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