info(device, "Last syncUUID did not get through, corrected:\n"); 3318b30ab791SAndreas Gruenbacher drbd_uuid_dump(device, "self", device->ldev->md.uuid, 3319b30ab791SAndreas Gruenbacher device->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(device) : 0, 0); 3320b411b363SPhilipp Reisner 3321b411b363SPhilipp Reisner return 1; 3322b411b363SPhilipp Reisner } 3323b411b363SPhilipp Reisner } 3324b411b363SPhilipp Reisner 3325b411b363SPhilipp Reisner 3326b411b363SPhilipp Reisner *rule_nr = 80; 3327b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_CURRENT] & ~((u64)1); 3328b411b363SPhilipp Reisner for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 3329b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[i] & ~((u64)1); 3330b411b363SPhilipp Reisner if (self == peer) 3331b411b363SPhilipp Reisner return 2; 3332b411b363SPhilipp Reisner } 3333b411b363SPhilipp Reisner 3334b411b363SPhilipp Reisner *rule_nr = 90; 3335b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[UI_BITMAP] & ~((u64)1); 3336b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_BITMAP] & ~((u64)1); 3337b411b363SPhilipp Reisner if (self == peer && self != ((u64)0)) 3338b411b363SPhilipp Reisner return 100; 3339b411b363SPhilipp Reisner 3340b411b363SPhilipp Reisner *rule_nr = 100; 3341b411b363SPhilipp Reisner for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 3342b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[i] & ~((u64)1); 3343b411b363SPhilipp Reisner for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { 3344b30ab791SAndreas Gruenbacher peer = device->p_uuid[j] & ~((u64)1); 3345b411b363SPhilipp Reisner if (self == peer) 3346b411b363SPhilipp Reisner return -100; 3347b411b363SPhilipp Reisner } 3348b411b363SPhilipp Reisner } 3349b411b363SPhilipp Reisner 3350b411b363SPhilipp Reisner return -1000; 3351b411b363SPhilipp Reisner } 3352b411b363SPhilipp Reisner 3353b411b363SPhilipp Reisner /* drbd_sync_handshake() returns the new conn state on success, or 3354b411b363SPhilipp Reisner CONN_MASK (-1) on failure. 3355b411b363SPhilipp Reisner */ 335669a22773SAndreas Gruenbacher static enum drbd_conns drbd_sync_handshake(struct drbd_peer_device *peer_device, 335769a22773SAndreas Gruenbacher enum drbd_role peer_role, 3358b411b363SPhilipp Reisner enum drbd_disk_state peer_disk) __must_hold(local) 3359b411b363SPhilipp Reisner { 336069a22773SAndreas Gruenbacher struct drbd_device *device = peer_device->device; 3361b411b363SPhilipp Reisner enum drbd_conns rv = C_MASK; 3362b411b363SPhilipp Reisner enum drbd_disk_state mydisk; 336344ed167dSPhilipp Reisner struct net_conf *nc; 33646dff2902SAndreas Gruenbacher int hg, rule_nr, rr_conflict, tentative; 3365b411b363SPhilipp Reisner 3366b30ab791SAndreas Gruenbacher mydisk = device->state.disk; 3367b411b363SPhilipp Reisner if (mydisk == D_NEGOTIATING) 3368b30ab791SAndreas Gruenbacher mydisk = device->new_state_tmp.disk; 3369b411b363SPhilipp Reisner 3370d0180171SAndreas Gruenbacher drbd_info(device, "drbd_sync_handshake:\n"); 33719f2247bbSPhilipp Reisner 3372b30ab791SAndreas Gruenbacher spin_lock_irq(&device->ldev->md.uuid_lock); 3373b30ab791SAndreas Gruenbacher drbd_uuid_dump(device, "self", device->ldev->md.uuid, device->comm_bm_set, 0); 3374b30ab791SAndreas Gruenbacher drbd_uuid_dump(device, "peer", device->p_uuid, 3375b30ab791SAndreas Gruenbacher device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); 3376b411b363SPhilipp Reisner 3377f2d3d75bSLars Ellenberg hg = drbd_uuid_compare(device, peer_role, &rule_nr); 3378b30ab791SAndreas Gruenbacher spin_unlock_irq(&device->ldev->md.uuid_lock); 3379b411b363SPhilipp Reisner 3380d0180171SAndreas Gruenbacher drbd_info(device, "uuid_compare()=%d by rule %d\n", hg, rule_nr); 3381b411b363SPhilipp Reisner 3382b411b363SPhilipp Reisner if (hg == -1000) { 3383d0180171SAndreas Gruenbacher drbd_alert(device, "Unrelated data, aborting!\n"); 3384b411b363SPhilipp Reisner return C_MASK; 3385b411b363SPhilipp Reisner } 3386f2d3d75bSLars Ellenberg if (hg < -0x10000) { 3387f2d3d75bSLars Ellenberg int proto, fflags; 3388f2d3d75bSLars Ellenberg hg = -hg; 3389f2d3d75bSLars Ellenberg proto = hg & 0xff; 3390f2d3d75bSLars Ellenberg fflags = (hg >> 8) & 0xff; 3391f2d3d75bSLars Ellenberg drbd_alert(device, "To resolve this both sides have to support at least protocol %d and feature flags 0x%x\n", 3392f2d3d75bSLars Ellenberg proto, fflags); 3393f2d3d75bSLars Ellenberg return C_MASK; 3394f2d3d75bSLars Ellenberg } 33954a23f264SPhilipp Reisner if (hg < -1000) { 3396d0180171SAndreas Gruenbacher drbd_alert(device, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); 3397b411b363SPhilipp Reisner return C_MASK; 3398b411b363SPhilipp Reisner } 3399b411b363SPhilipp Reisner 3400b411b363SPhilipp Reisner if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || 3401b411b363SPhilipp Reisner (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { 3402b411b363SPhilipp Reisner int f = (hg == -100) || abs(hg) == 2; 3403b411b363SPhilipp Reisner hg = mydisk > D_INCONSISTENT ? 1 : -1; 3404b411b363SPhilipp Reisner if (f) 3405b411b363SPhilipp Reisner hg = hg*2; 3406d0180171SAndreas Gruenbacher drbd_info(device, "Becoming sync %s due to disk states.\n", 3407b411b363SPhilipp Reisner hg > 0 ? "source" : "target"); 3408b411b363SPhilipp Reisner } 3409b411b363SPhilipp Reisner 34103a11a487SAdam Gandelman if (abs(hg) == 100) 3411b30ab791SAndreas Gruenbacher drbd_khelper(device, "initial-split-brain"); 34123a11a487SAdam Gandelman 341344ed167dSPhilipp Reisner rcu_read_lock(); 341469a22773SAndreas Gruenbacher nc = rcu_dereference(peer_device->connection->net_conf); 341544ed167dSPhilipp Reisner 341644ed167dSPhilipp Reisner if (hg == 100 || (hg == -100 && nc->always_asbp)) { 3417b30ab791SAndreas Gruenbacher int pcount = (device->state.role == R_PRIMARY) 3418b411b363SPhilipp Reisner + (peer_role == R_PRIMARY); 3419b411b363SPhilipp Reisner int forced = (hg == -100); 3420b411b363SPhilipp Reisner 3421b411b363SPhilipp Reisner switch (pcount) { 3422b411b363SPhilipp Reisner case 0: 342369a22773SAndreas Gruenbacher hg = drbd_asb_recover_0p(peer_device); 3424b411b363SPhilipp Reisner break; 3425b411b363SPhilipp Reisner case 1: 342669a22773SAndreas Gruenbacher hg = drbd_asb_recover_1p(peer_device); 3427b411b363SPhilipp Reisner break; 3428b411b363SPhilipp Reisner case 2: 342969a22773SAndreas Gruenbacher hg = drbd_asb_recover_2p(peer_device); 3430b411b363SPhilipp Reisner break; 3431b411b363SPhilipp Reisner } 3432b411b363SPhilipp Reisner if (abs(hg) < 100) { 3433d0180171SAndreas Gruenbacher drbd_warn(device, "Split-Brain detected, %d primaries, " 3434b411b363SPhilipp Reisner "automatically solved. Sync from %s node\n", 3435b411b363SPhilipp Reisner pcount, (hg < 0) ? "peer" : "this"); 3436b411b363SPhilipp Reisner if (forced) { 3437d0180171SAndreas Gruenbacher drbd_warn(device, "Doing a full sync, since" 3438b411b363SPhilipp Reisner " UUIDs where ambiguous.\n"); 3439b411b363SPhilipp Reisner hg = hg*2; 3440b411b363SPhilipp Reisner } 3441b411b363SPhilipp Reisner } 3442b411b363SPhilipp Reisner } 3443b411b363SPhilipp Reisner 3444b411b363SPhilipp Reisner if (hg == -100) { 3445b30ab791SAndreas Gruenbacher if (test_bit(DISCARD_MY_DATA, &device->flags) && !(device->p_uuid[UI_FLAGS]&1)) 3446b411b363SPhilipp Reisner hg = -1; 3447b30ab791SAndreas Gruenbacher if (!test_bit(DISCARD_MY_DATA, &device->flags) && (device->p_uuid[UI_FLAGS]&1)) 3448b411b363SPhilipp Reisner hg = 1; 3449b411b363SPhilipp Reisner 3450b411b363SPhilipp Reisner if (abs(hg) < 100) 3451d0180171SAndreas Gruenbacher drbd_warn(device, "Split-Brain detected, manually solved. " 3452b411b363SPhilipp Reisner "Sync from %s node\n", 3453b411b363SPhilipp Reisner (hg < 0) ? "peer" : "this"); 3454b411b363SPhilipp Reisner } 345544ed167dSPhilipp Reisner rr_conflict = nc->rr_conflict; 34566dff2902SAndreas Gruenbacher tentative = nc->tentative; 345744ed167dSPhilipp Reisner rcu_read_unlock(); 3458b411b363SPhilipp Reisner 3459b411b363SPhilipp Reisner if (hg == -100) { 3460580b9767SLars Ellenberg /* FIXME this log message is not correct if we end up here 3461580b9767SLars Ellenberg * after an attempted attach on a diskless node. 3462580b9767SLars Ellenberg * We just refuse to attach -- well, we drop the "connection" 3463580b9767SLars Ellenberg * to that disk, in a way... */ 3464d0180171SAndreas Gruenbacher drbd_alert(device, "Split-Brain detected but unresolved, dropping connection!\n"); 3465b30ab791SAndreas Gruenbacher drbd_khelper(device, "split-brain"); 3466b411b363SPhilipp Reisner return C_MASK; 3467b411b363SPhilipp Reisner } 3468b411b363SPhilipp Reisner 3469b411b363SPhilipp Reisner if (hg > 0 && mydisk <= D_INCONSISTENT) { 3470d0180171SAndreas Gruenbacher drbd_err(device, "I shall become SyncSource, but I am inconsistent!\n"); 3471b411b363SPhilipp Reisner return C_MASK; 3472b411b363SPhilipp Reisner } 3473b411b363SPhilipp Reisner 3474b411b363SPhilipp Reisner if (hg < 0 && /* by intention we do not use mydisk here. */ 3475b30ab791SAndreas Gruenbacher device->state.role == R_PRIMARY && device->state.disk >= D_CONSISTENT) { 347644ed167dSPhilipp Reisner switch (rr_conflict) { 3477b411b363SPhilipp Reisner case ASB_CALL_HELPER: 3478b30ab791SAndreas Gruenbacher drbd_khelper(device, "pri-lost"); 3479b411b363SPhilipp Reisner /* fall through */ 3480b411b363SPhilipp Reisner case ASB_DISCONNECT: 3481d0180171SAndreas Gruenbacher drbd_err(device, "I shall become SyncTarget, but I am primary!\n"); 3482b411b363SPhilipp Reisner return C_MASK; 3483b411b363SPhilipp Reisner case ASB_VIOLENTLY: 3484d0180171SAndreas Gruenbacher drbd_warn(device, "Becoming SyncTarget, violating the stable-data" 3485b411b363SPhilipp Reisner "assumption\n"); 3486b411b363SPhilipp Reisner } 3487b411b363SPhilipp Reisner } 3488b411b363SPhilipp Reisner 348969a22773SAndreas Gruenbacher if (tentative || test_bit(CONN_DRY_RUN, &peer_device->connection->flags)) { 3490cf14c2e9SPhilipp Reisner if (hg == 0) 3491d0180171SAndreas Gruenbacher drbd_info(device, "dry-run connect: No resync, would become Connected immediately.\n"); 3492cf14c2e9SPhilipp Reisner else 3493d0180171SAndreas Gruenbacher drbd_info(device, "dry-run connect: Would become %s, doing a %s resync.", 3494cf14c2e9SPhilipp Reisner drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), 3495cf14c2e9SPhilipp Reisner abs(hg) >= 2 ? "full" : "bit-map based"); 3496cf14c2e9SPhilipp Reisner return C_MASK; 3497cf14c2e9SPhilipp Reisner } 3498cf14c2e9SPhilipp Reisner 3499b411b363SPhilipp Reisner if (abs(hg) >= 2) { 3500d0180171SAndreas Gruenbacher drbd_info(device, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); 3501b30ab791SAndreas Gruenbacher if (drbd_bitmap_io(device, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", 350220ceb2b2SLars Ellenberg BM_LOCKED_SET_ALLOWED)) 3503b411b363SPhilipp Reisner return C_MASK; 3504b411b363SPhilipp Reisner } 3505b411b363SPhilipp Reisner 3506b411b363SPhilipp Reisner if (hg > 0) { /* become sync source. */ 3507b411b363SPhilipp Reisner rv = C_WF_BITMAP_S; 3508b411b363SPhilipp Reisner } else if (hg < 0) { /* become sync target */ 3509b411b363SPhilipp Reisner rv = C_WF_BITMAP_T; 3510b411b363SPhilipp Reisner } else { 3511b411b363SPhilipp Reisner rv = C_CONNECTED; 3512b30ab791SAndreas Gruenbacher if (drbd_bm_total_weight(device)) { 3513d0180171SAndreas Gruenbacher drbd_info(device, "No resync, but %lu bits in bitmap!\n", 3514b30ab791SAndreas Gruenbacher drbd_bm_total_weight(device)); 3515b411b363SPhilipp Reisner } 3516b411b363SPhilipp Reisner } 3517b411b363SPhilipp Reisner 3518b411b363SPhilipp Reisner return rv; 3519b411b363SPhilipp Reisner } 3520b411b363SPhilipp Reisner 3521f179d76dSPhilipp Reisner static enum drbd_after_sb_p convert_after_sb(enum drbd_after_sb_p peer) 3522b411b363SPhilipp Reisner { 3523b411b363SPhilipp Reisner /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ 3524f179d76dSPhilipp Reisner if (peer == ASB_DISCARD_REMOTE) 3525f179d76dSPhilipp Reisner return ASB_DISCARD_LOCAL; 3526b411b363SPhilipp Reisner 3527b411b363SPhilipp Reisner /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ 3528f179d76dSPhilipp Reisner if (peer == ASB_DISCARD_LOCAL) 3529f179d76dSPhilipp Reisner return ASB_DISCARD_REMOTE; 3530b411b363SPhilipp Reisner 3531b411b363SPhilipp Reisner /* everything else is valid if they are equal on both sides. */ 3532f179d76dSPhilipp Reisner return peer; 3533b411b363SPhilipp Reisner } 3534b411b363SPhilipp Reisner 3535bde89a9eSAndreas Gruenbacher static int receive_protocol(struct drbd_connection *connection, struct packet_info *pi) 3536b411b363SPhilipp Reisner { 3537e658983aSAndreas Gruenbacher struct p_protocol *p = pi->data; 3538036b17eaSPhilipp Reisner enum drbd_after_sb_p p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; 3539036b17eaSPhilipp Reisner int p_proto, p_discard_my_data, p_two_primaries, cf; 3540036b17eaSPhilipp Reisner struct net_conf *nc, *old_net_conf, *new_net_conf = NULL; 3541036b17eaSPhilipp Reisner char integrity_alg[SHARED_SECRET_MAX] = ""; 35429534d671SHerbert Xu struct crypto_ahash *peer_integrity_tfm = NULL; 35437aca6c75SPhilipp Reisner void *int_dig_in = NULL, *int_dig_vv = NULL; 3544b411b363SPhilipp Reisner 3545b411b363SPhilipp Reisner p_proto = be32_to_cpu(p->protocol); 3546b411b363SPhilipp Reisner p_after_sb_0p = be32_to_cpu(p->after_sb_0p); 3547b411b363SPhilipp Reisner p_after_sb_1p = be32_to_cpu(p->after_sb_1p); 3548b411b363SPhilipp Reisner p_after_sb_2p = be32_to_cpu(p->after_sb_2p); 3549b411b363SPhilipp Reisner p_two_primaries = be32_to_cpu(p->two_primaries); 3550cf14c2e9SPhilipp Reisner cf = be32_to_cpu(p->conn_flags); 35516139f60dSAndreas Gruenbacher p_discard_my_data = cf & CF_DISCARD_MY_DATA; 3552cf14c2e9SPhilipp Reisner 3553bde89a9eSAndreas Gruenbacher if (connection->agreed_pro_version >= 87) { 355486db0618SAndreas Gruenbacher int err; 355586db0618SAndreas Gruenbacher 355688104ca4SAndreas Gruenbacher if (pi->size > sizeof(integrity_alg)) 355786db0618SAndreas Gruenbacher return -EIO; 3558bde89a9eSAndreas Gruenbacher err = drbd_recv_all(connection, integrity_alg, pi->size); 355986db0618SAndreas Gruenbacher if (err) 356086db0618SAndreas Gruenbacher return err; 356188104ca4SAndreas Gruenbacher integrity_alg[SHARED_SECRET_MAX - 1] = 0; 3562036b17eaSPhilipp Reisner } 356386db0618SAndreas Gruenbacher 35647d4c782cSAndreas Gruenbacher if (pi->cmd != P_PROTOCOL_UPDATE) { 3565bde89a9eSAndreas Gruenbacher clear_bit(CONN_DRY_RUN, &connection->flags); 3566cf14c2e9SPhilipp Reisner 3567cf14c2e9SPhilipp Reisner if (cf & CF_DRY_RUN) 3568bde89a9eSAndreas Gruenbacher set_bit(CONN_DRY_RUN, &connection->flags); 3569b411b363SPhilipp Reisner 357044ed167dSPhilipp Reisner rcu_read_lock(); 3571bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 357244ed167dSPhilipp Reisner 3573036b17eaSPhilipp Reisner if (p_proto != nc->wire_protocol) { 35741ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "protocol"); 357544ed167dSPhilipp Reisner goto disconnect_rcu_unlock; 3576b411b363SPhilipp Reisner } 3577b411b363SPhilipp Reisner 3578f179d76dSPhilipp Reisner if (convert_after_sb(p_after_sb_0p) != nc->after_sb_0p) { 35791ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "after-sb-0pri"); 358044ed167dSPhilipp Reisner goto disconnect_rcu_unlock; 3581b411b363SPhilipp Reisner } 3582b411b363SPhilipp Reisner 3583f179d76dSPhilipp Reisner if (convert_after_sb(p_after_sb_1p) != nc->after_sb_1p) { 35841ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "after-sb-1pri"); 358544ed167dSPhilipp Reisner goto disconnect_rcu_unlock; 3586b411b363SPhilipp Reisner } 3587b411b363SPhilipp Reisner 3588f179d76dSPhilipp Reisner if (convert_after_sb(p_after_sb_2p) != nc->after_sb_2p) { 35891ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "after-sb-2pri"); 359044ed167dSPhilipp Reisner goto disconnect_rcu_unlock; 3591b411b363SPhilipp Reisner } 3592b411b363SPhilipp Reisner 35936139f60dSAndreas Gruenbacher if (p_discard_my_data && nc->discard_my_data) { 35941ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "discard-my-data"); 359544ed167dSPhilipp Reisner goto disconnect_rcu_unlock; 3596b411b363SPhilipp Reisner } 3597b411b363SPhilipp Reisner 359844ed167dSPhilipp Reisner if (p_two_primaries != nc->two_primaries) { 35991ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "allow-two-primaries"); 360044ed167dSPhilipp Reisner goto disconnect_rcu_unlock; 3601b411b363SPhilipp Reisner } 3602b411b363SPhilipp Reisner 3603036b17eaSPhilipp Reisner if (strcmp(integrity_alg, nc->integrity_alg)) { 36041ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "data-integrity-alg"); 3605036b17eaSPhilipp Reisner goto disconnect_rcu_unlock; 3606036b17eaSPhilipp Reisner } 3607036b17eaSPhilipp Reisner 360886db0618SAndreas Gruenbacher rcu_read_unlock(); 3609fbc12f45SAndreas Gruenbacher } 36107d4c782cSAndreas Gruenbacher 36117d4c782cSAndreas Gruenbacher if (integrity_alg[0]) { 36127d4c782cSAndreas Gruenbacher int hash_size; 36137d4c782cSAndreas Gruenbacher 36147d4c782cSAndreas Gruenbacher /* 36157d4c782cSAndreas Gruenbacher * We can only change the peer data integrity algorithm 36167d4c782cSAndreas Gruenbacher * here. Changing our own data integrity algorithm 36177d4c782cSAndreas Gruenbacher * requires that we send a P_PROTOCOL_UPDATE packet at 36187d4c782cSAndreas Gruenbacher * the same time; otherwise, the peer has no way to 36197d4c782cSAndreas Gruenbacher * tell between which packets the algorithm should 36207d4c782cSAndreas Gruenbacher * change. 36217d4c782cSAndreas Gruenbacher */ 36227d4c782cSAndreas Gruenbacher 36239534d671SHerbert Xu peer_integrity_tfm = crypto_alloc_ahash(integrity_alg, 0, CRYPTO_ALG_ASYNC); 36241b57e663SLars Ellenberg if (IS_ERR(peer_integrity_tfm)) { 36251b57e663SLars Ellenberg peer_integrity_tfm = NULL; 36261ec861ebSAndreas Gruenbacher drbd_err(connection, "peer data-integrity-alg %s not supported\n", 36277d4c782cSAndreas Gruenbacher integrity_alg); 3628b411b363SPhilipp Reisner goto disconnect; 3629b411b363SPhilipp Reisner } 3630b411b363SPhilipp Reisner 36319534d671SHerbert Xu hash_size = crypto_ahash_digestsize(peer_integrity_tfm); 36327d4c782cSAndreas Gruenbacher int_dig_in = kmalloc(hash_size, GFP_KERNEL); 36337d4c782cSAndreas Gruenbacher int_dig_vv = kmalloc(hash_size, GFP_KERNEL); 36347d4c782cSAndreas Gruenbacher if (!(int_dig_in && int_dig_vv)) { 36351ec861ebSAndreas Gruenbacher drbd_err(connection, "Allocation of buffers for data integrity checking failed\n"); 36367d4c782cSAndreas Gruenbacher goto disconnect; 36377d4c782cSAndreas Gruenbacher } 36387d4c782cSAndreas Gruenbacher } 36397d4c782cSAndreas Gruenbacher 36407d4c782cSAndreas Gruenbacher new_net_conf = kmalloc(sizeof(struct net_conf), GFP_KERNEL); 36417d4c782cSAndreas Gruenbacher if (!new_net_conf) { 36421ec861ebSAndreas Gruenbacher drbd_err(connection, "Allocation of new net_conf failed\n"); 3643b411b363SPhilipp Reisner goto disconnect; 3644b411b363SPhilipp Reisner } 3645b411b363SPhilipp Reisner 3646bde89a9eSAndreas Gruenbacher mutex_lock(&connection->data.mutex); 36470500813fSAndreas Gruenbacher mutex_lock(&connection->resource->conf_update); 3648bde89a9eSAndreas Gruenbacher old_net_conf = connection->net_conf; 36497d4c782cSAndreas Gruenbacher *new_net_conf = *old_net_conf; 3650b411b363SPhilipp Reisner 36517d4c782cSAndreas Gruenbacher new_net_conf->wire_protocol = p_proto; 36527d4c782cSAndreas Gruenbacher new_net_conf->after_sb_0p = convert_after_sb(p_after_sb_0p); 36537d4c782cSAndreas Gruenbacher new_net_conf->after_sb_1p = convert_after_sb(p_after_sb_1p); 36547d4c782cSAndreas Gruenbacher new_net_conf->after_sb_2p = convert_after_sb(p_after_sb_2p); 36557d4c782cSAndreas Gruenbacher new_net_conf->two_primaries = p_two_primaries; 3656b411b363SPhilipp Reisner 3657bde89a9eSAndreas Gruenbacher rcu_assign_pointer(connection->net_conf, new_net_conf); 36580500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 3659bde89a9eSAndreas Gruenbacher mutex_unlock(&connection->data.mutex); 3660b411b363SPhilipp Reisner 36619534d671SHerbert Xu crypto_free_ahash(connection->peer_integrity_tfm); 3662bde89a9eSAndreas Gruenbacher kfree(connection->int_dig_in); 3663bde89a9eSAndreas Gruenbacher kfree(connection->int_dig_vv); 3664bde89a9eSAndreas Gruenbacher connection->peer_integrity_tfm = peer_integrity_tfm; 3665bde89a9eSAndreas Gruenbacher connection->int_dig_in = int_dig_in; 3666bde89a9eSAndreas Gruenbacher connection->int_dig_vv = int_dig_vv; 3667b411b363SPhilipp Reisner 36687d4c782cSAndreas Gruenbacher if (strcmp(old_net_conf->integrity_alg, integrity_alg)) 36691ec861ebSAndreas Gruenbacher drbd_info(connection, "peer data-integrity-alg: %s\n", 36707d4c782cSAndreas Gruenbacher integrity_alg[0] ? integrity_alg : "(none)"); 3671b411b363SPhilipp Reisner 36727d4c782cSAndreas Gruenbacher synchronize_rcu(); 36737d4c782cSAndreas Gruenbacher kfree(old_net_conf); 367482bc0194SAndreas Gruenbacher return 0; 3675b411b363SPhilipp Reisner 367644ed167dSPhilipp Reisner disconnect_rcu_unlock: 367744ed167dSPhilipp Reisner rcu_read_unlock(); 3678b411b363SPhilipp Reisner disconnect: 36799534d671SHerbert Xu crypto_free_ahash(peer_integrity_tfm); 3680036b17eaSPhilipp Reisner kfree(int_dig_in); 3681036b17eaSPhilipp Reisner kfree(int_dig_vv); 3682bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); 368382bc0194SAndreas Gruenbacher return -EIO; 3684b411b363SPhilipp Reisner } 3685b411b363SPhilipp Reisner 3686b411b363SPhilipp Reisner /* helper function 3687b411b363SPhilipp Reisner * input: alg name, feature name 3688b411b363SPhilipp Reisner * return: NULL (alg name was "") 3689b411b363SPhilipp Reisner * ERR_PTR(error) if something goes wrong 3690b411b363SPhilipp Reisner * or the crypto hash ptr, if it worked out ok. */ 36919534d671SHerbert Xu static struct crypto_ahash *drbd_crypto_alloc_digest_safe(const struct drbd_device *device, 3692b411b363SPhilipp Reisner const char *alg, const char *name) 3693b411b363SPhilipp Reisner { 36949534d671SHerbert Xu struct crypto_ahash *tfm; 3695b411b363SPhilipp Reisner 3696b411b363SPhilipp Reisner if (!alg[0]) 3697b411b363SPhilipp Reisner return NULL; 3698b411b363SPhilipp Reisner 36999534d671SHerbert Xu tfm = crypto_alloc_ahash(alg, 0, CRYPTO_ALG_ASYNC); 3700b411b363SPhilipp Reisner if (IS_ERR(tfm)) { 3701d0180171SAndreas Gruenbacher drbd_err(device, "Can not allocate \"%s\" as %s (reason: %ld)\n", 3702b411b363SPhilipp Reisner alg, name, PTR_ERR(tfm)); 3703b411b363SPhilipp Reisner return tfm; 3704b411b363SPhilipp Reisner } 3705b411b363SPhilipp Reisner return tfm; 3706b411b363SPhilipp Reisner } 3707b411b363SPhilipp Reisner 3708bde89a9eSAndreas Gruenbacher static int ignore_remaining_packet(struct drbd_connection *connection, struct packet_info *pi) 3709b411b363SPhilipp Reisner { 3710bde89a9eSAndreas Gruenbacher void *buffer = connection->data.rbuf; 37114a76b161SAndreas Gruenbacher int size = pi->size; 37124a76b161SAndreas Gruenbacher 37134a76b161SAndreas Gruenbacher while (size) { 37144a76b161SAndreas Gruenbacher int s = min_t(int, size, DRBD_SOCKET_BUFFER_SIZE); 3715bde89a9eSAndreas Gruenbacher s = drbd_recv(connection, buffer, s); 37164a76b161SAndreas Gruenbacher if (s <= 0) { 37174a76b161SAndreas Gruenbacher if (s < 0) 37184a76b161SAndreas Gruenbacher return s; 37194a76b161SAndreas Gruenbacher break; 37204a76b161SAndreas Gruenbacher } 37214a76b161SAndreas Gruenbacher size -= s; 37224a76b161SAndreas Gruenbacher } 37234a76b161SAndreas Gruenbacher if (size) 37244a76b161SAndreas Gruenbacher return -EIO; 37254a76b161SAndreas Gruenbacher return 0; 37264a76b161SAndreas Gruenbacher } 37274a76b161SAndreas Gruenbacher 37284a76b161SAndreas Gruenbacher /* 37294a76b161SAndreas Gruenbacher * config_unknown_volume - device configuration command for unknown volume 37304a76b161SAndreas Gruenbacher * 37314a76b161SAndreas Gruenbacher * When a device is added to an existing connection, the node on which the 37324a76b161SAndreas Gruenbacher * device is added first will send configuration commands to its peer but the 37334a76b161SAndreas Gruenbacher * peer will not know about the device yet. It will warn and ignore these 37344a76b161SAndreas Gruenbacher * commands. Once the device is added on the second node, the second node will 37354a76b161SAndreas Gruenbacher * send the same device configuration commands, but in the other direction. 37364a76b161SAndreas Gruenbacher * 37374a76b161SAndreas Gruenbacher * (We can also end up here if drbd is misconfigured.) 37384a76b161SAndreas Gruenbacher */ 3739bde89a9eSAndreas Gruenbacher static int config_unknown_volume(struct drbd_connection *connection, struct packet_info *pi) 37404a76b161SAndreas Gruenbacher { 37411ec861ebSAndreas Gruenbacher drbd_warn(connection, "%s packet received for volume %u, which is not configured locally\n", 37422fcb8f30SAndreas Gruenbacher cmdname(pi->cmd), pi->vnr); 3743bde89a9eSAndreas Gruenbacher return ignore_remaining_packet(connection, pi); 37444a76b161SAndreas Gruenbacher } 37454a76b161SAndreas Gruenbacher 3746bde89a9eSAndreas Gruenbacher static int receive_SyncParam(struct drbd_connection *connection, struct packet_info *pi) 37474a76b161SAndreas Gruenbacher { 37489f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 3749b30ab791SAndreas Gruenbacher struct drbd_device *device; 3750e658983aSAndreas Gruenbacher struct p_rs_param_95 *p; 3751b411b363SPhilipp Reisner unsigned int header_size, data_size, exp_max_sz; 37529534d671SHerbert Xu struct crypto_ahash *verify_tfm = NULL; 37539534d671SHerbert Xu struct crypto_ahash *csums_tfm = NULL; 37542ec91e0eSPhilipp Reisner struct net_conf *old_net_conf, *new_net_conf = NULL; 3755813472ceSPhilipp Reisner struct disk_conf *old_disk_conf = NULL, *new_disk_conf = NULL; 3756bde89a9eSAndreas Gruenbacher const int apv = connection->agreed_pro_version; 3757813472ceSPhilipp Reisner struct fifo_buffer *old_plan = NULL, *new_plan = NULL; 3758778f271dSPhilipp Reisner int fifo_size = 0; 375982bc0194SAndreas Gruenbacher int err; 3760b411b363SPhilipp Reisner 37619f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 37629f4fe9adSAndreas Gruenbacher if (!peer_device) 3763bde89a9eSAndreas Gruenbacher return config_unknown_volume(connection, pi); 37649f4fe9adSAndreas Gruenbacher device = peer_device->device; 3765b411b363SPhilipp Reisner 3766b411b363SPhilipp Reisner exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) 3767b411b363SPhilipp Reisner : apv == 88 ? sizeof(struct p_rs_param) 3768b411b363SPhilipp Reisner + SHARED_SECRET_MAX 37698e26f9ccSPhilipp Reisner : apv <= 94 ? sizeof(struct p_rs_param_89) 37708e26f9ccSPhilipp Reisner : /* apv >= 95 */ sizeof(struct p_rs_param_95); 3771b411b363SPhilipp Reisner 3772e2857216SAndreas Gruenbacher if (pi->size > exp_max_sz) { 3773d0180171SAndreas Gruenbacher drbd_err(device, "SyncParam packet too long: received %u, expected <= %u bytes\n", 3774e2857216SAndreas Gruenbacher pi->size, exp_max_sz); 377582bc0194SAndreas Gruenbacher return -EIO; 3776b411b363SPhilipp Reisner } 3777b411b363SPhilipp Reisner 3778b411b363SPhilipp Reisner if (apv <= 88) { 3779e658983aSAndreas Gruenbacher header_size = sizeof(struct p_rs_param); 3780e2857216SAndreas Gruenbacher data_size = pi->size - header_size; 37818e26f9ccSPhilipp Reisner } else if (apv <= 94) { 3782e658983aSAndreas Gruenbacher header_size = sizeof(struct p_rs_param_89); 3783e2857216SAndreas Gruenbacher data_size = pi->size - header_size; 37840b0ba1efSAndreas Gruenbacher D_ASSERT(device, data_size == 0); 37858e26f9ccSPhilipp Reisner } else { 3786e658983aSAndreas Gruenbacher header_size = sizeof(struct p_rs_param_95); 3787e2857216SAndreas Gruenbacher data_size = pi->size - header_size; 37880b0ba1efSAndreas Gruenbacher D_ASSERT(device, data_size == 0); 3789b411b363SPhilipp Reisner } 3790b411b363SPhilipp Reisner 3791b411b363SPhilipp Reisner /* initialize verify_alg and csums_alg */ 3792e658983aSAndreas Gruenbacher p = pi->data; 3793b411b363SPhilipp Reisner memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); 3794b411b363SPhilipp Reisner 37959f4fe9adSAndreas Gruenbacher err = drbd_recv_all(peer_device->connection, p, header_size); 379682bc0194SAndreas Gruenbacher if (err) 379782bc0194SAndreas Gruenbacher return err; 3798b411b363SPhilipp Reisner 37990500813fSAndreas Gruenbacher mutex_lock(&connection->resource->conf_update); 38009f4fe9adSAndreas Gruenbacher old_net_conf = peer_device->connection->net_conf; 3801b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 3802daeda1ccSPhilipp Reisner new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); 3803daeda1ccSPhilipp Reisner if (!new_disk_conf) { 3804b30ab791SAndreas Gruenbacher put_ldev(device); 38050500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 3806d0180171SAndreas Gruenbacher drbd_err(device, "Allocation of new disk_conf failed\n"); 3807daeda1ccSPhilipp Reisner return -ENOMEM; 3808f399002eSLars Ellenberg } 3809b411b363SPhilipp Reisner 3810b30ab791SAndreas Gruenbacher old_disk_conf = device->ldev->disk_conf; 3811daeda1ccSPhilipp Reisner *new_disk_conf = *old_disk_conf; 3812daeda1ccSPhilipp Reisner 38136394b935SAndreas Gruenbacher new_disk_conf->resync_rate = be32_to_cpu(p->resync_rate); 3814813472ceSPhilipp Reisner } 3815b411b363SPhilipp Reisner 3816b411b363SPhilipp Reisner if (apv >= 88) { 3817b411b363SPhilipp Reisner if (apv == 88) { 38185de73827SPhilipp Reisner if (data_size > SHARED_SECRET_MAX || data_size == 0) { 3819d0180171SAndreas Gruenbacher drbd_err(device, "verify-alg of wrong size, " 38205de73827SPhilipp Reisner "peer wants %u, accepting only up to %u byte\n", 3821b411b363SPhilipp Reisner data_size, SHARED_SECRET_MAX); 3822813472ceSPhilipp Reisner err = -EIO; 3823813472ceSPhilipp Reisner goto reconnect; 3824b411b363SPhilipp Reisner } 3825b411b363SPhilipp Reisner 38269f4fe9adSAndreas Gruenbacher err = drbd_recv_all(peer_device->connection, p->verify_alg, data_size); 3827813472ceSPhilipp Reisner if (err) 3828813472ceSPhilipp Reisner goto reconnect; 3829b411b363SPhilipp Reisner /* we expect NUL terminated string */ 3830b411b363SPhilipp Reisner /* but just in case someone tries to be evil */ 38310b0ba1efSAndreas Gruenbacher D_ASSERT(device, p->verify_alg[data_size-1] == 0); 3832b411b363SPhilipp Reisner p->verify_alg[data_size-1] = 0; 3833b411b363SPhilipp Reisner 3834b411b363SPhilipp Reisner } else /* apv >= 89 */ { 3835b411b363SPhilipp Reisner /* we still expect NUL terminated strings */ 3836b411b363SPhilipp Reisner /* but just in case someone tries to be evil */ 38370b0ba1efSAndreas Gruenbacher D_ASSERT(device, p->verify_alg[SHARED_SECRET_MAX-1] == 0); 38380b0ba1efSAndreas Gruenbacher D_ASSERT(device, p->csums_alg[SHARED_SECRET_MAX-1] == 0); 3839b411b363SPhilipp Reisner p->verify_alg[SHARED_SECRET_MAX-1] = 0; 3840b411b363SPhilipp Reisner p->csums_alg[SHARED_SECRET_MAX-1] = 0; 3841b411b363SPhilipp Reisner } 3842b411b363SPhilipp Reisner 38432ec91e0eSPhilipp Reisner if (strcmp(old_net_conf->verify_alg, p->verify_alg)) { 3844b30ab791SAndreas Gruenbacher if (device->state.conn == C_WF_REPORT_PARAMS) { 3845d0180171SAndreas Gruenbacher drbd_err(device, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", 38462ec91e0eSPhilipp Reisner old_net_conf->verify_alg, p->verify_alg); 3847b411b363SPhilipp Reisner goto disconnect; 3848b411b363SPhilipp Reisner } 3849b30ab791SAndreas Gruenbacher verify_tfm = drbd_crypto_alloc_digest_safe(device, 3850b411b363SPhilipp Reisner p->verify_alg, "verify-alg"); 3851b411b363SPhilipp Reisner if (IS_ERR(verify_tfm)) { 3852b411b363SPhilipp Reisner verify_tfm = NULL; 3853b411b363SPhilipp Reisner goto disconnect; 3854b411b363SPhilipp Reisner } 3855b411b363SPhilipp Reisner } 3856b411b363SPhilipp Reisner 38572ec91e0eSPhilipp Reisner if (apv >= 89 && strcmp(old_net_conf->csums_alg, p->csums_alg)) { 3858b30ab791SAndreas Gruenbacher if (device->state.conn == C_WF_REPORT_PARAMS) { 3859d0180171SAndreas Gruenbacher drbd_err(device, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", 38602ec91e0eSPhilipp Reisner old_net_conf->csums_alg, p->csums_alg); 3861b411b363SPhilipp Reisner goto disconnect; 3862b411b363SPhilipp Reisner } 3863b30ab791SAndreas Gruenbacher csums_tfm = drbd_crypto_alloc_digest_safe(device, 3864b411b363SPhilipp Reisner p->csums_alg, "csums-alg"); 3865b411b363SPhilipp Reisner if (IS_ERR(csums_tfm)) { 3866b411b363SPhilipp Reisner csums_tfm = NULL; 3867b411b363SPhilipp Reisner goto disconnect; 3868b411b363SPhilipp Reisner } 3869b411b363SPhilipp Reisner } 3870b411b363SPhilipp Reisner 3871813472ceSPhilipp Reisner if (apv > 94 && new_disk_conf) { 3872daeda1ccSPhilipp Reisner new_disk_conf->c_plan_ahead = be32_to_cpu(p->c_plan_ahead); 3873daeda1ccSPhilipp Reisner new_disk_conf->c_delay_target = be32_to_cpu(p->c_delay_target); 3874daeda1ccSPhilipp Reisner new_disk_conf->c_fill_target = be32_to_cpu(p->c_fill_target); 3875daeda1ccSPhilipp Reisner new_disk_conf->c_max_rate = be32_to_cpu(p->c_max_rate); 3876778f271dSPhilipp Reisner 3877daeda1ccSPhilipp Reisner fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ; 3878b30ab791SAndreas Gruenbacher if (fifo_size != device->rs_plan_s->size) { 3879813472ceSPhilipp Reisner new_plan = fifo_alloc(fifo_size); 3880813472ceSPhilipp Reisner if (!new_plan) { 3881d0180171SAndreas Gruenbacher drbd_err(device, "kmalloc of fifo_buffer failed"); 3882b30ab791SAndreas Gruenbacher put_ldev(device); 3883778f271dSPhilipp Reisner goto disconnect; 3884778f271dSPhilipp Reisner } 3885778f271dSPhilipp Reisner } 38868e26f9ccSPhilipp Reisner } 3887b411b363SPhilipp Reisner 388891fd4dadSPhilipp Reisner if (verify_tfm || csums_tfm) { 38892ec91e0eSPhilipp Reisner new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL); 38902ec91e0eSPhilipp Reisner if (!new_net_conf) { 3891d0180171SAndreas Gruenbacher drbd_err(device, "Allocation of new net_conf failed\n"); 389291fd4dadSPhilipp Reisner goto disconnect; 389391fd4dadSPhilipp Reisner } 389491fd4dadSPhilipp Reisner 38952ec91e0eSPhilipp Reisner *new_net_conf = *old_net_conf; 389691fd4dadSPhilipp Reisner 3897b411b363SPhilipp Reisner if (verify_tfm) { 38982ec91e0eSPhilipp Reisner strcpy(new_net_conf->verify_alg, p->verify_alg); 38992ec91e0eSPhilipp Reisner new_net_conf->verify_alg_len = strlen(p->verify_alg) + 1; 39009534d671SHerbert Xu crypto_free_ahash(peer_device->connection->verify_tfm); 39019f4fe9adSAndreas Gruenbacher peer_device->connection->verify_tfm = verify_tfm; 3902d0180171SAndreas Gruenbacher drbd_info(device, "using verify-alg: \"%s\"\n", p->verify_alg); 3903b411b363SPhilipp Reisner } 3904b411b363SPhilipp Reisner if (csums_tfm) { 39052ec91e0eSPhilipp Reisner strcpy(new_net_conf->csums_alg, p->csums_alg); 39062ec91e0eSPhilipp Reisner new_net_conf->csums_alg_len = strlen(p->csums_alg) + 1; 39079534d671SHerbert Xu crypto_free_ahash(peer_device->connection->csums_tfm); 39089f4fe9adSAndreas Gruenbacher peer_device->connection->csums_tfm = csums_tfm; 3909d0180171SAndreas Gruenbacher drbd_info(device, "using csums-alg: \"%s\"\n", p->csums_alg); 3910b411b363SPhilipp Reisner } 3911bde89a9eSAndreas Gruenbacher rcu_assign_pointer(connection->net_conf, new_net_conf); 3912778f271dSPhilipp Reisner } 3913b411b363SPhilipp Reisner } 3914b411b363SPhilipp Reisner 3915813472ceSPhilipp Reisner if (new_disk_conf) { 3916b30ab791SAndreas Gruenbacher rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); 3917b30ab791SAndreas Gruenbacher put_ldev(device); 3918b411b363SPhilipp Reisner } 3919813472ceSPhilipp Reisner 3920813472ceSPhilipp Reisner if (new_plan) { 3921b30ab791SAndreas Gruenbacher old_plan = device->rs_plan_s; 3922b30ab791SAndreas Gruenbacher rcu_assign_pointer(device->rs_plan_s, new_plan); 3923813472ceSPhilipp Reisner } 3924daeda1ccSPhilipp Reisner 39250500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 3926daeda1ccSPhilipp Reisner synchronize_rcu(); 3927daeda1ccSPhilipp Reisner if (new_net_conf) 3928daeda1ccSPhilipp Reisner kfree(old_net_conf); 3929daeda1ccSPhilipp Reisner kfree(old_disk_conf); 3930813472ceSPhilipp Reisner kfree(old_plan); 3931daeda1ccSPhilipp Reisner 393282bc0194SAndreas Gruenbacher return 0; 3933b411b363SPhilipp Reisner 3934813472ceSPhilipp Reisner reconnect: 3935813472ceSPhilipp Reisner if (new_disk_conf) { 3936b30ab791SAndreas Gruenbacher put_ldev(device); 3937813472ceSPhilipp Reisner kfree(new_disk_conf); 3938813472ceSPhilipp Reisner } 39390500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 3940813472ceSPhilipp Reisner return -EIO; 3941813472ceSPhilipp Reisner 3942b411b363SPhilipp Reisner disconnect: 3943813472ceSPhilipp Reisner kfree(new_plan); 3944813472ceSPhilipp Reisner if (new_disk_conf) { 3945b30ab791SAndreas Gruenbacher put_ldev(device); 3946813472ceSPhilipp Reisner kfree(new_disk_conf); 3947813472ceSPhilipp Reisner } 39480500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 3949b411b363SPhilipp Reisner /* just for completeness: actually not needed, 3950b411b363SPhilipp Reisner * as this is not reached if csums_tfm was ok. */ 39519534d671SHerbert Xu crypto_free_ahash(csums_tfm); 3952b411b363SPhilipp Reisner /* but free the verify_tfm again, if csums_tfm did not work out */ 39539534d671SHerbert Xu crypto_free_ahash(verify_tfm); 39549f4fe9adSAndreas Gruenbacher conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); 395582bc0194SAndreas Gruenbacher return -EIO; 3956b411b363SPhilipp Reisner } 3957b411b363SPhilipp Reisner 3958b411b363SPhilipp Reisner /* warn if the arguments differ by more than 12.5% */ 3959b30ab791SAndreas Gruenbacher static void warn_if_differ_considerably(struct drbd_device *device, 3960b411b363SPhilipp Reisner const char *s, sector_t a, sector_t b) 3961b411b363SPhilipp Reisner { 3962b411b363SPhilipp Reisner sector_t d; 3963b411b363SPhilipp Reisner if (a == 0 || b == 0) 3964b411b363SPhilipp Reisner return; 3965b411b363SPhilipp Reisner d = (a > b) ? (a - b) : (b - a); 3966b411b363SPhilipp Reisner if (d > (a>>3) || d > (b>>3)) 3967d0180171SAndreas Gruenbacher drbd_warn(device, "Considerable difference in %s: %llus vs. %llus\n", s, 3968b411b363SPhilipp Reisner (unsigned long long)a, (unsigned long long)b); 3969b411b363SPhilipp Reisner } 3970b411b363SPhilipp Reisner 3971bde89a9eSAndreas Gruenbacher static int receive_sizes(struct drbd_connection *connection, struct packet_info *pi) 3972b411b363SPhilipp Reisner { 39739f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 3974b30ab791SAndreas Gruenbacher struct drbd_device *device; 3975e658983aSAndreas Gruenbacher struct p_sizes *p = pi->data; 39769104d31aSLars Ellenberg struct o_qlim *o = (connection->agreed_features & DRBD_FF_WSAME) ? p->qlim : NULL; 3977e96c9633SPhilipp Reisner enum determine_dev_size dd = DS_UNCHANGED; 39786a8d68b1SLars Ellenberg sector_t p_size, p_usize, p_csize, my_usize; 3979b411b363SPhilipp Reisner int ldsc = 0; /* local disk size changed */ 3980e89b591cSPhilipp Reisner enum dds_flags ddsf; 3981b411b363SPhilipp Reisner 39829f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 39839f4fe9adSAndreas Gruenbacher if (!peer_device) 3984bde89a9eSAndreas Gruenbacher return config_unknown_volume(connection, pi); 39859f4fe9adSAndreas Gruenbacher device = peer_device->device; 39864a76b161SAndreas Gruenbacher 3987b411b363SPhilipp Reisner p_size = be64_to_cpu(p->d_size); 3988b411b363SPhilipp Reisner p_usize = be64_to_cpu(p->u_size); 39896a8d68b1SLars Ellenberg p_csize = be64_to_cpu(p->c_size); 3990b411b363SPhilipp Reisner 3991b411b363SPhilipp Reisner /* just store the peer's disk size for now. 3992b411b363SPhilipp Reisner * we still need to figure out whether we accept that. */ 3993b30ab791SAndreas Gruenbacher device->p_size = p_size; 3994b411b363SPhilipp Reisner 3995b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 399660bac040SLars Ellenberg sector_t new_size, cur_size; 3997daeda1ccSPhilipp Reisner rcu_read_lock(); 3998b30ab791SAndreas Gruenbacher my_usize = rcu_dereference(device->ldev->disk_conf)->disk_size; 3999daeda1ccSPhilipp Reisner rcu_read_unlock(); 4000daeda1ccSPhilipp Reisner 4001b30ab791SAndreas Gruenbacher warn_if_differ_considerably(device, "lower level device sizes", 4002b30ab791SAndreas Gruenbacher p_size, drbd_get_max_capacity(device->ldev)); 4003b30ab791SAndreas Gruenbacher warn_if_differ_considerably(device, "user requested size", 4004daeda1ccSPhilipp Reisner p_usize, my_usize); 4005b411b363SPhilipp Reisner 4006b411b363SPhilipp Reisner /* if this is the first connect, or an otherwise expected 4007b411b363SPhilipp Reisner * param exchange, choose the minimum */ 4008b30ab791SAndreas Gruenbacher if (device->state.conn == C_WF_REPORT_PARAMS) 4009daeda1ccSPhilipp Reisner p_usize = min_not_zero(my_usize, p_usize); 4010b411b363SPhilipp Reisner 4011b411b363SPhilipp Reisner /* Never shrink a device with usable data during connect. 4012b411b363SPhilipp Reisner But allow online shrinking if we are connected. */ 401360bac040SLars Ellenberg new_size = drbd_new_dev_size(device, device->ldev, p_usize, 0); 401460bac040SLars Ellenberg cur_size = drbd_get_capacity(device->this_bdev); 401560bac040SLars Ellenberg if (new_size < cur_size && 4016b30ab791SAndreas Gruenbacher device->state.disk >= D_OUTDATED && 4017b30ab791SAndreas Gruenbacher device->state.conn < C_CONNECTED) { 401860bac040SLars Ellenberg drbd_err(device, "The peer's disk size is too small! (%llu < %llu sectors)\n", 401960bac040SLars Ellenberg (unsigned long long)new_size, (unsigned long long)cur_size); 40209f4fe9adSAndreas Gruenbacher conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); 4021b30ab791SAndreas Gruenbacher put_ldev(device); 402282bc0194SAndreas Gruenbacher return -EIO; 4023b411b363SPhilipp Reisner } 4024daeda1ccSPhilipp Reisner 4025daeda1ccSPhilipp Reisner if (my_usize != p_usize) { 4026daeda1ccSPhilipp Reisner struct disk_conf *old_disk_conf, *new_disk_conf = NULL; 4027daeda1ccSPhilipp Reisner 4028daeda1ccSPhilipp Reisner new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); 4029daeda1ccSPhilipp Reisner if (!new_disk_conf) { 4030d0180171SAndreas Gruenbacher drbd_err(device, "Allocation of new disk_conf failed\n"); 4031b30ab791SAndreas Gruenbacher put_ldev(device); 4032daeda1ccSPhilipp Reisner return -ENOMEM; 4033daeda1ccSPhilipp Reisner } 4034daeda1ccSPhilipp Reisner 40350500813fSAndreas Gruenbacher mutex_lock(&connection->resource->conf_update); 4036b30ab791SAndreas Gruenbacher old_disk_conf = device->ldev->disk_conf; 4037daeda1ccSPhilipp Reisner *new_disk_conf = *old_disk_conf; 4038daeda1ccSPhilipp Reisner new_disk_conf->disk_size = p_usize; 4039daeda1ccSPhilipp Reisner 4040b30ab791SAndreas Gruenbacher rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); 40410500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 4042daeda1ccSPhilipp Reisner synchronize_rcu(); 4043daeda1ccSPhilipp Reisner kfree(old_disk_conf); 4044daeda1ccSPhilipp Reisner 4045d0180171SAndreas Gruenbacher drbd_info(device, "Peer sets u_size to %lu sectors\n", 4046daeda1ccSPhilipp Reisner (unsigned long)my_usize); 4047daeda1ccSPhilipp Reisner } 4048daeda1ccSPhilipp Reisner 4049b30ab791SAndreas Gruenbacher put_ldev(device); 4050b411b363SPhilipp Reisner } 4051b411b363SPhilipp Reisner 405220c68fdeSLars Ellenberg device->peer_max_bio_size = be32_to_cpu(p->max_bio_size); 4053dd4f699dSLars Ellenberg /* Leave drbd_reconsider_queue_parameters() before drbd_determine_dev_size(). 405420c68fdeSLars Ellenberg In case we cleared the QUEUE_FLAG_DISCARD from our queue in 4055dd4f699dSLars Ellenberg drbd_reconsider_queue_parameters(), we can be sure that after 405620c68fdeSLars Ellenberg drbd_determine_dev_size() no REQ_DISCARDs are in the queue. */ 405720c68fdeSLars Ellenberg 4058e89b591cSPhilipp Reisner ddsf = be16_to_cpu(p->dds_flags); 4059b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 40609104d31aSLars Ellenberg drbd_reconsider_queue_parameters(device, device->ldev, o); 4061b30ab791SAndreas Gruenbacher dd = drbd_determine_dev_size(device, ddsf, NULL); 4062b30ab791SAndreas Gruenbacher put_ldev(device); 4063e96c9633SPhilipp Reisner if (dd == DS_ERROR) 406482bc0194SAndreas Gruenbacher return -EIO; 4065b30ab791SAndreas Gruenbacher drbd_md_sync(device); 4066b411b363SPhilipp Reisner } else { 40676a8d68b1SLars Ellenberg /* 40686a8d68b1SLars Ellenberg * I am diskless, need to accept the peer's *current* size. 40696a8d68b1SLars Ellenberg * I must NOT accept the peers backing disk size, 40706a8d68b1SLars Ellenberg * it may have been larger than mine all along... 40716a8d68b1SLars Ellenberg * 40726a8d68b1SLars Ellenberg * At this point, the peer knows more about my disk, or at 40736a8d68b1SLars Ellenberg * least about what we last agreed upon, than myself. 40746a8d68b1SLars Ellenberg * So if his c_size is less than his d_size, the most likely 40756a8d68b1SLars Ellenberg * reason is that *my* d_size was smaller last time we checked. 40766a8d68b1SLars Ellenberg * 40776a8d68b1SLars Ellenberg * However, if he sends a zero current size, 40786a8d68b1SLars Ellenberg * take his (user-capped or) backing disk size anyways. 40796a8d68b1SLars Ellenberg */ 40809104d31aSLars Ellenberg drbd_reconsider_queue_parameters(device, NULL, o); 40816a8d68b1SLars Ellenberg drbd_set_my_capacity(device, p_csize ?: p_usize ?: p_size); 4082b411b363SPhilipp Reisner } 4083b411b363SPhilipp Reisner 4084b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 4085b30ab791SAndreas Gruenbacher if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev)) { 4086b30ab791SAndreas Gruenbacher device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev); 4087b411b363SPhilipp Reisner ldsc = 1; 4088b411b363SPhilipp Reisner } 4089b411b363SPhilipp Reisner 4090b30ab791SAndreas Gruenbacher put_ldev(device); 4091b411b363SPhilipp Reisner } 4092b411b363SPhilipp Reisner 4093b30ab791SAndreas Gruenbacher if (device->state.conn > C_WF_REPORT_PARAMS) { 4094b411b363SPhilipp Reisner if (be64_to_cpu(p->c_size) != 4095b30ab791SAndreas Gruenbacher drbd_get_capacity(device->this_bdev) || ldsc) { 4096b411b363SPhilipp Reisner /* we have different sizes, probably peer 4097b411b363SPhilipp Reisner * needs to know my new size... */ 409869a22773SAndreas Gruenbacher drbd_send_sizes(peer_device, 0, ddsf); 4099b411b363SPhilipp Reisner } 4100b30ab791SAndreas Gruenbacher if (test_and_clear_bit(RESIZE_PENDING, &device->flags) || 4101b30ab791SAndreas Gruenbacher (dd == DS_GREW && device->state.conn == C_CONNECTED)) { 4102b30ab791SAndreas Gruenbacher if (device->state.pdsk >= D_INCONSISTENT && 4103b30ab791SAndreas Gruenbacher device->state.disk >= D_INCONSISTENT) { 4104e89b591cSPhilipp Reisner if (ddsf & DDSF_NO_RESYNC) 4105d0180171SAndreas Gruenbacher drbd_info(device, "Resync of new storage suppressed with --assume-clean\n"); 4106b411b363SPhilipp Reisner else 4107b30ab791SAndreas Gruenbacher resync_after_online_grow(device); 4108e89b591cSPhilipp Reisner } else 4109b30ab791SAndreas Gruenbacher set_bit(RESYNC_AFTER_NEG, &device->flags); 4110b411b363SPhilipp Reisner } 4111b411b363SPhilipp Reisner } 4112b411b363SPhilipp Reisner 411382bc0194SAndreas Gruenbacher return 0; 4114b411b363SPhilipp Reisner } 4115b411b363SPhilipp Reisner 4116bde89a9eSAndreas Gruenbacher static int receive_uuids(struct drbd_connection *connection, struct packet_info *pi) 4117b411b363SPhilipp Reisner { 41189f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 4119b30ab791SAndreas Gruenbacher struct drbd_device *device; 4120e658983aSAndreas Gruenbacher struct p_uuids *p = pi->data; 4121b411b363SPhilipp Reisner u64 *p_uuid; 412262b0da3aSLars Ellenberg int i, updated_uuids = 0; 4123b411b363SPhilipp Reisner 41249f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 41259f4fe9adSAndreas Gruenbacher if (!peer_device) 4126bde89a9eSAndreas Gruenbacher return config_unknown_volume(connection, pi); 41279f4fe9adSAndreas Gruenbacher device = peer_device->device; 41284a76b161SAndreas Gruenbacher 4129b411b363SPhilipp Reisner p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); 4130063eacf8SJing Wang if (!p_uuid) { 4131d0180171SAndreas Gruenbacher drbd_err(device, "kmalloc of p_uuid failed\n"); 4132063eacf8SJing Wang return false; 4133063eacf8SJing Wang } 4134b411b363SPhilipp Reisner 4135b411b363SPhilipp Reisner for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) 4136b411b363SPhilipp Reisner p_uuid[i] = be64_to_cpu(p->uuid[i]); 4137b411b363SPhilipp Reisner 4138b30ab791SAndreas Gruenbacher kfree(device->p_uuid); 4139b30ab791SAndreas Gruenbacher device->p_uuid = p_uuid; 4140b411b363SPhilipp Reisner 4141b30ab791SAndreas Gruenbacher if (device->state.conn < C_CONNECTED && 4142b30ab791SAndreas Gruenbacher device->state.disk < D_INCONSISTENT && 4143b30ab791SAndreas Gruenbacher device->state.role == R_PRIMARY && 4144b30ab791SAndreas Gruenbacher (device->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { 4145d0180171SAndreas Gruenbacher drbd_err(device, "Can only connect to data with current UUID=%016llX\n", 4146b30ab791SAndreas Gruenbacher (unsigned long long)device->ed_uuid); 41479f4fe9adSAndreas Gruenbacher conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); 414882bc0194SAndreas Gruenbacher return -EIO; 4149b411b363SPhilipp Reisner } 4150b411b363SPhilipp Reisner 4151b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 4152b411b363SPhilipp Reisner int skip_initial_sync = 4153b30ab791SAndreas Gruenbacher device->state.conn == C_CONNECTED && 41549f4fe9adSAndreas Gruenbacher peer_device->connection->agreed_pro_version >= 90 && 4155b30ab791SAndreas Gruenbacher device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && 4156b411b363SPhilipp Reisner (p_uuid[UI_FLAGS] & 8); 4157b411b363SPhilipp Reisner if (skip_initial_sync) { 4158d0180171SAndreas Gruenbacher drbd_info(device, "Accepted new current UUID, preparing to skip initial sync\n"); 4159b30ab791SAndreas Gruenbacher drbd_bitmap_io(device, &drbd_bmio_clear_n_write, 416020ceb2b2SLars Ellenberg "clear_n_write from receive_uuids", 416120ceb2b2SLars Ellenberg BM_LOCKED_TEST_ALLOWED); 4162b30ab791SAndreas Gruenbacher _drbd_uuid_set(device, UI_CURRENT, p_uuid[UI_CURRENT]); 4163b30ab791SAndreas Gruenbacher _drbd_uuid_set(device, UI_BITMAP, 0); 4164b30ab791SAndreas Gruenbacher _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), 4165b411b363SPhilipp Reisner CS_VERBOSE, NULL); 4166b30ab791SAndreas Gruenbacher drbd_md_sync(device); 416762b0da3aSLars Ellenberg updated_uuids = 1; 4168b411b363SPhilipp Reisner } 4169b30ab791SAndreas Gruenbacher put_ldev(device); 4170b30ab791SAndreas Gruenbacher } else if (device->state.disk < D_INCONSISTENT && 4171b30ab791SAndreas Gruenbacher device->state.role == R_PRIMARY) { 417218a50fa2SPhilipp Reisner /* I am a diskless primary, the peer just created a new current UUID 417318a50fa2SPhilipp Reisner for me. */ 4174b30ab791SAndreas Gruenbacher updated_uuids = drbd_set_ed_uuid(device, p_uuid[UI_CURRENT]); 4175b411b363SPhilipp Reisner } 4176b411b363SPhilipp Reisner 4177b411b363SPhilipp Reisner /* Before we test for the disk state, we should wait until an eventually 4178b411b363SPhilipp Reisner ongoing cluster wide state change is finished. That is important if 4179b411b363SPhilipp Reisner we are primary and are detaching from our disk. We need to see the 4180b411b363SPhilipp Reisner new disk state... */ 4181b30ab791SAndreas Gruenbacher mutex_lock(device->state_mutex); 4182b30ab791SAndreas Gruenbacher mutex_unlock(device->state_mutex); 4183b30ab791SAndreas Gruenbacher if (device->state.conn >= C_CONNECTED && device->state.disk < D_INCONSISTENT) 4184b30ab791SAndreas Gruenbacher updated_uuids |= drbd_set_ed_uuid(device, p_uuid[UI_CURRENT]); 418562b0da3aSLars Ellenberg 418662b0da3aSLars Ellenberg if (updated_uuids) 4187b30ab791SAndreas Gruenbacher drbd_print_uuids(device, "receiver updated UUIDs to"); 4188b411b363SPhilipp Reisner 418982bc0194SAndreas Gruenbacher return 0; 4190b411b363SPhilipp Reisner } 4191b411b363SPhilipp Reisner 4192b411b363SPhilipp Reisner /** 4193b411b363SPhilipp Reisner * convert_state() - Converts the peer's view of the cluster state to our point of view 4194b411b363SPhilipp Reisner * @ps: The state as seen by the peer. 4195b411b363SPhilipp Reisner */ 4196b411b363SPhilipp Reisner static union drbd_state convert_state(union drbd_state ps) 4197b411b363SPhilipp Reisner { 4198b411b363SPhilipp Reisner union drbd_state ms; 4199b411b363SPhilipp Reisner 4200b411b363SPhilipp Reisner static enum drbd_conns c_tab[] = { 4201369bea63SPhilipp Reisner [C_WF_REPORT_PARAMS] = C_WF_REPORT_PARAMS, 4202b411b363SPhilipp Reisner [C_CONNECTED] = C_CONNECTED, 4203b411b363SPhilipp Reisner 4204b411b363SPhilipp Reisner [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, 4205b411b363SPhilipp Reisner [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, 4206b411b363SPhilipp Reisner [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ 4207b411b363SPhilipp Reisner [C_VERIFY_S] = C_VERIFY_T, 4208b411b363SPhilipp Reisner [C_MASK] = C_MASK, 4209b411b363SPhilipp Reisner }; 4210b411b363SPhilipp Reisner 4211b411b363SPhilipp Reisner ms.i = ps.i; 4212b411b363SPhilipp Reisner 4213b411b363SPhilipp Reisner ms.conn = c_tab[ps.conn]; 4214b411b363SPhilipp Reisner ms.peer = ps.role; 4215b411b363SPhilipp Reisner ms.role = ps.peer; 4216b411b363SPhilipp Reisner ms.pdsk = ps.disk; 4217b411b363SPhilipp Reisner ms.disk = ps.pdsk; 4218b411b363SPhilipp Reisner ms.peer_isp = (ps.aftr_isp | ps.user_isp); 4219b411b363SPhilipp Reisner 4220b411b363SPhilipp Reisner return ms; 4221b411b363SPhilipp Reisner } 4222b411b363SPhilipp Reisner 4223bde89a9eSAndreas Gruenbacher static int receive_req_state(struct drbd_connection *connection, struct packet_info *pi) 4224b411b363SPhilipp Reisner { 42259f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 4226b30ab791SAndreas Gruenbacher struct drbd_device *device; 4227e658983aSAndreas Gruenbacher struct p_req_state *p = pi->data; 4228b411b363SPhilipp Reisner union drbd_state mask, val; 4229bf885f8aSAndreas Gruenbacher enum drbd_state_rv rv; 4230b411b363SPhilipp Reisner 42319f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 42329f4fe9adSAndreas Gruenbacher if (!peer_device) 42334a76b161SAndreas Gruenbacher return -EIO; 42349f4fe9adSAndreas Gruenbacher device = peer_device->device; 42354a76b161SAndreas Gruenbacher 4236b411b363SPhilipp Reisner mask.i = be32_to_cpu(p->mask); 4237b411b363SPhilipp Reisner val.i = be32_to_cpu(p->val); 4238b411b363SPhilipp Reisner 42399f4fe9adSAndreas Gruenbacher if (test_bit(RESOLVE_CONFLICTS, &peer_device->connection->flags) && 4240b30ab791SAndreas Gruenbacher mutex_is_locked(device->state_mutex)) { 424169a22773SAndreas Gruenbacher drbd_send_sr_reply(peer_device, SS_CONCURRENT_ST_CHG); 424282bc0194SAndreas Gruenbacher return 0; 4243b411b363SPhilipp Reisner } 4244b411b363SPhilipp Reisner 4245b411b363SPhilipp Reisner mask = convert_state(mask); 4246b411b363SPhilipp Reisner val = convert_state(val); 4247b411b363SPhilipp Reisner 4248b30ab791SAndreas Gruenbacher rv = drbd_change_state(device, CS_VERBOSE, mask, val); 424969a22773SAndreas Gruenbacher drbd_send_sr_reply(peer_device, rv); 4250047cd4a6SPhilipp Reisner 4251b30ab791SAndreas Gruenbacher drbd_md_sync(device); 4252b411b363SPhilipp Reisner 425382bc0194SAndreas Gruenbacher return 0; 4254b411b363SPhilipp Reisner } 4255b411b363SPhilipp Reisner 4256bde89a9eSAndreas Gruenbacher static int receive_req_conn_state(struct drbd_connection *connection, struct packet_info *pi) 4257b411b363SPhilipp Reisner { 4258e658983aSAndreas Gruenbacher struct p_req_state *p = pi->data; 4259dfafcc8aSPhilipp Reisner union drbd_state mask, val; 4260dfafcc8aSPhilipp Reisner enum drbd_state_rv rv; 4261dfafcc8aSPhilipp Reisner 4262dfafcc8aSPhilipp Reisner mask.i = be32_to_cpu(p->mask); 4263dfafcc8aSPhilipp Reisner val.i = be32_to_cpu(p->val); 4264dfafcc8aSPhilipp Reisner 4265bde89a9eSAndreas Gruenbacher if (test_bit(RESOLVE_CONFLICTS, &connection->flags) && 4266bde89a9eSAndreas Gruenbacher mutex_is_locked(&connection->cstate_mutex)) { 4267bde89a9eSAndreas Gruenbacher conn_send_sr_reply(connection, SS_CONCURRENT_ST_CHG); 426882bc0194SAndreas Gruenbacher return 0; 4269dfafcc8aSPhilipp Reisner } 4270dfafcc8aSPhilipp Reisner 4271dfafcc8aSPhilipp Reisner mask = convert_state(mask); 4272dfafcc8aSPhilipp Reisner val = convert_state(val); 4273dfafcc8aSPhilipp Reisner 4274bde89a9eSAndreas Gruenbacher rv = conn_request_state(connection, mask, val, CS_VERBOSE | CS_LOCAL_ONLY | CS_IGN_OUTD_FAIL); 4275bde89a9eSAndreas Gruenbacher conn_send_sr_reply(connection, rv); 4276dfafcc8aSPhilipp Reisner 427782bc0194SAndreas Gruenbacher return 0; 4278dfafcc8aSPhilipp Reisner } 4279dfafcc8aSPhilipp Reisner 4280bde89a9eSAndreas Gruenbacher static int receive_state(struct drbd_connection *connection, struct packet_info *pi) 4281b411b363SPhilipp Reisner { 42829f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 4283b30ab791SAndreas Gruenbacher struct drbd_device *device; 4284e658983aSAndreas Gruenbacher struct p_state *p = pi->data; 42854ac4aadaSLars Ellenberg union drbd_state os, ns, peer_state; 4286b411b363SPhilipp Reisner enum drbd_disk_state real_peer_disk; 428765d922c3SPhilipp Reisner enum chg_state_flags cs_flags; 4288b411b363SPhilipp Reisner int rv; 4289b411b363SPhilipp Reisner 42909f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 42919f4fe9adSAndreas Gruenbacher if (!peer_device) 4292bde89a9eSAndreas Gruenbacher return config_unknown_volume(connection, pi); 42939f4fe9adSAndreas Gruenbacher device = peer_device->device; 42944a76b161SAndreas Gruenbacher 4295b411b363SPhilipp Reisner peer_state.i = be32_to_cpu(p->state); 4296b411b363SPhilipp Reisner 4297b411b363SPhilipp Reisner real_peer_disk = peer_state.disk; 4298b411b363SPhilipp Reisner if (peer_state.disk == D_NEGOTIATING) { 4299b30ab791SAndreas Gruenbacher real_peer_disk = device->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; 4300d0180171SAndreas Gruenbacher drbd_info(device, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); 4301b411b363SPhilipp Reisner } 4302b411b363SPhilipp Reisner 43030500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 4304b411b363SPhilipp Reisner retry: 4305b30ab791SAndreas Gruenbacher os = ns = drbd_read_state(device); 43060500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 4307b411b363SPhilipp Reisner 4308668700b4SPhilipp Reisner /* If some other part of the code (ack_receiver thread, timeout) 4309545752d5SLars Ellenberg * already decided to close the connection again, 4310545752d5SLars Ellenberg * we must not "re-establish" it here. */ 4311545752d5SLars Ellenberg if (os.conn <= C_TEAR_DOWN) 431258ffa580SLars Ellenberg return -ECONNRESET; 4313545752d5SLars Ellenberg 431440424e4aSLars Ellenberg /* If this is the "end of sync" confirmation, usually the peer disk 431540424e4aSLars Ellenberg * transitions from D_INCONSISTENT to D_UP_TO_DATE. For empty (0 bits 431640424e4aSLars Ellenberg * set) resync started in PausedSyncT, or if the timing of pause-/ 431740424e4aSLars Ellenberg * unpause-sync events has been "just right", the peer disk may 431840424e4aSLars Ellenberg * transition from D_CONSISTENT to D_UP_TO_DATE as well. 431940424e4aSLars Ellenberg */ 432040424e4aSLars Ellenberg if ((os.pdsk == D_INCONSISTENT || os.pdsk == D_CONSISTENT) && 432140424e4aSLars Ellenberg real_peer_disk == D_UP_TO_DATE && 4322e9ef7bb6SLars Ellenberg os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { 4323e9ef7bb6SLars Ellenberg /* If we are (becoming) SyncSource, but peer is still in sync 4324e9ef7bb6SLars Ellenberg * preparation, ignore its uptodate-ness to avoid flapping, it 4325e9ef7bb6SLars Ellenberg * will change to inconsistent once the peer reaches active 4326e9ef7bb6SLars Ellenberg * syncing states. 4327e9ef7bb6SLars Ellenberg * It may have changed syncer-paused flags, however, so we 4328e9ef7bb6SLars Ellenberg * cannot ignore this completely. */ 4329e9ef7bb6SLars Ellenberg if (peer_state.conn > C_CONNECTED && 4330e9ef7bb6SLars Ellenberg peer_state.conn < C_SYNC_SOURCE) 4331e9ef7bb6SLars Ellenberg real_peer_disk = D_INCONSISTENT; 4332e9ef7bb6SLars Ellenberg 4333e9ef7bb6SLars Ellenberg /* if peer_state changes to connected at the same time, 4334e9ef7bb6SLars Ellenberg * it explicitly notifies us that it finished resync. 4335e9ef7bb6SLars Ellenberg * Maybe we should finish it up, too? */ 4336e9ef7bb6SLars Ellenberg else if (os.conn >= C_SYNC_SOURCE && 4337e9ef7bb6SLars Ellenberg peer_state.conn == C_CONNECTED) { 4338b30ab791SAndreas Gruenbacher if (drbd_bm_total_weight(device) <= device->rs_failed) 4339b30ab791SAndreas Gruenbacher drbd_resync_finished(device); 434082bc0194SAndreas Gruenbacher return 0; 4341e9ef7bb6SLars Ellenberg } 4342e9ef7bb6SLars Ellenberg } 4343e9ef7bb6SLars Ellenberg 434402b91b55SLars Ellenberg /* explicit verify finished notification, stop sector reached. */ 434502b91b55SLars Ellenberg if (os.conn == C_VERIFY_T && os.disk == D_UP_TO_DATE && 434602b91b55SLars Ellenberg peer_state.conn == C_CONNECTED && real_peer_disk == D_UP_TO_DATE) { 4347b30ab791SAndreas Gruenbacher ov_out_of_sync_print(device); 4348b30ab791SAndreas Gruenbacher drbd_resync_finished(device); 434958ffa580SLars Ellenberg return 0; 435002b91b55SLars Ellenberg } 435102b91b55SLars Ellenberg 4352e9ef7bb6SLars Ellenberg /* peer says his disk is inconsistent, while we think it is uptodate, 4353e9ef7bb6SLars Ellenberg * and this happens while the peer still thinks we have a sync going on, 4354e9ef7bb6SLars Ellenberg * but we think we are already done with the sync. 4355e9ef7bb6SLars Ellenberg * We ignore this to avoid flapping pdsk. 4356e9ef7bb6SLars Ellenberg * This should not happen, if the peer is a recent version of drbd. */ 4357e9ef7bb6SLars Ellenberg if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && 4358e9ef7bb6SLars Ellenberg os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) 4359e9ef7bb6SLars Ellenberg real_peer_disk = D_UP_TO_DATE; 4360e9ef7bb6SLars Ellenberg 43614ac4aadaSLars Ellenberg if (ns.conn == C_WF_REPORT_PARAMS) 43624ac4aadaSLars Ellenberg ns.conn = C_CONNECTED; 4363b411b363SPhilipp Reisner 436467531718SPhilipp Reisner if (peer_state.conn == C_AHEAD) 436567531718SPhilipp Reisner ns.conn = C_BEHIND; 436667531718SPhilipp Reisner 4367b30ab791SAndreas Gruenbacher if (device->p_uuid && peer_state.disk >= D_NEGOTIATING && 4368b30ab791SAndreas Gruenbacher get_ldev_if_state(device, D_NEGOTIATING)) { 4369b411b363SPhilipp Reisner int cr; /* consider resync */ 4370b411b363SPhilipp Reisner 4371b411b363SPhilipp Reisner /* if we established a new connection */ 43724ac4aadaSLars Ellenberg cr = (os.conn < C_CONNECTED); 4373b411b363SPhilipp Reisner /* if we had an established connection 4374b411b363SPhilipp Reisner * and one of the nodes newly attaches a disk */ 43754ac4aadaSLars Ellenberg cr |= (os.conn == C_CONNECTED && 4376b411b363SPhilipp Reisner (peer_state.disk == D_NEGOTIATING || 43774ac4aadaSLars Ellenberg os.disk == D_NEGOTIATING)); 4378b411b363SPhilipp Reisner /* if we have both been inconsistent, and the peer has been 4379b411b363SPhilipp Reisner * forced to be UpToDate with --overwrite-data */ 4380b30ab791SAndreas Gruenbacher cr |= test_bit(CONSIDER_RESYNC, &device->flags); 4381b411b363SPhilipp Reisner /* if we had been plain connected, and the admin requested to 4382b411b363SPhilipp Reisner * start a sync by "invalidate" or "invalidate-remote" */ 43834ac4aadaSLars Ellenberg cr |= (os.conn == C_CONNECTED && 4384b411b363SPhilipp Reisner (peer_state.conn >= C_STARTING_SYNC_S && 4385b411b363SPhilipp Reisner peer_state.conn <= C_WF_BITMAP_T)); 4386b411b363SPhilipp Reisner 4387b411b363SPhilipp Reisner if (cr) 438869a22773SAndreas Gruenbacher ns.conn = drbd_sync_handshake(peer_device, peer_state.role, real_peer_disk); 4389b411b363SPhilipp Reisner 4390b30ab791SAndreas Gruenbacher put_ldev(device); 43914ac4aadaSLars Ellenberg if (ns.conn == C_MASK) { 43924ac4aadaSLars Ellenberg ns.conn = C_CONNECTED; 4393b30ab791SAndreas Gruenbacher if (device->state.disk == D_NEGOTIATING) { 4394b30ab791SAndreas Gruenbacher drbd_force_state(device, NS(disk, D_FAILED)); 4395b411b363SPhilipp Reisner } else if (peer_state.disk == D_NEGOTIATING) { 4396d0180171SAndreas Gruenbacher drbd_err(device, "Disk attach process on the peer node was aborted.\n"); 4397b411b363SPhilipp Reisner peer_state.disk = D_DISKLESS; 4398580b9767SLars Ellenberg real_peer_disk = D_DISKLESS; 4399b411b363SPhilipp Reisner } else { 44009f4fe9adSAndreas Gruenbacher if (test_and_clear_bit(CONN_DRY_RUN, &peer_device->connection->flags)) 440182bc0194SAndreas Gruenbacher return -EIO; 44020b0ba1efSAndreas Gruenbacher D_ASSERT(device, os.conn == C_WF_REPORT_PARAMS); 44039f4fe9adSAndreas Gruenbacher conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); 440482bc0194SAndreas Gruenbacher return -EIO; 4405b411b363SPhilipp Reisner } 4406b411b363SPhilipp Reisner } 4407b411b363SPhilipp Reisner } 4408b411b363SPhilipp Reisner 44090500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 4410b30ab791SAndreas Gruenbacher if (os.i != drbd_read_state(device).i) 4411b411b363SPhilipp Reisner goto retry; 4412b30ab791SAndreas Gruenbacher clear_bit(CONSIDER_RESYNC, &device->flags); 4413b411b363SPhilipp Reisner ns.peer = peer_state.role; 4414b411b363SPhilipp Reisner ns.pdsk = real_peer_disk; 4415b411b363SPhilipp Reisner ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); 44164ac4aadaSLars Ellenberg if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) 4417b30ab791SAndreas Gruenbacher ns.disk = device->new_state_tmp.disk; 44184ac4aadaSLars Ellenberg cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); 4419b30ab791SAndreas Gruenbacher if (ns.pdsk == D_CONSISTENT && drbd_suspended(device) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && 4420b30ab791SAndreas Gruenbacher test_bit(NEW_CUR_UUID, &device->flags)) { 44218554df1cSAndreas Gruenbacher /* Do not allow tl_restart(RESEND) for a rebooted peer. We can only allow this 4422481c6f50SPhilipp Reisner for temporal network outages! */ 44230500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 4424d0180171SAndreas Gruenbacher drbd_err(device, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); 44259f4fe9adSAndreas Gruenbacher tl_clear(peer_device->connection); 4426b30ab791SAndreas Gruenbacher drbd_uuid_new_current(device); 4427b30ab791SAndreas Gruenbacher clear_bit(NEW_CUR_UUID, &device->flags); 44289f4fe9adSAndreas Gruenbacher conn_request_state(peer_device->connection, NS2(conn, C_PROTOCOL_ERROR, susp, 0), CS_HARD); 442982bc0194SAndreas Gruenbacher return -EIO; 4430481c6f50SPhilipp Reisner } 4431b30ab791SAndreas Gruenbacher rv = _drbd_set_state(device, ns, cs_flags, NULL); 4432b30ab791SAndreas Gruenbacher ns = drbd_read_state(device); 44330500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 4434b411b363SPhilipp Reisner 4435b411b363SPhilipp Reisner if (rv < SS_SUCCESS) { 44369f4fe9adSAndreas Gruenbacher conn_request_state(peer_device->connection, NS(conn, C_DISCONNECTING), CS_HARD); 443782bc0194SAndreas Gruenbacher return -EIO; 4438b411b363SPhilipp Reisner } 4439b411b363SPhilipp Reisner 44404ac4aadaSLars Ellenberg if (os.conn > C_WF_REPORT_PARAMS) { 44414ac4aadaSLars Ellenberg if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && 4442b411b363SPhilipp Reisner peer_state.disk != D_NEGOTIATING ) { 4443b411b363SPhilipp Reisner /* we want resync, peer has not yet decided to sync... */ 4444b411b363SPhilipp Reisner /* Nowadays only used when forcing a node into primary role and 4445b411b363SPhilipp Reisner setting its disk to UpToDate with that */ 444669a22773SAndreas Gruenbacher drbd_send_uuids(peer_device); 444769a22773SAndreas Gruenbacher drbd_send_current_state(peer_device); 4448b411b363SPhilipp Reisner } 4449b411b363SPhilipp Reisner } 4450b411b363SPhilipp Reisner 4451b30ab791SAndreas Gruenbacher clear_bit(DISCARD_MY_DATA, &device->flags); 4452b411b363SPhilipp Reisner 4453b30ab791SAndreas Gruenbacher drbd_md_sync(device); /* update connected indicator, la_size_sect, ... */ 4454b411b363SPhilipp Reisner 445582bc0194SAndreas Gruenbacher return 0; 4456b411b363SPhilipp Reisner } 4457b411b363SPhilipp Reisner 4458bde89a9eSAndreas Gruenbacher static int receive_sync_uuid(struct drbd_connection *connection, struct packet_info *pi) 4459b411b363SPhilipp Reisner { 44609f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 4461b30ab791SAndreas Gruenbacher struct drbd_device *device; 4462e658983aSAndreas Gruenbacher struct p_rs_uuid *p = pi->data; 44634a76b161SAndreas Gruenbacher 44649f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 44659f4fe9adSAndreas Gruenbacher if (!peer_device) 44664a76b161SAndreas Gruenbacher return -EIO; 44679f4fe9adSAndreas Gruenbacher device = peer_device->device; 4468b411b363SPhilipp Reisner 4469b30ab791SAndreas Gruenbacher wait_event(device->misc_wait, 4470b30ab791SAndreas Gruenbacher device->state.conn == C_WF_SYNC_UUID || 4471b30ab791SAndreas Gruenbacher device->state.conn == C_BEHIND || 4472b30ab791SAndreas Gruenbacher device->state.conn < C_CONNECTED || 4473b30ab791SAndreas Gruenbacher device->state.disk < D_NEGOTIATING); 4474b411b363SPhilipp Reisner 44750b0ba1efSAndreas Gruenbacher /* D_ASSERT(device, device->state.conn == C_WF_SYNC_UUID ); */ 4476b411b363SPhilipp Reisner 4477b411b363SPhilipp Reisner /* Here the _drbd_uuid_ functions are right, current should 4478b411b363SPhilipp Reisner _not_ be rotated into the history */ 4479b30ab791SAndreas Gruenbacher if (get_ldev_if_state(device, D_NEGOTIATING)) { 4480b30ab791SAndreas Gruenbacher _drbd_uuid_set(device, UI_CURRENT, be64_to_cpu(p->uuid)); 4481b30ab791SAndreas Gruenbacher _drbd_uuid_set(device, UI_BITMAP, 0UL); 4482b411b363SPhilipp Reisner 4483b30ab791SAndreas Gruenbacher drbd_print_uuids(device, "updated sync uuid"); 4484b30ab791SAndreas Gruenbacher drbd_start_resync(device, C_SYNC_TARGET); 4485b411b363SPhilipp Reisner 4486b30ab791SAndreas Gruenbacher put_ldev(device); 4487b411b363SPhilipp Reisner } else 4488d0180171SAndreas Gruenbacher drbd_err(device, "Ignoring SyncUUID packet!\n"); 4489b411b363SPhilipp Reisner 449082bc0194SAndreas Gruenbacher return 0; 4491b411b363SPhilipp Reisner } 4492b411b363SPhilipp Reisner 44932c46407dSAndreas Gruenbacher /** 44942c46407dSAndreas Gruenbacher * receive_bitmap_plain 44952c46407dSAndreas Gruenbacher * 44962c46407dSAndreas Gruenbacher * Return 0 when done, 1 when another iteration is needed, and a negative error 44972c46407dSAndreas Gruenbacher * code upon failure. 44982c46407dSAndreas Gruenbacher */ 44992c46407dSAndreas Gruenbacher static int 450069a22773SAndreas Gruenbacher receive_bitmap_plain(struct drbd_peer_device *peer_device, unsigned int size, 4501e658983aSAndreas Gruenbacher unsigned long *p, struct bm_xfer_ctx *c) 4502b411b363SPhilipp Reisner { 450350d0b1adSAndreas Gruenbacher unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - 450469a22773SAndreas Gruenbacher drbd_header_size(peer_device->connection); 4505e658983aSAndreas Gruenbacher unsigned int num_words = min_t(size_t, data_size / sizeof(*p), 450650d0b1adSAndreas Gruenbacher c->bm_words - c->word_offset); 4507e658983aSAndreas Gruenbacher unsigned int want = num_words * sizeof(*p); 45082c46407dSAndreas Gruenbacher int err; 4509b411b363SPhilipp Reisner 451050d0b1adSAndreas Gruenbacher if (want != size) { 451169a22773SAndreas Gruenbacher drbd_err(peer_device, "%s:want (%u) != size (%u)\n", __func__, want, size); 45122c46407dSAndreas Gruenbacher return -EIO; 4513b411b363SPhilipp Reisner } 4514b411b363SPhilipp Reisner if (want == 0) 45152c46407dSAndreas Gruenbacher return 0; 451669a22773SAndreas Gruenbacher err = drbd_recv_all(peer_device->connection, p, want); 451782bc0194SAndreas Gruenbacher if (err) 45182c46407dSAndreas Gruenbacher return err; 4519b411b363SPhilipp Reisner 452069a22773SAndreas Gruenbacher drbd_bm_merge_lel(peer_device->device, c->word_offset, num_words, p); 4521b411b363SPhilipp Reisner 4522b411b363SPhilipp Reisner c->word_offset += num_words; 4523b411b363SPhilipp Reisner c->bit_offset = c->word_offset * BITS_PER_LONG; 4524b411b363SPhilipp Reisner if (c->bit_offset > c->bm_bits) 4525b411b363SPhilipp Reisner c->bit_offset = c->bm_bits; 4526b411b363SPhilipp Reisner 45272c46407dSAndreas Gruenbacher return 1; 4528b411b363SPhilipp Reisner } 4529b411b363SPhilipp Reisner 4530a02d1240SAndreas Gruenbacher static enum drbd_bitmap_code dcbp_get_code(struct p_compressed_bm *p) 4531a02d1240SAndreas Gruenbacher { 4532a02d1240SAndreas Gruenbacher return (enum drbd_bitmap_code)(p->encoding & 0x0f); 4533a02d1240SAndreas Gruenbacher } 4534a02d1240SAndreas Gruenbacher 4535a02d1240SAndreas Gruenbacher static int dcbp_get_start(struct p_compressed_bm *p) 4536a02d1240SAndreas Gruenbacher { 4537a02d1240SAndreas Gruenbacher return (p->encoding & 0x80) != 0; 4538a02d1240SAndreas Gruenbacher } 4539a02d1240SAndreas Gruenbacher 4540a02d1240SAndreas Gruenbacher static int dcbp_get_pad_bits(struct p_compressed_bm *p) 4541a02d1240SAndreas Gruenbacher { 4542a02d1240SAndreas Gruenbacher return (p->encoding >> 4) & 0x7; 4543a02d1240SAndreas Gruenbacher } 4544a02d1240SAndreas Gruenbacher 45452c46407dSAndreas Gruenbacher /** 45462c46407dSAndreas Gruenbacher * recv_bm_rle_bits 45472c46407dSAndreas Gruenbacher * 45482c46407dSAndreas Gruenbacher * Return 0 when done, 1 when another iteration is needed, and a negative error 45492c46407dSAndreas Gruenbacher * code upon failure. 45502c46407dSAndreas Gruenbacher */ 45512c46407dSAndreas Gruenbacher static int 455269a22773SAndreas Gruenbacher recv_bm_rle_bits(struct drbd_peer_device *peer_device, 4553b411b363SPhilipp Reisner struct p_compressed_bm *p, 4554c6d25cfeSPhilipp Reisner struct bm_xfer_ctx *c, 4555c6d25cfeSPhilipp Reisner unsigned int len) 4556b411b363SPhilipp Reisner { 4557b411b363SPhilipp Reisner struct bitstream bs; 4558b411b363SPhilipp Reisner u64 look_ahead; 4559b411b363SPhilipp Reisner u64 rl; 4560b411b363SPhilipp Reisner u64 tmp; 4561b411b363SPhilipp Reisner unsigned long s = c->bit_offset; 4562b411b363SPhilipp Reisner unsigned long e; 4563a02d1240SAndreas Gruenbacher int toggle = dcbp_get_start(p); 4564b411b363SPhilipp Reisner int have; 4565b411b363SPhilipp Reisner int bits; 4566b411b363SPhilipp Reisner 4567a02d1240SAndreas Gruenbacher bitstream_init(&bs, p->code, len, dcbp_get_pad_bits(p)); 4568b411b363SPhilipp Reisner 4569b411b363SPhilipp Reisner bits = bitstream_get_bits(&bs, &look_ahead, 64); 4570b411b363SPhilipp Reisner if (bits < 0) 45712c46407dSAndreas Gruenbacher return -EIO; 4572b411b363SPhilipp Reisner 4573b411b363SPhilipp Reisner for (have = bits; have > 0; s += rl, toggle = !toggle) { 4574b411b363SPhilipp Reisner bits = vli_decode_bits(&rl, look_ahead); 4575b411b363SPhilipp Reisner if (bits <= 0) 45762c46407dSAndreas Gruenbacher return -EIO; 4577b411b363SPhilipp Reisner 4578b411b363SPhilipp Reisner if (toggle) { 4579b411b363SPhilipp Reisner e = s + rl -1; 4580b411b363SPhilipp Reisner if (e >= c->bm_bits) { 458169a22773SAndreas Gruenbacher drbd_err(peer_device, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); 45822c46407dSAndreas Gruenbacher return -EIO; 4583b411b363SPhilipp Reisner } 458469a22773SAndreas Gruenbacher _drbd_bm_set_bits(peer_device->device, s, e); 4585b411b363SPhilipp Reisner } 4586b411b363SPhilipp Reisner 4587b411b363SPhilipp Reisner if (have < bits) { 458869a22773SAndreas Gruenbacher drbd_err(peer_device, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", 4589b411b363SPhilipp Reisner have, bits, look_ahead, 4590b411b363SPhilipp Reisner (unsigned int)(bs.cur.b - p->code), 4591b411b363SPhilipp Reisner (unsigned int)bs.buf_len); 45922c46407dSAndreas Gruenbacher return -EIO; 4593b411b363SPhilipp Reisner } 4594d2da5b0cSLars Ellenberg /* if we consumed all 64 bits, assign 0; >> 64 is "undefined"; */ 4595d2da5b0cSLars Ellenberg if (likely(bits < 64)) 4596b411b363SPhilipp Reisner look_ahead >>= bits; 4597d2da5b0cSLars Ellenberg else 4598d2da5b0cSLars Ellenberg look_ahead = 0; 4599b411b363SPhilipp Reisner have -= bits; 4600b411b363SPhilipp Reisner 4601b411b363SPhilipp Reisner bits = bitstream_get_bits(&bs, &tmp, 64 - have); 4602b411b363SPhilipp Reisner if (bits < 0) 46032c46407dSAndreas Gruenbacher return -EIO; 4604b411b363SPhilipp Reisner look_ahead |= tmp << have; 4605b411b363SPhilipp Reisner have += bits; 4606b411b363SPhilipp Reisner } 4607b411b363SPhilipp Reisner 4608b411b363SPhilipp Reisner c->bit_offset = s; 4609b411b363SPhilipp Reisner bm_xfer_ctx_bit_to_word_offset(c); 4610b411b363SPhilipp Reisner 46112c46407dSAndreas Gruenbacher return (s != c->bm_bits); 4612b411b363SPhilipp Reisner } 4613b411b363SPhilipp Reisner 46142c46407dSAndreas Gruenbacher /** 46152c46407dSAndreas Gruenbacher * decode_bitmap_c 46162c46407dSAndreas Gruenbacher * 46172c46407dSAndreas Gruenbacher * Return 0 when done, 1 when another iteration is needed, and a negative error 46182c46407dSAndreas Gruenbacher * code upon failure. 46192c46407dSAndreas Gruenbacher */ 46202c46407dSAndreas Gruenbacher static int 462169a22773SAndreas Gruenbacher decode_bitmap_c(struct drbd_peer_device *peer_device, 4622b411b363SPhilipp Reisner struct p_compressed_bm *p, 4623c6d25cfeSPhilipp Reisner struct bm_xfer_ctx *c, 4624c6d25cfeSPhilipp Reisner unsigned int len) 4625b411b363SPhilipp Reisner { 4626a02d1240SAndreas Gruenbacher if (dcbp_get_code(p) == RLE_VLI_Bits) 462769a22773SAndreas Gruenbacher return recv_bm_rle_bits(peer_device, p, c, len - sizeof(*p)); 4628b411b363SPhilipp Reisner 4629b411b363SPhilipp Reisner /* other variants had been implemented for evaluation, 4630b411b363SPhilipp Reisner * but have been dropped as this one turned out to be "best" 4631b411b363SPhilipp Reisner * during all our tests. */ 4632b411b363SPhilipp Reisner 463369a22773SAndreas Gruenbacher drbd_err(peer_device, "receive_bitmap_c: unknown encoding %u\n", p->encoding); 463469a22773SAndreas Gruenbacher conn_request_state(peer_device->connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); 46352c46407dSAndreas Gruenbacher return -EIO; 4636b411b363SPhilipp Reisner } 4637b411b363SPhilipp Reisner 4638b30ab791SAndreas Gruenbacher void INFO_bm_xfer_stats(struct drbd_device *device, 4639b411b363SPhilipp Reisner const char *direction, struct bm_xfer_ctx *c) 4640b411b363SPhilipp Reisner { 4641b411b363SPhilipp Reisner /* what would it take to transfer it "plaintext" */ 4642a6b32bc3SAndreas Gruenbacher unsigned int header_size = drbd_header_size(first_peer_device(device)->connection); 464350d0b1adSAndreas Gruenbacher unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; 464450d0b1adSAndreas Gruenbacher unsigned int plain = 464550d0b1adSAndreas Gruenbacher header_size * (DIV_ROUND_UP(c->bm_words, data_size) + 1) + 464650d0b1adSAndreas Gruenbacher c->bm_words * sizeof(unsigned long); 464750d0b1adSAndreas Gruenbacher unsigned int total = c->bytes[0] + c->bytes[1]; 464850d0b1adSAndreas Gruenbacher unsigned int r; 4649b411b363SPhilipp Reisner 4650b411b363SPhilipp Reisner /* total can not be zero. but just in case: */ 4651b411b363SPhilipp Reisner if (total == 0) 4652b411b363SPhilipp Reisner return; 4653b411b363SPhilipp Reisner 4654b411b363SPhilipp Reisner /* don't report if not compressed */ 4655b411b363SPhilipp Reisner if (total >= plain) 4656b411b363SPhilipp Reisner return; 4657b411b363SPhilipp Reisner 4658b411b363SPhilipp Reisner /* total < plain. check for overflow, still */ 4659b411b363SPhilipp Reisner r = (total > UINT_MAX/1000) ? (total / (plain/1000)) 4660b411b363SPhilipp Reisner : (1000 * total / plain); 4661b411b363SPhilipp Reisner 4662b411b363SPhilipp Reisner if (r > 1000) 4663b411b363SPhilipp Reisner r = 1000; 4664b411b363SPhilipp Reisner 4665b411b363SPhilipp Reisner r = 1000 - r; 4666d0180171SAndreas Gruenbacher drbd_info(device, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " 4667b411b363SPhilipp Reisner "total %u; compression: %u.%u%%\n", 4668b411b363SPhilipp Reisner direction, 4669b411b363SPhilipp Reisner c->bytes[1], c->packets[1], 4670b411b363SPhilipp Reisner c->bytes[0], c->packets[0], 4671b411b363SPhilipp Reisner total, r/10, r % 10); 4672b411b363SPhilipp Reisner } 4673b411b363SPhilipp Reisner 4674b411b363SPhilipp Reisner /* Since we are processing the bitfield from lower addresses to higher, 4675b411b363SPhilipp Reisner it does not matter if the process it in 32 bit chunks or 64 bit 4676b411b363SPhilipp Reisner chunks as long as it is little endian. (Understand it as byte stream, 4677b411b363SPhilipp Reisner beginning with the lowest byte...) If we would use big endian 4678b411b363SPhilipp Reisner we would need to process it from the highest address to the lowest, 4679b411b363SPhilipp Reisner in order to be agnostic to the 32 vs 64 bits issue. 4680b411b363SPhilipp Reisner 4681b411b363SPhilipp Reisner returns 0 on failure, 1 if we successfully received it. */ 4682bde89a9eSAndreas Gruenbacher static int receive_bitmap(struct drbd_connection *connection, struct packet_info *pi) 4683b411b363SPhilipp Reisner { 46849f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 4685b30ab791SAndreas Gruenbacher struct drbd_device *device; 4686b411b363SPhilipp Reisner struct bm_xfer_ctx c; 46872c46407dSAndreas Gruenbacher int err; 46884a76b161SAndreas Gruenbacher 46899f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 46909f4fe9adSAndreas Gruenbacher if (!peer_device) 46914a76b161SAndreas Gruenbacher return -EIO; 46929f4fe9adSAndreas Gruenbacher device = peer_device->device; 4693b411b363SPhilipp Reisner 4694b30ab791SAndreas Gruenbacher drbd_bm_lock(device, "receive bitmap", BM_LOCKED_SET_ALLOWED); 469520ceb2b2SLars Ellenberg /* you are supposed to send additional out-of-sync information 469620ceb2b2SLars Ellenberg * if you actually set bits during this phase */ 4697b411b363SPhilipp Reisner 4698b411b363SPhilipp Reisner c = (struct bm_xfer_ctx) { 4699b30ab791SAndreas Gruenbacher .bm_bits = drbd_bm_bits(device), 4700b30ab791SAndreas Gruenbacher .bm_words = drbd_bm_words(device), 4701b411b363SPhilipp Reisner }; 4702b411b363SPhilipp Reisner 47032c46407dSAndreas Gruenbacher for(;;) { 4704e658983aSAndreas Gruenbacher if (pi->cmd == P_BITMAP) 470569a22773SAndreas Gruenbacher err = receive_bitmap_plain(peer_device, pi->size, pi->data, &c); 4706e658983aSAndreas Gruenbacher else if (pi->cmd == P_COMPRESSED_BITMAP) { 4707b411b363SPhilipp Reisner /* MAYBE: sanity check that we speak proto >= 90, 4708b411b363SPhilipp Reisner * and the feature is enabled! */ 4709e658983aSAndreas Gruenbacher struct p_compressed_bm *p = pi->data; 4710b411b363SPhilipp Reisner 4711bde89a9eSAndreas Gruenbacher if (pi->size > DRBD_SOCKET_BUFFER_SIZE - drbd_header_size(connection)) { 4712d0180171SAndreas Gruenbacher drbd_err(device, "ReportCBitmap packet too large\n"); 471382bc0194SAndreas Gruenbacher err = -EIO; 4714b411b363SPhilipp Reisner goto out; 4715b411b363SPhilipp Reisner } 4716e658983aSAndreas Gruenbacher if (pi->size <= sizeof(*p)) { 4717d0180171SAndreas Gruenbacher drbd_err(device, "ReportCBitmap packet too small (l:%u)\n", pi->size); 471882bc0194SAndreas Gruenbacher err = -EIO; 471978fcbdaeSAndreas Gruenbacher goto out; 4720b411b363SPhilipp Reisner } 47219f4fe9adSAndreas Gruenbacher err = drbd_recv_all(peer_device->connection, p, pi->size); 4722e658983aSAndreas Gruenbacher if (err) 4723e658983aSAndreas Gruenbacher goto out; 472469a22773SAndreas Gruenbacher err = decode_bitmap_c(peer_device, p, &c, pi->size); 4725b411b363SPhilipp Reisner } else { 4726d0180171SAndreas Gruenbacher drbd_warn(device, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", pi->cmd); 472782bc0194SAndreas Gruenbacher err = -EIO; 4728b411b363SPhilipp Reisner goto out; 4729b411b363SPhilipp Reisner } 4730b411b363SPhilipp Reisner 4731e2857216SAndreas Gruenbacher c.packets[pi->cmd == P_BITMAP]++; 4732bde89a9eSAndreas Gruenbacher c.bytes[pi->cmd == P_BITMAP] += drbd_header_size(connection) + pi->size; 4733b411b363SPhilipp Reisner 47342c46407dSAndreas Gruenbacher if (err <= 0) { 47352c46407dSAndreas Gruenbacher if (err < 0) 47362c46407dSAndreas Gruenbacher goto out; 4737b411b363SPhilipp Reisner break; 47382c46407dSAndreas Gruenbacher } 47399f4fe9adSAndreas Gruenbacher err = drbd_recv_header(peer_device->connection, pi); 474082bc0194SAndreas Gruenbacher if (err) 4741b411b363SPhilipp Reisner goto out; 47422c46407dSAndreas Gruenbacher } 4743b411b363SPhilipp Reisner 4744b30ab791SAndreas Gruenbacher INFO_bm_xfer_stats(device, "receive", &c); 4745b411b363SPhilipp Reisner 4746b30ab791SAndreas Gruenbacher if (device->state.conn == C_WF_BITMAP_T) { 4747de1f8e4aSAndreas Gruenbacher enum drbd_state_rv rv; 4748de1f8e4aSAndreas Gruenbacher 4749b30ab791SAndreas Gruenbacher err = drbd_send_bitmap(device); 475082bc0194SAndreas Gruenbacher if (err) 4751b411b363SPhilipp Reisner goto out; 4752b411b363SPhilipp Reisner /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ 4753b30ab791SAndreas Gruenbacher rv = _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); 47540b0ba1efSAndreas Gruenbacher D_ASSERT(device, rv == SS_SUCCESS); 4755b30ab791SAndreas Gruenbacher } else if (device->state.conn != C_WF_BITMAP_S) { 4756b411b363SPhilipp Reisner /* admin may have requested C_DISCONNECTING, 4757b411b363SPhilipp Reisner * other threads may have noticed network errors */ 4758d0180171SAndreas Gruenbacher drbd_info(device, "unexpected cstate (%s) in receive_bitmap\n", 4759b30ab791SAndreas Gruenbacher drbd_conn_str(device->state.conn)); 4760b411b363SPhilipp Reisner } 476182bc0194SAndreas Gruenbacher err = 0; 4762b411b363SPhilipp Reisner 4763b411b363SPhilipp Reisner out: 4764b30ab791SAndreas Gruenbacher drbd_bm_unlock(device); 4765b30ab791SAndreas Gruenbacher if (!err && device->state.conn == C_WF_BITMAP_S) 4766b30ab791SAndreas Gruenbacher drbd_start_resync(device, C_SYNC_SOURCE); 476782bc0194SAndreas Gruenbacher return err; 4768b411b363SPhilipp Reisner } 4769b411b363SPhilipp Reisner 4770bde89a9eSAndreas Gruenbacher static int receive_skip(struct drbd_connection *connection, struct packet_info *pi) 4771b411b363SPhilipp Reisner { 47721ec861ebSAndreas Gruenbacher drbd_warn(connection, "skipping unknown optional packet type %d, l: %d!\n", 4773e2857216SAndreas Gruenbacher pi->cmd, pi->size); 4774b411b363SPhilipp Reisner 4775bde89a9eSAndreas Gruenbacher return ignore_remaining_packet(connection, pi); 4776b411b363SPhilipp Reisner } 4777b411b363SPhilipp Reisner 4778bde89a9eSAndreas Gruenbacher static int receive_UnplugRemote(struct drbd_connection *connection, struct packet_info *pi) 4779b411b363SPhilipp Reisner { 4780b411b363SPhilipp Reisner /* Make sure we've acked all the TCP data associated 4781b411b363SPhilipp Reisner * with the data requests being unplugged */ 4782bde89a9eSAndreas Gruenbacher drbd_tcp_quickack(connection->data.socket); 4783b411b363SPhilipp Reisner 478482bc0194SAndreas Gruenbacher return 0; 4785b411b363SPhilipp Reisner } 4786b411b363SPhilipp Reisner 4787bde89a9eSAndreas Gruenbacher static int receive_out_of_sync(struct drbd_connection *connection, struct packet_info *pi) 478873a01a18SPhilipp Reisner { 47899f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 4790b30ab791SAndreas Gruenbacher struct drbd_device *device; 4791e658983aSAndreas Gruenbacher struct p_block_desc *p = pi->data; 47924a76b161SAndreas Gruenbacher 47939f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 47949f4fe9adSAndreas Gruenbacher if (!peer_device) 47954a76b161SAndreas Gruenbacher return -EIO; 47969f4fe9adSAndreas Gruenbacher device = peer_device->device; 479773a01a18SPhilipp Reisner 4798b30ab791SAndreas Gruenbacher switch (device->state.conn) { 4799f735e363SLars Ellenberg case C_WF_SYNC_UUID: 4800f735e363SLars Ellenberg case C_WF_BITMAP_T: 4801f735e363SLars Ellenberg case C_BEHIND: 4802f735e363SLars Ellenberg break; 4803f735e363SLars Ellenberg default: 4804d0180171SAndreas Gruenbacher drbd_err(device, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", 4805b30ab791SAndreas Gruenbacher drbd_conn_str(device->state.conn)); 4806f735e363SLars Ellenberg } 4807f735e363SLars Ellenberg 4808b30ab791SAndreas Gruenbacher drbd_set_out_of_sync(device, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); 480973a01a18SPhilipp Reisner 481082bc0194SAndreas Gruenbacher return 0; 481173a01a18SPhilipp Reisner } 481273a01a18SPhilipp Reisner 4813700ca8c0SPhilipp Reisner static int receive_rs_deallocated(struct drbd_connection *connection, struct packet_info *pi) 4814700ca8c0SPhilipp Reisner { 4815700ca8c0SPhilipp Reisner struct drbd_peer_device *peer_device; 4816700ca8c0SPhilipp Reisner struct p_block_desc *p = pi->data; 4817700ca8c0SPhilipp Reisner struct drbd_device *device; 4818700ca8c0SPhilipp Reisner sector_t sector; 4819700ca8c0SPhilipp Reisner int size, err = 0; 4820700ca8c0SPhilipp Reisner 4821700ca8c0SPhilipp Reisner peer_device = conn_peer_device(connection, pi->vnr); 4822700ca8c0SPhilipp Reisner if (!peer_device) 4823700ca8c0SPhilipp Reisner return -EIO; 4824700ca8c0SPhilipp Reisner device = peer_device->device; 4825700ca8c0SPhilipp Reisner 4826700ca8c0SPhilipp Reisner sector = be64_to_cpu(p->sector); 4827700ca8c0SPhilipp Reisner size = be32_to_cpu(p->blksize); 4828700ca8c0SPhilipp Reisner 4829700ca8c0SPhilipp Reisner dec_rs_pending(device); 4830700ca8c0SPhilipp Reisner 4831700ca8c0SPhilipp Reisner if (get_ldev(device)) { 4832700ca8c0SPhilipp Reisner struct drbd_peer_request *peer_req; 483345c21793SChristoph Hellwig const int op = REQ_OP_WRITE_ZEROES; 4834700ca8c0SPhilipp Reisner 4835700ca8c0SPhilipp Reisner peer_req = drbd_alloc_peer_req(peer_device, ID_SYNCER, sector, 48369104d31aSLars Ellenberg size, 0, GFP_NOIO); 4837700ca8c0SPhilipp Reisner if (!peer_req) { 4838700ca8c0SPhilipp Reisner put_ldev(device); 4839700ca8c0SPhilipp Reisner return -ENOMEM; 4840700ca8c0SPhilipp Reisner } 4841700ca8c0SPhilipp Reisner 4842700ca8c0SPhilipp Reisner peer_req->w.cb = e_end_resync_block; 4843700ca8c0SPhilipp Reisner peer_req->submit_jif = jiffies; 4844700ca8c0SPhilipp Reisner peer_req->flags |= EE_IS_TRIM; 4845700ca8c0SPhilipp Reisner 4846700ca8c0SPhilipp Reisner spin_lock_irq(&device->resource->req_lock); 4847700ca8c0SPhilipp Reisner list_add_tail(&peer_req->w.list, &device->sync_ee); 4848700ca8c0SPhilipp Reisner spin_unlock_irq(&device->resource->req_lock); 4849700ca8c0SPhilipp Reisner 4850700ca8c0SPhilipp Reisner atomic_add(pi->size >> 9, &device->rs_sect_ev); 4851700ca8c0SPhilipp Reisner err = drbd_submit_peer_request(device, peer_req, op, 0, DRBD_FAULT_RS_WR); 4852700ca8c0SPhilipp Reisner 4853700ca8c0SPhilipp Reisner if (err) { 4854700ca8c0SPhilipp Reisner spin_lock_irq(&device->resource->req_lock); 4855700ca8c0SPhilipp Reisner list_del(&peer_req->w.list); 4856700ca8c0SPhilipp Reisner spin_unlock_irq(&device->resource->req_lock); 4857700ca8c0SPhilipp Reisner 4858700ca8c0SPhilipp Reisner drbd_free_peer_req(device, peer_req); 4859700ca8c0SPhilipp Reisner put_ldev(device); 4860700ca8c0SPhilipp Reisner err = 0; 4861700ca8c0SPhilipp Reisner goto fail; 4862700ca8c0SPhilipp Reisner } 4863700ca8c0SPhilipp Reisner 4864700ca8c0SPhilipp Reisner inc_unacked(device); 4865700ca8c0SPhilipp Reisner 4866700ca8c0SPhilipp Reisner /* No put_ldev() here. Gets called in drbd_endio_write_sec_final(), 4867700ca8c0SPhilipp Reisner as well as drbd_rs_complete_io() */ 4868700ca8c0SPhilipp Reisner } else { 4869700ca8c0SPhilipp Reisner fail: 4870700ca8c0SPhilipp Reisner drbd_rs_complete_io(device, sector); 4871700ca8c0SPhilipp Reisner drbd_send_ack_ex(peer_device, P_NEG_ACK, sector, size, ID_SYNCER); 4872700ca8c0SPhilipp Reisner } 4873700ca8c0SPhilipp Reisner 4874700ca8c0SPhilipp Reisner atomic_add(size >> 9, &device->rs_sect_in); 4875700ca8c0SPhilipp Reisner 4876700ca8c0SPhilipp Reisner return err; 4877700ca8c0SPhilipp Reisner } 4878700ca8c0SPhilipp Reisner 487902918be2SPhilipp Reisner struct data_cmd { 488002918be2SPhilipp Reisner int expect_payload; 48819104d31aSLars Ellenberg unsigned int pkt_size; 4882bde89a9eSAndreas Gruenbacher int (*fn)(struct drbd_connection *, struct packet_info *); 4883b411b363SPhilipp Reisner }; 4884b411b363SPhilipp Reisner 488502918be2SPhilipp Reisner static struct data_cmd drbd_cmd_handler[] = { 488602918be2SPhilipp Reisner [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, 488702918be2SPhilipp Reisner [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, 488802918be2SPhilipp Reisner [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , 488902918be2SPhilipp Reisner [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , 4890e658983aSAndreas Gruenbacher [P_BITMAP] = { 1, 0, receive_bitmap } , 4891e658983aSAndreas Gruenbacher [P_COMPRESSED_BITMAP] = { 1, 0, receive_bitmap } , 4892e658983aSAndreas Gruenbacher [P_UNPLUG_REMOTE] = { 0, 0, receive_UnplugRemote }, 489302918be2SPhilipp Reisner [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, 489402918be2SPhilipp Reisner [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, 4895e658983aSAndreas Gruenbacher [P_SYNC_PARAM] = { 1, 0, receive_SyncParam }, 4896e658983aSAndreas Gruenbacher [P_SYNC_PARAM89] = { 1, 0, receive_SyncParam }, 489702918be2SPhilipp Reisner [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, 489802918be2SPhilipp Reisner [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, 489902918be2SPhilipp Reisner [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, 490002918be2SPhilipp Reisner [P_STATE] = { 0, sizeof(struct p_state), receive_state }, 490102918be2SPhilipp Reisner [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, 490202918be2SPhilipp Reisner [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, 490302918be2SPhilipp Reisner [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, 490402918be2SPhilipp Reisner [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, 490502918be2SPhilipp Reisner [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, 4906700ca8c0SPhilipp Reisner [P_RS_THIN_REQ] = { 0, sizeof(struct p_block_req), receive_DataRequest }, 490702918be2SPhilipp Reisner [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, 490873a01a18SPhilipp Reisner [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync }, 49094a76b161SAndreas Gruenbacher [P_CONN_ST_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_conn_state }, 4910036b17eaSPhilipp Reisner [P_PROTOCOL_UPDATE] = { 1, sizeof(struct p_protocol), receive_protocol }, 4911a0fb3c47SLars Ellenberg [P_TRIM] = { 0, sizeof(struct p_trim), receive_Data }, 4912700ca8c0SPhilipp Reisner [P_RS_DEALLOCATED] = { 0, sizeof(struct p_block_desc), receive_rs_deallocated }, 49139104d31aSLars Ellenberg [P_WSAME] = { 1, sizeof(struct p_wsame), receive_Data }, 491402918be2SPhilipp Reisner }; 491502918be2SPhilipp Reisner 4916bde89a9eSAndreas Gruenbacher static void drbdd(struct drbd_connection *connection) 4917b411b363SPhilipp Reisner { 491877351055SPhilipp Reisner struct packet_info pi; 491902918be2SPhilipp Reisner size_t shs; /* sub header size */ 492082bc0194SAndreas Gruenbacher int err; 4921b411b363SPhilipp Reisner 4922bde89a9eSAndreas Gruenbacher while (get_t_state(&connection->receiver) == RUNNING) { 49239104d31aSLars Ellenberg struct data_cmd const *cmd; 4924deebe195SAndreas Gruenbacher 4925bde89a9eSAndreas Gruenbacher drbd_thread_current_set_cpu(&connection->receiver); 4926c51a0ef3SLars Ellenberg update_receiver_timing_details(connection, drbd_recv_header_maybe_unplug); 4927c51a0ef3SLars Ellenberg if (drbd_recv_header_maybe_unplug(connection, &pi)) 492802918be2SPhilipp Reisner goto err_out; 492902918be2SPhilipp Reisner 4930deebe195SAndreas Gruenbacher cmd = &drbd_cmd_handler[pi.cmd]; 49314a76b161SAndreas Gruenbacher if (unlikely(pi.cmd >= ARRAY_SIZE(drbd_cmd_handler) || !cmd->fn)) { 49321ec861ebSAndreas Gruenbacher drbd_err(connection, "Unexpected data packet %s (0x%04x)", 49332fcb8f30SAndreas Gruenbacher cmdname(pi.cmd), pi.cmd); 493402918be2SPhilipp Reisner goto err_out; 49350b33a916SLars Ellenberg } 4936b411b363SPhilipp Reisner 4937e658983aSAndreas Gruenbacher shs = cmd->pkt_size; 49389104d31aSLars Ellenberg if (pi.cmd == P_SIZES && connection->agreed_features & DRBD_FF_WSAME) 49399104d31aSLars Ellenberg shs += sizeof(struct o_qlim); 4940e658983aSAndreas Gruenbacher if (pi.size > shs && !cmd->expect_payload) { 49411ec861ebSAndreas Gruenbacher drbd_err(connection, "No payload expected %s l:%d\n", 49422fcb8f30SAndreas Gruenbacher cmdname(pi.cmd), pi.size); 4943c13f7e1aSLars Ellenberg goto err_out; 4944c13f7e1aSLars Ellenberg } 49459104d31aSLars Ellenberg if (pi.size < shs) { 49469104d31aSLars Ellenberg drbd_err(connection, "%s: unexpected packet size, expected:%d received:%d\n", 49479104d31aSLars Ellenberg cmdname(pi.cmd), (int)shs, pi.size); 49489104d31aSLars Ellenberg goto err_out; 49499104d31aSLars Ellenberg } 4950c13f7e1aSLars Ellenberg 4951c13f7e1aSLars Ellenberg if (shs) { 4952944410e9SLars Ellenberg update_receiver_timing_details(connection, drbd_recv_all_warn); 4953bde89a9eSAndreas Gruenbacher err = drbd_recv_all_warn(connection, pi.data, shs); 4954a5c31904SAndreas Gruenbacher if (err) 495502918be2SPhilipp Reisner goto err_out; 4956e2857216SAndreas Gruenbacher pi.size -= shs; 4957b411b363SPhilipp Reisner } 495802918be2SPhilipp Reisner 4959944410e9SLars Ellenberg update_receiver_timing_details(connection, cmd->fn); 4960bde89a9eSAndreas Gruenbacher err = cmd->fn(connection, &pi); 49614a76b161SAndreas Gruenbacher if (err) { 49621ec861ebSAndreas Gruenbacher drbd_err(connection, "error receiving %s, e: %d l: %d!\n", 49639f5bdc33SAndreas Gruenbacher cmdname(pi.cmd), err, pi.size); 496402918be2SPhilipp Reisner goto err_out; 496502918be2SPhilipp Reisner } 496602918be2SPhilipp Reisner } 496782bc0194SAndreas Gruenbacher return; 496802918be2SPhilipp Reisner 496902918be2SPhilipp Reisner err_out: 4970bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); 4971b411b363SPhilipp Reisner } 4972b411b363SPhilipp Reisner 4973bde89a9eSAndreas Gruenbacher static void conn_disconnect(struct drbd_connection *connection) 4974f70b3511SPhilipp Reisner { 4975c06ece6bSAndreas Gruenbacher struct drbd_peer_device *peer_device; 4976bbeb641cSPhilipp Reisner enum drbd_conns oc; 4977376694a0SPhilipp Reisner int vnr; 4978f70b3511SPhilipp Reisner 4979bde89a9eSAndreas Gruenbacher if (connection->cstate == C_STANDALONE) 4980b411b363SPhilipp Reisner return; 4981b411b363SPhilipp Reisner 4982545752d5SLars Ellenberg /* We are about to start the cleanup after connection loss. 4983545752d5SLars Ellenberg * Make sure drbd_make_request knows about that. 4984545752d5SLars Ellenberg * Usually we should be in some network failure state already, 4985545752d5SLars Ellenberg * but just in case we are not, we fix it up here. 4986545752d5SLars Ellenberg */ 4987bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); 4988545752d5SLars Ellenberg 4989668700b4SPhilipp Reisner /* ack_receiver does not clean up anything. it must not interfere, either */ 49901c03e520SPhilipp Reisner drbd_thread_stop(&connection->ack_receiver); 4991668700b4SPhilipp Reisner if (connection->ack_sender) { 4992668700b4SPhilipp Reisner destroy_workqueue(connection->ack_sender); 4993668700b4SPhilipp Reisner connection->ack_sender = NULL; 4994668700b4SPhilipp Reisner } 4995bde89a9eSAndreas Gruenbacher drbd_free_sock(connection); 4996360cc740SPhilipp Reisner 4997c141ebdaSPhilipp Reisner rcu_read_lock(); 4998c06ece6bSAndreas Gruenbacher idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 4999c06ece6bSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 5000b30ab791SAndreas Gruenbacher kref_get(&device->kref); 5001c141ebdaSPhilipp Reisner rcu_read_unlock(); 500269a22773SAndreas Gruenbacher drbd_disconnected(peer_device); 5003c06ece6bSAndreas Gruenbacher kref_put(&device->kref, drbd_destroy_device); 5004c141ebdaSPhilipp Reisner rcu_read_lock(); 5005c141ebdaSPhilipp Reisner } 5006c141ebdaSPhilipp Reisner rcu_read_unlock(); 5007c141ebdaSPhilipp Reisner 5008bde89a9eSAndreas Gruenbacher if (!list_empty(&connection->current_epoch->list)) 50091ec861ebSAndreas Gruenbacher drbd_err(connection, "ASSERTION FAILED: connection->current_epoch->list not empty\n"); 501012038a3aSPhilipp Reisner /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ 5011bde89a9eSAndreas Gruenbacher atomic_set(&connection->current_epoch->epoch_size, 0); 5012bde89a9eSAndreas Gruenbacher connection->send.seen_any_write_yet = false; 501312038a3aSPhilipp Reisner 50141ec861ebSAndreas Gruenbacher drbd_info(connection, "Connection closed\n"); 5015360cc740SPhilipp Reisner 5016bde89a9eSAndreas Gruenbacher if (conn_highest_role(connection) == R_PRIMARY && conn_highest_pdsk(connection) >= D_UNKNOWN) 5017bde89a9eSAndreas Gruenbacher conn_try_outdate_peer_async(connection); 5018cb703454SPhilipp Reisner 50190500813fSAndreas Gruenbacher spin_lock_irq(&connection->resource->req_lock); 5020bde89a9eSAndreas Gruenbacher oc = connection->cstate; 5021bbeb641cSPhilipp Reisner if (oc >= C_UNCONNECTED) 5022bde89a9eSAndreas Gruenbacher _conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE); 5023bbeb641cSPhilipp Reisner 50240500813fSAndreas Gruenbacher spin_unlock_irq(&connection->resource->req_lock); 5025360cc740SPhilipp Reisner 5026f3dfa40aSLars Ellenberg if (oc == C_DISCONNECTING) 5027bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_STANDALONE), CS_VERBOSE | CS_HARD); 5028360cc740SPhilipp Reisner } 5029360cc740SPhilipp Reisner 503069a22773SAndreas Gruenbacher static int drbd_disconnected(struct drbd_peer_device *peer_device) 5031360cc740SPhilipp Reisner { 503269a22773SAndreas Gruenbacher struct drbd_device *device = peer_device->device; 5033360cc740SPhilipp Reisner unsigned int i; 5034b411b363SPhilipp Reisner 503585719573SPhilipp Reisner /* wait for current activity to cease. */ 50360500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 5037b30ab791SAndreas Gruenbacher _drbd_wait_ee_list_empty(device, &device->active_ee); 5038b30ab791SAndreas Gruenbacher _drbd_wait_ee_list_empty(device, &device->sync_ee); 5039b30ab791SAndreas Gruenbacher _drbd_wait_ee_list_empty(device, &device->read_ee); 50400500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 5041b411b363SPhilipp Reisner 5042b411b363SPhilipp Reisner /* We do not have data structures that would allow us to 5043b411b363SPhilipp Reisner * get the rs_pending_cnt down to 0 again. 5044b411b363SPhilipp Reisner * * On C_SYNC_TARGET we do not have any data structures describing 5045b411b363SPhilipp Reisner * the pending RSDataRequest's we have sent. 5046b411b363SPhilipp Reisner * * On C_SYNC_SOURCE there is no data structure that tracks 5047b411b363SPhilipp Reisner * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. 5048b411b363SPhilipp Reisner * And no, it is not the sum of the reference counts in the 5049b411b363SPhilipp Reisner * resync_LRU. The resync_LRU tracks the whole operation including 5050b411b363SPhilipp Reisner * the disk-IO, while the rs_pending_cnt only tracks the blocks 5051b411b363SPhilipp Reisner * on the fly. */ 5052b30ab791SAndreas Gruenbacher drbd_rs_cancel_all(device); 5053b30ab791SAndreas Gruenbacher device->rs_total = 0; 5054b30ab791SAndreas Gruenbacher device->rs_failed = 0; 5055b30ab791SAndreas Gruenbacher atomic_set(&device->rs_pending_cnt, 0); 5056b30ab791SAndreas Gruenbacher wake_up(&device->misc_wait); 5057b411b363SPhilipp Reisner 5058b30ab791SAndreas Gruenbacher del_timer_sync(&device->resync_timer); 5059b30ab791SAndreas Gruenbacher resync_timer_fn((unsigned long)device); 5060b411b363SPhilipp Reisner 5061b411b363SPhilipp Reisner /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, 5062b411b363SPhilipp Reisner * w_make_resync_request etc. which may still be on the worker queue 5063b411b363SPhilipp Reisner * to be "canceled" */ 5064b5043c5eSAndreas Gruenbacher drbd_flush_workqueue(&peer_device->connection->sender_work); 5065b411b363SPhilipp Reisner 5066b30ab791SAndreas Gruenbacher drbd_finish_peer_reqs(device); 5067b411b363SPhilipp Reisner 5068d10b4ea3SPhilipp Reisner /* This second workqueue flush is necessary, since drbd_finish_peer_reqs() 5069d10b4ea3SPhilipp Reisner might have issued a work again. The one before drbd_finish_peer_reqs() is 5070d10b4ea3SPhilipp Reisner necessary to reclain net_ee in drbd_finish_peer_reqs(). */ 5071b5043c5eSAndreas Gruenbacher drbd_flush_workqueue(&peer_device->connection->sender_work); 5072d10b4ea3SPhilipp Reisner 507308332d73SLars Ellenberg /* need to do it again, drbd_finish_peer_reqs() may have populated it 507408332d73SLars Ellenberg * again via drbd_try_clear_on_disk_bm(). */ 5075b30ab791SAndreas Gruenbacher drbd_rs_cancel_all(device); 5076b411b363SPhilipp Reisner 5077b30ab791SAndreas Gruenbacher kfree(device->p_uuid); 5078b30ab791SAndreas Gruenbacher device->p_uuid = NULL; 5079b411b363SPhilipp Reisner 5080b30ab791SAndreas Gruenbacher if (!drbd_suspended(device)) 508169a22773SAndreas Gruenbacher tl_clear(peer_device->connection); 5082b411b363SPhilipp Reisner 5083b30ab791SAndreas Gruenbacher drbd_md_sync(device); 5084b411b363SPhilipp Reisner 5085be115b69SLars Ellenberg if (get_ldev(device)) { 5086be115b69SLars Ellenberg drbd_bitmap_io(device, &drbd_bm_write_copy_pages, 5087be115b69SLars Ellenberg "write from disconnected", BM_LOCKED_CHANGE_ALLOWED); 5088be115b69SLars Ellenberg put_ldev(device); 5089be115b69SLars Ellenberg } 509020ceb2b2SLars Ellenberg 5091b411b363SPhilipp Reisner /* tcp_close and release of sendpage pages can be deferred. I don't 5092b411b363SPhilipp Reisner * want to use SO_LINGER, because apparently it can be deferred for 5093b411b363SPhilipp Reisner * more than 20 seconds (longest time I checked). 5094b411b363SPhilipp Reisner * 5095b411b363SPhilipp Reisner * Actually we don't care for exactly when the network stack does its 5096b411b363SPhilipp Reisner * put_page(), but release our reference on these pages right here. 5097b411b363SPhilipp Reisner */ 5098b30ab791SAndreas Gruenbacher i = drbd_free_peer_reqs(device, &device->net_ee); 5099b411b363SPhilipp Reisner if (i) 5100d0180171SAndreas Gruenbacher drbd_info(device, "net_ee not empty, killed %u entries\n", i); 5101b30ab791SAndreas Gruenbacher i = atomic_read(&device->pp_in_use_by_net); 5102435f0740SLars Ellenberg if (i) 5103d0180171SAndreas Gruenbacher drbd_info(device, "pp_in_use_by_net = %d, expected 0\n", i); 5104b30ab791SAndreas Gruenbacher i = atomic_read(&device->pp_in_use); 5105b411b363SPhilipp Reisner if (i) 5106d0180171SAndreas Gruenbacher drbd_info(device, "pp_in_use = %d, expected 0\n", i); 5107b411b363SPhilipp Reisner 51080b0ba1efSAndreas Gruenbacher D_ASSERT(device, list_empty(&device->read_ee)); 51090b0ba1efSAndreas Gruenbacher D_ASSERT(device, list_empty(&device->active_ee)); 51100b0ba1efSAndreas Gruenbacher D_ASSERT(device, list_empty(&device->sync_ee)); 51110b0ba1efSAndreas Gruenbacher D_ASSERT(device, list_empty(&device->done_ee)); 5112b411b363SPhilipp Reisner 5113360cc740SPhilipp Reisner return 0; 5114b411b363SPhilipp Reisner } 5115b411b363SPhilipp Reisner 5116b411b363SPhilipp Reisner /* 5117b411b363SPhilipp Reisner * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version 5118b411b363SPhilipp Reisner * we can agree on is stored in agreed_pro_version. 5119b411b363SPhilipp Reisner * 5120b411b363SPhilipp Reisner * feature flags and the reserved array should be enough room for future 5121b411b363SPhilipp Reisner * enhancements of the handshake protocol, and possible plugins... 5122b411b363SPhilipp Reisner * 5123b411b363SPhilipp Reisner * for now, they are expected to be zero, but ignored. 5124b411b363SPhilipp Reisner */ 5125bde89a9eSAndreas Gruenbacher static int drbd_send_features(struct drbd_connection *connection) 5126b411b363SPhilipp Reisner { 51279f5bdc33SAndreas Gruenbacher struct drbd_socket *sock; 51289f5bdc33SAndreas Gruenbacher struct p_connection_features *p; 5129b411b363SPhilipp Reisner 5130bde89a9eSAndreas Gruenbacher sock = &connection->data; 5131bde89a9eSAndreas Gruenbacher p = conn_prepare_command(connection, sock); 51329f5bdc33SAndreas Gruenbacher if (!p) 5133e8d17b01SAndreas Gruenbacher return -EIO; 5134b411b363SPhilipp Reisner memset(p, 0, sizeof(*p)); 5135b411b363SPhilipp Reisner p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); 5136b411b363SPhilipp Reisner p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); 513720c68fdeSLars Ellenberg p->feature_flags = cpu_to_be32(PRO_FEATURES); 5138bde89a9eSAndreas Gruenbacher return conn_send_command(connection, sock, P_CONNECTION_FEATURES, sizeof(*p), NULL, 0); 5139b411b363SPhilipp Reisner } 5140b411b363SPhilipp Reisner 5141b411b363SPhilipp Reisner /* 5142b411b363SPhilipp Reisner * return values: 5143b411b363SPhilipp Reisner * 1 yes, we have a valid connection 5144b411b363SPhilipp Reisner * 0 oops, did not work out, please try again 5145b411b363SPhilipp Reisner * -1 peer talks different language, 5146b411b363SPhilipp Reisner * no point in trying again, please go standalone. 5147b411b363SPhilipp Reisner */ 5148bde89a9eSAndreas Gruenbacher static int drbd_do_features(struct drbd_connection *connection) 5149b411b363SPhilipp Reisner { 5150bde89a9eSAndreas Gruenbacher /* ASSERT current == connection->receiver ... */ 5151e658983aSAndreas Gruenbacher struct p_connection_features *p; 5152e658983aSAndreas Gruenbacher const int expect = sizeof(struct p_connection_features); 515377351055SPhilipp Reisner struct packet_info pi; 5154a5c31904SAndreas Gruenbacher int err; 5155b411b363SPhilipp Reisner 5156bde89a9eSAndreas Gruenbacher err = drbd_send_features(connection); 5157e8d17b01SAndreas Gruenbacher if (err) 5158b411b363SPhilipp Reisner return 0; 5159b411b363SPhilipp Reisner 5160bde89a9eSAndreas Gruenbacher err = drbd_recv_header(connection, &pi); 516169bc7bc3SAndreas Gruenbacher if (err) 5162b411b363SPhilipp Reisner return 0; 5163b411b363SPhilipp Reisner 51646038178eSAndreas Gruenbacher if (pi.cmd != P_CONNECTION_FEATURES) { 51651ec861ebSAndreas Gruenbacher drbd_err(connection, "expected ConnectionFeatures packet, received: %s (0x%04x)\n", 516677351055SPhilipp Reisner cmdname(pi.cmd), pi.cmd); 5167b411b363SPhilipp Reisner return -1; 5168b411b363SPhilipp Reisner } 5169b411b363SPhilipp Reisner 517077351055SPhilipp Reisner if (pi.size != expect) { 51711ec861ebSAndreas Gruenbacher drbd_err(connection, "expected ConnectionFeatures length: %u, received: %u\n", 517277351055SPhilipp Reisner expect, pi.size); 5173b411b363SPhilipp Reisner return -1; 5174b411b363SPhilipp Reisner } 5175b411b363SPhilipp Reisner 5176e658983aSAndreas Gruenbacher p = pi.data; 5177bde89a9eSAndreas Gruenbacher err = drbd_recv_all_warn(connection, p, expect); 5178a5c31904SAndreas Gruenbacher if (err) 5179b411b363SPhilipp Reisner return 0; 5180b411b363SPhilipp Reisner 5181b411b363SPhilipp Reisner p->protocol_min = be32_to_cpu(p->protocol_min); 5182b411b363SPhilipp Reisner p->protocol_max = be32_to_cpu(p->protocol_max); 5183b411b363SPhilipp Reisner if (p->protocol_max == 0) 5184b411b363SPhilipp Reisner p->protocol_max = p->protocol_min; 5185b411b363SPhilipp Reisner 5186b411b363SPhilipp Reisner if (PRO_VERSION_MAX < p->protocol_min || 5187b411b363SPhilipp Reisner PRO_VERSION_MIN > p->protocol_max) 5188b411b363SPhilipp Reisner goto incompat; 5189b411b363SPhilipp Reisner 5190bde89a9eSAndreas Gruenbacher connection->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); 519120c68fdeSLars Ellenberg connection->agreed_features = PRO_FEATURES & be32_to_cpu(p->feature_flags); 5192b411b363SPhilipp Reisner 51931ec861ebSAndreas Gruenbacher drbd_info(connection, "Handshake successful: " 5194bde89a9eSAndreas Gruenbacher "Agreed network protocol version %d\n", connection->agreed_pro_version); 5195b411b363SPhilipp Reisner 51969104d31aSLars Ellenberg drbd_info(connection, "Feature flags enabled on protocol level: 0x%x%s%s%s.\n", 51979104d31aSLars Ellenberg connection->agreed_features, 51989104d31aSLars Ellenberg connection->agreed_features & DRBD_FF_TRIM ? " TRIM" : "", 51999104d31aSLars Ellenberg connection->agreed_features & DRBD_FF_THIN_RESYNC ? " THIN_RESYNC" : "", 52009104d31aSLars Ellenberg connection->agreed_features & DRBD_FF_WSAME ? " WRITE_SAME" : 52019104d31aSLars Ellenberg connection->agreed_features ? "" : " none"); 520292d94ae6SPhilipp Reisner 5203b411b363SPhilipp Reisner return 1; 5204b411b363SPhilipp Reisner 5205b411b363SPhilipp Reisner incompat: 52061ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible DRBD dialects: " 5207b411b363SPhilipp Reisner "I support %d-%d, peer supports %d-%d\n", 5208b411b363SPhilipp Reisner PRO_VERSION_MIN, PRO_VERSION_MAX, 5209b411b363SPhilipp Reisner p->protocol_min, p->protocol_max); 5210b411b363SPhilipp Reisner return -1; 5211b411b363SPhilipp Reisner } 5212b411b363SPhilipp Reisner 5213b411b363SPhilipp Reisner #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) 5214bde89a9eSAndreas Gruenbacher static int drbd_do_auth(struct drbd_connection *connection) 5215b411b363SPhilipp Reisner { 52161ec861ebSAndreas Gruenbacher drbd_err(connection, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); 52171ec861ebSAndreas Gruenbacher drbd_err(connection, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); 5218b10d96cbSJohannes Thoma return -1; 5219b411b363SPhilipp Reisner } 5220b411b363SPhilipp Reisner #else 5221b411b363SPhilipp Reisner #define CHALLENGE_LEN 64 5222b10d96cbSJohannes Thoma 5223b10d96cbSJohannes Thoma /* Return value: 5224b10d96cbSJohannes Thoma 1 - auth succeeded, 5225b10d96cbSJohannes Thoma 0 - failed, try again (network error), 5226b10d96cbSJohannes Thoma -1 - auth failed, don't try again. 5227b10d96cbSJohannes Thoma */ 5228b10d96cbSJohannes Thoma 5229bde89a9eSAndreas Gruenbacher static int drbd_do_auth(struct drbd_connection *connection) 5230b411b363SPhilipp Reisner { 52319f5bdc33SAndreas Gruenbacher struct drbd_socket *sock; 5232b411b363SPhilipp Reisner char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ 5233b411b363SPhilipp Reisner char *response = NULL; 5234b411b363SPhilipp Reisner char *right_response = NULL; 5235b411b363SPhilipp Reisner char *peers_ch = NULL; 523644ed167dSPhilipp Reisner unsigned int key_len; 523744ed167dSPhilipp Reisner char secret[SHARED_SECRET_MAX]; /* 64 byte */ 5238b411b363SPhilipp Reisner unsigned int resp_size; 52399534d671SHerbert Xu SHASH_DESC_ON_STACK(desc, connection->cram_hmac_tfm); 524077351055SPhilipp Reisner struct packet_info pi; 524144ed167dSPhilipp Reisner struct net_conf *nc; 524269bc7bc3SAndreas Gruenbacher int err, rv; 5243b411b363SPhilipp Reisner 52449f5bdc33SAndreas Gruenbacher /* FIXME: Put the challenge/response into the preallocated socket buffer. */ 52459f5bdc33SAndreas Gruenbacher 524644ed167dSPhilipp Reisner rcu_read_lock(); 5247bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 524844ed167dSPhilipp Reisner key_len = strlen(nc->shared_secret); 524944ed167dSPhilipp Reisner memcpy(secret, nc->shared_secret, key_len); 525044ed167dSPhilipp Reisner rcu_read_unlock(); 525144ed167dSPhilipp Reisner 52529534d671SHerbert Xu desc->tfm = connection->cram_hmac_tfm; 52539534d671SHerbert Xu desc->flags = 0; 5254b411b363SPhilipp Reisner 52559534d671SHerbert Xu rv = crypto_shash_setkey(connection->cram_hmac_tfm, (u8 *)secret, key_len); 5256b411b363SPhilipp Reisner if (rv) { 52579534d671SHerbert Xu drbd_err(connection, "crypto_shash_setkey() failed with %d\n", rv); 5258b10d96cbSJohannes Thoma rv = -1; 5259b411b363SPhilipp Reisner goto fail; 5260b411b363SPhilipp Reisner } 5261b411b363SPhilipp Reisner 5262b411b363SPhilipp Reisner get_random_bytes(my_challenge, CHALLENGE_LEN); 5263b411b363SPhilipp Reisner 5264bde89a9eSAndreas Gruenbacher sock = &connection->data; 5265bde89a9eSAndreas Gruenbacher if (!conn_prepare_command(connection, sock)) { 52669f5bdc33SAndreas Gruenbacher rv = 0; 52679f5bdc33SAndreas Gruenbacher goto fail; 52689f5bdc33SAndreas Gruenbacher } 5269bde89a9eSAndreas Gruenbacher rv = !conn_send_command(connection, sock, P_AUTH_CHALLENGE, 0, 52709f5bdc33SAndreas Gruenbacher my_challenge, CHALLENGE_LEN); 5271b411b363SPhilipp Reisner if (!rv) 5272b411b363SPhilipp Reisner goto fail; 5273b411b363SPhilipp Reisner 5274bde89a9eSAndreas Gruenbacher err = drbd_recv_header(connection, &pi); 527569bc7bc3SAndreas Gruenbacher if (err) { 5276b411b363SPhilipp Reisner rv = 0; 5277b411b363SPhilipp Reisner goto fail; 5278b411b363SPhilipp Reisner } 5279b411b363SPhilipp Reisner 528077351055SPhilipp Reisner if (pi.cmd != P_AUTH_CHALLENGE) { 52811ec861ebSAndreas Gruenbacher drbd_err(connection, "expected AuthChallenge packet, received: %s (0x%04x)\n", 528277351055SPhilipp Reisner cmdname(pi.cmd), pi.cmd); 5283b411b363SPhilipp Reisner rv = 0; 5284b411b363SPhilipp Reisner goto fail; 5285b411b363SPhilipp Reisner } 5286b411b363SPhilipp Reisner 528777351055SPhilipp Reisner if (pi.size > CHALLENGE_LEN * 2) { 52881ec861ebSAndreas Gruenbacher drbd_err(connection, "expected AuthChallenge payload too big.\n"); 5289b10d96cbSJohannes Thoma rv = -1; 5290b411b363SPhilipp Reisner goto fail; 5291b411b363SPhilipp Reisner } 5292b411b363SPhilipp Reisner 529367cca286SPhilipp Reisner if (pi.size < CHALLENGE_LEN) { 529467cca286SPhilipp Reisner drbd_err(connection, "AuthChallenge payload too small.\n"); 529567cca286SPhilipp Reisner rv = -1; 529667cca286SPhilipp Reisner goto fail; 529767cca286SPhilipp Reisner } 529867cca286SPhilipp Reisner 529977351055SPhilipp Reisner peers_ch = kmalloc(pi.size, GFP_NOIO); 5300b411b363SPhilipp Reisner if (peers_ch == NULL) { 53011ec861ebSAndreas Gruenbacher drbd_err(connection, "kmalloc of peers_ch failed\n"); 5302b10d96cbSJohannes Thoma rv = -1; 5303b411b363SPhilipp Reisner goto fail; 5304b411b363SPhilipp Reisner } 5305b411b363SPhilipp Reisner 5306bde89a9eSAndreas Gruenbacher err = drbd_recv_all_warn(connection, peers_ch, pi.size); 5307a5c31904SAndreas Gruenbacher if (err) { 5308b411b363SPhilipp Reisner rv = 0; 5309b411b363SPhilipp Reisner goto fail; 5310b411b363SPhilipp Reisner } 5311b411b363SPhilipp Reisner 531267cca286SPhilipp Reisner if (!memcmp(my_challenge, peers_ch, CHALLENGE_LEN)) { 531367cca286SPhilipp Reisner drbd_err(connection, "Peer presented the same challenge!\n"); 531467cca286SPhilipp Reisner rv = -1; 531567cca286SPhilipp Reisner goto fail; 531667cca286SPhilipp Reisner } 531767cca286SPhilipp Reisner 53189534d671SHerbert Xu resp_size = crypto_shash_digestsize(connection->cram_hmac_tfm); 5319b411b363SPhilipp Reisner response = kmalloc(resp_size, GFP_NOIO); 5320b411b363SPhilipp Reisner if (response == NULL) { 53211ec861ebSAndreas Gruenbacher drbd_err(connection, "kmalloc of response failed\n"); 5322b10d96cbSJohannes Thoma rv = -1; 5323b411b363SPhilipp Reisner goto fail; 5324b411b363SPhilipp Reisner } 5325b411b363SPhilipp Reisner 53269534d671SHerbert Xu rv = crypto_shash_digest(desc, peers_ch, pi.size, response); 5327b411b363SPhilipp Reisner if (rv) { 53281ec861ebSAndreas Gruenbacher drbd_err(connection, "crypto_hash_digest() failed with %d\n", rv); 5329b10d96cbSJohannes Thoma rv = -1; 5330b411b363SPhilipp Reisner goto fail; 5331b411b363SPhilipp Reisner } 5332b411b363SPhilipp Reisner 5333bde89a9eSAndreas Gruenbacher if (!conn_prepare_command(connection, sock)) { 53349f5bdc33SAndreas Gruenbacher rv = 0; 53359f5bdc33SAndreas Gruenbacher goto fail; 53369f5bdc33SAndreas Gruenbacher } 5337bde89a9eSAndreas Gruenbacher rv = !conn_send_command(connection, sock, P_AUTH_RESPONSE, 0, 53389f5bdc33SAndreas Gruenbacher response, resp_size); 5339b411b363SPhilipp Reisner if (!rv) 5340b411b363SPhilipp Reisner goto fail; 5341b411b363SPhilipp Reisner 5342bde89a9eSAndreas Gruenbacher err = drbd_recv_header(connection, &pi); 534369bc7bc3SAndreas Gruenbacher if (err) { 5344b411b363SPhilipp Reisner rv = 0; 5345b411b363SPhilipp Reisner goto fail; 5346b411b363SPhilipp Reisner } 5347b411b363SPhilipp Reisner 534877351055SPhilipp Reisner if (pi.cmd != P_AUTH_RESPONSE) { 53491ec861ebSAndreas Gruenbacher drbd_err(connection, "expected AuthResponse packet, received: %s (0x%04x)\n", 535077351055SPhilipp Reisner cmdname(pi.cmd), pi.cmd); 5351b411b363SPhilipp Reisner rv = 0; 5352b411b363SPhilipp Reisner goto fail; 5353b411b363SPhilipp Reisner } 5354b411b363SPhilipp Reisner 535577351055SPhilipp Reisner if (pi.size != resp_size) { 53561ec861ebSAndreas Gruenbacher drbd_err(connection, "expected AuthResponse payload of wrong size\n"); 5357b411b363SPhilipp Reisner rv = 0; 5358b411b363SPhilipp Reisner goto fail; 5359b411b363SPhilipp Reisner } 5360b411b363SPhilipp Reisner 5361bde89a9eSAndreas Gruenbacher err = drbd_recv_all_warn(connection, response , resp_size); 5362a5c31904SAndreas Gruenbacher if (err) { 5363b411b363SPhilipp Reisner rv = 0; 5364b411b363SPhilipp Reisner goto fail; 5365b411b363SPhilipp Reisner } 5366b411b363SPhilipp Reisner 5367b411b363SPhilipp Reisner right_response = kmalloc(resp_size, GFP_NOIO); 53682d1ee87dSJulia Lawall if (right_response == NULL) { 53691ec861ebSAndreas Gruenbacher drbd_err(connection, "kmalloc of right_response failed\n"); 5370b10d96cbSJohannes Thoma rv = -1; 5371b411b363SPhilipp Reisner goto fail; 5372b411b363SPhilipp Reisner } 5373b411b363SPhilipp Reisner 53749534d671SHerbert Xu rv = crypto_shash_digest(desc, my_challenge, CHALLENGE_LEN, 53759534d671SHerbert Xu right_response); 5376b411b363SPhilipp Reisner if (rv) { 53771ec861ebSAndreas Gruenbacher drbd_err(connection, "crypto_hash_digest() failed with %d\n", rv); 5378b10d96cbSJohannes Thoma rv = -1; 5379b411b363SPhilipp Reisner goto fail; 5380b411b363SPhilipp Reisner } 5381b411b363SPhilipp Reisner 5382b411b363SPhilipp Reisner rv = !memcmp(response, right_response, resp_size); 5383b411b363SPhilipp Reisner 5384b411b363SPhilipp Reisner if (rv) 53851ec861ebSAndreas Gruenbacher drbd_info(connection, "Peer authenticated using %d bytes HMAC\n", 538644ed167dSPhilipp Reisner resp_size); 5387b10d96cbSJohannes Thoma else 5388b10d96cbSJohannes Thoma rv = -1; 5389b411b363SPhilipp Reisner 5390b411b363SPhilipp Reisner fail: 5391b411b363SPhilipp Reisner kfree(peers_ch); 5392b411b363SPhilipp Reisner kfree(response); 5393b411b363SPhilipp Reisner kfree(right_response); 53949534d671SHerbert Xu shash_desc_zero(desc); 5395b411b363SPhilipp Reisner 5396b411b363SPhilipp Reisner return rv; 5397b411b363SPhilipp Reisner } 5398b411b363SPhilipp Reisner #endif 5399b411b363SPhilipp Reisner 54008fe60551SAndreas Gruenbacher int drbd_receiver(struct drbd_thread *thi) 5401b411b363SPhilipp Reisner { 5402bde89a9eSAndreas Gruenbacher struct drbd_connection *connection = thi->connection; 5403b411b363SPhilipp Reisner int h; 5404b411b363SPhilipp Reisner 54051ec861ebSAndreas Gruenbacher drbd_info(connection, "receiver (re)started\n"); 5406b411b363SPhilipp Reisner 5407b411b363SPhilipp Reisner do { 5408bde89a9eSAndreas Gruenbacher h = conn_connect(connection); 5409b411b363SPhilipp Reisner if (h == 0) { 5410bde89a9eSAndreas Gruenbacher conn_disconnect(connection); 541120ee6390SPhilipp Reisner schedule_timeout_interruptible(HZ); 5412b411b363SPhilipp Reisner } 5413b411b363SPhilipp Reisner if (h == -1) { 54141ec861ebSAndreas Gruenbacher drbd_warn(connection, "Discarding network configuration.\n"); 5415bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); 5416b411b363SPhilipp Reisner } 5417b411b363SPhilipp Reisner } while (h == 0); 5418b411b363SPhilipp Reisner 5419c51a0ef3SLars Ellenberg if (h > 0) { 5420c51a0ef3SLars Ellenberg blk_start_plug(&connection->receiver_plug); 5421bde89a9eSAndreas Gruenbacher drbdd(connection); 5422c51a0ef3SLars Ellenberg blk_finish_plug(&connection->receiver_plug); 5423c51a0ef3SLars Ellenberg } 5424b411b363SPhilipp Reisner 5425bde89a9eSAndreas Gruenbacher conn_disconnect(connection); 5426b411b363SPhilipp Reisner 54271ec861ebSAndreas Gruenbacher drbd_info(connection, "receiver terminated\n"); 5428b411b363SPhilipp Reisner return 0; 5429b411b363SPhilipp Reisner } 5430b411b363SPhilipp Reisner 5431b411b363SPhilipp Reisner /* ********* acknowledge sender ******** */ 5432b411b363SPhilipp Reisner 5433bde89a9eSAndreas Gruenbacher static int got_conn_RqSReply(struct drbd_connection *connection, struct packet_info *pi) 5434b411b363SPhilipp Reisner { 5435e658983aSAndreas Gruenbacher struct p_req_state_reply *p = pi->data; 5436b411b363SPhilipp Reisner int retcode = be32_to_cpu(p->retcode); 5437b411b363SPhilipp Reisner 5438b411b363SPhilipp Reisner if (retcode >= SS_SUCCESS) { 5439bde89a9eSAndreas Gruenbacher set_bit(CONN_WD_ST_CHG_OKAY, &connection->flags); 5440b411b363SPhilipp Reisner } else { 5441bde89a9eSAndreas Gruenbacher set_bit(CONN_WD_ST_CHG_FAIL, &connection->flags); 54421ec861ebSAndreas Gruenbacher drbd_err(connection, "Requested state change failed by peer: %s (%d)\n", 5443fc3b10a4SPhilipp Reisner drbd_set_st_err_str(retcode), retcode); 5444fc3b10a4SPhilipp Reisner } 5445bde89a9eSAndreas Gruenbacher wake_up(&connection->ping_wait); 5446e4f78edeSPhilipp Reisner 54472735a594SAndreas Gruenbacher return 0; 5448fc3b10a4SPhilipp Reisner } 5449e4f78edeSPhilipp Reisner 5450bde89a9eSAndreas Gruenbacher static int got_RqSReply(struct drbd_connection *connection, struct packet_info *pi) 5451e4f78edeSPhilipp Reisner { 54529f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 5453b30ab791SAndreas Gruenbacher struct drbd_device *device; 5454e658983aSAndreas Gruenbacher struct p_req_state_reply *p = pi->data; 5455e4f78edeSPhilipp Reisner int retcode = be32_to_cpu(p->retcode); 5456e4f78edeSPhilipp Reisner 54579f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 54589f4fe9adSAndreas Gruenbacher if (!peer_device) 54592735a594SAndreas Gruenbacher return -EIO; 54609f4fe9adSAndreas Gruenbacher device = peer_device->device; 54611952e916SAndreas Gruenbacher 5462bde89a9eSAndreas Gruenbacher if (test_bit(CONN_WD_ST_CHG_REQ, &connection->flags)) { 54630b0ba1efSAndreas Gruenbacher D_ASSERT(device, connection->agreed_pro_version < 100); 5464bde89a9eSAndreas Gruenbacher return got_conn_RqSReply(connection, pi); 54654d0fc3fdSPhilipp Reisner } 54664d0fc3fdSPhilipp Reisner 5467e4f78edeSPhilipp Reisner if (retcode >= SS_SUCCESS) { 5468b30ab791SAndreas Gruenbacher set_bit(CL_ST_CHG_SUCCESS, &device->flags); 5469e4f78edeSPhilipp Reisner } else { 5470b30ab791SAndreas Gruenbacher set_bit(CL_ST_CHG_FAIL, &device->flags); 5471d0180171SAndreas Gruenbacher drbd_err(device, "Requested state change failed by peer: %s (%d)\n", 5472b411b363SPhilipp Reisner drbd_set_st_err_str(retcode), retcode); 5473b411b363SPhilipp Reisner } 5474b30ab791SAndreas Gruenbacher wake_up(&device->state_wait); 5475b411b363SPhilipp Reisner 54762735a594SAndreas Gruenbacher return 0; 5477b411b363SPhilipp Reisner } 5478b411b363SPhilipp Reisner 5479bde89a9eSAndreas Gruenbacher static int got_Ping(struct drbd_connection *connection, struct packet_info *pi) 5480b411b363SPhilipp Reisner { 5481bde89a9eSAndreas Gruenbacher return drbd_send_ping_ack(connection); 5482b411b363SPhilipp Reisner 5483b411b363SPhilipp Reisner } 5484b411b363SPhilipp Reisner 5485bde89a9eSAndreas Gruenbacher static int got_PingAck(struct drbd_connection *connection, struct packet_info *pi) 5486b411b363SPhilipp Reisner { 5487b411b363SPhilipp Reisner /* restore idle timeout */ 5488bde89a9eSAndreas Gruenbacher connection->meta.socket->sk->sk_rcvtimeo = connection->net_conf->ping_int*HZ; 5489bde89a9eSAndreas Gruenbacher if (!test_and_set_bit(GOT_PING_ACK, &connection->flags)) 5490bde89a9eSAndreas Gruenbacher wake_up(&connection->ping_wait); 5491b411b363SPhilipp Reisner 54922735a594SAndreas Gruenbacher return 0; 5493b411b363SPhilipp Reisner } 5494b411b363SPhilipp Reisner 5495bde89a9eSAndreas Gruenbacher static int got_IsInSync(struct drbd_connection *connection, struct packet_info *pi) 5496b411b363SPhilipp Reisner { 54979f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 5498b30ab791SAndreas Gruenbacher struct drbd_device *device; 5499e658983aSAndreas Gruenbacher struct p_block_ack *p = pi->data; 5500b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 5501b411b363SPhilipp Reisner int blksize = be32_to_cpu(p->blksize); 5502b411b363SPhilipp Reisner 55039f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 55049f4fe9adSAndreas Gruenbacher if (!peer_device) 55052735a594SAndreas Gruenbacher return -EIO; 55069f4fe9adSAndreas Gruenbacher device = peer_device->device; 55071952e916SAndreas Gruenbacher 55089f4fe9adSAndreas Gruenbacher D_ASSERT(device, peer_device->connection->agreed_pro_version >= 89); 5509b411b363SPhilipp Reisner 551069a22773SAndreas Gruenbacher update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); 5511b411b363SPhilipp Reisner 5512b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 5513b30ab791SAndreas Gruenbacher drbd_rs_complete_io(device, sector); 5514b30ab791SAndreas Gruenbacher drbd_set_in_sync(device, sector, blksize); 5515b411b363SPhilipp Reisner /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ 5516b30ab791SAndreas Gruenbacher device->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); 5517b30ab791SAndreas Gruenbacher put_ldev(device); 55181d53f09eSLars Ellenberg } 5519b30ab791SAndreas Gruenbacher dec_rs_pending(device); 5520b30ab791SAndreas Gruenbacher atomic_add(blksize >> 9, &device->rs_sect_in); 5521b411b363SPhilipp Reisner 55222735a594SAndreas Gruenbacher return 0; 5523b411b363SPhilipp Reisner } 5524b411b363SPhilipp Reisner 5525bc9c5c41SAndreas Gruenbacher static int 5526b30ab791SAndreas Gruenbacher validate_req_change_req_state(struct drbd_device *device, u64 id, sector_t sector, 5527bc9c5c41SAndreas Gruenbacher struct rb_root *root, const char *func, 5528bc9c5c41SAndreas Gruenbacher enum drbd_req_event what, bool missing_ok) 5529b411b363SPhilipp Reisner { 5530b411b363SPhilipp Reisner struct drbd_request *req; 5531b411b363SPhilipp Reisner struct bio_and_error m; 5532b411b363SPhilipp Reisner 55330500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 5534b30ab791SAndreas Gruenbacher req = find_request(device, root, id, sector, missing_ok, func); 5535b411b363SPhilipp Reisner if (unlikely(!req)) { 55360500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 553785997675SAndreas Gruenbacher return -EIO; 5538b411b363SPhilipp Reisner } 5539b411b363SPhilipp Reisner __req_mod(req, what, &m); 55400500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 5541b411b363SPhilipp Reisner 5542b411b363SPhilipp Reisner if (m.bio) 5543b30ab791SAndreas Gruenbacher complete_master_bio(device, &m); 554485997675SAndreas Gruenbacher return 0; 5545b411b363SPhilipp Reisner } 5546b411b363SPhilipp Reisner 5547bde89a9eSAndreas Gruenbacher static int got_BlockAck(struct drbd_connection *connection, struct packet_info *pi) 5548b411b363SPhilipp Reisner { 55499f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 5550b30ab791SAndreas Gruenbacher struct drbd_device *device; 5551e658983aSAndreas Gruenbacher struct p_block_ack *p = pi->data; 5552b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 5553b411b363SPhilipp Reisner int blksize = be32_to_cpu(p->blksize); 5554b411b363SPhilipp Reisner enum drbd_req_event what; 5555b411b363SPhilipp Reisner 55569f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 55579f4fe9adSAndreas Gruenbacher if (!peer_device) 55582735a594SAndreas Gruenbacher return -EIO; 55599f4fe9adSAndreas Gruenbacher device = peer_device->device; 55601952e916SAndreas Gruenbacher 556169a22773SAndreas Gruenbacher update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); 5562b411b363SPhilipp Reisner 5563579b57edSAndreas Gruenbacher if (p->block_id == ID_SYNCER) { 5564b30ab791SAndreas Gruenbacher drbd_set_in_sync(device, sector, blksize); 5565b30ab791SAndreas Gruenbacher dec_rs_pending(device); 55662735a594SAndreas Gruenbacher return 0; 5567b411b363SPhilipp Reisner } 5568e05e1e59SAndreas Gruenbacher switch (pi->cmd) { 5569b411b363SPhilipp Reisner case P_RS_WRITE_ACK: 55708554df1cSAndreas Gruenbacher what = WRITE_ACKED_BY_PEER_AND_SIS; 5571b411b363SPhilipp Reisner break; 5572b411b363SPhilipp Reisner case P_WRITE_ACK: 55738554df1cSAndreas Gruenbacher what = WRITE_ACKED_BY_PEER; 5574b411b363SPhilipp Reisner break; 5575b411b363SPhilipp Reisner case P_RECV_ACK: 55768554df1cSAndreas Gruenbacher what = RECV_ACKED_BY_PEER; 5577b411b363SPhilipp Reisner break; 5578d4dabbe2SLars Ellenberg case P_SUPERSEDED: 5579d4dabbe2SLars Ellenberg what = CONFLICT_RESOLVED; 55807be8da07SAndreas Gruenbacher break; 55817be8da07SAndreas Gruenbacher case P_RETRY_WRITE: 55827be8da07SAndreas Gruenbacher what = POSTPONE_WRITE; 5583b411b363SPhilipp Reisner break; 5584b411b363SPhilipp Reisner default: 55852735a594SAndreas Gruenbacher BUG(); 5586b411b363SPhilipp Reisner } 5587b411b363SPhilipp Reisner 5588b30ab791SAndreas Gruenbacher return validate_req_change_req_state(device, p->block_id, sector, 5589b30ab791SAndreas Gruenbacher &device->write_requests, __func__, 5590bc9c5c41SAndreas Gruenbacher what, false); 5591b411b363SPhilipp Reisner } 5592b411b363SPhilipp Reisner 5593bde89a9eSAndreas Gruenbacher static int got_NegAck(struct drbd_connection *connection, struct packet_info *pi) 5594b411b363SPhilipp Reisner { 55959f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 5596b30ab791SAndreas Gruenbacher struct drbd_device *device; 5597e658983aSAndreas Gruenbacher struct p_block_ack *p = pi->data; 5598b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 55992deb8336SPhilipp Reisner int size = be32_to_cpu(p->blksize); 560085997675SAndreas Gruenbacher int err; 5601b411b363SPhilipp Reisner 56029f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 56039f4fe9adSAndreas Gruenbacher if (!peer_device) 56042735a594SAndreas Gruenbacher return -EIO; 56059f4fe9adSAndreas Gruenbacher device = peer_device->device; 5606b411b363SPhilipp Reisner 560769a22773SAndreas Gruenbacher update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); 5608b411b363SPhilipp Reisner 5609579b57edSAndreas Gruenbacher if (p->block_id == ID_SYNCER) { 5610b30ab791SAndreas Gruenbacher dec_rs_pending(device); 5611b30ab791SAndreas Gruenbacher drbd_rs_failed_io(device, sector, size); 56122735a594SAndreas Gruenbacher return 0; 5613b411b363SPhilipp Reisner } 56142deb8336SPhilipp Reisner 5615b30ab791SAndreas Gruenbacher err = validate_req_change_req_state(device, p->block_id, sector, 5616b30ab791SAndreas Gruenbacher &device->write_requests, __func__, 5617303d1448SPhilipp Reisner NEG_ACKED, true); 561885997675SAndreas Gruenbacher if (err) { 56192deb8336SPhilipp Reisner /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. 56202deb8336SPhilipp Reisner The master bio might already be completed, therefore the 5621c3afd8f5SAndreas Gruenbacher request is no longer in the collision hash. */ 56222deb8336SPhilipp Reisner /* In Protocol B we might already have got a P_RECV_ACK 56232deb8336SPhilipp Reisner but then get a P_NEG_ACK afterwards. */ 5624b30ab791SAndreas Gruenbacher drbd_set_out_of_sync(device, sector, size); 56252deb8336SPhilipp Reisner } 56262735a594SAndreas Gruenbacher return 0; 5627b411b363SPhilipp Reisner } 5628b411b363SPhilipp Reisner 5629bde89a9eSAndreas Gruenbacher static int got_NegDReply(struct drbd_connection *connection, struct packet_info *pi) 5630b411b363SPhilipp Reisner { 56319f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 5632b30ab791SAndreas Gruenbacher struct drbd_device *device; 5633e658983aSAndreas Gruenbacher struct p_block_ack *p = pi->data; 5634b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 5635b411b363SPhilipp Reisner 56369f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 56379f4fe9adSAndreas Gruenbacher if (!peer_device) 56382735a594SAndreas Gruenbacher return -EIO; 56399f4fe9adSAndreas Gruenbacher device = peer_device->device; 56401952e916SAndreas Gruenbacher 564169a22773SAndreas Gruenbacher update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); 56427be8da07SAndreas Gruenbacher 5643d0180171SAndreas Gruenbacher drbd_err(device, "Got NegDReply; Sector %llus, len %u.\n", 5644b411b363SPhilipp Reisner (unsigned long long)sector, be32_to_cpu(p->blksize)); 5645b411b363SPhilipp Reisner 5646b30ab791SAndreas Gruenbacher return validate_req_change_req_state(device, p->block_id, sector, 5647b30ab791SAndreas Gruenbacher &device->read_requests, __func__, 56488554df1cSAndreas Gruenbacher NEG_ACKED, false); 5649b411b363SPhilipp Reisner } 5650b411b363SPhilipp Reisner 5651bde89a9eSAndreas Gruenbacher static int got_NegRSDReply(struct drbd_connection *connection, struct packet_info *pi) 5652b411b363SPhilipp Reisner { 56539f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 5654b30ab791SAndreas Gruenbacher struct drbd_device *device; 5655b411b363SPhilipp Reisner sector_t sector; 5656b411b363SPhilipp Reisner int size; 5657e658983aSAndreas Gruenbacher struct p_block_ack *p = pi->data; 56581952e916SAndreas Gruenbacher 56599f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 56609f4fe9adSAndreas Gruenbacher if (!peer_device) 56612735a594SAndreas Gruenbacher return -EIO; 56629f4fe9adSAndreas Gruenbacher device = peer_device->device; 5663b411b363SPhilipp Reisner 5664b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 5665b411b363SPhilipp Reisner size = be32_to_cpu(p->blksize); 5666b411b363SPhilipp Reisner 566769a22773SAndreas Gruenbacher update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); 5668b411b363SPhilipp Reisner 5669b30ab791SAndreas Gruenbacher dec_rs_pending(device); 5670b411b363SPhilipp Reisner 5671b30ab791SAndreas Gruenbacher if (get_ldev_if_state(device, D_FAILED)) { 5672b30ab791SAndreas Gruenbacher drbd_rs_complete_io(device, sector); 5673e05e1e59SAndreas Gruenbacher switch (pi->cmd) { 5674d612d309SPhilipp Reisner case P_NEG_RS_DREPLY: 5675b30ab791SAndreas Gruenbacher drbd_rs_failed_io(device, sector, size); 5676d612d309SPhilipp Reisner case P_RS_CANCEL: 5677d612d309SPhilipp Reisner break; 5678d612d309SPhilipp Reisner default: 56792735a594SAndreas Gruenbacher BUG(); 5680d612d309SPhilipp Reisner } 5681b30ab791SAndreas Gruenbacher put_ldev(device); 5682b411b363SPhilipp Reisner } 5683b411b363SPhilipp Reisner 56842735a594SAndreas Gruenbacher return 0; 5685b411b363SPhilipp Reisner } 5686b411b363SPhilipp Reisner 5687bde89a9eSAndreas Gruenbacher static int got_BarrierAck(struct drbd_connection *connection, struct packet_info *pi) 5688b411b363SPhilipp Reisner { 5689e658983aSAndreas Gruenbacher struct p_barrier_ack *p = pi->data; 5690c06ece6bSAndreas Gruenbacher struct drbd_peer_device *peer_device; 56919ed57dcbSLars Ellenberg int vnr; 5692b411b363SPhilipp Reisner 5693bde89a9eSAndreas Gruenbacher tl_release(connection, p->barrier, be32_to_cpu(p->set_size)); 5694b411b363SPhilipp Reisner 56959ed57dcbSLars Ellenberg rcu_read_lock(); 5696c06ece6bSAndreas Gruenbacher idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 5697c06ece6bSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 5698c06ece6bSAndreas Gruenbacher 5699b30ab791SAndreas Gruenbacher if (device->state.conn == C_AHEAD && 5700b30ab791SAndreas Gruenbacher atomic_read(&device->ap_in_flight) == 0 && 5701b30ab791SAndreas Gruenbacher !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &device->flags)) { 5702b30ab791SAndreas Gruenbacher device->start_resync_timer.expires = jiffies + HZ; 5703b30ab791SAndreas Gruenbacher add_timer(&device->start_resync_timer); 5704c4752ef1SPhilipp Reisner } 57059ed57dcbSLars Ellenberg } 57069ed57dcbSLars Ellenberg rcu_read_unlock(); 5707c4752ef1SPhilipp Reisner 57082735a594SAndreas Gruenbacher return 0; 5709b411b363SPhilipp Reisner } 5710b411b363SPhilipp Reisner 5711bde89a9eSAndreas Gruenbacher static int got_OVResult(struct drbd_connection *connection, struct packet_info *pi) 5712b411b363SPhilipp Reisner { 57139f4fe9adSAndreas Gruenbacher struct drbd_peer_device *peer_device; 5714b30ab791SAndreas Gruenbacher struct drbd_device *device; 5715e658983aSAndreas Gruenbacher struct p_block_ack *p = pi->data; 571684b8c06bSAndreas Gruenbacher struct drbd_device_work *dw; 5717b411b363SPhilipp Reisner sector_t sector; 5718b411b363SPhilipp Reisner int size; 5719b411b363SPhilipp Reisner 57209f4fe9adSAndreas Gruenbacher peer_device = conn_peer_device(connection, pi->vnr); 57219f4fe9adSAndreas Gruenbacher if (!peer_device) 57222735a594SAndreas Gruenbacher return -EIO; 57239f4fe9adSAndreas Gruenbacher device = peer_device->device; 57241952e916SAndreas Gruenbacher 5725b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 5726b411b363SPhilipp Reisner size = be32_to_cpu(p->blksize); 5727b411b363SPhilipp Reisner 572869a22773SAndreas Gruenbacher update_peer_seq(peer_device, be32_to_cpu(p->seq_num)); 5729b411b363SPhilipp Reisner 5730b411b363SPhilipp Reisner if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) 5731b30ab791SAndreas Gruenbacher drbd_ov_out_of_sync_found(device, sector, size); 5732b411b363SPhilipp Reisner else 5733b30ab791SAndreas Gruenbacher ov_out_of_sync_print(device); 5734b411b363SPhilipp Reisner 5735b30ab791SAndreas Gruenbacher if (!get_ldev(device)) 57362735a594SAndreas Gruenbacher return 0; 57371d53f09eSLars Ellenberg 5738b30ab791SAndreas Gruenbacher drbd_rs_complete_io(device, sector); 5739b30ab791SAndreas Gruenbacher dec_rs_pending(device); 5740b411b363SPhilipp Reisner 5741b30ab791SAndreas Gruenbacher --device->ov_left; 5742ea5442afSLars Ellenberg 5743ea5442afSLars Ellenberg /* let's advance progress step marks only for every other megabyte */ 5744b30ab791SAndreas Gruenbacher if ((device->ov_left & 0x200) == 0x200) 5745b30ab791SAndreas Gruenbacher drbd_advance_rs_marks(device, device->ov_left); 5746ea5442afSLars Ellenberg 5747b30ab791SAndreas Gruenbacher if (device->ov_left == 0) { 574884b8c06bSAndreas Gruenbacher dw = kmalloc(sizeof(*dw), GFP_NOIO); 574984b8c06bSAndreas Gruenbacher if (dw) { 575084b8c06bSAndreas Gruenbacher dw->w.cb = w_ov_finished; 575184b8c06bSAndreas Gruenbacher dw->device = device; 575284b8c06bSAndreas Gruenbacher drbd_queue_work(&peer_device->connection->sender_work, &dw->w); 5753b411b363SPhilipp Reisner } else { 575484b8c06bSAndreas Gruenbacher drbd_err(device, "kmalloc(dw) failed."); 5755b30ab791SAndreas Gruenbacher ov_out_of_sync_print(device); 5756b30ab791SAndreas Gruenbacher drbd_resync_finished(device); 5757b411b363SPhilipp Reisner } 5758b411b363SPhilipp Reisner } 5759b30ab791SAndreas Gruenbacher put_ldev(device); 57602735a594SAndreas Gruenbacher return 0; 5761b411b363SPhilipp Reisner } 5762b411b363SPhilipp Reisner 5763bde89a9eSAndreas Gruenbacher static int got_skip(struct drbd_connection *connection, struct packet_info *pi) 57640ced55a3SPhilipp Reisner { 57652735a594SAndreas Gruenbacher return 0; 57660ced55a3SPhilipp Reisner } 57670ced55a3SPhilipp Reisner 5768668700b4SPhilipp Reisner struct meta_sock_cmd { 5769b411b363SPhilipp Reisner size_t pkt_size; 5770bde89a9eSAndreas Gruenbacher int (*fn)(struct drbd_connection *connection, struct packet_info *); 5771b411b363SPhilipp Reisner }; 5772b411b363SPhilipp Reisner 5773668700b4SPhilipp Reisner static void set_rcvtimeo(struct drbd_connection *connection, bool ping_timeout) 5774668700b4SPhilipp Reisner { 5775668700b4SPhilipp Reisner long t; 5776668700b4SPhilipp Reisner struct net_conf *nc; 5777668700b4SPhilipp Reisner 5778668700b4SPhilipp Reisner rcu_read_lock(); 5779668700b4SPhilipp Reisner nc = rcu_dereference(connection->net_conf); 5780668700b4SPhilipp Reisner t = ping_timeout ? nc->ping_timeo : nc->ping_int; 5781668700b4SPhilipp Reisner rcu_read_unlock(); 5782668700b4SPhilipp Reisner 5783668700b4SPhilipp Reisner t *= HZ; 5784668700b4SPhilipp Reisner if (ping_timeout) 5785668700b4SPhilipp Reisner t /= 10; 5786668700b4SPhilipp Reisner 5787668700b4SPhilipp Reisner connection->meta.socket->sk->sk_rcvtimeo = t; 5788668700b4SPhilipp Reisner } 5789668700b4SPhilipp Reisner 5790668700b4SPhilipp Reisner static void set_ping_timeout(struct drbd_connection *connection) 5791668700b4SPhilipp Reisner { 5792668700b4SPhilipp Reisner set_rcvtimeo(connection, 1); 5793668700b4SPhilipp Reisner } 5794668700b4SPhilipp Reisner 5795668700b4SPhilipp Reisner static void set_idle_timeout(struct drbd_connection *connection) 5796668700b4SPhilipp Reisner { 5797668700b4SPhilipp Reisner set_rcvtimeo(connection, 0); 5798668700b4SPhilipp Reisner } 5799668700b4SPhilipp Reisner 5800668700b4SPhilipp Reisner static struct meta_sock_cmd ack_receiver_tbl[] = { 5801e658983aSAndreas Gruenbacher [P_PING] = { 0, got_Ping }, 5802e658983aSAndreas Gruenbacher [P_PING_ACK] = { 0, got_PingAck }, 5803b411b363SPhilipp Reisner [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 5804b411b363SPhilipp Reisner [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 5805b411b363SPhilipp Reisner [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 5806d4dabbe2SLars Ellenberg [P_SUPERSEDED] = { sizeof(struct p_block_ack), got_BlockAck }, 5807b411b363SPhilipp Reisner [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, 5808b411b363SPhilipp Reisner [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, 5809b411b363SPhilipp Reisner [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply }, 5810b411b363SPhilipp Reisner [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, 5811b411b363SPhilipp Reisner [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, 5812b411b363SPhilipp Reisner [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, 5813b411b363SPhilipp Reisner [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, 581402918be2SPhilipp Reisner [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, 5815d612d309SPhilipp Reisner [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply }, 58161952e916SAndreas Gruenbacher [P_CONN_ST_CHG_REPLY]={ sizeof(struct p_req_state_reply), got_conn_RqSReply }, 58171952e916SAndreas Gruenbacher [P_RETRY_WRITE] = { sizeof(struct p_block_ack), got_BlockAck }, 5818b411b363SPhilipp Reisner }; 5819b411b363SPhilipp Reisner 58201c03e520SPhilipp Reisner int drbd_ack_receiver(struct drbd_thread *thi) 5821b411b363SPhilipp Reisner { 5822bde89a9eSAndreas Gruenbacher struct drbd_connection *connection = thi->connection; 5823668700b4SPhilipp Reisner struct meta_sock_cmd *cmd = NULL; 582477351055SPhilipp Reisner struct packet_info pi; 5825668700b4SPhilipp Reisner unsigned long pre_recv_jif; 5826257d0af6SPhilipp Reisner int rv; 5827bde89a9eSAndreas Gruenbacher void *buf = connection->meta.rbuf; 5828b411b363SPhilipp Reisner int received = 0; 5829bde89a9eSAndreas Gruenbacher unsigned int header_size = drbd_header_size(connection); 583052b061a4SAndreas Gruenbacher int expect = header_size; 583144ed167dSPhilipp Reisner bool ping_timeout_active = false; 58323990e04dSPhilipp Reisner struct sched_param param = { .sched_priority = 2 }; 5833b411b363SPhilipp Reisner 58343990e04dSPhilipp Reisner rv = sched_setscheduler(current, SCHED_RR, ¶m); 58353990e04dSPhilipp Reisner if (rv < 0) 5836668700b4SPhilipp Reisner drbd_err(connection, "drbd_ack_receiver: ERROR set priority, ret=%d\n", rv); 5837b411b363SPhilipp Reisner 5838e77a0a5cSAndreas Gruenbacher while (get_t_state(thi) == RUNNING) { 583980822284SPhilipp Reisner drbd_thread_current_set_cpu(thi); 584044ed167dSPhilipp Reisner 5841668700b4SPhilipp Reisner conn_reclaim_net_peer_reqs(connection); 584244ed167dSPhilipp Reisner 5843bde89a9eSAndreas Gruenbacher if (test_and_clear_bit(SEND_PING, &connection->flags)) { 5844bde89a9eSAndreas Gruenbacher if (drbd_send_ping(connection)) { 58451ec861ebSAndreas Gruenbacher drbd_err(connection, "drbd_send_ping has failed\n"); 5846841ce241SAndreas Gruenbacher goto reconnect; 5847841ce241SAndreas Gruenbacher } 5848668700b4SPhilipp Reisner set_ping_timeout(connection); 584944ed167dSPhilipp Reisner ping_timeout_active = true; 5850b411b363SPhilipp Reisner } 5851b411b363SPhilipp Reisner 5852668700b4SPhilipp Reisner pre_recv_jif = jiffies; 5853bde89a9eSAndreas Gruenbacher rv = drbd_recv_short(connection->meta.socket, buf, expect-received, 0); 5854b411b363SPhilipp Reisner 5855b411b363SPhilipp Reisner /* Note: 5856b411b363SPhilipp Reisner * -EINTR (on meta) we got a signal 5857b411b363SPhilipp Reisner * -EAGAIN (on meta) rcvtimeo expired 5858b411b363SPhilipp Reisner * -ECONNRESET other side closed the connection 5859b411b363SPhilipp Reisner * -ERESTARTSYS (on data) we got a signal 5860b411b363SPhilipp Reisner * rv < 0 other than above: unexpected error! 5861b411b363SPhilipp Reisner * rv == expected: full header or command 5862b411b363SPhilipp Reisner * rv < expected: "woken" by signal during receive 5863b411b363SPhilipp Reisner * rv == 0 : "connection shut down by peer" 5864b411b363SPhilipp Reisner */ 5865b411b363SPhilipp Reisner if (likely(rv > 0)) { 5866b411b363SPhilipp Reisner received += rv; 5867b411b363SPhilipp Reisner buf += rv; 5868b411b363SPhilipp Reisner } else if (rv == 0) { 5869bde89a9eSAndreas Gruenbacher if (test_bit(DISCONNECT_SENT, &connection->flags)) { 5870b66623e3SPhilipp Reisner long t; 5871b66623e3SPhilipp Reisner rcu_read_lock(); 5872bde89a9eSAndreas Gruenbacher t = rcu_dereference(connection->net_conf)->ping_timeo * HZ/10; 5873b66623e3SPhilipp Reisner rcu_read_unlock(); 5874b66623e3SPhilipp Reisner 5875bde89a9eSAndreas Gruenbacher t = wait_event_timeout(connection->ping_wait, 5876bde89a9eSAndreas Gruenbacher connection->cstate < C_WF_REPORT_PARAMS, 5877b66623e3SPhilipp Reisner t); 5878599377acSPhilipp Reisner if (t) 5879599377acSPhilipp Reisner break; 5880599377acSPhilipp Reisner } 58811ec861ebSAndreas Gruenbacher drbd_err(connection, "meta connection shut down by peer.\n"); 5882b411b363SPhilipp Reisner goto reconnect; 5883b411b363SPhilipp Reisner } else if (rv == -EAGAIN) { 5884cb6518cbSLars Ellenberg /* If the data socket received something meanwhile, 5885cb6518cbSLars Ellenberg * that is good enough: peer is still alive. */ 5886668700b4SPhilipp Reisner if (time_after(connection->last_received, pre_recv_jif)) 5887cb6518cbSLars Ellenberg continue; 5888f36af18cSLars Ellenberg if (ping_timeout_active) { 58891ec861ebSAndreas Gruenbacher drbd_err(connection, "PingAck did not arrive in time.\n"); 5890b411b363SPhilipp Reisner goto reconnect; 5891b411b363SPhilipp Reisner } 5892bde89a9eSAndreas Gruenbacher set_bit(SEND_PING, &connection->flags); 5893b411b363SPhilipp Reisner continue; 5894b411b363SPhilipp Reisner } else if (rv == -EINTR) { 5895668700b4SPhilipp Reisner /* maybe drbd_thread_stop(): the while condition will notice. 5896668700b4SPhilipp Reisner * maybe woken for send_ping: we'll send a ping above, 5897668700b4SPhilipp Reisner * and change the rcvtimeo */ 5898668700b4SPhilipp Reisner flush_signals(current); 5899b411b363SPhilipp Reisner continue; 5900b411b363SPhilipp Reisner } else { 59011ec861ebSAndreas Gruenbacher drbd_err(connection, "sock_recvmsg returned %d\n", rv); 5902b411b363SPhilipp Reisner goto reconnect; 5903b411b363SPhilipp Reisner } 5904b411b363SPhilipp Reisner 5905b411b363SPhilipp Reisner if (received == expect && cmd == NULL) { 5906bde89a9eSAndreas Gruenbacher if (decode_header(connection, connection->meta.rbuf, &pi)) 5907b411b363SPhilipp Reisner goto reconnect; 5908668700b4SPhilipp Reisner cmd = &ack_receiver_tbl[pi.cmd]; 5909668700b4SPhilipp Reisner if (pi.cmd >= ARRAY_SIZE(ack_receiver_tbl) || !cmd->fn) { 59101ec861ebSAndreas Gruenbacher drbd_err(connection, "Unexpected meta packet %s (0x%04x)\n", 59112fcb8f30SAndreas Gruenbacher cmdname(pi.cmd), pi.cmd); 5912b411b363SPhilipp Reisner goto disconnect; 5913b411b363SPhilipp Reisner } 5914e658983aSAndreas Gruenbacher expect = header_size + cmd->pkt_size; 591552b061a4SAndreas Gruenbacher if (pi.size != expect - header_size) { 59161ec861ebSAndreas Gruenbacher drbd_err(connection, "Wrong packet size on meta (c: %d, l: %d)\n", 591777351055SPhilipp Reisner pi.cmd, pi.size); 5918b411b363SPhilipp Reisner goto reconnect; 5919b411b363SPhilipp Reisner } 5920257d0af6SPhilipp Reisner } 5921b411b363SPhilipp Reisner if (received == expect) { 59222735a594SAndreas Gruenbacher bool err; 5923a4fbda8eSPhilipp Reisner 5924bde89a9eSAndreas Gruenbacher err = cmd->fn(connection, &pi); 59252735a594SAndreas Gruenbacher if (err) { 59261ec861ebSAndreas Gruenbacher drbd_err(connection, "%pf failed\n", cmd->fn); 5927b411b363SPhilipp Reisner goto reconnect; 59281952e916SAndreas Gruenbacher } 5929b411b363SPhilipp Reisner 5930bde89a9eSAndreas Gruenbacher connection->last_received = jiffies; 5931f36af18cSLars Ellenberg 5932668700b4SPhilipp Reisner if (cmd == &ack_receiver_tbl[P_PING_ACK]) { 5933668700b4SPhilipp Reisner set_idle_timeout(connection); 593444ed167dSPhilipp Reisner ping_timeout_active = false; 593544ed167dSPhilipp Reisner } 5936b411b363SPhilipp Reisner 5937bde89a9eSAndreas Gruenbacher buf = connection->meta.rbuf; 5938b411b363SPhilipp Reisner received = 0; 593952b061a4SAndreas Gruenbacher expect = header_size; 5940b411b363SPhilipp Reisner cmd = NULL; 5941b411b363SPhilipp Reisner } 5942b411b363SPhilipp Reisner } 5943b411b363SPhilipp Reisner 5944b411b363SPhilipp Reisner if (0) { 5945b411b363SPhilipp Reisner reconnect: 5946bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); 5947bde89a9eSAndreas Gruenbacher conn_md_sync(connection); 5948b411b363SPhilipp Reisner } 5949b411b363SPhilipp Reisner if (0) { 5950b411b363SPhilipp Reisner disconnect: 5951bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); 5952b411b363SPhilipp Reisner } 5953b411b363SPhilipp Reisner 5954668700b4SPhilipp Reisner drbd_info(connection, "ack_receiver terminated\n"); 5955b411b363SPhilipp Reisner 5956b411b363SPhilipp Reisner return 0; 5957b411b363SPhilipp Reisner } 5958668700b4SPhilipp Reisner 5959668700b4SPhilipp Reisner void drbd_send_acks_wf(struct work_struct *ws) 5960668700b4SPhilipp Reisner { 5961668700b4SPhilipp Reisner struct drbd_peer_device *peer_device = 5962668700b4SPhilipp Reisner container_of(ws, struct drbd_peer_device, send_acks_work); 5963668700b4SPhilipp Reisner struct drbd_connection *connection = peer_device->connection; 5964668700b4SPhilipp Reisner struct drbd_device *device = peer_device->device; 5965668700b4SPhilipp Reisner struct net_conf *nc; 5966668700b4SPhilipp Reisner int tcp_cork, err; 5967668700b4SPhilipp Reisner 5968668700b4SPhilipp Reisner rcu_read_lock(); 5969668700b4SPhilipp Reisner nc = rcu_dereference(connection->net_conf); 5970668700b4SPhilipp Reisner tcp_cork = nc->tcp_cork; 5971668700b4SPhilipp Reisner rcu_read_unlock(); 5972668700b4SPhilipp Reisner 5973668700b4SPhilipp Reisner if (tcp_cork) 5974668700b4SPhilipp Reisner drbd_tcp_cork(connection->meta.socket); 5975668700b4SPhilipp Reisner 5976668700b4SPhilipp Reisner err = drbd_finish_peer_reqs(device); 5977668700b4SPhilipp Reisner kref_put(&device->kref, drbd_destroy_device); 5978668700b4SPhilipp Reisner /* get is in drbd_endio_write_sec_final(). That is necessary to keep the 5979668700b4SPhilipp Reisner struct work_struct send_acks_work alive, which is in the peer_device object */ 5980668700b4SPhilipp Reisner 5981668700b4SPhilipp Reisner if (err) { 5982668700b4SPhilipp Reisner conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); 5983668700b4SPhilipp Reisner return; 5984668700b4SPhilipp Reisner } 5985668700b4SPhilipp Reisner 5986668700b4SPhilipp Reisner if (tcp_cork) 5987668700b4SPhilipp Reisner drbd_tcp_uncork(connection->meta.socket); 5988668700b4SPhilipp Reisner 5989668700b4SPhilipp Reisner return; 5990668700b4SPhilipp Reisner } 5991