drivers/block/drbd/drbd_main.c

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