1b411b363SPhilipp Reisner /* 2b411b363SPhilipp Reisner drbd_receiver.c 3b411b363SPhilipp Reisner 4b411b363SPhilipp Reisner This file is part of DRBD by Philipp Reisner and Lars Ellenberg. 5b411b363SPhilipp Reisner 6b411b363SPhilipp Reisner Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. 7b411b363SPhilipp Reisner Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. 8b411b363SPhilipp Reisner Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. 9b411b363SPhilipp Reisner 10b411b363SPhilipp Reisner drbd is free software; you can redistribute it and/or modify 11b411b363SPhilipp Reisner it under the terms of the GNU General Public License as published by 12b411b363SPhilipp Reisner the Free Software Foundation; either version 2, or (at your option) 13b411b363SPhilipp Reisner any later version. 14b411b363SPhilipp Reisner 15b411b363SPhilipp Reisner drbd is distributed in the hope that it will be useful, 16b411b363SPhilipp Reisner but WITHOUT ANY WARRANTY; without even the implied warranty of 17b411b363SPhilipp Reisner MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 18b411b363SPhilipp Reisner GNU General Public License for more details. 19b411b363SPhilipp Reisner 20b411b363SPhilipp Reisner You should have received a copy of the GNU General Public License 21b411b363SPhilipp Reisner along with drbd; see the file COPYING. If not, write to 22b411b363SPhilipp Reisner the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. 23b411b363SPhilipp Reisner */ 24b411b363SPhilipp Reisner 25b411b363SPhilipp Reisner 26b411b363SPhilipp Reisner #include <linux/module.h> 27b411b363SPhilipp Reisner 28b411b363SPhilipp Reisner #include <asm/uaccess.h> 29b411b363SPhilipp Reisner #include <net/sock.h> 30b411b363SPhilipp Reisner 31b411b363SPhilipp Reisner #include <linux/drbd.h> 32b411b363SPhilipp Reisner #include <linux/fs.h> 33b411b363SPhilipp Reisner #include <linux/file.h> 34b411b363SPhilipp Reisner #include <linux/in.h> 35b411b363SPhilipp Reisner #include <linux/mm.h> 36b411b363SPhilipp Reisner #include <linux/memcontrol.h> 37b411b363SPhilipp Reisner #include <linux/mm_inline.h> 38b411b363SPhilipp Reisner #include <linux/slab.h> 39b411b363SPhilipp Reisner #include <linux/pkt_sched.h> 40b411b363SPhilipp Reisner #define __KERNEL_SYSCALLS__ 41b411b363SPhilipp Reisner #include <linux/unistd.h> 42b411b363SPhilipp Reisner #include <linux/vmalloc.h> 43b411b363SPhilipp Reisner #include <linux/random.h> 44b411b363SPhilipp Reisner #include <linux/string.h> 45b411b363SPhilipp Reisner #include <linux/scatterlist.h> 46b411b363SPhilipp Reisner #include "drbd_int.h" 47a3603a6eSAndreas Gruenbacher #include "drbd_protocol.h" 48b411b363SPhilipp Reisner #include "drbd_req.h" 49b411b363SPhilipp Reisner 50b411b363SPhilipp Reisner #include "drbd_vli.h" 51b411b363SPhilipp Reisner 5277351055SPhilipp Reisner struct packet_info { 5377351055SPhilipp Reisner enum drbd_packet cmd; 54e2857216SAndreas Gruenbacher unsigned int size; 55e2857216SAndreas Gruenbacher unsigned int vnr; 56e658983aSAndreas Gruenbacher void *data; 5777351055SPhilipp Reisner }; 5877351055SPhilipp Reisner 59b411b363SPhilipp Reisner enum finish_epoch { 60b411b363SPhilipp Reisner FE_STILL_LIVE, 61b411b363SPhilipp Reisner FE_DESTROYED, 62b411b363SPhilipp Reisner FE_RECYCLED, 63b411b363SPhilipp Reisner }; 64b411b363SPhilipp Reisner 65bde89a9eSAndreas Gruenbacher static int drbd_do_features(struct drbd_connection *connection); 66bde89a9eSAndreas Gruenbacher static int drbd_do_auth(struct drbd_connection *connection); 67b30ab791SAndreas Gruenbacher static int drbd_disconnected(struct drbd_device *device); 68b411b363SPhilipp Reisner 69bde89a9eSAndreas Gruenbacher static enum finish_epoch drbd_may_finish_epoch(struct drbd_connection *, struct drbd_epoch *, enum epoch_event); 7099920dc5SAndreas Gruenbacher static int e_end_block(struct drbd_work *, int); 71b411b363SPhilipp Reisner 72b411b363SPhilipp Reisner 73b411b363SPhilipp Reisner #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) 74b411b363SPhilipp Reisner 7545bb912bSLars Ellenberg /* 7645bb912bSLars Ellenberg * some helper functions to deal with single linked page lists, 7745bb912bSLars Ellenberg * page->private being our "next" pointer. 7845bb912bSLars Ellenberg */ 7945bb912bSLars Ellenberg 8045bb912bSLars Ellenberg /* If at least n pages are linked at head, get n pages off. 8145bb912bSLars Ellenberg * Otherwise, don't modify head, and return NULL. 8245bb912bSLars Ellenberg * Locking is the responsibility of the caller. 8345bb912bSLars Ellenberg */ 8445bb912bSLars Ellenberg static struct page *page_chain_del(struct page **head, int n) 8545bb912bSLars Ellenberg { 8645bb912bSLars Ellenberg struct page *page; 8745bb912bSLars Ellenberg struct page *tmp; 8845bb912bSLars Ellenberg 8945bb912bSLars Ellenberg BUG_ON(!n); 9045bb912bSLars Ellenberg BUG_ON(!head); 9145bb912bSLars Ellenberg 9245bb912bSLars Ellenberg page = *head; 9323ce4227SPhilipp Reisner 9423ce4227SPhilipp Reisner if (!page) 9523ce4227SPhilipp Reisner return NULL; 9623ce4227SPhilipp Reisner 9745bb912bSLars Ellenberg while (page) { 9845bb912bSLars Ellenberg tmp = page_chain_next(page); 9945bb912bSLars Ellenberg if (--n == 0) 10045bb912bSLars Ellenberg break; /* found sufficient pages */ 10145bb912bSLars Ellenberg if (tmp == NULL) 10245bb912bSLars Ellenberg /* insufficient pages, don't use any of them. */ 10345bb912bSLars Ellenberg return NULL; 10445bb912bSLars Ellenberg page = tmp; 10545bb912bSLars Ellenberg } 10645bb912bSLars Ellenberg 10745bb912bSLars Ellenberg /* add end of list marker for the returned list */ 10845bb912bSLars Ellenberg set_page_private(page, 0); 10945bb912bSLars Ellenberg /* actual return value, and adjustment of head */ 11045bb912bSLars Ellenberg page = *head; 11145bb912bSLars Ellenberg *head = tmp; 11245bb912bSLars Ellenberg return page; 11345bb912bSLars Ellenberg } 11445bb912bSLars Ellenberg 11545bb912bSLars Ellenberg /* may be used outside of locks to find the tail of a (usually short) 11645bb912bSLars Ellenberg * "private" page chain, before adding it back to a global chain head 11745bb912bSLars Ellenberg * with page_chain_add() under a spinlock. */ 11845bb912bSLars Ellenberg static struct page *page_chain_tail(struct page *page, int *len) 11945bb912bSLars Ellenberg { 12045bb912bSLars Ellenberg struct page *tmp; 12145bb912bSLars Ellenberg int i = 1; 12245bb912bSLars Ellenberg while ((tmp = page_chain_next(page))) 12345bb912bSLars Ellenberg ++i, page = tmp; 12445bb912bSLars Ellenberg if (len) 12545bb912bSLars Ellenberg *len = i; 12645bb912bSLars Ellenberg return page; 12745bb912bSLars Ellenberg } 12845bb912bSLars Ellenberg 12945bb912bSLars Ellenberg static int page_chain_free(struct page *page) 13045bb912bSLars Ellenberg { 13145bb912bSLars Ellenberg struct page *tmp; 13245bb912bSLars Ellenberg int i = 0; 13345bb912bSLars Ellenberg page_chain_for_each_safe(page, tmp) { 13445bb912bSLars Ellenberg put_page(page); 13545bb912bSLars Ellenberg ++i; 13645bb912bSLars Ellenberg } 13745bb912bSLars Ellenberg return i; 13845bb912bSLars Ellenberg } 13945bb912bSLars Ellenberg 14045bb912bSLars Ellenberg static void page_chain_add(struct page **head, 14145bb912bSLars Ellenberg struct page *chain_first, struct page *chain_last) 14245bb912bSLars Ellenberg { 14345bb912bSLars Ellenberg #if 1 14445bb912bSLars Ellenberg struct page *tmp; 14545bb912bSLars Ellenberg tmp = page_chain_tail(chain_first, NULL); 14645bb912bSLars Ellenberg BUG_ON(tmp != chain_last); 14745bb912bSLars Ellenberg #endif 14845bb912bSLars Ellenberg 14945bb912bSLars Ellenberg /* add chain to head */ 15045bb912bSLars Ellenberg set_page_private(chain_last, (unsigned long)*head); 15145bb912bSLars Ellenberg *head = chain_first; 15245bb912bSLars Ellenberg } 15345bb912bSLars Ellenberg 154b30ab791SAndreas Gruenbacher static struct page *__drbd_alloc_pages(struct drbd_device *device, 15518c2d522SAndreas Gruenbacher unsigned int number) 156b411b363SPhilipp Reisner { 157b411b363SPhilipp Reisner struct page *page = NULL; 15845bb912bSLars Ellenberg struct page *tmp = NULL; 15918c2d522SAndreas Gruenbacher unsigned int i = 0; 160b411b363SPhilipp Reisner 161b411b363SPhilipp Reisner /* Yes, testing drbd_pp_vacant outside the lock is racy. 162b411b363SPhilipp Reisner * So what. It saves a spin_lock. */ 16345bb912bSLars Ellenberg if (drbd_pp_vacant >= number) { 164b411b363SPhilipp Reisner spin_lock(&drbd_pp_lock); 16545bb912bSLars Ellenberg page = page_chain_del(&drbd_pp_pool, number); 16645bb912bSLars Ellenberg if (page) 16745bb912bSLars Ellenberg drbd_pp_vacant -= number; 168b411b363SPhilipp Reisner spin_unlock(&drbd_pp_lock); 16945bb912bSLars Ellenberg if (page) 17045bb912bSLars Ellenberg return page; 171b411b363SPhilipp Reisner } 17245bb912bSLars Ellenberg 173b411b363SPhilipp Reisner /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD 174b411b363SPhilipp Reisner * "criss-cross" setup, that might cause write-out on some other DRBD, 175b411b363SPhilipp Reisner * which in turn might block on the other node at this very place. */ 17645bb912bSLars Ellenberg for (i = 0; i < number; i++) { 17745bb912bSLars Ellenberg tmp = alloc_page(GFP_TRY); 17845bb912bSLars Ellenberg if (!tmp) 17945bb912bSLars Ellenberg break; 18045bb912bSLars Ellenberg set_page_private(tmp, (unsigned long)page); 18145bb912bSLars Ellenberg page = tmp; 18245bb912bSLars Ellenberg } 18345bb912bSLars Ellenberg 18445bb912bSLars Ellenberg if (i == number) 185b411b363SPhilipp Reisner return page; 18645bb912bSLars Ellenberg 18745bb912bSLars Ellenberg /* Not enough pages immediately available this time. 188c37c8ecfSAndreas Gruenbacher * No need to jump around here, drbd_alloc_pages will retry this 18945bb912bSLars Ellenberg * function "soon". */ 19045bb912bSLars Ellenberg if (page) { 19145bb912bSLars Ellenberg tmp = page_chain_tail(page, NULL); 19245bb912bSLars Ellenberg spin_lock(&drbd_pp_lock); 19345bb912bSLars Ellenberg page_chain_add(&drbd_pp_pool, page, tmp); 19445bb912bSLars Ellenberg drbd_pp_vacant += i; 19545bb912bSLars Ellenberg spin_unlock(&drbd_pp_lock); 19645bb912bSLars Ellenberg } 19745bb912bSLars Ellenberg return NULL; 198b411b363SPhilipp Reisner } 199b411b363SPhilipp Reisner 200b30ab791SAndreas Gruenbacher static void reclaim_finished_net_peer_reqs(struct drbd_device *device, 201a990be46SAndreas Gruenbacher struct list_head *to_be_freed) 202b411b363SPhilipp Reisner { 203db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req; 204b411b363SPhilipp Reisner struct list_head *le, *tle; 205b411b363SPhilipp Reisner 206b411b363SPhilipp Reisner /* The EEs are always appended to the end of the list. Since 207b411b363SPhilipp Reisner they are sent in order over the wire, they have to finish 208b411b363SPhilipp Reisner in order. As soon as we see the first not finished we can 209b411b363SPhilipp Reisner stop to examine the list... */ 210b411b363SPhilipp Reisner 211b30ab791SAndreas Gruenbacher list_for_each_safe(le, tle, &device->net_ee) { 212db830c46SAndreas Gruenbacher peer_req = list_entry(le, struct drbd_peer_request, w.list); 213045417f7SAndreas Gruenbacher if (drbd_peer_req_has_active_page(peer_req)) 214b411b363SPhilipp Reisner break; 215b411b363SPhilipp Reisner list_move(le, to_be_freed); 216b411b363SPhilipp Reisner } 217b411b363SPhilipp Reisner } 218b411b363SPhilipp Reisner 219b30ab791SAndreas Gruenbacher static void drbd_kick_lo_and_reclaim_net(struct drbd_device *device) 220b411b363SPhilipp Reisner { 221b411b363SPhilipp Reisner LIST_HEAD(reclaimed); 222db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req, *t; 223b411b363SPhilipp Reisner 2240500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 225b30ab791SAndreas Gruenbacher reclaim_finished_net_peer_reqs(device, &reclaimed); 2260500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 227b411b363SPhilipp Reisner 228db830c46SAndreas 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 232b411b363SPhilipp Reisner /** 233c37c8ecfSAndreas Gruenbacher * drbd_alloc_pages() - Returns @number pages, retries forever (or until signalled) 234b30ab791SAndreas Gruenbacher * @device: DRBD device. 23545bb912bSLars Ellenberg * @number: number of pages requested 23645bb912bSLars Ellenberg * @retry: whether to retry, if not enough pages are available right now 237b411b363SPhilipp Reisner * 23845bb912bSLars Ellenberg * Tries to allocate number pages, first from our own page pool, then from 23945bb912bSLars Ellenberg * the kernel, unless this allocation would exceed the max_buffers setting. 24045bb912bSLars Ellenberg * Possibly retry until DRBD frees sufficient pages somewhere else. 24145bb912bSLars Ellenberg * 24245bb912bSLars Ellenberg * Returns a page chain linked via page->private. 243b411b363SPhilipp Reisner */ 244b30ab791SAndreas Gruenbacher struct page *drbd_alloc_pages(struct drbd_device *device, unsigned int number, 245c37c8ecfSAndreas Gruenbacher bool retry) 246b411b363SPhilipp Reisner { 247b411b363SPhilipp Reisner struct page *page = NULL; 24844ed167dSPhilipp Reisner struct net_conf *nc; 249b411b363SPhilipp Reisner DEFINE_WAIT(wait); 25044ed167dSPhilipp Reisner int mxb; 251b411b363SPhilipp Reisner 25245bb912bSLars Ellenberg /* Yes, we may run up to @number over max_buffers. If we 25345bb912bSLars Ellenberg * follow it strictly, the admin will get it wrong anyways. */ 25444ed167dSPhilipp Reisner rcu_read_lock(); 255a6b32bc3SAndreas Gruenbacher nc = rcu_dereference(first_peer_device(device)->connection->net_conf); 25644ed167dSPhilipp Reisner mxb = nc ? nc->max_buffers : 1000000; 25744ed167dSPhilipp Reisner rcu_read_unlock(); 25844ed167dSPhilipp Reisner 259b30ab791SAndreas Gruenbacher if (atomic_read(&device->pp_in_use) < mxb) 260b30ab791SAndreas Gruenbacher page = __drbd_alloc_pages(device, number); 261b411b363SPhilipp Reisner 26245bb912bSLars Ellenberg while (page == NULL) { 263b411b363SPhilipp Reisner prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); 264b411b363SPhilipp Reisner 265b30ab791SAndreas Gruenbacher drbd_kick_lo_and_reclaim_net(device); 266b411b363SPhilipp Reisner 267b30ab791SAndreas Gruenbacher if (atomic_read(&device->pp_in_use) < mxb) { 268b30ab791SAndreas Gruenbacher page = __drbd_alloc_pages(device, number); 269b411b363SPhilipp Reisner if (page) 270b411b363SPhilipp Reisner break; 271b411b363SPhilipp Reisner } 272b411b363SPhilipp Reisner 273b411b363SPhilipp Reisner if (!retry) 274b411b363SPhilipp Reisner break; 275b411b363SPhilipp Reisner 276b411b363SPhilipp Reisner if (signal_pending(current)) { 277d0180171SAndreas Gruenbacher drbd_warn(device, "drbd_alloc_pages interrupted!\n"); 278b411b363SPhilipp Reisner break; 279b411b363SPhilipp Reisner } 280b411b363SPhilipp Reisner 281b411b363SPhilipp Reisner schedule(); 282b411b363SPhilipp Reisner } 283b411b363SPhilipp Reisner finish_wait(&drbd_pp_wait, &wait); 284b411b363SPhilipp Reisner 28545bb912bSLars Ellenberg if (page) 286b30ab791SAndreas Gruenbacher atomic_add(number, &device->pp_in_use); 287b411b363SPhilipp Reisner return page; 288b411b363SPhilipp Reisner } 289b411b363SPhilipp Reisner 290c37c8ecfSAndreas Gruenbacher /* Must not be used from irq, as that may deadlock: see drbd_alloc_pages. 2910500813fSAndreas Gruenbacher * Is also used from inside an other spin_lock_irq(&resource->req_lock); 29245bb912bSLars Ellenberg * Either links the page chain back to the global pool, 29345bb912bSLars Ellenberg * or returns all pages to the system. */ 294b30ab791SAndreas Gruenbacher static void drbd_free_pages(struct drbd_device *device, struct page *page, int is_net) 295b411b363SPhilipp Reisner { 296b30ab791SAndreas Gruenbacher atomic_t *a = is_net ? &device->pp_in_use_by_net : &device->pp_in_use; 297b411b363SPhilipp Reisner int i; 298435f0740SLars Ellenberg 299a73ff323SLars Ellenberg if (page == NULL) 300a73ff323SLars Ellenberg return; 301a73ff323SLars Ellenberg 3021816a2b4SLars Ellenberg if (drbd_pp_vacant > (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count) 30345bb912bSLars Ellenberg i = page_chain_free(page); 30445bb912bSLars Ellenberg else { 30545bb912bSLars Ellenberg struct page *tmp; 30645bb912bSLars Ellenberg tmp = page_chain_tail(page, &i); 307b411b363SPhilipp Reisner spin_lock(&drbd_pp_lock); 30845bb912bSLars Ellenberg page_chain_add(&drbd_pp_pool, page, tmp); 30945bb912bSLars Ellenberg drbd_pp_vacant += i; 310b411b363SPhilipp Reisner spin_unlock(&drbd_pp_lock); 311b411b363SPhilipp Reisner } 312435f0740SLars Ellenberg i = atomic_sub_return(i, a); 31345bb912bSLars Ellenberg if (i < 0) 314d0180171SAndreas Gruenbacher drbd_warn(device, "ASSERTION FAILED: %s: %d < 0\n", 315435f0740SLars Ellenberg is_net ? "pp_in_use_by_net" : "pp_in_use", i); 316b411b363SPhilipp Reisner wake_up(&drbd_pp_wait); 317b411b363SPhilipp Reisner } 318b411b363SPhilipp Reisner 319b411b363SPhilipp Reisner /* 320b411b363SPhilipp Reisner You need to hold the req_lock: 321b411b363SPhilipp Reisner _drbd_wait_ee_list_empty() 322b411b363SPhilipp Reisner 323b411b363SPhilipp Reisner You must not have the req_lock: 3243967deb1SAndreas Gruenbacher drbd_free_peer_req() 3250db55363SAndreas Gruenbacher drbd_alloc_peer_req() 3267721f567SAndreas Gruenbacher drbd_free_peer_reqs() 327b411b363SPhilipp Reisner drbd_ee_fix_bhs() 328a990be46SAndreas Gruenbacher drbd_finish_peer_reqs() 329b411b363SPhilipp Reisner drbd_clear_done_ee() 330b411b363SPhilipp Reisner drbd_wait_ee_list_empty() 331b411b363SPhilipp Reisner */ 332b411b363SPhilipp Reisner 333f6ffca9fSAndreas Gruenbacher struct drbd_peer_request * 334b30ab791SAndreas Gruenbacher drbd_alloc_peer_req(struct drbd_device *device, u64 id, sector_t sector, 335f6ffca9fSAndreas Gruenbacher unsigned int data_size, gfp_t gfp_mask) __must_hold(local) 336b411b363SPhilipp Reisner { 337db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req; 338a73ff323SLars Ellenberg struct page *page = NULL; 33945bb912bSLars Ellenberg unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; 340b411b363SPhilipp Reisner 341b30ab791SAndreas Gruenbacher if (drbd_insert_fault(device, DRBD_FAULT_AL_EE)) 342b411b363SPhilipp Reisner return NULL; 343b411b363SPhilipp Reisner 344db830c46SAndreas Gruenbacher peer_req = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); 345db830c46SAndreas Gruenbacher if (!peer_req) { 346b411b363SPhilipp Reisner if (!(gfp_mask & __GFP_NOWARN)) 347d0180171SAndreas Gruenbacher drbd_err(device, "%s: allocation failed\n", __func__); 348b411b363SPhilipp Reisner return NULL; 349b411b363SPhilipp Reisner } 350b411b363SPhilipp Reisner 351a73ff323SLars Ellenberg if (data_size) { 352b30ab791SAndreas Gruenbacher page = drbd_alloc_pages(device, nr_pages, (gfp_mask & __GFP_WAIT)); 35345bb912bSLars Ellenberg if (!page) 35445bb912bSLars Ellenberg goto fail; 355a73ff323SLars Ellenberg } 356b411b363SPhilipp Reisner 357db830c46SAndreas Gruenbacher drbd_clear_interval(&peer_req->i); 358db830c46SAndreas Gruenbacher peer_req->i.size = data_size; 359db830c46SAndreas Gruenbacher peer_req->i.sector = sector; 360db830c46SAndreas Gruenbacher peer_req->i.local = false; 361db830c46SAndreas Gruenbacher peer_req->i.waiting = false; 362b411b363SPhilipp Reisner 363db830c46SAndreas Gruenbacher peer_req->epoch = NULL; 364b30ab791SAndreas Gruenbacher peer_req->w.device = device; 365db830c46SAndreas Gruenbacher peer_req->pages = page; 366db830c46SAndreas Gruenbacher atomic_set(&peer_req->pending_bios, 0); 367db830c46SAndreas Gruenbacher peer_req->flags = 0; 3689a8e7753SAndreas Gruenbacher /* 3699a8e7753SAndreas Gruenbacher * The block_id is opaque to the receiver. It is not endianness 3709a8e7753SAndreas Gruenbacher * converted, and sent back to the sender unchanged. 3719a8e7753SAndreas Gruenbacher */ 372db830c46SAndreas Gruenbacher peer_req->block_id = id; 373b411b363SPhilipp Reisner 374db830c46SAndreas Gruenbacher return peer_req; 375b411b363SPhilipp Reisner 37645bb912bSLars Ellenberg fail: 377db830c46SAndreas Gruenbacher mempool_free(peer_req, drbd_ee_mempool); 378b411b363SPhilipp Reisner return NULL; 379b411b363SPhilipp Reisner } 380b411b363SPhilipp Reisner 381b30ab791SAndreas Gruenbacher void __drbd_free_peer_req(struct drbd_device *device, struct drbd_peer_request *peer_req, 382f6ffca9fSAndreas Gruenbacher int is_net) 383b411b363SPhilipp Reisner { 384db830c46SAndreas Gruenbacher if (peer_req->flags & EE_HAS_DIGEST) 385db830c46SAndreas Gruenbacher kfree(peer_req->digest); 386b30ab791SAndreas Gruenbacher drbd_free_pages(device, peer_req->pages, is_net); 3870b0ba1efSAndreas Gruenbacher D_ASSERT(device, atomic_read(&peer_req->pending_bios) == 0); 3880b0ba1efSAndreas Gruenbacher D_ASSERT(device, drbd_interval_empty(&peer_req->i)); 389db830c46SAndreas Gruenbacher mempool_free(peer_req, drbd_ee_mempool); 390b411b363SPhilipp Reisner } 391b411b363SPhilipp Reisner 392b30ab791SAndreas Gruenbacher int drbd_free_peer_reqs(struct drbd_device *device, struct list_head *list) 393b411b363SPhilipp Reisner { 394b411b363SPhilipp Reisner LIST_HEAD(work_list); 395db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req, *t; 396b411b363SPhilipp Reisner int count = 0; 397b30ab791SAndreas Gruenbacher int is_net = list == &device->net_ee; 398b411b363SPhilipp Reisner 3990500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 400b411b363SPhilipp Reisner list_splice_init(list, &work_list); 4010500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 402b411b363SPhilipp Reisner 403db830c46SAndreas Gruenbacher list_for_each_entry_safe(peer_req, t, &work_list, w.list) { 404b30ab791SAndreas Gruenbacher __drbd_free_peer_req(device, peer_req, is_net); 405b411b363SPhilipp Reisner count++; 406b411b363SPhilipp Reisner } 407b411b363SPhilipp Reisner return count; 408b411b363SPhilipp Reisner } 409b411b363SPhilipp Reisner 410b411b363SPhilipp Reisner /* 411a990be46SAndreas Gruenbacher * See also comments in _req_mod(,BARRIER_ACKED) and receive_Barrier. 412b411b363SPhilipp Reisner */ 413b30ab791SAndreas Gruenbacher static int drbd_finish_peer_reqs(struct drbd_device *device) 414b411b363SPhilipp Reisner { 415b411b363SPhilipp Reisner LIST_HEAD(work_list); 416b411b363SPhilipp Reisner LIST_HEAD(reclaimed); 417db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req, *t; 418e2b3032bSAndreas Gruenbacher int err = 0; 419b411b363SPhilipp Reisner 4200500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 421b30ab791SAndreas Gruenbacher reclaim_finished_net_peer_reqs(device, &reclaimed); 422b30ab791SAndreas Gruenbacher list_splice_init(&device->done_ee, &work_list); 4230500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 424b411b363SPhilipp Reisner 425db830c46SAndreas Gruenbacher list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) 426b30ab791SAndreas Gruenbacher drbd_free_net_peer_req(device, peer_req); 427b411b363SPhilipp Reisner 428b411b363SPhilipp Reisner /* possible callbacks here: 429d4dabbe2SLars Ellenberg * e_end_block, and e_end_resync_block, e_send_superseded. 430b411b363SPhilipp Reisner * all ignore the last argument. 431b411b363SPhilipp Reisner */ 432db830c46SAndreas Gruenbacher list_for_each_entry_safe(peer_req, t, &work_list, w.list) { 433e2b3032bSAndreas Gruenbacher int err2; 434e2b3032bSAndreas Gruenbacher 435b411b363SPhilipp Reisner /* list_del not necessary, next/prev members not touched */ 436e2b3032bSAndreas Gruenbacher err2 = peer_req->w.cb(&peer_req->w, !!err); 437e2b3032bSAndreas Gruenbacher if (!err) 438e2b3032bSAndreas Gruenbacher err = err2; 439b30ab791SAndreas Gruenbacher drbd_free_peer_req(device, peer_req); 440b411b363SPhilipp Reisner } 441b30ab791SAndreas Gruenbacher wake_up(&device->ee_wait); 442b411b363SPhilipp Reisner 443e2b3032bSAndreas Gruenbacher return err; 444b411b363SPhilipp Reisner } 445b411b363SPhilipp Reisner 446b30ab791SAndreas Gruenbacher static void _drbd_wait_ee_list_empty(struct drbd_device *device, 447d4da1537SAndreas Gruenbacher struct list_head *head) 448b411b363SPhilipp Reisner { 449b411b363SPhilipp Reisner DEFINE_WAIT(wait); 450b411b363SPhilipp Reisner 451b411b363SPhilipp Reisner /* avoids spin_lock/unlock 452b411b363SPhilipp Reisner * and calling prepare_to_wait in the fast path */ 453b411b363SPhilipp Reisner while (!list_empty(head)) { 454b30ab791SAndreas Gruenbacher prepare_to_wait(&device->ee_wait, &wait, TASK_UNINTERRUPTIBLE); 4550500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 4567eaceaccSJens Axboe io_schedule(); 457b30ab791SAndreas Gruenbacher finish_wait(&device->ee_wait, &wait); 4580500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 459b411b363SPhilipp Reisner } 460b411b363SPhilipp Reisner } 461b411b363SPhilipp Reisner 462b30ab791SAndreas Gruenbacher static void drbd_wait_ee_list_empty(struct drbd_device *device, 463d4da1537SAndreas Gruenbacher struct list_head *head) 464b411b363SPhilipp Reisner { 4650500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 466b30ab791SAndreas Gruenbacher _drbd_wait_ee_list_empty(device, head); 4670500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 468b411b363SPhilipp Reisner } 469b411b363SPhilipp Reisner 470dbd9eea0SPhilipp Reisner static int drbd_recv_short(struct socket *sock, void *buf, size_t size, int flags) 471b411b363SPhilipp Reisner { 472b411b363SPhilipp Reisner mm_segment_t oldfs; 473b411b363SPhilipp Reisner struct kvec iov = { 474b411b363SPhilipp Reisner .iov_base = buf, 475b411b363SPhilipp Reisner .iov_len = size, 476b411b363SPhilipp Reisner }; 477b411b363SPhilipp Reisner struct msghdr msg = { 478b411b363SPhilipp Reisner .msg_iovlen = 1, 479b411b363SPhilipp Reisner .msg_iov = (struct iovec *)&iov, 480b411b363SPhilipp Reisner .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) 481b411b363SPhilipp Reisner }; 482b411b363SPhilipp Reisner int rv; 483b411b363SPhilipp Reisner 484b411b363SPhilipp Reisner oldfs = get_fs(); 485b411b363SPhilipp Reisner set_fs(KERNEL_DS); 486b411b363SPhilipp Reisner rv = sock_recvmsg(sock, &msg, size, msg.msg_flags); 487b411b363SPhilipp Reisner set_fs(oldfs); 488b411b363SPhilipp Reisner 489b411b363SPhilipp Reisner return rv; 490b411b363SPhilipp Reisner } 491b411b363SPhilipp Reisner 492bde89a9eSAndreas Gruenbacher static int drbd_recv(struct drbd_connection *connection, void *buf, size_t size) 493b411b363SPhilipp Reisner { 494b411b363SPhilipp Reisner int rv; 495b411b363SPhilipp Reisner 496bde89a9eSAndreas Gruenbacher rv = drbd_recv_short(connection->data.socket, buf, size, 0); 497b411b363SPhilipp Reisner 498b411b363SPhilipp Reisner if (rv < 0) { 499b411b363SPhilipp Reisner if (rv == -ECONNRESET) 5001ec861ebSAndreas Gruenbacher drbd_info(connection, "sock was reset by peer\n"); 501b411b363SPhilipp Reisner else if (rv != -ERESTARTSYS) 5021ec861ebSAndreas Gruenbacher drbd_err(connection, "sock_recvmsg returned %d\n", rv); 503b411b363SPhilipp Reisner } else if (rv == 0) { 504bde89a9eSAndreas Gruenbacher if (test_bit(DISCONNECT_SENT, &connection->flags)) { 505b66623e3SPhilipp Reisner long t; 506b66623e3SPhilipp Reisner rcu_read_lock(); 507bde89a9eSAndreas Gruenbacher t = rcu_dereference(connection->net_conf)->ping_timeo * HZ/10; 508b66623e3SPhilipp Reisner rcu_read_unlock(); 509b66623e3SPhilipp Reisner 510bde89a9eSAndreas Gruenbacher t = wait_event_timeout(connection->ping_wait, connection->cstate < C_WF_REPORT_PARAMS, t); 511b66623e3SPhilipp Reisner 512599377acSPhilipp Reisner if (t) 513599377acSPhilipp Reisner goto out; 514599377acSPhilipp Reisner } 5151ec861ebSAndreas Gruenbacher drbd_info(connection, "sock was shut down by peer\n"); 516599377acSPhilipp Reisner } 517599377acSPhilipp Reisner 518b411b363SPhilipp Reisner if (rv != size) 519bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD); 520b411b363SPhilipp Reisner 521599377acSPhilipp Reisner out: 522b411b363SPhilipp Reisner return rv; 523b411b363SPhilipp Reisner } 524b411b363SPhilipp Reisner 525bde89a9eSAndreas Gruenbacher static int drbd_recv_all(struct drbd_connection *connection, void *buf, size_t size) 526c6967746SAndreas Gruenbacher { 527c6967746SAndreas Gruenbacher int err; 528c6967746SAndreas Gruenbacher 529bde89a9eSAndreas Gruenbacher err = drbd_recv(connection, buf, size); 530c6967746SAndreas Gruenbacher if (err != size) { 531c6967746SAndreas Gruenbacher if (err >= 0) 532c6967746SAndreas Gruenbacher err = -EIO; 533c6967746SAndreas Gruenbacher } else 534c6967746SAndreas Gruenbacher err = 0; 535c6967746SAndreas Gruenbacher return err; 536c6967746SAndreas Gruenbacher } 537c6967746SAndreas Gruenbacher 538bde89a9eSAndreas Gruenbacher static int drbd_recv_all_warn(struct drbd_connection *connection, void *buf, size_t size) 539a5c31904SAndreas Gruenbacher { 540a5c31904SAndreas Gruenbacher int err; 541a5c31904SAndreas Gruenbacher 542bde89a9eSAndreas Gruenbacher err = drbd_recv_all(connection, buf, size); 543a5c31904SAndreas Gruenbacher if (err && !signal_pending(current)) 5441ec861ebSAndreas Gruenbacher drbd_warn(connection, "short read (expected size %d)\n", (int)size); 545a5c31904SAndreas Gruenbacher return err; 546a5c31904SAndreas Gruenbacher } 547a5c31904SAndreas Gruenbacher 5485dbf1673SLars Ellenberg /* quoting tcp(7): 5495dbf1673SLars Ellenberg * On individual connections, the socket buffer size must be set prior to the 5505dbf1673SLars Ellenberg * listen(2) or connect(2) calls in order to have it take effect. 5515dbf1673SLars Ellenberg * This is our wrapper to do so. 5525dbf1673SLars Ellenberg */ 5535dbf1673SLars Ellenberg static void drbd_setbufsize(struct socket *sock, unsigned int snd, 5545dbf1673SLars Ellenberg unsigned int rcv) 5555dbf1673SLars Ellenberg { 5565dbf1673SLars Ellenberg /* open coded SO_SNDBUF, SO_RCVBUF */ 5575dbf1673SLars Ellenberg if (snd) { 5585dbf1673SLars Ellenberg sock->sk->sk_sndbuf = snd; 5595dbf1673SLars Ellenberg sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 5605dbf1673SLars Ellenberg } 5615dbf1673SLars Ellenberg if (rcv) { 5625dbf1673SLars Ellenberg sock->sk->sk_rcvbuf = rcv; 5635dbf1673SLars Ellenberg sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 5645dbf1673SLars Ellenberg } 5655dbf1673SLars Ellenberg } 5665dbf1673SLars Ellenberg 567bde89a9eSAndreas Gruenbacher static struct socket *drbd_try_connect(struct drbd_connection *connection) 568b411b363SPhilipp Reisner { 569b411b363SPhilipp Reisner const char *what; 570b411b363SPhilipp Reisner struct socket *sock; 571b411b363SPhilipp Reisner struct sockaddr_in6 src_in6; 57244ed167dSPhilipp Reisner struct sockaddr_in6 peer_in6; 57344ed167dSPhilipp Reisner struct net_conf *nc; 57444ed167dSPhilipp Reisner int err, peer_addr_len, my_addr_len; 57569ef82deSAndreas Gruenbacher int sndbuf_size, rcvbuf_size, connect_int; 576b411b363SPhilipp Reisner int disconnect_on_error = 1; 577b411b363SPhilipp Reisner 57844ed167dSPhilipp Reisner rcu_read_lock(); 579bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 58044ed167dSPhilipp Reisner if (!nc) { 58144ed167dSPhilipp Reisner rcu_read_unlock(); 582b411b363SPhilipp Reisner return NULL; 58344ed167dSPhilipp Reisner } 58444ed167dSPhilipp Reisner sndbuf_size = nc->sndbuf_size; 58544ed167dSPhilipp Reisner rcvbuf_size = nc->rcvbuf_size; 58669ef82deSAndreas Gruenbacher connect_int = nc->connect_int; 587089c075dSAndreas Gruenbacher rcu_read_unlock(); 58844ed167dSPhilipp Reisner 589bde89a9eSAndreas Gruenbacher my_addr_len = min_t(int, connection->my_addr_len, sizeof(src_in6)); 590bde89a9eSAndreas Gruenbacher memcpy(&src_in6, &connection->my_addr, my_addr_len); 59144ed167dSPhilipp Reisner 592bde89a9eSAndreas Gruenbacher if (((struct sockaddr *)&connection->my_addr)->sa_family == AF_INET6) 59344ed167dSPhilipp Reisner src_in6.sin6_port = 0; 59444ed167dSPhilipp Reisner else 59544ed167dSPhilipp Reisner ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ 59644ed167dSPhilipp Reisner 597bde89a9eSAndreas Gruenbacher peer_addr_len = min_t(int, connection->peer_addr_len, sizeof(src_in6)); 598bde89a9eSAndreas Gruenbacher memcpy(&peer_in6, &connection->peer_addr, peer_addr_len); 599b411b363SPhilipp Reisner 600b411b363SPhilipp Reisner what = "sock_create_kern"; 60144ed167dSPhilipp Reisner err = sock_create_kern(((struct sockaddr *)&src_in6)->sa_family, 602b411b363SPhilipp Reisner SOCK_STREAM, IPPROTO_TCP, &sock); 603b411b363SPhilipp Reisner if (err < 0) { 604b411b363SPhilipp Reisner sock = NULL; 605b411b363SPhilipp Reisner goto out; 606b411b363SPhilipp Reisner } 607b411b363SPhilipp Reisner 608b411b363SPhilipp Reisner sock->sk->sk_rcvtimeo = 60969ef82deSAndreas Gruenbacher sock->sk->sk_sndtimeo = connect_int * HZ; 61044ed167dSPhilipp Reisner drbd_setbufsize(sock, sndbuf_size, rcvbuf_size); 611b411b363SPhilipp Reisner 612b411b363SPhilipp Reisner /* explicitly bind to the configured IP as source IP 613b411b363SPhilipp Reisner * for the outgoing connections. 614b411b363SPhilipp Reisner * This is needed for multihomed hosts and to be 615b411b363SPhilipp Reisner * able to use lo: interfaces for drbd. 616b411b363SPhilipp Reisner * Make sure to use 0 as port number, so linux selects 617b411b363SPhilipp Reisner * a free one dynamically. 618b411b363SPhilipp Reisner */ 619b411b363SPhilipp Reisner what = "bind before connect"; 62044ed167dSPhilipp Reisner err = sock->ops->bind(sock, (struct sockaddr *) &src_in6, my_addr_len); 621b411b363SPhilipp Reisner if (err < 0) 622b411b363SPhilipp Reisner goto out; 623b411b363SPhilipp Reisner 624b411b363SPhilipp Reisner /* connect may fail, peer not yet available. 625b411b363SPhilipp Reisner * stay C_WF_CONNECTION, don't go Disconnecting! */ 626b411b363SPhilipp Reisner disconnect_on_error = 0; 627b411b363SPhilipp Reisner what = "connect"; 62844ed167dSPhilipp Reisner err = sock->ops->connect(sock, (struct sockaddr *) &peer_in6, peer_addr_len, 0); 629b411b363SPhilipp Reisner 630b411b363SPhilipp Reisner out: 631b411b363SPhilipp Reisner if (err < 0) { 632b411b363SPhilipp Reisner if (sock) { 633b411b363SPhilipp Reisner sock_release(sock); 634b411b363SPhilipp Reisner sock = NULL; 635b411b363SPhilipp Reisner } 636b411b363SPhilipp Reisner switch (-err) { 637b411b363SPhilipp Reisner /* timeout, busy, signal pending */ 638b411b363SPhilipp Reisner case ETIMEDOUT: case EAGAIN: case EINPROGRESS: 639b411b363SPhilipp Reisner case EINTR: case ERESTARTSYS: 640b411b363SPhilipp Reisner /* peer not (yet) available, network problem */ 641b411b363SPhilipp Reisner case ECONNREFUSED: case ENETUNREACH: 642b411b363SPhilipp Reisner case EHOSTDOWN: case EHOSTUNREACH: 643b411b363SPhilipp Reisner disconnect_on_error = 0; 644b411b363SPhilipp Reisner break; 645b411b363SPhilipp Reisner default: 6461ec861ebSAndreas Gruenbacher drbd_err(connection, "%s failed, err = %d\n", what, err); 647b411b363SPhilipp Reisner } 648b411b363SPhilipp Reisner if (disconnect_on_error) 649bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); 650b411b363SPhilipp Reisner } 65144ed167dSPhilipp Reisner 652b411b363SPhilipp Reisner return sock; 653b411b363SPhilipp Reisner } 654b411b363SPhilipp Reisner 6557a426fd8SPhilipp Reisner struct accept_wait_data { 656bde89a9eSAndreas Gruenbacher struct drbd_connection *connection; 6577a426fd8SPhilipp Reisner struct socket *s_listen; 6587a426fd8SPhilipp Reisner struct completion door_bell; 6597a426fd8SPhilipp Reisner void (*original_sk_state_change)(struct sock *sk); 6607a426fd8SPhilipp Reisner 6617a426fd8SPhilipp Reisner }; 6627a426fd8SPhilipp Reisner 663715306f6SAndreas Gruenbacher static void drbd_incoming_connection(struct sock *sk) 664b411b363SPhilipp Reisner { 6657a426fd8SPhilipp Reisner struct accept_wait_data *ad = sk->sk_user_data; 666715306f6SAndreas Gruenbacher void (*state_change)(struct sock *sk); 6677a426fd8SPhilipp Reisner 668715306f6SAndreas Gruenbacher state_change = ad->original_sk_state_change; 669715306f6SAndreas Gruenbacher if (sk->sk_state == TCP_ESTABLISHED) 6707a426fd8SPhilipp Reisner complete(&ad->door_bell); 671715306f6SAndreas Gruenbacher state_change(sk); 6727a426fd8SPhilipp Reisner } 6737a426fd8SPhilipp Reisner 674bde89a9eSAndreas Gruenbacher static int prepare_listen_socket(struct drbd_connection *connection, struct accept_wait_data *ad) 675b411b363SPhilipp Reisner { 6761f3e509bSPhilipp Reisner int err, sndbuf_size, rcvbuf_size, my_addr_len; 67744ed167dSPhilipp Reisner struct sockaddr_in6 my_addr; 6781f3e509bSPhilipp Reisner struct socket *s_listen; 67944ed167dSPhilipp Reisner struct net_conf *nc; 680b411b363SPhilipp Reisner const char *what; 681b411b363SPhilipp Reisner 68244ed167dSPhilipp Reisner rcu_read_lock(); 683bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 68444ed167dSPhilipp Reisner if (!nc) { 68544ed167dSPhilipp Reisner rcu_read_unlock(); 6867a426fd8SPhilipp Reisner return -EIO; 68744ed167dSPhilipp Reisner } 68844ed167dSPhilipp Reisner sndbuf_size = nc->sndbuf_size; 68944ed167dSPhilipp Reisner rcvbuf_size = nc->rcvbuf_size; 69044ed167dSPhilipp Reisner rcu_read_unlock(); 691b411b363SPhilipp Reisner 692bde89a9eSAndreas Gruenbacher my_addr_len = min_t(int, connection->my_addr_len, sizeof(struct sockaddr_in6)); 693bde89a9eSAndreas Gruenbacher memcpy(&my_addr, &connection->my_addr, my_addr_len); 694b411b363SPhilipp Reisner 695b411b363SPhilipp Reisner what = "sock_create_kern"; 69644ed167dSPhilipp Reisner err = sock_create_kern(((struct sockaddr *)&my_addr)->sa_family, 697b411b363SPhilipp Reisner SOCK_STREAM, IPPROTO_TCP, &s_listen); 698b411b363SPhilipp Reisner if (err) { 699b411b363SPhilipp Reisner s_listen = NULL; 700b411b363SPhilipp Reisner goto out; 701b411b363SPhilipp Reisner } 702b411b363SPhilipp Reisner 7034a17fd52SPavel Emelyanov s_listen->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ 70444ed167dSPhilipp Reisner drbd_setbufsize(s_listen, sndbuf_size, rcvbuf_size); 705b411b363SPhilipp Reisner 706b411b363SPhilipp Reisner what = "bind before listen"; 70744ed167dSPhilipp Reisner err = s_listen->ops->bind(s_listen, (struct sockaddr *)&my_addr, my_addr_len); 708b411b363SPhilipp Reisner if (err < 0) 709b411b363SPhilipp Reisner goto out; 710b411b363SPhilipp Reisner 7117a426fd8SPhilipp Reisner ad->s_listen = s_listen; 7127a426fd8SPhilipp Reisner write_lock_bh(&s_listen->sk->sk_callback_lock); 7137a426fd8SPhilipp Reisner ad->original_sk_state_change = s_listen->sk->sk_state_change; 714715306f6SAndreas Gruenbacher s_listen->sk->sk_state_change = drbd_incoming_connection; 7157a426fd8SPhilipp Reisner s_listen->sk->sk_user_data = ad; 7167a426fd8SPhilipp Reisner write_unlock_bh(&s_listen->sk->sk_callback_lock); 717b411b363SPhilipp Reisner 7182820fd39SPhilipp Reisner what = "listen"; 7192820fd39SPhilipp Reisner err = s_listen->ops->listen(s_listen, 5); 7202820fd39SPhilipp Reisner if (err < 0) 7212820fd39SPhilipp Reisner goto out; 7222820fd39SPhilipp Reisner 7237a426fd8SPhilipp Reisner return 0; 724b411b363SPhilipp Reisner out: 725b411b363SPhilipp Reisner if (s_listen) 726b411b363SPhilipp Reisner sock_release(s_listen); 727b411b363SPhilipp Reisner if (err < 0) { 728b411b363SPhilipp Reisner if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { 7291ec861ebSAndreas Gruenbacher drbd_err(connection, "%s failed, err = %d\n", what, err); 730bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); 731b411b363SPhilipp Reisner } 732b411b363SPhilipp Reisner } 7331f3e509bSPhilipp Reisner 7347a426fd8SPhilipp Reisner return -EIO; 7351f3e509bSPhilipp Reisner } 7361f3e509bSPhilipp Reisner 737715306f6SAndreas Gruenbacher static void unregister_state_change(struct sock *sk, struct accept_wait_data *ad) 738715306f6SAndreas Gruenbacher { 739715306f6SAndreas Gruenbacher write_lock_bh(&sk->sk_callback_lock); 740715306f6SAndreas Gruenbacher sk->sk_state_change = ad->original_sk_state_change; 741715306f6SAndreas Gruenbacher sk->sk_user_data = NULL; 742715306f6SAndreas Gruenbacher write_unlock_bh(&sk->sk_callback_lock); 743715306f6SAndreas Gruenbacher } 744715306f6SAndreas Gruenbacher 745bde89a9eSAndreas Gruenbacher static struct socket *drbd_wait_for_connect(struct drbd_connection *connection, struct accept_wait_data *ad) 7461f3e509bSPhilipp Reisner { 7471f3e509bSPhilipp Reisner int timeo, connect_int, err = 0; 7481f3e509bSPhilipp Reisner struct socket *s_estab = NULL; 7491f3e509bSPhilipp Reisner struct net_conf *nc; 7501f3e509bSPhilipp Reisner 7511f3e509bSPhilipp Reisner rcu_read_lock(); 752bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 7531f3e509bSPhilipp Reisner if (!nc) { 7541f3e509bSPhilipp Reisner rcu_read_unlock(); 7551f3e509bSPhilipp Reisner return NULL; 7561f3e509bSPhilipp Reisner } 7571f3e509bSPhilipp Reisner connect_int = nc->connect_int; 7581f3e509bSPhilipp Reisner rcu_read_unlock(); 7591f3e509bSPhilipp Reisner 7601f3e509bSPhilipp Reisner timeo = connect_int * HZ; 76138b682b2SAkinobu Mita /* 28.5% random jitter */ 76238b682b2SAkinobu Mita timeo += (prandom_u32() & 1) ? timeo / 7 : -timeo / 7; 7631f3e509bSPhilipp Reisner 7647a426fd8SPhilipp Reisner err = wait_for_completion_interruptible_timeout(&ad->door_bell, timeo); 7657a426fd8SPhilipp Reisner if (err <= 0) 7667a426fd8SPhilipp Reisner return NULL; 7671f3e509bSPhilipp Reisner 7687a426fd8SPhilipp Reisner err = kernel_accept(ad->s_listen, &s_estab, 0); 769b411b363SPhilipp Reisner if (err < 0) { 770b411b363SPhilipp Reisner if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { 7711ec861ebSAndreas Gruenbacher drbd_err(connection, "accept failed, err = %d\n", err); 772bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); 773b411b363SPhilipp Reisner } 774b411b363SPhilipp Reisner } 775b411b363SPhilipp Reisner 776715306f6SAndreas Gruenbacher if (s_estab) 777715306f6SAndreas Gruenbacher unregister_state_change(s_estab->sk, ad); 778b411b363SPhilipp Reisner 779b411b363SPhilipp Reisner return s_estab; 780b411b363SPhilipp Reisner } 781b411b363SPhilipp Reisner 782bde89a9eSAndreas Gruenbacher static int decode_header(struct drbd_connection *, void *, struct packet_info *); 783b411b363SPhilipp Reisner 784bde89a9eSAndreas Gruenbacher static int send_first_packet(struct drbd_connection *connection, struct drbd_socket *sock, 7859f5bdc33SAndreas Gruenbacher enum drbd_packet cmd) 7869f5bdc33SAndreas Gruenbacher { 787bde89a9eSAndreas Gruenbacher if (!conn_prepare_command(connection, sock)) 7889f5bdc33SAndreas Gruenbacher return -EIO; 789bde89a9eSAndreas Gruenbacher return conn_send_command(connection, sock, cmd, 0, NULL, 0); 790b411b363SPhilipp Reisner } 791b411b363SPhilipp Reisner 792bde89a9eSAndreas Gruenbacher static int receive_first_packet(struct drbd_connection *connection, struct socket *sock) 793b411b363SPhilipp Reisner { 794bde89a9eSAndreas Gruenbacher unsigned int header_size = drbd_header_size(connection); 7959f5bdc33SAndreas Gruenbacher struct packet_info pi; 7969f5bdc33SAndreas Gruenbacher int err; 797b411b363SPhilipp Reisner 798bde89a9eSAndreas Gruenbacher err = drbd_recv_short(sock, connection->data.rbuf, header_size, 0); 7999f5bdc33SAndreas Gruenbacher if (err != header_size) { 8009f5bdc33SAndreas Gruenbacher if (err >= 0) 8019f5bdc33SAndreas Gruenbacher err = -EIO; 8029f5bdc33SAndreas Gruenbacher return err; 8039f5bdc33SAndreas Gruenbacher } 804bde89a9eSAndreas Gruenbacher err = decode_header(connection, connection->data.rbuf, &pi); 8059f5bdc33SAndreas Gruenbacher if (err) 8069f5bdc33SAndreas Gruenbacher return err; 8079f5bdc33SAndreas Gruenbacher return pi.cmd; 808b411b363SPhilipp Reisner } 809b411b363SPhilipp Reisner 810b411b363SPhilipp Reisner /** 811b411b363SPhilipp Reisner * drbd_socket_okay() - Free the socket if its connection is not okay 812b411b363SPhilipp Reisner * @sock: pointer to the pointer to the socket. 813b411b363SPhilipp Reisner */ 814dbd9eea0SPhilipp Reisner static int drbd_socket_okay(struct socket **sock) 815b411b363SPhilipp Reisner { 816b411b363SPhilipp Reisner int rr; 817b411b363SPhilipp Reisner char tb[4]; 818b411b363SPhilipp Reisner 819b411b363SPhilipp Reisner if (!*sock) 82081e84650SAndreas Gruenbacher return false; 821b411b363SPhilipp Reisner 822dbd9eea0SPhilipp Reisner rr = drbd_recv_short(*sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); 823b411b363SPhilipp Reisner 824b411b363SPhilipp Reisner if (rr > 0 || rr == -EAGAIN) { 82581e84650SAndreas Gruenbacher return true; 826b411b363SPhilipp Reisner } else { 827b411b363SPhilipp Reisner sock_release(*sock); 828b411b363SPhilipp Reisner *sock = NULL; 82981e84650SAndreas Gruenbacher return false; 830b411b363SPhilipp Reisner } 831b411b363SPhilipp Reisner } 8322325eb66SPhilipp Reisner /* Gets called if a connection is established, or if a new minor gets created 8332325eb66SPhilipp Reisner in a connection */ 834b30ab791SAndreas Gruenbacher int drbd_connected(struct drbd_device *device) 835907599e0SPhilipp Reisner { 8360829f5edSAndreas Gruenbacher int err; 837907599e0SPhilipp Reisner 838b30ab791SAndreas Gruenbacher atomic_set(&device->packet_seq, 0); 839b30ab791SAndreas Gruenbacher device->peer_seq = 0; 840907599e0SPhilipp Reisner 841a6b32bc3SAndreas Gruenbacher device->state_mutex = first_peer_device(device)->connection->agreed_pro_version < 100 ? 842a6b32bc3SAndreas Gruenbacher &first_peer_device(device)->connection->cstate_mutex : 843b30ab791SAndreas Gruenbacher &device->own_state_mutex; 8448410da8fSPhilipp Reisner 845b30ab791SAndreas Gruenbacher err = drbd_send_sync_param(device); 8460829f5edSAndreas Gruenbacher if (!err) 847b30ab791SAndreas Gruenbacher err = drbd_send_sizes(device, 0, 0); 8480829f5edSAndreas Gruenbacher if (!err) 849b30ab791SAndreas Gruenbacher err = drbd_send_uuids(device); 8500829f5edSAndreas Gruenbacher if (!err) 851b30ab791SAndreas Gruenbacher err = drbd_send_current_state(device); 852b30ab791SAndreas Gruenbacher clear_bit(USE_DEGR_WFC_T, &device->flags); 853b30ab791SAndreas Gruenbacher clear_bit(RESIZE_PENDING, &device->flags); 854b30ab791SAndreas Gruenbacher atomic_set(&device->ap_in_flight, 0); 855b30ab791SAndreas Gruenbacher mod_timer(&device->request_timer, jiffies + HZ); /* just start it here. */ 8560829f5edSAndreas Gruenbacher return err; 857907599e0SPhilipp Reisner } 858b411b363SPhilipp Reisner 859b411b363SPhilipp Reisner /* 860b411b363SPhilipp Reisner * return values: 861b411b363SPhilipp Reisner * 1 yes, we have a valid connection 862b411b363SPhilipp Reisner * 0 oops, did not work out, please try again 863b411b363SPhilipp Reisner * -1 peer talks different language, 864b411b363SPhilipp Reisner * no point in trying again, please go standalone. 865b411b363SPhilipp Reisner * -2 We do not have a network config... 866b411b363SPhilipp Reisner */ 867bde89a9eSAndreas Gruenbacher static int conn_connect(struct drbd_connection *connection) 868b411b363SPhilipp Reisner { 8697da35862SPhilipp Reisner struct drbd_socket sock, msock; 870c06ece6bSAndreas Gruenbacher struct drbd_peer_device *peer_device; 87144ed167dSPhilipp Reisner struct net_conf *nc; 87292f14951SPhilipp Reisner int vnr, timeout, h, ok; 87308b165baSPhilipp Reisner bool discard_my_data; 874197296ffSPhilipp Reisner enum drbd_state_rv rv; 8757a426fd8SPhilipp Reisner struct accept_wait_data ad = { 876bde89a9eSAndreas Gruenbacher .connection = connection, 8777a426fd8SPhilipp Reisner .door_bell = COMPLETION_INITIALIZER_ONSTACK(ad.door_bell), 8787a426fd8SPhilipp Reisner }; 879b411b363SPhilipp Reisner 880bde89a9eSAndreas Gruenbacher clear_bit(DISCONNECT_SENT, &connection->flags); 881bde89a9eSAndreas Gruenbacher if (conn_request_state(connection, NS(conn, C_WF_CONNECTION), CS_VERBOSE) < SS_SUCCESS) 882b411b363SPhilipp Reisner return -2; 883b411b363SPhilipp Reisner 8847da35862SPhilipp Reisner mutex_init(&sock.mutex); 885bde89a9eSAndreas Gruenbacher sock.sbuf = connection->data.sbuf; 886bde89a9eSAndreas Gruenbacher sock.rbuf = connection->data.rbuf; 8877da35862SPhilipp Reisner sock.socket = NULL; 8887da35862SPhilipp Reisner mutex_init(&msock.mutex); 889bde89a9eSAndreas Gruenbacher msock.sbuf = connection->meta.sbuf; 890bde89a9eSAndreas Gruenbacher msock.rbuf = connection->meta.rbuf; 8917da35862SPhilipp Reisner msock.socket = NULL; 8927da35862SPhilipp Reisner 8930916e0e3SAndreas Gruenbacher /* Assume that the peer only understands protocol 80 until we know better. */ 894bde89a9eSAndreas Gruenbacher connection->agreed_pro_version = 80; 895b411b363SPhilipp Reisner 896bde89a9eSAndreas Gruenbacher if (prepare_listen_socket(connection, &ad)) 8977a426fd8SPhilipp Reisner return 0; 898b411b363SPhilipp Reisner 899b411b363SPhilipp Reisner do { 9002bf89621SAndreas Gruenbacher struct socket *s; 901b411b363SPhilipp Reisner 902bde89a9eSAndreas Gruenbacher s = drbd_try_connect(connection); 903b411b363SPhilipp Reisner if (s) { 9047da35862SPhilipp Reisner if (!sock.socket) { 9057da35862SPhilipp Reisner sock.socket = s; 906bde89a9eSAndreas Gruenbacher send_first_packet(connection, &sock, P_INITIAL_DATA); 9077da35862SPhilipp Reisner } else if (!msock.socket) { 908bde89a9eSAndreas Gruenbacher clear_bit(RESOLVE_CONFLICTS, &connection->flags); 9097da35862SPhilipp Reisner msock.socket = s; 910bde89a9eSAndreas Gruenbacher send_first_packet(connection, &msock, P_INITIAL_META); 911b411b363SPhilipp Reisner } else { 9121ec861ebSAndreas Gruenbacher drbd_err(connection, "Logic error in conn_connect()\n"); 913b411b363SPhilipp Reisner goto out_release_sockets; 914b411b363SPhilipp Reisner } 915b411b363SPhilipp Reisner } 916b411b363SPhilipp Reisner 9177da35862SPhilipp Reisner if (sock.socket && msock.socket) { 9187da35862SPhilipp Reisner rcu_read_lock(); 919bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 9207da35862SPhilipp Reisner timeout = nc->ping_timeo * HZ / 10; 9217da35862SPhilipp Reisner rcu_read_unlock(); 9227da35862SPhilipp Reisner schedule_timeout_interruptible(timeout); 9237da35862SPhilipp Reisner ok = drbd_socket_okay(&sock.socket); 9247da35862SPhilipp Reisner ok = drbd_socket_okay(&msock.socket) && ok; 925b411b363SPhilipp Reisner if (ok) 926b411b363SPhilipp Reisner break; 927b411b363SPhilipp Reisner } 928b411b363SPhilipp Reisner 929b411b363SPhilipp Reisner retry: 930bde89a9eSAndreas Gruenbacher s = drbd_wait_for_connect(connection, &ad); 931b411b363SPhilipp Reisner if (s) { 932bde89a9eSAndreas Gruenbacher int fp = receive_first_packet(connection, s); 9337da35862SPhilipp Reisner drbd_socket_okay(&sock.socket); 9347da35862SPhilipp Reisner drbd_socket_okay(&msock.socket); 93592f14951SPhilipp Reisner switch (fp) { 936e5d6f33aSAndreas Gruenbacher case P_INITIAL_DATA: 9377da35862SPhilipp Reisner if (sock.socket) { 9381ec861ebSAndreas Gruenbacher drbd_warn(connection, "initial packet S crossed\n"); 9397da35862SPhilipp Reisner sock_release(sock.socket); 94080c6eed4SPhilipp Reisner sock.socket = s; 94180c6eed4SPhilipp Reisner goto randomize; 942b411b363SPhilipp Reisner } 9437da35862SPhilipp Reisner sock.socket = s; 944b411b363SPhilipp Reisner break; 945e5d6f33aSAndreas Gruenbacher case P_INITIAL_META: 946bde89a9eSAndreas Gruenbacher set_bit(RESOLVE_CONFLICTS, &connection->flags); 9477da35862SPhilipp Reisner if (msock.socket) { 9481ec861ebSAndreas Gruenbacher drbd_warn(connection, "initial packet M crossed\n"); 9497da35862SPhilipp Reisner sock_release(msock.socket); 95080c6eed4SPhilipp Reisner msock.socket = s; 95180c6eed4SPhilipp Reisner goto randomize; 952b411b363SPhilipp Reisner } 9537da35862SPhilipp Reisner msock.socket = s; 954b411b363SPhilipp Reisner break; 955b411b363SPhilipp Reisner default: 9561ec861ebSAndreas Gruenbacher drbd_warn(connection, "Error receiving initial packet\n"); 957b411b363SPhilipp Reisner sock_release(s); 95880c6eed4SPhilipp Reisner randomize: 95938b682b2SAkinobu Mita if (prandom_u32() & 1) 960b411b363SPhilipp Reisner goto retry; 961b411b363SPhilipp Reisner } 962b411b363SPhilipp Reisner } 963b411b363SPhilipp Reisner 964bde89a9eSAndreas Gruenbacher if (connection->cstate <= C_DISCONNECTING) 965b411b363SPhilipp Reisner goto out_release_sockets; 966b411b363SPhilipp Reisner if (signal_pending(current)) { 967b411b363SPhilipp Reisner flush_signals(current); 968b411b363SPhilipp Reisner smp_rmb(); 969bde89a9eSAndreas Gruenbacher if (get_t_state(&connection->receiver) == EXITING) 970b411b363SPhilipp Reisner goto out_release_sockets; 971b411b363SPhilipp Reisner } 972b411b363SPhilipp Reisner 9737da35862SPhilipp Reisner ok = drbd_socket_okay(&sock.socket); 9747da35862SPhilipp Reisner ok = drbd_socket_okay(&msock.socket) && ok; 975b666dbf8SPhilipp Reisner } while (!ok); 976b411b363SPhilipp Reisner 9777a426fd8SPhilipp Reisner if (ad.s_listen) 9787a426fd8SPhilipp Reisner sock_release(ad.s_listen); 979b411b363SPhilipp Reisner 98098683650SPhilipp Reisner sock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ 98198683650SPhilipp Reisner msock.socket->sk->sk_reuse = SK_CAN_REUSE; /* SO_REUSEADDR */ 982b411b363SPhilipp Reisner 9837da35862SPhilipp Reisner sock.socket->sk->sk_allocation = GFP_NOIO; 9847da35862SPhilipp Reisner msock.socket->sk->sk_allocation = GFP_NOIO; 985b411b363SPhilipp Reisner 9867da35862SPhilipp Reisner sock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; 9877da35862SPhilipp Reisner msock.socket->sk->sk_priority = TC_PRIO_INTERACTIVE; 988b411b363SPhilipp Reisner 989b411b363SPhilipp Reisner /* NOT YET ... 990bde89a9eSAndreas Gruenbacher * sock.socket->sk->sk_sndtimeo = connection->net_conf->timeout*HZ/10; 9917da35862SPhilipp Reisner * sock.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 9926038178eSAndreas Gruenbacher * first set it to the P_CONNECTION_FEATURES timeout, 993b411b363SPhilipp Reisner * which we set to 4x the configured ping_timeout. */ 99444ed167dSPhilipp Reisner rcu_read_lock(); 995bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 996b411b363SPhilipp Reisner 9977da35862SPhilipp Reisner sock.socket->sk->sk_sndtimeo = 9987da35862SPhilipp Reisner sock.socket->sk->sk_rcvtimeo = nc->ping_timeo*4*HZ/10; 99944ed167dSPhilipp Reisner 10007da35862SPhilipp Reisner msock.socket->sk->sk_rcvtimeo = nc->ping_int*HZ; 100144ed167dSPhilipp Reisner timeout = nc->timeout * HZ / 10; 100208b165baSPhilipp Reisner discard_my_data = nc->discard_my_data; 100344ed167dSPhilipp Reisner rcu_read_unlock(); 100444ed167dSPhilipp Reisner 10057da35862SPhilipp Reisner msock.socket->sk->sk_sndtimeo = timeout; 1006b411b363SPhilipp Reisner 1007b411b363SPhilipp Reisner /* we don't want delays. 100825985edcSLucas De Marchi * we use TCP_CORK where appropriate, though */ 10097da35862SPhilipp Reisner drbd_tcp_nodelay(sock.socket); 10107da35862SPhilipp Reisner drbd_tcp_nodelay(msock.socket); 1011b411b363SPhilipp Reisner 1012bde89a9eSAndreas Gruenbacher connection->data.socket = sock.socket; 1013bde89a9eSAndreas Gruenbacher connection->meta.socket = msock.socket; 1014bde89a9eSAndreas Gruenbacher connection->last_received = jiffies; 1015b411b363SPhilipp Reisner 1016bde89a9eSAndreas Gruenbacher h = drbd_do_features(connection); 1017b411b363SPhilipp Reisner if (h <= 0) 1018b411b363SPhilipp Reisner return h; 1019b411b363SPhilipp Reisner 1020bde89a9eSAndreas Gruenbacher if (connection->cram_hmac_tfm) { 1021b30ab791SAndreas Gruenbacher /* drbd_request_state(device, NS(conn, WFAuth)); */ 1022bde89a9eSAndreas Gruenbacher switch (drbd_do_auth(connection)) { 1023b10d96cbSJohannes Thoma case -1: 10241ec861ebSAndreas Gruenbacher drbd_err(connection, "Authentication of peer failed\n"); 1025b411b363SPhilipp Reisner return -1; 1026b10d96cbSJohannes Thoma case 0: 10271ec861ebSAndreas Gruenbacher drbd_err(connection, "Authentication of peer failed, trying again.\n"); 1028b10d96cbSJohannes Thoma return 0; 1029b411b363SPhilipp Reisner } 1030b411b363SPhilipp Reisner } 1031b411b363SPhilipp Reisner 1032bde89a9eSAndreas Gruenbacher connection->data.socket->sk->sk_sndtimeo = timeout; 1033bde89a9eSAndreas Gruenbacher connection->data.socket->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 1034b411b363SPhilipp Reisner 1035bde89a9eSAndreas Gruenbacher if (drbd_send_protocol(connection) == -EOPNOTSUPP) 10367e2455c1SPhilipp Reisner return -1; 10371e86ac48SPhilipp Reisner 1038bde89a9eSAndreas Gruenbacher set_bit(STATE_SENT, &connection->flags); 1039197296ffSPhilipp Reisner 1040c141ebdaSPhilipp Reisner rcu_read_lock(); 1041c06ece6bSAndreas Gruenbacher idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 1042c06ece6bSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 1043b30ab791SAndreas Gruenbacher kref_get(&device->kref); 104426ea8f92SAndreas Gruenbacher rcu_read_unlock(); 104526ea8f92SAndreas Gruenbacher 104613c76abaSPhilipp Reisner /* Prevent a race between resync-handshake and 104713c76abaSPhilipp Reisner * being promoted to Primary. 104813c76abaSPhilipp Reisner * 104913c76abaSPhilipp Reisner * Grab and release the state mutex, so we know that any current 105013c76abaSPhilipp Reisner * drbd_set_role() is finished, and any incoming drbd_set_role 105113c76abaSPhilipp Reisner * will see the STATE_SENT flag, and wait for it to be cleared. 105213c76abaSPhilipp Reisner */ 1053b30ab791SAndreas Gruenbacher mutex_lock(device->state_mutex); 1054b30ab791SAndreas Gruenbacher mutex_unlock(device->state_mutex); 105513c76abaSPhilipp Reisner 105608b165baSPhilipp Reisner if (discard_my_data) 1057b30ab791SAndreas Gruenbacher set_bit(DISCARD_MY_DATA, &device->flags); 105808b165baSPhilipp Reisner else 1059b30ab791SAndreas Gruenbacher clear_bit(DISCARD_MY_DATA, &device->flags); 106008b165baSPhilipp Reisner 1061b30ab791SAndreas Gruenbacher drbd_connected(device); 106205a10ec7SAndreas Gruenbacher kref_put(&device->kref, drbd_destroy_device); 1063c141ebdaSPhilipp Reisner rcu_read_lock(); 1064c141ebdaSPhilipp Reisner } 1065c141ebdaSPhilipp Reisner rcu_read_unlock(); 1066c141ebdaSPhilipp Reisner 1067bde89a9eSAndreas Gruenbacher rv = conn_request_state(connection, NS(conn, C_WF_REPORT_PARAMS), CS_VERBOSE); 1068bde89a9eSAndreas Gruenbacher if (rv < SS_SUCCESS || connection->cstate != C_WF_REPORT_PARAMS) { 1069bde89a9eSAndreas Gruenbacher clear_bit(STATE_SENT, &connection->flags); 10701e86ac48SPhilipp Reisner return 0; 1071a1096a6eSPhilipp Reisner } 10721e86ac48SPhilipp Reisner 1073bde89a9eSAndreas Gruenbacher drbd_thread_start(&connection->asender); 1074b411b363SPhilipp Reisner 10750500813fSAndreas Gruenbacher mutex_lock(&connection->resource->conf_update); 107608b165baSPhilipp Reisner /* The discard_my_data flag is a single-shot modifier to the next 107708b165baSPhilipp Reisner * connection attempt, the handshake of which is now well underway. 107808b165baSPhilipp Reisner * No need for rcu style copying of the whole struct 107908b165baSPhilipp Reisner * just to clear a single value. */ 1080bde89a9eSAndreas Gruenbacher connection->net_conf->discard_my_data = 0; 10810500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 108208b165baSPhilipp Reisner 1083d3fcb490SPhilipp Reisner return h; 1084b411b363SPhilipp Reisner 1085b411b363SPhilipp Reisner out_release_sockets: 10867a426fd8SPhilipp Reisner if (ad.s_listen) 10877a426fd8SPhilipp Reisner sock_release(ad.s_listen); 10887da35862SPhilipp Reisner if (sock.socket) 10897da35862SPhilipp Reisner sock_release(sock.socket); 10907da35862SPhilipp Reisner if (msock.socket) 10917da35862SPhilipp Reisner sock_release(msock.socket); 1092b411b363SPhilipp Reisner return -1; 1093b411b363SPhilipp Reisner } 1094b411b363SPhilipp Reisner 1095bde89a9eSAndreas Gruenbacher static int decode_header(struct drbd_connection *connection, void *header, struct packet_info *pi) 1096b411b363SPhilipp Reisner { 1097bde89a9eSAndreas Gruenbacher unsigned int header_size = drbd_header_size(connection); 1098b411b363SPhilipp Reisner 10990c8e36d9SAndreas Gruenbacher if (header_size == sizeof(struct p_header100) && 11000c8e36d9SAndreas Gruenbacher *(__be32 *)header == cpu_to_be32(DRBD_MAGIC_100)) { 11010c8e36d9SAndreas Gruenbacher struct p_header100 *h = header; 11020c8e36d9SAndreas Gruenbacher if (h->pad != 0) { 11031ec861ebSAndreas Gruenbacher drbd_err(connection, "Header padding is not zero\n"); 11040c8e36d9SAndreas Gruenbacher return -EINVAL; 110502918be2SPhilipp Reisner } 11060c8e36d9SAndreas Gruenbacher pi->vnr = be16_to_cpu(h->volume); 11070c8e36d9SAndreas Gruenbacher pi->cmd = be16_to_cpu(h->command); 11080c8e36d9SAndreas Gruenbacher pi->size = be32_to_cpu(h->length); 11090c8e36d9SAndreas Gruenbacher } else if (header_size == sizeof(struct p_header95) && 1110e658983aSAndreas Gruenbacher *(__be16 *)header == cpu_to_be16(DRBD_MAGIC_BIG)) { 1111e658983aSAndreas Gruenbacher struct p_header95 *h = header; 1112e658983aSAndreas Gruenbacher pi->cmd = be16_to_cpu(h->command); 1113b55d84baSAndreas Gruenbacher pi->size = be32_to_cpu(h->length); 1114eefc2f7dSPhilipp Reisner pi->vnr = 0; 1115e658983aSAndreas Gruenbacher } else if (header_size == sizeof(struct p_header80) && 1116e658983aSAndreas Gruenbacher *(__be32 *)header == cpu_to_be32(DRBD_MAGIC)) { 1117e658983aSAndreas Gruenbacher struct p_header80 *h = header; 1118e658983aSAndreas Gruenbacher pi->cmd = be16_to_cpu(h->command); 1119e658983aSAndreas Gruenbacher pi->size = be16_to_cpu(h->length); 112077351055SPhilipp Reisner pi->vnr = 0; 112102918be2SPhilipp Reisner } else { 11221ec861ebSAndreas Gruenbacher drbd_err(connection, "Wrong magic value 0x%08x in protocol version %d\n", 1123e658983aSAndreas Gruenbacher be32_to_cpu(*(__be32 *)header), 1124bde89a9eSAndreas Gruenbacher connection->agreed_pro_version); 11258172f3e9SAndreas Gruenbacher return -EINVAL; 1126b411b363SPhilipp Reisner } 1127e658983aSAndreas Gruenbacher pi->data = header + header_size; 11288172f3e9SAndreas Gruenbacher return 0; 1129b411b363SPhilipp Reisner } 1130b411b363SPhilipp Reisner 1131bde89a9eSAndreas Gruenbacher static int drbd_recv_header(struct drbd_connection *connection, struct packet_info *pi) 1132257d0af6SPhilipp Reisner { 1133bde89a9eSAndreas Gruenbacher void *buffer = connection->data.rbuf; 113469bc7bc3SAndreas Gruenbacher int err; 1135257d0af6SPhilipp Reisner 1136bde89a9eSAndreas Gruenbacher err = drbd_recv_all_warn(connection, buffer, drbd_header_size(connection)); 1137a5c31904SAndreas Gruenbacher if (err) 113869bc7bc3SAndreas Gruenbacher return err; 1139257d0af6SPhilipp Reisner 1140bde89a9eSAndreas Gruenbacher err = decode_header(connection, buffer, pi); 1141bde89a9eSAndreas Gruenbacher connection->last_received = jiffies; 1142b411b363SPhilipp Reisner 114369bc7bc3SAndreas Gruenbacher return err; 1144b411b363SPhilipp Reisner } 1145b411b363SPhilipp Reisner 1146bde89a9eSAndreas Gruenbacher static void drbd_flush(struct drbd_connection *connection) 1147b411b363SPhilipp Reisner { 1148b411b363SPhilipp Reisner int rv; 1149c06ece6bSAndreas Gruenbacher struct drbd_peer_device *peer_device; 11504b0007c0SPhilipp Reisner int vnr; 1151b411b363SPhilipp Reisner 1152bde89a9eSAndreas Gruenbacher if (connection->write_ordering >= WO_bdev_flush) { 1153615e087fSLars Ellenberg rcu_read_lock(); 1154c06ece6bSAndreas Gruenbacher idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 1155c06ece6bSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 1156c06ece6bSAndreas Gruenbacher 1157b30ab791SAndreas Gruenbacher if (!get_ldev(device)) 1158615e087fSLars Ellenberg continue; 1159b30ab791SAndreas Gruenbacher kref_get(&device->kref); 1160615e087fSLars Ellenberg rcu_read_unlock(); 11614b0007c0SPhilipp Reisner 1162b30ab791SAndreas Gruenbacher rv = blkdev_issue_flush(device->ldev->backing_bdev, 1163615e087fSLars Ellenberg GFP_NOIO, NULL); 1164b411b363SPhilipp Reisner if (rv) { 1165d0180171SAndreas Gruenbacher drbd_info(device, "local disk flush failed with status %d\n", rv); 1166b411b363SPhilipp Reisner /* would rather check on EOPNOTSUPP, but that is not reliable. 1167b411b363SPhilipp Reisner * don't try again for ANY return value != 0 1168b411b363SPhilipp Reisner * if (rv == -EOPNOTSUPP) */ 1169bde89a9eSAndreas Gruenbacher drbd_bump_write_ordering(connection, WO_drain_io); 1170b411b363SPhilipp Reisner } 1171b30ab791SAndreas Gruenbacher put_ldev(device); 117205a10ec7SAndreas Gruenbacher kref_put(&device->kref, drbd_destroy_device); 1173615e087fSLars Ellenberg 1174615e087fSLars Ellenberg rcu_read_lock(); 1175615e087fSLars Ellenberg if (rv) 11764b0007c0SPhilipp Reisner break; 1177b411b363SPhilipp Reisner } 1178615e087fSLars Ellenberg rcu_read_unlock(); 1179b411b363SPhilipp Reisner } 1180b411b363SPhilipp Reisner } 1181b411b363SPhilipp Reisner 1182b411b363SPhilipp Reisner /** 1183b411b363SPhilipp Reisner * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. 1184b30ab791SAndreas Gruenbacher * @device: DRBD device. 1185b411b363SPhilipp Reisner * @epoch: Epoch object. 1186b411b363SPhilipp Reisner * @ev: Epoch event. 1187b411b363SPhilipp Reisner */ 1188bde89a9eSAndreas Gruenbacher static enum finish_epoch drbd_may_finish_epoch(struct drbd_connection *connection, 1189b411b363SPhilipp Reisner struct drbd_epoch *epoch, 1190b411b363SPhilipp Reisner enum epoch_event ev) 1191b411b363SPhilipp Reisner { 11922451fc3bSPhilipp Reisner int epoch_size; 1193b411b363SPhilipp Reisner struct drbd_epoch *next_epoch; 1194b411b363SPhilipp Reisner enum finish_epoch rv = FE_STILL_LIVE; 1195b411b363SPhilipp Reisner 1196bde89a9eSAndreas Gruenbacher spin_lock(&connection->epoch_lock); 1197b411b363SPhilipp Reisner do { 1198b411b363SPhilipp Reisner next_epoch = NULL; 1199b411b363SPhilipp Reisner 1200b411b363SPhilipp Reisner epoch_size = atomic_read(&epoch->epoch_size); 1201b411b363SPhilipp Reisner 1202b411b363SPhilipp Reisner switch (ev & ~EV_CLEANUP) { 1203b411b363SPhilipp Reisner case EV_PUT: 1204b411b363SPhilipp Reisner atomic_dec(&epoch->active); 1205b411b363SPhilipp Reisner break; 1206b411b363SPhilipp Reisner case EV_GOT_BARRIER_NR: 1207b411b363SPhilipp Reisner set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); 1208b411b363SPhilipp Reisner break; 1209b411b363SPhilipp Reisner case EV_BECAME_LAST: 1210b411b363SPhilipp Reisner /* nothing to do*/ 1211b411b363SPhilipp Reisner break; 1212b411b363SPhilipp Reisner } 1213b411b363SPhilipp Reisner 1214b411b363SPhilipp Reisner if (epoch_size != 0 && 1215b411b363SPhilipp Reisner atomic_read(&epoch->active) == 0 && 121680f9fd55SPhilipp Reisner (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) || ev & EV_CLEANUP)) { 1217b411b363SPhilipp Reisner if (!(ev & EV_CLEANUP)) { 1218bde89a9eSAndreas Gruenbacher spin_unlock(&connection->epoch_lock); 1219bde89a9eSAndreas Gruenbacher drbd_send_b_ack(epoch->connection, epoch->barrier_nr, epoch_size); 1220bde89a9eSAndreas Gruenbacher spin_lock(&connection->epoch_lock); 1221b411b363SPhilipp Reisner } 12229ed57dcbSLars Ellenberg #if 0 12239ed57dcbSLars Ellenberg /* FIXME: dec unacked on connection, once we have 12249ed57dcbSLars Ellenberg * something to count pending connection packets in. */ 122580f9fd55SPhilipp Reisner if (test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) 1226bde89a9eSAndreas Gruenbacher dec_unacked(epoch->connection); 12279ed57dcbSLars Ellenberg #endif 1228b411b363SPhilipp Reisner 1229bde89a9eSAndreas Gruenbacher if (connection->current_epoch != epoch) { 1230b411b363SPhilipp Reisner next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); 1231b411b363SPhilipp Reisner list_del(&epoch->list); 1232b411b363SPhilipp Reisner ev = EV_BECAME_LAST | (ev & EV_CLEANUP); 1233bde89a9eSAndreas Gruenbacher connection->epochs--; 1234b411b363SPhilipp Reisner kfree(epoch); 1235b411b363SPhilipp Reisner 1236b411b363SPhilipp Reisner if (rv == FE_STILL_LIVE) 1237b411b363SPhilipp Reisner rv = FE_DESTROYED; 1238b411b363SPhilipp Reisner } else { 1239b411b363SPhilipp Reisner epoch->flags = 0; 1240b411b363SPhilipp Reisner atomic_set(&epoch->epoch_size, 0); 1241698f9315SUwe Kleine-König /* atomic_set(&epoch->active, 0); is already zero */ 1242b411b363SPhilipp Reisner if (rv == FE_STILL_LIVE) 1243b411b363SPhilipp Reisner rv = FE_RECYCLED; 1244b411b363SPhilipp Reisner } 1245b411b363SPhilipp Reisner } 1246b411b363SPhilipp Reisner 1247b411b363SPhilipp Reisner if (!next_epoch) 1248b411b363SPhilipp Reisner break; 1249b411b363SPhilipp Reisner 1250b411b363SPhilipp Reisner epoch = next_epoch; 1251b411b363SPhilipp Reisner } while (1); 1252b411b363SPhilipp Reisner 1253bde89a9eSAndreas Gruenbacher spin_unlock(&connection->epoch_lock); 1254b411b363SPhilipp Reisner 1255b411b363SPhilipp Reisner return rv; 1256b411b363SPhilipp Reisner } 1257b411b363SPhilipp Reisner 1258b411b363SPhilipp Reisner /** 1259b411b363SPhilipp Reisner * drbd_bump_write_ordering() - Fall back to an other write ordering method 1260bde89a9eSAndreas Gruenbacher * @connection: DRBD connection. 1261b411b363SPhilipp Reisner * @wo: Write ordering method to try. 1262b411b363SPhilipp Reisner */ 1263bde89a9eSAndreas Gruenbacher void drbd_bump_write_ordering(struct drbd_connection *connection, enum write_ordering_e wo) 1264b411b363SPhilipp Reisner { 1265daeda1ccSPhilipp Reisner struct disk_conf *dc; 1266c06ece6bSAndreas Gruenbacher struct drbd_peer_device *peer_device; 1267b411b363SPhilipp Reisner enum write_ordering_e pwo; 12684b0007c0SPhilipp Reisner int vnr; 1269b411b363SPhilipp Reisner static char *write_ordering_str[] = { 1270b411b363SPhilipp Reisner [WO_none] = "none", 1271b411b363SPhilipp Reisner [WO_drain_io] = "drain", 1272b411b363SPhilipp Reisner [WO_bdev_flush] = "flush", 1273b411b363SPhilipp Reisner }; 1274b411b363SPhilipp Reisner 1275bde89a9eSAndreas Gruenbacher pwo = connection->write_ordering; 1276b411b363SPhilipp Reisner wo = min(pwo, wo); 1277daeda1ccSPhilipp Reisner rcu_read_lock(); 1278c06ece6bSAndreas Gruenbacher idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 1279c06ece6bSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 1280c06ece6bSAndreas Gruenbacher 1281b30ab791SAndreas Gruenbacher if (!get_ldev_if_state(device, D_ATTACHING)) 12824b0007c0SPhilipp Reisner continue; 1283b30ab791SAndreas Gruenbacher dc = rcu_dereference(device->ldev->disk_conf); 1284daeda1ccSPhilipp Reisner 128566b2f6b9SAndreas Gruenbacher if (wo == WO_bdev_flush && !dc->disk_flushes) 1286b411b363SPhilipp Reisner wo = WO_drain_io; 1287d0c980e2SAndreas Gruenbacher if (wo == WO_drain_io && !dc->disk_drain) 1288b411b363SPhilipp Reisner wo = WO_none; 1289b30ab791SAndreas Gruenbacher put_ldev(device); 12904b0007c0SPhilipp Reisner } 1291daeda1ccSPhilipp Reisner rcu_read_unlock(); 1292bde89a9eSAndreas Gruenbacher connection->write_ordering = wo; 1293bde89a9eSAndreas Gruenbacher if (pwo != connection->write_ordering || wo == WO_bdev_flush) 12941ec861ebSAndreas Gruenbacher drbd_info(connection, "Method to ensure write ordering: %s\n", write_ordering_str[connection->write_ordering]); 1295b411b363SPhilipp Reisner } 1296b411b363SPhilipp Reisner 1297b411b363SPhilipp Reisner /** 1298fbe29decSAndreas Gruenbacher * drbd_submit_peer_request() 1299b30ab791SAndreas Gruenbacher * @device: DRBD device. 1300db830c46SAndreas Gruenbacher * @peer_req: peer request 130145bb912bSLars Ellenberg * @rw: flag field, see bio->bi_rw 130210f6d992SLars Ellenberg * 130310f6d992SLars Ellenberg * May spread the pages to multiple bios, 130410f6d992SLars Ellenberg * depending on bio_add_page restrictions. 130510f6d992SLars Ellenberg * 130610f6d992SLars Ellenberg * Returns 0 if all bios have been submitted, 130710f6d992SLars Ellenberg * -ENOMEM if we could not allocate enough bios, 130810f6d992SLars Ellenberg * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a 130910f6d992SLars Ellenberg * single page to an empty bio (which should never happen and likely indicates 131010f6d992SLars Ellenberg * that the lower level IO stack is in some way broken). This has been observed 131110f6d992SLars Ellenberg * on certain Xen deployments. 131245bb912bSLars Ellenberg */ 131345bb912bSLars Ellenberg /* TODO allocate from our own bio_set. */ 1314b30ab791SAndreas Gruenbacher int drbd_submit_peer_request(struct drbd_device *device, 1315fbe29decSAndreas Gruenbacher struct drbd_peer_request *peer_req, 131645bb912bSLars Ellenberg const unsigned rw, const int fault_type) 131745bb912bSLars Ellenberg { 131845bb912bSLars Ellenberg struct bio *bios = NULL; 131945bb912bSLars Ellenberg struct bio *bio; 1320db830c46SAndreas Gruenbacher struct page *page = peer_req->pages; 1321db830c46SAndreas Gruenbacher sector_t sector = peer_req->i.sector; 1322db830c46SAndreas Gruenbacher unsigned ds = peer_req->i.size; 132345bb912bSLars Ellenberg unsigned n_bios = 0; 132445bb912bSLars Ellenberg unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT; 132510f6d992SLars Ellenberg int err = -ENOMEM; 132645bb912bSLars Ellenberg 132745bb912bSLars Ellenberg /* In most cases, we will only need one bio. But in case the lower 132845bb912bSLars Ellenberg * level restrictions happen to be different at this offset on this 132945bb912bSLars Ellenberg * side than those of the sending peer, we may need to submit the 13309476f39dSLars Ellenberg * request in more than one bio. 13319476f39dSLars Ellenberg * 13329476f39dSLars Ellenberg * Plain bio_alloc is good enough here, this is no DRBD internally 13339476f39dSLars Ellenberg * generated bio, but a bio allocated on behalf of the peer. 13349476f39dSLars Ellenberg */ 133545bb912bSLars Ellenberg next_bio: 133645bb912bSLars Ellenberg bio = bio_alloc(GFP_NOIO, nr_pages); 133745bb912bSLars Ellenberg if (!bio) { 1338d0180171SAndreas Gruenbacher drbd_err(device, "submit_ee: Allocation of a bio failed\n"); 133945bb912bSLars Ellenberg goto fail; 134045bb912bSLars Ellenberg } 1341db830c46SAndreas Gruenbacher /* > peer_req->i.sector, unless this is the first bio */ 13424f024f37SKent Overstreet bio->bi_iter.bi_sector = sector; 1343b30ab791SAndreas Gruenbacher bio->bi_bdev = device->ldev->backing_bdev; 134445bb912bSLars Ellenberg bio->bi_rw = rw; 1345db830c46SAndreas Gruenbacher bio->bi_private = peer_req; 1346fcefa62eSAndreas Gruenbacher bio->bi_end_io = drbd_peer_request_endio; 134745bb912bSLars Ellenberg 134845bb912bSLars Ellenberg bio->bi_next = bios; 134945bb912bSLars Ellenberg bios = bio; 135045bb912bSLars Ellenberg ++n_bios; 135145bb912bSLars Ellenberg 135245bb912bSLars Ellenberg page_chain_for_each(page) { 135345bb912bSLars Ellenberg unsigned len = min_t(unsigned, ds, PAGE_SIZE); 135445bb912bSLars Ellenberg if (!bio_add_page(bio, page, len, 0)) { 135510f6d992SLars Ellenberg /* A single page must always be possible! 135610f6d992SLars Ellenberg * But in case it fails anyways, 135710f6d992SLars Ellenberg * we deal with it, and complain (below). */ 135810f6d992SLars Ellenberg if (bio->bi_vcnt == 0) { 1359d0180171SAndreas Gruenbacher drbd_err(device, 136010f6d992SLars Ellenberg "bio_add_page failed for len=%u, " 136110f6d992SLars Ellenberg "bi_vcnt=0 (bi_sector=%llu)\n", 13624f024f37SKent Overstreet len, (uint64_t)bio->bi_iter.bi_sector); 136310f6d992SLars Ellenberg err = -ENOSPC; 136410f6d992SLars Ellenberg goto fail; 136510f6d992SLars Ellenberg } 136645bb912bSLars Ellenberg goto next_bio; 136745bb912bSLars Ellenberg } 136845bb912bSLars Ellenberg ds -= len; 136945bb912bSLars Ellenberg sector += len >> 9; 137045bb912bSLars Ellenberg --nr_pages; 137145bb912bSLars Ellenberg } 13720b0ba1efSAndreas Gruenbacher D_ASSERT(device, page == NULL); 13730b0ba1efSAndreas Gruenbacher D_ASSERT(device, ds == 0); 137445bb912bSLars Ellenberg 1375db830c46SAndreas Gruenbacher atomic_set(&peer_req->pending_bios, n_bios); 137645bb912bSLars Ellenberg do { 137745bb912bSLars Ellenberg bio = bios; 137845bb912bSLars Ellenberg bios = bios->bi_next; 137945bb912bSLars Ellenberg bio->bi_next = NULL; 138045bb912bSLars Ellenberg 1381b30ab791SAndreas Gruenbacher drbd_generic_make_request(device, fault_type, bio); 138245bb912bSLars Ellenberg } while (bios); 138345bb912bSLars Ellenberg return 0; 138445bb912bSLars Ellenberg 138545bb912bSLars Ellenberg fail: 138645bb912bSLars Ellenberg while (bios) { 138745bb912bSLars Ellenberg bio = bios; 138845bb912bSLars Ellenberg bios = bios->bi_next; 138945bb912bSLars Ellenberg bio_put(bio); 139045bb912bSLars Ellenberg } 139110f6d992SLars Ellenberg return err; 139245bb912bSLars Ellenberg } 139345bb912bSLars Ellenberg 1394b30ab791SAndreas Gruenbacher static void drbd_remove_epoch_entry_interval(struct drbd_device *device, 1395db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req) 139653840641SAndreas Gruenbacher { 1397db830c46SAndreas Gruenbacher struct drbd_interval *i = &peer_req->i; 139853840641SAndreas Gruenbacher 1399b30ab791SAndreas Gruenbacher drbd_remove_interval(&device->write_requests, i); 140053840641SAndreas Gruenbacher drbd_clear_interval(i); 140153840641SAndreas Gruenbacher 14026c852becSAndreas Gruenbacher /* Wake up any processes waiting for this peer request to complete. */ 140353840641SAndreas Gruenbacher if (i->waiting) 1404b30ab791SAndreas Gruenbacher wake_up(&device->misc_wait); 140553840641SAndreas Gruenbacher } 140653840641SAndreas Gruenbacher 1407bde89a9eSAndreas Gruenbacher static void conn_wait_active_ee_empty(struct drbd_connection *connection) 140877fede51SPhilipp Reisner { 1409c06ece6bSAndreas Gruenbacher struct drbd_peer_device *peer_device; 141077fede51SPhilipp Reisner int vnr; 141177fede51SPhilipp Reisner 141277fede51SPhilipp Reisner rcu_read_lock(); 1413c06ece6bSAndreas Gruenbacher idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 1414c06ece6bSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 1415c06ece6bSAndreas Gruenbacher 1416b30ab791SAndreas Gruenbacher kref_get(&device->kref); 141777fede51SPhilipp Reisner rcu_read_unlock(); 1418b30ab791SAndreas Gruenbacher drbd_wait_ee_list_empty(device, &device->active_ee); 141905a10ec7SAndreas Gruenbacher kref_put(&device->kref, drbd_destroy_device); 142077fede51SPhilipp Reisner rcu_read_lock(); 142177fede51SPhilipp Reisner } 142277fede51SPhilipp Reisner rcu_read_unlock(); 142377fede51SPhilipp Reisner } 142477fede51SPhilipp Reisner 1425bde89a9eSAndreas Gruenbacher static int receive_Barrier(struct drbd_connection *connection, struct packet_info *pi) 1426b411b363SPhilipp Reisner { 14272451fc3bSPhilipp Reisner int rv; 1428e658983aSAndreas Gruenbacher struct p_barrier *p = pi->data; 1429b411b363SPhilipp Reisner struct drbd_epoch *epoch; 1430b411b363SPhilipp Reisner 14319ed57dcbSLars Ellenberg /* FIXME these are unacked on connection, 14329ed57dcbSLars Ellenberg * not a specific (peer)device. 14339ed57dcbSLars Ellenberg */ 1434bde89a9eSAndreas Gruenbacher connection->current_epoch->barrier_nr = p->barrier; 1435bde89a9eSAndreas Gruenbacher connection->current_epoch->connection = connection; 1436bde89a9eSAndreas Gruenbacher rv = drbd_may_finish_epoch(connection, connection->current_epoch, EV_GOT_BARRIER_NR); 1437b411b363SPhilipp Reisner 1438b411b363SPhilipp Reisner /* P_BARRIER_ACK may imply that the corresponding extent is dropped from 1439b411b363SPhilipp Reisner * the activity log, which means it would not be resynced in case the 1440b411b363SPhilipp Reisner * R_PRIMARY crashes now. 1441b411b363SPhilipp Reisner * Therefore we must send the barrier_ack after the barrier request was 1442b411b363SPhilipp Reisner * completed. */ 1443bde89a9eSAndreas Gruenbacher switch (connection->write_ordering) { 1444b411b363SPhilipp Reisner case WO_none: 1445b411b363SPhilipp Reisner if (rv == FE_RECYCLED) 144682bc0194SAndreas Gruenbacher return 0; 1447b411b363SPhilipp Reisner 1448b411b363SPhilipp Reisner /* receiver context, in the writeout path of the other node. 1449b411b363SPhilipp Reisner * avoid potential distributed deadlock */ 1450b411b363SPhilipp Reisner epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); 14512451fc3bSPhilipp Reisner if (epoch) 14522451fc3bSPhilipp Reisner break; 14532451fc3bSPhilipp Reisner else 14541ec861ebSAndreas Gruenbacher drbd_warn(connection, "Allocation of an epoch failed, slowing down\n"); 14552451fc3bSPhilipp Reisner /* Fall through */ 14562451fc3bSPhilipp Reisner 14572451fc3bSPhilipp Reisner case WO_bdev_flush: 14582451fc3bSPhilipp Reisner case WO_drain_io: 1459bde89a9eSAndreas Gruenbacher conn_wait_active_ee_empty(connection); 1460bde89a9eSAndreas Gruenbacher drbd_flush(connection); 14612451fc3bSPhilipp Reisner 1462bde89a9eSAndreas Gruenbacher if (atomic_read(&connection->current_epoch->epoch_size)) { 14632451fc3bSPhilipp Reisner epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); 14642451fc3bSPhilipp Reisner if (epoch) 14652451fc3bSPhilipp Reisner break; 1466b411b363SPhilipp Reisner } 1467b411b363SPhilipp Reisner 146882bc0194SAndreas Gruenbacher return 0; 14692451fc3bSPhilipp Reisner default: 14701ec861ebSAndreas Gruenbacher drbd_err(connection, "Strangeness in connection->write_ordering %d\n", connection->write_ordering); 147182bc0194SAndreas Gruenbacher return -EIO; 1472b411b363SPhilipp Reisner } 1473b411b363SPhilipp Reisner 1474b411b363SPhilipp Reisner epoch->flags = 0; 1475b411b363SPhilipp Reisner atomic_set(&epoch->epoch_size, 0); 1476b411b363SPhilipp Reisner atomic_set(&epoch->active, 0); 1477b411b363SPhilipp Reisner 1478bde89a9eSAndreas Gruenbacher spin_lock(&connection->epoch_lock); 1479bde89a9eSAndreas Gruenbacher if (atomic_read(&connection->current_epoch->epoch_size)) { 1480bde89a9eSAndreas Gruenbacher list_add(&epoch->list, &connection->current_epoch->list); 1481bde89a9eSAndreas Gruenbacher connection->current_epoch = epoch; 1482bde89a9eSAndreas Gruenbacher connection->epochs++; 1483b411b363SPhilipp Reisner } else { 1484b411b363SPhilipp Reisner /* The current_epoch got recycled while we allocated this one... */ 1485b411b363SPhilipp Reisner kfree(epoch); 1486b411b363SPhilipp Reisner } 1487bde89a9eSAndreas Gruenbacher spin_unlock(&connection->epoch_lock); 1488b411b363SPhilipp Reisner 148982bc0194SAndreas Gruenbacher return 0; 1490b411b363SPhilipp Reisner } 1491b411b363SPhilipp Reisner 1492b411b363SPhilipp Reisner /* used from receive_RSDataReply (recv_resync_read) 1493b411b363SPhilipp Reisner * and from receive_Data */ 1494f6ffca9fSAndreas Gruenbacher static struct drbd_peer_request * 1495b30ab791SAndreas Gruenbacher read_in_block(struct drbd_device *device, u64 id, sector_t sector, 1496f6ffca9fSAndreas Gruenbacher int data_size) __must_hold(local) 1497b411b363SPhilipp Reisner { 1498b30ab791SAndreas Gruenbacher const sector_t capacity = drbd_get_capacity(device->this_bdev); 1499db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req; 1500b411b363SPhilipp Reisner struct page *page; 1501a5c31904SAndreas Gruenbacher int dgs, ds, err; 1502a6b32bc3SAndreas Gruenbacher void *dig_in = first_peer_device(device)->connection->int_dig_in; 1503a6b32bc3SAndreas Gruenbacher void *dig_vv = first_peer_device(device)->connection->int_dig_vv; 15046b4388acSPhilipp Reisner unsigned long *data; 1505b411b363SPhilipp Reisner 150688104ca4SAndreas Gruenbacher dgs = 0; 1507a6b32bc3SAndreas Gruenbacher if (first_peer_device(device)->connection->peer_integrity_tfm) { 1508a6b32bc3SAndreas Gruenbacher dgs = crypto_hash_digestsize(first_peer_device(device)->connection->peer_integrity_tfm); 15099f5bdc33SAndreas Gruenbacher /* 15109f5bdc33SAndreas Gruenbacher * FIXME: Receive the incoming digest into the receive buffer 15119f5bdc33SAndreas Gruenbacher * here, together with its struct p_data? 15129f5bdc33SAndreas Gruenbacher */ 1513a6b32bc3SAndreas Gruenbacher err = drbd_recv_all_warn(first_peer_device(device)->connection, dig_in, dgs); 1514a5c31904SAndreas Gruenbacher if (err) 1515b411b363SPhilipp Reisner return NULL; 1516b411b363SPhilipp Reisner data_size -= dgs; 151788104ca4SAndreas Gruenbacher } 1518b411b363SPhilipp Reisner 1519841ce241SAndreas Gruenbacher if (!expect(IS_ALIGNED(data_size, 512))) 1520841ce241SAndreas Gruenbacher return NULL; 1521841ce241SAndreas Gruenbacher if (!expect(data_size <= DRBD_MAX_BIO_SIZE)) 1522841ce241SAndreas Gruenbacher return NULL; 1523b411b363SPhilipp Reisner 15246666032aSLars Ellenberg /* even though we trust out peer, 15256666032aSLars Ellenberg * we sometimes have to double check. */ 15266666032aSLars Ellenberg if (sector + (data_size>>9) > capacity) { 1527d0180171SAndreas Gruenbacher drbd_err(device, "request from peer beyond end of local disk: " 1528fdda6544SLars Ellenberg "capacity: %llus < sector: %llus + size: %u\n", 15296666032aSLars Ellenberg (unsigned long long)capacity, 15306666032aSLars Ellenberg (unsigned long long)sector, data_size); 15316666032aSLars Ellenberg return NULL; 15326666032aSLars Ellenberg } 15336666032aSLars Ellenberg 1534b411b363SPhilipp Reisner /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD 1535b411b363SPhilipp Reisner * "criss-cross" setup, that might cause write-out on some other DRBD, 1536b411b363SPhilipp Reisner * which in turn might block on the other node at this very place. */ 1537b30ab791SAndreas Gruenbacher peer_req = drbd_alloc_peer_req(device, id, sector, data_size, GFP_NOIO); 1538db830c46SAndreas Gruenbacher if (!peer_req) 1539b411b363SPhilipp Reisner return NULL; 154045bb912bSLars Ellenberg 1541a73ff323SLars Ellenberg if (!data_size) 154281a3537aSLars Ellenberg return peer_req; 1543a73ff323SLars Ellenberg 1544b411b363SPhilipp Reisner ds = data_size; 1545db830c46SAndreas Gruenbacher page = peer_req->pages; 154645bb912bSLars Ellenberg page_chain_for_each(page) { 154745bb912bSLars Ellenberg unsigned len = min_t(int, ds, PAGE_SIZE); 15486b4388acSPhilipp Reisner data = kmap(page); 1549a6b32bc3SAndreas Gruenbacher err = drbd_recv_all_warn(first_peer_device(device)->connection, data, len); 1550b30ab791SAndreas Gruenbacher if (drbd_insert_fault(device, DRBD_FAULT_RECEIVE)) { 1551d0180171SAndreas Gruenbacher drbd_err(device, "Fault injection: Corrupting data on receive\n"); 15526b4388acSPhilipp Reisner data[0] = data[0] ^ (unsigned long)-1; 15536b4388acSPhilipp Reisner } 1554b411b363SPhilipp Reisner kunmap(page); 1555a5c31904SAndreas Gruenbacher if (err) { 1556b30ab791SAndreas Gruenbacher drbd_free_peer_req(device, peer_req); 1557b411b363SPhilipp Reisner return NULL; 1558b411b363SPhilipp Reisner } 1559a5c31904SAndreas Gruenbacher ds -= len; 1560b411b363SPhilipp Reisner } 1561b411b363SPhilipp Reisner 1562b411b363SPhilipp Reisner if (dgs) { 1563a6b32bc3SAndreas Gruenbacher drbd_csum_ee(device, first_peer_device(device)->connection->peer_integrity_tfm, peer_req, dig_vv); 1564b411b363SPhilipp Reisner if (memcmp(dig_in, dig_vv, dgs)) { 1565d0180171SAndreas Gruenbacher drbd_err(device, "Digest integrity check FAILED: %llus +%u\n", 1566470be44aSLars Ellenberg (unsigned long long)sector, data_size); 1567b30ab791SAndreas Gruenbacher drbd_free_peer_req(device, peer_req); 1568b411b363SPhilipp Reisner return NULL; 1569b411b363SPhilipp Reisner } 1570b411b363SPhilipp Reisner } 1571b30ab791SAndreas Gruenbacher device->recv_cnt += data_size>>9; 1572db830c46SAndreas Gruenbacher return peer_req; 1573b411b363SPhilipp Reisner } 1574b411b363SPhilipp Reisner 1575b411b363SPhilipp Reisner /* drbd_drain_block() just takes a data block 1576b411b363SPhilipp Reisner * out of the socket input buffer, and discards it. 1577b411b363SPhilipp Reisner */ 1578b30ab791SAndreas Gruenbacher static int drbd_drain_block(struct drbd_device *device, int data_size) 1579b411b363SPhilipp Reisner { 1580b411b363SPhilipp Reisner struct page *page; 1581a5c31904SAndreas Gruenbacher int err = 0; 1582b411b363SPhilipp Reisner void *data; 1583b411b363SPhilipp Reisner 1584c3470cdeSLars Ellenberg if (!data_size) 1585fc5be839SAndreas Gruenbacher return 0; 1586c3470cdeSLars Ellenberg 1587b30ab791SAndreas Gruenbacher page = drbd_alloc_pages(device, 1, 1); 1588b411b363SPhilipp Reisner 1589b411b363SPhilipp Reisner data = kmap(page); 1590b411b363SPhilipp Reisner while (data_size) { 1591fc5be839SAndreas Gruenbacher unsigned int len = min_t(int, data_size, PAGE_SIZE); 1592fc5be839SAndreas Gruenbacher 1593a6b32bc3SAndreas Gruenbacher err = drbd_recv_all_warn(first_peer_device(device)->connection, data, len); 1594a5c31904SAndreas Gruenbacher if (err) 1595b411b363SPhilipp Reisner break; 1596a5c31904SAndreas Gruenbacher data_size -= len; 1597b411b363SPhilipp Reisner } 1598b411b363SPhilipp Reisner kunmap(page); 1599b30ab791SAndreas Gruenbacher drbd_free_pages(device, page, 0); 1600fc5be839SAndreas Gruenbacher return err; 1601b411b363SPhilipp Reisner } 1602b411b363SPhilipp Reisner 1603b30ab791SAndreas Gruenbacher static int recv_dless_read(struct drbd_device *device, struct drbd_request *req, 1604b411b363SPhilipp Reisner sector_t sector, int data_size) 1605b411b363SPhilipp Reisner { 16067988613bSKent Overstreet struct bio_vec bvec; 16077988613bSKent Overstreet struct bvec_iter iter; 1608b411b363SPhilipp Reisner struct bio *bio; 16097988613bSKent Overstreet int dgs, err, expect; 1610a6b32bc3SAndreas Gruenbacher void *dig_in = first_peer_device(device)->connection->int_dig_in; 1611a6b32bc3SAndreas Gruenbacher void *dig_vv = first_peer_device(device)->connection->int_dig_vv; 1612b411b363SPhilipp Reisner 161388104ca4SAndreas Gruenbacher dgs = 0; 1614a6b32bc3SAndreas Gruenbacher if (first_peer_device(device)->connection->peer_integrity_tfm) { 1615a6b32bc3SAndreas Gruenbacher dgs = crypto_hash_digestsize(first_peer_device(device)->connection->peer_integrity_tfm); 1616a6b32bc3SAndreas Gruenbacher err = drbd_recv_all_warn(first_peer_device(device)->connection, dig_in, dgs); 1617a5c31904SAndreas Gruenbacher if (err) 1618a5c31904SAndreas Gruenbacher return err; 1619b411b363SPhilipp Reisner data_size -= dgs; 162088104ca4SAndreas Gruenbacher } 1621b411b363SPhilipp Reisner 1622b411b363SPhilipp Reisner /* optimistically update recv_cnt. if receiving fails below, 1623b411b363SPhilipp Reisner * we disconnect anyways, and counters will be reset. */ 1624b30ab791SAndreas Gruenbacher device->recv_cnt += data_size>>9; 1625b411b363SPhilipp Reisner 1626b411b363SPhilipp Reisner bio = req->master_bio; 16270b0ba1efSAndreas Gruenbacher D_ASSERT(device, sector == bio->bi_iter.bi_sector); 1628b411b363SPhilipp Reisner 16297988613bSKent Overstreet bio_for_each_segment(bvec, bio, iter) { 16307988613bSKent Overstreet void *mapped = kmap(bvec.bv_page) + bvec.bv_offset; 16317988613bSKent Overstreet expect = min_t(int, data_size, bvec.bv_len); 1632a6b32bc3SAndreas Gruenbacher err = drbd_recv_all_warn(first_peer_device(device)->connection, mapped, expect); 16337988613bSKent Overstreet kunmap(bvec.bv_page); 1634a5c31904SAndreas Gruenbacher if (err) 1635a5c31904SAndreas Gruenbacher return err; 1636a5c31904SAndreas Gruenbacher data_size -= expect; 1637b411b363SPhilipp Reisner } 1638b411b363SPhilipp Reisner 1639b411b363SPhilipp Reisner if (dgs) { 1640a6b32bc3SAndreas Gruenbacher drbd_csum_bio(device, first_peer_device(device)->connection->peer_integrity_tfm, bio, dig_vv); 1641b411b363SPhilipp Reisner if (memcmp(dig_in, dig_vv, dgs)) { 1642d0180171SAndreas Gruenbacher drbd_err(device, "Digest integrity check FAILED. Broken NICs?\n"); 164328284cefSAndreas Gruenbacher return -EINVAL; 1644b411b363SPhilipp Reisner } 1645b411b363SPhilipp Reisner } 1646b411b363SPhilipp Reisner 16470b0ba1efSAndreas Gruenbacher D_ASSERT(device, data_size == 0); 164828284cefSAndreas Gruenbacher return 0; 1649b411b363SPhilipp Reisner } 1650b411b363SPhilipp Reisner 1651a990be46SAndreas Gruenbacher /* 1652a990be46SAndreas Gruenbacher * e_end_resync_block() is called in asender context via 1653a990be46SAndreas Gruenbacher * drbd_finish_peer_reqs(). 1654a990be46SAndreas Gruenbacher */ 165599920dc5SAndreas Gruenbacher static int e_end_resync_block(struct drbd_work *w, int unused) 1656b411b363SPhilipp Reisner { 16578050e6d0SAndreas Gruenbacher struct drbd_peer_request *peer_req = 16588050e6d0SAndreas Gruenbacher container_of(w, struct drbd_peer_request, w); 1659b30ab791SAndreas Gruenbacher struct drbd_device *device = w->device; 1660db830c46SAndreas Gruenbacher sector_t sector = peer_req->i.sector; 166199920dc5SAndreas Gruenbacher int err; 1662b411b363SPhilipp Reisner 16630b0ba1efSAndreas Gruenbacher D_ASSERT(device, drbd_interval_empty(&peer_req->i)); 1664b411b363SPhilipp Reisner 1665db830c46SAndreas Gruenbacher if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { 1666b30ab791SAndreas Gruenbacher drbd_set_in_sync(device, sector, peer_req->i.size); 1667b30ab791SAndreas Gruenbacher err = drbd_send_ack(device, P_RS_WRITE_ACK, peer_req); 1668b411b363SPhilipp Reisner } else { 1669b411b363SPhilipp Reisner /* Record failure to sync */ 1670b30ab791SAndreas Gruenbacher drbd_rs_failed_io(device, sector, peer_req->i.size); 1671b411b363SPhilipp Reisner 1672b30ab791SAndreas Gruenbacher err = drbd_send_ack(device, P_NEG_ACK, peer_req); 1673b411b363SPhilipp Reisner } 1674b30ab791SAndreas Gruenbacher dec_unacked(device); 1675b411b363SPhilipp Reisner 167699920dc5SAndreas Gruenbacher return err; 1677b411b363SPhilipp Reisner } 1678b411b363SPhilipp Reisner 1679b30ab791SAndreas Gruenbacher static int recv_resync_read(struct drbd_device *device, sector_t sector, int data_size) __releases(local) 1680b411b363SPhilipp Reisner { 1681db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req; 1682b411b363SPhilipp Reisner 1683b30ab791SAndreas Gruenbacher peer_req = read_in_block(device, ID_SYNCER, sector, data_size); 1684db830c46SAndreas Gruenbacher if (!peer_req) 168545bb912bSLars Ellenberg goto fail; 1686b411b363SPhilipp Reisner 1687b30ab791SAndreas Gruenbacher dec_rs_pending(device); 1688b411b363SPhilipp Reisner 1689b30ab791SAndreas Gruenbacher inc_unacked(device); 1690b411b363SPhilipp Reisner /* corresponding dec_unacked() in e_end_resync_block() 1691b411b363SPhilipp Reisner * respective _drbd_clear_done_ee */ 1692b411b363SPhilipp Reisner 1693db830c46SAndreas Gruenbacher peer_req->w.cb = e_end_resync_block; 169445bb912bSLars Ellenberg 16950500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 1696b30ab791SAndreas Gruenbacher list_add(&peer_req->w.list, &device->sync_ee); 16970500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 1698b411b363SPhilipp Reisner 1699b30ab791SAndreas Gruenbacher atomic_add(data_size >> 9, &device->rs_sect_ev); 1700b30ab791SAndreas Gruenbacher if (drbd_submit_peer_request(device, peer_req, WRITE, DRBD_FAULT_RS_WR) == 0) 1701e1c1b0fcSAndreas Gruenbacher return 0; 170245bb912bSLars Ellenberg 170310f6d992SLars Ellenberg /* don't care for the reason here */ 1704d0180171SAndreas Gruenbacher drbd_err(device, "submit failed, triggering re-connect\n"); 17050500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 1706db830c46SAndreas Gruenbacher list_del(&peer_req->w.list); 17070500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 170822cc37a9SLars Ellenberg 1709b30ab791SAndreas Gruenbacher drbd_free_peer_req(device, peer_req); 171045bb912bSLars Ellenberg fail: 1711b30ab791SAndreas Gruenbacher put_ldev(device); 1712e1c1b0fcSAndreas Gruenbacher return -EIO; 1713b411b363SPhilipp Reisner } 1714b411b363SPhilipp Reisner 1715668eebc6SAndreas Gruenbacher static struct drbd_request * 1716b30ab791SAndreas Gruenbacher find_request(struct drbd_device *device, struct rb_root *root, u64 id, 1717bc9c5c41SAndreas Gruenbacher sector_t sector, bool missing_ok, const char *func) 1718b411b363SPhilipp Reisner { 1719b411b363SPhilipp Reisner struct drbd_request *req; 1720668eebc6SAndreas Gruenbacher 1721bc9c5c41SAndreas Gruenbacher /* Request object according to our peer */ 1722bc9c5c41SAndreas Gruenbacher req = (struct drbd_request *)(unsigned long)id; 17235e472264SAndreas Gruenbacher if (drbd_contains_interval(root, sector, &req->i) && req->i.local) 1724668eebc6SAndreas Gruenbacher return req; 1725c3afd8f5SAndreas Gruenbacher if (!missing_ok) { 1726d0180171SAndreas Gruenbacher drbd_err(device, "%s: failed to find request 0x%lx, sector %llus\n", func, 1727c3afd8f5SAndreas Gruenbacher (unsigned long)id, (unsigned long long)sector); 1728c3afd8f5SAndreas Gruenbacher } 1729668eebc6SAndreas Gruenbacher return NULL; 1730668eebc6SAndreas Gruenbacher } 1731668eebc6SAndreas Gruenbacher 1732bde89a9eSAndreas Gruenbacher static int receive_DataReply(struct drbd_connection *connection, struct packet_info *pi) 1733b411b363SPhilipp Reisner { 1734b30ab791SAndreas Gruenbacher struct drbd_device *device; 1735b411b363SPhilipp Reisner struct drbd_request *req; 1736b411b363SPhilipp Reisner sector_t sector; 173782bc0194SAndreas Gruenbacher int err; 1738e658983aSAndreas Gruenbacher struct p_data *p = pi->data; 17394a76b161SAndreas Gruenbacher 1740bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 1741b30ab791SAndreas Gruenbacher if (!device) 17424a76b161SAndreas Gruenbacher return -EIO; 1743b411b363SPhilipp Reisner 1744b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 1745b411b363SPhilipp Reisner 17460500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 1747b30ab791SAndreas Gruenbacher req = find_request(device, &device->read_requests, p->block_id, sector, false, __func__); 17480500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 1749c3afd8f5SAndreas Gruenbacher if (unlikely(!req)) 175082bc0194SAndreas Gruenbacher return -EIO; 1751b411b363SPhilipp Reisner 175224c4830cSBart Van Assche /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid 1753b411b363SPhilipp Reisner * special casing it there for the various failure cases. 1754b411b363SPhilipp Reisner * still no race with drbd_fail_pending_reads */ 1755b30ab791SAndreas Gruenbacher err = recv_dless_read(device, req, sector, pi->size); 175682bc0194SAndreas Gruenbacher if (!err) 17578554df1cSAndreas Gruenbacher req_mod(req, DATA_RECEIVED); 1758b411b363SPhilipp Reisner /* else: nothing. handled from drbd_disconnect... 1759b411b363SPhilipp Reisner * I don't think we may complete this just yet 1760b411b363SPhilipp Reisner * in case we are "on-disconnect: freeze" */ 1761b411b363SPhilipp Reisner 176282bc0194SAndreas Gruenbacher return err; 1763b411b363SPhilipp Reisner } 1764b411b363SPhilipp Reisner 1765bde89a9eSAndreas Gruenbacher static int receive_RSDataReply(struct drbd_connection *connection, struct packet_info *pi) 1766b411b363SPhilipp Reisner { 1767b30ab791SAndreas Gruenbacher struct drbd_device *device; 1768b411b363SPhilipp Reisner sector_t sector; 176982bc0194SAndreas Gruenbacher int err; 1770e658983aSAndreas Gruenbacher struct p_data *p = pi->data; 17714a76b161SAndreas Gruenbacher 1772bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 1773b30ab791SAndreas Gruenbacher if (!device) 17744a76b161SAndreas Gruenbacher return -EIO; 1775b411b363SPhilipp Reisner 1776b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 17770b0ba1efSAndreas Gruenbacher D_ASSERT(device, p->block_id == ID_SYNCER); 1778b411b363SPhilipp Reisner 1779b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 1780b411b363SPhilipp Reisner /* data is submitted to disk within recv_resync_read. 1781b411b363SPhilipp Reisner * corresponding put_ldev done below on error, 1782fcefa62eSAndreas Gruenbacher * or in drbd_peer_request_endio. */ 1783b30ab791SAndreas Gruenbacher err = recv_resync_read(device, sector, pi->size); 1784b411b363SPhilipp Reisner } else { 1785b411b363SPhilipp Reisner if (__ratelimit(&drbd_ratelimit_state)) 1786d0180171SAndreas Gruenbacher drbd_err(device, "Can not write resync data to local disk.\n"); 1787b411b363SPhilipp Reisner 1788b30ab791SAndreas Gruenbacher err = drbd_drain_block(device, pi->size); 1789b411b363SPhilipp Reisner 1790b30ab791SAndreas Gruenbacher drbd_send_ack_dp(device, P_NEG_ACK, p, pi->size); 1791b411b363SPhilipp Reisner } 1792b411b363SPhilipp Reisner 1793b30ab791SAndreas Gruenbacher atomic_add(pi->size >> 9, &device->rs_sect_in); 1794778f271dSPhilipp Reisner 179582bc0194SAndreas Gruenbacher return err; 1796b411b363SPhilipp Reisner } 1797b411b363SPhilipp Reisner 1798b30ab791SAndreas Gruenbacher static void restart_conflicting_writes(struct drbd_device *device, 17997be8da07SAndreas Gruenbacher sector_t sector, int size) 1800b411b363SPhilipp Reisner { 18017be8da07SAndreas Gruenbacher struct drbd_interval *i; 18027be8da07SAndreas Gruenbacher struct drbd_request *req; 1803b411b363SPhilipp Reisner 1804b30ab791SAndreas Gruenbacher drbd_for_each_overlap(i, &device->write_requests, sector, size) { 18057be8da07SAndreas Gruenbacher if (!i->local) 18067be8da07SAndreas Gruenbacher continue; 18077be8da07SAndreas Gruenbacher req = container_of(i, struct drbd_request, i); 18087be8da07SAndreas Gruenbacher if (req->rq_state & RQ_LOCAL_PENDING || 18097be8da07SAndreas Gruenbacher !(req->rq_state & RQ_POSTPONED)) 18107be8da07SAndreas Gruenbacher continue; 18112312f0b3SLars Ellenberg /* as it is RQ_POSTPONED, this will cause it to 18122312f0b3SLars Ellenberg * be queued on the retry workqueue. */ 1813d4dabbe2SLars Ellenberg __req_mod(req, CONFLICT_RESOLVED, NULL); 18147be8da07SAndreas Gruenbacher } 18157be8da07SAndreas Gruenbacher } 18167be8da07SAndreas Gruenbacher 1817a990be46SAndreas Gruenbacher /* 1818a990be46SAndreas Gruenbacher * e_end_block() is called in asender context via drbd_finish_peer_reqs(). 1819b411b363SPhilipp Reisner */ 182099920dc5SAndreas Gruenbacher static int e_end_block(struct drbd_work *w, int cancel) 1821b411b363SPhilipp Reisner { 18228050e6d0SAndreas Gruenbacher struct drbd_peer_request *peer_req = 18238050e6d0SAndreas Gruenbacher container_of(w, struct drbd_peer_request, w); 1824b30ab791SAndreas Gruenbacher struct drbd_device *device = w->device; 1825db830c46SAndreas Gruenbacher sector_t sector = peer_req->i.sector; 182699920dc5SAndreas Gruenbacher int err = 0, pcmd; 1827b411b363SPhilipp Reisner 1828303d1448SPhilipp Reisner if (peer_req->flags & EE_SEND_WRITE_ACK) { 1829db830c46SAndreas Gruenbacher if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { 1830b30ab791SAndreas Gruenbacher pcmd = (device->state.conn >= C_SYNC_SOURCE && 1831b30ab791SAndreas Gruenbacher device->state.conn <= C_PAUSED_SYNC_T && 1832db830c46SAndreas Gruenbacher peer_req->flags & EE_MAY_SET_IN_SYNC) ? 1833b411b363SPhilipp Reisner P_RS_WRITE_ACK : P_WRITE_ACK; 1834b30ab791SAndreas Gruenbacher err = drbd_send_ack(device, pcmd, peer_req); 1835b411b363SPhilipp Reisner if (pcmd == P_RS_WRITE_ACK) 1836b30ab791SAndreas Gruenbacher drbd_set_in_sync(device, sector, peer_req->i.size); 1837b411b363SPhilipp Reisner } else { 1838b30ab791SAndreas Gruenbacher err = drbd_send_ack(device, P_NEG_ACK, peer_req); 1839b411b363SPhilipp Reisner /* we expect it to be marked out of sync anyways... 1840b411b363SPhilipp Reisner * maybe assert this? */ 1841b411b363SPhilipp Reisner } 1842b30ab791SAndreas Gruenbacher dec_unacked(device); 1843b411b363SPhilipp Reisner } 1844b411b363SPhilipp Reisner /* we delete from the conflict detection hash _after_ we sent out the 1845b411b363SPhilipp Reisner * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ 1846302bdeaeSPhilipp Reisner if (peer_req->flags & EE_IN_INTERVAL_TREE) { 18470500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 18480b0ba1efSAndreas Gruenbacher D_ASSERT(device, !drbd_interval_empty(&peer_req->i)); 1849b30ab791SAndreas Gruenbacher drbd_remove_epoch_entry_interval(device, peer_req); 18507be8da07SAndreas Gruenbacher if (peer_req->flags & EE_RESTART_REQUESTS) 1851b30ab791SAndreas Gruenbacher restart_conflicting_writes(device, sector, peer_req->i.size); 18520500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 1853bb3bfe96SAndreas Gruenbacher } else 18540b0ba1efSAndreas Gruenbacher D_ASSERT(device, drbd_interval_empty(&peer_req->i)); 1855b411b363SPhilipp Reisner 1856a6b32bc3SAndreas Gruenbacher drbd_may_finish_epoch(first_peer_device(device)->connection, peer_req->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); 1857b411b363SPhilipp Reisner 185899920dc5SAndreas Gruenbacher return err; 1859b411b363SPhilipp Reisner } 1860b411b363SPhilipp Reisner 18617be8da07SAndreas Gruenbacher static int e_send_ack(struct drbd_work *w, enum drbd_packet ack) 1862b411b363SPhilipp Reisner { 1863b30ab791SAndreas Gruenbacher struct drbd_device *device = w->device; 18648050e6d0SAndreas Gruenbacher struct drbd_peer_request *peer_req = 18658050e6d0SAndreas Gruenbacher container_of(w, struct drbd_peer_request, w); 186699920dc5SAndreas Gruenbacher int err; 1867b411b363SPhilipp Reisner 1868b30ab791SAndreas Gruenbacher err = drbd_send_ack(device, ack, peer_req); 1869b30ab791SAndreas Gruenbacher dec_unacked(device); 1870b411b363SPhilipp Reisner 187199920dc5SAndreas Gruenbacher return err; 1872b411b363SPhilipp Reisner } 1873b411b363SPhilipp Reisner 1874d4dabbe2SLars Ellenberg static int e_send_superseded(struct drbd_work *w, int unused) 1875b6a370baSPhilipp Reisner { 1876d4dabbe2SLars Ellenberg return e_send_ack(w, P_SUPERSEDED); 18777be8da07SAndreas Gruenbacher } 1878b6a370baSPhilipp Reisner 187999920dc5SAndreas Gruenbacher static int e_send_retry_write(struct drbd_work *w, int unused) 18807be8da07SAndreas Gruenbacher { 1881a6b32bc3SAndreas Gruenbacher struct drbd_connection *connection = first_peer_device(w->device)->connection; 18827be8da07SAndreas Gruenbacher 1883bde89a9eSAndreas Gruenbacher return e_send_ack(w, connection->agreed_pro_version >= 100 ? 1884d4dabbe2SLars Ellenberg P_RETRY_WRITE : P_SUPERSEDED); 18857be8da07SAndreas Gruenbacher } 18867be8da07SAndreas Gruenbacher 18873e394da1SAndreas Gruenbacher static bool seq_greater(u32 a, u32 b) 18883e394da1SAndreas Gruenbacher { 18893e394da1SAndreas Gruenbacher /* 18903e394da1SAndreas Gruenbacher * We assume 32-bit wrap-around here. 18913e394da1SAndreas Gruenbacher * For 24-bit wrap-around, we would have to shift: 18923e394da1SAndreas Gruenbacher * a <<= 8; b <<= 8; 18933e394da1SAndreas Gruenbacher */ 18943e394da1SAndreas Gruenbacher return (s32)a - (s32)b > 0; 18953e394da1SAndreas Gruenbacher } 18963e394da1SAndreas Gruenbacher 18973e394da1SAndreas Gruenbacher static u32 seq_max(u32 a, u32 b) 18983e394da1SAndreas Gruenbacher { 18993e394da1SAndreas Gruenbacher return seq_greater(a, b) ? a : b; 19003e394da1SAndreas Gruenbacher } 19013e394da1SAndreas Gruenbacher 1902b30ab791SAndreas Gruenbacher static void update_peer_seq(struct drbd_device *device, unsigned int peer_seq) 19033e394da1SAndreas Gruenbacher { 19043c13b680SLars Ellenberg unsigned int newest_peer_seq; 19053e394da1SAndreas Gruenbacher 1906a6b32bc3SAndreas Gruenbacher if (test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags)) { 1907b30ab791SAndreas Gruenbacher spin_lock(&device->peer_seq_lock); 1908b30ab791SAndreas Gruenbacher newest_peer_seq = seq_max(device->peer_seq, peer_seq); 1909b30ab791SAndreas Gruenbacher device->peer_seq = newest_peer_seq; 1910b30ab791SAndreas Gruenbacher spin_unlock(&device->peer_seq_lock); 1911b30ab791SAndreas Gruenbacher /* wake up only if we actually changed device->peer_seq */ 19123c13b680SLars Ellenberg if (peer_seq == newest_peer_seq) 1913b30ab791SAndreas Gruenbacher wake_up(&device->seq_wait); 19143e394da1SAndreas Gruenbacher } 19157be8da07SAndreas Gruenbacher } 19163e394da1SAndreas Gruenbacher 1917d93f6302SLars Ellenberg static inline int overlaps(sector_t s1, int l1, sector_t s2, int l2) 1918d93f6302SLars Ellenberg { 1919d93f6302SLars Ellenberg return !((s1 + (l1>>9) <= s2) || (s1 >= s2 + (l2>>9))); 1920d93f6302SLars Ellenberg } 1921d93f6302SLars Ellenberg 1922d93f6302SLars Ellenberg /* maybe change sync_ee into interval trees as well? */ 1923b30ab791SAndreas Gruenbacher static bool overlapping_resync_write(struct drbd_device *device, struct drbd_peer_request *peer_req) 1924d93f6302SLars Ellenberg { 1925d93f6302SLars Ellenberg struct drbd_peer_request *rs_req; 1926b6a370baSPhilipp Reisner bool rv = 0; 1927b6a370baSPhilipp Reisner 19280500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 1929b30ab791SAndreas Gruenbacher list_for_each_entry(rs_req, &device->sync_ee, w.list) { 1930d93f6302SLars Ellenberg if (overlaps(peer_req->i.sector, peer_req->i.size, 1931d93f6302SLars Ellenberg rs_req->i.sector, rs_req->i.size)) { 1932b6a370baSPhilipp Reisner rv = 1; 1933b6a370baSPhilipp Reisner break; 1934b6a370baSPhilipp Reisner } 1935b6a370baSPhilipp Reisner } 19360500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 1937b6a370baSPhilipp Reisner 1938b6a370baSPhilipp Reisner return rv; 1939b6a370baSPhilipp Reisner } 1940b6a370baSPhilipp Reisner 1941b411b363SPhilipp Reisner /* Called from receive_Data. 1942b411b363SPhilipp Reisner * Synchronize packets on sock with packets on msock. 1943b411b363SPhilipp Reisner * 1944b411b363SPhilipp Reisner * This is here so even when a P_DATA packet traveling via sock overtook an Ack 1945b411b363SPhilipp Reisner * packet traveling on msock, they are still processed in the order they have 1946b411b363SPhilipp Reisner * been sent. 1947b411b363SPhilipp Reisner * 1948b411b363SPhilipp Reisner * Note: we don't care for Ack packets overtaking P_DATA packets. 1949b411b363SPhilipp Reisner * 1950b30ab791SAndreas Gruenbacher * In case packet_seq is larger than device->peer_seq number, there are 1951b411b363SPhilipp Reisner * outstanding packets on the msock. We wait for them to arrive. 1952b30ab791SAndreas Gruenbacher * In case we are the logically next packet, we update device->peer_seq 1953b411b363SPhilipp Reisner * ourselves. Correctly handles 32bit wrap around. 1954b411b363SPhilipp Reisner * 1955b411b363SPhilipp Reisner * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, 1956b411b363SPhilipp Reisner * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds 1957b411b363SPhilipp Reisner * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have 1958b411b363SPhilipp Reisner * 1<<9 == 512 seconds aka ages for the 32bit wrap around... 1959b411b363SPhilipp Reisner * 1960b411b363SPhilipp Reisner * returns 0 if we may process the packet, 1961b411b363SPhilipp Reisner * -ERESTARTSYS if we were interrupted (by disconnect signal). */ 1962b30ab791SAndreas Gruenbacher static int wait_for_and_update_peer_seq(struct drbd_device *device, const u32 peer_seq) 1963b411b363SPhilipp Reisner { 1964b411b363SPhilipp Reisner DEFINE_WAIT(wait); 1965b411b363SPhilipp Reisner long timeout; 1966b874d231SPhilipp Reisner int ret = 0, tp; 19677be8da07SAndreas Gruenbacher 1968a6b32bc3SAndreas Gruenbacher if (!test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags)) 19697be8da07SAndreas Gruenbacher return 0; 19707be8da07SAndreas Gruenbacher 1971b30ab791SAndreas Gruenbacher spin_lock(&device->peer_seq_lock); 1972b411b363SPhilipp Reisner for (;;) { 1973b30ab791SAndreas Gruenbacher if (!seq_greater(peer_seq - 1, device->peer_seq)) { 1974b30ab791SAndreas Gruenbacher device->peer_seq = seq_max(device->peer_seq, peer_seq); 1975b411b363SPhilipp Reisner break; 19767be8da07SAndreas Gruenbacher } 1977b874d231SPhilipp Reisner 1978b411b363SPhilipp Reisner if (signal_pending(current)) { 1979b411b363SPhilipp Reisner ret = -ERESTARTSYS; 1980b411b363SPhilipp Reisner break; 1981b411b363SPhilipp Reisner } 1982b874d231SPhilipp Reisner 1983b874d231SPhilipp Reisner rcu_read_lock(); 1984a6b32bc3SAndreas Gruenbacher tp = rcu_dereference(first_peer_device(device)->connection->net_conf)->two_primaries; 1985b874d231SPhilipp Reisner rcu_read_unlock(); 1986b874d231SPhilipp Reisner 1987b874d231SPhilipp Reisner if (!tp) 1988b874d231SPhilipp Reisner break; 1989b874d231SPhilipp Reisner 1990b874d231SPhilipp Reisner /* Only need to wait if two_primaries is enabled */ 1991b30ab791SAndreas Gruenbacher prepare_to_wait(&device->seq_wait, &wait, TASK_INTERRUPTIBLE); 1992b30ab791SAndreas Gruenbacher spin_unlock(&device->peer_seq_lock); 199344ed167dSPhilipp Reisner rcu_read_lock(); 1994a6b32bc3SAndreas Gruenbacher timeout = rcu_dereference(first_peer_device(device)->connection->net_conf)->ping_timeo*HZ/10; 199544ed167dSPhilipp Reisner rcu_read_unlock(); 199671b1c1ebSAndreas Gruenbacher timeout = schedule_timeout(timeout); 1997b30ab791SAndreas Gruenbacher spin_lock(&device->peer_seq_lock); 19987be8da07SAndreas Gruenbacher if (!timeout) { 1999b411b363SPhilipp Reisner ret = -ETIMEDOUT; 2000d0180171SAndreas Gruenbacher drbd_err(device, "Timed out waiting for missing ack packets; disconnecting\n"); 2001b411b363SPhilipp Reisner break; 2002b411b363SPhilipp Reisner } 2003b411b363SPhilipp Reisner } 2004b30ab791SAndreas Gruenbacher spin_unlock(&device->peer_seq_lock); 2005b30ab791SAndreas Gruenbacher finish_wait(&device->seq_wait, &wait); 2006b411b363SPhilipp Reisner return ret; 2007b411b363SPhilipp Reisner } 2008b411b363SPhilipp Reisner 2009688593c5SLars Ellenberg /* see also bio_flags_to_wire() 2010688593c5SLars Ellenberg * DRBD_REQ_*, because we need to semantically map the flags to data packet 2011688593c5SLars Ellenberg * flags and back. We may replicate to other kernel versions. */ 2012b30ab791SAndreas Gruenbacher static unsigned long wire_flags_to_bio(struct drbd_device *device, u32 dpf) 201376d2e7ecSPhilipp Reisner { 201476d2e7ecSPhilipp Reisner return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | 201576d2e7ecSPhilipp Reisner (dpf & DP_FUA ? REQ_FUA : 0) | 2016688593c5SLars Ellenberg (dpf & DP_FLUSH ? REQ_FLUSH : 0) | 201776d2e7ecSPhilipp Reisner (dpf & DP_DISCARD ? REQ_DISCARD : 0); 201876d2e7ecSPhilipp Reisner } 201976d2e7ecSPhilipp Reisner 2020b30ab791SAndreas Gruenbacher static void fail_postponed_requests(struct drbd_device *device, sector_t sector, 20217be8da07SAndreas Gruenbacher unsigned int size) 2022b411b363SPhilipp Reisner { 20237be8da07SAndreas Gruenbacher struct drbd_interval *i; 20247be8da07SAndreas Gruenbacher 20257be8da07SAndreas Gruenbacher repeat: 2026b30ab791SAndreas Gruenbacher drbd_for_each_overlap(i, &device->write_requests, sector, size) { 20277be8da07SAndreas Gruenbacher struct drbd_request *req; 20287be8da07SAndreas Gruenbacher struct bio_and_error m; 20297be8da07SAndreas Gruenbacher 20307be8da07SAndreas Gruenbacher if (!i->local) 20317be8da07SAndreas Gruenbacher continue; 20327be8da07SAndreas Gruenbacher req = container_of(i, struct drbd_request, i); 20337be8da07SAndreas Gruenbacher if (!(req->rq_state & RQ_POSTPONED)) 20347be8da07SAndreas Gruenbacher continue; 20357be8da07SAndreas Gruenbacher req->rq_state &= ~RQ_POSTPONED; 20367be8da07SAndreas Gruenbacher __req_mod(req, NEG_ACKED, &m); 20370500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 20387be8da07SAndreas Gruenbacher if (m.bio) 2039b30ab791SAndreas Gruenbacher complete_master_bio(device, &m); 20400500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 20417be8da07SAndreas Gruenbacher goto repeat; 20427be8da07SAndreas Gruenbacher } 20437be8da07SAndreas Gruenbacher } 20447be8da07SAndreas Gruenbacher 2045b30ab791SAndreas Gruenbacher static int handle_write_conflicts(struct drbd_device *device, 20467be8da07SAndreas Gruenbacher struct drbd_peer_request *peer_req) 20477be8da07SAndreas Gruenbacher { 2048a6b32bc3SAndreas Gruenbacher struct drbd_connection *connection = first_peer_device(device)->connection; 2049bde89a9eSAndreas Gruenbacher bool resolve_conflicts = test_bit(RESOLVE_CONFLICTS, &connection->flags); 20507be8da07SAndreas Gruenbacher sector_t sector = peer_req->i.sector; 20517be8da07SAndreas Gruenbacher const unsigned int size = peer_req->i.size; 20527be8da07SAndreas Gruenbacher struct drbd_interval *i; 20537be8da07SAndreas Gruenbacher bool equal; 20547be8da07SAndreas Gruenbacher int err; 20557be8da07SAndreas Gruenbacher 20567be8da07SAndreas Gruenbacher /* 20577be8da07SAndreas Gruenbacher * Inserting the peer request into the write_requests tree will prevent 20587be8da07SAndreas Gruenbacher * new conflicting local requests from being added. 20597be8da07SAndreas Gruenbacher */ 2060b30ab791SAndreas Gruenbacher drbd_insert_interval(&device->write_requests, &peer_req->i); 20617be8da07SAndreas Gruenbacher 20627be8da07SAndreas Gruenbacher repeat: 2063b30ab791SAndreas Gruenbacher drbd_for_each_overlap(i, &device->write_requests, sector, size) { 20647be8da07SAndreas Gruenbacher if (i == &peer_req->i) 20657be8da07SAndreas Gruenbacher continue; 20667be8da07SAndreas Gruenbacher 20677be8da07SAndreas Gruenbacher if (!i->local) { 20687be8da07SAndreas Gruenbacher /* 20697be8da07SAndreas Gruenbacher * Our peer has sent a conflicting remote request; this 20707be8da07SAndreas Gruenbacher * should not happen in a two-node setup. Wait for the 20717be8da07SAndreas Gruenbacher * earlier peer request to complete. 20727be8da07SAndreas Gruenbacher */ 2073b30ab791SAndreas Gruenbacher err = drbd_wait_misc(device, i); 20747be8da07SAndreas Gruenbacher if (err) 20757be8da07SAndreas Gruenbacher goto out; 20767be8da07SAndreas Gruenbacher goto repeat; 20777be8da07SAndreas Gruenbacher } 20787be8da07SAndreas Gruenbacher 20797be8da07SAndreas Gruenbacher equal = i->sector == sector && i->size == size; 20807be8da07SAndreas Gruenbacher if (resolve_conflicts) { 20817be8da07SAndreas Gruenbacher /* 20827be8da07SAndreas Gruenbacher * If the peer request is fully contained within the 2083d4dabbe2SLars Ellenberg * overlapping request, it can be considered overwritten 2084d4dabbe2SLars Ellenberg * and thus superseded; otherwise, it will be retried 2085d4dabbe2SLars Ellenberg * once all overlapping requests have completed. 20867be8da07SAndreas Gruenbacher */ 2087d4dabbe2SLars Ellenberg bool superseded = i->sector <= sector && i->sector + 20887be8da07SAndreas Gruenbacher (i->size >> 9) >= sector + (size >> 9); 20897be8da07SAndreas Gruenbacher 20907be8da07SAndreas Gruenbacher if (!equal) 2091d0180171SAndreas Gruenbacher drbd_alert(device, "Concurrent writes detected: " 20927be8da07SAndreas Gruenbacher "local=%llus +%u, remote=%llus +%u, " 20937be8da07SAndreas Gruenbacher "assuming %s came first\n", 20947be8da07SAndreas Gruenbacher (unsigned long long)i->sector, i->size, 20957be8da07SAndreas Gruenbacher (unsigned long long)sector, size, 2096d4dabbe2SLars Ellenberg superseded ? "local" : "remote"); 20977be8da07SAndreas Gruenbacher 2098b30ab791SAndreas Gruenbacher inc_unacked(device); 2099d4dabbe2SLars Ellenberg peer_req->w.cb = superseded ? e_send_superseded : 21007be8da07SAndreas Gruenbacher e_send_retry_write; 2101b30ab791SAndreas Gruenbacher list_add_tail(&peer_req->w.list, &device->done_ee); 2102a6b32bc3SAndreas Gruenbacher wake_asender(first_peer_device(device)->connection); 21037be8da07SAndreas Gruenbacher 21047be8da07SAndreas Gruenbacher err = -ENOENT; 21057be8da07SAndreas Gruenbacher goto out; 21067be8da07SAndreas Gruenbacher } else { 21077be8da07SAndreas Gruenbacher struct drbd_request *req = 21087be8da07SAndreas Gruenbacher container_of(i, struct drbd_request, i); 21097be8da07SAndreas Gruenbacher 21107be8da07SAndreas Gruenbacher if (!equal) 2111d0180171SAndreas Gruenbacher drbd_alert(device, "Concurrent writes detected: " 21127be8da07SAndreas Gruenbacher "local=%llus +%u, remote=%llus +%u\n", 21137be8da07SAndreas Gruenbacher (unsigned long long)i->sector, i->size, 21147be8da07SAndreas Gruenbacher (unsigned long long)sector, size); 21157be8da07SAndreas Gruenbacher 21167be8da07SAndreas Gruenbacher if (req->rq_state & RQ_LOCAL_PENDING || 21177be8da07SAndreas Gruenbacher !(req->rq_state & RQ_POSTPONED)) { 21187be8da07SAndreas Gruenbacher /* 21197be8da07SAndreas Gruenbacher * Wait for the node with the discard flag to 2120d4dabbe2SLars Ellenberg * decide if this request has been superseded 2121d4dabbe2SLars Ellenberg * or needs to be retried. 2122d4dabbe2SLars Ellenberg * Requests that have been superseded will 21237be8da07SAndreas Gruenbacher * disappear from the write_requests tree. 21247be8da07SAndreas Gruenbacher * 21257be8da07SAndreas Gruenbacher * In addition, wait for the conflicting 21267be8da07SAndreas Gruenbacher * request to finish locally before submitting 21277be8da07SAndreas Gruenbacher * the conflicting peer request. 21287be8da07SAndreas Gruenbacher */ 2129b30ab791SAndreas Gruenbacher err = drbd_wait_misc(device, &req->i); 21307be8da07SAndreas Gruenbacher if (err) { 2131a6b32bc3SAndreas Gruenbacher _conn_request_state(first_peer_device(device)->connection, 21327be8da07SAndreas Gruenbacher NS(conn, C_TIMEOUT), 21337be8da07SAndreas Gruenbacher CS_HARD); 2134b30ab791SAndreas Gruenbacher fail_postponed_requests(device, sector, size); 21357be8da07SAndreas Gruenbacher goto out; 21367be8da07SAndreas Gruenbacher } 21377be8da07SAndreas Gruenbacher goto repeat; 21387be8da07SAndreas Gruenbacher } 21397be8da07SAndreas Gruenbacher /* 21407be8da07SAndreas Gruenbacher * Remember to restart the conflicting requests after 21417be8da07SAndreas Gruenbacher * the new peer request has completed. 21427be8da07SAndreas Gruenbacher */ 21437be8da07SAndreas Gruenbacher peer_req->flags |= EE_RESTART_REQUESTS; 21447be8da07SAndreas Gruenbacher } 21457be8da07SAndreas Gruenbacher } 21467be8da07SAndreas Gruenbacher err = 0; 21477be8da07SAndreas Gruenbacher 21487be8da07SAndreas Gruenbacher out: 21497be8da07SAndreas Gruenbacher if (err) 2150b30ab791SAndreas Gruenbacher drbd_remove_epoch_entry_interval(device, peer_req); 21517be8da07SAndreas Gruenbacher return err; 21527be8da07SAndreas Gruenbacher } 21537be8da07SAndreas Gruenbacher 2154b411b363SPhilipp Reisner /* mirrored write */ 2155bde89a9eSAndreas Gruenbacher static int receive_Data(struct drbd_connection *connection, struct packet_info *pi) 2156b411b363SPhilipp Reisner { 2157b30ab791SAndreas Gruenbacher struct drbd_device *device; 2158b411b363SPhilipp Reisner sector_t sector; 2159db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req; 2160e658983aSAndreas Gruenbacher struct p_data *p = pi->data; 21617be8da07SAndreas Gruenbacher u32 peer_seq = be32_to_cpu(p->seq_num); 2162b411b363SPhilipp Reisner int rw = WRITE; 2163b411b363SPhilipp Reisner u32 dp_flags; 2164302bdeaeSPhilipp Reisner int err, tp; 21657be8da07SAndreas Gruenbacher 2166bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 2167b30ab791SAndreas Gruenbacher if (!device) 21684a76b161SAndreas Gruenbacher return -EIO; 2169b411b363SPhilipp Reisner 2170b30ab791SAndreas Gruenbacher if (!get_ldev(device)) { 217182bc0194SAndreas Gruenbacher int err2; 2172b411b363SPhilipp Reisner 2173b30ab791SAndreas Gruenbacher err = wait_for_and_update_peer_seq(device, peer_seq); 2174b30ab791SAndreas Gruenbacher drbd_send_ack_dp(device, P_NEG_ACK, p, pi->size); 2175bde89a9eSAndreas Gruenbacher atomic_inc(&connection->current_epoch->epoch_size); 2176b30ab791SAndreas Gruenbacher err2 = drbd_drain_block(device, pi->size); 217782bc0194SAndreas Gruenbacher if (!err) 217882bc0194SAndreas Gruenbacher err = err2; 217982bc0194SAndreas Gruenbacher return err; 2180b411b363SPhilipp Reisner } 2181b411b363SPhilipp Reisner 2182fcefa62eSAndreas Gruenbacher /* 2183fcefa62eSAndreas Gruenbacher * Corresponding put_ldev done either below (on various errors), or in 2184fcefa62eSAndreas Gruenbacher * drbd_peer_request_endio, if we successfully submit the data at the 2185fcefa62eSAndreas Gruenbacher * end of this function. 2186fcefa62eSAndreas Gruenbacher */ 2187b411b363SPhilipp Reisner 2188b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 2189b30ab791SAndreas Gruenbacher peer_req = read_in_block(device, p->block_id, sector, pi->size); 2190db830c46SAndreas Gruenbacher if (!peer_req) { 2191b30ab791SAndreas Gruenbacher put_ldev(device); 219282bc0194SAndreas Gruenbacher return -EIO; 2193b411b363SPhilipp Reisner } 2194b411b363SPhilipp Reisner 2195db830c46SAndreas Gruenbacher peer_req->w.cb = e_end_block; 2196b411b363SPhilipp Reisner 2197688593c5SLars Ellenberg dp_flags = be32_to_cpu(p->dp_flags); 2198b30ab791SAndreas Gruenbacher rw |= wire_flags_to_bio(device, dp_flags); 219981a3537aSLars Ellenberg if (peer_req->pages == NULL) { 22000b0ba1efSAndreas Gruenbacher D_ASSERT(device, peer_req->i.size == 0); 22010b0ba1efSAndreas Gruenbacher D_ASSERT(device, dp_flags & DP_FLUSH); 2202a73ff323SLars Ellenberg } 2203688593c5SLars Ellenberg 2204688593c5SLars Ellenberg if (dp_flags & DP_MAY_SET_IN_SYNC) 2205db830c46SAndreas Gruenbacher peer_req->flags |= EE_MAY_SET_IN_SYNC; 2206688593c5SLars Ellenberg 2207bde89a9eSAndreas Gruenbacher spin_lock(&connection->epoch_lock); 2208bde89a9eSAndreas Gruenbacher peer_req->epoch = connection->current_epoch; 2209db830c46SAndreas Gruenbacher atomic_inc(&peer_req->epoch->epoch_size); 2210db830c46SAndreas Gruenbacher atomic_inc(&peer_req->epoch->active); 2211bde89a9eSAndreas Gruenbacher spin_unlock(&connection->epoch_lock); 2212b411b363SPhilipp Reisner 2213302bdeaeSPhilipp Reisner rcu_read_lock(); 2214a6b32bc3SAndreas Gruenbacher tp = rcu_dereference(first_peer_device(device)->connection->net_conf)->two_primaries; 2215302bdeaeSPhilipp Reisner rcu_read_unlock(); 2216302bdeaeSPhilipp Reisner if (tp) { 2217302bdeaeSPhilipp Reisner peer_req->flags |= EE_IN_INTERVAL_TREE; 2218b30ab791SAndreas Gruenbacher err = wait_for_and_update_peer_seq(device, peer_seq); 22197be8da07SAndreas Gruenbacher if (err) 2220b411b363SPhilipp Reisner goto out_interrupted; 22210500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 2222b30ab791SAndreas Gruenbacher err = handle_write_conflicts(device, peer_req); 22237be8da07SAndreas Gruenbacher if (err) { 22240500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 22257be8da07SAndreas Gruenbacher if (err == -ENOENT) { 2226b30ab791SAndreas Gruenbacher put_ldev(device); 222782bc0194SAndreas Gruenbacher return 0; 2228b411b363SPhilipp Reisner } 2229b411b363SPhilipp Reisner goto out_interrupted; 2230b411b363SPhilipp Reisner } 2231b874d231SPhilipp Reisner } else { 2232b30ab791SAndreas Gruenbacher update_peer_seq(device, peer_seq); 22330500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 2234b874d231SPhilipp Reisner } 2235b30ab791SAndreas Gruenbacher list_add(&peer_req->w.list, &device->active_ee); 22360500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 2237b411b363SPhilipp Reisner 2238b30ab791SAndreas Gruenbacher if (device->state.conn == C_SYNC_TARGET) 2239b30ab791SAndreas Gruenbacher wait_event(device->ee_wait, !overlapping_resync_write(device, peer_req)); 2240b6a370baSPhilipp Reisner 2241a6b32bc3SAndreas Gruenbacher if (first_peer_device(device)->connection->agreed_pro_version < 100) { 224244ed167dSPhilipp Reisner rcu_read_lock(); 2243a6b32bc3SAndreas Gruenbacher switch (rcu_dereference(first_peer_device(device)->connection->net_conf)->wire_protocol) { 2244b411b363SPhilipp Reisner case DRBD_PROT_C: 2245303d1448SPhilipp Reisner dp_flags |= DP_SEND_WRITE_ACK; 2246303d1448SPhilipp Reisner break; 2247303d1448SPhilipp Reisner case DRBD_PROT_B: 2248303d1448SPhilipp Reisner dp_flags |= DP_SEND_RECEIVE_ACK; 2249303d1448SPhilipp Reisner break; 2250303d1448SPhilipp Reisner } 225144ed167dSPhilipp Reisner rcu_read_unlock(); 2252303d1448SPhilipp Reisner } 2253303d1448SPhilipp Reisner 2254303d1448SPhilipp Reisner if (dp_flags & DP_SEND_WRITE_ACK) { 2255303d1448SPhilipp Reisner peer_req->flags |= EE_SEND_WRITE_ACK; 2256b30ab791SAndreas Gruenbacher inc_unacked(device); 2257b411b363SPhilipp Reisner /* corresponding dec_unacked() in e_end_block() 2258b411b363SPhilipp Reisner * respective _drbd_clear_done_ee */ 2259303d1448SPhilipp Reisner } 2260303d1448SPhilipp Reisner 2261303d1448SPhilipp Reisner if (dp_flags & DP_SEND_RECEIVE_ACK) { 2262b411b363SPhilipp Reisner /* I really don't like it that the receiver thread 2263b411b363SPhilipp Reisner * sends on the msock, but anyways */ 2264b30ab791SAndreas Gruenbacher drbd_send_ack(device, P_RECV_ACK, peer_req); 2265b411b363SPhilipp Reisner } 2266b411b363SPhilipp Reisner 2267b30ab791SAndreas Gruenbacher if (device->state.pdsk < D_INCONSISTENT) { 2268b411b363SPhilipp Reisner /* In case we have the only disk of the cluster, */ 2269b30ab791SAndreas Gruenbacher drbd_set_out_of_sync(device, peer_req->i.sector, peer_req->i.size); 2270db830c46SAndreas Gruenbacher peer_req->flags |= EE_CALL_AL_COMPLETE_IO; 2271db830c46SAndreas Gruenbacher peer_req->flags &= ~EE_MAY_SET_IN_SYNC; 2272b30ab791SAndreas Gruenbacher drbd_al_begin_io(device, &peer_req->i, true); 2273b411b363SPhilipp Reisner } 2274b411b363SPhilipp Reisner 2275b30ab791SAndreas Gruenbacher err = drbd_submit_peer_request(device, peer_req, rw, DRBD_FAULT_DT_WR); 227682bc0194SAndreas Gruenbacher if (!err) 227782bc0194SAndreas Gruenbacher return 0; 2278b411b363SPhilipp Reisner 227910f6d992SLars Ellenberg /* don't care for the reason here */ 2280d0180171SAndreas Gruenbacher drbd_err(device, "submit failed, triggering re-connect\n"); 22810500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 2282db830c46SAndreas Gruenbacher list_del(&peer_req->w.list); 2283b30ab791SAndreas Gruenbacher drbd_remove_epoch_entry_interval(device, peer_req); 22840500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 2285db830c46SAndreas Gruenbacher if (peer_req->flags & EE_CALL_AL_COMPLETE_IO) 2286b30ab791SAndreas Gruenbacher drbd_al_complete_io(device, &peer_req->i); 228722cc37a9SLars Ellenberg 2288b411b363SPhilipp Reisner out_interrupted: 2289bde89a9eSAndreas Gruenbacher drbd_may_finish_epoch(connection, peer_req->epoch, EV_PUT + EV_CLEANUP); 2290b30ab791SAndreas Gruenbacher put_ldev(device); 2291b30ab791SAndreas Gruenbacher drbd_free_peer_req(device, peer_req); 229282bc0194SAndreas Gruenbacher return err; 2293b411b363SPhilipp Reisner } 2294b411b363SPhilipp Reisner 22950f0601f4SLars Ellenberg /* We may throttle resync, if the lower device seems to be busy, 22960f0601f4SLars Ellenberg * and current sync rate is above c_min_rate. 22970f0601f4SLars Ellenberg * 22980f0601f4SLars Ellenberg * To decide whether or not the lower device is busy, we use a scheme similar 22990f0601f4SLars Ellenberg * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" 23000f0601f4SLars Ellenberg * (more than 64 sectors) of activity we cannot account for with our own resync 23010f0601f4SLars Ellenberg * activity, it obviously is "busy". 23020f0601f4SLars Ellenberg * 23030f0601f4SLars Ellenberg * The current sync rate used here uses only the most recent two step marks, 23040f0601f4SLars Ellenberg * to have a short time average so we can react faster. 23050f0601f4SLars Ellenberg */ 2306b30ab791SAndreas Gruenbacher int drbd_rs_should_slow_down(struct drbd_device *device, sector_t sector) 23070f0601f4SLars Ellenberg { 2308b30ab791SAndreas Gruenbacher struct gendisk *disk = device->ldev->backing_bdev->bd_contains->bd_disk; 23090f0601f4SLars Ellenberg unsigned long db, dt, dbdt; 2310e3555d85SPhilipp Reisner struct lc_element *tmp; 23110f0601f4SLars Ellenberg int curr_events; 23120f0601f4SLars Ellenberg int throttle = 0; 2313daeda1ccSPhilipp Reisner unsigned int c_min_rate; 2314daeda1ccSPhilipp Reisner 2315daeda1ccSPhilipp Reisner rcu_read_lock(); 2316b30ab791SAndreas Gruenbacher c_min_rate = rcu_dereference(device->ldev->disk_conf)->c_min_rate; 2317daeda1ccSPhilipp Reisner rcu_read_unlock(); 23180f0601f4SLars Ellenberg 23190f0601f4SLars Ellenberg /* feature disabled? */ 2320daeda1ccSPhilipp Reisner if (c_min_rate == 0) 23210f0601f4SLars Ellenberg return 0; 23220f0601f4SLars Ellenberg 2323b30ab791SAndreas Gruenbacher spin_lock_irq(&device->al_lock); 2324b30ab791SAndreas Gruenbacher tmp = lc_find(device->resync, BM_SECT_TO_EXT(sector)); 2325e3555d85SPhilipp Reisner if (tmp) { 2326e3555d85SPhilipp Reisner struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); 2327e3555d85SPhilipp Reisner if (test_bit(BME_PRIORITY, &bm_ext->flags)) { 2328b30ab791SAndreas Gruenbacher spin_unlock_irq(&device->al_lock); 2329e3555d85SPhilipp Reisner return 0; 2330e3555d85SPhilipp Reisner } 2331e3555d85SPhilipp Reisner /* Do not slow down if app IO is already waiting for this extent */ 2332e3555d85SPhilipp Reisner } 2333b30ab791SAndreas Gruenbacher spin_unlock_irq(&device->al_lock); 2334e3555d85SPhilipp Reisner 23350f0601f4SLars Ellenberg curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + 23360f0601f4SLars Ellenberg (int)part_stat_read(&disk->part0, sectors[1]) - 2337b30ab791SAndreas Gruenbacher atomic_read(&device->rs_sect_ev); 2338e3555d85SPhilipp Reisner 2339b30ab791SAndreas Gruenbacher if (!device->rs_last_events || curr_events - device->rs_last_events > 64) { 23400f0601f4SLars Ellenberg unsigned long rs_left; 23410f0601f4SLars Ellenberg int i; 23420f0601f4SLars Ellenberg 2343b30ab791SAndreas Gruenbacher device->rs_last_events = curr_events; 23440f0601f4SLars Ellenberg 23450f0601f4SLars Ellenberg /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, 23460f0601f4SLars Ellenberg * approx. */ 2347b30ab791SAndreas Gruenbacher i = (device->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; 23482649f080SLars Ellenberg 2349b30ab791SAndreas Gruenbacher if (device->state.conn == C_VERIFY_S || device->state.conn == C_VERIFY_T) 2350b30ab791SAndreas Gruenbacher rs_left = device->ov_left; 23512649f080SLars Ellenberg else 2352b30ab791SAndreas Gruenbacher rs_left = drbd_bm_total_weight(device) - device->rs_failed; 23530f0601f4SLars Ellenberg 2354b30ab791SAndreas Gruenbacher dt = ((long)jiffies - (long)device->rs_mark_time[i]) / HZ; 23550f0601f4SLars Ellenberg if (!dt) 23560f0601f4SLars Ellenberg dt++; 2357b30ab791SAndreas Gruenbacher db = device->rs_mark_left[i] - rs_left; 23580f0601f4SLars Ellenberg dbdt = Bit2KB(db/dt); 23590f0601f4SLars Ellenberg 2360daeda1ccSPhilipp Reisner if (dbdt > c_min_rate) 23610f0601f4SLars Ellenberg throttle = 1; 23620f0601f4SLars Ellenberg } 23630f0601f4SLars Ellenberg return throttle; 23640f0601f4SLars Ellenberg } 23650f0601f4SLars Ellenberg 23660f0601f4SLars Ellenberg 2367bde89a9eSAndreas Gruenbacher static int receive_DataRequest(struct drbd_connection *connection, struct packet_info *pi) 2368b411b363SPhilipp Reisner { 2369b30ab791SAndreas Gruenbacher struct drbd_device *device; 2370b411b363SPhilipp Reisner sector_t sector; 23714a76b161SAndreas Gruenbacher sector_t capacity; 2372db830c46SAndreas Gruenbacher struct drbd_peer_request *peer_req; 2373b411b363SPhilipp Reisner struct digest_info *di = NULL; 2374b18b37beSPhilipp Reisner int size, verb; 2375b411b363SPhilipp Reisner unsigned int fault_type; 2376e658983aSAndreas Gruenbacher struct p_block_req *p = pi->data; 23774a76b161SAndreas Gruenbacher 2378bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 2379b30ab791SAndreas Gruenbacher if (!device) 23804a76b161SAndreas Gruenbacher return -EIO; 2381b30ab791SAndreas Gruenbacher capacity = drbd_get_capacity(device->this_bdev); 2382b411b363SPhilipp Reisner 2383b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 2384b411b363SPhilipp Reisner size = be32_to_cpu(p->blksize); 2385b411b363SPhilipp Reisner 2386c670a398SAndreas Gruenbacher if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { 2387d0180171SAndreas Gruenbacher drbd_err(device, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, 2388b411b363SPhilipp Reisner (unsigned long long)sector, size); 238982bc0194SAndreas Gruenbacher return -EINVAL; 2390b411b363SPhilipp Reisner } 2391b411b363SPhilipp Reisner if (sector + (size>>9) > capacity) { 2392d0180171SAndreas Gruenbacher drbd_err(device, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, 2393b411b363SPhilipp Reisner (unsigned long long)sector, size); 239482bc0194SAndreas Gruenbacher return -EINVAL; 2395b411b363SPhilipp Reisner } 2396b411b363SPhilipp Reisner 2397b30ab791SAndreas Gruenbacher if (!get_ldev_if_state(device, D_UP_TO_DATE)) { 2398b18b37beSPhilipp Reisner verb = 1; 2399e2857216SAndreas Gruenbacher switch (pi->cmd) { 2400b18b37beSPhilipp Reisner case P_DATA_REQUEST: 2401b30ab791SAndreas Gruenbacher drbd_send_ack_rp(device, P_NEG_DREPLY, p); 2402b18b37beSPhilipp Reisner break; 2403b18b37beSPhilipp Reisner case P_RS_DATA_REQUEST: 2404b18b37beSPhilipp Reisner case P_CSUM_RS_REQUEST: 2405b18b37beSPhilipp Reisner case P_OV_REQUEST: 2406b30ab791SAndreas Gruenbacher drbd_send_ack_rp(device, P_NEG_RS_DREPLY , p); 2407b18b37beSPhilipp Reisner break; 2408b18b37beSPhilipp Reisner case P_OV_REPLY: 2409b18b37beSPhilipp Reisner verb = 0; 2410b30ab791SAndreas Gruenbacher dec_rs_pending(device); 2411b30ab791SAndreas Gruenbacher drbd_send_ack_ex(device, P_OV_RESULT, sector, size, ID_IN_SYNC); 2412b18b37beSPhilipp Reisner break; 2413b18b37beSPhilipp Reisner default: 241449ba9b1bSAndreas Gruenbacher BUG(); 2415b18b37beSPhilipp Reisner } 2416b18b37beSPhilipp Reisner if (verb && __ratelimit(&drbd_ratelimit_state)) 2417d0180171SAndreas Gruenbacher drbd_err(device, "Can not satisfy peer's read request, " 2418b411b363SPhilipp Reisner "no local data.\n"); 2419b18b37beSPhilipp Reisner 2420a821cc4aSLars Ellenberg /* drain possibly payload */ 2421b30ab791SAndreas Gruenbacher return drbd_drain_block(device, pi->size); 2422b411b363SPhilipp Reisner } 2423b411b363SPhilipp Reisner 2424b411b363SPhilipp Reisner /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD 2425b411b363SPhilipp Reisner * "criss-cross" setup, that might cause write-out on some other DRBD, 2426b411b363SPhilipp Reisner * which in turn might block on the other node at this very place. */ 2427b30ab791SAndreas Gruenbacher peer_req = drbd_alloc_peer_req(device, p->block_id, sector, size, GFP_NOIO); 2428db830c46SAndreas Gruenbacher if (!peer_req) { 2429b30ab791SAndreas Gruenbacher put_ldev(device); 243082bc0194SAndreas Gruenbacher return -ENOMEM; 2431b411b363SPhilipp Reisner } 2432b411b363SPhilipp Reisner 2433e2857216SAndreas Gruenbacher switch (pi->cmd) { 2434b411b363SPhilipp Reisner case P_DATA_REQUEST: 2435db830c46SAndreas Gruenbacher peer_req->w.cb = w_e_end_data_req; 2436b411b363SPhilipp Reisner fault_type = DRBD_FAULT_DT_RD; 243780a40e43SLars Ellenberg /* application IO, don't drbd_rs_begin_io */ 243880a40e43SLars Ellenberg goto submit; 243980a40e43SLars Ellenberg 2440b411b363SPhilipp Reisner case P_RS_DATA_REQUEST: 2441db830c46SAndreas Gruenbacher peer_req->w.cb = w_e_end_rsdata_req; 2442b411b363SPhilipp Reisner fault_type = DRBD_FAULT_RS_RD; 24435f9915bbSLars Ellenberg /* used in the sector offset progress display */ 2444b30ab791SAndreas Gruenbacher device->bm_resync_fo = BM_SECT_TO_BIT(sector); 2445b411b363SPhilipp Reisner break; 2446b411b363SPhilipp Reisner 2447b411b363SPhilipp Reisner case P_OV_REPLY: 2448b411b363SPhilipp Reisner case P_CSUM_RS_REQUEST: 2449b411b363SPhilipp Reisner fault_type = DRBD_FAULT_RS_RD; 2450e2857216SAndreas Gruenbacher di = kmalloc(sizeof(*di) + pi->size, GFP_NOIO); 2451b411b363SPhilipp Reisner if (!di) 2452b411b363SPhilipp Reisner goto out_free_e; 2453b411b363SPhilipp Reisner 2454e2857216SAndreas Gruenbacher di->digest_size = pi->size; 2455b411b363SPhilipp Reisner di->digest = (((char *)di)+sizeof(struct digest_info)); 2456b411b363SPhilipp Reisner 2457db830c46SAndreas Gruenbacher peer_req->digest = di; 2458db830c46SAndreas Gruenbacher peer_req->flags |= EE_HAS_DIGEST; 2459c36c3cedSLars Ellenberg 2460a6b32bc3SAndreas Gruenbacher if (drbd_recv_all(first_peer_device(device)->connection, di->digest, pi->size)) 2461b411b363SPhilipp Reisner goto out_free_e; 2462b411b363SPhilipp Reisner 2463e2857216SAndreas Gruenbacher if (pi->cmd == P_CSUM_RS_REQUEST) { 24640b0ba1efSAndreas Gruenbacher D_ASSERT(device, first_peer_device(device)->connection->agreed_pro_version >= 89); 2465db830c46SAndreas Gruenbacher peer_req->w.cb = w_e_end_csum_rs_req; 24665f9915bbSLars Ellenberg /* used in the sector offset progress display */ 2467b30ab791SAndreas Gruenbacher device->bm_resync_fo = BM_SECT_TO_BIT(sector); 2468e2857216SAndreas Gruenbacher } else if (pi->cmd == P_OV_REPLY) { 24692649f080SLars Ellenberg /* track progress, we may need to throttle */ 2470b30ab791SAndreas Gruenbacher atomic_add(size >> 9, &device->rs_sect_in); 2471db830c46SAndreas Gruenbacher peer_req->w.cb = w_e_end_ov_reply; 2472b30ab791SAndreas Gruenbacher dec_rs_pending(device); 24730f0601f4SLars Ellenberg /* drbd_rs_begin_io done when we sent this request, 24740f0601f4SLars Ellenberg * but accounting still needs to be done. */ 24750f0601f4SLars Ellenberg goto submit_for_resync; 2476b411b363SPhilipp Reisner } 2477b411b363SPhilipp Reisner break; 2478b411b363SPhilipp Reisner 2479b411b363SPhilipp Reisner case P_OV_REQUEST: 2480b30ab791SAndreas Gruenbacher if (device->ov_start_sector == ~(sector_t)0 && 2481a6b32bc3SAndreas Gruenbacher first_peer_device(device)->connection->agreed_pro_version >= 90) { 2482de228bbaSLars Ellenberg unsigned long now = jiffies; 2483de228bbaSLars Ellenberg int i; 2484b30ab791SAndreas Gruenbacher device->ov_start_sector = sector; 2485b30ab791SAndreas Gruenbacher device->ov_position = sector; 2486b30ab791SAndreas Gruenbacher device->ov_left = drbd_bm_bits(device) - BM_SECT_TO_BIT(sector); 2487b30ab791SAndreas Gruenbacher device->rs_total = device->ov_left; 2488de228bbaSLars Ellenberg for (i = 0; i < DRBD_SYNC_MARKS; i++) { 2489b30ab791SAndreas Gruenbacher device->rs_mark_left[i] = device->ov_left; 2490b30ab791SAndreas Gruenbacher device->rs_mark_time[i] = now; 2491de228bbaSLars Ellenberg } 2492d0180171SAndreas Gruenbacher drbd_info(device, "Online Verify start sector: %llu\n", 2493b411b363SPhilipp Reisner (unsigned long long)sector); 2494b411b363SPhilipp Reisner } 2495db830c46SAndreas Gruenbacher peer_req->w.cb = w_e_end_ov_req; 2496b411b363SPhilipp Reisner fault_type = DRBD_FAULT_RS_RD; 2497b411b363SPhilipp Reisner break; 2498b411b363SPhilipp Reisner 2499b411b363SPhilipp Reisner default: 250049ba9b1bSAndreas Gruenbacher BUG(); 2501b411b363SPhilipp Reisner } 2502b411b363SPhilipp Reisner 25030f0601f4SLars Ellenberg /* Throttle, drbd_rs_begin_io and submit should become asynchronous 25040f0601f4SLars Ellenberg * wrt the receiver, but it is not as straightforward as it may seem. 25050f0601f4SLars Ellenberg * Various places in the resync start and stop logic assume resync 25060f0601f4SLars Ellenberg * requests are processed in order, requeuing this on the worker thread 25070f0601f4SLars Ellenberg * introduces a bunch of new code for synchronization between threads. 25080f0601f4SLars Ellenberg * 25090f0601f4SLars Ellenberg * Unlimited throttling before drbd_rs_begin_io may stall the resync 25100f0601f4SLars Ellenberg * "forever", throttling after drbd_rs_begin_io will lock that extent 25110f0601f4SLars Ellenberg * for application writes for the same time. For now, just throttle 25120f0601f4SLars Ellenberg * here, where the rest of the code expects the receiver to sleep for 25130f0601f4SLars Ellenberg * a while, anyways. 25140f0601f4SLars Ellenberg */ 2515b411b363SPhilipp Reisner 25160f0601f4SLars Ellenberg /* Throttle before drbd_rs_begin_io, as that locks out application IO; 25170f0601f4SLars Ellenberg * this defers syncer requests for some time, before letting at least 25180f0601f4SLars Ellenberg * on request through. The resync controller on the receiving side 25190f0601f4SLars Ellenberg * will adapt to the incoming rate accordingly. 25200f0601f4SLars Ellenberg * 25210f0601f4SLars Ellenberg * We cannot throttle here if remote is Primary/SyncTarget: 25220f0601f4SLars Ellenberg * we would also throttle its application reads. 25230f0601f4SLars Ellenberg * In that case, throttling is done on the SyncTarget only. 25240f0601f4SLars Ellenberg */ 2525b30ab791SAndreas Gruenbacher if (device->state.peer != R_PRIMARY && drbd_rs_should_slow_down(device, sector)) 2526e3555d85SPhilipp Reisner schedule_timeout_uninterruptible(HZ/10); 2527b30ab791SAndreas Gruenbacher if (drbd_rs_begin_io(device, sector)) 252880a40e43SLars Ellenberg goto out_free_e; 2529b411b363SPhilipp Reisner 25300f0601f4SLars Ellenberg submit_for_resync: 2531b30ab791SAndreas Gruenbacher atomic_add(size >> 9, &device->rs_sect_ev); 25320f0601f4SLars Ellenberg 253380a40e43SLars Ellenberg submit: 2534b30ab791SAndreas Gruenbacher inc_unacked(device); 25350500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 2536b30ab791SAndreas Gruenbacher list_add_tail(&peer_req->w.list, &device->read_ee); 25370500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 2538b411b363SPhilipp Reisner 2539b30ab791SAndreas Gruenbacher if (drbd_submit_peer_request(device, peer_req, READ, fault_type) == 0) 254082bc0194SAndreas Gruenbacher return 0; 2541b411b363SPhilipp Reisner 254210f6d992SLars Ellenberg /* don't care for the reason here */ 2543d0180171SAndreas Gruenbacher drbd_err(device, "submit failed, triggering re-connect\n"); 25440500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 2545db830c46SAndreas Gruenbacher list_del(&peer_req->w.list); 25460500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 254722cc37a9SLars Ellenberg /* no drbd_rs_complete_io(), we are dropping the connection anyways */ 254822cc37a9SLars Ellenberg 2549b411b363SPhilipp Reisner out_free_e: 2550b30ab791SAndreas Gruenbacher put_ldev(device); 2551b30ab791SAndreas Gruenbacher drbd_free_peer_req(device, peer_req); 255282bc0194SAndreas Gruenbacher return -EIO; 2553b411b363SPhilipp Reisner } 2554b411b363SPhilipp Reisner 2555b30ab791SAndreas Gruenbacher static int drbd_asb_recover_0p(struct drbd_device *device) __must_hold(local) 2556b411b363SPhilipp Reisner { 2557b411b363SPhilipp Reisner int self, peer, rv = -100; 2558b411b363SPhilipp Reisner unsigned long ch_self, ch_peer; 255944ed167dSPhilipp Reisner enum drbd_after_sb_p after_sb_0p; 2560b411b363SPhilipp Reisner 2561b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[UI_BITMAP] & 1; 2562b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_BITMAP] & 1; 2563b411b363SPhilipp Reisner 2564b30ab791SAndreas Gruenbacher ch_peer = device->p_uuid[UI_SIZE]; 2565b30ab791SAndreas Gruenbacher ch_self = device->comm_bm_set; 2566b411b363SPhilipp Reisner 256744ed167dSPhilipp Reisner rcu_read_lock(); 2568a6b32bc3SAndreas Gruenbacher after_sb_0p = rcu_dereference(first_peer_device(device)->connection->net_conf)->after_sb_0p; 256944ed167dSPhilipp Reisner rcu_read_unlock(); 257044ed167dSPhilipp Reisner switch (after_sb_0p) { 2571b411b363SPhilipp Reisner case ASB_CONSENSUS: 2572b411b363SPhilipp Reisner case ASB_DISCARD_SECONDARY: 2573b411b363SPhilipp Reisner case ASB_CALL_HELPER: 257444ed167dSPhilipp Reisner case ASB_VIOLENTLY: 2575d0180171SAndreas Gruenbacher drbd_err(device, "Configuration error.\n"); 2576b411b363SPhilipp Reisner break; 2577b411b363SPhilipp Reisner case ASB_DISCONNECT: 2578b411b363SPhilipp Reisner break; 2579b411b363SPhilipp Reisner case ASB_DISCARD_YOUNGER_PRI: 2580b411b363SPhilipp Reisner if (self == 0 && peer == 1) { 2581b411b363SPhilipp Reisner rv = -1; 2582b411b363SPhilipp Reisner break; 2583b411b363SPhilipp Reisner } 2584b411b363SPhilipp Reisner if (self == 1 && peer == 0) { 2585b411b363SPhilipp Reisner rv = 1; 2586b411b363SPhilipp Reisner break; 2587b411b363SPhilipp Reisner } 2588b411b363SPhilipp Reisner /* Else fall through to one of the other strategies... */ 2589b411b363SPhilipp Reisner case ASB_DISCARD_OLDER_PRI: 2590b411b363SPhilipp Reisner if (self == 0 && peer == 1) { 2591b411b363SPhilipp Reisner rv = 1; 2592b411b363SPhilipp Reisner break; 2593b411b363SPhilipp Reisner } 2594b411b363SPhilipp Reisner if (self == 1 && peer == 0) { 2595b411b363SPhilipp Reisner rv = -1; 2596b411b363SPhilipp Reisner break; 2597b411b363SPhilipp Reisner } 2598b411b363SPhilipp Reisner /* Else fall through to one of the other strategies... */ 2599d0180171SAndreas Gruenbacher drbd_warn(device, "Discard younger/older primary did not find a decision\n" 2600b411b363SPhilipp Reisner "Using discard-least-changes instead\n"); 2601b411b363SPhilipp Reisner case ASB_DISCARD_ZERO_CHG: 2602b411b363SPhilipp Reisner if (ch_peer == 0 && ch_self == 0) { 2603a6b32bc3SAndreas Gruenbacher rv = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags) 2604b411b363SPhilipp Reisner ? -1 : 1; 2605b411b363SPhilipp Reisner break; 2606b411b363SPhilipp Reisner } else { 2607b411b363SPhilipp Reisner if (ch_peer == 0) { rv = 1; break; } 2608b411b363SPhilipp Reisner if (ch_self == 0) { rv = -1; break; } 2609b411b363SPhilipp Reisner } 261044ed167dSPhilipp Reisner if (after_sb_0p == ASB_DISCARD_ZERO_CHG) 2611b411b363SPhilipp Reisner break; 2612b411b363SPhilipp Reisner case ASB_DISCARD_LEAST_CHG: 2613b411b363SPhilipp Reisner if (ch_self < ch_peer) 2614b411b363SPhilipp Reisner rv = -1; 2615b411b363SPhilipp Reisner else if (ch_self > ch_peer) 2616b411b363SPhilipp Reisner rv = 1; 2617b411b363SPhilipp Reisner else /* ( ch_self == ch_peer ) */ 2618b411b363SPhilipp Reisner /* Well, then use something else. */ 2619a6b32bc3SAndreas Gruenbacher rv = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags) 2620b411b363SPhilipp Reisner ? -1 : 1; 2621b411b363SPhilipp Reisner break; 2622b411b363SPhilipp Reisner case ASB_DISCARD_LOCAL: 2623b411b363SPhilipp Reisner rv = -1; 2624b411b363SPhilipp Reisner break; 2625b411b363SPhilipp Reisner case ASB_DISCARD_REMOTE: 2626b411b363SPhilipp Reisner rv = 1; 2627b411b363SPhilipp Reisner } 2628b411b363SPhilipp Reisner 2629b411b363SPhilipp Reisner return rv; 2630b411b363SPhilipp Reisner } 2631b411b363SPhilipp Reisner 2632b30ab791SAndreas Gruenbacher static int drbd_asb_recover_1p(struct drbd_device *device) __must_hold(local) 2633b411b363SPhilipp Reisner { 26346184ea21SAndreas Gruenbacher int hg, rv = -100; 263544ed167dSPhilipp Reisner enum drbd_after_sb_p after_sb_1p; 2636b411b363SPhilipp Reisner 263744ed167dSPhilipp Reisner rcu_read_lock(); 2638a6b32bc3SAndreas Gruenbacher after_sb_1p = rcu_dereference(first_peer_device(device)->connection->net_conf)->after_sb_1p; 263944ed167dSPhilipp Reisner rcu_read_unlock(); 264044ed167dSPhilipp Reisner switch (after_sb_1p) { 2641b411b363SPhilipp Reisner case ASB_DISCARD_YOUNGER_PRI: 2642b411b363SPhilipp Reisner case ASB_DISCARD_OLDER_PRI: 2643b411b363SPhilipp Reisner case ASB_DISCARD_LEAST_CHG: 2644b411b363SPhilipp Reisner case ASB_DISCARD_LOCAL: 2645b411b363SPhilipp Reisner case ASB_DISCARD_REMOTE: 264644ed167dSPhilipp Reisner case ASB_DISCARD_ZERO_CHG: 2647d0180171SAndreas Gruenbacher drbd_err(device, "Configuration error.\n"); 2648b411b363SPhilipp Reisner break; 2649b411b363SPhilipp Reisner case ASB_DISCONNECT: 2650b411b363SPhilipp Reisner break; 2651b411b363SPhilipp Reisner case ASB_CONSENSUS: 2652b30ab791SAndreas Gruenbacher hg = drbd_asb_recover_0p(device); 2653b30ab791SAndreas Gruenbacher if (hg == -1 && device->state.role == R_SECONDARY) 2654b411b363SPhilipp Reisner rv = hg; 2655b30ab791SAndreas Gruenbacher if (hg == 1 && device->state.role == R_PRIMARY) 2656b411b363SPhilipp Reisner rv = hg; 2657b411b363SPhilipp Reisner break; 2658b411b363SPhilipp Reisner case ASB_VIOLENTLY: 2659b30ab791SAndreas Gruenbacher rv = drbd_asb_recover_0p(device); 2660b411b363SPhilipp Reisner break; 2661b411b363SPhilipp Reisner case ASB_DISCARD_SECONDARY: 2662b30ab791SAndreas Gruenbacher return device->state.role == R_PRIMARY ? 1 : -1; 2663b411b363SPhilipp Reisner case ASB_CALL_HELPER: 2664b30ab791SAndreas Gruenbacher hg = drbd_asb_recover_0p(device); 2665b30ab791SAndreas Gruenbacher if (hg == -1 && device->state.role == R_PRIMARY) { 2666bb437946SAndreas Gruenbacher enum drbd_state_rv rv2; 2667bb437946SAndreas Gruenbacher 2668b411b363SPhilipp Reisner /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, 2669b411b363SPhilipp Reisner * we might be here in C_WF_REPORT_PARAMS which is transient. 2670b411b363SPhilipp Reisner * we do not need to wait for the after state change work either. */ 2671b30ab791SAndreas Gruenbacher rv2 = drbd_change_state(device, CS_VERBOSE, NS(role, R_SECONDARY)); 2672bb437946SAndreas Gruenbacher if (rv2 != SS_SUCCESS) { 2673b30ab791SAndreas Gruenbacher drbd_khelper(device, "pri-lost-after-sb"); 2674b411b363SPhilipp Reisner } else { 2675d0180171SAndreas Gruenbacher drbd_warn(device, "Successfully gave up primary role.\n"); 2676b411b363SPhilipp Reisner rv = hg; 2677b411b363SPhilipp Reisner } 2678b411b363SPhilipp Reisner } else 2679b411b363SPhilipp Reisner rv = hg; 2680b411b363SPhilipp Reisner } 2681b411b363SPhilipp Reisner 2682b411b363SPhilipp Reisner return rv; 2683b411b363SPhilipp Reisner } 2684b411b363SPhilipp Reisner 2685b30ab791SAndreas Gruenbacher static int drbd_asb_recover_2p(struct drbd_device *device) __must_hold(local) 2686b411b363SPhilipp Reisner { 26876184ea21SAndreas Gruenbacher int hg, rv = -100; 268844ed167dSPhilipp Reisner enum drbd_after_sb_p after_sb_2p; 2689b411b363SPhilipp Reisner 269044ed167dSPhilipp Reisner rcu_read_lock(); 2691a6b32bc3SAndreas Gruenbacher after_sb_2p = rcu_dereference(first_peer_device(device)->connection->net_conf)->after_sb_2p; 269244ed167dSPhilipp Reisner rcu_read_unlock(); 269344ed167dSPhilipp Reisner switch (after_sb_2p) { 2694b411b363SPhilipp Reisner case ASB_DISCARD_YOUNGER_PRI: 2695b411b363SPhilipp Reisner case ASB_DISCARD_OLDER_PRI: 2696b411b363SPhilipp Reisner case ASB_DISCARD_LEAST_CHG: 2697b411b363SPhilipp Reisner case ASB_DISCARD_LOCAL: 2698b411b363SPhilipp Reisner case ASB_DISCARD_REMOTE: 2699b411b363SPhilipp Reisner case ASB_CONSENSUS: 2700b411b363SPhilipp Reisner case ASB_DISCARD_SECONDARY: 270144ed167dSPhilipp Reisner case ASB_DISCARD_ZERO_CHG: 2702d0180171SAndreas Gruenbacher drbd_err(device, "Configuration error.\n"); 2703b411b363SPhilipp Reisner break; 2704b411b363SPhilipp Reisner case ASB_VIOLENTLY: 2705b30ab791SAndreas Gruenbacher rv = drbd_asb_recover_0p(device); 2706b411b363SPhilipp Reisner break; 2707b411b363SPhilipp Reisner case ASB_DISCONNECT: 2708b411b363SPhilipp Reisner break; 2709b411b363SPhilipp Reisner case ASB_CALL_HELPER: 2710b30ab791SAndreas Gruenbacher hg = drbd_asb_recover_0p(device); 2711b411b363SPhilipp Reisner if (hg == -1) { 2712bb437946SAndreas Gruenbacher enum drbd_state_rv rv2; 2713bb437946SAndreas Gruenbacher 2714b411b363SPhilipp Reisner /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, 2715b411b363SPhilipp Reisner * we might be here in C_WF_REPORT_PARAMS which is transient. 2716b411b363SPhilipp Reisner * we do not need to wait for the after state change work either. */ 2717b30ab791SAndreas Gruenbacher rv2 = drbd_change_state(device, CS_VERBOSE, NS(role, R_SECONDARY)); 2718bb437946SAndreas Gruenbacher if (rv2 != SS_SUCCESS) { 2719b30ab791SAndreas Gruenbacher drbd_khelper(device, "pri-lost-after-sb"); 2720b411b363SPhilipp Reisner } else { 2721d0180171SAndreas Gruenbacher drbd_warn(device, "Successfully gave up primary role.\n"); 2722b411b363SPhilipp Reisner rv = hg; 2723b411b363SPhilipp Reisner } 2724b411b363SPhilipp Reisner } else 2725b411b363SPhilipp Reisner rv = hg; 2726b411b363SPhilipp Reisner } 2727b411b363SPhilipp Reisner 2728b411b363SPhilipp Reisner return rv; 2729b411b363SPhilipp Reisner } 2730b411b363SPhilipp Reisner 2731b30ab791SAndreas Gruenbacher static void drbd_uuid_dump(struct drbd_device *device, char *text, u64 *uuid, 2732b411b363SPhilipp Reisner u64 bits, u64 flags) 2733b411b363SPhilipp Reisner { 2734b411b363SPhilipp Reisner if (!uuid) { 2735d0180171SAndreas Gruenbacher drbd_info(device, "%s uuid info vanished while I was looking!\n", text); 2736b411b363SPhilipp Reisner return; 2737b411b363SPhilipp Reisner } 2738d0180171SAndreas Gruenbacher drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", 2739b411b363SPhilipp Reisner text, 2740b411b363SPhilipp Reisner (unsigned long long)uuid[UI_CURRENT], 2741b411b363SPhilipp Reisner (unsigned long long)uuid[UI_BITMAP], 2742b411b363SPhilipp Reisner (unsigned long long)uuid[UI_HISTORY_START], 2743b411b363SPhilipp Reisner (unsigned long long)uuid[UI_HISTORY_END], 2744b411b363SPhilipp Reisner (unsigned long long)bits, 2745b411b363SPhilipp Reisner (unsigned long long)flags); 2746b411b363SPhilipp Reisner } 2747b411b363SPhilipp Reisner 2748b411b363SPhilipp Reisner /* 2749b411b363SPhilipp Reisner 100 after split brain try auto recover 2750b411b363SPhilipp Reisner 2 C_SYNC_SOURCE set BitMap 2751b411b363SPhilipp Reisner 1 C_SYNC_SOURCE use BitMap 2752b411b363SPhilipp Reisner 0 no Sync 2753b411b363SPhilipp Reisner -1 C_SYNC_TARGET use BitMap 2754b411b363SPhilipp Reisner -2 C_SYNC_TARGET set BitMap 2755b411b363SPhilipp Reisner -100 after split brain, disconnect 2756b411b363SPhilipp Reisner -1000 unrelated data 27574a23f264SPhilipp Reisner -1091 requires proto 91 27584a23f264SPhilipp Reisner -1096 requires proto 96 2759b411b363SPhilipp Reisner */ 2760b30ab791SAndreas Gruenbacher static int drbd_uuid_compare(struct drbd_device *device, int *rule_nr) __must_hold(local) 2761b411b363SPhilipp Reisner { 2762b411b363SPhilipp Reisner u64 self, peer; 2763b411b363SPhilipp Reisner int i, j; 2764b411b363SPhilipp Reisner 2765b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[UI_CURRENT] & ~((u64)1); 2766b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_CURRENT] & ~((u64)1); 2767b411b363SPhilipp Reisner 2768b411b363SPhilipp Reisner *rule_nr = 10; 2769b411b363SPhilipp Reisner if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) 2770b411b363SPhilipp Reisner return 0; 2771b411b363SPhilipp Reisner 2772b411b363SPhilipp Reisner *rule_nr = 20; 2773b411b363SPhilipp Reisner if ((self == UUID_JUST_CREATED || self == (u64)0) && 2774b411b363SPhilipp Reisner peer != UUID_JUST_CREATED) 2775b411b363SPhilipp Reisner return -2; 2776b411b363SPhilipp Reisner 2777b411b363SPhilipp Reisner *rule_nr = 30; 2778b411b363SPhilipp Reisner if (self != UUID_JUST_CREATED && 2779b411b363SPhilipp Reisner (peer == UUID_JUST_CREATED || peer == (u64)0)) 2780b411b363SPhilipp Reisner return 2; 2781b411b363SPhilipp Reisner 2782b411b363SPhilipp Reisner if (self == peer) { 2783b411b363SPhilipp Reisner int rct, dc; /* roles at crash time */ 2784b411b363SPhilipp Reisner 2785b30ab791SAndreas Gruenbacher if (device->p_uuid[UI_BITMAP] == (u64)0 && device->ldev->md.uuid[UI_BITMAP] != (u64)0) { 2786b411b363SPhilipp Reisner 2787a6b32bc3SAndreas Gruenbacher if (first_peer_device(device)->connection->agreed_pro_version < 91) 27884a23f264SPhilipp Reisner return -1091; 2789b411b363SPhilipp Reisner 2790b30ab791SAndreas Gruenbacher if ((device->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START] & ~((u64)1)) && 2791b30ab791SAndreas Gruenbacher (device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { 2792d0180171SAndreas Gruenbacher drbd_info(device, "was SyncSource, missed the resync finished event, corrected myself:\n"); 2793b30ab791SAndreas Gruenbacher drbd_uuid_move_history(device); 2794b30ab791SAndreas Gruenbacher device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP]; 2795b30ab791SAndreas Gruenbacher device->ldev->md.uuid[UI_BITMAP] = 0; 2796b411b363SPhilipp Reisner 2797b30ab791SAndreas Gruenbacher drbd_uuid_dump(device, "self", device->ldev->md.uuid, 2798b30ab791SAndreas Gruenbacher device->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(device) : 0, 0); 2799b411b363SPhilipp Reisner *rule_nr = 34; 2800b411b363SPhilipp Reisner } else { 2801d0180171SAndreas Gruenbacher drbd_info(device, "was SyncSource (peer failed to write sync_uuid)\n"); 2802b411b363SPhilipp Reisner *rule_nr = 36; 2803b411b363SPhilipp Reisner } 2804b411b363SPhilipp Reisner 2805b411b363SPhilipp Reisner return 1; 2806b411b363SPhilipp Reisner } 2807b411b363SPhilipp Reisner 2808b30ab791SAndreas Gruenbacher if (device->ldev->md.uuid[UI_BITMAP] == (u64)0 && device->p_uuid[UI_BITMAP] != (u64)0) { 2809b411b363SPhilipp Reisner 2810a6b32bc3SAndreas Gruenbacher if (first_peer_device(device)->connection->agreed_pro_version < 91) 28114a23f264SPhilipp Reisner return -1091; 2812b411b363SPhilipp Reisner 2813b30ab791SAndreas Gruenbacher if ((device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (device->p_uuid[UI_BITMAP] & ~((u64)1)) && 2814b30ab791SAndreas Gruenbacher (device->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (device->p_uuid[UI_HISTORY_START] & ~((u64)1))) { 2815d0180171SAndreas Gruenbacher drbd_info(device, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); 2816b411b363SPhilipp Reisner 2817b30ab791SAndreas Gruenbacher device->p_uuid[UI_HISTORY_START + 1] = device->p_uuid[UI_HISTORY_START]; 2818b30ab791SAndreas Gruenbacher device->p_uuid[UI_HISTORY_START] = device->p_uuid[UI_BITMAP]; 2819b30ab791SAndreas Gruenbacher device->p_uuid[UI_BITMAP] = 0UL; 2820b411b363SPhilipp Reisner 2821b30ab791SAndreas Gruenbacher drbd_uuid_dump(device, "peer", device->p_uuid, device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); 2822b411b363SPhilipp Reisner *rule_nr = 35; 2823b411b363SPhilipp Reisner } else { 2824d0180171SAndreas Gruenbacher drbd_info(device, "was SyncTarget (failed to write sync_uuid)\n"); 2825b411b363SPhilipp Reisner *rule_nr = 37; 2826b411b363SPhilipp Reisner } 2827b411b363SPhilipp Reisner 2828b411b363SPhilipp Reisner return -1; 2829b411b363SPhilipp Reisner } 2830b411b363SPhilipp Reisner 2831b411b363SPhilipp Reisner /* Common power [off|failure] */ 2832b30ab791SAndreas Gruenbacher rct = (test_bit(CRASHED_PRIMARY, &device->flags) ? 1 : 0) + 2833b30ab791SAndreas Gruenbacher (device->p_uuid[UI_FLAGS] & 2); 2834b411b363SPhilipp Reisner /* lowest bit is set when we were primary, 2835b411b363SPhilipp Reisner * next bit (weight 2) is set when peer was primary */ 2836b411b363SPhilipp Reisner *rule_nr = 40; 2837b411b363SPhilipp Reisner 2838b411b363SPhilipp Reisner switch (rct) { 2839b411b363SPhilipp Reisner case 0: /* !self_pri && !peer_pri */ return 0; 2840b411b363SPhilipp Reisner case 1: /* self_pri && !peer_pri */ return 1; 2841b411b363SPhilipp Reisner case 2: /* !self_pri && peer_pri */ return -1; 2842b411b363SPhilipp Reisner case 3: /* self_pri && peer_pri */ 2843a6b32bc3SAndreas Gruenbacher dc = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags); 2844b411b363SPhilipp Reisner return dc ? -1 : 1; 2845b411b363SPhilipp Reisner } 2846b411b363SPhilipp Reisner } 2847b411b363SPhilipp Reisner 2848b411b363SPhilipp Reisner *rule_nr = 50; 2849b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_BITMAP] & ~((u64)1); 2850b411b363SPhilipp Reisner if (self == peer) 2851b411b363SPhilipp Reisner return -1; 2852b411b363SPhilipp Reisner 2853b411b363SPhilipp Reisner *rule_nr = 51; 2854b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_HISTORY_START] & ~((u64)1); 2855b411b363SPhilipp Reisner if (self == peer) { 2856a6b32bc3SAndreas Gruenbacher if (first_peer_device(device)->connection->agreed_pro_version < 96 ? 2857b30ab791SAndreas Gruenbacher (device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == 2858b30ab791SAndreas Gruenbacher (device->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : 2859b30ab791SAndreas Gruenbacher peer + UUID_NEW_BM_OFFSET == (device->p_uuid[UI_BITMAP] & ~((u64)1))) { 2860b411b363SPhilipp Reisner /* The last P_SYNC_UUID did not get though. Undo the last start of 2861b411b363SPhilipp Reisner resync as sync source modifications of the peer's UUIDs. */ 2862b411b363SPhilipp Reisner 2863a6b32bc3SAndreas Gruenbacher if (first_peer_device(device)->connection->agreed_pro_version < 91) 28644a23f264SPhilipp Reisner return -1091; 2865b411b363SPhilipp Reisner 2866b30ab791SAndreas Gruenbacher device->p_uuid[UI_BITMAP] = device->p_uuid[UI_HISTORY_START]; 2867b30ab791SAndreas Gruenbacher device->p_uuid[UI_HISTORY_START] = device->p_uuid[UI_HISTORY_START + 1]; 28684a23f264SPhilipp Reisner 2869d0180171SAndreas Gruenbacher drbd_info(device, "Lost last syncUUID packet, corrected:\n"); 2870b30ab791SAndreas Gruenbacher drbd_uuid_dump(device, "peer", device->p_uuid, device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); 28714a23f264SPhilipp Reisner 2872b411b363SPhilipp Reisner return -1; 2873b411b363SPhilipp Reisner } 2874b411b363SPhilipp Reisner } 2875b411b363SPhilipp Reisner 2876b411b363SPhilipp Reisner *rule_nr = 60; 2877b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[UI_CURRENT] & ~((u64)1); 2878b411b363SPhilipp Reisner for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 2879b30ab791SAndreas Gruenbacher peer = device->p_uuid[i] & ~((u64)1); 2880b411b363SPhilipp Reisner if (self == peer) 2881b411b363SPhilipp Reisner return -2; 2882b411b363SPhilipp Reisner } 2883b411b363SPhilipp Reisner 2884b411b363SPhilipp Reisner *rule_nr = 70; 2885b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[UI_BITMAP] & ~((u64)1); 2886b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_CURRENT] & ~((u64)1); 2887b411b363SPhilipp Reisner if (self == peer) 2888b411b363SPhilipp Reisner return 1; 2889b411b363SPhilipp Reisner 2890b411b363SPhilipp Reisner *rule_nr = 71; 2891b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); 2892b411b363SPhilipp Reisner if (self == peer) { 2893a6b32bc3SAndreas Gruenbacher if (first_peer_device(device)->connection->agreed_pro_version < 96 ? 2894b30ab791SAndreas Gruenbacher (device->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == 2895b30ab791SAndreas Gruenbacher (device->p_uuid[UI_HISTORY_START] & ~((u64)1)) : 2896b30ab791SAndreas Gruenbacher self + UUID_NEW_BM_OFFSET == (device->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { 2897b411b363SPhilipp Reisner /* The last P_SYNC_UUID did not get though. Undo the last start of 2898b411b363SPhilipp Reisner resync as sync source modifications of our UUIDs. */ 2899b411b363SPhilipp Reisner 2900a6b32bc3SAndreas Gruenbacher if (first_peer_device(device)->connection->agreed_pro_version < 91) 29014a23f264SPhilipp Reisner return -1091; 2902b411b363SPhilipp Reisner 2903b30ab791SAndreas Gruenbacher __drbd_uuid_set(device, UI_BITMAP, device->ldev->md.uuid[UI_HISTORY_START]); 2904b30ab791SAndreas Gruenbacher __drbd_uuid_set(device, UI_HISTORY_START, device->ldev->md.uuid[UI_HISTORY_START + 1]); 2905b411b363SPhilipp Reisner 2906d0180171SAndreas Gruenbacher drbd_info(device, "Last syncUUID did not get through, corrected:\n"); 2907b30ab791SAndreas Gruenbacher drbd_uuid_dump(device, "self", device->ldev->md.uuid, 2908b30ab791SAndreas Gruenbacher device->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(device) : 0, 0); 2909b411b363SPhilipp Reisner 2910b411b363SPhilipp Reisner return 1; 2911b411b363SPhilipp Reisner } 2912b411b363SPhilipp Reisner } 2913b411b363SPhilipp Reisner 2914b411b363SPhilipp Reisner 2915b411b363SPhilipp Reisner *rule_nr = 80; 2916b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_CURRENT] & ~((u64)1); 2917b411b363SPhilipp Reisner for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 2918b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[i] & ~((u64)1); 2919b411b363SPhilipp Reisner if (self == peer) 2920b411b363SPhilipp Reisner return 2; 2921b411b363SPhilipp Reisner } 2922b411b363SPhilipp Reisner 2923b411b363SPhilipp Reisner *rule_nr = 90; 2924b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[UI_BITMAP] & ~((u64)1); 2925b30ab791SAndreas Gruenbacher peer = device->p_uuid[UI_BITMAP] & ~((u64)1); 2926b411b363SPhilipp Reisner if (self == peer && self != ((u64)0)) 2927b411b363SPhilipp Reisner return 100; 2928b411b363SPhilipp Reisner 2929b411b363SPhilipp Reisner *rule_nr = 100; 2930b411b363SPhilipp Reisner for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { 2931b30ab791SAndreas Gruenbacher self = device->ldev->md.uuid[i] & ~((u64)1); 2932b411b363SPhilipp Reisner for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { 2933b30ab791SAndreas Gruenbacher peer = device->p_uuid[j] & ~((u64)1); 2934b411b363SPhilipp Reisner if (self == peer) 2935b411b363SPhilipp Reisner return -100; 2936b411b363SPhilipp Reisner } 2937b411b363SPhilipp Reisner } 2938b411b363SPhilipp Reisner 2939b411b363SPhilipp Reisner return -1000; 2940b411b363SPhilipp Reisner } 2941b411b363SPhilipp Reisner 2942b411b363SPhilipp Reisner /* drbd_sync_handshake() returns the new conn state on success, or 2943b411b363SPhilipp Reisner CONN_MASK (-1) on failure. 2944b411b363SPhilipp Reisner */ 2945b30ab791SAndreas Gruenbacher static enum drbd_conns drbd_sync_handshake(struct drbd_device *device, enum drbd_role peer_role, 2946b411b363SPhilipp Reisner enum drbd_disk_state peer_disk) __must_hold(local) 2947b411b363SPhilipp Reisner { 2948b411b363SPhilipp Reisner enum drbd_conns rv = C_MASK; 2949b411b363SPhilipp Reisner enum drbd_disk_state mydisk; 295044ed167dSPhilipp Reisner struct net_conf *nc; 29516dff2902SAndreas Gruenbacher int hg, rule_nr, rr_conflict, tentative; 2952b411b363SPhilipp Reisner 2953b30ab791SAndreas Gruenbacher mydisk = device->state.disk; 2954b411b363SPhilipp Reisner if (mydisk == D_NEGOTIATING) 2955b30ab791SAndreas Gruenbacher mydisk = device->new_state_tmp.disk; 2956b411b363SPhilipp Reisner 2957d0180171SAndreas Gruenbacher drbd_info(device, "drbd_sync_handshake:\n"); 29589f2247bbSPhilipp Reisner 2959b30ab791SAndreas Gruenbacher spin_lock_irq(&device->ldev->md.uuid_lock); 2960b30ab791SAndreas Gruenbacher drbd_uuid_dump(device, "self", device->ldev->md.uuid, device->comm_bm_set, 0); 2961b30ab791SAndreas Gruenbacher drbd_uuid_dump(device, "peer", device->p_uuid, 2962b30ab791SAndreas Gruenbacher device->p_uuid[UI_SIZE], device->p_uuid[UI_FLAGS]); 2963b411b363SPhilipp Reisner 2964b30ab791SAndreas Gruenbacher hg = drbd_uuid_compare(device, &rule_nr); 2965b30ab791SAndreas Gruenbacher spin_unlock_irq(&device->ldev->md.uuid_lock); 2966b411b363SPhilipp Reisner 2967d0180171SAndreas Gruenbacher drbd_info(device, "uuid_compare()=%d by rule %d\n", hg, rule_nr); 2968b411b363SPhilipp Reisner 2969b411b363SPhilipp Reisner if (hg == -1000) { 2970d0180171SAndreas Gruenbacher drbd_alert(device, "Unrelated data, aborting!\n"); 2971b411b363SPhilipp Reisner return C_MASK; 2972b411b363SPhilipp Reisner } 29734a23f264SPhilipp Reisner if (hg < -1000) { 2974d0180171SAndreas Gruenbacher drbd_alert(device, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); 2975b411b363SPhilipp Reisner return C_MASK; 2976b411b363SPhilipp Reisner } 2977b411b363SPhilipp Reisner 2978b411b363SPhilipp Reisner if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || 2979b411b363SPhilipp Reisner (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { 2980b411b363SPhilipp Reisner int f = (hg == -100) || abs(hg) == 2; 2981b411b363SPhilipp Reisner hg = mydisk > D_INCONSISTENT ? 1 : -1; 2982b411b363SPhilipp Reisner if (f) 2983b411b363SPhilipp Reisner hg = hg*2; 2984d0180171SAndreas Gruenbacher drbd_info(device, "Becoming sync %s due to disk states.\n", 2985b411b363SPhilipp Reisner hg > 0 ? "source" : "target"); 2986b411b363SPhilipp Reisner } 2987b411b363SPhilipp Reisner 29883a11a487SAdam Gandelman if (abs(hg) == 100) 2989b30ab791SAndreas Gruenbacher drbd_khelper(device, "initial-split-brain"); 29903a11a487SAdam Gandelman 299144ed167dSPhilipp Reisner rcu_read_lock(); 2992a6b32bc3SAndreas Gruenbacher nc = rcu_dereference(first_peer_device(device)->connection->net_conf); 299344ed167dSPhilipp Reisner 299444ed167dSPhilipp Reisner if (hg == 100 || (hg == -100 && nc->always_asbp)) { 2995b30ab791SAndreas Gruenbacher int pcount = (device->state.role == R_PRIMARY) 2996b411b363SPhilipp Reisner + (peer_role == R_PRIMARY); 2997b411b363SPhilipp Reisner int forced = (hg == -100); 2998b411b363SPhilipp Reisner 2999b411b363SPhilipp Reisner switch (pcount) { 3000b411b363SPhilipp Reisner case 0: 3001b30ab791SAndreas Gruenbacher hg = drbd_asb_recover_0p(device); 3002b411b363SPhilipp Reisner break; 3003b411b363SPhilipp Reisner case 1: 3004b30ab791SAndreas Gruenbacher hg = drbd_asb_recover_1p(device); 3005b411b363SPhilipp Reisner break; 3006b411b363SPhilipp Reisner case 2: 3007b30ab791SAndreas Gruenbacher hg = drbd_asb_recover_2p(device); 3008b411b363SPhilipp Reisner break; 3009b411b363SPhilipp Reisner } 3010b411b363SPhilipp Reisner if (abs(hg) < 100) { 3011d0180171SAndreas Gruenbacher drbd_warn(device, "Split-Brain detected, %d primaries, " 3012b411b363SPhilipp Reisner "automatically solved. Sync from %s node\n", 3013b411b363SPhilipp Reisner pcount, (hg < 0) ? "peer" : "this"); 3014b411b363SPhilipp Reisner if (forced) { 3015d0180171SAndreas Gruenbacher drbd_warn(device, "Doing a full sync, since" 3016b411b363SPhilipp Reisner " UUIDs where ambiguous.\n"); 3017b411b363SPhilipp Reisner hg = hg*2; 3018b411b363SPhilipp Reisner } 3019b411b363SPhilipp Reisner } 3020b411b363SPhilipp Reisner } 3021b411b363SPhilipp Reisner 3022b411b363SPhilipp Reisner if (hg == -100) { 3023b30ab791SAndreas Gruenbacher if (test_bit(DISCARD_MY_DATA, &device->flags) && !(device->p_uuid[UI_FLAGS]&1)) 3024b411b363SPhilipp Reisner hg = -1; 3025b30ab791SAndreas Gruenbacher if (!test_bit(DISCARD_MY_DATA, &device->flags) && (device->p_uuid[UI_FLAGS]&1)) 3026b411b363SPhilipp Reisner hg = 1; 3027b411b363SPhilipp Reisner 3028b411b363SPhilipp Reisner if (abs(hg) < 100) 3029d0180171SAndreas Gruenbacher drbd_warn(device, "Split-Brain detected, manually solved. " 3030b411b363SPhilipp Reisner "Sync from %s node\n", 3031b411b363SPhilipp Reisner (hg < 0) ? "peer" : "this"); 3032b411b363SPhilipp Reisner } 303344ed167dSPhilipp Reisner rr_conflict = nc->rr_conflict; 30346dff2902SAndreas Gruenbacher tentative = nc->tentative; 303544ed167dSPhilipp Reisner rcu_read_unlock(); 3036b411b363SPhilipp Reisner 3037b411b363SPhilipp Reisner if (hg == -100) { 3038580b9767SLars Ellenberg /* FIXME this log message is not correct if we end up here 3039580b9767SLars Ellenberg * after an attempted attach on a diskless node. 3040580b9767SLars Ellenberg * We just refuse to attach -- well, we drop the "connection" 3041580b9767SLars Ellenberg * to that disk, in a way... */ 3042d0180171SAndreas Gruenbacher drbd_alert(device, "Split-Brain detected but unresolved, dropping connection!\n"); 3043b30ab791SAndreas Gruenbacher drbd_khelper(device, "split-brain"); 3044b411b363SPhilipp Reisner return C_MASK; 3045b411b363SPhilipp Reisner } 3046b411b363SPhilipp Reisner 3047b411b363SPhilipp Reisner if (hg > 0 && mydisk <= D_INCONSISTENT) { 3048d0180171SAndreas Gruenbacher drbd_err(device, "I shall become SyncSource, but I am inconsistent!\n"); 3049b411b363SPhilipp Reisner return C_MASK; 3050b411b363SPhilipp Reisner } 3051b411b363SPhilipp Reisner 3052b411b363SPhilipp Reisner if (hg < 0 && /* by intention we do not use mydisk here. */ 3053b30ab791SAndreas Gruenbacher device->state.role == R_PRIMARY && device->state.disk >= D_CONSISTENT) { 305444ed167dSPhilipp Reisner switch (rr_conflict) { 3055b411b363SPhilipp Reisner case ASB_CALL_HELPER: 3056b30ab791SAndreas Gruenbacher drbd_khelper(device, "pri-lost"); 3057b411b363SPhilipp Reisner /* fall through */ 3058b411b363SPhilipp Reisner case ASB_DISCONNECT: 3059d0180171SAndreas Gruenbacher drbd_err(device, "I shall become SyncTarget, but I am primary!\n"); 3060b411b363SPhilipp Reisner return C_MASK; 3061b411b363SPhilipp Reisner case ASB_VIOLENTLY: 3062d0180171SAndreas Gruenbacher drbd_warn(device, "Becoming SyncTarget, violating the stable-data" 3063b411b363SPhilipp Reisner "assumption\n"); 3064b411b363SPhilipp Reisner } 3065b411b363SPhilipp Reisner } 3066b411b363SPhilipp Reisner 3067a6b32bc3SAndreas Gruenbacher if (tentative || test_bit(CONN_DRY_RUN, &first_peer_device(device)->connection->flags)) { 3068cf14c2e9SPhilipp Reisner if (hg == 0) 3069d0180171SAndreas Gruenbacher drbd_info(device, "dry-run connect: No resync, would become Connected immediately.\n"); 3070cf14c2e9SPhilipp Reisner else 3071d0180171SAndreas Gruenbacher drbd_info(device, "dry-run connect: Would become %s, doing a %s resync.", 3072cf14c2e9SPhilipp Reisner drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), 3073cf14c2e9SPhilipp Reisner abs(hg) >= 2 ? "full" : "bit-map based"); 3074cf14c2e9SPhilipp Reisner return C_MASK; 3075cf14c2e9SPhilipp Reisner } 3076cf14c2e9SPhilipp Reisner 3077b411b363SPhilipp Reisner if (abs(hg) >= 2) { 3078d0180171SAndreas Gruenbacher drbd_info(device, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); 3079b30ab791SAndreas Gruenbacher if (drbd_bitmap_io(device, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", 308020ceb2b2SLars Ellenberg BM_LOCKED_SET_ALLOWED)) 3081b411b363SPhilipp Reisner return C_MASK; 3082b411b363SPhilipp Reisner } 3083b411b363SPhilipp Reisner 3084b411b363SPhilipp Reisner if (hg > 0) { /* become sync source. */ 3085b411b363SPhilipp Reisner rv = C_WF_BITMAP_S; 3086b411b363SPhilipp Reisner } else if (hg < 0) { /* become sync target */ 3087b411b363SPhilipp Reisner rv = C_WF_BITMAP_T; 3088b411b363SPhilipp Reisner } else { 3089b411b363SPhilipp Reisner rv = C_CONNECTED; 3090b30ab791SAndreas Gruenbacher if (drbd_bm_total_weight(device)) { 3091d0180171SAndreas Gruenbacher drbd_info(device, "No resync, but %lu bits in bitmap!\n", 3092b30ab791SAndreas Gruenbacher drbd_bm_total_weight(device)); 3093b411b363SPhilipp Reisner } 3094b411b363SPhilipp Reisner } 3095b411b363SPhilipp Reisner 3096b411b363SPhilipp Reisner return rv; 3097b411b363SPhilipp Reisner } 3098b411b363SPhilipp Reisner 3099f179d76dSPhilipp Reisner static enum drbd_after_sb_p convert_after_sb(enum drbd_after_sb_p peer) 3100b411b363SPhilipp Reisner { 3101b411b363SPhilipp Reisner /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ 3102f179d76dSPhilipp Reisner if (peer == ASB_DISCARD_REMOTE) 3103f179d76dSPhilipp Reisner return ASB_DISCARD_LOCAL; 3104b411b363SPhilipp Reisner 3105b411b363SPhilipp Reisner /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ 3106f179d76dSPhilipp Reisner if (peer == ASB_DISCARD_LOCAL) 3107f179d76dSPhilipp Reisner return ASB_DISCARD_REMOTE; 3108b411b363SPhilipp Reisner 3109b411b363SPhilipp Reisner /* everything else is valid if they are equal on both sides. */ 3110f179d76dSPhilipp Reisner return peer; 3111b411b363SPhilipp Reisner } 3112b411b363SPhilipp Reisner 3113bde89a9eSAndreas Gruenbacher static int receive_protocol(struct drbd_connection *connection, struct packet_info *pi) 3114b411b363SPhilipp Reisner { 3115e658983aSAndreas Gruenbacher struct p_protocol *p = pi->data; 3116036b17eaSPhilipp Reisner enum drbd_after_sb_p p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; 3117036b17eaSPhilipp Reisner int p_proto, p_discard_my_data, p_two_primaries, cf; 3118036b17eaSPhilipp Reisner struct net_conf *nc, *old_net_conf, *new_net_conf = NULL; 3119036b17eaSPhilipp Reisner char integrity_alg[SHARED_SECRET_MAX] = ""; 3120accdbcc5SAndreas Gruenbacher struct crypto_hash *peer_integrity_tfm = NULL; 31217aca6c75SPhilipp Reisner void *int_dig_in = NULL, *int_dig_vv = NULL; 3122b411b363SPhilipp Reisner 3123b411b363SPhilipp Reisner p_proto = be32_to_cpu(p->protocol); 3124b411b363SPhilipp Reisner p_after_sb_0p = be32_to_cpu(p->after_sb_0p); 3125b411b363SPhilipp Reisner p_after_sb_1p = be32_to_cpu(p->after_sb_1p); 3126b411b363SPhilipp Reisner p_after_sb_2p = be32_to_cpu(p->after_sb_2p); 3127b411b363SPhilipp Reisner p_two_primaries = be32_to_cpu(p->two_primaries); 3128cf14c2e9SPhilipp Reisner cf = be32_to_cpu(p->conn_flags); 31296139f60dSAndreas Gruenbacher p_discard_my_data = cf & CF_DISCARD_MY_DATA; 3130cf14c2e9SPhilipp Reisner 3131bde89a9eSAndreas Gruenbacher if (connection->agreed_pro_version >= 87) { 313286db0618SAndreas Gruenbacher int err; 313386db0618SAndreas Gruenbacher 313488104ca4SAndreas Gruenbacher if (pi->size > sizeof(integrity_alg)) 313586db0618SAndreas Gruenbacher return -EIO; 3136bde89a9eSAndreas Gruenbacher err = drbd_recv_all(connection, integrity_alg, pi->size); 313786db0618SAndreas Gruenbacher if (err) 313886db0618SAndreas Gruenbacher return err; 313988104ca4SAndreas Gruenbacher integrity_alg[SHARED_SECRET_MAX - 1] = 0; 3140036b17eaSPhilipp Reisner } 314186db0618SAndreas Gruenbacher 31427d4c782cSAndreas Gruenbacher if (pi->cmd != P_PROTOCOL_UPDATE) { 3143bde89a9eSAndreas Gruenbacher clear_bit(CONN_DRY_RUN, &connection->flags); 3144cf14c2e9SPhilipp Reisner 3145cf14c2e9SPhilipp Reisner if (cf & CF_DRY_RUN) 3146bde89a9eSAndreas Gruenbacher set_bit(CONN_DRY_RUN, &connection->flags); 3147b411b363SPhilipp Reisner 314844ed167dSPhilipp Reisner rcu_read_lock(); 3149bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 315044ed167dSPhilipp Reisner 3151036b17eaSPhilipp Reisner if (p_proto != nc->wire_protocol) { 31521ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "protocol"); 315344ed167dSPhilipp Reisner goto disconnect_rcu_unlock; 3154b411b363SPhilipp Reisner } 3155b411b363SPhilipp Reisner 3156f179d76dSPhilipp Reisner if (convert_after_sb(p_after_sb_0p) != nc->after_sb_0p) { 31571ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "after-sb-0pri"); 315844ed167dSPhilipp Reisner goto disconnect_rcu_unlock; 3159b411b363SPhilipp Reisner } 3160b411b363SPhilipp Reisner 3161f179d76dSPhilipp Reisner if (convert_after_sb(p_after_sb_1p) != nc->after_sb_1p) { 31621ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "after-sb-1pri"); 316344ed167dSPhilipp Reisner goto disconnect_rcu_unlock; 3164b411b363SPhilipp Reisner } 3165b411b363SPhilipp Reisner 3166f179d76dSPhilipp Reisner if (convert_after_sb(p_after_sb_2p) != nc->after_sb_2p) { 31671ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "after-sb-2pri"); 316844ed167dSPhilipp Reisner goto disconnect_rcu_unlock; 3169b411b363SPhilipp Reisner } 3170b411b363SPhilipp Reisner 31716139f60dSAndreas Gruenbacher if (p_discard_my_data && nc->discard_my_data) { 31721ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "discard-my-data"); 317344ed167dSPhilipp Reisner goto disconnect_rcu_unlock; 3174b411b363SPhilipp Reisner } 3175b411b363SPhilipp Reisner 317644ed167dSPhilipp Reisner if (p_two_primaries != nc->two_primaries) { 31771ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "allow-two-primaries"); 317844ed167dSPhilipp Reisner goto disconnect_rcu_unlock; 3179b411b363SPhilipp Reisner } 3180b411b363SPhilipp Reisner 3181036b17eaSPhilipp Reisner if (strcmp(integrity_alg, nc->integrity_alg)) { 31821ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible %s settings\n", "data-integrity-alg"); 3183036b17eaSPhilipp Reisner goto disconnect_rcu_unlock; 3184036b17eaSPhilipp Reisner } 3185036b17eaSPhilipp Reisner 318686db0618SAndreas Gruenbacher rcu_read_unlock(); 3187fbc12f45SAndreas Gruenbacher } 31887d4c782cSAndreas Gruenbacher 31897d4c782cSAndreas Gruenbacher if (integrity_alg[0]) { 31907d4c782cSAndreas Gruenbacher int hash_size; 31917d4c782cSAndreas Gruenbacher 31927d4c782cSAndreas Gruenbacher /* 31937d4c782cSAndreas Gruenbacher * We can only change the peer data integrity algorithm 31947d4c782cSAndreas Gruenbacher * here. Changing our own data integrity algorithm 31957d4c782cSAndreas Gruenbacher * requires that we send a P_PROTOCOL_UPDATE packet at 31967d4c782cSAndreas Gruenbacher * the same time; otherwise, the peer has no way to 31977d4c782cSAndreas Gruenbacher * tell between which packets the algorithm should 31987d4c782cSAndreas Gruenbacher * change. 31997d4c782cSAndreas Gruenbacher */ 32007d4c782cSAndreas Gruenbacher 32017d4c782cSAndreas Gruenbacher peer_integrity_tfm = crypto_alloc_hash(integrity_alg, 0, CRYPTO_ALG_ASYNC); 32027d4c782cSAndreas Gruenbacher if (!peer_integrity_tfm) { 32031ec861ebSAndreas Gruenbacher drbd_err(connection, "peer data-integrity-alg %s not supported\n", 32047d4c782cSAndreas Gruenbacher integrity_alg); 3205b411b363SPhilipp Reisner goto disconnect; 3206b411b363SPhilipp Reisner } 3207b411b363SPhilipp Reisner 32087d4c782cSAndreas Gruenbacher hash_size = crypto_hash_digestsize(peer_integrity_tfm); 32097d4c782cSAndreas Gruenbacher int_dig_in = kmalloc(hash_size, GFP_KERNEL); 32107d4c782cSAndreas Gruenbacher int_dig_vv = kmalloc(hash_size, GFP_KERNEL); 32117d4c782cSAndreas Gruenbacher if (!(int_dig_in && int_dig_vv)) { 32121ec861ebSAndreas Gruenbacher drbd_err(connection, "Allocation of buffers for data integrity checking failed\n"); 32137d4c782cSAndreas Gruenbacher goto disconnect; 32147d4c782cSAndreas Gruenbacher } 32157d4c782cSAndreas Gruenbacher } 32167d4c782cSAndreas Gruenbacher 32177d4c782cSAndreas Gruenbacher new_net_conf = kmalloc(sizeof(struct net_conf), GFP_KERNEL); 32187d4c782cSAndreas Gruenbacher if (!new_net_conf) { 32191ec861ebSAndreas Gruenbacher drbd_err(connection, "Allocation of new net_conf failed\n"); 3220b411b363SPhilipp Reisner goto disconnect; 3221b411b363SPhilipp Reisner } 3222b411b363SPhilipp Reisner 3223bde89a9eSAndreas Gruenbacher mutex_lock(&connection->data.mutex); 32240500813fSAndreas Gruenbacher mutex_lock(&connection->resource->conf_update); 3225bde89a9eSAndreas Gruenbacher old_net_conf = connection->net_conf; 32267d4c782cSAndreas Gruenbacher *new_net_conf = *old_net_conf; 3227b411b363SPhilipp Reisner 32287d4c782cSAndreas Gruenbacher new_net_conf->wire_protocol = p_proto; 32297d4c782cSAndreas Gruenbacher new_net_conf->after_sb_0p = convert_after_sb(p_after_sb_0p); 32307d4c782cSAndreas Gruenbacher new_net_conf->after_sb_1p = convert_after_sb(p_after_sb_1p); 32317d4c782cSAndreas Gruenbacher new_net_conf->after_sb_2p = convert_after_sb(p_after_sb_2p); 32327d4c782cSAndreas Gruenbacher new_net_conf->two_primaries = p_two_primaries; 3233b411b363SPhilipp Reisner 3234bde89a9eSAndreas Gruenbacher rcu_assign_pointer(connection->net_conf, new_net_conf); 32350500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 3236bde89a9eSAndreas Gruenbacher mutex_unlock(&connection->data.mutex); 3237b411b363SPhilipp Reisner 3238bde89a9eSAndreas Gruenbacher crypto_free_hash(connection->peer_integrity_tfm); 3239bde89a9eSAndreas Gruenbacher kfree(connection->int_dig_in); 3240bde89a9eSAndreas Gruenbacher kfree(connection->int_dig_vv); 3241bde89a9eSAndreas Gruenbacher connection->peer_integrity_tfm = peer_integrity_tfm; 3242bde89a9eSAndreas Gruenbacher connection->int_dig_in = int_dig_in; 3243bde89a9eSAndreas Gruenbacher connection->int_dig_vv = int_dig_vv; 3244b411b363SPhilipp Reisner 32457d4c782cSAndreas Gruenbacher if (strcmp(old_net_conf->integrity_alg, integrity_alg)) 32461ec861ebSAndreas Gruenbacher drbd_info(connection, "peer data-integrity-alg: %s\n", 32477d4c782cSAndreas Gruenbacher integrity_alg[0] ? integrity_alg : "(none)"); 3248b411b363SPhilipp Reisner 32497d4c782cSAndreas Gruenbacher synchronize_rcu(); 32507d4c782cSAndreas Gruenbacher kfree(old_net_conf); 325182bc0194SAndreas Gruenbacher return 0; 3252b411b363SPhilipp Reisner 325344ed167dSPhilipp Reisner disconnect_rcu_unlock: 325444ed167dSPhilipp Reisner rcu_read_unlock(); 3255b411b363SPhilipp Reisner disconnect: 3256b792c35cSAndreas Gruenbacher crypto_free_hash(peer_integrity_tfm); 3257036b17eaSPhilipp Reisner kfree(int_dig_in); 3258036b17eaSPhilipp Reisner kfree(int_dig_vv); 3259bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); 326082bc0194SAndreas Gruenbacher return -EIO; 3261b411b363SPhilipp Reisner } 3262b411b363SPhilipp Reisner 3263b411b363SPhilipp Reisner /* helper function 3264b411b363SPhilipp Reisner * input: alg name, feature name 3265b411b363SPhilipp Reisner * return: NULL (alg name was "") 3266b411b363SPhilipp Reisner * ERR_PTR(error) if something goes wrong 3267b411b363SPhilipp Reisner * or the crypto hash ptr, if it worked out ok. */ 3268f63e631aSRashika Kheria static 3269b30ab791SAndreas Gruenbacher struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_device *device, 3270b411b363SPhilipp Reisner const char *alg, const char *name) 3271b411b363SPhilipp Reisner { 3272b411b363SPhilipp Reisner struct crypto_hash *tfm; 3273b411b363SPhilipp Reisner 3274b411b363SPhilipp Reisner if (!alg[0]) 3275b411b363SPhilipp Reisner return NULL; 3276b411b363SPhilipp Reisner 3277b411b363SPhilipp Reisner tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); 3278b411b363SPhilipp Reisner if (IS_ERR(tfm)) { 3279d0180171SAndreas Gruenbacher drbd_err(device, "Can not allocate \"%s\" as %s (reason: %ld)\n", 3280b411b363SPhilipp Reisner alg, name, PTR_ERR(tfm)); 3281b411b363SPhilipp Reisner return tfm; 3282b411b363SPhilipp Reisner } 3283b411b363SPhilipp Reisner return tfm; 3284b411b363SPhilipp Reisner } 3285b411b363SPhilipp Reisner 3286bde89a9eSAndreas Gruenbacher static int ignore_remaining_packet(struct drbd_connection *connection, struct packet_info *pi) 3287b411b363SPhilipp Reisner { 3288bde89a9eSAndreas Gruenbacher void *buffer = connection->data.rbuf; 32894a76b161SAndreas Gruenbacher int size = pi->size; 32904a76b161SAndreas Gruenbacher 32914a76b161SAndreas Gruenbacher while (size) { 32924a76b161SAndreas Gruenbacher int s = min_t(int, size, DRBD_SOCKET_BUFFER_SIZE); 3293bde89a9eSAndreas Gruenbacher s = drbd_recv(connection, buffer, s); 32944a76b161SAndreas Gruenbacher if (s <= 0) { 32954a76b161SAndreas Gruenbacher if (s < 0) 32964a76b161SAndreas Gruenbacher return s; 32974a76b161SAndreas Gruenbacher break; 32984a76b161SAndreas Gruenbacher } 32994a76b161SAndreas Gruenbacher size -= s; 33004a76b161SAndreas Gruenbacher } 33014a76b161SAndreas Gruenbacher if (size) 33024a76b161SAndreas Gruenbacher return -EIO; 33034a76b161SAndreas Gruenbacher return 0; 33044a76b161SAndreas Gruenbacher } 33054a76b161SAndreas Gruenbacher 33064a76b161SAndreas Gruenbacher /* 33074a76b161SAndreas Gruenbacher * config_unknown_volume - device configuration command for unknown volume 33084a76b161SAndreas Gruenbacher * 33094a76b161SAndreas Gruenbacher * When a device is added to an existing connection, the node on which the 33104a76b161SAndreas Gruenbacher * device is added first will send configuration commands to its peer but the 33114a76b161SAndreas Gruenbacher * peer will not know about the device yet. It will warn and ignore these 33124a76b161SAndreas Gruenbacher * commands. Once the device is added on the second node, the second node will 33134a76b161SAndreas Gruenbacher * send the same device configuration commands, but in the other direction. 33144a76b161SAndreas Gruenbacher * 33154a76b161SAndreas Gruenbacher * (We can also end up here if drbd is misconfigured.) 33164a76b161SAndreas Gruenbacher */ 3317bde89a9eSAndreas Gruenbacher static int config_unknown_volume(struct drbd_connection *connection, struct packet_info *pi) 33184a76b161SAndreas Gruenbacher { 33191ec861ebSAndreas Gruenbacher drbd_warn(connection, "%s packet received for volume %u, which is not configured locally\n", 33202fcb8f30SAndreas Gruenbacher cmdname(pi->cmd), pi->vnr); 3321bde89a9eSAndreas Gruenbacher return ignore_remaining_packet(connection, pi); 33224a76b161SAndreas Gruenbacher } 33234a76b161SAndreas Gruenbacher 3324bde89a9eSAndreas Gruenbacher static int receive_SyncParam(struct drbd_connection *connection, struct packet_info *pi) 33254a76b161SAndreas Gruenbacher { 3326b30ab791SAndreas Gruenbacher struct drbd_device *device; 3327e658983aSAndreas Gruenbacher struct p_rs_param_95 *p; 3328b411b363SPhilipp Reisner unsigned int header_size, data_size, exp_max_sz; 3329b411b363SPhilipp Reisner struct crypto_hash *verify_tfm = NULL; 3330b411b363SPhilipp Reisner struct crypto_hash *csums_tfm = NULL; 33312ec91e0eSPhilipp Reisner struct net_conf *old_net_conf, *new_net_conf = NULL; 3332813472ceSPhilipp Reisner struct disk_conf *old_disk_conf = NULL, *new_disk_conf = NULL; 3333bde89a9eSAndreas Gruenbacher const int apv = connection->agreed_pro_version; 3334813472ceSPhilipp Reisner struct fifo_buffer *old_plan = NULL, *new_plan = NULL; 3335778f271dSPhilipp Reisner int fifo_size = 0; 333682bc0194SAndreas Gruenbacher int err; 3337b411b363SPhilipp Reisner 3338bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 3339b30ab791SAndreas Gruenbacher if (!device) 3340bde89a9eSAndreas Gruenbacher return config_unknown_volume(connection, pi); 3341b411b363SPhilipp Reisner 3342b411b363SPhilipp Reisner exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) 3343b411b363SPhilipp Reisner : apv == 88 ? sizeof(struct p_rs_param) 3344b411b363SPhilipp Reisner + SHARED_SECRET_MAX 33458e26f9ccSPhilipp Reisner : apv <= 94 ? sizeof(struct p_rs_param_89) 33468e26f9ccSPhilipp Reisner : /* apv >= 95 */ sizeof(struct p_rs_param_95); 3347b411b363SPhilipp Reisner 3348e2857216SAndreas Gruenbacher if (pi->size > exp_max_sz) { 3349d0180171SAndreas Gruenbacher drbd_err(device, "SyncParam packet too long: received %u, expected <= %u bytes\n", 3350e2857216SAndreas Gruenbacher pi->size, exp_max_sz); 335182bc0194SAndreas Gruenbacher return -EIO; 3352b411b363SPhilipp Reisner } 3353b411b363SPhilipp Reisner 3354b411b363SPhilipp Reisner if (apv <= 88) { 3355e658983aSAndreas Gruenbacher header_size = sizeof(struct p_rs_param); 3356e2857216SAndreas Gruenbacher data_size = pi->size - header_size; 33578e26f9ccSPhilipp Reisner } else if (apv <= 94) { 3358e658983aSAndreas Gruenbacher header_size = sizeof(struct p_rs_param_89); 3359e2857216SAndreas Gruenbacher data_size = pi->size - header_size; 33600b0ba1efSAndreas Gruenbacher D_ASSERT(device, data_size == 0); 33618e26f9ccSPhilipp Reisner } else { 3362e658983aSAndreas Gruenbacher header_size = sizeof(struct p_rs_param_95); 3363e2857216SAndreas Gruenbacher data_size = pi->size - header_size; 33640b0ba1efSAndreas Gruenbacher D_ASSERT(device, data_size == 0); 3365b411b363SPhilipp Reisner } 3366b411b363SPhilipp Reisner 3367b411b363SPhilipp Reisner /* initialize verify_alg and csums_alg */ 3368e658983aSAndreas Gruenbacher p = pi->data; 3369b411b363SPhilipp Reisner memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); 3370b411b363SPhilipp Reisner 3371a6b32bc3SAndreas Gruenbacher err = drbd_recv_all(first_peer_device(device)->connection, p, header_size); 337282bc0194SAndreas Gruenbacher if (err) 337382bc0194SAndreas Gruenbacher return err; 3374b411b363SPhilipp Reisner 33750500813fSAndreas Gruenbacher mutex_lock(&connection->resource->conf_update); 3376a6b32bc3SAndreas Gruenbacher old_net_conf = first_peer_device(device)->connection->net_conf; 3377b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 3378daeda1ccSPhilipp Reisner new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); 3379daeda1ccSPhilipp Reisner if (!new_disk_conf) { 3380b30ab791SAndreas Gruenbacher put_ldev(device); 33810500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 3382d0180171SAndreas Gruenbacher drbd_err(device, "Allocation of new disk_conf failed\n"); 3383daeda1ccSPhilipp Reisner return -ENOMEM; 3384f399002eSLars Ellenberg } 3385b411b363SPhilipp Reisner 3386b30ab791SAndreas Gruenbacher old_disk_conf = device->ldev->disk_conf; 3387daeda1ccSPhilipp Reisner *new_disk_conf = *old_disk_conf; 3388daeda1ccSPhilipp Reisner 33896394b935SAndreas Gruenbacher new_disk_conf->resync_rate = be32_to_cpu(p->resync_rate); 3390813472ceSPhilipp Reisner } 3391b411b363SPhilipp Reisner 3392b411b363SPhilipp Reisner if (apv >= 88) { 3393b411b363SPhilipp Reisner if (apv == 88) { 33945de73827SPhilipp Reisner if (data_size > SHARED_SECRET_MAX || data_size == 0) { 3395d0180171SAndreas Gruenbacher drbd_err(device, "verify-alg of wrong size, " 33965de73827SPhilipp Reisner "peer wants %u, accepting only up to %u byte\n", 3397b411b363SPhilipp Reisner data_size, SHARED_SECRET_MAX); 3398813472ceSPhilipp Reisner err = -EIO; 3399813472ceSPhilipp Reisner goto reconnect; 3400b411b363SPhilipp Reisner } 3401b411b363SPhilipp Reisner 3402a6b32bc3SAndreas Gruenbacher err = drbd_recv_all(first_peer_device(device)->connection, p->verify_alg, data_size); 3403813472ceSPhilipp Reisner if (err) 3404813472ceSPhilipp Reisner goto reconnect; 3405b411b363SPhilipp Reisner /* we expect NUL terminated string */ 3406b411b363SPhilipp Reisner /* but just in case someone tries to be evil */ 34070b0ba1efSAndreas Gruenbacher D_ASSERT(device, p->verify_alg[data_size-1] == 0); 3408b411b363SPhilipp Reisner p->verify_alg[data_size-1] = 0; 3409b411b363SPhilipp Reisner 3410b411b363SPhilipp Reisner } else /* apv >= 89 */ { 3411b411b363SPhilipp Reisner /* we still expect NUL terminated strings */ 3412b411b363SPhilipp Reisner /* but just in case someone tries to be evil */ 34130b0ba1efSAndreas Gruenbacher D_ASSERT(device, p->verify_alg[SHARED_SECRET_MAX-1] == 0); 34140b0ba1efSAndreas Gruenbacher D_ASSERT(device, p->csums_alg[SHARED_SECRET_MAX-1] == 0); 3415b411b363SPhilipp Reisner p->verify_alg[SHARED_SECRET_MAX-1] = 0; 3416b411b363SPhilipp Reisner p->csums_alg[SHARED_SECRET_MAX-1] = 0; 3417b411b363SPhilipp Reisner } 3418b411b363SPhilipp Reisner 34192ec91e0eSPhilipp Reisner if (strcmp(old_net_conf->verify_alg, p->verify_alg)) { 3420b30ab791SAndreas Gruenbacher if (device->state.conn == C_WF_REPORT_PARAMS) { 3421d0180171SAndreas Gruenbacher drbd_err(device, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", 34222ec91e0eSPhilipp Reisner old_net_conf->verify_alg, p->verify_alg); 3423b411b363SPhilipp Reisner goto disconnect; 3424b411b363SPhilipp Reisner } 3425b30ab791SAndreas Gruenbacher verify_tfm = drbd_crypto_alloc_digest_safe(device, 3426b411b363SPhilipp Reisner p->verify_alg, "verify-alg"); 3427b411b363SPhilipp Reisner if (IS_ERR(verify_tfm)) { 3428b411b363SPhilipp Reisner verify_tfm = NULL; 3429b411b363SPhilipp Reisner goto disconnect; 3430b411b363SPhilipp Reisner } 3431b411b363SPhilipp Reisner } 3432b411b363SPhilipp Reisner 34332ec91e0eSPhilipp Reisner if (apv >= 89 && strcmp(old_net_conf->csums_alg, p->csums_alg)) { 3434b30ab791SAndreas Gruenbacher if (device->state.conn == C_WF_REPORT_PARAMS) { 3435d0180171SAndreas Gruenbacher drbd_err(device, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", 34362ec91e0eSPhilipp Reisner old_net_conf->csums_alg, p->csums_alg); 3437b411b363SPhilipp Reisner goto disconnect; 3438b411b363SPhilipp Reisner } 3439b30ab791SAndreas Gruenbacher csums_tfm = drbd_crypto_alloc_digest_safe(device, 3440b411b363SPhilipp Reisner p->csums_alg, "csums-alg"); 3441b411b363SPhilipp Reisner if (IS_ERR(csums_tfm)) { 3442b411b363SPhilipp Reisner csums_tfm = NULL; 3443b411b363SPhilipp Reisner goto disconnect; 3444b411b363SPhilipp Reisner } 3445b411b363SPhilipp Reisner } 3446b411b363SPhilipp Reisner 3447813472ceSPhilipp Reisner if (apv > 94 && new_disk_conf) { 3448daeda1ccSPhilipp Reisner new_disk_conf->c_plan_ahead = be32_to_cpu(p->c_plan_ahead); 3449daeda1ccSPhilipp Reisner new_disk_conf->c_delay_target = be32_to_cpu(p->c_delay_target); 3450daeda1ccSPhilipp Reisner new_disk_conf->c_fill_target = be32_to_cpu(p->c_fill_target); 3451daeda1ccSPhilipp Reisner new_disk_conf->c_max_rate = be32_to_cpu(p->c_max_rate); 3452778f271dSPhilipp Reisner 3453daeda1ccSPhilipp Reisner fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ; 3454b30ab791SAndreas Gruenbacher if (fifo_size != device->rs_plan_s->size) { 3455813472ceSPhilipp Reisner new_plan = fifo_alloc(fifo_size); 3456813472ceSPhilipp Reisner if (!new_plan) { 3457d0180171SAndreas Gruenbacher drbd_err(device, "kmalloc of fifo_buffer failed"); 3458b30ab791SAndreas Gruenbacher put_ldev(device); 3459778f271dSPhilipp Reisner goto disconnect; 3460778f271dSPhilipp Reisner } 3461778f271dSPhilipp Reisner } 34628e26f9ccSPhilipp Reisner } 3463b411b363SPhilipp Reisner 346491fd4dadSPhilipp Reisner if (verify_tfm || csums_tfm) { 34652ec91e0eSPhilipp Reisner new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL); 34662ec91e0eSPhilipp Reisner if (!new_net_conf) { 3467d0180171SAndreas Gruenbacher drbd_err(device, "Allocation of new net_conf failed\n"); 346891fd4dadSPhilipp Reisner goto disconnect; 346991fd4dadSPhilipp Reisner } 347091fd4dadSPhilipp Reisner 34712ec91e0eSPhilipp Reisner *new_net_conf = *old_net_conf; 347291fd4dadSPhilipp Reisner 3473b411b363SPhilipp Reisner if (verify_tfm) { 34742ec91e0eSPhilipp Reisner strcpy(new_net_conf->verify_alg, p->verify_alg); 34752ec91e0eSPhilipp Reisner new_net_conf->verify_alg_len = strlen(p->verify_alg) + 1; 3476a6b32bc3SAndreas Gruenbacher crypto_free_hash(first_peer_device(device)->connection->verify_tfm); 3477a6b32bc3SAndreas Gruenbacher first_peer_device(device)->connection->verify_tfm = verify_tfm; 3478d0180171SAndreas Gruenbacher drbd_info(device, "using verify-alg: \"%s\"\n", p->verify_alg); 3479b411b363SPhilipp Reisner } 3480b411b363SPhilipp Reisner if (csums_tfm) { 34812ec91e0eSPhilipp Reisner strcpy(new_net_conf->csums_alg, p->csums_alg); 34822ec91e0eSPhilipp Reisner new_net_conf->csums_alg_len = strlen(p->csums_alg) + 1; 3483a6b32bc3SAndreas Gruenbacher crypto_free_hash(first_peer_device(device)->connection->csums_tfm); 3484a6b32bc3SAndreas Gruenbacher first_peer_device(device)->connection->csums_tfm = csums_tfm; 3485d0180171SAndreas Gruenbacher drbd_info(device, "using csums-alg: \"%s\"\n", p->csums_alg); 3486b411b363SPhilipp Reisner } 3487bde89a9eSAndreas Gruenbacher rcu_assign_pointer(connection->net_conf, new_net_conf); 3488778f271dSPhilipp Reisner } 3489b411b363SPhilipp Reisner } 3490b411b363SPhilipp Reisner 3491813472ceSPhilipp Reisner if (new_disk_conf) { 3492b30ab791SAndreas Gruenbacher rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); 3493b30ab791SAndreas Gruenbacher put_ldev(device); 3494b411b363SPhilipp Reisner } 3495813472ceSPhilipp Reisner 3496813472ceSPhilipp Reisner if (new_plan) { 3497b30ab791SAndreas Gruenbacher old_plan = device->rs_plan_s; 3498b30ab791SAndreas Gruenbacher rcu_assign_pointer(device->rs_plan_s, new_plan); 3499813472ceSPhilipp Reisner } 3500daeda1ccSPhilipp Reisner 35010500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 3502daeda1ccSPhilipp Reisner synchronize_rcu(); 3503daeda1ccSPhilipp Reisner if (new_net_conf) 3504daeda1ccSPhilipp Reisner kfree(old_net_conf); 3505daeda1ccSPhilipp Reisner kfree(old_disk_conf); 3506813472ceSPhilipp Reisner kfree(old_plan); 3507daeda1ccSPhilipp Reisner 350882bc0194SAndreas Gruenbacher return 0; 3509b411b363SPhilipp Reisner 3510813472ceSPhilipp Reisner reconnect: 3511813472ceSPhilipp Reisner if (new_disk_conf) { 3512b30ab791SAndreas Gruenbacher put_ldev(device); 3513813472ceSPhilipp Reisner kfree(new_disk_conf); 3514813472ceSPhilipp Reisner } 35150500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 3516813472ceSPhilipp Reisner return -EIO; 3517813472ceSPhilipp Reisner 3518b411b363SPhilipp Reisner disconnect: 3519813472ceSPhilipp Reisner kfree(new_plan); 3520813472ceSPhilipp Reisner if (new_disk_conf) { 3521b30ab791SAndreas Gruenbacher put_ldev(device); 3522813472ceSPhilipp Reisner kfree(new_disk_conf); 3523813472ceSPhilipp Reisner } 35240500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 3525b411b363SPhilipp Reisner /* just for completeness: actually not needed, 3526b411b363SPhilipp Reisner * as this is not reached if csums_tfm was ok. */ 3527b411b363SPhilipp Reisner crypto_free_hash(csums_tfm); 3528b411b363SPhilipp Reisner /* but free the verify_tfm again, if csums_tfm did not work out */ 3529b411b363SPhilipp Reisner crypto_free_hash(verify_tfm); 3530a6b32bc3SAndreas Gruenbacher conn_request_state(first_peer_device(device)->connection, NS(conn, C_DISCONNECTING), CS_HARD); 353182bc0194SAndreas Gruenbacher return -EIO; 3532b411b363SPhilipp Reisner } 3533b411b363SPhilipp Reisner 3534b411b363SPhilipp Reisner /* warn if the arguments differ by more than 12.5% */ 3535b30ab791SAndreas Gruenbacher static void warn_if_differ_considerably(struct drbd_device *device, 3536b411b363SPhilipp Reisner const char *s, sector_t a, sector_t b) 3537b411b363SPhilipp Reisner { 3538b411b363SPhilipp Reisner sector_t d; 3539b411b363SPhilipp Reisner if (a == 0 || b == 0) 3540b411b363SPhilipp Reisner return; 3541b411b363SPhilipp Reisner d = (a > b) ? (a - b) : (b - a); 3542b411b363SPhilipp Reisner if (d > (a>>3) || d > (b>>3)) 3543d0180171SAndreas Gruenbacher drbd_warn(device, "Considerable difference in %s: %llus vs. %llus\n", s, 3544b411b363SPhilipp Reisner (unsigned long long)a, (unsigned long long)b); 3545b411b363SPhilipp Reisner } 3546b411b363SPhilipp Reisner 3547bde89a9eSAndreas Gruenbacher static int receive_sizes(struct drbd_connection *connection, struct packet_info *pi) 3548b411b363SPhilipp Reisner { 3549b30ab791SAndreas Gruenbacher struct drbd_device *device; 3550e658983aSAndreas Gruenbacher struct p_sizes *p = pi->data; 3551e96c9633SPhilipp Reisner enum determine_dev_size dd = DS_UNCHANGED; 3552b411b363SPhilipp Reisner sector_t p_size, p_usize, my_usize; 3553b411b363SPhilipp Reisner int ldsc = 0; /* local disk size changed */ 3554e89b591cSPhilipp Reisner enum dds_flags ddsf; 3555b411b363SPhilipp Reisner 3556bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 3557b30ab791SAndreas Gruenbacher if (!device) 3558bde89a9eSAndreas Gruenbacher return config_unknown_volume(connection, pi); 35594a76b161SAndreas Gruenbacher 3560b411b363SPhilipp Reisner p_size = be64_to_cpu(p->d_size); 3561b411b363SPhilipp Reisner p_usize = be64_to_cpu(p->u_size); 3562b411b363SPhilipp Reisner 3563b411b363SPhilipp Reisner /* just store the peer's disk size for now. 3564b411b363SPhilipp Reisner * we still need to figure out whether we accept that. */ 3565b30ab791SAndreas Gruenbacher device->p_size = p_size; 3566b411b363SPhilipp Reisner 3567b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 3568daeda1ccSPhilipp Reisner rcu_read_lock(); 3569b30ab791SAndreas Gruenbacher my_usize = rcu_dereference(device->ldev->disk_conf)->disk_size; 3570daeda1ccSPhilipp Reisner rcu_read_unlock(); 3571daeda1ccSPhilipp Reisner 3572b30ab791SAndreas Gruenbacher warn_if_differ_considerably(device, "lower level device sizes", 3573b30ab791SAndreas Gruenbacher p_size, drbd_get_max_capacity(device->ldev)); 3574b30ab791SAndreas Gruenbacher warn_if_differ_considerably(device, "user requested size", 3575daeda1ccSPhilipp Reisner p_usize, my_usize); 3576b411b363SPhilipp Reisner 3577b411b363SPhilipp Reisner /* if this is the first connect, or an otherwise expected 3578b411b363SPhilipp Reisner * param exchange, choose the minimum */ 3579b30ab791SAndreas Gruenbacher if (device->state.conn == C_WF_REPORT_PARAMS) 3580daeda1ccSPhilipp Reisner p_usize = min_not_zero(my_usize, p_usize); 3581b411b363SPhilipp Reisner 3582b411b363SPhilipp Reisner /* Never shrink a device with usable data during connect. 3583b411b363SPhilipp Reisner But allow online shrinking if we are connected. */ 3584b30ab791SAndreas Gruenbacher if (drbd_new_dev_size(device, device->ldev, p_usize, 0) < 3585b30ab791SAndreas Gruenbacher drbd_get_capacity(device->this_bdev) && 3586b30ab791SAndreas Gruenbacher device->state.disk >= D_OUTDATED && 3587b30ab791SAndreas Gruenbacher device->state.conn < C_CONNECTED) { 3588d0180171SAndreas Gruenbacher drbd_err(device, "The peer's disk size is too small!\n"); 3589a6b32bc3SAndreas Gruenbacher conn_request_state(first_peer_device(device)->connection, NS(conn, C_DISCONNECTING), CS_HARD); 3590b30ab791SAndreas Gruenbacher put_ldev(device); 359182bc0194SAndreas Gruenbacher return -EIO; 3592b411b363SPhilipp Reisner } 3593daeda1ccSPhilipp Reisner 3594daeda1ccSPhilipp Reisner if (my_usize != p_usize) { 3595daeda1ccSPhilipp Reisner struct disk_conf *old_disk_conf, *new_disk_conf = NULL; 3596daeda1ccSPhilipp Reisner 3597daeda1ccSPhilipp Reisner new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); 3598daeda1ccSPhilipp Reisner if (!new_disk_conf) { 3599d0180171SAndreas Gruenbacher drbd_err(device, "Allocation of new disk_conf failed\n"); 3600b30ab791SAndreas Gruenbacher put_ldev(device); 3601daeda1ccSPhilipp Reisner return -ENOMEM; 3602daeda1ccSPhilipp Reisner } 3603daeda1ccSPhilipp Reisner 36040500813fSAndreas Gruenbacher mutex_lock(&connection->resource->conf_update); 3605b30ab791SAndreas Gruenbacher old_disk_conf = device->ldev->disk_conf; 3606daeda1ccSPhilipp Reisner *new_disk_conf = *old_disk_conf; 3607daeda1ccSPhilipp Reisner new_disk_conf->disk_size = p_usize; 3608daeda1ccSPhilipp Reisner 3609b30ab791SAndreas Gruenbacher rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); 36100500813fSAndreas Gruenbacher mutex_unlock(&connection->resource->conf_update); 3611daeda1ccSPhilipp Reisner synchronize_rcu(); 3612daeda1ccSPhilipp Reisner kfree(old_disk_conf); 3613daeda1ccSPhilipp Reisner 3614d0180171SAndreas Gruenbacher drbd_info(device, "Peer sets u_size to %lu sectors\n", 3615daeda1ccSPhilipp Reisner (unsigned long)my_usize); 3616daeda1ccSPhilipp Reisner } 3617daeda1ccSPhilipp Reisner 3618b30ab791SAndreas Gruenbacher put_ldev(device); 3619b411b363SPhilipp Reisner } 3620b411b363SPhilipp Reisner 3621e89b591cSPhilipp Reisner ddsf = be16_to_cpu(p->dds_flags); 3622b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 3623b30ab791SAndreas Gruenbacher dd = drbd_determine_dev_size(device, ddsf, NULL); 3624b30ab791SAndreas Gruenbacher put_ldev(device); 3625e96c9633SPhilipp Reisner if (dd == DS_ERROR) 362682bc0194SAndreas Gruenbacher return -EIO; 3627b30ab791SAndreas Gruenbacher drbd_md_sync(device); 3628b411b363SPhilipp Reisner } else { 3629b411b363SPhilipp Reisner /* I am diskless, need to accept the peer's size. */ 3630b30ab791SAndreas Gruenbacher drbd_set_my_capacity(device, p_size); 3631b411b363SPhilipp Reisner } 3632b411b363SPhilipp Reisner 3633b30ab791SAndreas Gruenbacher device->peer_max_bio_size = be32_to_cpu(p->max_bio_size); 3634b30ab791SAndreas Gruenbacher drbd_reconsider_max_bio_size(device); 363599432fccSPhilipp Reisner 3636b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 3637b30ab791SAndreas Gruenbacher if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev)) { 3638b30ab791SAndreas Gruenbacher device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev); 3639b411b363SPhilipp Reisner ldsc = 1; 3640b411b363SPhilipp Reisner } 3641b411b363SPhilipp Reisner 3642b30ab791SAndreas Gruenbacher put_ldev(device); 3643b411b363SPhilipp Reisner } 3644b411b363SPhilipp Reisner 3645b30ab791SAndreas Gruenbacher if (device->state.conn > C_WF_REPORT_PARAMS) { 3646b411b363SPhilipp Reisner if (be64_to_cpu(p->c_size) != 3647b30ab791SAndreas Gruenbacher drbd_get_capacity(device->this_bdev) || ldsc) { 3648b411b363SPhilipp Reisner /* we have different sizes, probably peer 3649b411b363SPhilipp Reisner * needs to know my new size... */ 3650b30ab791SAndreas Gruenbacher drbd_send_sizes(device, 0, ddsf); 3651b411b363SPhilipp Reisner } 3652b30ab791SAndreas Gruenbacher if (test_and_clear_bit(RESIZE_PENDING, &device->flags) || 3653b30ab791SAndreas Gruenbacher (dd == DS_GREW && device->state.conn == C_CONNECTED)) { 3654b30ab791SAndreas Gruenbacher if (device->state.pdsk >= D_INCONSISTENT && 3655b30ab791SAndreas Gruenbacher device->state.disk >= D_INCONSISTENT) { 3656e89b591cSPhilipp Reisner if (ddsf & DDSF_NO_RESYNC) 3657d0180171SAndreas Gruenbacher drbd_info(device, "Resync of new storage suppressed with --assume-clean\n"); 3658b411b363SPhilipp Reisner else 3659b30ab791SAndreas Gruenbacher resync_after_online_grow(device); 3660e89b591cSPhilipp Reisner } else 3661b30ab791SAndreas Gruenbacher set_bit(RESYNC_AFTER_NEG, &device->flags); 3662b411b363SPhilipp Reisner } 3663b411b363SPhilipp Reisner } 3664b411b363SPhilipp Reisner 366582bc0194SAndreas Gruenbacher return 0; 3666b411b363SPhilipp Reisner } 3667b411b363SPhilipp Reisner 3668bde89a9eSAndreas Gruenbacher static int receive_uuids(struct drbd_connection *connection, struct packet_info *pi) 3669b411b363SPhilipp Reisner { 3670b30ab791SAndreas Gruenbacher struct drbd_device *device; 3671e658983aSAndreas Gruenbacher struct p_uuids *p = pi->data; 3672b411b363SPhilipp Reisner u64 *p_uuid; 367362b0da3aSLars Ellenberg int i, updated_uuids = 0; 3674b411b363SPhilipp Reisner 3675bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 3676b30ab791SAndreas Gruenbacher if (!device) 3677bde89a9eSAndreas Gruenbacher return config_unknown_volume(connection, pi); 36784a76b161SAndreas Gruenbacher 3679b411b363SPhilipp Reisner p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); 3680063eacf8SJing Wang if (!p_uuid) { 3681d0180171SAndreas Gruenbacher drbd_err(device, "kmalloc of p_uuid failed\n"); 3682063eacf8SJing Wang return false; 3683063eacf8SJing Wang } 3684b411b363SPhilipp Reisner 3685b411b363SPhilipp Reisner for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) 3686b411b363SPhilipp Reisner p_uuid[i] = be64_to_cpu(p->uuid[i]); 3687b411b363SPhilipp Reisner 3688b30ab791SAndreas Gruenbacher kfree(device->p_uuid); 3689b30ab791SAndreas Gruenbacher device->p_uuid = p_uuid; 3690b411b363SPhilipp Reisner 3691b30ab791SAndreas Gruenbacher if (device->state.conn < C_CONNECTED && 3692b30ab791SAndreas Gruenbacher device->state.disk < D_INCONSISTENT && 3693b30ab791SAndreas Gruenbacher device->state.role == R_PRIMARY && 3694b30ab791SAndreas Gruenbacher (device->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { 3695d0180171SAndreas Gruenbacher drbd_err(device, "Can only connect to data with current UUID=%016llX\n", 3696b30ab791SAndreas Gruenbacher (unsigned long long)device->ed_uuid); 3697a6b32bc3SAndreas Gruenbacher conn_request_state(first_peer_device(device)->connection, NS(conn, C_DISCONNECTING), CS_HARD); 369882bc0194SAndreas Gruenbacher return -EIO; 3699b411b363SPhilipp Reisner } 3700b411b363SPhilipp Reisner 3701b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 3702b411b363SPhilipp Reisner int skip_initial_sync = 3703b30ab791SAndreas Gruenbacher device->state.conn == C_CONNECTED && 3704a6b32bc3SAndreas Gruenbacher first_peer_device(device)->connection->agreed_pro_version >= 90 && 3705b30ab791SAndreas Gruenbacher device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && 3706b411b363SPhilipp Reisner (p_uuid[UI_FLAGS] & 8); 3707b411b363SPhilipp Reisner if (skip_initial_sync) { 3708d0180171SAndreas Gruenbacher drbd_info(device, "Accepted new current UUID, preparing to skip initial sync\n"); 3709b30ab791SAndreas Gruenbacher drbd_bitmap_io(device, &drbd_bmio_clear_n_write, 371020ceb2b2SLars Ellenberg "clear_n_write from receive_uuids", 371120ceb2b2SLars Ellenberg BM_LOCKED_TEST_ALLOWED); 3712b30ab791SAndreas Gruenbacher _drbd_uuid_set(device, UI_CURRENT, p_uuid[UI_CURRENT]); 3713b30ab791SAndreas Gruenbacher _drbd_uuid_set(device, UI_BITMAP, 0); 3714b30ab791SAndreas Gruenbacher _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), 3715b411b363SPhilipp Reisner CS_VERBOSE, NULL); 3716b30ab791SAndreas Gruenbacher drbd_md_sync(device); 371762b0da3aSLars Ellenberg updated_uuids = 1; 3718b411b363SPhilipp Reisner } 3719b30ab791SAndreas Gruenbacher put_ldev(device); 3720b30ab791SAndreas Gruenbacher } else if (device->state.disk < D_INCONSISTENT && 3721b30ab791SAndreas Gruenbacher device->state.role == R_PRIMARY) { 372218a50fa2SPhilipp Reisner /* I am a diskless primary, the peer just created a new current UUID 372318a50fa2SPhilipp Reisner for me. */ 3724b30ab791SAndreas Gruenbacher updated_uuids = drbd_set_ed_uuid(device, p_uuid[UI_CURRENT]); 3725b411b363SPhilipp Reisner } 3726b411b363SPhilipp Reisner 3727b411b363SPhilipp Reisner /* Before we test for the disk state, we should wait until an eventually 3728b411b363SPhilipp Reisner ongoing cluster wide state change is finished. That is important if 3729b411b363SPhilipp Reisner we are primary and are detaching from our disk. We need to see the 3730b411b363SPhilipp Reisner new disk state... */ 3731b30ab791SAndreas Gruenbacher mutex_lock(device->state_mutex); 3732b30ab791SAndreas Gruenbacher mutex_unlock(device->state_mutex); 3733b30ab791SAndreas Gruenbacher if (device->state.conn >= C_CONNECTED && device->state.disk < D_INCONSISTENT) 3734b30ab791SAndreas Gruenbacher updated_uuids |= drbd_set_ed_uuid(device, p_uuid[UI_CURRENT]); 373562b0da3aSLars Ellenberg 373662b0da3aSLars Ellenberg if (updated_uuids) 3737b30ab791SAndreas Gruenbacher drbd_print_uuids(device, "receiver updated UUIDs to"); 3738b411b363SPhilipp Reisner 373982bc0194SAndreas Gruenbacher return 0; 3740b411b363SPhilipp Reisner } 3741b411b363SPhilipp Reisner 3742b411b363SPhilipp Reisner /** 3743b411b363SPhilipp Reisner * convert_state() - Converts the peer's view of the cluster state to our point of view 3744b411b363SPhilipp Reisner * @ps: The state as seen by the peer. 3745b411b363SPhilipp Reisner */ 3746b411b363SPhilipp Reisner static union drbd_state convert_state(union drbd_state ps) 3747b411b363SPhilipp Reisner { 3748b411b363SPhilipp Reisner union drbd_state ms; 3749b411b363SPhilipp Reisner 3750b411b363SPhilipp Reisner static enum drbd_conns c_tab[] = { 3751369bea63SPhilipp Reisner [C_WF_REPORT_PARAMS] = C_WF_REPORT_PARAMS, 3752b411b363SPhilipp Reisner [C_CONNECTED] = C_CONNECTED, 3753b411b363SPhilipp Reisner 3754b411b363SPhilipp Reisner [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, 3755b411b363SPhilipp Reisner [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, 3756b411b363SPhilipp Reisner [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ 3757b411b363SPhilipp Reisner [C_VERIFY_S] = C_VERIFY_T, 3758b411b363SPhilipp Reisner [C_MASK] = C_MASK, 3759b411b363SPhilipp Reisner }; 3760b411b363SPhilipp Reisner 3761b411b363SPhilipp Reisner ms.i = ps.i; 3762b411b363SPhilipp Reisner 3763b411b363SPhilipp Reisner ms.conn = c_tab[ps.conn]; 3764b411b363SPhilipp Reisner ms.peer = ps.role; 3765b411b363SPhilipp Reisner ms.role = ps.peer; 3766b411b363SPhilipp Reisner ms.pdsk = ps.disk; 3767b411b363SPhilipp Reisner ms.disk = ps.pdsk; 3768b411b363SPhilipp Reisner ms.peer_isp = (ps.aftr_isp | ps.user_isp); 3769b411b363SPhilipp Reisner 3770b411b363SPhilipp Reisner return ms; 3771b411b363SPhilipp Reisner } 3772b411b363SPhilipp Reisner 3773bde89a9eSAndreas Gruenbacher static int receive_req_state(struct drbd_connection *connection, struct packet_info *pi) 3774b411b363SPhilipp Reisner { 3775b30ab791SAndreas Gruenbacher struct drbd_device *device; 3776e658983aSAndreas Gruenbacher struct p_req_state *p = pi->data; 3777b411b363SPhilipp Reisner union drbd_state mask, val; 3778bf885f8aSAndreas Gruenbacher enum drbd_state_rv rv; 3779b411b363SPhilipp Reisner 3780bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 3781b30ab791SAndreas Gruenbacher if (!device) 37824a76b161SAndreas Gruenbacher return -EIO; 37834a76b161SAndreas Gruenbacher 3784b411b363SPhilipp Reisner mask.i = be32_to_cpu(p->mask); 3785b411b363SPhilipp Reisner val.i = be32_to_cpu(p->val); 3786b411b363SPhilipp Reisner 3787a6b32bc3SAndreas Gruenbacher if (test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags) && 3788b30ab791SAndreas Gruenbacher mutex_is_locked(device->state_mutex)) { 3789b30ab791SAndreas Gruenbacher drbd_send_sr_reply(device, SS_CONCURRENT_ST_CHG); 379082bc0194SAndreas Gruenbacher return 0; 3791b411b363SPhilipp Reisner } 3792b411b363SPhilipp Reisner 3793b411b363SPhilipp Reisner mask = convert_state(mask); 3794b411b363SPhilipp Reisner val = convert_state(val); 3795b411b363SPhilipp Reisner 3796b30ab791SAndreas Gruenbacher rv = drbd_change_state(device, CS_VERBOSE, mask, val); 3797b30ab791SAndreas Gruenbacher drbd_send_sr_reply(device, rv); 3798047cd4a6SPhilipp Reisner 3799b30ab791SAndreas Gruenbacher drbd_md_sync(device); 3800b411b363SPhilipp Reisner 380182bc0194SAndreas Gruenbacher return 0; 3802b411b363SPhilipp Reisner } 3803b411b363SPhilipp Reisner 3804bde89a9eSAndreas Gruenbacher static int receive_req_conn_state(struct drbd_connection *connection, struct packet_info *pi) 3805b411b363SPhilipp Reisner { 3806e658983aSAndreas Gruenbacher struct p_req_state *p = pi->data; 3807dfafcc8aSPhilipp Reisner union drbd_state mask, val; 3808dfafcc8aSPhilipp Reisner enum drbd_state_rv rv; 3809dfafcc8aSPhilipp Reisner 3810dfafcc8aSPhilipp Reisner mask.i = be32_to_cpu(p->mask); 3811dfafcc8aSPhilipp Reisner val.i = be32_to_cpu(p->val); 3812dfafcc8aSPhilipp Reisner 3813bde89a9eSAndreas Gruenbacher if (test_bit(RESOLVE_CONFLICTS, &connection->flags) && 3814bde89a9eSAndreas Gruenbacher mutex_is_locked(&connection->cstate_mutex)) { 3815bde89a9eSAndreas Gruenbacher conn_send_sr_reply(connection, SS_CONCURRENT_ST_CHG); 381682bc0194SAndreas Gruenbacher return 0; 3817dfafcc8aSPhilipp Reisner } 3818dfafcc8aSPhilipp Reisner 3819dfafcc8aSPhilipp Reisner mask = convert_state(mask); 3820dfafcc8aSPhilipp Reisner val = convert_state(val); 3821dfafcc8aSPhilipp Reisner 3822bde89a9eSAndreas Gruenbacher rv = conn_request_state(connection, mask, val, CS_VERBOSE | CS_LOCAL_ONLY | CS_IGN_OUTD_FAIL); 3823bde89a9eSAndreas Gruenbacher conn_send_sr_reply(connection, rv); 3824dfafcc8aSPhilipp Reisner 382582bc0194SAndreas Gruenbacher return 0; 3826dfafcc8aSPhilipp Reisner } 3827dfafcc8aSPhilipp Reisner 3828bde89a9eSAndreas Gruenbacher static int receive_state(struct drbd_connection *connection, struct packet_info *pi) 3829b411b363SPhilipp Reisner { 3830b30ab791SAndreas Gruenbacher struct drbd_device *device; 3831e658983aSAndreas Gruenbacher struct p_state *p = pi->data; 38324ac4aadaSLars Ellenberg union drbd_state os, ns, peer_state; 3833b411b363SPhilipp Reisner enum drbd_disk_state real_peer_disk; 383465d922c3SPhilipp Reisner enum chg_state_flags cs_flags; 3835b411b363SPhilipp Reisner int rv; 3836b411b363SPhilipp Reisner 3837bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 3838b30ab791SAndreas Gruenbacher if (!device) 3839bde89a9eSAndreas Gruenbacher return config_unknown_volume(connection, pi); 38404a76b161SAndreas Gruenbacher 3841b411b363SPhilipp Reisner peer_state.i = be32_to_cpu(p->state); 3842b411b363SPhilipp Reisner 3843b411b363SPhilipp Reisner real_peer_disk = peer_state.disk; 3844b411b363SPhilipp Reisner if (peer_state.disk == D_NEGOTIATING) { 3845b30ab791SAndreas Gruenbacher real_peer_disk = device->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; 3846d0180171SAndreas Gruenbacher drbd_info(device, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); 3847b411b363SPhilipp Reisner } 3848b411b363SPhilipp Reisner 38490500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 3850b411b363SPhilipp Reisner retry: 3851b30ab791SAndreas Gruenbacher os = ns = drbd_read_state(device); 38520500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 3853b411b363SPhilipp Reisner 3854545752d5SLars Ellenberg /* If some other part of the code (asender thread, timeout) 3855545752d5SLars Ellenberg * already decided to close the connection again, 3856545752d5SLars Ellenberg * we must not "re-establish" it here. */ 3857545752d5SLars Ellenberg if (os.conn <= C_TEAR_DOWN) 385858ffa580SLars Ellenberg return -ECONNRESET; 3859545752d5SLars Ellenberg 386040424e4aSLars Ellenberg /* If this is the "end of sync" confirmation, usually the peer disk 386140424e4aSLars Ellenberg * transitions from D_INCONSISTENT to D_UP_TO_DATE. For empty (0 bits 386240424e4aSLars Ellenberg * set) resync started in PausedSyncT, or if the timing of pause-/ 386340424e4aSLars Ellenberg * unpause-sync events has been "just right", the peer disk may 386440424e4aSLars Ellenberg * transition from D_CONSISTENT to D_UP_TO_DATE as well. 386540424e4aSLars Ellenberg */ 386640424e4aSLars Ellenberg if ((os.pdsk == D_INCONSISTENT || os.pdsk == D_CONSISTENT) && 386740424e4aSLars Ellenberg real_peer_disk == D_UP_TO_DATE && 3868e9ef7bb6SLars Ellenberg os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { 3869e9ef7bb6SLars Ellenberg /* If we are (becoming) SyncSource, but peer is still in sync 3870e9ef7bb6SLars Ellenberg * preparation, ignore its uptodate-ness to avoid flapping, it 3871e9ef7bb6SLars Ellenberg * will change to inconsistent once the peer reaches active 3872e9ef7bb6SLars Ellenberg * syncing states. 3873e9ef7bb6SLars Ellenberg * It may have changed syncer-paused flags, however, so we 3874e9ef7bb6SLars Ellenberg * cannot ignore this completely. */ 3875e9ef7bb6SLars Ellenberg if (peer_state.conn > C_CONNECTED && 3876e9ef7bb6SLars Ellenberg peer_state.conn < C_SYNC_SOURCE) 3877e9ef7bb6SLars Ellenberg real_peer_disk = D_INCONSISTENT; 3878e9ef7bb6SLars Ellenberg 3879e9ef7bb6SLars Ellenberg /* if peer_state changes to connected at the same time, 3880e9ef7bb6SLars Ellenberg * it explicitly notifies us that it finished resync. 3881e9ef7bb6SLars Ellenberg * Maybe we should finish it up, too? */ 3882e9ef7bb6SLars Ellenberg else if (os.conn >= C_SYNC_SOURCE && 3883e9ef7bb6SLars Ellenberg peer_state.conn == C_CONNECTED) { 3884b30ab791SAndreas Gruenbacher if (drbd_bm_total_weight(device) <= device->rs_failed) 3885b30ab791SAndreas Gruenbacher drbd_resync_finished(device); 388682bc0194SAndreas Gruenbacher return 0; 3887e9ef7bb6SLars Ellenberg } 3888e9ef7bb6SLars Ellenberg } 3889e9ef7bb6SLars Ellenberg 389002b91b55SLars Ellenberg /* explicit verify finished notification, stop sector reached. */ 389102b91b55SLars Ellenberg if (os.conn == C_VERIFY_T && os.disk == D_UP_TO_DATE && 389202b91b55SLars Ellenberg peer_state.conn == C_CONNECTED && real_peer_disk == D_UP_TO_DATE) { 3893b30ab791SAndreas Gruenbacher ov_out_of_sync_print(device); 3894b30ab791SAndreas Gruenbacher drbd_resync_finished(device); 389558ffa580SLars Ellenberg return 0; 389602b91b55SLars Ellenberg } 389702b91b55SLars Ellenberg 3898e9ef7bb6SLars Ellenberg /* peer says his disk is inconsistent, while we think it is uptodate, 3899e9ef7bb6SLars Ellenberg * and this happens while the peer still thinks we have a sync going on, 3900e9ef7bb6SLars Ellenberg * but we think we are already done with the sync. 3901e9ef7bb6SLars Ellenberg * We ignore this to avoid flapping pdsk. 3902e9ef7bb6SLars Ellenberg * This should not happen, if the peer is a recent version of drbd. */ 3903e9ef7bb6SLars Ellenberg if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && 3904e9ef7bb6SLars Ellenberg os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) 3905e9ef7bb6SLars Ellenberg real_peer_disk = D_UP_TO_DATE; 3906e9ef7bb6SLars Ellenberg 39074ac4aadaSLars Ellenberg if (ns.conn == C_WF_REPORT_PARAMS) 39084ac4aadaSLars Ellenberg ns.conn = C_CONNECTED; 3909b411b363SPhilipp Reisner 391067531718SPhilipp Reisner if (peer_state.conn == C_AHEAD) 391167531718SPhilipp Reisner ns.conn = C_BEHIND; 391267531718SPhilipp Reisner 3913b30ab791SAndreas Gruenbacher if (device->p_uuid && peer_state.disk >= D_NEGOTIATING && 3914b30ab791SAndreas Gruenbacher get_ldev_if_state(device, D_NEGOTIATING)) { 3915b411b363SPhilipp Reisner int cr; /* consider resync */ 3916b411b363SPhilipp Reisner 3917b411b363SPhilipp Reisner /* if we established a new connection */ 39184ac4aadaSLars Ellenberg cr = (os.conn < C_CONNECTED); 3919b411b363SPhilipp Reisner /* if we had an established connection 3920b411b363SPhilipp Reisner * and one of the nodes newly attaches a disk */ 39214ac4aadaSLars Ellenberg cr |= (os.conn == C_CONNECTED && 3922b411b363SPhilipp Reisner (peer_state.disk == D_NEGOTIATING || 39234ac4aadaSLars Ellenberg os.disk == D_NEGOTIATING)); 3924b411b363SPhilipp Reisner /* if we have both been inconsistent, and the peer has been 3925b411b363SPhilipp Reisner * forced to be UpToDate with --overwrite-data */ 3926b30ab791SAndreas Gruenbacher cr |= test_bit(CONSIDER_RESYNC, &device->flags); 3927b411b363SPhilipp Reisner /* if we had been plain connected, and the admin requested to 3928b411b363SPhilipp Reisner * start a sync by "invalidate" or "invalidate-remote" */ 39294ac4aadaSLars Ellenberg cr |= (os.conn == C_CONNECTED && 3930b411b363SPhilipp Reisner (peer_state.conn >= C_STARTING_SYNC_S && 3931b411b363SPhilipp Reisner peer_state.conn <= C_WF_BITMAP_T)); 3932b411b363SPhilipp Reisner 3933b411b363SPhilipp Reisner if (cr) 3934b30ab791SAndreas Gruenbacher ns.conn = drbd_sync_handshake(device, peer_state.role, real_peer_disk); 3935b411b363SPhilipp Reisner 3936b30ab791SAndreas Gruenbacher put_ldev(device); 39374ac4aadaSLars Ellenberg if (ns.conn == C_MASK) { 39384ac4aadaSLars Ellenberg ns.conn = C_CONNECTED; 3939b30ab791SAndreas Gruenbacher if (device->state.disk == D_NEGOTIATING) { 3940b30ab791SAndreas Gruenbacher drbd_force_state(device, NS(disk, D_FAILED)); 3941b411b363SPhilipp Reisner } else if (peer_state.disk == D_NEGOTIATING) { 3942d0180171SAndreas Gruenbacher drbd_err(device, "Disk attach process on the peer node was aborted.\n"); 3943b411b363SPhilipp Reisner peer_state.disk = D_DISKLESS; 3944580b9767SLars Ellenberg real_peer_disk = D_DISKLESS; 3945b411b363SPhilipp Reisner } else { 3946a6b32bc3SAndreas Gruenbacher if (test_and_clear_bit(CONN_DRY_RUN, &first_peer_device(device)->connection->flags)) 394782bc0194SAndreas Gruenbacher return -EIO; 39480b0ba1efSAndreas Gruenbacher D_ASSERT(device, os.conn == C_WF_REPORT_PARAMS); 3949a6b32bc3SAndreas Gruenbacher conn_request_state(first_peer_device(device)->connection, NS(conn, C_DISCONNECTING), CS_HARD); 395082bc0194SAndreas Gruenbacher return -EIO; 3951b411b363SPhilipp Reisner } 3952b411b363SPhilipp Reisner } 3953b411b363SPhilipp Reisner } 3954b411b363SPhilipp Reisner 39550500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 3956b30ab791SAndreas Gruenbacher if (os.i != drbd_read_state(device).i) 3957b411b363SPhilipp Reisner goto retry; 3958b30ab791SAndreas Gruenbacher clear_bit(CONSIDER_RESYNC, &device->flags); 3959b411b363SPhilipp Reisner ns.peer = peer_state.role; 3960b411b363SPhilipp Reisner ns.pdsk = real_peer_disk; 3961b411b363SPhilipp Reisner ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); 39624ac4aadaSLars Ellenberg if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) 3963b30ab791SAndreas Gruenbacher ns.disk = device->new_state_tmp.disk; 39644ac4aadaSLars Ellenberg cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); 3965b30ab791SAndreas Gruenbacher if (ns.pdsk == D_CONSISTENT && drbd_suspended(device) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && 3966b30ab791SAndreas Gruenbacher test_bit(NEW_CUR_UUID, &device->flags)) { 39678554df1cSAndreas Gruenbacher /* Do not allow tl_restart(RESEND) for a rebooted peer. We can only allow this 3968481c6f50SPhilipp Reisner for temporal network outages! */ 39690500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 3970d0180171SAndreas Gruenbacher drbd_err(device, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); 3971a6b32bc3SAndreas Gruenbacher tl_clear(first_peer_device(device)->connection); 3972b30ab791SAndreas Gruenbacher drbd_uuid_new_current(device); 3973b30ab791SAndreas Gruenbacher clear_bit(NEW_CUR_UUID, &device->flags); 3974a6b32bc3SAndreas Gruenbacher conn_request_state(first_peer_device(device)->connection, NS2(conn, C_PROTOCOL_ERROR, susp, 0), CS_HARD); 397582bc0194SAndreas Gruenbacher return -EIO; 3976481c6f50SPhilipp Reisner } 3977b30ab791SAndreas Gruenbacher rv = _drbd_set_state(device, ns, cs_flags, NULL); 3978b30ab791SAndreas Gruenbacher ns = drbd_read_state(device); 39790500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 3980b411b363SPhilipp Reisner 3981b411b363SPhilipp Reisner if (rv < SS_SUCCESS) { 3982a6b32bc3SAndreas Gruenbacher conn_request_state(first_peer_device(device)->connection, NS(conn, C_DISCONNECTING), CS_HARD); 398382bc0194SAndreas Gruenbacher return -EIO; 3984b411b363SPhilipp Reisner } 3985b411b363SPhilipp Reisner 39864ac4aadaSLars Ellenberg if (os.conn > C_WF_REPORT_PARAMS) { 39874ac4aadaSLars Ellenberg if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && 3988b411b363SPhilipp Reisner peer_state.disk != D_NEGOTIATING ) { 3989b411b363SPhilipp Reisner /* we want resync, peer has not yet decided to sync... */ 3990b411b363SPhilipp Reisner /* Nowadays only used when forcing a node into primary role and 3991b411b363SPhilipp Reisner setting its disk to UpToDate with that */ 3992b30ab791SAndreas Gruenbacher drbd_send_uuids(device); 3993b30ab791SAndreas Gruenbacher drbd_send_current_state(device); 3994b411b363SPhilipp Reisner } 3995b411b363SPhilipp Reisner } 3996b411b363SPhilipp Reisner 3997b30ab791SAndreas Gruenbacher clear_bit(DISCARD_MY_DATA, &device->flags); 3998b411b363SPhilipp Reisner 3999b30ab791SAndreas Gruenbacher drbd_md_sync(device); /* update connected indicator, la_size_sect, ... */ 4000b411b363SPhilipp Reisner 400182bc0194SAndreas Gruenbacher return 0; 4002b411b363SPhilipp Reisner } 4003b411b363SPhilipp Reisner 4004bde89a9eSAndreas Gruenbacher static int receive_sync_uuid(struct drbd_connection *connection, struct packet_info *pi) 4005b411b363SPhilipp Reisner { 4006b30ab791SAndreas Gruenbacher struct drbd_device *device; 4007e658983aSAndreas Gruenbacher struct p_rs_uuid *p = pi->data; 40084a76b161SAndreas Gruenbacher 4009bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 4010b30ab791SAndreas Gruenbacher if (!device) 40114a76b161SAndreas Gruenbacher return -EIO; 4012b411b363SPhilipp Reisner 4013b30ab791SAndreas Gruenbacher wait_event(device->misc_wait, 4014b30ab791SAndreas Gruenbacher device->state.conn == C_WF_SYNC_UUID || 4015b30ab791SAndreas Gruenbacher device->state.conn == C_BEHIND || 4016b30ab791SAndreas Gruenbacher device->state.conn < C_CONNECTED || 4017b30ab791SAndreas Gruenbacher device->state.disk < D_NEGOTIATING); 4018b411b363SPhilipp Reisner 40190b0ba1efSAndreas Gruenbacher /* D_ASSERT(device, device->state.conn == C_WF_SYNC_UUID ); */ 4020b411b363SPhilipp Reisner 4021b411b363SPhilipp Reisner /* Here the _drbd_uuid_ functions are right, current should 4022b411b363SPhilipp Reisner _not_ be rotated into the history */ 4023b30ab791SAndreas Gruenbacher if (get_ldev_if_state(device, D_NEGOTIATING)) { 4024b30ab791SAndreas Gruenbacher _drbd_uuid_set(device, UI_CURRENT, be64_to_cpu(p->uuid)); 4025b30ab791SAndreas Gruenbacher _drbd_uuid_set(device, UI_BITMAP, 0UL); 4026b411b363SPhilipp Reisner 4027b30ab791SAndreas Gruenbacher drbd_print_uuids(device, "updated sync uuid"); 4028b30ab791SAndreas Gruenbacher drbd_start_resync(device, C_SYNC_TARGET); 4029b411b363SPhilipp Reisner 4030b30ab791SAndreas Gruenbacher put_ldev(device); 4031b411b363SPhilipp Reisner } else 4032d0180171SAndreas Gruenbacher drbd_err(device, "Ignoring SyncUUID packet!\n"); 4033b411b363SPhilipp Reisner 403482bc0194SAndreas Gruenbacher return 0; 4035b411b363SPhilipp Reisner } 4036b411b363SPhilipp Reisner 40372c46407dSAndreas Gruenbacher /** 40382c46407dSAndreas Gruenbacher * receive_bitmap_plain 40392c46407dSAndreas Gruenbacher * 40402c46407dSAndreas Gruenbacher * Return 0 when done, 1 when another iteration is needed, and a negative error 40412c46407dSAndreas Gruenbacher * code upon failure. 40422c46407dSAndreas Gruenbacher */ 40432c46407dSAndreas Gruenbacher static int 4044b30ab791SAndreas Gruenbacher receive_bitmap_plain(struct drbd_device *device, unsigned int size, 4045e658983aSAndreas Gruenbacher unsigned long *p, struct bm_xfer_ctx *c) 4046b411b363SPhilipp Reisner { 404750d0b1adSAndreas Gruenbacher unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - 4048a6b32bc3SAndreas Gruenbacher drbd_header_size(first_peer_device(device)->connection); 4049e658983aSAndreas Gruenbacher unsigned int num_words = min_t(size_t, data_size / sizeof(*p), 405050d0b1adSAndreas Gruenbacher c->bm_words - c->word_offset); 4051e658983aSAndreas Gruenbacher unsigned int want = num_words * sizeof(*p); 40522c46407dSAndreas Gruenbacher int err; 4053b411b363SPhilipp Reisner 405450d0b1adSAndreas Gruenbacher if (want != size) { 4055d0180171SAndreas Gruenbacher drbd_err(device, "%s:want (%u) != size (%u)\n", __func__, want, size); 40562c46407dSAndreas Gruenbacher return -EIO; 4057b411b363SPhilipp Reisner } 4058b411b363SPhilipp Reisner if (want == 0) 40592c46407dSAndreas Gruenbacher return 0; 4060a6b32bc3SAndreas Gruenbacher err = drbd_recv_all(first_peer_device(device)->connection, p, want); 406182bc0194SAndreas Gruenbacher if (err) 40622c46407dSAndreas Gruenbacher return err; 4063b411b363SPhilipp Reisner 4064b30ab791SAndreas Gruenbacher drbd_bm_merge_lel(device, c->word_offset, num_words, p); 4065b411b363SPhilipp Reisner 4066b411b363SPhilipp Reisner c->word_offset += num_words; 4067b411b363SPhilipp Reisner c->bit_offset = c->word_offset * BITS_PER_LONG; 4068b411b363SPhilipp Reisner if (c->bit_offset > c->bm_bits) 4069b411b363SPhilipp Reisner c->bit_offset = c->bm_bits; 4070b411b363SPhilipp Reisner 40712c46407dSAndreas Gruenbacher return 1; 4072b411b363SPhilipp Reisner } 4073b411b363SPhilipp Reisner 4074a02d1240SAndreas Gruenbacher static enum drbd_bitmap_code dcbp_get_code(struct p_compressed_bm *p) 4075a02d1240SAndreas Gruenbacher { 4076a02d1240SAndreas Gruenbacher return (enum drbd_bitmap_code)(p->encoding & 0x0f); 4077a02d1240SAndreas Gruenbacher } 4078a02d1240SAndreas Gruenbacher 4079a02d1240SAndreas Gruenbacher static int dcbp_get_start(struct p_compressed_bm *p) 4080a02d1240SAndreas Gruenbacher { 4081a02d1240SAndreas Gruenbacher return (p->encoding & 0x80) != 0; 4082a02d1240SAndreas Gruenbacher } 4083a02d1240SAndreas Gruenbacher 4084a02d1240SAndreas Gruenbacher static int dcbp_get_pad_bits(struct p_compressed_bm *p) 4085a02d1240SAndreas Gruenbacher { 4086a02d1240SAndreas Gruenbacher return (p->encoding >> 4) & 0x7; 4087a02d1240SAndreas Gruenbacher } 4088a02d1240SAndreas Gruenbacher 40892c46407dSAndreas Gruenbacher /** 40902c46407dSAndreas Gruenbacher * recv_bm_rle_bits 40912c46407dSAndreas Gruenbacher * 40922c46407dSAndreas Gruenbacher * Return 0 when done, 1 when another iteration is needed, and a negative error 40932c46407dSAndreas Gruenbacher * code upon failure. 40942c46407dSAndreas Gruenbacher */ 40952c46407dSAndreas Gruenbacher static int 4096b30ab791SAndreas Gruenbacher recv_bm_rle_bits(struct drbd_device *device, 4097b411b363SPhilipp Reisner struct p_compressed_bm *p, 4098c6d25cfeSPhilipp Reisner struct bm_xfer_ctx *c, 4099c6d25cfeSPhilipp Reisner unsigned int len) 4100b411b363SPhilipp Reisner { 4101b411b363SPhilipp Reisner struct bitstream bs; 4102b411b363SPhilipp Reisner u64 look_ahead; 4103b411b363SPhilipp Reisner u64 rl; 4104b411b363SPhilipp Reisner u64 tmp; 4105b411b363SPhilipp Reisner unsigned long s = c->bit_offset; 4106b411b363SPhilipp Reisner unsigned long e; 4107a02d1240SAndreas Gruenbacher int toggle = dcbp_get_start(p); 4108b411b363SPhilipp Reisner int have; 4109b411b363SPhilipp Reisner int bits; 4110b411b363SPhilipp Reisner 4111a02d1240SAndreas Gruenbacher bitstream_init(&bs, p->code, len, dcbp_get_pad_bits(p)); 4112b411b363SPhilipp Reisner 4113b411b363SPhilipp Reisner bits = bitstream_get_bits(&bs, &look_ahead, 64); 4114b411b363SPhilipp Reisner if (bits < 0) 41152c46407dSAndreas Gruenbacher return -EIO; 4116b411b363SPhilipp Reisner 4117b411b363SPhilipp Reisner for (have = bits; have > 0; s += rl, toggle = !toggle) { 4118b411b363SPhilipp Reisner bits = vli_decode_bits(&rl, look_ahead); 4119b411b363SPhilipp Reisner if (bits <= 0) 41202c46407dSAndreas Gruenbacher return -EIO; 4121b411b363SPhilipp Reisner 4122b411b363SPhilipp Reisner if (toggle) { 4123b411b363SPhilipp Reisner e = s + rl -1; 4124b411b363SPhilipp Reisner if (e >= c->bm_bits) { 4125d0180171SAndreas Gruenbacher drbd_err(device, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); 41262c46407dSAndreas Gruenbacher return -EIO; 4127b411b363SPhilipp Reisner } 4128b30ab791SAndreas Gruenbacher _drbd_bm_set_bits(device, s, e); 4129b411b363SPhilipp Reisner } 4130b411b363SPhilipp Reisner 4131b411b363SPhilipp Reisner if (have < bits) { 4132d0180171SAndreas Gruenbacher drbd_err(device, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", 4133b411b363SPhilipp Reisner have, bits, look_ahead, 4134b411b363SPhilipp Reisner (unsigned int)(bs.cur.b - p->code), 4135b411b363SPhilipp Reisner (unsigned int)bs.buf_len); 41362c46407dSAndreas Gruenbacher return -EIO; 4137b411b363SPhilipp Reisner } 4138d2da5b0cSLars Ellenberg /* if we consumed all 64 bits, assign 0; >> 64 is "undefined"; */ 4139d2da5b0cSLars Ellenberg if (likely(bits < 64)) 4140b411b363SPhilipp Reisner look_ahead >>= bits; 4141d2da5b0cSLars Ellenberg else 4142d2da5b0cSLars Ellenberg look_ahead = 0; 4143b411b363SPhilipp Reisner have -= bits; 4144b411b363SPhilipp Reisner 4145b411b363SPhilipp Reisner bits = bitstream_get_bits(&bs, &tmp, 64 - have); 4146b411b363SPhilipp Reisner if (bits < 0) 41472c46407dSAndreas Gruenbacher return -EIO; 4148b411b363SPhilipp Reisner look_ahead |= tmp << have; 4149b411b363SPhilipp Reisner have += bits; 4150b411b363SPhilipp Reisner } 4151b411b363SPhilipp Reisner 4152b411b363SPhilipp Reisner c->bit_offset = s; 4153b411b363SPhilipp Reisner bm_xfer_ctx_bit_to_word_offset(c); 4154b411b363SPhilipp Reisner 41552c46407dSAndreas Gruenbacher return (s != c->bm_bits); 4156b411b363SPhilipp Reisner } 4157b411b363SPhilipp Reisner 41582c46407dSAndreas Gruenbacher /** 41592c46407dSAndreas Gruenbacher * decode_bitmap_c 41602c46407dSAndreas Gruenbacher * 41612c46407dSAndreas Gruenbacher * Return 0 when done, 1 when another iteration is needed, and a negative error 41622c46407dSAndreas Gruenbacher * code upon failure. 41632c46407dSAndreas Gruenbacher */ 41642c46407dSAndreas Gruenbacher static int 4165b30ab791SAndreas Gruenbacher decode_bitmap_c(struct drbd_device *device, 4166b411b363SPhilipp Reisner struct p_compressed_bm *p, 4167c6d25cfeSPhilipp Reisner struct bm_xfer_ctx *c, 4168c6d25cfeSPhilipp Reisner unsigned int len) 4169b411b363SPhilipp Reisner { 4170a02d1240SAndreas Gruenbacher if (dcbp_get_code(p) == RLE_VLI_Bits) 4171b30ab791SAndreas Gruenbacher return recv_bm_rle_bits(device, p, c, len - sizeof(*p)); 4172b411b363SPhilipp Reisner 4173b411b363SPhilipp Reisner /* other variants had been implemented for evaluation, 4174b411b363SPhilipp Reisner * but have been dropped as this one turned out to be "best" 4175b411b363SPhilipp Reisner * during all our tests. */ 4176b411b363SPhilipp Reisner 4177d0180171SAndreas Gruenbacher drbd_err(device, "receive_bitmap_c: unknown encoding %u\n", p->encoding); 4178a6b32bc3SAndreas Gruenbacher conn_request_state(first_peer_device(device)->connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); 41792c46407dSAndreas Gruenbacher return -EIO; 4180b411b363SPhilipp Reisner } 4181b411b363SPhilipp Reisner 4182b30ab791SAndreas Gruenbacher void INFO_bm_xfer_stats(struct drbd_device *device, 4183b411b363SPhilipp Reisner const char *direction, struct bm_xfer_ctx *c) 4184b411b363SPhilipp Reisner { 4185b411b363SPhilipp Reisner /* what would it take to transfer it "plaintext" */ 4186a6b32bc3SAndreas Gruenbacher unsigned int header_size = drbd_header_size(first_peer_device(device)->connection); 418750d0b1adSAndreas Gruenbacher unsigned int data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; 418850d0b1adSAndreas Gruenbacher unsigned int plain = 418950d0b1adSAndreas Gruenbacher header_size * (DIV_ROUND_UP(c->bm_words, data_size) + 1) + 419050d0b1adSAndreas Gruenbacher c->bm_words * sizeof(unsigned long); 419150d0b1adSAndreas Gruenbacher unsigned int total = c->bytes[0] + c->bytes[1]; 419250d0b1adSAndreas Gruenbacher unsigned int r; 4193b411b363SPhilipp Reisner 4194b411b363SPhilipp Reisner /* total can not be zero. but just in case: */ 4195b411b363SPhilipp Reisner if (total == 0) 4196b411b363SPhilipp Reisner return; 4197b411b363SPhilipp Reisner 4198b411b363SPhilipp Reisner /* don't report if not compressed */ 4199b411b363SPhilipp Reisner if (total >= plain) 4200b411b363SPhilipp Reisner return; 4201b411b363SPhilipp Reisner 4202b411b363SPhilipp Reisner /* total < plain. check for overflow, still */ 4203b411b363SPhilipp Reisner r = (total > UINT_MAX/1000) ? (total / (plain/1000)) 4204b411b363SPhilipp Reisner : (1000 * total / plain); 4205b411b363SPhilipp Reisner 4206b411b363SPhilipp Reisner if (r > 1000) 4207b411b363SPhilipp Reisner r = 1000; 4208b411b363SPhilipp Reisner 4209b411b363SPhilipp Reisner r = 1000 - r; 4210d0180171SAndreas Gruenbacher drbd_info(device, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " 4211b411b363SPhilipp Reisner "total %u; compression: %u.%u%%\n", 4212b411b363SPhilipp Reisner direction, 4213b411b363SPhilipp Reisner c->bytes[1], c->packets[1], 4214b411b363SPhilipp Reisner c->bytes[0], c->packets[0], 4215b411b363SPhilipp Reisner total, r/10, r % 10); 4216b411b363SPhilipp Reisner } 4217b411b363SPhilipp Reisner 4218b411b363SPhilipp Reisner /* Since we are processing the bitfield from lower addresses to higher, 4219b411b363SPhilipp Reisner it does not matter if the process it in 32 bit chunks or 64 bit 4220b411b363SPhilipp Reisner chunks as long as it is little endian. (Understand it as byte stream, 4221b411b363SPhilipp Reisner beginning with the lowest byte...) If we would use big endian 4222b411b363SPhilipp Reisner we would need to process it from the highest address to the lowest, 4223b411b363SPhilipp Reisner in order to be agnostic to the 32 vs 64 bits issue. 4224b411b363SPhilipp Reisner 4225b411b363SPhilipp Reisner returns 0 on failure, 1 if we successfully received it. */ 4226bde89a9eSAndreas Gruenbacher static int receive_bitmap(struct drbd_connection *connection, struct packet_info *pi) 4227b411b363SPhilipp Reisner { 4228b30ab791SAndreas Gruenbacher struct drbd_device *device; 4229b411b363SPhilipp Reisner struct bm_xfer_ctx c; 42302c46407dSAndreas Gruenbacher int err; 42314a76b161SAndreas Gruenbacher 4232bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 4233b30ab791SAndreas Gruenbacher if (!device) 42344a76b161SAndreas Gruenbacher return -EIO; 4235b411b363SPhilipp Reisner 4236b30ab791SAndreas Gruenbacher drbd_bm_lock(device, "receive bitmap", BM_LOCKED_SET_ALLOWED); 423720ceb2b2SLars Ellenberg /* you are supposed to send additional out-of-sync information 423820ceb2b2SLars Ellenberg * if you actually set bits during this phase */ 4239b411b363SPhilipp Reisner 4240b411b363SPhilipp Reisner c = (struct bm_xfer_ctx) { 4241b30ab791SAndreas Gruenbacher .bm_bits = drbd_bm_bits(device), 4242b30ab791SAndreas Gruenbacher .bm_words = drbd_bm_words(device), 4243b411b363SPhilipp Reisner }; 4244b411b363SPhilipp Reisner 42452c46407dSAndreas Gruenbacher for(;;) { 4246e658983aSAndreas Gruenbacher if (pi->cmd == P_BITMAP) 4247b30ab791SAndreas Gruenbacher err = receive_bitmap_plain(device, pi->size, pi->data, &c); 4248e658983aSAndreas Gruenbacher else if (pi->cmd == P_COMPRESSED_BITMAP) { 4249b411b363SPhilipp Reisner /* MAYBE: sanity check that we speak proto >= 90, 4250b411b363SPhilipp Reisner * and the feature is enabled! */ 4251e658983aSAndreas Gruenbacher struct p_compressed_bm *p = pi->data; 4252b411b363SPhilipp Reisner 4253bde89a9eSAndreas Gruenbacher if (pi->size > DRBD_SOCKET_BUFFER_SIZE - drbd_header_size(connection)) { 4254d0180171SAndreas Gruenbacher drbd_err(device, "ReportCBitmap packet too large\n"); 425582bc0194SAndreas Gruenbacher err = -EIO; 4256b411b363SPhilipp Reisner goto out; 4257b411b363SPhilipp Reisner } 4258e658983aSAndreas Gruenbacher if (pi->size <= sizeof(*p)) { 4259d0180171SAndreas Gruenbacher drbd_err(device, "ReportCBitmap packet too small (l:%u)\n", pi->size); 426082bc0194SAndreas Gruenbacher err = -EIO; 426178fcbdaeSAndreas Gruenbacher goto out; 4262b411b363SPhilipp Reisner } 4263a6b32bc3SAndreas Gruenbacher err = drbd_recv_all(first_peer_device(device)->connection, p, pi->size); 4264e658983aSAndreas Gruenbacher if (err) 4265e658983aSAndreas Gruenbacher goto out; 4266b30ab791SAndreas Gruenbacher err = decode_bitmap_c(device, p, &c, pi->size); 4267b411b363SPhilipp Reisner } else { 4268d0180171SAndreas Gruenbacher drbd_warn(device, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", pi->cmd); 426982bc0194SAndreas Gruenbacher err = -EIO; 4270b411b363SPhilipp Reisner goto out; 4271b411b363SPhilipp Reisner } 4272b411b363SPhilipp Reisner 4273e2857216SAndreas Gruenbacher c.packets[pi->cmd == P_BITMAP]++; 4274bde89a9eSAndreas Gruenbacher c.bytes[pi->cmd == P_BITMAP] += drbd_header_size(connection) + pi->size; 4275b411b363SPhilipp Reisner 42762c46407dSAndreas Gruenbacher if (err <= 0) { 42772c46407dSAndreas Gruenbacher if (err < 0) 42782c46407dSAndreas Gruenbacher goto out; 4279b411b363SPhilipp Reisner break; 42802c46407dSAndreas Gruenbacher } 4281a6b32bc3SAndreas Gruenbacher err = drbd_recv_header(first_peer_device(device)->connection, pi); 428282bc0194SAndreas Gruenbacher if (err) 4283b411b363SPhilipp Reisner goto out; 42842c46407dSAndreas Gruenbacher } 4285b411b363SPhilipp Reisner 4286b30ab791SAndreas Gruenbacher INFO_bm_xfer_stats(device, "receive", &c); 4287b411b363SPhilipp Reisner 4288b30ab791SAndreas Gruenbacher if (device->state.conn == C_WF_BITMAP_T) { 4289de1f8e4aSAndreas Gruenbacher enum drbd_state_rv rv; 4290de1f8e4aSAndreas Gruenbacher 4291b30ab791SAndreas Gruenbacher err = drbd_send_bitmap(device); 429282bc0194SAndreas Gruenbacher if (err) 4293b411b363SPhilipp Reisner goto out; 4294b411b363SPhilipp Reisner /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ 4295b30ab791SAndreas Gruenbacher rv = _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); 42960b0ba1efSAndreas Gruenbacher D_ASSERT(device, rv == SS_SUCCESS); 4297b30ab791SAndreas Gruenbacher } else if (device->state.conn != C_WF_BITMAP_S) { 4298b411b363SPhilipp Reisner /* admin may have requested C_DISCONNECTING, 4299b411b363SPhilipp Reisner * other threads may have noticed network errors */ 4300d0180171SAndreas Gruenbacher drbd_info(device, "unexpected cstate (%s) in receive_bitmap\n", 4301b30ab791SAndreas Gruenbacher drbd_conn_str(device->state.conn)); 4302b411b363SPhilipp Reisner } 430382bc0194SAndreas Gruenbacher err = 0; 4304b411b363SPhilipp Reisner 4305b411b363SPhilipp Reisner out: 4306b30ab791SAndreas Gruenbacher drbd_bm_unlock(device); 4307b30ab791SAndreas Gruenbacher if (!err && device->state.conn == C_WF_BITMAP_S) 4308b30ab791SAndreas Gruenbacher drbd_start_resync(device, C_SYNC_SOURCE); 430982bc0194SAndreas Gruenbacher return err; 4310b411b363SPhilipp Reisner } 4311b411b363SPhilipp Reisner 4312bde89a9eSAndreas Gruenbacher static int receive_skip(struct drbd_connection *connection, struct packet_info *pi) 4313b411b363SPhilipp Reisner { 43141ec861ebSAndreas Gruenbacher drbd_warn(connection, "skipping unknown optional packet type %d, l: %d!\n", 4315e2857216SAndreas Gruenbacher pi->cmd, pi->size); 4316b411b363SPhilipp Reisner 4317bde89a9eSAndreas Gruenbacher return ignore_remaining_packet(connection, pi); 4318b411b363SPhilipp Reisner } 4319b411b363SPhilipp Reisner 4320bde89a9eSAndreas Gruenbacher static int receive_UnplugRemote(struct drbd_connection *connection, struct packet_info *pi) 4321b411b363SPhilipp Reisner { 4322b411b363SPhilipp Reisner /* Make sure we've acked all the TCP data associated 4323b411b363SPhilipp Reisner * with the data requests being unplugged */ 4324bde89a9eSAndreas Gruenbacher drbd_tcp_quickack(connection->data.socket); 4325b411b363SPhilipp Reisner 432682bc0194SAndreas Gruenbacher return 0; 4327b411b363SPhilipp Reisner } 4328b411b363SPhilipp Reisner 4329bde89a9eSAndreas Gruenbacher static int receive_out_of_sync(struct drbd_connection *connection, struct packet_info *pi) 433073a01a18SPhilipp Reisner { 4331b30ab791SAndreas Gruenbacher struct drbd_device *device; 4332e658983aSAndreas Gruenbacher struct p_block_desc *p = pi->data; 43334a76b161SAndreas Gruenbacher 4334bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 4335b30ab791SAndreas Gruenbacher if (!device) 43364a76b161SAndreas Gruenbacher return -EIO; 433773a01a18SPhilipp Reisner 4338b30ab791SAndreas Gruenbacher switch (device->state.conn) { 4339f735e363SLars Ellenberg case C_WF_SYNC_UUID: 4340f735e363SLars Ellenberg case C_WF_BITMAP_T: 4341f735e363SLars Ellenberg case C_BEHIND: 4342f735e363SLars Ellenberg break; 4343f735e363SLars Ellenberg default: 4344d0180171SAndreas Gruenbacher drbd_err(device, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", 4345b30ab791SAndreas Gruenbacher drbd_conn_str(device->state.conn)); 4346f735e363SLars Ellenberg } 4347f735e363SLars Ellenberg 4348b30ab791SAndreas Gruenbacher drbd_set_out_of_sync(device, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); 434973a01a18SPhilipp Reisner 435082bc0194SAndreas Gruenbacher return 0; 435173a01a18SPhilipp Reisner } 435273a01a18SPhilipp Reisner 435302918be2SPhilipp Reisner struct data_cmd { 435402918be2SPhilipp Reisner int expect_payload; 435502918be2SPhilipp Reisner size_t pkt_size; 4356bde89a9eSAndreas Gruenbacher int (*fn)(struct drbd_connection *, struct packet_info *); 4357b411b363SPhilipp Reisner }; 4358b411b363SPhilipp Reisner 435902918be2SPhilipp Reisner static struct data_cmd drbd_cmd_handler[] = { 436002918be2SPhilipp Reisner [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, 436102918be2SPhilipp Reisner [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, 436202918be2SPhilipp Reisner [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , 436302918be2SPhilipp Reisner [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , 4364e658983aSAndreas Gruenbacher [P_BITMAP] = { 1, 0, receive_bitmap } , 4365e658983aSAndreas Gruenbacher [P_COMPRESSED_BITMAP] = { 1, 0, receive_bitmap } , 4366e658983aSAndreas Gruenbacher [P_UNPLUG_REMOTE] = { 0, 0, receive_UnplugRemote }, 436702918be2SPhilipp Reisner [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, 436802918be2SPhilipp Reisner [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, 4369e658983aSAndreas Gruenbacher [P_SYNC_PARAM] = { 1, 0, receive_SyncParam }, 4370e658983aSAndreas Gruenbacher [P_SYNC_PARAM89] = { 1, 0, receive_SyncParam }, 437102918be2SPhilipp Reisner [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, 437202918be2SPhilipp Reisner [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, 437302918be2SPhilipp Reisner [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, 437402918be2SPhilipp Reisner [P_STATE] = { 0, sizeof(struct p_state), receive_state }, 437502918be2SPhilipp Reisner [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, 437602918be2SPhilipp Reisner [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, 437702918be2SPhilipp Reisner [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, 437802918be2SPhilipp Reisner [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, 437902918be2SPhilipp Reisner [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, 438002918be2SPhilipp Reisner [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, 438173a01a18SPhilipp Reisner [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync }, 43824a76b161SAndreas Gruenbacher [P_CONN_ST_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_conn_state }, 4383036b17eaSPhilipp Reisner [P_PROTOCOL_UPDATE] = { 1, sizeof(struct p_protocol), receive_protocol }, 438402918be2SPhilipp Reisner }; 438502918be2SPhilipp Reisner 4386bde89a9eSAndreas Gruenbacher static void drbdd(struct drbd_connection *connection) 4387b411b363SPhilipp Reisner { 438877351055SPhilipp Reisner struct packet_info pi; 438902918be2SPhilipp Reisner size_t shs; /* sub header size */ 439082bc0194SAndreas Gruenbacher int err; 4391b411b363SPhilipp Reisner 4392bde89a9eSAndreas Gruenbacher while (get_t_state(&connection->receiver) == RUNNING) { 4393deebe195SAndreas Gruenbacher struct data_cmd *cmd; 4394deebe195SAndreas Gruenbacher 4395bde89a9eSAndreas Gruenbacher drbd_thread_current_set_cpu(&connection->receiver); 4396bde89a9eSAndreas Gruenbacher if (drbd_recv_header(connection, &pi)) 439702918be2SPhilipp Reisner goto err_out; 439802918be2SPhilipp Reisner 4399deebe195SAndreas Gruenbacher cmd = &drbd_cmd_handler[pi.cmd]; 44004a76b161SAndreas Gruenbacher if (unlikely(pi.cmd >= ARRAY_SIZE(drbd_cmd_handler) || !cmd->fn)) { 44011ec861ebSAndreas Gruenbacher drbd_err(connection, "Unexpected data packet %s (0x%04x)", 44022fcb8f30SAndreas Gruenbacher cmdname(pi.cmd), pi.cmd); 440302918be2SPhilipp Reisner goto err_out; 44040b33a916SLars Ellenberg } 4405b411b363SPhilipp Reisner 4406e658983aSAndreas Gruenbacher shs = cmd->pkt_size; 4407e658983aSAndreas Gruenbacher if (pi.size > shs && !cmd->expect_payload) { 44081ec861ebSAndreas Gruenbacher drbd_err(connection, "No payload expected %s l:%d\n", 44092fcb8f30SAndreas Gruenbacher cmdname(pi.cmd), pi.size); 4410c13f7e1aSLars Ellenberg goto err_out; 4411c13f7e1aSLars Ellenberg } 4412c13f7e1aSLars Ellenberg 4413c13f7e1aSLars Ellenberg if (shs) { 4414bde89a9eSAndreas Gruenbacher err = drbd_recv_all_warn(connection, pi.data, shs); 4415a5c31904SAndreas Gruenbacher if (err) 441602918be2SPhilipp Reisner goto err_out; 4417e2857216SAndreas Gruenbacher pi.size -= shs; 4418b411b363SPhilipp Reisner } 441902918be2SPhilipp Reisner 4420bde89a9eSAndreas Gruenbacher err = cmd->fn(connection, &pi); 44214a76b161SAndreas Gruenbacher if (err) { 44221ec861ebSAndreas Gruenbacher drbd_err(connection, "error receiving %s, e: %d l: %d!\n", 44239f5bdc33SAndreas Gruenbacher cmdname(pi.cmd), err, pi.size); 442402918be2SPhilipp Reisner goto err_out; 442502918be2SPhilipp Reisner } 442602918be2SPhilipp Reisner } 442782bc0194SAndreas Gruenbacher return; 442802918be2SPhilipp Reisner 442902918be2SPhilipp Reisner err_out: 4430bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); 4431b411b363SPhilipp Reisner } 4432b411b363SPhilipp Reisner 4433bde89a9eSAndreas Gruenbacher void conn_flush_workqueue(struct drbd_connection *connection) 4434b411b363SPhilipp Reisner { 4435b411b363SPhilipp Reisner struct drbd_wq_barrier barr; 4436b411b363SPhilipp Reisner 4437b411b363SPhilipp Reisner barr.w.cb = w_prev_work_done; 4438bde89a9eSAndreas Gruenbacher barr.w.connection = connection; 4439b411b363SPhilipp Reisner init_completion(&barr.done); 4440bde89a9eSAndreas Gruenbacher drbd_queue_work(&connection->sender_work, &barr.w); 4441b411b363SPhilipp Reisner wait_for_completion(&barr.done); 4442b411b363SPhilipp Reisner } 4443b411b363SPhilipp Reisner 4444bde89a9eSAndreas Gruenbacher static void conn_disconnect(struct drbd_connection *connection) 4445f70b3511SPhilipp Reisner { 4446c06ece6bSAndreas Gruenbacher struct drbd_peer_device *peer_device; 4447bbeb641cSPhilipp Reisner enum drbd_conns oc; 4448376694a0SPhilipp Reisner int vnr; 4449f70b3511SPhilipp Reisner 4450bde89a9eSAndreas Gruenbacher if (connection->cstate == C_STANDALONE) 4451b411b363SPhilipp Reisner return; 4452b411b363SPhilipp Reisner 4453545752d5SLars Ellenberg /* We are about to start the cleanup after connection loss. 4454545752d5SLars Ellenberg * Make sure drbd_make_request knows about that. 4455545752d5SLars Ellenberg * Usually we should be in some network failure state already, 4456545752d5SLars Ellenberg * but just in case we are not, we fix it up here. 4457545752d5SLars Ellenberg */ 4458bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); 4459545752d5SLars Ellenberg 4460b411b363SPhilipp Reisner /* asender does not clean up anything. it must not interfere, either */ 4461bde89a9eSAndreas Gruenbacher drbd_thread_stop(&connection->asender); 4462bde89a9eSAndreas Gruenbacher drbd_free_sock(connection); 4463360cc740SPhilipp Reisner 4464c141ebdaSPhilipp Reisner rcu_read_lock(); 4465c06ece6bSAndreas Gruenbacher idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 4466c06ece6bSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 4467b30ab791SAndreas Gruenbacher kref_get(&device->kref); 4468c141ebdaSPhilipp Reisner rcu_read_unlock(); 4469b30ab791SAndreas Gruenbacher drbd_disconnected(device); 4470c06ece6bSAndreas Gruenbacher kref_put(&device->kref, drbd_destroy_device); 4471c141ebdaSPhilipp Reisner rcu_read_lock(); 4472c141ebdaSPhilipp Reisner } 4473c141ebdaSPhilipp Reisner rcu_read_unlock(); 4474c141ebdaSPhilipp Reisner 4475bde89a9eSAndreas Gruenbacher if (!list_empty(&connection->current_epoch->list)) 44761ec861ebSAndreas Gruenbacher drbd_err(connection, "ASSERTION FAILED: connection->current_epoch->list not empty\n"); 447712038a3aSPhilipp Reisner /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ 4478bde89a9eSAndreas Gruenbacher atomic_set(&connection->current_epoch->epoch_size, 0); 4479bde89a9eSAndreas Gruenbacher connection->send.seen_any_write_yet = false; 448012038a3aSPhilipp Reisner 44811ec861ebSAndreas Gruenbacher drbd_info(connection, "Connection closed\n"); 4482360cc740SPhilipp Reisner 4483bde89a9eSAndreas Gruenbacher if (conn_highest_role(connection) == R_PRIMARY && conn_highest_pdsk(connection) >= D_UNKNOWN) 4484bde89a9eSAndreas Gruenbacher conn_try_outdate_peer_async(connection); 4485cb703454SPhilipp Reisner 44860500813fSAndreas Gruenbacher spin_lock_irq(&connection->resource->req_lock); 4487bde89a9eSAndreas Gruenbacher oc = connection->cstate; 4488bbeb641cSPhilipp Reisner if (oc >= C_UNCONNECTED) 4489bde89a9eSAndreas Gruenbacher _conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE); 4490bbeb641cSPhilipp Reisner 44910500813fSAndreas Gruenbacher spin_unlock_irq(&connection->resource->req_lock); 4492360cc740SPhilipp Reisner 4493f3dfa40aSLars Ellenberg if (oc == C_DISCONNECTING) 4494bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_STANDALONE), CS_VERBOSE | CS_HARD); 4495360cc740SPhilipp Reisner } 4496360cc740SPhilipp Reisner 4497b30ab791SAndreas Gruenbacher static int drbd_disconnected(struct drbd_device *device) 4498360cc740SPhilipp Reisner { 4499360cc740SPhilipp Reisner unsigned int i; 4500b411b363SPhilipp Reisner 450185719573SPhilipp Reisner /* wait for current activity to cease. */ 45020500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 4503b30ab791SAndreas Gruenbacher _drbd_wait_ee_list_empty(device, &device->active_ee); 4504b30ab791SAndreas Gruenbacher _drbd_wait_ee_list_empty(device, &device->sync_ee); 4505b30ab791SAndreas Gruenbacher _drbd_wait_ee_list_empty(device, &device->read_ee); 45060500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 4507b411b363SPhilipp Reisner 4508b411b363SPhilipp Reisner /* We do not have data structures that would allow us to 4509b411b363SPhilipp Reisner * get the rs_pending_cnt down to 0 again. 4510b411b363SPhilipp Reisner * * On C_SYNC_TARGET we do not have any data structures describing 4511b411b363SPhilipp Reisner * the pending RSDataRequest's we have sent. 4512b411b363SPhilipp Reisner * * On C_SYNC_SOURCE there is no data structure that tracks 4513b411b363SPhilipp Reisner * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. 4514b411b363SPhilipp Reisner * And no, it is not the sum of the reference counts in the 4515b411b363SPhilipp Reisner * resync_LRU. The resync_LRU tracks the whole operation including 4516b411b363SPhilipp Reisner * the disk-IO, while the rs_pending_cnt only tracks the blocks 4517b411b363SPhilipp Reisner * on the fly. */ 4518b30ab791SAndreas Gruenbacher drbd_rs_cancel_all(device); 4519b30ab791SAndreas Gruenbacher device->rs_total = 0; 4520b30ab791SAndreas Gruenbacher device->rs_failed = 0; 4521b30ab791SAndreas Gruenbacher atomic_set(&device->rs_pending_cnt, 0); 4522b30ab791SAndreas Gruenbacher wake_up(&device->misc_wait); 4523b411b363SPhilipp Reisner 4524b30ab791SAndreas Gruenbacher del_timer_sync(&device->resync_timer); 4525b30ab791SAndreas Gruenbacher resync_timer_fn((unsigned long)device); 4526b411b363SPhilipp Reisner 4527b411b363SPhilipp Reisner /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, 4528b411b363SPhilipp Reisner * w_make_resync_request etc. which may still be on the worker queue 4529b411b363SPhilipp Reisner * to be "canceled" */ 4530b30ab791SAndreas Gruenbacher drbd_flush_workqueue(device); 4531b411b363SPhilipp Reisner 4532b30ab791SAndreas Gruenbacher drbd_finish_peer_reqs(device); 4533b411b363SPhilipp Reisner 4534d10b4ea3SPhilipp Reisner /* This second workqueue flush is necessary, since drbd_finish_peer_reqs() 4535d10b4ea3SPhilipp Reisner might have issued a work again. The one before drbd_finish_peer_reqs() is 4536d10b4ea3SPhilipp Reisner necessary to reclain net_ee in drbd_finish_peer_reqs(). */ 4537b30ab791SAndreas Gruenbacher drbd_flush_workqueue(device); 4538d10b4ea3SPhilipp Reisner 453908332d73SLars Ellenberg /* need to do it again, drbd_finish_peer_reqs() may have populated it 454008332d73SLars Ellenberg * again via drbd_try_clear_on_disk_bm(). */ 4541b30ab791SAndreas Gruenbacher drbd_rs_cancel_all(device); 4542b411b363SPhilipp Reisner 4543b30ab791SAndreas Gruenbacher kfree(device->p_uuid); 4544b30ab791SAndreas Gruenbacher device->p_uuid = NULL; 4545b411b363SPhilipp Reisner 4546b30ab791SAndreas Gruenbacher if (!drbd_suspended(device)) 4547a6b32bc3SAndreas Gruenbacher tl_clear(first_peer_device(device)->connection); 4548b411b363SPhilipp Reisner 4549b30ab791SAndreas Gruenbacher drbd_md_sync(device); 4550b411b363SPhilipp Reisner 455120ceb2b2SLars Ellenberg /* serialize with bitmap writeout triggered by the state change, 455220ceb2b2SLars Ellenberg * if any. */ 4553b30ab791SAndreas Gruenbacher wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags)); 455420ceb2b2SLars Ellenberg 4555b411b363SPhilipp Reisner /* tcp_close and release of sendpage pages can be deferred. I don't 4556b411b363SPhilipp Reisner * want to use SO_LINGER, because apparently it can be deferred for 4557b411b363SPhilipp Reisner * more than 20 seconds (longest time I checked). 4558b411b363SPhilipp Reisner * 4559b411b363SPhilipp Reisner * Actually we don't care for exactly when the network stack does its 4560b411b363SPhilipp Reisner * put_page(), but release our reference on these pages right here. 4561b411b363SPhilipp Reisner */ 4562b30ab791SAndreas Gruenbacher i = drbd_free_peer_reqs(device, &device->net_ee); 4563b411b363SPhilipp Reisner if (i) 4564d0180171SAndreas Gruenbacher drbd_info(device, "net_ee not empty, killed %u entries\n", i); 4565b30ab791SAndreas Gruenbacher i = atomic_read(&device->pp_in_use_by_net); 4566435f0740SLars Ellenberg if (i) 4567d0180171SAndreas Gruenbacher drbd_info(device, "pp_in_use_by_net = %d, expected 0\n", i); 4568b30ab791SAndreas Gruenbacher i = atomic_read(&device->pp_in_use); 4569b411b363SPhilipp Reisner if (i) 4570d0180171SAndreas Gruenbacher drbd_info(device, "pp_in_use = %d, expected 0\n", i); 4571b411b363SPhilipp Reisner 45720b0ba1efSAndreas Gruenbacher D_ASSERT(device, list_empty(&device->read_ee)); 45730b0ba1efSAndreas Gruenbacher D_ASSERT(device, list_empty(&device->active_ee)); 45740b0ba1efSAndreas Gruenbacher D_ASSERT(device, list_empty(&device->sync_ee)); 45750b0ba1efSAndreas Gruenbacher D_ASSERT(device, list_empty(&device->done_ee)); 4576b411b363SPhilipp Reisner 4577360cc740SPhilipp Reisner return 0; 4578b411b363SPhilipp Reisner } 4579b411b363SPhilipp Reisner 4580b411b363SPhilipp Reisner /* 4581b411b363SPhilipp Reisner * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version 4582b411b363SPhilipp Reisner * we can agree on is stored in agreed_pro_version. 4583b411b363SPhilipp Reisner * 4584b411b363SPhilipp Reisner * feature flags and the reserved array should be enough room for future 4585b411b363SPhilipp Reisner * enhancements of the handshake protocol, and possible plugins... 4586b411b363SPhilipp Reisner * 4587b411b363SPhilipp Reisner * for now, they are expected to be zero, but ignored. 4588b411b363SPhilipp Reisner */ 4589bde89a9eSAndreas Gruenbacher static int drbd_send_features(struct drbd_connection *connection) 4590b411b363SPhilipp Reisner { 45919f5bdc33SAndreas Gruenbacher struct drbd_socket *sock; 45929f5bdc33SAndreas Gruenbacher struct p_connection_features *p; 4593b411b363SPhilipp Reisner 4594bde89a9eSAndreas Gruenbacher sock = &connection->data; 4595bde89a9eSAndreas Gruenbacher p = conn_prepare_command(connection, sock); 45969f5bdc33SAndreas Gruenbacher if (!p) 4597e8d17b01SAndreas Gruenbacher return -EIO; 4598b411b363SPhilipp Reisner memset(p, 0, sizeof(*p)); 4599b411b363SPhilipp Reisner p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); 4600b411b363SPhilipp Reisner p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); 4601bde89a9eSAndreas Gruenbacher return conn_send_command(connection, sock, P_CONNECTION_FEATURES, sizeof(*p), NULL, 0); 4602b411b363SPhilipp Reisner } 4603b411b363SPhilipp Reisner 4604b411b363SPhilipp Reisner /* 4605b411b363SPhilipp Reisner * return values: 4606b411b363SPhilipp Reisner * 1 yes, we have a valid connection 4607b411b363SPhilipp Reisner * 0 oops, did not work out, please try again 4608b411b363SPhilipp Reisner * -1 peer talks different language, 4609b411b363SPhilipp Reisner * no point in trying again, please go standalone. 4610b411b363SPhilipp Reisner */ 4611bde89a9eSAndreas Gruenbacher static int drbd_do_features(struct drbd_connection *connection) 4612b411b363SPhilipp Reisner { 4613bde89a9eSAndreas Gruenbacher /* ASSERT current == connection->receiver ... */ 4614e658983aSAndreas Gruenbacher struct p_connection_features *p; 4615e658983aSAndreas Gruenbacher const int expect = sizeof(struct p_connection_features); 461677351055SPhilipp Reisner struct packet_info pi; 4617a5c31904SAndreas Gruenbacher int err; 4618b411b363SPhilipp Reisner 4619bde89a9eSAndreas Gruenbacher err = drbd_send_features(connection); 4620e8d17b01SAndreas Gruenbacher if (err) 4621b411b363SPhilipp Reisner return 0; 4622b411b363SPhilipp Reisner 4623bde89a9eSAndreas Gruenbacher err = drbd_recv_header(connection, &pi); 462469bc7bc3SAndreas Gruenbacher if (err) 4625b411b363SPhilipp Reisner return 0; 4626b411b363SPhilipp Reisner 46276038178eSAndreas Gruenbacher if (pi.cmd != P_CONNECTION_FEATURES) { 46281ec861ebSAndreas Gruenbacher drbd_err(connection, "expected ConnectionFeatures packet, received: %s (0x%04x)\n", 462977351055SPhilipp Reisner cmdname(pi.cmd), pi.cmd); 4630b411b363SPhilipp Reisner return -1; 4631b411b363SPhilipp Reisner } 4632b411b363SPhilipp Reisner 463377351055SPhilipp Reisner if (pi.size != expect) { 46341ec861ebSAndreas Gruenbacher drbd_err(connection, "expected ConnectionFeatures length: %u, received: %u\n", 463577351055SPhilipp Reisner expect, pi.size); 4636b411b363SPhilipp Reisner return -1; 4637b411b363SPhilipp Reisner } 4638b411b363SPhilipp Reisner 4639e658983aSAndreas Gruenbacher p = pi.data; 4640bde89a9eSAndreas Gruenbacher err = drbd_recv_all_warn(connection, p, expect); 4641a5c31904SAndreas Gruenbacher if (err) 4642b411b363SPhilipp Reisner return 0; 4643b411b363SPhilipp Reisner 4644b411b363SPhilipp Reisner p->protocol_min = be32_to_cpu(p->protocol_min); 4645b411b363SPhilipp Reisner p->protocol_max = be32_to_cpu(p->protocol_max); 4646b411b363SPhilipp Reisner if (p->protocol_max == 0) 4647b411b363SPhilipp Reisner p->protocol_max = p->protocol_min; 4648b411b363SPhilipp Reisner 4649b411b363SPhilipp Reisner if (PRO_VERSION_MAX < p->protocol_min || 4650b411b363SPhilipp Reisner PRO_VERSION_MIN > p->protocol_max) 4651b411b363SPhilipp Reisner goto incompat; 4652b411b363SPhilipp Reisner 4653bde89a9eSAndreas Gruenbacher connection->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); 4654b411b363SPhilipp Reisner 46551ec861ebSAndreas Gruenbacher drbd_info(connection, "Handshake successful: " 4656bde89a9eSAndreas Gruenbacher "Agreed network protocol version %d\n", connection->agreed_pro_version); 4657b411b363SPhilipp Reisner 4658b411b363SPhilipp Reisner return 1; 4659b411b363SPhilipp Reisner 4660b411b363SPhilipp Reisner incompat: 46611ec861ebSAndreas Gruenbacher drbd_err(connection, "incompatible DRBD dialects: " 4662b411b363SPhilipp Reisner "I support %d-%d, peer supports %d-%d\n", 4663b411b363SPhilipp Reisner PRO_VERSION_MIN, PRO_VERSION_MAX, 4664b411b363SPhilipp Reisner p->protocol_min, p->protocol_max); 4665b411b363SPhilipp Reisner return -1; 4666b411b363SPhilipp Reisner } 4667b411b363SPhilipp Reisner 4668b411b363SPhilipp Reisner #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) 4669bde89a9eSAndreas Gruenbacher static int drbd_do_auth(struct drbd_connection *connection) 4670b411b363SPhilipp Reisner { 46711ec861ebSAndreas Gruenbacher drbd_err(connection, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); 46721ec861ebSAndreas Gruenbacher drbd_err(connection, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); 4673b10d96cbSJohannes Thoma return -1; 4674b411b363SPhilipp Reisner } 4675b411b363SPhilipp Reisner #else 4676b411b363SPhilipp Reisner #define CHALLENGE_LEN 64 4677b10d96cbSJohannes Thoma 4678b10d96cbSJohannes Thoma /* Return value: 4679b10d96cbSJohannes Thoma 1 - auth succeeded, 4680b10d96cbSJohannes Thoma 0 - failed, try again (network error), 4681b10d96cbSJohannes Thoma -1 - auth failed, don't try again. 4682b10d96cbSJohannes Thoma */ 4683b10d96cbSJohannes Thoma 4684bde89a9eSAndreas Gruenbacher static int drbd_do_auth(struct drbd_connection *connection) 4685b411b363SPhilipp Reisner { 46869f5bdc33SAndreas Gruenbacher struct drbd_socket *sock; 4687b411b363SPhilipp Reisner char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ 4688b411b363SPhilipp Reisner struct scatterlist sg; 4689b411b363SPhilipp Reisner char *response = NULL; 4690b411b363SPhilipp Reisner char *right_response = NULL; 4691b411b363SPhilipp Reisner char *peers_ch = NULL; 469244ed167dSPhilipp Reisner unsigned int key_len; 469344ed167dSPhilipp Reisner char secret[SHARED_SECRET_MAX]; /* 64 byte */ 4694b411b363SPhilipp Reisner unsigned int resp_size; 4695b411b363SPhilipp Reisner struct hash_desc desc; 469677351055SPhilipp Reisner struct packet_info pi; 469744ed167dSPhilipp Reisner struct net_conf *nc; 469869bc7bc3SAndreas Gruenbacher int err, rv; 4699b411b363SPhilipp Reisner 47009f5bdc33SAndreas Gruenbacher /* FIXME: Put the challenge/response into the preallocated socket buffer. */ 47019f5bdc33SAndreas Gruenbacher 470244ed167dSPhilipp Reisner rcu_read_lock(); 4703bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 470444ed167dSPhilipp Reisner key_len = strlen(nc->shared_secret); 470544ed167dSPhilipp Reisner memcpy(secret, nc->shared_secret, key_len); 470644ed167dSPhilipp Reisner rcu_read_unlock(); 470744ed167dSPhilipp Reisner 4708bde89a9eSAndreas Gruenbacher desc.tfm = connection->cram_hmac_tfm; 4709b411b363SPhilipp Reisner desc.flags = 0; 4710b411b363SPhilipp Reisner 4711bde89a9eSAndreas Gruenbacher rv = crypto_hash_setkey(connection->cram_hmac_tfm, (u8 *)secret, key_len); 4712b411b363SPhilipp Reisner if (rv) { 47131ec861ebSAndreas Gruenbacher drbd_err(connection, "crypto_hash_setkey() failed with %d\n", rv); 4714b10d96cbSJohannes Thoma rv = -1; 4715b411b363SPhilipp Reisner goto fail; 4716b411b363SPhilipp Reisner } 4717b411b363SPhilipp Reisner 4718b411b363SPhilipp Reisner get_random_bytes(my_challenge, CHALLENGE_LEN); 4719b411b363SPhilipp Reisner 4720bde89a9eSAndreas Gruenbacher sock = &connection->data; 4721bde89a9eSAndreas Gruenbacher if (!conn_prepare_command(connection, sock)) { 47229f5bdc33SAndreas Gruenbacher rv = 0; 47239f5bdc33SAndreas Gruenbacher goto fail; 47249f5bdc33SAndreas Gruenbacher } 4725bde89a9eSAndreas Gruenbacher rv = !conn_send_command(connection, sock, P_AUTH_CHALLENGE, 0, 47269f5bdc33SAndreas Gruenbacher my_challenge, CHALLENGE_LEN); 4727b411b363SPhilipp Reisner if (!rv) 4728b411b363SPhilipp Reisner goto fail; 4729b411b363SPhilipp Reisner 4730bde89a9eSAndreas Gruenbacher err = drbd_recv_header(connection, &pi); 473169bc7bc3SAndreas Gruenbacher if (err) { 4732b411b363SPhilipp Reisner rv = 0; 4733b411b363SPhilipp Reisner goto fail; 4734b411b363SPhilipp Reisner } 4735b411b363SPhilipp Reisner 473677351055SPhilipp Reisner if (pi.cmd != P_AUTH_CHALLENGE) { 47371ec861ebSAndreas Gruenbacher drbd_err(connection, "expected AuthChallenge packet, received: %s (0x%04x)\n", 473877351055SPhilipp Reisner cmdname(pi.cmd), pi.cmd); 4739b411b363SPhilipp Reisner rv = 0; 4740b411b363SPhilipp Reisner goto fail; 4741b411b363SPhilipp Reisner } 4742b411b363SPhilipp Reisner 474377351055SPhilipp Reisner if (pi.size > CHALLENGE_LEN * 2) { 47441ec861ebSAndreas Gruenbacher drbd_err(connection, "expected AuthChallenge payload too big.\n"); 4745b10d96cbSJohannes Thoma rv = -1; 4746b411b363SPhilipp Reisner goto fail; 4747b411b363SPhilipp Reisner } 4748b411b363SPhilipp Reisner 474977351055SPhilipp Reisner peers_ch = kmalloc(pi.size, GFP_NOIO); 4750b411b363SPhilipp Reisner if (peers_ch == NULL) { 47511ec861ebSAndreas Gruenbacher drbd_err(connection, "kmalloc of peers_ch failed\n"); 4752b10d96cbSJohannes Thoma rv = -1; 4753b411b363SPhilipp Reisner goto fail; 4754b411b363SPhilipp Reisner } 4755b411b363SPhilipp Reisner 4756bde89a9eSAndreas Gruenbacher err = drbd_recv_all_warn(connection, peers_ch, pi.size); 4757a5c31904SAndreas Gruenbacher if (err) { 4758b411b363SPhilipp Reisner rv = 0; 4759b411b363SPhilipp Reisner goto fail; 4760b411b363SPhilipp Reisner } 4761b411b363SPhilipp Reisner 4762bde89a9eSAndreas Gruenbacher resp_size = crypto_hash_digestsize(connection->cram_hmac_tfm); 4763b411b363SPhilipp Reisner response = kmalloc(resp_size, GFP_NOIO); 4764b411b363SPhilipp Reisner if (response == NULL) { 47651ec861ebSAndreas Gruenbacher drbd_err(connection, "kmalloc of response failed\n"); 4766b10d96cbSJohannes Thoma rv = -1; 4767b411b363SPhilipp Reisner goto fail; 4768b411b363SPhilipp Reisner } 4769b411b363SPhilipp Reisner 4770b411b363SPhilipp Reisner sg_init_table(&sg, 1); 477177351055SPhilipp Reisner sg_set_buf(&sg, peers_ch, pi.size); 4772b411b363SPhilipp Reisner 4773b411b363SPhilipp Reisner rv = crypto_hash_digest(&desc, &sg, sg.length, response); 4774b411b363SPhilipp Reisner if (rv) { 47751ec861ebSAndreas Gruenbacher drbd_err(connection, "crypto_hash_digest() failed with %d\n", rv); 4776b10d96cbSJohannes Thoma rv = -1; 4777b411b363SPhilipp Reisner goto fail; 4778b411b363SPhilipp Reisner } 4779b411b363SPhilipp Reisner 4780bde89a9eSAndreas Gruenbacher if (!conn_prepare_command(connection, sock)) { 47819f5bdc33SAndreas Gruenbacher rv = 0; 47829f5bdc33SAndreas Gruenbacher goto fail; 47839f5bdc33SAndreas Gruenbacher } 4784bde89a9eSAndreas Gruenbacher rv = !conn_send_command(connection, sock, P_AUTH_RESPONSE, 0, 47859f5bdc33SAndreas Gruenbacher response, resp_size); 4786b411b363SPhilipp Reisner if (!rv) 4787b411b363SPhilipp Reisner goto fail; 4788b411b363SPhilipp Reisner 4789bde89a9eSAndreas Gruenbacher err = drbd_recv_header(connection, &pi); 479069bc7bc3SAndreas Gruenbacher if (err) { 4791b411b363SPhilipp Reisner rv = 0; 4792b411b363SPhilipp Reisner goto fail; 4793b411b363SPhilipp Reisner } 4794b411b363SPhilipp Reisner 479577351055SPhilipp Reisner if (pi.cmd != P_AUTH_RESPONSE) { 47961ec861ebSAndreas Gruenbacher drbd_err(connection, "expected AuthResponse packet, received: %s (0x%04x)\n", 479777351055SPhilipp Reisner cmdname(pi.cmd), pi.cmd); 4798b411b363SPhilipp Reisner rv = 0; 4799b411b363SPhilipp Reisner goto fail; 4800b411b363SPhilipp Reisner } 4801b411b363SPhilipp Reisner 480277351055SPhilipp Reisner if (pi.size != resp_size) { 48031ec861ebSAndreas Gruenbacher drbd_err(connection, "expected AuthResponse payload of wrong size\n"); 4804b411b363SPhilipp Reisner rv = 0; 4805b411b363SPhilipp Reisner goto fail; 4806b411b363SPhilipp Reisner } 4807b411b363SPhilipp Reisner 4808bde89a9eSAndreas Gruenbacher err = drbd_recv_all_warn(connection, response , resp_size); 4809a5c31904SAndreas Gruenbacher if (err) { 4810b411b363SPhilipp Reisner rv = 0; 4811b411b363SPhilipp Reisner goto fail; 4812b411b363SPhilipp Reisner } 4813b411b363SPhilipp Reisner 4814b411b363SPhilipp Reisner right_response = kmalloc(resp_size, GFP_NOIO); 48152d1ee87dSJulia Lawall if (right_response == NULL) { 48161ec861ebSAndreas Gruenbacher drbd_err(connection, "kmalloc of right_response failed\n"); 4817b10d96cbSJohannes Thoma rv = -1; 4818b411b363SPhilipp Reisner goto fail; 4819b411b363SPhilipp Reisner } 4820b411b363SPhilipp Reisner 4821b411b363SPhilipp Reisner sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); 4822b411b363SPhilipp Reisner 4823b411b363SPhilipp Reisner rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); 4824b411b363SPhilipp Reisner if (rv) { 48251ec861ebSAndreas Gruenbacher drbd_err(connection, "crypto_hash_digest() failed with %d\n", rv); 4826b10d96cbSJohannes Thoma rv = -1; 4827b411b363SPhilipp Reisner goto fail; 4828b411b363SPhilipp Reisner } 4829b411b363SPhilipp Reisner 4830b411b363SPhilipp Reisner rv = !memcmp(response, right_response, resp_size); 4831b411b363SPhilipp Reisner 4832b411b363SPhilipp Reisner if (rv) 48331ec861ebSAndreas Gruenbacher drbd_info(connection, "Peer authenticated using %d bytes HMAC\n", 483444ed167dSPhilipp Reisner resp_size); 4835b10d96cbSJohannes Thoma else 4836b10d96cbSJohannes Thoma rv = -1; 4837b411b363SPhilipp Reisner 4838b411b363SPhilipp Reisner fail: 4839b411b363SPhilipp Reisner kfree(peers_ch); 4840b411b363SPhilipp Reisner kfree(response); 4841b411b363SPhilipp Reisner kfree(right_response); 4842b411b363SPhilipp Reisner 4843b411b363SPhilipp Reisner return rv; 4844b411b363SPhilipp Reisner } 4845b411b363SPhilipp Reisner #endif 4846b411b363SPhilipp Reisner 4847b411b363SPhilipp Reisner int drbdd_init(struct drbd_thread *thi) 4848b411b363SPhilipp Reisner { 4849bde89a9eSAndreas Gruenbacher struct drbd_connection *connection = thi->connection; 4850b411b363SPhilipp Reisner int h; 4851b411b363SPhilipp Reisner 48521ec861ebSAndreas Gruenbacher drbd_info(connection, "receiver (re)started\n"); 4853b411b363SPhilipp Reisner 4854b411b363SPhilipp Reisner do { 4855bde89a9eSAndreas Gruenbacher h = conn_connect(connection); 4856b411b363SPhilipp Reisner if (h == 0) { 4857bde89a9eSAndreas Gruenbacher conn_disconnect(connection); 485820ee6390SPhilipp Reisner schedule_timeout_interruptible(HZ); 4859b411b363SPhilipp Reisner } 4860b411b363SPhilipp Reisner if (h == -1) { 48611ec861ebSAndreas Gruenbacher drbd_warn(connection, "Discarding network configuration.\n"); 4862bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); 4863b411b363SPhilipp Reisner } 4864b411b363SPhilipp Reisner } while (h == 0); 4865b411b363SPhilipp Reisner 486691fd4dadSPhilipp Reisner if (h > 0) 4867bde89a9eSAndreas Gruenbacher drbdd(connection); 4868b411b363SPhilipp Reisner 4869bde89a9eSAndreas Gruenbacher conn_disconnect(connection); 4870b411b363SPhilipp Reisner 48711ec861ebSAndreas Gruenbacher drbd_info(connection, "receiver terminated\n"); 4872b411b363SPhilipp Reisner return 0; 4873b411b363SPhilipp Reisner } 4874b411b363SPhilipp Reisner 4875b411b363SPhilipp Reisner /* ********* acknowledge sender ******** */ 4876b411b363SPhilipp Reisner 4877bde89a9eSAndreas Gruenbacher static int got_conn_RqSReply(struct drbd_connection *connection, struct packet_info *pi) 4878b411b363SPhilipp Reisner { 4879e658983aSAndreas Gruenbacher struct p_req_state_reply *p = pi->data; 4880b411b363SPhilipp Reisner int retcode = be32_to_cpu(p->retcode); 4881b411b363SPhilipp Reisner 4882b411b363SPhilipp Reisner if (retcode >= SS_SUCCESS) { 4883bde89a9eSAndreas Gruenbacher set_bit(CONN_WD_ST_CHG_OKAY, &connection->flags); 4884b411b363SPhilipp Reisner } else { 4885bde89a9eSAndreas Gruenbacher set_bit(CONN_WD_ST_CHG_FAIL, &connection->flags); 48861ec861ebSAndreas Gruenbacher drbd_err(connection, "Requested state change failed by peer: %s (%d)\n", 4887fc3b10a4SPhilipp Reisner drbd_set_st_err_str(retcode), retcode); 4888fc3b10a4SPhilipp Reisner } 4889bde89a9eSAndreas Gruenbacher wake_up(&connection->ping_wait); 4890e4f78edeSPhilipp Reisner 48912735a594SAndreas Gruenbacher return 0; 4892fc3b10a4SPhilipp Reisner } 4893e4f78edeSPhilipp Reisner 4894bde89a9eSAndreas Gruenbacher static int got_RqSReply(struct drbd_connection *connection, struct packet_info *pi) 4895e4f78edeSPhilipp Reisner { 4896b30ab791SAndreas Gruenbacher struct drbd_device *device; 4897e658983aSAndreas Gruenbacher struct p_req_state_reply *p = pi->data; 4898e4f78edeSPhilipp Reisner int retcode = be32_to_cpu(p->retcode); 4899e4f78edeSPhilipp Reisner 4900bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 4901b30ab791SAndreas Gruenbacher if (!device) 49022735a594SAndreas Gruenbacher return -EIO; 49031952e916SAndreas Gruenbacher 4904bde89a9eSAndreas Gruenbacher if (test_bit(CONN_WD_ST_CHG_REQ, &connection->flags)) { 49050b0ba1efSAndreas Gruenbacher D_ASSERT(device, connection->agreed_pro_version < 100); 4906bde89a9eSAndreas Gruenbacher return got_conn_RqSReply(connection, pi); 49074d0fc3fdSPhilipp Reisner } 49084d0fc3fdSPhilipp Reisner 4909e4f78edeSPhilipp Reisner if (retcode >= SS_SUCCESS) { 4910b30ab791SAndreas Gruenbacher set_bit(CL_ST_CHG_SUCCESS, &device->flags); 4911e4f78edeSPhilipp Reisner } else { 4912b30ab791SAndreas Gruenbacher set_bit(CL_ST_CHG_FAIL, &device->flags); 4913d0180171SAndreas Gruenbacher drbd_err(device, "Requested state change failed by peer: %s (%d)\n", 4914b411b363SPhilipp Reisner drbd_set_st_err_str(retcode), retcode); 4915b411b363SPhilipp Reisner } 4916b30ab791SAndreas Gruenbacher wake_up(&device->state_wait); 4917b411b363SPhilipp Reisner 49182735a594SAndreas Gruenbacher return 0; 4919b411b363SPhilipp Reisner } 4920b411b363SPhilipp Reisner 4921bde89a9eSAndreas Gruenbacher static int got_Ping(struct drbd_connection *connection, struct packet_info *pi) 4922b411b363SPhilipp Reisner { 4923bde89a9eSAndreas Gruenbacher return drbd_send_ping_ack(connection); 4924b411b363SPhilipp Reisner 4925b411b363SPhilipp Reisner } 4926b411b363SPhilipp Reisner 4927bde89a9eSAndreas Gruenbacher static int got_PingAck(struct drbd_connection *connection, struct packet_info *pi) 4928b411b363SPhilipp Reisner { 4929b411b363SPhilipp Reisner /* restore idle timeout */ 4930bde89a9eSAndreas Gruenbacher connection->meta.socket->sk->sk_rcvtimeo = connection->net_conf->ping_int*HZ; 4931bde89a9eSAndreas Gruenbacher if (!test_and_set_bit(GOT_PING_ACK, &connection->flags)) 4932bde89a9eSAndreas Gruenbacher wake_up(&connection->ping_wait); 4933b411b363SPhilipp Reisner 49342735a594SAndreas Gruenbacher return 0; 4935b411b363SPhilipp Reisner } 4936b411b363SPhilipp Reisner 4937bde89a9eSAndreas Gruenbacher static int got_IsInSync(struct drbd_connection *connection, struct packet_info *pi) 4938b411b363SPhilipp Reisner { 4939b30ab791SAndreas Gruenbacher struct drbd_device *device; 4940e658983aSAndreas Gruenbacher struct p_block_ack *p = pi->data; 4941b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 4942b411b363SPhilipp Reisner int blksize = be32_to_cpu(p->blksize); 4943b411b363SPhilipp Reisner 4944bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 4945b30ab791SAndreas Gruenbacher if (!device) 49462735a594SAndreas Gruenbacher return -EIO; 49471952e916SAndreas Gruenbacher 49480b0ba1efSAndreas Gruenbacher D_ASSERT(device, first_peer_device(device)->connection->agreed_pro_version >= 89); 4949b411b363SPhilipp Reisner 4950b30ab791SAndreas Gruenbacher update_peer_seq(device, be32_to_cpu(p->seq_num)); 4951b411b363SPhilipp Reisner 4952b30ab791SAndreas Gruenbacher if (get_ldev(device)) { 4953b30ab791SAndreas Gruenbacher drbd_rs_complete_io(device, sector); 4954b30ab791SAndreas Gruenbacher drbd_set_in_sync(device, sector, blksize); 4955b411b363SPhilipp Reisner /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ 4956b30ab791SAndreas Gruenbacher device->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); 4957b30ab791SAndreas Gruenbacher put_ldev(device); 49581d53f09eSLars Ellenberg } 4959b30ab791SAndreas Gruenbacher dec_rs_pending(device); 4960b30ab791SAndreas Gruenbacher atomic_add(blksize >> 9, &device->rs_sect_in); 4961b411b363SPhilipp Reisner 49622735a594SAndreas Gruenbacher return 0; 4963b411b363SPhilipp Reisner } 4964b411b363SPhilipp Reisner 4965bc9c5c41SAndreas Gruenbacher static int 4966b30ab791SAndreas Gruenbacher validate_req_change_req_state(struct drbd_device *device, u64 id, sector_t sector, 4967bc9c5c41SAndreas Gruenbacher struct rb_root *root, const char *func, 4968bc9c5c41SAndreas Gruenbacher enum drbd_req_event what, bool missing_ok) 4969b411b363SPhilipp Reisner { 4970b411b363SPhilipp Reisner struct drbd_request *req; 4971b411b363SPhilipp Reisner struct bio_and_error m; 4972b411b363SPhilipp Reisner 49730500813fSAndreas Gruenbacher spin_lock_irq(&device->resource->req_lock); 4974b30ab791SAndreas Gruenbacher req = find_request(device, root, id, sector, missing_ok, func); 4975b411b363SPhilipp Reisner if (unlikely(!req)) { 49760500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 497785997675SAndreas Gruenbacher return -EIO; 4978b411b363SPhilipp Reisner } 4979b411b363SPhilipp Reisner __req_mod(req, what, &m); 49800500813fSAndreas Gruenbacher spin_unlock_irq(&device->resource->req_lock); 4981b411b363SPhilipp Reisner 4982b411b363SPhilipp Reisner if (m.bio) 4983b30ab791SAndreas Gruenbacher complete_master_bio(device, &m); 498485997675SAndreas Gruenbacher return 0; 4985b411b363SPhilipp Reisner } 4986b411b363SPhilipp Reisner 4987bde89a9eSAndreas Gruenbacher static int got_BlockAck(struct drbd_connection *connection, struct packet_info *pi) 4988b411b363SPhilipp Reisner { 4989b30ab791SAndreas Gruenbacher struct drbd_device *device; 4990e658983aSAndreas Gruenbacher struct p_block_ack *p = pi->data; 4991b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 4992b411b363SPhilipp Reisner int blksize = be32_to_cpu(p->blksize); 4993b411b363SPhilipp Reisner enum drbd_req_event what; 4994b411b363SPhilipp Reisner 4995bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 4996b30ab791SAndreas Gruenbacher if (!device) 49972735a594SAndreas Gruenbacher return -EIO; 49981952e916SAndreas Gruenbacher 4999b30ab791SAndreas Gruenbacher update_peer_seq(device, be32_to_cpu(p->seq_num)); 5000b411b363SPhilipp Reisner 5001579b57edSAndreas Gruenbacher if (p->block_id == ID_SYNCER) { 5002b30ab791SAndreas Gruenbacher drbd_set_in_sync(device, sector, blksize); 5003b30ab791SAndreas Gruenbacher dec_rs_pending(device); 50042735a594SAndreas Gruenbacher return 0; 5005b411b363SPhilipp Reisner } 5006e05e1e59SAndreas Gruenbacher switch (pi->cmd) { 5007b411b363SPhilipp Reisner case P_RS_WRITE_ACK: 50088554df1cSAndreas Gruenbacher what = WRITE_ACKED_BY_PEER_AND_SIS; 5009b411b363SPhilipp Reisner break; 5010b411b363SPhilipp Reisner case P_WRITE_ACK: 50118554df1cSAndreas Gruenbacher what = WRITE_ACKED_BY_PEER; 5012b411b363SPhilipp Reisner break; 5013b411b363SPhilipp Reisner case P_RECV_ACK: 50148554df1cSAndreas Gruenbacher what = RECV_ACKED_BY_PEER; 5015b411b363SPhilipp Reisner break; 5016d4dabbe2SLars Ellenberg case P_SUPERSEDED: 5017d4dabbe2SLars Ellenberg what = CONFLICT_RESOLVED; 50187be8da07SAndreas Gruenbacher break; 50197be8da07SAndreas Gruenbacher case P_RETRY_WRITE: 50207be8da07SAndreas Gruenbacher what = POSTPONE_WRITE; 5021b411b363SPhilipp Reisner break; 5022b411b363SPhilipp Reisner default: 50232735a594SAndreas Gruenbacher BUG(); 5024b411b363SPhilipp Reisner } 5025b411b363SPhilipp Reisner 5026b30ab791SAndreas Gruenbacher return validate_req_change_req_state(device, p->block_id, sector, 5027b30ab791SAndreas Gruenbacher &device->write_requests, __func__, 5028bc9c5c41SAndreas Gruenbacher what, false); 5029b411b363SPhilipp Reisner } 5030b411b363SPhilipp Reisner 5031bde89a9eSAndreas Gruenbacher static int got_NegAck(struct drbd_connection *connection, struct packet_info *pi) 5032b411b363SPhilipp Reisner { 5033b30ab791SAndreas Gruenbacher struct drbd_device *device; 5034e658983aSAndreas Gruenbacher struct p_block_ack *p = pi->data; 5035b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 50362deb8336SPhilipp Reisner int size = be32_to_cpu(p->blksize); 503785997675SAndreas Gruenbacher int err; 5038b411b363SPhilipp Reisner 5039bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 5040b30ab791SAndreas Gruenbacher if (!device) 50412735a594SAndreas Gruenbacher return -EIO; 5042b411b363SPhilipp Reisner 5043b30ab791SAndreas Gruenbacher update_peer_seq(device, be32_to_cpu(p->seq_num)); 5044b411b363SPhilipp Reisner 5045579b57edSAndreas Gruenbacher if (p->block_id == ID_SYNCER) { 5046b30ab791SAndreas Gruenbacher dec_rs_pending(device); 5047b30ab791SAndreas Gruenbacher drbd_rs_failed_io(device, sector, size); 50482735a594SAndreas Gruenbacher return 0; 5049b411b363SPhilipp Reisner } 50502deb8336SPhilipp Reisner 5051b30ab791SAndreas Gruenbacher err = validate_req_change_req_state(device, p->block_id, sector, 5052b30ab791SAndreas Gruenbacher &device->write_requests, __func__, 5053303d1448SPhilipp Reisner NEG_ACKED, true); 505485997675SAndreas Gruenbacher if (err) { 50552deb8336SPhilipp Reisner /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. 50562deb8336SPhilipp Reisner The master bio might already be completed, therefore the 5057c3afd8f5SAndreas Gruenbacher request is no longer in the collision hash. */ 50582deb8336SPhilipp Reisner /* In Protocol B we might already have got a P_RECV_ACK 50592deb8336SPhilipp Reisner but then get a P_NEG_ACK afterwards. */ 5060b30ab791SAndreas Gruenbacher drbd_set_out_of_sync(device, sector, size); 50612deb8336SPhilipp Reisner } 50622735a594SAndreas Gruenbacher return 0; 5063b411b363SPhilipp Reisner } 5064b411b363SPhilipp Reisner 5065bde89a9eSAndreas Gruenbacher static int got_NegDReply(struct drbd_connection *connection, struct packet_info *pi) 5066b411b363SPhilipp Reisner { 5067b30ab791SAndreas Gruenbacher struct drbd_device *device; 5068e658983aSAndreas Gruenbacher struct p_block_ack *p = pi->data; 5069b411b363SPhilipp Reisner sector_t sector = be64_to_cpu(p->sector); 5070b411b363SPhilipp Reisner 5071bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 5072b30ab791SAndreas Gruenbacher if (!device) 50732735a594SAndreas Gruenbacher return -EIO; 50741952e916SAndreas Gruenbacher 5075b30ab791SAndreas Gruenbacher update_peer_seq(device, be32_to_cpu(p->seq_num)); 50767be8da07SAndreas Gruenbacher 5077d0180171SAndreas Gruenbacher drbd_err(device, "Got NegDReply; Sector %llus, len %u.\n", 5078b411b363SPhilipp Reisner (unsigned long long)sector, be32_to_cpu(p->blksize)); 5079b411b363SPhilipp Reisner 5080b30ab791SAndreas Gruenbacher return validate_req_change_req_state(device, p->block_id, sector, 5081b30ab791SAndreas Gruenbacher &device->read_requests, __func__, 50828554df1cSAndreas Gruenbacher NEG_ACKED, false); 5083b411b363SPhilipp Reisner } 5084b411b363SPhilipp Reisner 5085bde89a9eSAndreas Gruenbacher static int got_NegRSDReply(struct drbd_connection *connection, struct packet_info *pi) 5086b411b363SPhilipp Reisner { 5087b30ab791SAndreas Gruenbacher struct drbd_device *device; 5088b411b363SPhilipp Reisner sector_t sector; 5089b411b363SPhilipp Reisner int size; 5090e658983aSAndreas Gruenbacher struct p_block_ack *p = pi->data; 50911952e916SAndreas Gruenbacher 5092bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 5093b30ab791SAndreas Gruenbacher if (!device) 50942735a594SAndreas Gruenbacher return -EIO; 5095b411b363SPhilipp Reisner 5096b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 5097b411b363SPhilipp Reisner size = be32_to_cpu(p->blksize); 5098b411b363SPhilipp Reisner 5099b30ab791SAndreas Gruenbacher update_peer_seq(device, be32_to_cpu(p->seq_num)); 5100b411b363SPhilipp Reisner 5101b30ab791SAndreas Gruenbacher dec_rs_pending(device); 5102b411b363SPhilipp Reisner 5103b30ab791SAndreas Gruenbacher if (get_ldev_if_state(device, D_FAILED)) { 5104b30ab791SAndreas Gruenbacher drbd_rs_complete_io(device, sector); 5105e05e1e59SAndreas Gruenbacher switch (pi->cmd) { 5106d612d309SPhilipp Reisner case P_NEG_RS_DREPLY: 5107b30ab791SAndreas Gruenbacher drbd_rs_failed_io(device, sector, size); 5108d612d309SPhilipp Reisner case P_RS_CANCEL: 5109d612d309SPhilipp Reisner break; 5110d612d309SPhilipp Reisner default: 51112735a594SAndreas Gruenbacher BUG(); 5112d612d309SPhilipp Reisner } 5113b30ab791SAndreas Gruenbacher put_ldev(device); 5114b411b363SPhilipp Reisner } 5115b411b363SPhilipp Reisner 51162735a594SAndreas Gruenbacher return 0; 5117b411b363SPhilipp Reisner } 5118b411b363SPhilipp Reisner 5119bde89a9eSAndreas Gruenbacher static int got_BarrierAck(struct drbd_connection *connection, struct packet_info *pi) 5120b411b363SPhilipp Reisner { 5121e658983aSAndreas Gruenbacher struct p_barrier_ack *p = pi->data; 5122c06ece6bSAndreas Gruenbacher struct drbd_peer_device *peer_device; 51239ed57dcbSLars Ellenberg int vnr; 5124b411b363SPhilipp Reisner 5125bde89a9eSAndreas Gruenbacher tl_release(connection, p->barrier, be32_to_cpu(p->set_size)); 5126b411b363SPhilipp Reisner 51279ed57dcbSLars Ellenberg rcu_read_lock(); 5128c06ece6bSAndreas Gruenbacher idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 5129c06ece6bSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 5130c06ece6bSAndreas Gruenbacher 5131b30ab791SAndreas Gruenbacher if (device->state.conn == C_AHEAD && 5132b30ab791SAndreas Gruenbacher atomic_read(&device->ap_in_flight) == 0 && 5133b30ab791SAndreas Gruenbacher !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &device->flags)) { 5134b30ab791SAndreas Gruenbacher device->start_resync_timer.expires = jiffies + HZ; 5135b30ab791SAndreas Gruenbacher add_timer(&device->start_resync_timer); 5136c4752ef1SPhilipp Reisner } 51379ed57dcbSLars Ellenberg } 51389ed57dcbSLars Ellenberg rcu_read_unlock(); 5139c4752ef1SPhilipp Reisner 51402735a594SAndreas Gruenbacher return 0; 5141b411b363SPhilipp Reisner } 5142b411b363SPhilipp Reisner 5143bde89a9eSAndreas Gruenbacher static int got_OVResult(struct drbd_connection *connection, struct packet_info *pi) 5144b411b363SPhilipp Reisner { 5145b30ab791SAndreas Gruenbacher struct drbd_device *device; 5146e658983aSAndreas Gruenbacher struct p_block_ack *p = pi->data; 5147b411b363SPhilipp Reisner struct drbd_work *w; 5148b411b363SPhilipp Reisner sector_t sector; 5149b411b363SPhilipp Reisner int size; 5150b411b363SPhilipp Reisner 5151bde89a9eSAndreas Gruenbacher device = vnr_to_device(connection, pi->vnr); 5152b30ab791SAndreas Gruenbacher if (!device) 51532735a594SAndreas Gruenbacher return -EIO; 51541952e916SAndreas Gruenbacher 5155b411b363SPhilipp Reisner sector = be64_to_cpu(p->sector); 5156b411b363SPhilipp Reisner size = be32_to_cpu(p->blksize); 5157b411b363SPhilipp Reisner 5158b30ab791SAndreas Gruenbacher update_peer_seq(device, be32_to_cpu(p->seq_num)); 5159b411b363SPhilipp Reisner 5160b411b363SPhilipp Reisner if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) 5161b30ab791SAndreas Gruenbacher drbd_ov_out_of_sync_found(device, sector, size); 5162b411b363SPhilipp Reisner else 5163b30ab791SAndreas Gruenbacher ov_out_of_sync_print(device); 5164b411b363SPhilipp Reisner 5165b30ab791SAndreas Gruenbacher if (!get_ldev(device)) 51662735a594SAndreas Gruenbacher return 0; 51671d53f09eSLars Ellenberg 5168b30ab791SAndreas Gruenbacher drbd_rs_complete_io(device, sector); 5169b30ab791SAndreas Gruenbacher dec_rs_pending(device); 5170b411b363SPhilipp Reisner 5171b30ab791SAndreas Gruenbacher --device->ov_left; 5172ea5442afSLars Ellenberg 5173ea5442afSLars Ellenberg /* let's advance progress step marks only for every other megabyte */ 5174b30ab791SAndreas Gruenbacher if ((device->ov_left & 0x200) == 0x200) 5175b30ab791SAndreas Gruenbacher drbd_advance_rs_marks(device, device->ov_left); 5176ea5442afSLars Ellenberg 5177b30ab791SAndreas Gruenbacher if (device->ov_left == 0) { 5178b411b363SPhilipp Reisner w = kmalloc(sizeof(*w), GFP_NOIO); 5179b411b363SPhilipp Reisner if (w) { 5180b411b363SPhilipp Reisner w->cb = w_ov_finished; 5181b30ab791SAndreas Gruenbacher w->device = device; 5182a6b32bc3SAndreas Gruenbacher drbd_queue_work(&first_peer_device(device)->connection->sender_work, w); 5183b411b363SPhilipp Reisner } else { 5184d0180171SAndreas Gruenbacher drbd_err(device, "kmalloc(w) failed."); 5185b30ab791SAndreas Gruenbacher ov_out_of_sync_print(device); 5186b30ab791SAndreas Gruenbacher drbd_resync_finished(device); 5187b411b363SPhilipp Reisner } 5188b411b363SPhilipp Reisner } 5189b30ab791SAndreas Gruenbacher put_ldev(device); 51902735a594SAndreas Gruenbacher return 0; 5191b411b363SPhilipp Reisner } 5192b411b363SPhilipp Reisner 5193bde89a9eSAndreas Gruenbacher static int got_skip(struct drbd_connection *connection, struct packet_info *pi) 51940ced55a3SPhilipp Reisner { 51952735a594SAndreas Gruenbacher return 0; 51960ced55a3SPhilipp Reisner } 51970ced55a3SPhilipp Reisner 5198bde89a9eSAndreas Gruenbacher static int connection_finish_peer_reqs(struct drbd_connection *connection) 519932862ec7SPhilipp Reisner { 5200c06ece6bSAndreas Gruenbacher struct drbd_peer_device *peer_device; 5201c141ebdaSPhilipp Reisner int vnr, not_empty = 0; 520232862ec7SPhilipp Reisner 520332862ec7SPhilipp Reisner do { 5204bde89a9eSAndreas Gruenbacher clear_bit(SIGNAL_ASENDER, &connection->flags); 520532862ec7SPhilipp Reisner flush_signals(current); 5206c141ebdaSPhilipp Reisner 5207c141ebdaSPhilipp Reisner rcu_read_lock(); 5208c06ece6bSAndreas Gruenbacher idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 5209c06ece6bSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 5210b30ab791SAndreas Gruenbacher kref_get(&device->kref); 5211c141ebdaSPhilipp Reisner rcu_read_unlock(); 5212b30ab791SAndreas Gruenbacher if (drbd_finish_peer_reqs(device)) { 521305a10ec7SAndreas Gruenbacher kref_put(&device->kref, drbd_destroy_device); 5214c141ebdaSPhilipp Reisner return 1; 5215082a3439SPhilipp Reisner } 521605a10ec7SAndreas Gruenbacher kref_put(&device->kref, drbd_destroy_device); 5217c141ebdaSPhilipp Reisner rcu_read_lock(); 5218d3fcb490SPhilipp Reisner } 5219bde89a9eSAndreas Gruenbacher set_bit(SIGNAL_ASENDER, &connection->flags); 5220082a3439SPhilipp Reisner 52210500813fSAndreas Gruenbacher spin_lock_irq(&connection->resource->req_lock); 5222c06ece6bSAndreas Gruenbacher idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 5223c06ece6bSAndreas Gruenbacher struct drbd_device *device = peer_device->device; 5224b30ab791SAndreas Gruenbacher not_empty = !list_empty(&device->done_ee); 5225082a3439SPhilipp Reisner if (not_empty) 5226082a3439SPhilipp Reisner break; 5227082a3439SPhilipp Reisner } 52280500813fSAndreas Gruenbacher spin_unlock_irq(&connection->resource->req_lock); 5229c141ebdaSPhilipp Reisner rcu_read_unlock(); 523032862ec7SPhilipp Reisner } while (not_empty); 523132862ec7SPhilipp Reisner 523232862ec7SPhilipp Reisner return 0; 5233b411b363SPhilipp Reisner } 5234b411b363SPhilipp Reisner 5235b411b363SPhilipp Reisner struct asender_cmd { 5236b411b363SPhilipp Reisner size_t pkt_size; 5237bde89a9eSAndreas Gruenbacher int (*fn)(struct drbd_connection *connection, struct packet_info *); 5238b411b363SPhilipp Reisner }; 5239b411b363SPhilipp Reisner 5240b411b363SPhilipp Reisner static struct asender_cmd asender_tbl[] = { 5241e658983aSAndreas Gruenbacher [P_PING] = { 0, got_Ping }, 5242e658983aSAndreas Gruenbacher [P_PING_ACK] = { 0, got_PingAck }, 5243b411b363SPhilipp Reisner [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 5244b411b363SPhilipp Reisner [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 5245b411b363SPhilipp Reisner [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, 5246d4dabbe2SLars Ellenberg [P_SUPERSEDED] = { sizeof(struct p_block_ack), got_BlockAck }, 5247b411b363SPhilipp Reisner [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, 5248b411b363SPhilipp Reisner [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, 5249b411b363SPhilipp Reisner [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply }, 5250b411b363SPhilipp Reisner [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, 5251b411b363SPhilipp Reisner [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, 5252b411b363SPhilipp Reisner [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, 5253b411b363SPhilipp Reisner [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, 525402918be2SPhilipp Reisner [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, 5255d612d309SPhilipp Reisner [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply }, 52561952e916SAndreas Gruenbacher [P_CONN_ST_CHG_REPLY]={ sizeof(struct p_req_state_reply), got_conn_RqSReply }, 52571952e916SAndreas Gruenbacher [P_RETRY_WRITE] = { sizeof(struct p_block_ack), got_BlockAck }, 5258b411b363SPhilipp Reisner }; 5259b411b363SPhilipp Reisner 5260b411b363SPhilipp Reisner int drbd_asender(struct drbd_thread *thi) 5261b411b363SPhilipp Reisner { 5262bde89a9eSAndreas Gruenbacher struct drbd_connection *connection = thi->connection; 5263b411b363SPhilipp Reisner struct asender_cmd *cmd = NULL; 526477351055SPhilipp Reisner struct packet_info pi; 5265257d0af6SPhilipp Reisner int rv; 5266bde89a9eSAndreas Gruenbacher void *buf = connection->meta.rbuf; 5267b411b363SPhilipp Reisner int received = 0; 5268bde89a9eSAndreas Gruenbacher unsigned int header_size = drbd_header_size(connection); 526952b061a4SAndreas Gruenbacher int expect = header_size; 527044ed167dSPhilipp Reisner bool ping_timeout_active = false; 527144ed167dSPhilipp Reisner struct net_conf *nc; 5272bb77d34eSAndreas Gruenbacher int ping_timeo, tcp_cork, ping_int; 52733990e04dSPhilipp Reisner struct sched_param param = { .sched_priority = 2 }; 5274b411b363SPhilipp Reisner 52753990e04dSPhilipp Reisner rv = sched_setscheduler(current, SCHED_RR, ¶m); 52763990e04dSPhilipp Reisner if (rv < 0) 52771ec861ebSAndreas Gruenbacher drbd_err(connection, "drbd_asender: ERROR set priority, ret=%d\n", rv); 5278b411b363SPhilipp Reisner 5279e77a0a5cSAndreas Gruenbacher while (get_t_state(thi) == RUNNING) { 528080822284SPhilipp Reisner drbd_thread_current_set_cpu(thi); 528144ed167dSPhilipp Reisner 528244ed167dSPhilipp Reisner rcu_read_lock(); 5283bde89a9eSAndreas Gruenbacher nc = rcu_dereference(connection->net_conf); 528444ed167dSPhilipp Reisner ping_timeo = nc->ping_timeo; 5285bb77d34eSAndreas Gruenbacher tcp_cork = nc->tcp_cork; 528644ed167dSPhilipp Reisner ping_int = nc->ping_int; 528744ed167dSPhilipp Reisner rcu_read_unlock(); 528844ed167dSPhilipp Reisner 5289bde89a9eSAndreas Gruenbacher if (test_and_clear_bit(SEND_PING, &connection->flags)) { 5290bde89a9eSAndreas Gruenbacher if (drbd_send_ping(connection)) { 52911ec861ebSAndreas Gruenbacher drbd_err(connection, "drbd_send_ping has failed\n"); 5292841ce241SAndreas Gruenbacher goto reconnect; 5293841ce241SAndreas Gruenbacher } 5294bde89a9eSAndreas Gruenbacher connection->meta.socket->sk->sk_rcvtimeo = ping_timeo * HZ / 10; 529544ed167dSPhilipp Reisner ping_timeout_active = true; 5296b411b363SPhilipp Reisner } 5297b411b363SPhilipp Reisner 529832862ec7SPhilipp Reisner /* TODO: conditionally cork; it may hurt latency if we cork without 529932862ec7SPhilipp Reisner much to send */ 5300bb77d34eSAndreas Gruenbacher if (tcp_cork) 5301bde89a9eSAndreas Gruenbacher drbd_tcp_cork(connection->meta.socket); 5302bde89a9eSAndreas Gruenbacher if (connection_finish_peer_reqs(connection)) { 53031ec861ebSAndreas Gruenbacher drbd_err(connection, "connection_finish_peer_reqs() failed\n"); 5304b411b363SPhilipp Reisner goto reconnect; 5305b411b363SPhilipp Reisner } 5306b411b363SPhilipp Reisner /* but unconditionally uncork unless disabled */ 5307bb77d34eSAndreas Gruenbacher if (tcp_cork) 5308bde89a9eSAndreas Gruenbacher drbd_tcp_uncork(connection->meta.socket); 5309b411b363SPhilipp Reisner 5310b411b363SPhilipp Reisner /* short circuit, recv_msg would return EINTR anyways. */ 5311b411b363SPhilipp Reisner if (signal_pending(current)) 5312b411b363SPhilipp Reisner continue; 5313b411b363SPhilipp Reisner 5314bde89a9eSAndreas Gruenbacher rv = drbd_recv_short(connection->meta.socket, buf, expect-received, 0); 5315bde89a9eSAndreas Gruenbacher clear_bit(SIGNAL_ASENDER, &connection->flags); 5316b411b363SPhilipp Reisner 5317b411b363SPhilipp Reisner flush_signals(current); 5318b411b363SPhilipp Reisner 5319b411b363SPhilipp Reisner /* Note: 5320b411b363SPhilipp Reisner * -EINTR (on meta) we got a signal 5321b411b363SPhilipp Reisner * -EAGAIN (on meta) rcvtimeo expired 5322b411b363SPhilipp Reisner * -ECONNRESET other side closed the connection 5323b411b363SPhilipp Reisner * -ERESTARTSYS (on data) we got a signal 5324b411b363SPhilipp Reisner * rv < 0 other than above: unexpected error! 5325b411b363SPhilipp Reisner * rv == expected: full header or command 5326b411b363SPhilipp Reisner * rv < expected: "woken" by signal during receive 5327b411b363SPhilipp Reisner * rv == 0 : "connection shut down by peer" 5328b411b363SPhilipp Reisner */ 5329b411b363SPhilipp Reisner if (likely(rv > 0)) { 5330b411b363SPhilipp Reisner received += rv; 5331b411b363SPhilipp Reisner buf += rv; 5332b411b363SPhilipp Reisner } else if (rv == 0) { 5333bde89a9eSAndreas Gruenbacher if (test_bit(DISCONNECT_SENT, &connection->flags)) { 5334b66623e3SPhilipp Reisner long t; 5335b66623e3SPhilipp Reisner rcu_read_lock(); 5336bde89a9eSAndreas Gruenbacher t = rcu_dereference(connection->net_conf)->ping_timeo * HZ/10; 5337b66623e3SPhilipp Reisner rcu_read_unlock(); 5338b66623e3SPhilipp Reisner 5339bde89a9eSAndreas Gruenbacher t = wait_event_timeout(connection->ping_wait, 5340bde89a9eSAndreas Gruenbacher connection->cstate < C_WF_REPORT_PARAMS, 5341b66623e3SPhilipp Reisner t); 5342599377acSPhilipp Reisner if (t) 5343599377acSPhilipp Reisner break; 5344599377acSPhilipp Reisner } 53451ec861ebSAndreas Gruenbacher drbd_err(connection, "meta connection shut down by peer.\n"); 5346b411b363SPhilipp Reisner goto reconnect; 5347b411b363SPhilipp Reisner } else if (rv == -EAGAIN) { 5348cb6518cbSLars Ellenberg /* If the data socket received something meanwhile, 5349cb6518cbSLars Ellenberg * that is good enough: peer is still alive. */ 5350bde89a9eSAndreas Gruenbacher if (time_after(connection->last_received, 5351bde89a9eSAndreas Gruenbacher jiffies - connection->meta.socket->sk->sk_rcvtimeo)) 5352cb6518cbSLars Ellenberg continue; 5353f36af18cSLars Ellenberg if (ping_timeout_active) { 53541ec861ebSAndreas Gruenbacher drbd_err(connection, "PingAck did not arrive in time.\n"); 5355b411b363SPhilipp Reisner goto reconnect; 5356b411b363SPhilipp Reisner } 5357bde89a9eSAndreas Gruenbacher set_bit(SEND_PING, &connection->flags); 5358b411b363SPhilipp Reisner continue; 5359b411b363SPhilipp Reisner } else if (rv == -EINTR) { 5360b411b363SPhilipp Reisner continue; 5361b411b363SPhilipp Reisner } else { 53621ec861ebSAndreas Gruenbacher drbd_err(connection, "sock_recvmsg returned %d\n", rv); 5363b411b363SPhilipp Reisner goto reconnect; 5364b411b363SPhilipp Reisner } 5365b411b363SPhilipp Reisner 5366b411b363SPhilipp Reisner if (received == expect && cmd == NULL) { 5367bde89a9eSAndreas Gruenbacher if (decode_header(connection, connection->meta.rbuf, &pi)) 5368b411b363SPhilipp Reisner goto reconnect; 53697201b972SAndreas Gruenbacher cmd = &asender_tbl[pi.cmd]; 53701952e916SAndreas Gruenbacher if (pi.cmd >= ARRAY_SIZE(asender_tbl) || !cmd->fn) { 53711ec861ebSAndreas Gruenbacher drbd_err(connection, "Unexpected meta packet %s (0x%04x)\n", 53722fcb8f30SAndreas Gruenbacher cmdname(pi.cmd), pi.cmd); 5373b411b363SPhilipp Reisner goto disconnect; 5374b411b363SPhilipp Reisner } 5375e658983aSAndreas Gruenbacher expect = header_size + cmd->pkt_size; 537652b061a4SAndreas Gruenbacher if (pi.size != expect - header_size) { 53771ec861ebSAndreas Gruenbacher drbd_err(connection, "Wrong packet size on meta (c: %d, l: %d)\n", 537877351055SPhilipp Reisner pi.cmd, pi.size); 5379b411b363SPhilipp Reisner goto reconnect; 5380b411b363SPhilipp Reisner } 5381257d0af6SPhilipp Reisner } 5382b411b363SPhilipp Reisner if (received == expect) { 53832735a594SAndreas Gruenbacher bool err; 5384a4fbda8eSPhilipp Reisner 5385bde89a9eSAndreas Gruenbacher err = cmd->fn(connection, &pi); 53862735a594SAndreas Gruenbacher if (err) { 53871ec861ebSAndreas Gruenbacher drbd_err(connection, "%pf failed\n", cmd->fn); 5388b411b363SPhilipp Reisner goto reconnect; 53891952e916SAndreas Gruenbacher } 5390b411b363SPhilipp Reisner 5391bde89a9eSAndreas Gruenbacher connection->last_received = jiffies; 5392f36af18cSLars Ellenberg 539344ed167dSPhilipp Reisner if (cmd == &asender_tbl[P_PING_ACK]) { 539444ed167dSPhilipp Reisner /* restore idle timeout */ 5395bde89a9eSAndreas Gruenbacher connection->meta.socket->sk->sk_rcvtimeo = ping_int * HZ; 539644ed167dSPhilipp Reisner ping_timeout_active = false; 539744ed167dSPhilipp Reisner } 5398b411b363SPhilipp Reisner 5399bde89a9eSAndreas Gruenbacher buf = connection->meta.rbuf; 5400b411b363SPhilipp Reisner received = 0; 540152b061a4SAndreas Gruenbacher expect = header_size; 5402b411b363SPhilipp Reisner cmd = NULL; 5403b411b363SPhilipp Reisner } 5404b411b363SPhilipp Reisner } 5405b411b363SPhilipp Reisner 5406b411b363SPhilipp Reisner if (0) { 5407b411b363SPhilipp Reisner reconnect: 5408bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_NETWORK_FAILURE), CS_HARD); 5409bde89a9eSAndreas Gruenbacher conn_md_sync(connection); 5410b411b363SPhilipp Reisner } 5411b411b363SPhilipp Reisner if (0) { 5412b411b363SPhilipp Reisner disconnect: 5413bde89a9eSAndreas Gruenbacher conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_HARD); 5414b411b363SPhilipp Reisner } 5415bde89a9eSAndreas Gruenbacher clear_bit(SIGNAL_ASENDER, &connection->flags); 5416b411b363SPhilipp Reisner 54171ec861ebSAndreas Gruenbacher drbd_info(connection, "asender terminated\n"); 5418b411b363SPhilipp Reisner 5419b411b363SPhilipp Reisner return 0; 5420b411b363SPhilipp Reisner } 5421