1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 drbd.c 4 5 This file is part of DRBD by Philipp Reisner and Lars Ellenberg. 6 7 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. 8 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. 9 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. 10 11 Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev 12 from Logicworks, Inc. for making SDP replication support possible. 13 14 15 */ 16 17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 18 19 #include <linux/module.h> 20 #include <linux/jiffies.h> 21 #include <linux/drbd.h> 22 #include <linux/uaccess.h> 23 #include <asm/types.h> 24 #include <net/sock.h> 25 #include <linux/ctype.h> 26 #include <linux/mutex.h> 27 #include <linux/fs.h> 28 #include <linux/file.h> 29 #include <linux/proc_fs.h> 30 #include <linux/init.h> 31 #include <linux/mm.h> 32 #include <linux/memcontrol.h> 33 #include <linux/mm_inline.h> 34 #include <linux/slab.h> 35 #include <linux/random.h> 36 #include <linux/reboot.h> 37 #include <linux/notifier.h> 38 #include <linux/kthread.h> 39 #include <linux/workqueue.h> 40 #define __KERNEL_SYSCALLS__ 41 #include <linux/unistd.h> 42 #include <linux/vmalloc.h> 43 #include <linux/sched/signal.h> 44 45 #include <linux/drbd_limits.h> 46 #include "drbd_int.h" 47 #include "drbd_protocol.h" 48 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */ 49 #include "drbd_vli.h" 50 #include "drbd_debugfs.h" 51 52 static DEFINE_MUTEX(drbd_main_mutex); 53 static int drbd_open(struct block_device *bdev, fmode_t mode); 54 static void drbd_release(struct gendisk *gd, fmode_t mode); 55 static void md_sync_timer_fn(struct timer_list *t); 56 static int w_bitmap_io(struct drbd_work *w, int unused); 57 58 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, " 59 "Lars Ellenberg <lars@linbit.com>"); 60 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION); 61 MODULE_VERSION(REL_VERSION); 62 MODULE_LICENSE("GPL"); 63 MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices (" 64 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")"); 65 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR); 66 67 #include <linux/moduleparam.h> 68 /* thanks to these macros, if compiled into the kernel (not-module), 69 * these become boot parameters (e.g., drbd.minor_count) */ 70 71 #ifdef CONFIG_DRBD_FAULT_INJECTION 72 int drbd_enable_faults; 73 int drbd_fault_rate; 74 static int drbd_fault_count; 75 static int drbd_fault_devs; 76 /* bitmap of enabled faults */ 77 module_param_named(enable_faults, drbd_enable_faults, int, 0664); 78 /* fault rate % value - applies to all enabled faults */ 79 module_param_named(fault_rate, drbd_fault_rate, int, 0664); 80 /* count of faults inserted */ 81 module_param_named(fault_count, drbd_fault_count, int, 0664); 82 /* bitmap of devices to insert faults on */ 83 module_param_named(fault_devs, drbd_fault_devs, int, 0644); 84 #endif 85 86 /* module parameters we can keep static */ 87 static bool drbd_allow_oos; /* allow_open_on_secondary */ 88 static bool drbd_disable_sendpage; 89 MODULE_PARM_DESC(allow_oos, "DONT USE!"); 90 module_param_named(allow_oos, drbd_allow_oos, bool, 0); 91 module_param_named(disable_sendpage, drbd_disable_sendpage, bool, 0644); 92 93 /* module parameters we share */ 94 int drbd_proc_details; /* Detail level in proc drbd*/ 95 module_param_named(proc_details, drbd_proc_details, int, 0644); 96 /* module parameters shared with defaults */ 97 unsigned int drbd_minor_count = DRBD_MINOR_COUNT_DEF; 98 /* Module parameter for setting the user mode helper program 99 * to run. Default is /sbin/drbdadm */ 100 char drbd_usermode_helper[80] = "/sbin/drbdadm"; 101 module_param_named(minor_count, drbd_minor_count, uint, 0444); 102 module_param_string(usermode_helper, drbd_usermode_helper, sizeof(drbd_usermode_helper), 0644); 103 104 /* in 2.6.x, our device mapping and config info contains our virtual gendisks 105 * as member "struct gendisk *vdisk;" 106 */ 107 struct idr drbd_devices; 108 struct list_head drbd_resources; 109 struct mutex resources_mutex; 110 111 struct kmem_cache *drbd_request_cache; 112 struct kmem_cache *drbd_ee_cache; /* peer requests */ 113 struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */ 114 struct kmem_cache *drbd_al_ext_cache; /* activity log extents */ 115 mempool_t drbd_request_mempool; 116 mempool_t drbd_ee_mempool; 117 mempool_t drbd_md_io_page_pool; 118 struct bio_set drbd_md_io_bio_set; 119 struct bio_set drbd_io_bio_set; 120 121 /* I do not use a standard mempool, because: 122 1) I want to hand out the pre-allocated objects first. 123 2) I want to be able to interrupt sleeping allocation with a signal. 124 Note: This is a single linked list, the next pointer is the private 125 member of struct page. 126 */ 127 struct page *drbd_pp_pool; 128 spinlock_t drbd_pp_lock; 129 int drbd_pp_vacant; 130 wait_queue_head_t drbd_pp_wait; 131 132 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5); 133 134 static const struct block_device_operations drbd_ops = { 135 .owner = THIS_MODULE, 136 .submit_bio = drbd_submit_bio, 137 .open = drbd_open, 138 .release = drbd_release, 139 }; 140 141 struct bio *bio_alloc_drbd(gfp_t gfp_mask) 142 { 143 struct bio *bio; 144 145 if (!bioset_initialized(&drbd_md_io_bio_set)) 146 return bio_alloc(gfp_mask, 1); 147 148 bio = bio_alloc_bioset(gfp_mask, 1, &drbd_md_io_bio_set); 149 if (!bio) 150 return NULL; 151 return bio; 152 } 153 154 #ifdef __CHECKER__ 155 /* When checking with sparse, and this is an inline function, sparse will 156 give tons of false positives. When this is a real functions sparse works. 157 */ 158 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins) 159 { 160 int io_allowed; 161 162 atomic_inc(&device->local_cnt); 163 io_allowed = (device->state.disk >= mins); 164 if (!io_allowed) { 165 if (atomic_dec_and_test(&device->local_cnt)) 166 wake_up(&device->misc_wait); 167 } 168 return io_allowed; 169 } 170 171 #endif 172 173 /** 174 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch 175 * @connection: DRBD connection. 176 * @barrier_nr: Expected identifier of the DRBD write barrier packet. 177 * @set_size: Expected number of requests before that barrier. 178 * 179 * In case the passed barrier_nr or set_size does not match the oldest 180 * epoch of not yet barrier-acked requests, this function will cause a 181 * termination of the connection. 182 */ 183 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr, 184 unsigned int set_size) 185 { 186 struct drbd_request *r; 187 struct drbd_request *req = NULL; 188 int expect_epoch = 0; 189 int expect_size = 0; 190 191 spin_lock_irq(&connection->resource->req_lock); 192 193 /* find oldest not yet barrier-acked write request, 194 * count writes in its epoch. */ 195 list_for_each_entry(r, &connection->transfer_log, tl_requests) { 196 const unsigned s = r->rq_state; 197 if (!req) { 198 if (!(s & RQ_WRITE)) 199 continue; 200 if (!(s & RQ_NET_MASK)) 201 continue; 202 if (s & RQ_NET_DONE) 203 continue; 204 req = r; 205 expect_epoch = req->epoch; 206 expect_size ++; 207 } else { 208 if (r->epoch != expect_epoch) 209 break; 210 if (!(s & RQ_WRITE)) 211 continue; 212 /* if (s & RQ_DONE): not expected */ 213 /* if (!(s & RQ_NET_MASK)): not expected */ 214 expect_size++; 215 } 216 } 217 218 /* first some paranoia code */ 219 if (req == NULL) { 220 drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n", 221 barrier_nr); 222 goto bail; 223 } 224 if (expect_epoch != barrier_nr) { 225 drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n", 226 barrier_nr, expect_epoch); 227 goto bail; 228 } 229 230 if (expect_size != set_size) { 231 drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n", 232 barrier_nr, set_size, expect_size); 233 goto bail; 234 } 235 236 /* Clean up list of requests processed during current epoch. */ 237 /* this extra list walk restart is paranoia, 238 * to catch requests being barrier-acked "unexpectedly". 239 * It usually should find the same req again, or some READ preceding it. */ 240 list_for_each_entry(req, &connection->transfer_log, tl_requests) 241 if (req->epoch == expect_epoch) 242 break; 243 list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) { 244 if (req->epoch != expect_epoch) 245 break; 246 _req_mod(req, BARRIER_ACKED); 247 } 248 spin_unlock_irq(&connection->resource->req_lock); 249 250 return; 251 252 bail: 253 spin_unlock_irq(&connection->resource->req_lock); 254 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); 255 } 256 257 258 /** 259 * _tl_restart() - Walks the transfer log, and applies an action to all requests 260 * @connection: DRBD connection to operate on. 261 * @what: The action/event to perform with all request objects 262 * 263 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO, 264 * RESTART_FROZEN_DISK_IO. 265 */ 266 /* must hold resource->req_lock */ 267 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what) 268 { 269 struct drbd_request *req, *r; 270 271 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) 272 _req_mod(req, what); 273 } 274 275 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what) 276 { 277 spin_lock_irq(&connection->resource->req_lock); 278 _tl_restart(connection, what); 279 spin_unlock_irq(&connection->resource->req_lock); 280 } 281 282 /** 283 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL 284 * @device: DRBD device. 285 * 286 * This is called after the connection to the peer was lost. The storage covered 287 * by the requests on the transfer gets marked as our of sync. Called from the 288 * receiver thread and the worker thread. 289 */ 290 void tl_clear(struct drbd_connection *connection) 291 { 292 tl_restart(connection, CONNECTION_LOST_WHILE_PENDING); 293 } 294 295 /** 296 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL 297 * @device: DRBD device. 298 */ 299 void tl_abort_disk_io(struct drbd_device *device) 300 { 301 struct drbd_connection *connection = first_peer_device(device)->connection; 302 struct drbd_request *req, *r; 303 304 spin_lock_irq(&connection->resource->req_lock); 305 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) { 306 if (!(req->rq_state & RQ_LOCAL_PENDING)) 307 continue; 308 if (req->device != device) 309 continue; 310 _req_mod(req, ABORT_DISK_IO); 311 } 312 spin_unlock_irq(&connection->resource->req_lock); 313 } 314 315 static int drbd_thread_setup(void *arg) 316 { 317 struct drbd_thread *thi = (struct drbd_thread *) arg; 318 struct drbd_resource *resource = thi->resource; 319 unsigned long flags; 320 int retval; 321 322 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s", 323 thi->name[0], 324 resource->name); 325 326 allow_kernel_signal(DRBD_SIGKILL); 327 allow_kernel_signal(SIGXCPU); 328 restart: 329 retval = thi->function(thi); 330 331 spin_lock_irqsave(&thi->t_lock, flags); 332 333 /* if the receiver has been "EXITING", the last thing it did 334 * was set the conn state to "StandAlone", 335 * if now a re-connect request comes in, conn state goes C_UNCONNECTED, 336 * and receiver thread will be "started". 337 * drbd_thread_start needs to set "RESTARTING" in that case. 338 * t_state check and assignment needs to be within the same spinlock, 339 * so either thread_start sees EXITING, and can remap to RESTARTING, 340 * or thread_start see NONE, and can proceed as normal. 341 */ 342 343 if (thi->t_state == RESTARTING) { 344 drbd_info(resource, "Restarting %s thread\n", thi->name); 345 thi->t_state = RUNNING; 346 spin_unlock_irqrestore(&thi->t_lock, flags); 347 goto restart; 348 } 349 350 thi->task = NULL; 351 thi->t_state = NONE; 352 smp_mb(); 353 complete_all(&thi->stop); 354 spin_unlock_irqrestore(&thi->t_lock, flags); 355 356 drbd_info(resource, "Terminating %s\n", current->comm); 357 358 /* Release mod reference taken when thread was started */ 359 360 if (thi->connection) 361 kref_put(&thi->connection->kref, drbd_destroy_connection); 362 kref_put(&resource->kref, drbd_destroy_resource); 363 module_put(THIS_MODULE); 364 return retval; 365 } 366 367 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi, 368 int (*func) (struct drbd_thread *), const char *name) 369 { 370 spin_lock_init(&thi->t_lock); 371 thi->task = NULL; 372 thi->t_state = NONE; 373 thi->function = func; 374 thi->resource = resource; 375 thi->connection = NULL; 376 thi->name = name; 377 } 378 379 int drbd_thread_start(struct drbd_thread *thi) 380 { 381 struct drbd_resource *resource = thi->resource; 382 struct task_struct *nt; 383 unsigned long flags; 384 385 /* is used from state engine doing drbd_thread_stop_nowait, 386 * while holding the req lock irqsave */ 387 spin_lock_irqsave(&thi->t_lock, flags); 388 389 switch (thi->t_state) { 390 case NONE: 391 drbd_info(resource, "Starting %s thread (from %s [%d])\n", 392 thi->name, current->comm, current->pid); 393 394 /* Get ref on module for thread - this is released when thread exits */ 395 if (!try_module_get(THIS_MODULE)) { 396 drbd_err(resource, "Failed to get module reference in drbd_thread_start\n"); 397 spin_unlock_irqrestore(&thi->t_lock, flags); 398 return false; 399 } 400 401 kref_get(&resource->kref); 402 if (thi->connection) 403 kref_get(&thi->connection->kref); 404 405 init_completion(&thi->stop); 406 thi->reset_cpu_mask = 1; 407 thi->t_state = RUNNING; 408 spin_unlock_irqrestore(&thi->t_lock, flags); 409 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */ 410 411 nt = kthread_create(drbd_thread_setup, (void *) thi, 412 "drbd_%c_%s", thi->name[0], thi->resource->name); 413 414 if (IS_ERR(nt)) { 415 drbd_err(resource, "Couldn't start thread\n"); 416 417 if (thi->connection) 418 kref_put(&thi->connection->kref, drbd_destroy_connection); 419 kref_put(&resource->kref, drbd_destroy_resource); 420 module_put(THIS_MODULE); 421 return false; 422 } 423 spin_lock_irqsave(&thi->t_lock, flags); 424 thi->task = nt; 425 thi->t_state = RUNNING; 426 spin_unlock_irqrestore(&thi->t_lock, flags); 427 wake_up_process(nt); 428 break; 429 case EXITING: 430 thi->t_state = RESTARTING; 431 drbd_info(resource, "Restarting %s thread (from %s [%d])\n", 432 thi->name, current->comm, current->pid); 433 fallthrough; 434 case RUNNING: 435 case RESTARTING: 436 default: 437 spin_unlock_irqrestore(&thi->t_lock, flags); 438 break; 439 } 440 441 return true; 442 } 443 444 445 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait) 446 { 447 unsigned long flags; 448 449 enum drbd_thread_state ns = restart ? RESTARTING : EXITING; 450 451 /* may be called from state engine, holding the req lock irqsave */ 452 spin_lock_irqsave(&thi->t_lock, flags); 453 454 if (thi->t_state == NONE) { 455 spin_unlock_irqrestore(&thi->t_lock, flags); 456 if (restart) 457 drbd_thread_start(thi); 458 return; 459 } 460 461 if (thi->t_state != ns) { 462 if (thi->task == NULL) { 463 spin_unlock_irqrestore(&thi->t_lock, flags); 464 return; 465 } 466 467 thi->t_state = ns; 468 smp_mb(); 469 init_completion(&thi->stop); 470 if (thi->task != current) 471 send_sig(DRBD_SIGKILL, thi->task, 1); 472 } 473 474 spin_unlock_irqrestore(&thi->t_lock, flags); 475 476 if (wait) 477 wait_for_completion(&thi->stop); 478 } 479 480 int conn_lowest_minor(struct drbd_connection *connection) 481 { 482 struct drbd_peer_device *peer_device; 483 int vnr = 0, minor = -1; 484 485 rcu_read_lock(); 486 peer_device = idr_get_next(&connection->peer_devices, &vnr); 487 if (peer_device) 488 minor = device_to_minor(peer_device->device); 489 rcu_read_unlock(); 490 491 return minor; 492 } 493 494 #ifdef CONFIG_SMP 495 /** 496 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs 497 * 498 * Forces all threads of a resource onto the same CPU. This is beneficial for 499 * DRBD's performance. May be overwritten by user's configuration. 500 */ 501 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask) 502 { 503 unsigned int *resources_per_cpu, min_index = ~0; 504 505 resources_per_cpu = kcalloc(nr_cpu_ids, sizeof(*resources_per_cpu), 506 GFP_KERNEL); 507 if (resources_per_cpu) { 508 struct drbd_resource *resource; 509 unsigned int cpu, min = ~0; 510 511 rcu_read_lock(); 512 for_each_resource_rcu(resource, &drbd_resources) { 513 for_each_cpu(cpu, resource->cpu_mask) 514 resources_per_cpu[cpu]++; 515 } 516 rcu_read_unlock(); 517 for_each_online_cpu(cpu) { 518 if (resources_per_cpu[cpu] < min) { 519 min = resources_per_cpu[cpu]; 520 min_index = cpu; 521 } 522 } 523 kfree(resources_per_cpu); 524 } 525 if (min_index == ~0) { 526 cpumask_setall(*cpu_mask); 527 return; 528 } 529 cpumask_set_cpu(min_index, *cpu_mask); 530 } 531 532 /** 533 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread 534 * @device: DRBD device. 535 * @thi: drbd_thread object 536 * 537 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die 538 * prematurely. 539 */ 540 void drbd_thread_current_set_cpu(struct drbd_thread *thi) 541 { 542 struct drbd_resource *resource = thi->resource; 543 struct task_struct *p = current; 544 545 if (!thi->reset_cpu_mask) 546 return; 547 thi->reset_cpu_mask = 0; 548 set_cpus_allowed_ptr(p, resource->cpu_mask); 549 } 550 #else 551 #define drbd_calc_cpu_mask(A) ({}) 552 #endif 553 554 /** 555 * drbd_header_size - size of a packet header 556 * 557 * The header size is a multiple of 8, so any payload following the header is 558 * word aligned on 64-bit architectures. (The bitmap send and receive code 559 * relies on this.) 560 */ 561 unsigned int drbd_header_size(struct drbd_connection *connection) 562 { 563 if (connection->agreed_pro_version >= 100) { 564 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8)); 565 return sizeof(struct p_header100); 566 } else { 567 BUILD_BUG_ON(sizeof(struct p_header80) != 568 sizeof(struct p_header95)); 569 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8)); 570 return sizeof(struct p_header80); 571 } 572 } 573 574 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size) 575 { 576 h->magic = cpu_to_be32(DRBD_MAGIC); 577 h->command = cpu_to_be16(cmd); 578 h->length = cpu_to_be16(size); 579 return sizeof(struct p_header80); 580 } 581 582 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size) 583 { 584 h->magic = cpu_to_be16(DRBD_MAGIC_BIG); 585 h->command = cpu_to_be16(cmd); 586 h->length = cpu_to_be32(size); 587 return sizeof(struct p_header95); 588 } 589 590 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd, 591 int size, int vnr) 592 { 593 h->magic = cpu_to_be32(DRBD_MAGIC_100); 594 h->volume = cpu_to_be16(vnr); 595 h->command = cpu_to_be16(cmd); 596 h->length = cpu_to_be32(size); 597 h->pad = 0; 598 return sizeof(struct p_header100); 599 } 600 601 static unsigned int prepare_header(struct drbd_connection *connection, int vnr, 602 void *buffer, enum drbd_packet cmd, int size) 603 { 604 if (connection->agreed_pro_version >= 100) 605 return prepare_header100(buffer, cmd, size, vnr); 606 else if (connection->agreed_pro_version >= 95 && 607 size > DRBD_MAX_SIZE_H80_PACKET) 608 return prepare_header95(buffer, cmd, size); 609 else 610 return prepare_header80(buffer, cmd, size); 611 } 612 613 static void *__conn_prepare_command(struct drbd_connection *connection, 614 struct drbd_socket *sock) 615 { 616 if (!sock->socket) 617 return NULL; 618 return sock->sbuf + drbd_header_size(connection); 619 } 620 621 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock) 622 { 623 void *p; 624 625 mutex_lock(&sock->mutex); 626 p = __conn_prepare_command(connection, sock); 627 if (!p) 628 mutex_unlock(&sock->mutex); 629 630 return p; 631 } 632 633 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock) 634 { 635 return conn_prepare_command(peer_device->connection, sock); 636 } 637 638 static int __send_command(struct drbd_connection *connection, int vnr, 639 struct drbd_socket *sock, enum drbd_packet cmd, 640 unsigned int header_size, void *data, 641 unsigned int size) 642 { 643 int msg_flags; 644 int err; 645 646 /* 647 * Called with @data == NULL and the size of the data blocks in @size 648 * for commands that send data blocks. For those commands, omit the 649 * MSG_MORE flag: this will increase the likelihood that data blocks 650 * which are page aligned on the sender will end up page aligned on the 651 * receiver. 652 */ 653 msg_flags = data ? MSG_MORE : 0; 654 655 header_size += prepare_header(connection, vnr, sock->sbuf, cmd, 656 header_size + size); 657 err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size, 658 msg_flags); 659 if (data && !err) 660 err = drbd_send_all(connection, sock->socket, data, size, 0); 661 /* DRBD protocol "pings" are latency critical. 662 * This is supposed to trigger tcp_push_pending_frames() */ 663 if (!err && (cmd == P_PING || cmd == P_PING_ACK)) 664 tcp_sock_set_nodelay(sock->socket->sk); 665 666 return err; 667 } 668 669 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock, 670 enum drbd_packet cmd, unsigned int header_size, 671 void *data, unsigned int size) 672 { 673 return __send_command(connection, 0, sock, cmd, header_size, data, size); 674 } 675 676 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock, 677 enum drbd_packet cmd, unsigned int header_size, 678 void *data, unsigned int size) 679 { 680 int err; 681 682 err = __conn_send_command(connection, sock, cmd, header_size, data, size); 683 mutex_unlock(&sock->mutex); 684 return err; 685 } 686 687 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock, 688 enum drbd_packet cmd, unsigned int header_size, 689 void *data, unsigned int size) 690 { 691 int err; 692 693 err = __send_command(peer_device->connection, peer_device->device->vnr, 694 sock, cmd, header_size, data, size); 695 mutex_unlock(&sock->mutex); 696 return err; 697 } 698 699 int drbd_send_ping(struct drbd_connection *connection) 700 { 701 struct drbd_socket *sock; 702 703 sock = &connection->meta; 704 if (!conn_prepare_command(connection, sock)) 705 return -EIO; 706 return conn_send_command(connection, sock, P_PING, 0, NULL, 0); 707 } 708 709 int drbd_send_ping_ack(struct drbd_connection *connection) 710 { 711 struct drbd_socket *sock; 712 713 sock = &connection->meta; 714 if (!conn_prepare_command(connection, sock)) 715 return -EIO; 716 return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0); 717 } 718 719 int drbd_send_sync_param(struct drbd_peer_device *peer_device) 720 { 721 struct drbd_socket *sock; 722 struct p_rs_param_95 *p; 723 int size; 724 const int apv = peer_device->connection->agreed_pro_version; 725 enum drbd_packet cmd; 726 struct net_conf *nc; 727 struct disk_conf *dc; 728 729 sock = &peer_device->connection->data; 730 p = drbd_prepare_command(peer_device, sock); 731 if (!p) 732 return -EIO; 733 734 rcu_read_lock(); 735 nc = rcu_dereference(peer_device->connection->net_conf); 736 737 size = apv <= 87 ? sizeof(struct p_rs_param) 738 : apv == 88 ? sizeof(struct p_rs_param) 739 + strlen(nc->verify_alg) + 1 740 : apv <= 94 ? sizeof(struct p_rs_param_89) 741 : /* apv >= 95 */ sizeof(struct p_rs_param_95); 742 743 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM; 744 745 /* initialize verify_alg and csums_alg */ 746 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); 747 748 if (get_ldev(peer_device->device)) { 749 dc = rcu_dereference(peer_device->device->ldev->disk_conf); 750 p->resync_rate = cpu_to_be32(dc->resync_rate); 751 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead); 752 p->c_delay_target = cpu_to_be32(dc->c_delay_target); 753 p->c_fill_target = cpu_to_be32(dc->c_fill_target); 754 p->c_max_rate = cpu_to_be32(dc->c_max_rate); 755 put_ldev(peer_device->device); 756 } else { 757 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF); 758 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF); 759 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF); 760 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF); 761 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF); 762 } 763 764 if (apv >= 88) 765 strcpy(p->verify_alg, nc->verify_alg); 766 if (apv >= 89) 767 strcpy(p->csums_alg, nc->csums_alg); 768 rcu_read_unlock(); 769 770 return drbd_send_command(peer_device, sock, cmd, size, NULL, 0); 771 } 772 773 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd) 774 { 775 struct drbd_socket *sock; 776 struct p_protocol *p; 777 struct net_conf *nc; 778 int size, cf; 779 780 sock = &connection->data; 781 p = __conn_prepare_command(connection, sock); 782 if (!p) 783 return -EIO; 784 785 rcu_read_lock(); 786 nc = rcu_dereference(connection->net_conf); 787 788 if (nc->tentative && connection->agreed_pro_version < 92) { 789 rcu_read_unlock(); 790 drbd_err(connection, "--dry-run is not supported by peer"); 791 return -EOPNOTSUPP; 792 } 793 794 size = sizeof(*p); 795 if (connection->agreed_pro_version >= 87) 796 size += strlen(nc->integrity_alg) + 1; 797 798 p->protocol = cpu_to_be32(nc->wire_protocol); 799 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p); 800 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p); 801 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p); 802 p->two_primaries = cpu_to_be32(nc->two_primaries); 803 cf = 0; 804 if (nc->discard_my_data) 805 cf |= CF_DISCARD_MY_DATA; 806 if (nc->tentative) 807 cf |= CF_DRY_RUN; 808 p->conn_flags = cpu_to_be32(cf); 809 810 if (connection->agreed_pro_version >= 87) 811 strcpy(p->integrity_alg, nc->integrity_alg); 812 rcu_read_unlock(); 813 814 return __conn_send_command(connection, sock, cmd, size, NULL, 0); 815 } 816 817 int drbd_send_protocol(struct drbd_connection *connection) 818 { 819 int err; 820 821 mutex_lock(&connection->data.mutex); 822 err = __drbd_send_protocol(connection, P_PROTOCOL); 823 mutex_unlock(&connection->data.mutex); 824 825 return err; 826 } 827 828 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags) 829 { 830 struct drbd_device *device = peer_device->device; 831 struct drbd_socket *sock; 832 struct p_uuids *p; 833 int i; 834 835 if (!get_ldev_if_state(device, D_NEGOTIATING)) 836 return 0; 837 838 sock = &peer_device->connection->data; 839 p = drbd_prepare_command(peer_device, sock); 840 if (!p) { 841 put_ldev(device); 842 return -EIO; 843 } 844 spin_lock_irq(&device->ldev->md.uuid_lock); 845 for (i = UI_CURRENT; i < UI_SIZE; i++) 846 p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); 847 spin_unlock_irq(&device->ldev->md.uuid_lock); 848 849 device->comm_bm_set = drbd_bm_total_weight(device); 850 p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set); 851 rcu_read_lock(); 852 uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0; 853 rcu_read_unlock(); 854 uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0; 855 uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0; 856 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags); 857 858 put_ldev(device); 859 return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0); 860 } 861 862 int drbd_send_uuids(struct drbd_peer_device *peer_device) 863 { 864 return _drbd_send_uuids(peer_device, 0); 865 } 866 867 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device) 868 { 869 return _drbd_send_uuids(peer_device, 8); 870 } 871 872 void drbd_print_uuids(struct drbd_device *device, const char *text) 873 { 874 if (get_ldev_if_state(device, D_NEGOTIATING)) { 875 u64 *uuid = device->ldev->md.uuid; 876 drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n", 877 text, 878 (unsigned long long)uuid[UI_CURRENT], 879 (unsigned long long)uuid[UI_BITMAP], 880 (unsigned long long)uuid[UI_HISTORY_START], 881 (unsigned long long)uuid[UI_HISTORY_END]); 882 put_ldev(device); 883 } else { 884 drbd_info(device, "%s effective data uuid: %016llX\n", 885 text, 886 (unsigned long long)device->ed_uuid); 887 } 888 } 889 890 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device) 891 { 892 struct drbd_device *device = peer_device->device; 893 struct drbd_socket *sock; 894 struct p_rs_uuid *p; 895 u64 uuid; 896 897 D_ASSERT(device, device->state.disk == D_UP_TO_DATE); 898 899 uuid = device->ldev->md.uuid[UI_BITMAP]; 900 if (uuid && uuid != UUID_JUST_CREATED) 901 uuid = uuid + UUID_NEW_BM_OFFSET; 902 else 903 get_random_bytes(&uuid, sizeof(u64)); 904 drbd_uuid_set(device, UI_BITMAP, uuid); 905 drbd_print_uuids(device, "updated sync UUID"); 906 drbd_md_sync(device); 907 908 sock = &peer_device->connection->data; 909 p = drbd_prepare_command(peer_device, sock); 910 if (p) { 911 p->uuid = cpu_to_be64(uuid); 912 drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0); 913 } 914 } 915 916 /* communicated if (agreed_features & DRBD_FF_WSAME) */ 917 static void 918 assign_p_sizes_qlim(struct drbd_device *device, struct p_sizes *p, 919 struct request_queue *q) 920 { 921 if (q) { 922 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q)); 923 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q)); 924 p->qlim->alignment_offset = cpu_to_be32(queue_alignment_offset(q)); 925 p->qlim->io_min = cpu_to_be32(queue_io_min(q)); 926 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q)); 927 p->qlim->discard_enabled = blk_queue_discard(q); 928 p->qlim->write_same_capable = !!q->limits.max_write_same_sectors; 929 } else { 930 q = device->rq_queue; 931 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q)); 932 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q)); 933 p->qlim->alignment_offset = 0; 934 p->qlim->io_min = cpu_to_be32(queue_io_min(q)); 935 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q)); 936 p->qlim->discard_enabled = 0; 937 p->qlim->write_same_capable = 0; 938 } 939 } 940 941 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags) 942 { 943 struct drbd_device *device = peer_device->device; 944 struct drbd_socket *sock; 945 struct p_sizes *p; 946 sector_t d_size, u_size; 947 int q_order_type; 948 unsigned int max_bio_size; 949 unsigned int packet_size; 950 951 sock = &peer_device->connection->data; 952 p = drbd_prepare_command(peer_device, sock); 953 if (!p) 954 return -EIO; 955 956 packet_size = sizeof(*p); 957 if (peer_device->connection->agreed_features & DRBD_FF_WSAME) 958 packet_size += sizeof(p->qlim[0]); 959 960 memset(p, 0, packet_size); 961 if (get_ldev_if_state(device, D_NEGOTIATING)) { 962 struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev); 963 d_size = drbd_get_max_capacity(device->ldev); 964 rcu_read_lock(); 965 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size; 966 rcu_read_unlock(); 967 q_order_type = drbd_queue_order_type(device); 968 max_bio_size = queue_max_hw_sectors(q) << 9; 969 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE); 970 assign_p_sizes_qlim(device, p, q); 971 put_ldev(device); 972 } else { 973 d_size = 0; 974 u_size = 0; 975 q_order_type = QUEUE_ORDERED_NONE; 976 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */ 977 assign_p_sizes_qlim(device, p, NULL); 978 } 979 980 if (peer_device->connection->agreed_pro_version <= 94) 981 max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET); 982 else if (peer_device->connection->agreed_pro_version < 100) 983 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95); 984 985 p->d_size = cpu_to_be64(d_size); 986 p->u_size = cpu_to_be64(u_size); 987 if (trigger_reply) 988 p->c_size = 0; 989 else 990 p->c_size = cpu_to_be64(get_capacity(device->vdisk)); 991 p->max_bio_size = cpu_to_be32(max_bio_size); 992 p->queue_order_type = cpu_to_be16(q_order_type); 993 p->dds_flags = cpu_to_be16(flags); 994 995 return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0); 996 } 997 998 /** 999 * drbd_send_current_state() - Sends the drbd state to the peer 1000 * @peer_device: DRBD peer device. 1001 */ 1002 int drbd_send_current_state(struct drbd_peer_device *peer_device) 1003 { 1004 struct drbd_socket *sock; 1005 struct p_state *p; 1006 1007 sock = &peer_device->connection->data; 1008 p = drbd_prepare_command(peer_device, sock); 1009 if (!p) 1010 return -EIO; 1011 p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */ 1012 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0); 1013 } 1014 1015 /** 1016 * drbd_send_state() - After a state change, sends the new state to the peer 1017 * @peer_device: DRBD peer device. 1018 * @state: the state to send, not necessarily the current state. 1019 * 1020 * Each state change queues an "after_state_ch" work, which will eventually 1021 * send the resulting new state to the peer. If more state changes happen 1022 * between queuing and processing of the after_state_ch work, we still 1023 * want to send each intermediary state in the order it occurred. 1024 */ 1025 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state) 1026 { 1027 struct drbd_socket *sock; 1028 struct p_state *p; 1029 1030 sock = &peer_device->connection->data; 1031 p = drbd_prepare_command(peer_device, sock); 1032 if (!p) 1033 return -EIO; 1034 p->state = cpu_to_be32(state.i); /* Within the send mutex */ 1035 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0); 1036 } 1037 1038 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val) 1039 { 1040 struct drbd_socket *sock; 1041 struct p_req_state *p; 1042 1043 sock = &peer_device->connection->data; 1044 p = drbd_prepare_command(peer_device, sock); 1045 if (!p) 1046 return -EIO; 1047 p->mask = cpu_to_be32(mask.i); 1048 p->val = cpu_to_be32(val.i); 1049 return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0); 1050 } 1051 1052 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val) 1053 { 1054 enum drbd_packet cmd; 1055 struct drbd_socket *sock; 1056 struct p_req_state *p; 1057 1058 cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ; 1059 sock = &connection->data; 1060 p = conn_prepare_command(connection, sock); 1061 if (!p) 1062 return -EIO; 1063 p->mask = cpu_to_be32(mask.i); 1064 p->val = cpu_to_be32(val.i); 1065 return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); 1066 } 1067 1068 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode) 1069 { 1070 struct drbd_socket *sock; 1071 struct p_req_state_reply *p; 1072 1073 sock = &peer_device->connection->meta; 1074 p = drbd_prepare_command(peer_device, sock); 1075 if (p) { 1076 p->retcode = cpu_to_be32(retcode); 1077 drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0); 1078 } 1079 } 1080 1081 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode) 1082 { 1083 struct drbd_socket *sock; 1084 struct p_req_state_reply *p; 1085 enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY; 1086 1087 sock = &connection->meta; 1088 p = conn_prepare_command(connection, sock); 1089 if (p) { 1090 p->retcode = cpu_to_be32(retcode); 1091 conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); 1092 } 1093 } 1094 1095 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code) 1096 { 1097 BUG_ON(code & ~0xf); 1098 p->encoding = (p->encoding & ~0xf) | code; 1099 } 1100 1101 static void dcbp_set_start(struct p_compressed_bm *p, int set) 1102 { 1103 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0); 1104 } 1105 1106 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n) 1107 { 1108 BUG_ON(n & ~0x7); 1109 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4); 1110 } 1111 1112 static int fill_bitmap_rle_bits(struct drbd_device *device, 1113 struct p_compressed_bm *p, 1114 unsigned int size, 1115 struct bm_xfer_ctx *c) 1116 { 1117 struct bitstream bs; 1118 unsigned long plain_bits; 1119 unsigned long tmp; 1120 unsigned long rl; 1121 unsigned len; 1122 unsigned toggle; 1123 int bits, use_rle; 1124 1125 /* may we use this feature? */ 1126 rcu_read_lock(); 1127 use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle; 1128 rcu_read_unlock(); 1129 if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90) 1130 return 0; 1131 1132 if (c->bit_offset >= c->bm_bits) 1133 return 0; /* nothing to do. */ 1134 1135 /* use at most thus many bytes */ 1136 bitstream_init(&bs, p->code, size, 0); 1137 memset(p->code, 0, size); 1138 /* plain bits covered in this code string */ 1139 plain_bits = 0; 1140 1141 /* p->encoding & 0x80 stores whether the first run length is set. 1142 * bit offset is implicit. 1143 * start with toggle == 2 to be able to tell the first iteration */ 1144 toggle = 2; 1145 1146 /* see how much plain bits we can stuff into one packet 1147 * using RLE and VLI. */ 1148 do { 1149 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset) 1150 : _drbd_bm_find_next(device, c->bit_offset); 1151 if (tmp == -1UL) 1152 tmp = c->bm_bits; 1153 rl = tmp - c->bit_offset; 1154 1155 if (toggle == 2) { /* first iteration */ 1156 if (rl == 0) { 1157 /* the first checked bit was set, 1158 * store start value, */ 1159 dcbp_set_start(p, 1); 1160 /* but skip encoding of zero run length */ 1161 toggle = !toggle; 1162 continue; 1163 } 1164 dcbp_set_start(p, 0); 1165 } 1166 1167 /* paranoia: catch zero runlength. 1168 * can only happen if bitmap is modified while we scan it. */ 1169 if (rl == 0) { 1170 drbd_err(device, "unexpected zero runlength while encoding bitmap " 1171 "t:%u bo:%lu\n", toggle, c->bit_offset); 1172 return -1; 1173 } 1174 1175 bits = vli_encode_bits(&bs, rl); 1176 if (bits == -ENOBUFS) /* buffer full */ 1177 break; 1178 if (bits <= 0) { 1179 drbd_err(device, "error while encoding bitmap: %d\n", bits); 1180 return 0; 1181 } 1182 1183 toggle = !toggle; 1184 plain_bits += rl; 1185 c->bit_offset = tmp; 1186 } while (c->bit_offset < c->bm_bits); 1187 1188 len = bs.cur.b - p->code + !!bs.cur.bit; 1189 1190 if (plain_bits < (len << 3)) { 1191 /* incompressible with this method. 1192 * we need to rewind both word and bit position. */ 1193 c->bit_offset -= plain_bits; 1194 bm_xfer_ctx_bit_to_word_offset(c); 1195 c->bit_offset = c->word_offset * BITS_PER_LONG; 1196 return 0; 1197 } 1198 1199 /* RLE + VLI was able to compress it just fine. 1200 * update c->word_offset. */ 1201 bm_xfer_ctx_bit_to_word_offset(c); 1202 1203 /* store pad_bits */ 1204 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7); 1205 1206 return len; 1207 } 1208 1209 /** 1210 * send_bitmap_rle_or_plain 1211 * 1212 * Return 0 when done, 1 when another iteration is needed, and a negative error 1213 * code upon failure. 1214 */ 1215 static int 1216 send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c) 1217 { 1218 struct drbd_socket *sock = &first_peer_device(device)->connection->data; 1219 unsigned int header_size = drbd_header_size(first_peer_device(device)->connection); 1220 struct p_compressed_bm *p = sock->sbuf + header_size; 1221 int len, err; 1222 1223 len = fill_bitmap_rle_bits(device, p, 1224 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c); 1225 if (len < 0) 1226 return -EIO; 1227 1228 if (len) { 1229 dcbp_set_code(p, RLE_VLI_Bits); 1230 err = __send_command(first_peer_device(device)->connection, device->vnr, sock, 1231 P_COMPRESSED_BITMAP, sizeof(*p) + len, 1232 NULL, 0); 1233 c->packets[0]++; 1234 c->bytes[0] += header_size + sizeof(*p) + len; 1235 1236 if (c->bit_offset >= c->bm_bits) 1237 len = 0; /* DONE */ 1238 } else { 1239 /* was not compressible. 1240 * send a buffer full of plain text bits instead. */ 1241 unsigned int data_size; 1242 unsigned long num_words; 1243 unsigned long *p = sock->sbuf + header_size; 1244 1245 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; 1246 num_words = min_t(size_t, data_size / sizeof(*p), 1247 c->bm_words - c->word_offset); 1248 len = num_words * sizeof(*p); 1249 if (len) 1250 drbd_bm_get_lel(device, c->word_offset, num_words, p); 1251 err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0); 1252 c->word_offset += num_words; 1253 c->bit_offset = c->word_offset * BITS_PER_LONG; 1254 1255 c->packets[1]++; 1256 c->bytes[1] += header_size + len; 1257 1258 if (c->bit_offset > c->bm_bits) 1259 c->bit_offset = c->bm_bits; 1260 } 1261 if (!err) { 1262 if (len == 0) { 1263 INFO_bm_xfer_stats(device, "send", c); 1264 return 0; 1265 } else 1266 return 1; 1267 } 1268 return -EIO; 1269 } 1270 1271 /* See the comment at receive_bitmap() */ 1272 static int _drbd_send_bitmap(struct drbd_device *device) 1273 { 1274 struct bm_xfer_ctx c; 1275 int err; 1276 1277 if (!expect(device->bitmap)) 1278 return false; 1279 1280 if (get_ldev(device)) { 1281 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) { 1282 drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n"); 1283 drbd_bm_set_all(device); 1284 if (drbd_bm_write(device)) { 1285 /* write_bm did fail! Leave full sync flag set in Meta P_DATA 1286 * but otherwise process as per normal - need to tell other 1287 * side that a full resync is required! */ 1288 drbd_err(device, "Failed to write bitmap to disk!\n"); 1289 } else { 1290 drbd_md_clear_flag(device, MDF_FULL_SYNC); 1291 drbd_md_sync(device); 1292 } 1293 } 1294 put_ldev(device); 1295 } 1296 1297 c = (struct bm_xfer_ctx) { 1298 .bm_bits = drbd_bm_bits(device), 1299 .bm_words = drbd_bm_words(device), 1300 }; 1301 1302 do { 1303 err = send_bitmap_rle_or_plain(device, &c); 1304 } while (err > 0); 1305 1306 return err == 0; 1307 } 1308 1309 int drbd_send_bitmap(struct drbd_device *device) 1310 { 1311 struct drbd_socket *sock = &first_peer_device(device)->connection->data; 1312 int err = -1; 1313 1314 mutex_lock(&sock->mutex); 1315 if (sock->socket) 1316 err = !_drbd_send_bitmap(device); 1317 mutex_unlock(&sock->mutex); 1318 return err; 1319 } 1320 1321 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size) 1322 { 1323 struct drbd_socket *sock; 1324 struct p_barrier_ack *p; 1325 1326 if (connection->cstate < C_WF_REPORT_PARAMS) 1327 return; 1328 1329 sock = &connection->meta; 1330 p = conn_prepare_command(connection, sock); 1331 if (!p) 1332 return; 1333 p->barrier = barrier_nr; 1334 p->set_size = cpu_to_be32(set_size); 1335 conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0); 1336 } 1337 1338 /** 1339 * _drbd_send_ack() - Sends an ack packet 1340 * @device: DRBD device. 1341 * @cmd: Packet command code. 1342 * @sector: sector, needs to be in big endian byte order 1343 * @blksize: size in byte, needs to be in big endian byte order 1344 * @block_id: Id, big endian byte order 1345 */ 1346 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd, 1347 u64 sector, u32 blksize, u64 block_id) 1348 { 1349 struct drbd_socket *sock; 1350 struct p_block_ack *p; 1351 1352 if (peer_device->device->state.conn < C_CONNECTED) 1353 return -EIO; 1354 1355 sock = &peer_device->connection->meta; 1356 p = drbd_prepare_command(peer_device, sock); 1357 if (!p) 1358 return -EIO; 1359 p->sector = sector; 1360 p->block_id = block_id; 1361 p->blksize = blksize; 1362 p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq)); 1363 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0); 1364 } 1365 1366 /* dp->sector and dp->block_id already/still in network byte order, 1367 * data_size is payload size according to dp->head, 1368 * and may need to be corrected for digest size. */ 1369 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd, 1370 struct p_data *dp, int data_size) 1371 { 1372 if (peer_device->connection->peer_integrity_tfm) 1373 data_size -= crypto_shash_digestsize(peer_device->connection->peer_integrity_tfm); 1374 _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size), 1375 dp->block_id); 1376 } 1377 1378 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd, 1379 struct p_block_req *rp) 1380 { 1381 _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id); 1382 } 1383 1384 /** 1385 * drbd_send_ack() - Sends an ack packet 1386 * @device: DRBD device 1387 * @cmd: packet command code 1388 * @peer_req: peer request 1389 */ 1390 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd, 1391 struct drbd_peer_request *peer_req) 1392 { 1393 return _drbd_send_ack(peer_device, cmd, 1394 cpu_to_be64(peer_req->i.sector), 1395 cpu_to_be32(peer_req->i.size), 1396 peer_req->block_id); 1397 } 1398 1399 /* This function misuses the block_id field to signal if the blocks 1400 * are is sync or not. */ 1401 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd, 1402 sector_t sector, int blksize, u64 block_id) 1403 { 1404 return _drbd_send_ack(peer_device, cmd, 1405 cpu_to_be64(sector), 1406 cpu_to_be32(blksize), 1407 cpu_to_be64(block_id)); 1408 } 1409 1410 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device, 1411 struct drbd_peer_request *peer_req) 1412 { 1413 struct drbd_socket *sock; 1414 struct p_block_desc *p; 1415 1416 sock = &peer_device->connection->data; 1417 p = drbd_prepare_command(peer_device, sock); 1418 if (!p) 1419 return -EIO; 1420 p->sector = cpu_to_be64(peer_req->i.sector); 1421 p->blksize = cpu_to_be32(peer_req->i.size); 1422 p->pad = 0; 1423 return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0); 1424 } 1425 1426 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd, 1427 sector_t sector, int size, u64 block_id) 1428 { 1429 struct drbd_socket *sock; 1430 struct p_block_req *p; 1431 1432 sock = &peer_device->connection->data; 1433 p = drbd_prepare_command(peer_device, sock); 1434 if (!p) 1435 return -EIO; 1436 p->sector = cpu_to_be64(sector); 1437 p->block_id = block_id; 1438 p->blksize = cpu_to_be32(size); 1439 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0); 1440 } 1441 1442 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size, 1443 void *digest, int digest_size, enum drbd_packet cmd) 1444 { 1445 struct drbd_socket *sock; 1446 struct p_block_req *p; 1447 1448 /* FIXME: Put the digest into the preallocated socket buffer. */ 1449 1450 sock = &peer_device->connection->data; 1451 p = drbd_prepare_command(peer_device, sock); 1452 if (!p) 1453 return -EIO; 1454 p->sector = cpu_to_be64(sector); 1455 p->block_id = ID_SYNCER /* unused */; 1456 p->blksize = cpu_to_be32(size); 1457 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size); 1458 } 1459 1460 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size) 1461 { 1462 struct drbd_socket *sock; 1463 struct p_block_req *p; 1464 1465 sock = &peer_device->connection->data; 1466 p = drbd_prepare_command(peer_device, sock); 1467 if (!p) 1468 return -EIO; 1469 p->sector = cpu_to_be64(sector); 1470 p->block_id = ID_SYNCER /* unused */; 1471 p->blksize = cpu_to_be32(size); 1472 return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0); 1473 } 1474 1475 /* called on sndtimeo 1476 * returns false if we should retry, 1477 * true if we think connection is dead 1478 */ 1479 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock) 1480 { 1481 int drop_it; 1482 /* long elapsed = (long)(jiffies - device->last_received); */ 1483 1484 drop_it = connection->meta.socket == sock 1485 || !connection->ack_receiver.task 1486 || get_t_state(&connection->ack_receiver) != RUNNING 1487 || connection->cstate < C_WF_REPORT_PARAMS; 1488 1489 if (drop_it) 1490 return true; 1491 1492 drop_it = !--connection->ko_count; 1493 if (!drop_it) { 1494 drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n", 1495 current->comm, current->pid, connection->ko_count); 1496 request_ping(connection); 1497 } 1498 1499 return drop_it; /* && (device->state == R_PRIMARY) */; 1500 } 1501 1502 static void drbd_update_congested(struct drbd_connection *connection) 1503 { 1504 struct sock *sk = connection->data.socket->sk; 1505 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5) 1506 set_bit(NET_CONGESTED, &connection->flags); 1507 } 1508 1509 /* The idea of sendpage seems to be to put some kind of reference 1510 * to the page into the skb, and to hand it over to the NIC. In 1511 * this process get_page() gets called. 1512 * 1513 * As soon as the page was really sent over the network put_page() 1514 * gets called by some part of the network layer. [ NIC driver? ] 1515 * 1516 * [ get_page() / put_page() increment/decrement the count. If count 1517 * reaches 0 the page will be freed. ] 1518 * 1519 * This works nicely with pages from FSs. 1520 * But this means that in protocol A we might signal IO completion too early! 1521 * 1522 * In order not to corrupt data during a resync we must make sure 1523 * that we do not reuse our own buffer pages (EEs) to early, therefore 1524 * we have the net_ee list. 1525 * 1526 * XFS seems to have problems, still, it submits pages with page_count == 0! 1527 * As a workaround, we disable sendpage on pages 1528 * with page_count == 0 or PageSlab. 1529 */ 1530 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page, 1531 int offset, size_t size, unsigned msg_flags) 1532 { 1533 struct socket *socket; 1534 void *addr; 1535 int err; 1536 1537 socket = peer_device->connection->data.socket; 1538 addr = kmap(page) + offset; 1539 err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags); 1540 kunmap(page); 1541 if (!err) 1542 peer_device->device->send_cnt += size >> 9; 1543 return err; 1544 } 1545 1546 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page, 1547 int offset, size_t size, unsigned msg_flags) 1548 { 1549 struct socket *socket = peer_device->connection->data.socket; 1550 int len = size; 1551 int err = -EIO; 1552 1553 /* e.g. XFS meta- & log-data is in slab pages, which have a 1554 * page_count of 0 and/or have PageSlab() set. 1555 * we cannot use send_page for those, as that does get_page(); 1556 * put_page(); and would cause either a VM_BUG directly, or 1557 * __page_cache_release a page that would actually still be referenced 1558 * by someone, leading to some obscure delayed Oops somewhere else. */ 1559 if (drbd_disable_sendpage || !sendpage_ok(page)) 1560 return _drbd_no_send_page(peer_device, page, offset, size, msg_flags); 1561 1562 msg_flags |= MSG_NOSIGNAL; 1563 drbd_update_congested(peer_device->connection); 1564 do { 1565 int sent; 1566 1567 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags); 1568 if (sent <= 0) { 1569 if (sent == -EAGAIN) { 1570 if (we_should_drop_the_connection(peer_device->connection, socket)) 1571 break; 1572 continue; 1573 } 1574 drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n", 1575 __func__, (int)size, len, sent); 1576 if (sent < 0) 1577 err = sent; 1578 break; 1579 } 1580 len -= sent; 1581 offset += sent; 1582 } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/); 1583 clear_bit(NET_CONGESTED, &peer_device->connection->flags); 1584 1585 if (len == 0) { 1586 err = 0; 1587 peer_device->device->send_cnt += size >> 9; 1588 } 1589 return err; 1590 } 1591 1592 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio) 1593 { 1594 struct bio_vec bvec; 1595 struct bvec_iter iter; 1596 1597 /* hint all but last page with MSG_MORE */ 1598 bio_for_each_segment(bvec, bio, iter) { 1599 int err; 1600 1601 err = _drbd_no_send_page(peer_device, bvec.bv_page, 1602 bvec.bv_offset, bvec.bv_len, 1603 bio_iter_last(bvec, iter) 1604 ? 0 : MSG_MORE); 1605 if (err) 1606 return err; 1607 /* REQ_OP_WRITE_SAME has only one segment */ 1608 if (bio_op(bio) == REQ_OP_WRITE_SAME) 1609 break; 1610 } 1611 return 0; 1612 } 1613 1614 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio) 1615 { 1616 struct bio_vec bvec; 1617 struct bvec_iter iter; 1618 1619 /* hint all but last page with MSG_MORE */ 1620 bio_for_each_segment(bvec, bio, iter) { 1621 int err; 1622 1623 err = _drbd_send_page(peer_device, bvec.bv_page, 1624 bvec.bv_offset, bvec.bv_len, 1625 bio_iter_last(bvec, iter) ? 0 : MSG_MORE); 1626 if (err) 1627 return err; 1628 /* REQ_OP_WRITE_SAME has only one segment */ 1629 if (bio_op(bio) == REQ_OP_WRITE_SAME) 1630 break; 1631 } 1632 return 0; 1633 } 1634 1635 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device, 1636 struct drbd_peer_request *peer_req) 1637 { 1638 struct page *page = peer_req->pages; 1639 unsigned len = peer_req->i.size; 1640 int err; 1641 1642 /* hint all but last page with MSG_MORE */ 1643 page_chain_for_each(page) { 1644 unsigned l = min_t(unsigned, len, PAGE_SIZE); 1645 1646 err = _drbd_send_page(peer_device, page, 0, l, 1647 page_chain_next(page) ? MSG_MORE : 0); 1648 if (err) 1649 return err; 1650 len -= l; 1651 } 1652 return 0; 1653 } 1654 1655 static u32 bio_flags_to_wire(struct drbd_connection *connection, 1656 struct bio *bio) 1657 { 1658 if (connection->agreed_pro_version >= 95) 1659 return (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) | 1660 (bio->bi_opf & REQ_FUA ? DP_FUA : 0) | 1661 (bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) | 1662 (bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) | 1663 (bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) | 1664 (bio_op(bio) == REQ_OP_WRITE_ZEROES ? 1665 ((connection->agreed_features & DRBD_FF_WZEROES) ? 1666 (DP_ZEROES |(!(bio->bi_opf & REQ_NOUNMAP) ? DP_DISCARD : 0)) 1667 : DP_DISCARD) 1668 : 0); 1669 else 1670 return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0; 1671 } 1672 1673 /* Used to send write or TRIM aka REQ_OP_DISCARD requests 1674 * R_PRIMARY -> Peer (P_DATA, P_TRIM) 1675 */ 1676 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req) 1677 { 1678 struct drbd_device *device = peer_device->device; 1679 struct drbd_socket *sock; 1680 struct p_data *p; 1681 struct p_wsame *wsame = NULL; 1682 void *digest_out; 1683 unsigned int dp_flags = 0; 1684 int digest_size; 1685 int err; 1686 1687 sock = &peer_device->connection->data; 1688 p = drbd_prepare_command(peer_device, sock); 1689 digest_size = peer_device->connection->integrity_tfm ? 1690 crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0; 1691 1692 if (!p) 1693 return -EIO; 1694 p->sector = cpu_to_be64(req->i.sector); 1695 p->block_id = (unsigned long)req; 1696 p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq)); 1697 dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio); 1698 if (device->state.conn >= C_SYNC_SOURCE && 1699 device->state.conn <= C_PAUSED_SYNC_T) 1700 dp_flags |= DP_MAY_SET_IN_SYNC; 1701 if (peer_device->connection->agreed_pro_version >= 100) { 1702 if (req->rq_state & RQ_EXP_RECEIVE_ACK) 1703 dp_flags |= DP_SEND_RECEIVE_ACK; 1704 /* During resync, request an explicit write ack, 1705 * even in protocol != C */ 1706 if (req->rq_state & RQ_EXP_WRITE_ACK 1707 || (dp_flags & DP_MAY_SET_IN_SYNC)) 1708 dp_flags |= DP_SEND_WRITE_ACK; 1709 } 1710 p->dp_flags = cpu_to_be32(dp_flags); 1711 1712 if (dp_flags & (DP_DISCARD|DP_ZEROES)) { 1713 enum drbd_packet cmd = (dp_flags & DP_ZEROES) ? P_ZEROES : P_TRIM; 1714 struct p_trim *t = (struct p_trim*)p; 1715 t->size = cpu_to_be32(req->i.size); 1716 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*t), NULL, 0); 1717 goto out; 1718 } 1719 if (dp_flags & DP_WSAME) { 1720 /* this will only work if DRBD_FF_WSAME is set AND the 1721 * handshake agreed that all nodes and backend devices are 1722 * WRITE_SAME capable and agree on logical_block_size */ 1723 wsame = (struct p_wsame*)p; 1724 digest_out = wsame + 1; 1725 wsame->size = cpu_to_be32(req->i.size); 1726 } else 1727 digest_out = p + 1; 1728 1729 /* our digest is still only over the payload. 1730 * TRIM does not carry any payload. */ 1731 if (digest_size) 1732 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out); 1733 if (wsame) { 1734 err = 1735 __send_command(peer_device->connection, device->vnr, sock, P_WSAME, 1736 sizeof(*wsame) + digest_size, NULL, 1737 bio_iovec(req->master_bio).bv_len); 1738 } else 1739 err = 1740 __send_command(peer_device->connection, device->vnr, sock, P_DATA, 1741 sizeof(*p) + digest_size, NULL, req->i.size); 1742 if (!err) { 1743 /* For protocol A, we have to memcpy the payload into 1744 * socket buffers, as we may complete right away 1745 * as soon as we handed it over to tcp, at which point the data 1746 * pages may become invalid. 1747 * 1748 * For data-integrity enabled, we copy it as well, so we can be 1749 * sure that even if the bio pages may still be modified, it 1750 * won't change the data on the wire, thus if the digest checks 1751 * out ok after sending on this side, but does not fit on the 1752 * receiving side, we sure have detected corruption elsewhere. 1753 */ 1754 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size) 1755 err = _drbd_send_bio(peer_device, req->master_bio); 1756 else 1757 err = _drbd_send_zc_bio(peer_device, req->master_bio); 1758 1759 /* double check digest, sometimes buffers have been modified in flight. */ 1760 if (digest_size > 0 && digest_size <= 64) { 1761 /* 64 byte, 512 bit, is the largest digest size 1762 * currently supported in kernel crypto. */ 1763 unsigned char digest[64]; 1764 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest); 1765 if (memcmp(p + 1, digest, digest_size)) { 1766 drbd_warn(device, 1767 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n", 1768 (unsigned long long)req->i.sector, req->i.size); 1769 } 1770 } /* else if (digest_size > 64) { 1771 ... Be noisy about digest too large ... 1772 } */ 1773 } 1774 out: 1775 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ 1776 1777 return err; 1778 } 1779 1780 /* answer packet, used to send data back for read requests: 1781 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY) 1782 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY) 1783 */ 1784 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd, 1785 struct drbd_peer_request *peer_req) 1786 { 1787 struct drbd_device *device = peer_device->device; 1788 struct drbd_socket *sock; 1789 struct p_data *p; 1790 int err; 1791 int digest_size; 1792 1793 sock = &peer_device->connection->data; 1794 p = drbd_prepare_command(peer_device, sock); 1795 1796 digest_size = peer_device->connection->integrity_tfm ? 1797 crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0; 1798 1799 if (!p) 1800 return -EIO; 1801 p->sector = cpu_to_be64(peer_req->i.sector); 1802 p->block_id = peer_req->block_id; 1803 p->seq_num = 0; /* unused */ 1804 p->dp_flags = 0; 1805 if (digest_size) 1806 drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1); 1807 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size); 1808 if (!err) 1809 err = _drbd_send_zc_ee(peer_device, peer_req); 1810 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ 1811 1812 return err; 1813 } 1814 1815 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req) 1816 { 1817 struct drbd_socket *sock; 1818 struct p_block_desc *p; 1819 1820 sock = &peer_device->connection->data; 1821 p = drbd_prepare_command(peer_device, sock); 1822 if (!p) 1823 return -EIO; 1824 p->sector = cpu_to_be64(req->i.sector); 1825 p->blksize = cpu_to_be32(req->i.size); 1826 return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0); 1827 } 1828 1829 /* 1830 drbd_send distinguishes two cases: 1831 1832 Packets sent via the data socket "sock" 1833 and packets sent via the meta data socket "msock" 1834 1835 sock msock 1836 -----------------+-------------------------+------------------------------ 1837 timeout conf.timeout / 2 conf.timeout / 2 1838 timeout action send a ping via msock Abort communication 1839 and close all sockets 1840 */ 1841 1842 /* 1843 * you must have down()ed the appropriate [m]sock_mutex elsewhere! 1844 */ 1845 int drbd_send(struct drbd_connection *connection, struct socket *sock, 1846 void *buf, size_t size, unsigned msg_flags) 1847 { 1848 struct kvec iov = {.iov_base = buf, .iov_len = size}; 1849 struct msghdr msg = {.msg_flags = msg_flags | MSG_NOSIGNAL}; 1850 int rv, sent = 0; 1851 1852 if (!sock) 1853 return -EBADR; 1854 1855 /* THINK if (signal_pending) return ... ? */ 1856 1857 iov_iter_kvec(&msg.msg_iter, WRITE, &iov, 1, size); 1858 1859 if (sock == connection->data.socket) { 1860 rcu_read_lock(); 1861 connection->ko_count = rcu_dereference(connection->net_conf)->ko_count; 1862 rcu_read_unlock(); 1863 drbd_update_congested(connection); 1864 } 1865 do { 1866 rv = sock_sendmsg(sock, &msg); 1867 if (rv == -EAGAIN) { 1868 if (we_should_drop_the_connection(connection, sock)) 1869 break; 1870 else 1871 continue; 1872 } 1873 if (rv == -EINTR) { 1874 flush_signals(current); 1875 rv = 0; 1876 } 1877 if (rv < 0) 1878 break; 1879 sent += rv; 1880 } while (sent < size); 1881 1882 if (sock == connection->data.socket) 1883 clear_bit(NET_CONGESTED, &connection->flags); 1884 1885 if (rv <= 0) { 1886 if (rv != -EAGAIN) { 1887 drbd_err(connection, "%s_sendmsg returned %d\n", 1888 sock == connection->meta.socket ? "msock" : "sock", 1889 rv); 1890 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD); 1891 } else 1892 conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD); 1893 } 1894 1895 return sent; 1896 } 1897 1898 /** 1899 * drbd_send_all - Send an entire buffer 1900 * 1901 * Returns 0 upon success and a negative error value otherwise. 1902 */ 1903 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer, 1904 size_t size, unsigned msg_flags) 1905 { 1906 int err; 1907 1908 err = drbd_send(connection, sock, buffer, size, msg_flags); 1909 if (err < 0) 1910 return err; 1911 if (err != size) 1912 return -EIO; 1913 return 0; 1914 } 1915 1916 static int drbd_open(struct block_device *bdev, fmode_t mode) 1917 { 1918 struct drbd_device *device = bdev->bd_disk->private_data; 1919 unsigned long flags; 1920 int rv = 0; 1921 1922 mutex_lock(&drbd_main_mutex); 1923 spin_lock_irqsave(&device->resource->req_lock, flags); 1924 /* to have a stable device->state.role 1925 * and no race with updating open_cnt */ 1926 1927 if (device->state.role != R_PRIMARY) { 1928 if (mode & FMODE_WRITE) 1929 rv = -EROFS; 1930 else if (!drbd_allow_oos) 1931 rv = -EMEDIUMTYPE; 1932 } 1933 1934 if (!rv) 1935 device->open_cnt++; 1936 spin_unlock_irqrestore(&device->resource->req_lock, flags); 1937 mutex_unlock(&drbd_main_mutex); 1938 1939 return rv; 1940 } 1941 1942 static void drbd_release(struct gendisk *gd, fmode_t mode) 1943 { 1944 struct drbd_device *device = gd->private_data; 1945 mutex_lock(&drbd_main_mutex); 1946 device->open_cnt--; 1947 mutex_unlock(&drbd_main_mutex); 1948 } 1949 1950 /* need to hold resource->req_lock */ 1951 void drbd_queue_unplug(struct drbd_device *device) 1952 { 1953 if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) { 1954 D_ASSERT(device, device->state.role == R_PRIMARY); 1955 if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) { 1956 drbd_queue_work_if_unqueued( 1957 &first_peer_device(device)->connection->sender_work, 1958 &device->unplug_work); 1959 } 1960 } 1961 } 1962 1963 static void drbd_set_defaults(struct drbd_device *device) 1964 { 1965 /* Beware! The actual layout differs 1966 * between big endian and little endian */ 1967 device->state = (union drbd_dev_state) { 1968 { .role = R_SECONDARY, 1969 .peer = R_UNKNOWN, 1970 .conn = C_STANDALONE, 1971 .disk = D_DISKLESS, 1972 .pdsk = D_UNKNOWN, 1973 } }; 1974 } 1975 1976 void drbd_init_set_defaults(struct drbd_device *device) 1977 { 1978 /* the memset(,0,) did most of this. 1979 * note: only assignments, no allocation in here */ 1980 1981 drbd_set_defaults(device); 1982 1983 atomic_set(&device->ap_bio_cnt, 0); 1984 atomic_set(&device->ap_actlog_cnt, 0); 1985 atomic_set(&device->ap_pending_cnt, 0); 1986 atomic_set(&device->rs_pending_cnt, 0); 1987 atomic_set(&device->unacked_cnt, 0); 1988 atomic_set(&device->local_cnt, 0); 1989 atomic_set(&device->pp_in_use_by_net, 0); 1990 atomic_set(&device->rs_sect_in, 0); 1991 atomic_set(&device->rs_sect_ev, 0); 1992 atomic_set(&device->ap_in_flight, 0); 1993 atomic_set(&device->md_io.in_use, 0); 1994 1995 mutex_init(&device->own_state_mutex); 1996 device->state_mutex = &device->own_state_mutex; 1997 1998 spin_lock_init(&device->al_lock); 1999 spin_lock_init(&device->peer_seq_lock); 2000 2001 INIT_LIST_HEAD(&device->active_ee); 2002 INIT_LIST_HEAD(&device->sync_ee); 2003 INIT_LIST_HEAD(&device->done_ee); 2004 INIT_LIST_HEAD(&device->read_ee); 2005 INIT_LIST_HEAD(&device->net_ee); 2006 INIT_LIST_HEAD(&device->resync_reads); 2007 INIT_LIST_HEAD(&device->resync_work.list); 2008 INIT_LIST_HEAD(&device->unplug_work.list); 2009 INIT_LIST_HEAD(&device->bm_io_work.w.list); 2010 INIT_LIST_HEAD(&device->pending_master_completion[0]); 2011 INIT_LIST_HEAD(&device->pending_master_completion[1]); 2012 INIT_LIST_HEAD(&device->pending_completion[0]); 2013 INIT_LIST_HEAD(&device->pending_completion[1]); 2014 2015 device->resync_work.cb = w_resync_timer; 2016 device->unplug_work.cb = w_send_write_hint; 2017 device->bm_io_work.w.cb = w_bitmap_io; 2018 2019 timer_setup(&device->resync_timer, resync_timer_fn, 0); 2020 timer_setup(&device->md_sync_timer, md_sync_timer_fn, 0); 2021 timer_setup(&device->start_resync_timer, start_resync_timer_fn, 0); 2022 timer_setup(&device->request_timer, request_timer_fn, 0); 2023 2024 init_waitqueue_head(&device->misc_wait); 2025 init_waitqueue_head(&device->state_wait); 2026 init_waitqueue_head(&device->ee_wait); 2027 init_waitqueue_head(&device->al_wait); 2028 init_waitqueue_head(&device->seq_wait); 2029 2030 device->resync_wenr = LC_FREE; 2031 device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; 2032 device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; 2033 } 2034 2035 void drbd_set_my_capacity(struct drbd_device *device, sector_t size) 2036 { 2037 char ppb[10]; 2038 2039 set_capacity(device->vdisk, size); 2040 revalidate_disk_size(device->vdisk, false); 2041 2042 drbd_info(device, "size = %s (%llu KB)\n", 2043 ppsize(ppb, size>>1), (unsigned long long)size>>1); 2044 } 2045 2046 void drbd_device_cleanup(struct drbd_device *device) 2047 { 2048 int i; 2049 if (first_peer_device(device)->connection->receiver.t_state != NONE) 2050 drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n", 2051 first_peer_device(device)->connection->receiver.t_state); 2052 2053 device->al_writ_cnt = 2054 device->bm_writ_cnt = 2055 device->read_cnt = 2056 device->recv_cnt = 2057 device->send_cnt = 2058 device->writ_cnt = 2059 device->p_size = 2060 device->rs_start = 2061 device->rs_total = 2062 device->rs_failed = 0; 2063 device->rs_last_events = 0; 2064 device->rs_last_sect_ev = 0; 2065 for (i = 0; i < DRBD_SYNC_MARKS; i++) { 2066 device->rs_mark_left[i] = 0; 2067 device->rs_mark_time[i] = 0; 2068 } 2069 D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL); 2070 2071 set_capacity(device->vdisk, 0); 2072 revalidate_disk_size(device->vdisk, false); 2073 if (device->bitmap) { 2074 /* maybe never allocated. */ 2075 drbd_bm_resize(device, 0, 1); 2076 drbd_bm_cleanup(device); 2077 } 2078 2079 drbd_backing_dev_free(device, device->ldev); 2080 device->ldev = NULL; 2081 2082 clear_bit(AL_SUSPENDED, &device->flags); 2083 2084 D_ASSERT(device, list_empty(&device->active_ee)); 2085 D_ASSERT(device, list_empty(&device->sync_ee)); 2086 D_ASSERT(device, list_empty(&device->done_ee)); 2087 D_ASSERT(device, list_empty(&device->read_ee)); 2088 D_ASSERT(device, list_empty(&device->net_ee)); 2089 D_ASSERT(device, list_empty(&device->resync_reads)); 2090 D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q)); 2091 D_ASSERT(device, list_empty(&device->resync_work.list)); 2092 D_ASSERT(device, list_empty(&device->unplug_work.list)); 2093 2094 drbd_set_defaults(device); 2095 } 2096 2097 2098 static void drbd_destroy_mempools(void) 2099 { 2100 struct page *page; 2101 2102 while (drbd_pp_pool) { 2103 page = drbd_pp_pool; 2104 drbd_pp_pool = (struct page *)page_private(page); 2105 __free_page(page); 2106 drbd_pp_vacant--; 2107 } 2108 2109 /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */ 2110 2111 bioset_exit(&drbd_io_bio_set); 2112 bioset_exit(&drbd_md_io_bio_set); 2113 mempool_exit(&drbd_md_io_page_pool); 2114 mempool_exit(&drbd_ee_mempool); 2115 mempool_exit(&drbd_request_mempool); 2116 kmem_cache_destroy(drbd_ee_cache); 2117 kmem_cache_destroy(drbd_request_cache); 2118 kmem_cache_destroy(drbd_bm_ext_cache); 2119 kmem_cache_destroy(drbd_al_ext_cache); 2120 2121 drbd_ee_cache = NULL; 2122 drbd_request_cache = NULL; 2123 drbd_bm_ext_cache = NULL; 2124 drbd_al_ext_cache = NULL; 2125 2126 return; 2127 } 2128 2129 static int drbd_create_mempools(void) 2130 { 2131 struct page *page; 2132 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count; 2133 int i, ret; 2134 2135 /* caches */ 2136 drbd_request_cache = kmem_cache_create( 2137 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL); 2138 if (drbd_request_cache == NULL) 2139 goto Enomem; 2140 2141 drbd_ee_cache = kmem_cache_create( 2142 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL); 2143 if (drbd_ee_cache == NULL) 2144 goto Enomem; 2145 2146 drbd_bm_ext_cache = kmem_cache_create( 2147 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL); 2148 if (drbd_bm_ext_cache == NULL) 2149 goto Enomem; 2150 2151 drbd_al_ext_cache = kmem_cache_create( 2152 "drbd_al", sizeof(struct lc_element), 0, 0, NULL); 2153 if (drbd_al_ext_cache == NULL) 2154 goto Enomem; 2155 2156 /* mempools */ 2157 ret = bioset_init(&drbd_io_bio_set, BIO_POOL_SIZE, 0, 0); 2158 if (ret) 2159 goto Enomem; 2160 2161 ret = bioset_init(&drbd_md_io_bio_set, DRBD_MIN_POOL_PAGES, 0, 2162 BIOSET_NEED_BVECS); 2163 if (ret) 2164 goto Enomem; 2165 2166 ret = mempool_init_page_pool(&drbd_md_io_page_pool, DRBD_MIN_POOL_PAGES, 0); 2167 if (ret) 2168 goto Enomem; 2169 2170 ret = mempool_init_slab_pool(&drbd_request_mempool, number, 2171 drbd_request_cache); 2172 if (ret) 2173 goto Enomem; 2174 2175 ret = mempool_init_slab_pool(&drbd_ee_mempool, number, drbd_ee_cache); 2176 if (ret) 2177 goto Enomem; 2178 2179 /* drbd's page pool */ 2180 spin_lock_init(&drbd_pp_lock); 2181 2182 for (i = 0; i < number; i++) { 2183 page = alloc_page(GFP_HIGHUSER); 2184 if (!page) 2185 goto Enomem; 2186 set_page_private(page, (unsigned long)drbd_pp_pool); 2187 drbd_pp_pool = page; 2188 } 2189 drbd_pp_vacant = number; 2190 2191 return 0; 2192 2193 Enomem: 2194 drbd_destroy_mempools(); /* in case we allocated some */ 2195 return -ENOMEM; 2196 } 2197 2198 static void drbd_release_all_peer_reqs(struct drbd_device *device) 2199 { 2200 int rr; 2201 2202 rr = drbd_free_peer_reqs(device, &device->active_ee); 2203 if (rr) 2204 drbd_err(device, "%d EEs in active list found!\n", rr); 2205 2206 rr = drbd_free_peer_reqs(device, &device->sync_ee); 2207 if (rr) 2208 drbd_err(device, "%d EEs in sync list found!\n", rr); 2209 2210 rr = drbd_free_peer_reqs(device, &device->read_ee); 2211 if (rr) 2212 drbd_err(device, "%d EEs in read list found!\n", rr); 2213 2214 rr = drbd_free_peer_reqs(device, &device->done_ee); 2215 if (rr) 2216 drbd_err(device, "%d EEs in done list found!\n", rr); 2217 2218 rr = drbd_free_peer_reqs(device, &device->net_ee); 2219 if (rr) 2220 drbd_err(device, "%d EEs in net list found!\n", rr); 2221 } 2222 2223 /* caution. no locking. */ 2224 void drbd_destroy_device(struct kref *kref) 2225 { 2226 struct drbd_device *device = container_of(kref, struct drbd_device, kref); 2227 struct drbd_resource *resource = device->resource; 2228 struct drbd_peer_device *peer_device, *tmp_peer_device; 2229 2230 del_timer_sync(&device->request_timer); 2231 2232 /* paranoia asserts */ 2233 D_ASSERT(device, device->open_cnt == 0); 2234 /* end paranoia asserts */ 2235 2236 /* cleanup stuff that may have been allocated during 2237 * device (re-)configuration or state changes */ 2238 2239 drbd_backing_dev_free(device, device->ldev); 2240 device->ldev = NULL; 2241 2242 drbd_release_all_peer_reqs(device); 2243 2244 lc_destroy(device->act_log); 2245 lc_destroy(device->resync); 2246 2247 kfree(device->p_uuid); 2248 /* device->p_uuid = NULL; */ 2249 2250 if (device->bitmap) /* should no longer be there. */ 2251 drbd_bm_cleanup(device); 2252 __free_page(device->md_io.page); 2253 put_disk(device->vdisk); 2254 blk_cleanup_queue(device->rq_queue); 2255 kfree(device->rs_plan_s); 2256 2257 /* not for_each_connection(connection, resource): 2258 * those may have been cleaned up and disassociated already. 2259 */ 2260 for_each_peer_device_safe(peer_device, tmp_peer_device, device) { 2261 kref_put(&peer_device->connection->kref, drbd_destroy_connection); 2262 kfree(peer_device); 2263 } 2264 memset(device, 0xfd, sizeof(*device)); 2265 kfree(device); 2266 kref_put(&resource->kref, drbd_destroy_resource); 2267 } 2268 2269 /* One global retry thread, if we need to push back some bio and have it 2270 * reinserted through our make request function. 2271 */ 2272 static struct retry_worker { 2273 struct workqueue_struct *wq; 2274 struct work_struct worker; 2275 2276 spinlock_t lock; 2277 struct list_head writes; 2278 } retry; 2279 2280 static void do_retry(struct work_struct *ws) 2281 { 2282 struct retry_worker *retry = container_of(ws, struct retry_worker, worker); 2283 LIST_HEAD(writes); 2284 struct drbd_request *req, *tmp; 2285 2286 spin_lock_irq(&retry->lock); 2287 list_splice_init(&retry->writes, &writes); 2288 spin_unlock_irq(&retry->lock); 2289 2290 list_for_each_entry_safe(req, tmp, &writes, tl_requests) { 2291 struct drbd_device *device = req->device; 2292 struct bio *bio = req->master_bio; 2293 unsigned long start_jif = req->start_jif; 2294 bool expected; 2295 2296 expected = 2297 expect(atomic_read(&req->completion_ref) == 0) && 2298 expect(req->rq_state & RQ_POSTPONED) && 2299 expect((req->rq_state & RQ_LOCAL_PENDING) == 0 || 2300 (req->rq_state & RQ_LOCAL_ABORTED) != 0); 2301 2302 if (!expected) 2303 drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n", 2304 req, atomic_read(&req->completion_ref), 2305 req->rq_state); 2306 2307 /* We still need to put one kref associated with the 2308 * "completion_ref" going zero in the code path that queued it 2309 * here. The request object may still be referenced by a 2310 * frozen local req->private_bio, in case we force-detached. 2311 */ 2312 kref_put(&req->kref, drbd_req_destroy); 2313 2314 /* A single suspended or otherwise blocking device may stall 2315 * all others as well. Fortunately, this code path is to 2316 * recover from a situation that "should not happen": 2317 * concurrent writes in multi-primary setup. 2318 * In a "normal" lifecycle, this workqueue is supposed to be 2319 * destroyed without ever doing anything. 2320 * If it turns out to be an issue anyways, we can do per 2321 * resource (replication group) or per device (minor) retry 2322 * workqueues instead. 2323 */ 2324 2325 /* We are not just doing submit_bio_noacct(), 2326 * as we want to keep the start_time information. */ 2327 inc_ap_bio(device); 2328 __drbd_make_request(device, bio, start_jif); 2329 } 2330 } 2331 2332 /* called via drbd_req_put_completion_ref(), 2333 * holds resource->req_lock */ 2334 void drbd_restart_request(struct drbd_request *req) 2335 { 2336 unsigned long flags; 2337 spin_lock_irqsave(&retry.lock, flags); 2338 list_move_tail(&req->tl_requests, &retry.writes); 2339 spin_unlock_irqrestore(&retry.lock, flags); 2340 2341 /* Drop the extra reference that would otherwise 2342 * have been dropped by complete_master_bio. 2343 * do_retry() needs to grab a new one. */ 2344 dec_ap_bio(req->device); 2345 2346 queue_work(retry.wq, &retry.worker); 2347 } 2348 2349 void drbd_destroy_resource(struct kref *kref) 2350 { 2351 struct drbd_resource *resource = 2352 container_of(kref, struct drbd_resource, kref); 2353 2354 idr_destroy(&resource->devices); 2355 free_cpumask_var(resource->cpu_mask); 2356 kfree(resource->name); 2357 memset(resource, 0xf2, sizeof(*resource)); 2358 kfree(resource); 2359 } 2360 2361 void drbd_free_resource(struct drbd_resource *resource) 2362 { 2363 struct drbd_connection *connection, *tmp; 2364 2365 for_each_connection_safe(connection, tmp, resource) { 2366 list_del(&connection->connections); 2367 drbd_debugfs_connection_cleanup(connection); 2368 kref_put(&connection->kref, drbd_destroy_connection); 2369 } 2370 drbd_debugfs_resource_cleanup(resource); 2371 kref_put(&resource->kref, drbd_destroy_resource); 2372 } 2373 2374 static void drbd_cleanup(void) 2375 { 2376 unsigned int i; 2377 struct drbd_device *device; 2378 struct drbd_resource *resource, *tmp; 2379 2380 /* first remove proc, 2381 * drbdsetup uses it's presence to detect 2382 * whether DRBD is loaded. 2383 * If we would get stuck in proc removal, 2384 * but have netlink already deregistered, 2385 * some drbdsetup commands may wait forever 2386 * for an answer. 2387 */ 2388 if (drbd_proc) 2389 remove_proc_entry("drbd", NULL); 2390 2391 if (retry.wq) 2392 destroy_workqueue(retry.wq); 2393 2394 drbd_genl_unregister(); 2395 2396 idr_for_each_entry(&drbd_devices, device, i) 2397 drbd_delete_device(device); 2398 2399 /* not _rcu since, no other updater anymore. Genl already unregistered */ 2400 for_each_resource_safe(resource, tmp, &drbd_resources) { 2401 list_del(&resource->resources); 2402 drbd_free_resource(resource); 2403 } 2404 2405 drbd_debugfs_cleanup(); 2406 2407 drbd_destroy_mempools(); 2408 unregister_blkdev(DRBD_MAJOR, "drbd"); 2409 2410 idr_destroy(&drbd_devices); 2411 2412 pr_info("module cleanup done.\n"); 2413 } 2414 2415 static void drbd_init_workqueue(struct drbd_work_queue* wq) 2416 { 2417 spin_lock_init(&wq->q_lock); 2418 INIT_LIST_HEAD(&wq->q); 2419 init_waitqueue_head(&wq->q_wait); 2420 } 2421 2422 struct completion_work { 2423 struct drbd_work w; 2424 struct completion done; 2425 }; 2426 2427 static int w_complete(struct drbd_work *w, int cancel) 2428 { 2429 struct completion_work *completion_work = 2430 container_of(w, struct completion_work, w); 2431 2432 complete(&completion_work->done); 2433 return 0; 2434 } 2435 2436 void drbd_flush_workqueue(struct drbd_work_queue *work_queue) 2437 { 2438 struct completion_work completion_work; 2439 2440 completion_work.w.cb = w_complete; 2441 init_completion(&completion_work.done); 2442 drbd_queue_work(work_queue, &completion_work.w); 2443 wait_for_completion(&completion_work.done); 2444 } 2445 2446 struct drbd_resource *drbd_find_resource(const char *name) 2447 { 2448 struct drbd_resource *resource; 2449 2450 if (!name || !name[0]) 2451 return NULL; 2452 2453 rcu_read_lock(); 2454 for_each_resource_rcu(resource, &drbd_resources) { 2455 if (!strcmp(resource->name, name)) { 2456 kref_get(&resource->kref); 2457 goto found; 2458 } 2459 } 2460 resource = NULL; 2461 found: 2462 rcu_read_unlock(); 2463 return resource; 2464 } 2465 2466 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len, 2467 void *peer_addr, int peer_addr_len) 2468 { 2469 struct drbd_resource *resource; 2470 struct drbd_connection *connection; 2471 2472 rcu_read_lock(); 2473 for_each_resource_rcu(resource, &drbd_resources) { 2474 for_each_connection_rcu(connection, resource) { 2475 if (connection->my_addr_len == my_addr_len && 2476 connection->peer_addr_len == peer_addr_len && 2477 !memcmp(&connection->my_addr, my_addr, my_addr_len) && 2478 !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) { 2479 kref_get(&connection->kref); 2480 goto found; 2481 } 2482 } 2483 } 2484 connection = NULL; 2485 found: 2486 rcu_read_unlock(); 2487 return connection; 2488 } 2489 2490 static int drbd_alloc_socket(struct drbd_socket *socket) 2491 { 2492 socket->rbuf = (void *) __get_free_page(GFP_KERNEL); 2493 if (!socket->rbuf) 2494 return -ENOMEM; 2495 socket->sbuf = (void *) __get_free_page(GFP_KERNEL); 2496 if (!socket->sbuf) 2497 return -ENOMEM; 2498 return 0; 2499 } 2500 2501 static void drbd_free_socket(struct drbd_socket *socket) 2502 { 2503 free_page((unsigned long) socket->sbuf); 2504 free_page((unsigned long) socket->rbuf); 2505 } 2506 2507 void conn_free_crypto(struct drbd_connection *connection) 2508 { 2509 drbd_free_sock(connection); 2510 2511 crypto_free_shash(connection->csums_tfm); 2512 crypto_free_shash(connection->verify_tfm); 2513 crypto_free_shash(connection->cram_hmac_tfm); 2514 crypto_free_shash(connection->integrity_tfm); 2515 crypto_free_shash(connection->peer_integrity_tfm); 2516 kfree(connection->int_dig_in); 2517 kfree(connection->int_dig_vv); 2518 2519 connection->csums_tfm = NULL; 2520 connection->verify_tfm = NULL; 2521 connection->cram_hmac_tfm = NULL; 2522 connection->integrity_tfm = NULL; 2523 connection->peer_integrity_tfm = NULL; 2524 connection->int_dig_in = NULL; 2525 connection->int_dig_vv = NULL; 2526 } 2527 2528 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts) 2529 { 2530 struct drbd_connection *connection; 2531 cpumask_var_t new_cpu_mask; 2532 int err; 2533 2534 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) 2535 return -ENOMEM; 2536 2537 /* silently ignore cpu mask on UP kernel */ 2538 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) { 2539 err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE, 2540 cpumask_bits(new_cpu_mask), nr_cpu_ids); 2541 if (err == -EOVERFLOW) { 2542 /* So what. mask it out. */ 2543 cpumask_var_t tmp_cpu_mask; 2544 if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) { 2545 cpumask_setall(tmp_cpu_mask); 2546 cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask); 2547 drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n", 2548 res_opts->cpu_mask, 2549 strlen(res_opts->cpu_mask) > 12 ? "..." : "", 2550 nr_cpu_ids); 2551 free_cpumask_var(tmp_cpu_mask); 2552 err = 0; 2553 } 2554 } 2555 if (err) { 2556 drbd_warn(resource, "bitmap_parse() failed with %d\n", err); 2557 /* retcode = ERR_CPU_MASK_PARSE; */ 2558 goto fail; 2559 } 2560 } 2561 resource->res_opts = *res_opts; 2562 if (cpumask_empty(new_cpu_mask)) 2563 drbd_calc_cpu_mask(&new_cpu_mask); 2564 if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) { 2565 cpumask_copy(resource->cpu_mask, new_cpu_mask); 2566 for_each_connection_rcu(connection, resource) { 2567 connection->receiver.reset_cpu_mask = 1; 2568 connection->ack_receiver.reset_cpu_mask = 1; 2569 connection->worker.reset_cpu_mask = 1; 2570 } 2571 } 2572 err = 0; 2573 2574 fail: 2575 free_cpumask_var(new_cpu_mask); 2576 return err; 2577 2578 } 2579 2580 struct drbd_resource *drbd_create_resource(const char *name) 2581 { 2582 struct drbd_resource *resource; 2583 2584 resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL); 2585 if (!resource) 2586 goto fail; 2587 resource->name = kstrdup(name, GFP_KERNEL); 2588 if (!resource->name) 2589 goto fail_free_resource; 2590 if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL)) 2591 goto fail_free_name; 2592 kref_init(&resource->kref); 2593 idr_init(&resource->devices); 2594 INIT_LIST_HEAD(&resource->connections); 2595 resource->write_ordering = WO_BDEV_FLUSH; 2596 list_add_tail_rcu(&resource->resources, &drbd_resources); 2597 mutex_init(&resource->conf_update); 2598 mutex_init(&resource->adm_mutex); 2599 spin_lock_init(&resource->req_lock); 2600 drbd_debugfs_resource_add(resource); 2601 return resource; 2602 2603 fail_free_name: 2604 kfree(resource->name); 2605 fail_free_resource: 2606 kfree(resource); 2607 fail: 2608 return NULL; 2609 } 2610 2611 /* caller must be under adm_mutex */ 2612 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts) 2613 { 2614 struct drbd_resource *resource; 2615 struct drbd_connection *connection; 2616 2617 connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL); 2618 if (!connection) 2619 return NULL; 2620 2621 if (drbd_alloc_socket(&connection->data)) 2622 goto fail; 2623 if (drbd_alloc_socket(&connection->meta)) 2624 goto fail; 2625 2626 connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL); 2627 if (!connection->current_epoch) 2628 goto fail; 2629 2630 INIT_LIST_HEAD(&connection->transfer_log); 2631 2632 INIT_LIST_HEAD(&connection->current_epoch->list); 2633 connection->epochs = 1; 2634 spin_lock_init(&connection->epoch_lock); 2635 2636 connection->send.seen_any_write_yet = false; 2637 connection->send.current_epoch_nr = 0; 2638 connection->send.current_epoch_writes = 0; 2639 2640 resource = drbd_create_resource(name); 2641 if (!resource) 2642 goto fail; 2643 2644 connection->cstate = C_STANDALONE; 2645 mutex_init(&connection->cstate_mutex); 2646 init_waitqueue_head(&connection->ping_wait); 2647 idr_init(&connection->peer_devices); 2648 2649 drbd_init_workqueue(&connection->sender_work); 2650 mutex_init(&connection->data.mutex); 2651 mutex_init(&connection->meta.mutex); 2652 2653 drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver"); 2654 connection->receiver.connection = connection; 2655 drbd_thread_init(resource, &connection->worker, drbd_worker, "worker"); 2656 connection->worker.connection = connection; 2657 drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv"); 2658 connection->ack_receiver.connection = connection; 2659 2660 kref_init(&connection->kref); 2661 2662 connection->resource = resource; 2663 2664 if (set_resource_options(resource, res_opts)) 2665 goto fail_resource; 2666 2667 kref_get(&resource->kref); 2668 list_add_tail_rcu(&connection->connections, &resource->connections); 2669 drbd_debugfs_connection_add(connection); 2670 return connection; 2671 2672 fail_resource: 2673 list_del(&resource->resources); 2674 drbd_free_resource(resource); 2675 fail: 2676 kfree(connection->current_epoch); 2677 drbd_free_socket(&connection->meta); 2678 drbd_free_socket(&connection->data); 2679 kfree(connection); 2680 return NULL; 2681 } 2682 2683 void drbd_destroy_connection(struct kref *kref) 2684 { 2685 struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref); 2686 struct drbd_resource *resource = connection->resource; 2687 2688 if (atomic_read(&connection->current_epoch->epoch_size) != 0) 2689 drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size)); 2690 kfree(connection->current_epoch); 2691 2692 idr_destroy(&connection->peer_devices); 2693 2694 drbd_free_socket(&connection->meta); 2695 drbd_free_socket(&connection->data); 2696 kfree(connection->int_dig_in); 2697 kfree(connection->int_dig_vv); 2698 memset(connection, 0xfc, sizeof(*connection)); 2699 kfree(connection); 2700 kref_put(&resource->kref, drbd_destroy_resource); 2701 } 2702 2703 static int init_submitter(struct drbd_device *device) 2704 { 2705 /* opencoded create_singlethread_workqueue(), 2706 * to be able to say "drbd%d", ..., minor */ 2707 device->submit.wq = 2708 alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor); 2709 if (!device->submit.wq) 2710 return -ENOMEM; 2711 2712 INIT_WORK(&device->submit.worker, do_submit); 2713 INIT_LIST_HEAD(&device->submit.writes); 2714 return 0; 2715 } 2716 2717 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor) 2718 { 2719 struct drbd_resource *resource = adm_ctx->resource; 2720 struct drbd_connection *connection; 2721 struct drbd_device *device; 2722 struct drbd_peer_device *peer_device, *tmp_peer_device; 2723 struct gendisk *disk; 2724 struct request_queue *q; 2725 int id; 2726 int vnr = adm_ctx->volume; 2727 enum drbd_ret_code err = ERR_NOMEM; 2728 2729 device = minor_to_device(minor); 2730 if (device) 2731 return ERR_MINOR_OR_VOLUME_EXISTS; 2732 2733 /* GFP_KERNEL, we are outside of all write-out paths */ 2734 device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL); 2735 if (!device) 2736 return ERR_NOMEM; 2737 kref_init(&device->kref); 2738 2739 kref_get(&resource->kref); 2740 device->resource = resource; 2741 device->minor = minor; 2742 device->vnr = vnr; 2743 2744 drbd_init_set_defaults(device); 2745 2746 q = blk_alloc_queue(NUMA_NO_NODE); 2747 if (!q) 2748 goto out_no_q; 2749 device->rq_queue = q; 2750 2751 disk = alloc_disk(1); 2752 if (!disk) 2753 goto out_no_disk; 2754 device->vdisk = disk; 2755 2756 set_disk_ro(disk, true); 2757 2758 disk->queue = q; 2759 disk->major = DRBD_MAJOR; 2760 disk->first_minor = minor; 2761 disk->fops = &drbd_ops; 2762 sprintf(disk->disk_name, "drbd%d", minor); 2763 disk->private_data = device; 2764 2765 blk_queue_write_cache(q, true, true); 2766 /* Setting the max_hw_sectors to an odd value of 8kibyte here 2767 This triggers a max_bio_size message upon first attach or connect */ 2768 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8); 2769 2770 device->md_io.page = alloc_page(GFP_KERNEL); 2771 if (!device->md_io.page) 2772 goto out_no_io_page; 2773 2774 if (drbd_bm_init(device)) 2775 goto out_no_bitmap; 2776 device->read_requests = RB_ROOT; 2777 device->write_requests = RB_ROOT; 2778 2779 id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL); 2780 if (id < 0) { 2781 if (id == -ENOSPC) 2782 err = ERR_MINOR_OR_VOLUME_EXISTS; 2783 goto out_no_minor_idr; 2784 } 2785 kref_get(&device->kref); 2786 2787 id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL); 2788 if (id < 0) { 2789 if (id == -ENOSPC) 2790 err = ERR_MINOR_OR_VOLUME_EXISTS; 2791 goto out_idr_remove_minor; 2792 } 2793 kref_get(&device->kref); 2794 2795 INIT_LIST_HEAD(&device->peer_devices); 2796 INIT_LIST_HEAD(&device->pending_bitmap_io); 2797 for_each_connection(connection, resource) { 2798 peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL); 2799 if (!peer_device) 2800 goto out_idr_remove_from_resource; 2801 peer_device->connection = connection; 2802 peer_device->device = device; 2803 2804 list_add(&peer_device->peer_devices, &device->peer_devices); 2805 kref_get(&device->kref); 2806 2807 id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL); 2808 if (id < 0) { 2809 if (id == -ENOSPC) 2810 err = ERR_INVALID_REQUEST; 2811 goto out_idr_remove_from_resource; 2812 } 2813 kref_get(&connection->kref); 2814 INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf); 2815 } 2816 2817 if (init_submitter(device)) { 2818 err = ERR_NOMEM; 2819 goto out_idr_remove_vol; 2820 } 2821 2822 add_disk(disk); 2823 2824 /* inherit the connection state */ 2825 device->state.conn = first_connection(resource)->cstate; 2826 if (device->state.conn == C_WF_REPORT_PARAMS) { 2827 for_each_peer_device(peer_device, device) 2828 drbd_connected(peer_device); 2829 } 2830 /* move to create_peer_device() */ 2831 for_each_peer_device(peer_device, device) 2832 drbd_debugfs_peer_device_add(peer_device); 2833 drbd_debugfs_device_add(device); 2834 return NO_ERROR; 2835 2836 out_idr_remove_vol: 2837 idr_remove(&connection->peer_devices, vnr); 2838 out_idr_remove_from_resource: 2839 for_each_connection(connection, resource) { 2840 peer_device = idr_remove(&connection->peer_devices, vnr); 2841 if (peer_device) 2842 kref_put(&connection->kref, drbd_destroy_connection); 2843 } 2844 for_each_peer_device_safe(peer_device, tmp_peer_device, device) { 2845 list_del(&peer_device->peer_devices); 2846 kfree(peer_device); 2847 } 2848 idr_remove(&resource->devices, vnr); 2849 out_idr_remove_minor: 2850 idr_remove(&drbd_devices, minor); 2851 synchronize_rcu(); 2852 out_no_minor_idr: 2853 drbd_bm_cleanup(device); 2854 out_no_bitmap: 2855 __free_page(device->md_io.page); 2856 out_no_io_page: 2857 put_disk(disk); 2858 out_no_disk: 2859 blk_cleanup_queue(q); 2860 out_no_q: 2861 kref_put(&resource->kref, drbd_destroy_resource); 2862 kfree(device); 2863 return err; 2864 } 2865 2866 void drbd_delete_device(struct drbd_device *device) 2867 { 2868 struct drbd_resource *resource = device->resource; 2869 struct drbd_connection *connection; 2870 struct drbd_peer_device *peer_device; 2871 2872 /* move to free_peer_device() */ 2873 for_each_peer_device(peer_device, device) 2874 drbd_debugfs_peer_device_cleanup(peer_device); 2875 drbd_debugfs_device_cleanup(device); 2876 for_each_connection(connection, resource) { 2877 idr_remove(&connection->peer_devices, device->vnr); 2878 kref_put(&device->kref, drbd_destroy_device); 2879 } 2880 idr_remove(&resource->devices, device->vnr); 2881 kref_put(&device->kref, drbd_destroy_device); 2882 idr_remove(&drbd_devices, device_to_minor(device)); 2883 kref_put(&device->kref, drbd_destroy_device); 2884 del_gendisk(device->vdisk); 2885 synchronize_rcu(); 2886 kref_put(&device->kref, drbd_destroy_device); 2887 } 2888 2889 static int __init drbd_init(void) 2890 { 2891 int err; 2892 2893 if (drbd_minor_count < DRBD_MINOR_COUNT_MIN || drbd_minor_count > DRBD_MINOR_COUNT_MAX) { 2894 pr_err("invalid minor_count (%d)\n", drbd_minor_count); 2895 #ifdef MODULE 2896 return -EINVAL; 2897 #else 2898 drbd_minor_count = DRBD_MINOR_COUNT_DEF; 2899 #endif 2900 } 2901 2902 err = register_blkdev(DRBD_MAJOR, "drbd"); 2903 if (err) { 2904 pr_err("unable to register block device major %d\n", 2905 DRBD_MAJOR); 2906 return err; 2907 } 2908 2909 /* 2910 * allocate all necessary structs 2911 */ 2912 init_waitqueue_head(&drbd_pp_wait); 2913 2914 drbd_proc = NULL; /* play safe for drbd_cleanup */ 2915 idr_init(&drbd_devices); 2916 2917 mutex_init(&resources_mutex); 2918 INIT_LIST_HEAD(&drbd_resources); 2919 2920 err = drbd_genl_register(); 2921 if (err) { 2922 pr_err("unable to register generic netlink family\n"); 2923 goto fail; 2924 } 2925 2926 err = drbd_create_mempools(); 2927 if (err) 2928 goto fail; 2929 2930 err = -ENOMEM; 2931 drbd_proc = proc_create_single("drbd", S_IFREG | 0444 , NULL, drbd_seq_show); 2932 if (!drbd_proc) { 2933 pr_err("unable to register proc file\n"); 2934 goto fail; 2935 } 2936 2937 retry.wq = create_singlethread_workqueue("drbd-reissue"); 2938 if (!retry.wq) { 2939 pr_err("unable to create retry workqueue\n"); 2940 goto fail; 2941 } 2942 INIT_WORK(&retry.worker, do_retry); 2943 spin_lock_init(&retry.lock); 2944 INIT_LIST_HEAD(&retry.writes); 2945 2946 drbd_debugfs_init(); 2947 2948 pr_info("initialized. " 2949 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n", 2950 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX); 2951 pr_info("%s\n", drbd_buildtag()); 2952 pr_info("registered as block device major %d\n", DRBD_MAJOR); 2953 return 0; /* Success! */ 2954 2955 fail: 2956 drbd_cleanup(); 2957 if (err == -ENOMEM) 2958 pr_err("ran out of memory\n"); 2959 else 2960 pr_err("initialization failure\n"); 2961 return err; 2962 } 2963 2964 static void drbd_free_one_sock(struct drbd_socket *ds) 2965 { 2966 struct socket *s; 2967 mutex_lock(&ds->mutex); 2968 s = ds->socket; 2969 ds->socket = NULL; 2970 mutex_unlock(&ds->mutex); 2971 if (s) { 2972 /* so debugfs does not need to mutex_lock() */ 2973 synchronize_rcu(); 2974 kernel_sock_shutdown(s, SHUT_RDWR); 2975 sock_release(s); 2976 } 2977 } 2978 2979 void drbd_free_sock(struct drbd_connection *connection) 2980 { 2981 if (connection->data.socket) 2982 drbd_free_one_sock(&connection->data); 2983 if (connection->meta.socket) 2984 drbd_free_one_sock(&connection->meta); 2985 } 2986 2987 /* meta data management */ 2988 2989 void conn_md_sync(struct drbd_connection *connection) 2990 { 2991 struct drbd_peer_device *peer_device; 2992 int vnr; 2993 2994 rcu_read_lock(); 2995 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { 2996 struct drbd_device *device = peer_device->device; 2997 2998 kref_get(&device->kref); 2999 rcu_read_unlock(); 3000 drbd_md_sync(device); 3001 kref_put(&device->kref, drbd_destroy_device); 3002 rcu_read_lock(); 3003 } 3004 rcu_read_unlock(); 3005 } 3006 3007 /* aligned 4kByte */ 3008 struct meta_data_on_disk { 3009 u64 la_size_sect; /* last agreed size. */ 3010 u64 uuid[UI_SIZE]; /* UUIDs. */ 3011 u64 device_uuid; 3012 u64 reserved_u64_1; 3013 u32 flags; /* MDF */ 3014 u32 magic; 3015 u32 md_size_sect; 3016 u32 al_offset; /* offset to this block */ 3017 u32 al_nr_extents; /* important for restoring the AL (userspace) */ 3018 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */ 3019 u32 bm_offset; /* offset to the bitmap, from here */ 3020 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */ 3021 u32 la_peer_max_bio_size; /* last peer max_bio_size */ 3022 3023 /* see al_tr_number_to_on_disk_sector() */ 3024 u32 al_stripes; 3025 u32 al_stripe_size_4k; 3026 3027 u8 reserved_u8[4096 - (7*8 + 10*4)]; 3028 } __packed; 3029 3030 3031 3032 void drbd_md_write(struct drbd_device *device, void *b) 3033 { 3034 struct meta_data_on_disk *buffer = b; 3035 sector_t sector; 3036 int i; 3037 3038 memset(buffer, 0, sizeof(*buffer)); 3039 3040 buffer->la_size_sect = cpu_to_be64(get_capacity(device->vdisk)); 3041 for (i = UI_CURRENT; i < UI_SIZE; i++) 3042 buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); 3043 buffer->flags = cpu_to_be32(device->ldev->md.flags); 3044 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN); 3045 3046 buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect); 3047 buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset); 3048 buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements); 3049 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE); 3050 buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid); 3051 3052 buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset); 3053 buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size); 3054 3055 buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes); 3056 buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k); 3057 3058 D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset); 3059 sector = device->ldev->md.md_offset; 3060 3061 if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) { 3062 /* this was a try anyways ... */ 3063 drbd_err(device, "meta data update failed!\n"); 3064 drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR); 3065 } 3066 } 3067 3068 /** 3069 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set 3070 * @device: DRBD device. 3071 */ 3072 void drbd_md_sync(struct drbd_device *device) 3073 { 3074 struct meta_data_on_disk *buffer; 3075 3076 /* Don't accidentally change the DRBD meta data layout. */ 3077 BUILD_BUG_ON(UI_SIZE != 4); 3078 BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096); 3079 3080 del_timer(&device->md_sync_timer); 3081 /* timer may be rearmed by drbd_md_mark_dirty() now. */ 3082 if (!test_and_clear_bit(MD_DIRTY, &device->flags)) 3083 return; 3084 3085 /* We use here D_FAILED and not D_ATTACHING because we try to write 3086 * metadata even if we detach due to a disk failure! */ 3087 if (!get_ldev_if_state(device, D_FAILED)) 3088 return; 3089 3090 buffer = drbd_md_get_buffer(device, __func__); 3091 if (!buffer) 3092 goto out; 3093 3094 drbd_md_write(device, buffer); 3095 3096 /* Update device->ldev->md.la_size_sect, 3097 * since we updated it on metadata. */ 3098 device->ldev->md.la_size_sect = get_capacity(device->vdisk); 3099 3100 drbd_md_put_buffer(device); 3101 out: 3102 put_ldev(device); 3103 } 3104 3105 static int check_activity_log_stripe_size(struct drbd_device *device, 3106 struct meta_data_on_disk *on_disk, 3107 struct drbd_md *in_core) 3108 { 3109 u32 al_stripes = be32_to_cpu(on_disk->al_stripes); 3110 u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k); 3111 u64 al_size_4k; 3112 3113 /* both not set: default to old fixed size activity log */ 3114 if (al_stripes == 0 && al_stripe_size_4k == 0) { 3115 al_stripes = 1; 3116 al_stripe_size_4k = MD_32kB_SECT/8; 3117 } 3118 3119 /* some paranoia plausibility checks */ 3120 3121 /* we need both values to be set */ 3122 if (al_stripes == 0 || al_stripe_size_4k == 0) 3123 goto err; 3124 3125 al_size_4k = (u64)al_stripes * al_stripe_size_4k; 3126 3127 /* Upper limit of activity log area, to avoid potential overflow 3128 * problems in al_tr_number_to_on_disk_sector(). As right now, more 3129 * than 72 * 4k blocks total only increases the amount of history, 3130 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */ 3131 if (al_size_4k > (16 * 1024 * 1024/4)) 3132 goto err; 3133 3134 /* Lower limit: we need at least 8 transaction slots (32kB) 3135 * to not break existing setups */ 3136 if (al_size_4k < MD_32kB_SECT/8) 3137 goto err; 3138 3139 in_core->al_stripe_size_4k = al_stripe_size_4k; 3140 in_core->al_stripes = al_stripes; 3141 in_core->al_size_4k = al_size_4k; 3142 3143 return 0; 3144 err: 3145 drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n", 3146 al_stripes, al_stripe_size_4k); 3147 return -EINVAL; 3148 } 3149 3150 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev) 3151 { 3152 sector_t capacity = drbd_get_capacity(bdev->md_bdev); 3153 struct drbd_md *in_core = &bdev->md; 3154 s32 on_disk_al_sect; 3155 s32 on_disk_bm_sect; 3156 3157 /* The on-disk size of the activity log, calculated from offsets, and 3158 * the size of the activity log calculated from the stripe settings, 3159 * should match. 3160 * Though we could relax this a bit: it is ok, if the striped activity log 3161 * fits in the available on-disk activity log size. 3162 * Right now, that would break how resize is implemented. 3163 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware 3164 * of possible unused padding space in the on disk layout. */ 3165 if (in_core->al_offset < 0) { 3166 if (in_core->bm_offset > in_core->al_offset) 3167 goto err; 3168 on_disk_al_sect = -in_core->al_offset; 3169 on_disk_bm_sect = in_core->al_offset - in_core->bm_offset; 3170 } else { 3171 if (in_core->al_offset != MD_4kB_SECT) 3172 goto err; 3173 if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT) 3174 goto err; 3175 3176 on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT; 3177 on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset; 3178 } 3179 3180 /* old fixed size meta data is exactly that: fixed. */ 3181 if (in_core->meta_dev_idx >= 0) { 3182 if (in_core->md_size_sect != MD_128MB_SECT 3183 || in_core->al_offset != MD_4kB_SECT 3184 || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT 3185 || in_core->al_stripes != 1 3186 || in_core->al_stripe_size_4k != MD_32kB_SECT/8) 3187 goto err; 3188 } 3189 3190 if (capacity < in_core->md_size_sect) 3191 goto err; 3192 if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev)) 3193 goto err; 3194 3195 /* should be aligned, and at least 32k */ 3196 if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT)) 3197 goto err; 3198 3199 /* should fit (for now: exactly) into the available on-disk space; 3200 * overflow prevention is in check_activity_log_stripe_size() above. */ 3201 if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT) 3202 goto err; 3203 3204 /* again, should be aligned */ 3205 if (in_core->bm_offset & 7) 3206 goto err; 3207 3208 /* FIXME check for device grow with flex external meta data? */ 3209 3210 /* can the available bitmap space cover the last agreed device size? */ 3211 if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512) 3212 goto err; 3213 3214 return 0; 3215 3216 err: 3217 drbd_err(device, "meta data offsets don't make sense: idx=%d " 3218 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, " 3219 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n", 3220 in_core->meta_dev_idx, 3221 in_core->al_stripes, in_core->al_stripe_size_4k, 3222 in_core->al_offset, in_core->bm_offset, in_core->md_size_sect, 3223 (unsigned long long)in_core->la_size_sect, 3224 (unsigned long long)capacity); 3225 3226 return -EINVAL; 3227 } 3228 3229 3230 /** 3231 * drbd_md_read() - Reads in the meta data super block 3232 * @device: DRBD device. 3233 * @bdev: Device from which the meta data should be read in. 3234 * 3235 * Return NO_ERROR on success, and an enum drbd_ret_code in case 3236 * something goes wrong. 3237 * 3238 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS, 3239 * even before @bdev is assigned to @device->ldev. 3240 */ 3241 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev) 3242 { 3243 struct meta_data_on_disk *buffer; 3244 u32 magic, flags; 3245 int i, rv = NO_ERROR; 3246 3247 if (device->state.disk != D_DISKLESS) 3248 return ERR_DISK_CONFIGURED; 3249 3250 buffer = drbd_md_get_buffer(device, __func__); 3251 if (!buffer) 3252 return ERR_NOMEM; 3253 3254 /* First, figure out where our meta data superblock is located, 3255 * and read it. */ 3256 bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx; 3257 bdev->md.md_offset = drbd_md_ss(bdev); 3258 /* Even for (flexible or indexed) external meta data, 3259 * initially restrict us to the 4k superblock for now. 3260 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */ 3261 bdev->md.md_size_sect = 8; 3262 3263 if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset, 3264 REQ_OP_READ)) { 3265 /* NOTE: can't do normal error processing here as this is 3266 called BEFORE disk is attached */ 3267 drbd_err(device, "Error while reading metadata.\n"); 3268 rv = ERR_IO_MD_DISK; 3269 goto err; 3270 } 3271 3272 magic = be32_to_cpu(buffer->magic); 3273 flags = be32_to_cpu(buffer->flags); 3274 if (magic == DRBD_MD_MAGIC_84_UNCLEAN || 3275 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) { 3276 /* btw: that's Activity Log clean, not "all" clean. */ 3277 drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n"); 3278 rv = ERR_MD_UNCLEAN; 3279 goto err; 3280 } 3281 3282 rv = ERR_MD_INVALID; 3283 if (magic != DRBD_MD_MAGIC_08) { 3284 if (magic == DRBD_MD_MAGIC_07) 3285 drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n"); 3286 else 3287 drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n"); 3288 goto err; 3289 } 3290 3291 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) { 3292 drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n", 3293 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE); 3294 goto err; 3295 } 3296 3297 3298 /* convert to in_core endian */ 3299 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect); 3300 for (i = UI_CURRENT; i < UI_SIZE; i++) 3301 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]); 3302 bdev->md.flags = be32_to_cpu(buffer->flags); 3303 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid); 3304 3305 bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect); 3306 bdev->md.al_offset = be32_to_cpu(buffer->al_offset); 3307 bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset); 3308 3309 if (check_activity_log_stripe_size(device, buffer, &bdev->md)) 3310 goto err; 3311 if (check_offsets_and_sizes(device, bdev)) 3312 goto err; 3313 3314 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) { 3315 drbd_err(device, "unexpected bm_offset: %d (expected %d)\n", 3316 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset); 3317 goto err; 3318 } 3319 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) { 3320 drbd_err(device, "unexpected md_size: %u (expected %u)\n", 3321 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect); 3322 goto err; 3323 } 3324 3325 rv = NO_ERROR; 3326 3327 spin_lock_irq(&device->resource->req_lock); 3328 if (device->state.conn < C_CONNECTED) { 3329 unsigned int peer; 3330 peer = be32_to_cpu(buffer->la_peer_max_bio_size); 3331 peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE); 3332 device->peer_max_bio_size = peer; 3333 } 3334 spin_unlock_irq(&device->resource->req_lock); 3335 3336 err: 3337 drbd_md_put_buffer(device); 3338 3339 return rv; 3340 } 3341 3342 /** 3343 * drbd_md_mark_dirty() - Mark meta data super block as dirty 3344 * @device: DRBD device. 3345 * 3346 * Call this function if you change anything that should be written to 3347 * the meta-data super block. This function sets MD_DIRTY, and starts a 3348 * timer that ensures that within five seconds you have to call drbd_md_sync(). 3349 */ 3350 void drbd_md_mark_dirty(struct drbd_device *device) 3351 { 3352 if (!test_and_set_bit(MD_DIRTY, &device->flags)) 3353 mod_timer(&device->md_sync_timer, jiffies + 5*HZ); 3354 } 3355 3356 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local) 3357 { 3358 int i; 3359 3360 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++) 3361 device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i]; 3362 } 3363 3364 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) 3365 { 3366 if (idx == UI_CURRENT) { 3367 if (device->state.role == R_PRIMARY) 3368 val |= 1; 3369 else 3370 val &= ~((u64)1); 3371 3372 drbd_set_ed_uuid(device, val); 3373 } 3374 3375 device->ldev->md.uuid[idx] = val; 3376 drbd_md_mark_dirty(device); 3377 } 3378 3379 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) 3380 { 3381 unsigned long flags; 3382 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); 3383 __drbd_uuid_set(device, idx, val); 3384 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); 3385 } 3386 3387 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) 3388 { 3389 unsigned long flags; 3390 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); 3391 if (device->ldev->md.uuid[idx]) { 3392 drbd_uuid_move_history(device); 3393 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx]; 3394 } 3395 __drbd_uuid_set(device, idx, val); 3396 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); 3397 } 3398 3399 /** 3400 * drbd_uuid_new_current() - Creates a new current UUID 3401 * @device: DRBD device. 3402 * 3403 * Creates a new current UUID, and rotates the old current UUID into 3404 * the bitmap slot. Causes an incremental resync upon next connect. 3405 */ 3406 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local) 3407 { 3408 u64 val; 3409 unsigned long long bm_uuid; 3410 3411 get_random_bytes(&val, sizeof(u64)); 3412 3413 spin_lock_irq(&device->ldev->md.uuid_lock); 3414 bm_uuid = device->ldev->md.uuid[UI_BITMAP]; 3415 3416 if (bm_uuid) 3417 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid); 3418 3419 device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT]; 3420 __drbd_uuid_set(device, UI_CURRENT, val); 3421 spin_unlock_irq(&device->ldev->md.uuid_lock); 3422 3423 drbd_print_uuids(device, "new current UUID"); 3424 /* get it to stable storage _now_ */ 3425 drbd_md_sync(device); 3426 } 3427 3428 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local) 3429 { 3430 unsigned long flags; 3431 if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0) 3432 return; 3433 3434 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); 3435 if (val == 0) { 3436 drbd_uuid_move_history(device); 3437 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP]; 3438 device->ldev->md.uuid[UI_BITMAP] = 0; 3439 } else { 3440 unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP]; 3441 if (bm_uuid) 3442 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid); 3443 3444 device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1); 3445 } 3446 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); 3447 3448 drbd_md_mark_dirty(device); 3449 } 3450 3451 /** 3452 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() 3453 * @device: DRBD device. 3454 * 3455 * Sets all bits in the bitmap and writes the whole bitmap to stable storage. 3456 */ 3457 int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local) 3458 { 3459 int rv = -EIO; 3460 3461 drbd_md_set_flag(device, MDF_FULL_SYNC); 3462 drbd_md_sync(device); 3463 drbd_bm_set_all(device); 3464 3465 rv = drbd_bm_write(device); 3466 3467 if (!rv) { 3468 drbd_md_clear_flag(device, MDF_FULL_SYNC); 3469 drbd_md_sync(device); 3470 } 3471 3472 return rv; 3473 } 3474 3475 /** 3476 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() 3477 * @device: DRBD device. 3478 * 3479 * Clears all bits in the bitmap and writes the whole bitmap to stable storage. 3480 */ 3481 int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local) 3482 { 3483 drbd_resume_al(device); 3484 drbd_bm_clear_all(device); 3485 return drbd_bm_write(device); 3486 } 3487 3488 static int w_bitmap_io(struct drbd_work *w, int unused) 3489 { 3490 struct drbd_device *device = 3491 container_of(w, struct drbd_device, bm_io_work.w); 3492 struct bm_io_work *work = &device->bm_io_work; 3493 int rv = -EIO; 3494 3495 if (work->flags != BM_LOCKED_CHANGE_ALLOWED) { 3496 int cnt = atomic_read(&device->ap_bio_cnt); 3497 if (cnt) 3498 drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n", 3499 cnt, work->why); 3500 } 3501 3502 if (get_ldev(device)) { 3503 drbd_bm_lock(device, work->why, work->flags); 3504 rv = work->io_fn(device); 3505 drbd_bm_unlock(device); 3506 put_ldev(device); 3507 } 3508 3509 clear_bit_unlock(BITMAP_IO, &device->flags); 3510 wake_up(&device->misc_wait); 3511 3512 if (work->done) 3513 work->done(device, rv); 3514 3515 clear_bit(BITMAP_IO_QUEUED, &device->flags); 3516 work->why = NULL; 3517 work->flags = 0; 3518 3519 return 0; 3520 } 3521 3522 /** 3523 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap 3524 * @device: DRBD device. 3525 * @io_fn: IO callback to be called when bitmap IO is possible 3526 * @done: callback to be called after the bitmap IO was performed 3527 * @why: Descriptive text of the reason for doing the IO 3528 * 3529 * While IO on the bitmap happens we freeze application IO thus we ensure 3530 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be 3531 * called from worker context. It MUST NOT be used while a previous such 3532 * work is still pending! 3533 * 3534 * Its worker function encloses the call of io_fn() by get_ldev() and 3535 * put_ldev(). 3536 */ 3537 void drbd_queue_bitmap_io(struct drbd_device *device, 3538 int (*io_fn)(struct drbd_device *), 3539 void (*done)(struct drbd_device *, int), 3540 char *why, enum bm_flag flags) 3541 { 3542 D_ASSERT(device, current == first_peer_device(device)->connection->worker.task); 3543 3544 D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags)); 3545 D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags)); 3546 D_ASSERT(device, list_empty(&device->bm_io_work.w.list)); 3547 if (device->bm_io_work.why) 3548 drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n", 3549 why, device->bm_io_work.why); 3550 3551 device->bm_io_work.io_fn = io_fn; 3552 device->bm_io_work.done = done; 3553 device->bm_io_work.why = why; 3554 device->bm_io_work.flags = flags; 3555 3556 spin_lock_irq(&device->resource->req_lock); 3557 set_bit(BITMAP_IO, &device->flags); 3558 /* don't wait for pending application IO if the caller indicates that 3559 * application IO does not conflict anyways. */ 3560 if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) { 3561 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags)) 3562 drbd_queue_work(&first_peer_device(device)->connection->sender_work, 3563 &device->bm_io_work.w); 3564 } 3565 spin_unlock_irq(&device->resource->req_lock); 3566 } 3567 3568 /** 3569 * drbd_bitmap_io() - Does an IO operation on the whole bitmap 3570 * @device: DRBD device. 3571 * @io_fn: IO callback to be called when bitmap IO is possible 3572 * @why: Descriptive text of the reason for doing the IO 3573 * 3574 * freezes application IO while that the actual IO operations runs. This 3575 * functions MAY NOT be called from worker context. 3576 */ 3577 int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *), 3578 char *why, enum bm_flag flags) 3579 { 3580 /* Only suspend io, if some operation is supposed to be locked out */ 3581 const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST); 3582 int rv; 3583 3584 D_ASSERT(device, current != first_peer_device(device)->connection->worker.task); 3585 3586 if (do_suspend_io) 3587 drbd_suspend_io(device); 3588 3589 drbd_bm_lock(device, why, flags); 3590 rv = io_fn(device); 3591 drbd_bm_unlock(device); 3592 3593 if (do_suspend_io) 3594 drbd_resume_io(device); 3595 3596 return rv; 3597 } 3598 3599 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local) 3600 { 3601 if ((device->ldev->md.flags & flag) != flag) { 3602 drbd_md_mark_dirty(device); 3603 device->ldev->md.flags |= flag; 3604 } 3605 } 3606 3607 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local) 3608 { 3609 if ((device->ldev->md.flags & flag) != 0) { 3610 drbd_md_mark_dirty(device); 3611 device->ldev->md.flags &= ~flag; 3612 } 3613 } 3614 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag) 3615 { 3616 return (bdev->md.flags & flag) != 0; 3617 } 3618 3619 static void md_sync_timer_fn(struct timer_list *t) 3620 { 3621 struct drbd_device *device = from_timer(device, t, md_sync_timer); 3622 drbd_device_post_work(device, MD_SYNC); 3623 } 3624 3625 const char *cmdname(enum drbd_packet cmd) 3626 { 3627 /* THINK may need to become several global tables 3628 * when we want to support more than 3629 * one PRO_VERSION */ 3630 static const char *cmdnames[] = { 3631 [P_DATA] = "Data", 3632 [P_WSAME] = "WriteSame", 3633 [P_TRIM] = "Trim", 3634 [P_DATA_REPLY] = "DataReply", 3635 [P_RS_DATA_REPLY] = "RSDataReply", 3636 [P_BARRIER] = "Barrier", 3637 [P_BITMAP] = "ReportBitMap", 3638 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget", 3639 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource", 3640 [P_UNPLUG_REMOTE] = "UnplugRemote", 3641 [P_DATA_REQUEST] = "DataRequest", 3642 [P_RS_DATA_REQUEST] = "RSDataRequest", 3643 [P_SYNC_PARAM] = "SyncParam", 3644 [P_SYNC_PARAM89] = "SyncParam89", 3645 [P_PROTOCOL] = "ReportProtocol", 3646 [P_UUIDS] = "ReportUUIDs", 3647 [P_SIZES] = "ReportSizes", 3648 [P_STATE] = "ReportState", 3649 [P_SYNC_UUID] = "ReportSyncUUID", 3650 [P_AUTH_CHALLENGE] = "AuthChallenge", 3651 [P_AUTH_RESPONSE] = "AuthResponse", 3652 [P_PING] = "Ping", 3653 [P_PING_ACK] = "PingAck", 3654 [P_RECV_ACK] = "RecvAck", 3655 [P_WRITE_ACK] = "WriteAck", 3656 [P_RS_WRITE_ACK] = "RSWriteAck", 3657 [P_SUPERSEDED] = "Superseded", 3658 [P_NEG_ACK] = "NegAck", 3659 [P_NEG_DREPLY] = "NegDReply", 3660 [P_NEG_RS_DREPLY] = "NegRSDReply", 3661 [P_BARRIER_ACK] = "BarrierAck", 3662 [P_STATE_CHG_REQ] = "StateChgRequest", 3663 [P_STATE_CHG_REPLY] = "StateChgReply", 3664 [P_OV_REQUEST] = "OVRequest", 3665 [P_OV_REPLY] = "OVReply", 3666 [P_OV_RESULT] = "OVResult", 3667 [P_CSUM_RS_REQUEST] = "CsumRSRequest", 3668 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync", 3669 [P_COMPRESSED_BITMAP] = "CBitmap", 3670 [P_DELAY_PROBE] = "DelayProbe", 3671 [P_OUT_OF_SYNC] = "OutOfSync", 3672 [P_RETRY_WRITE] = "RetryWrite", 3673 [P_RS_CANCEL] = "RSCancel", 3674 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req", 3675 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply", 3676 [P_RETRY_WRITE] = "retry_write", 3677 [P_PROTOCOL_UPDATE] = "protocol_update", 3678 [P_RS_THIN_REQ] = "rs_thin_req", 3679 [P_RS_DEALLOCATED] = "rs_deallocated", 3680 3681 /* enum drbd_packet, but not commands - obsoleted flags: 3682 * P_MAY_IGNORE 3683 * P_MAX_OPT_CMD 3684 */ 3685 }; 3686 3687 /* too big for the array: 0xfffX */ 3688 if (cmd == P_INITIAL_META) 3689 return "InitialMeta"; 3690 if (cmd == P_INITIAL_DATA) 3691 return "InitialData"; 3692 if (cmd == P_CONNECTION_FEATURES) 3693 return "ConnectionFeatures"; 3694 if (cmd >= ARRAY_SIZE(cmdnames)) 3695 return "Unknown"; 3696 return cmdnames[cmd]; 3697 } 3698 3699 /** 3700 * drbd_wait_misc - wait for a request to make progress 3701 * @device: device associated with the request 3702 * @i: the struct drbd_interval embedded in struct drbd_request or 3703 * struct drbd_peer_request 3704 */ 3705 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i) 3706 { 3707 struct net_conf *nc; 3708 DEFINE_WAIT(wait); 3709 long timeout; 3710 3711 rcu_read_lock(); 3712 nc = rcu_dereference(first_peer_device(device)->connection->net_conf); 3713 if (!nc) { 3714 rcu_read_unlock(); 3715 return -ETIMEDOUT; 3716 } 3717 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT; 3718 rcu_read_unlock(); 3719 3720 /* Indicate to wake up device->misc_wait on progress. */ 3721 i->waiting = true; 3722 prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE); 3723 spin_unlock_irq(&device->resource->req_lock); 3724 timeout = schedule_timeout(timeout); 3725 finish_wait(&device->misc_wait, &wait); 3726 spin_lock_irq(&device->resource->req_lock); 3727 if (!timeout || device->state.conn < C_CONNECTED) 3728 return -ETIMEDOUT; 3729 if (signal_pending(current)) 3730 return -ERESTARTSYS; 3731 return 0; 3732 } 3733 3734 void lock_all_resources(void) 3735 { 3736 struct drbd_resource *resource; 3737 int __maybe_unused i = 0; 3738 3739 mutex_lock(&resources_mutex); 3740 local_irq_disable(); 3741 for_each_resource(resource, &drbd_resources) 3742 spin_lock_nested(&resource->req_lock, i++); 3743 } 3744 3745 void unlock_all_resources(void) 3746 { 3747 struct drbd_resource *resource; 3748 3749 for_each_resource(resource, &drbd_resources) 3750 spin_unlock(&resource->req_lock); 3751 local_irq_enable(); 3752 mutex_unlock(&resources_mutex); 3753 } 3754 3755 #ifdef CONFIG_DRBD_FAULT_INJECTION 3756 /* Fault insertion support including random number generator shamelessly 3757 * stolen from kernel/rcutorture.c */ 3758 struct fault_random_state { 3759 unsigned long state; 3760 unsigned long count; 3761 }; 3762 3763 #define FAULT_RANDOM_MULT 39916801 /* prime */ 3764 #define FAULT_RANDOM_ADD 479001701 /* prime */ 3765 #define FAULT_RANDOM_REFRESH 10000 3766 3767 /* 3768 * Crude but fast random-number generator. Uses a linear congruential 3769 * generator, with occasional help from get_random_bytes(). 3770 */ 3771 static unsigned long 3772 _drbd_fault_random(struct fault_random_state *rsp) 3773 { 3774 long refresh; 3775 3776 if (!rsp->count--) { 3777 get_random_bytes(&refresh, sizeof(refresh)); 3778 rsp->state += refresh; 3779 rsp->count = FAULT_RANDOM_REFRESH; 3780 } 3781 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD; 3782 return swahw32(rsp->state); 3783 } 3784 3785 static char * 3786 _drbd_fault_str(unsigned int type) { 3787 static char *_faults[] = { 3788 [DRBD_FAULT_MD_WR] = "Meta-data write", 3789 [DRBD_FAULT_MD_RD] = "Meta-data read", 3790 [DRBD_FAULT_RS_WR] = "Resync write", 3791 [DRBD_FAULT_RS_RD] = "Resync read", 3792 [DRBD_FAULT_DT_WR] = "Data write", 3793 [DRBD_FAULT_DT_RD] = "Data read", 3794 [DRBD_FAULT_DT_RA] = "Data read ahead", 3795 [DRBD_FAULT_BM_ALLOC] = "BM allocation", 3796 [DRBD_FAULT_AL_EE] = "EE allocation", 3797 [DRBD_FAULT_RECEIVE] = "receive data corruption", 3798 }; 3799 3800 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**"; 3801 } 3802 3803 unsigned int 3804 _drbd_insert_fault(struct drbd_device *device, unsigned int type) 3805 { 3806 static struct fault_random_state rrs = {0, 0}; 3807 3808 unsigned int ret = ( 3809 (drbd_fault_devs == 0 || 3810 ((1 << device_to_minor(device)) & drbd_fault_devs) != 0) && 3811 (((_drbd_fault_random(&rrs) % 100) + 1) <= drbd_fault_rate)); 3812 3813 if (ret) { 3814 drbd_fault_count++; 3815 3816 if (__ratelimit(&drbd_ratelimit_state)) 3817 drbd_warn(device, "***Simulating %s failure\n", 3818 _drbd_fault_str(type)); 3819 } 3820 3821 return ret; 3822 } 3823 #endif 3824 3825 const char *drbd_buildtag(void) 3826 { 3827 /* DRBD built from external sources has here a reference to the 3828 git hash of the source code. */ 3829 3830 static char buildtag[38] = "\0uilt-in"; 3831 3832 if (buildtag[0] == 0) { 3833 #ifdef MODULE 3834 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion); 3835 #else 3836 buildtag[0] = 'b'; 3837 #endif 3838 } 3839 3840 return buildtag; 3841 } 3842 3843 module_init(drbd_init) 3844 module_exit(drbd_cleanup) 3845 3846 EXPORT_SYMBOL(drbd_conn_str); 3847 EXPORT_SYMBOL(drbd_role_str); 3848 EXPORT_SYMBOL(drbd_disk_str); 3849 EXPORT_SYMBOL(drbd_set_st_err_str); 3850