drivers/md/md-cluster.c

   1 /*
   2  * Copyright (C) 2015, SUSE
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2, or (at your option)
   7  * any later version.
   8  *
   9  */
  10
  11
  12 #include <linux/module.h>
  13 #include <linux/dlm.h>
  14 #include <linux/sched.h>
  15 #include <linux/raid/md_p.h>
  16 #include "md.h"
  17 #include "bitmap.h"
  18 #include "md-cluster.h"
  19
  20 #define LVB_SIZE        64
  21 #define NEW_DEV_TIMEOUT 5000
  22
  23 struct dlm_lock_resource {
  24         dlm_lockspace_t *ls;
  25         struct dlm_lksb lksb;
  26         char *name; /* lock name. */
  27         uint32_t flags; /* flags to pass to dlm_lock() */
  28         struct completion completion; /* completion for synchronized locking */
  29         void (*bast)(void *arg, int mode); /* blocking AST function pointer*/
  30         struct mddev *mddev; /* pointing back to mddev. */
  31 };
  32
  33 struct suspend_info {
  34         int slot;
  35         sector_t lo;
  36         sector_t hi;
  37         struct list_head list;
  38 };
  39
  40 struct resync_info {
  41         __le64 lo;
  42         __le64 hi;
  43 };
  44
  45 /* md_cluster_info flags */
  46 #define         MD_CLUSTER_WAITING_FOR_NEWDISK          1
  47 #define         MD_CLUSTER_SUSPEND_READ_BALANCING       2
  48
  49
  50 struct md_cluster_info {
  51         /* dlm lock space and resources for clustered raid. */
  52         dlm_lockspace_t *lockspace;
  53         int slot_number;
  54         struct completion completion;
  55         struct dlm_lock_resource *sb_lock;
  56         struct mutex sb_mutex;
  57         struct dlm_lock_resource *bitmap_lockres;
  58         struct list_head suspend_list;
  59         spinlock_t suspend_lock;
  60         struct md_thread *recovery_thread;
  61         unsigned long recovery_map;
  62         /* communication loc resources */
  63         struct dlm_lock_resource *ack_lockres;
  64         struct dlm_lock_resource *message_lockres;
  65         struct dlm_lock_resource *token_lockres;
  66         struct dlm_lock_resource *no_new_dev_lockres;
  67         struct md_thread *recv_thread;
  68         struct completion newdisk_completion;
  69         unsigned long state;
  70 };
  71
  72 enum msg_type {
  73         METADATA_UPDATED = 0,
  74         RESYNCING,
  75         NEWDISK,
  76         REMOVE,
  77         RE_ADD,
  78 };
  79
  80 struct cluster_msg {
  81         int type;
  82         int slot;
  83         /* TODO: Unionize this for smaller footprint */
  84         sector_t low;
  85         sector_t high;
  86         char uuid[16];
  87         int raid_slot;
  88 };
  89
  90 static void sync_ast(void *arg)
  91 {
  92         struct dlm_lock_resource *res;
  93
  94         res = (struct dlm_lock_resource *) arg;
  95         complete(&res->completion);
  96 }
  97
  98 static int dlm_lock_sync(struct dlm_lock_resource *res, int mode)
  99 {
 100         int ret = 0;
 101
 102         init_completion(&res->completion);
 103         ret = dlm_lock(res->ls, mode, &res->lksb,
 104                         res->flags, res->name, strlen(res->name),
 105                         0, sync_ast, res, res->bast);
 106         if (ret)
 107                 return ret;
 108         wait_for_completion(&res->completion);
 109         return res->lksb.sb_status;
 110 }
 111
 112 static int dlm_unlock_sync(struct dlm_lock_resource *res)
 113 {
 114         return dlm_lock_sync(res, DLM_LOCK_NL);
 115 }
 116
 117 static struct dlm_lock_resource *lockres_init(struct mddev *mddev,
 118                 char *name, void (*bastfn)(void *arg, int mode), int with_lvb)
 119 {
 120         struct dlm_lock_resource *res = NULL;
 121         int ret, namelen;
 122         struct md_cluster_info *cinfo = mddev->cluster_info;
 123
 124         res = kzalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL);
 125         if (!res)
 126                 return NULL;
 127         res->ls = cinfo->lockspace;
 128         res->mddev = mddev;
 129         namelen = strlen(name);
 130         res->name = kzalloc(namelen + 1, GFP_KERNEL);
 131         if (!res->name) {
 132                 pr_err("md-cluster: Unable to allocate resource name for resource %s\n", name);
 133                 goto out_err;
 134         }
 135         strlcpy(res->name, name, namelen + 1);
 136         if (with_lvb) {
 137                 res->lksb.sb_lvbptr = kzalloc(LVB_SIZE, GFP_KERNEL);
 138                 if (!res->lksb.sb_lvbptr) {
 139                         pr_err("md-cluster: Unable to allocate LVB for resource %s\n", name);
 140                         goto out_err;
 141                 }
 142                 res->flags = DLM_LKF_VALBLK;
 143         }
 144
 145         if (bastfn)
 146                 res->bast = bastfn;
 147
 148         res->flags |= DLM_LKF_EXPEDITE;
 149
 150         ret = dlm_lock_sync(res, DLM_LOCK_NL);
 151         if (ret) {
 152                 pr_err("md-cluster: Unable to lock NL on new lock resource %s\n", name);
 153                 goto out_err;
 154         }
 155         res->flags &= ~DLM_LKF_EXPEDITE;
 156         res->flags |= DLM_LKF_CONVERT;
 157
 158         return res;
 159 out_err:
 160         kfree(res->lksb.sb_lvbptr);
 161         kfree(res->name);
 162         kfree(res);
 163         return NULL;
 164 }
 165
 166 static void lockres_free(struct dlm_lock_resource *res)
 167 {
 168         if (!res)
 169                 return;
 170
 171         init_completion(&res->completion);
 172         dlm_unlock(res->ls, res->lksb.sb_lkid, 0, &res->lksb, res);
 173         wait_for_completion(&res->completion);
 174
 175         kfree(res->name);
 176         kfree(res->lksb.sb_lvbptr);
 177         kfree(res);
 178 }
 179
 180 static char *pretty_uuid(char *dest, char *src)
 181 {
 182         int i, len = 0;
 183
 184         for (i = 0; i < 16; i++) {
 185                 if (i == 4 || i == 6 || i == 8 || i == 10)
 186                         len += sprintf(dest + len, "-");
 187                 len += sprintf(dest + len, "%02x", (__u8)src[i]);
 188         }
 189         return dest;
 190 }
 191
 192 static void add_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres,
 193                 sector_t lo, sector_t hi)
 194 {
 195         struct resync_info *ri;
 196
 197         ri = (struct resync_info *)lockres->lksb.sb_lvbptr;
 198         ri->lo = cpu_to_le64(lo);
 199         ri->hi = cpu_to_le64(hi);
 200 }
 201
 202 static struct suspend_info *read_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres)
 203 {
 204         struct resync_info ri;
 205         struct suspend_info *s = NULL;
 206         sector_t hi = 0;
 207
 208         dlm_lock_sync(lockres, DLM_LOCK_CR);
 209         memcpy(&ri, lockres->lksb.sb_lvbptr, sizeof(struct resync_info));
 210         hi = le64_to_cpu(ri.hi);
 211         if (ri.hi > 0) {
 212                 s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL);
 213                 if (!s)
 214                         goto out;
 215                 s->hi = hi;
 216                 s->lo = le64_to_cpu(ri.lo);
 217         }
 218         dlm_unlock_sync(lockres);
 219 out:
 220         return s;
 221 }
 222
 223 static void recover_bitmaps(struct md_thread *thread)
 224 {
 225         struct mddev *mddev = thread->mddev;
 226         struct md_cluster_info *cinfo = mddev->cluster_info;
 227         struct dlm_lock_resource *bm_lockres;
 228         char str[64];
 229         int slot, ret;
 230         struct suspend_info *s, *tmp;
 231         sector_t lo, hi;
 232
 233         while (cinfo->recovery_map) {
 234                 slot = fls64((u64)cinfo->recovery_map) - 1;
 235
 236                 /* Clear suspend_area associated with the bitmap */
 237                 spin_lock_irq(&cinfo->suspend_lock);
 238                 list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list)
 239                         if (slot == s->slot) {
 240                                 list_del(&s->list);
 241                                 kfree(s);
 242                         }
 243                 spin_unlock_irq(&cinfo->suspend_lock);
 244
 245                 snprintf(str, 64, "bitmap%04d", slot);
 246                 bm_lockres = lockres_init(mddev, str, NULL, 1);
 247                 if (!bm_lockres) {
 248                         pr_err("md-cluster: Cannot initialize bitmaps\n");
 249                         goto clear_bit;
 250                 }
 251
 252                 ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
 253                 if (ret) {
 254                         pr_err("md-cluster: Could not DLM lock %s: %d\n",
 255                                         str, ret);
 256                         goto clear_bit;
 257                 }
 258                 ret = bitmap_copy_from_slot(mddev, slot, &lo, &hi, true);
 259                 if (ret) {
 260                         pr_err("md-cluster: Could not copy data from bitmap %d\n", slot);
 261                         goto dlm_unlock;
 262                 }
 263                 if (hi > 0) {
 264                         /* TODO:Wait for current resync to get over */
 265                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 266                         if (lo < mddev->recovery_cp)
 267                                 mddev->recovery_cp = lo;
 268                         md_check_recovery(mddev);
 269                 }
 270 dlm_unlock:
 271                 dlm_unlock_sync(bm_lockres);
 272 clear_bit:
 273                 clear_bit(slot, &cinfo->recovery_map);
 274         }
 275 }
 276
 277 static void recover_prep(void *arg)
 278 {
 279         struct mddev *mddev = arg;
 280         struct md_cluster_info *cinfo = mddev->cluster_info;
 281         set_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state);
 282 }
 283
 284 static void recover_slot(void *arg, struct dlm_slot *slot)
 285 {
 286         struct mddev *mddev = arg;
 287         struct md_cluster_info *cinfo = mddev->cluster_info;
 288
 289         pr_info("md-cluster: %s Node %d/%d down. My slot: %d. Initiating recovery.\n",
 290                         mddev->bitmap_info.cluster_name,
 291                         slot->nodeid, slot->slot,
 292                         cinfo->slot_number);
 293         set_bit(slot->slot - 1, &cinfo->recovery_map);
 294         if (!cinfo->recovery_thread) {
 295                 cinfo->recovery_thread = md_register_thread(recover_bitmaps,
 296                                 mddev, "recover");
 297                 if (!cinfo->recovery_thread) {
 298                         pr_warn("md-cluster: Could not create recovery thread\n");
 299                         return;
 300                 }
 301         }
 302         md_wakeup_thread(cinfo->recovery_thread);
 303 }
 304
 305 static void recover_done(void *arg, struct dlm_slot *slots,
 306                 int num_slots, int our_slot,
 307                 uint32_t generation)
 308 {
 309         struct mddev *mddev = arg;
 310         struct md_cluster_info *cinfo = mddev->cluster_info;
 311
 312         cinfo->slot_number = our_slot;
 313         complete(&cinfo->completion);
 314         clear_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state);
 315 }
 316
 317 static const struct dlm_lockspace_ops md_ls_ops = {
 318         .recover_prep = recover_prep,
 319         .recover_slot = recover_slot,
 320         .recover_done = recover_done,
 321 };
 322
 323 /*
 324  * The BAST function for the ack lock resource
 325  * This function wakes up the receive thread in
 326  * order to receive and process the message.
 327  */
 328 static void ack_bast(void *arg, int mode)
 329 {
 330         struct dlm_lock_resource *res = (struct dlm_lock_resource *)arg;
 331         struct md_cluster_info *cinfo = res->mddev->cluster_info;
 332
 333         if (mode == DLM_LOCK_EX)
 334                 md_wakeup_thread(cinfo->recv_thread);
 335 }
 336
 337 static void __remove_suspend_info(struct md_cluster_info *cinfo, int slot)
 338 {
 339         struct suspend_info *s, *tmp;
 340
 341         list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list)
 342                 if (slot == s->slot) {
 343                         pr_info("%s:%d Deleting suspend_info: %d\n",
 344                                         __func__, __LINE__, slot);
 345                         list_del(&s->list);
 346                         kfree(s);
 347                         break;
 348                 }
 349 }
 350
 351 static void remove_suspend_info(struct md_cluster_info *cinfo, int slot)
 352 {
 353         spin_lock_irq(&cinfo->suspend_lock);
 354         __remove_suspend_info(cinfo, slot);
 355         spin_unlock_irq(&cinfo->suspend_lock);
 356 }
 357
 358
 359 static void process_suspend_info(struct md_cluster_info *cinfo,
 360                 int slot, sector_t lo, sector_t hi)
 361 {
 362         struct suspend_info *s;
 363
 364         if (!hi) {
 365                 remove_suspend_info(cinfo, slot);
 366                 return;
 367         }
 368         s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL);
 369         if (!s)
 370                 return;
 371         s->slot = slot;
 372         s->lo = lo;
 373         s->hi = hi;
 374         spin_lock_irq(&cinfo->suspend_lock);
 375         /* Remove existing entry (if exists) before adding */
 376         __remove_suspend_info(cinfo, slot);
 377         list_add(&s->list, &cinfo->suspend_list);
 378         spin_unlock_irq(&cinfo->suspend_lock);
 379 }
 380
 381 static void process_add_new_disk(struct mddev *mddev, struct cluster_msg *cmsg)
 382 {
 383         char disk_uuid[64];
 384         struct md_cluster_info *cinfo = mddev->cluster_info;
 385         char event_name[] = "EVENT=ADD_DEVICE";
 386         char raid_slot[16];
 387         char *envp[] = {event_name, disk_uuid, raid_slot, NULL};
 388         int len;
 389
 390         len = snprintf(disk_uuid, 64, "DEVICE_UUID=");
 391         pretty_uuid(disk_uuid + len, cmsg->uuid);
 392         snprintf(raid_slot, 16, "RAID_DISK=%d", cmsg->raid_slot);
 393         pr_info("%s:%d Sending kobject change with %s and %s\n", __func__, __LINE__, disk_uuid, raid_slot);
 394         init_completion(&cinfo->newdisk_completion);
 395         set_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state);
 396         kobject_uevent_env(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE, envp);
 397         wait_for_completion_timeout(&cinfo->newdisk_completion,
 398                         NEW_DEV_TIMEOUT);
 399         clear_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state);
 400 }
 401
 402
 403 static void process_metadata_update(struct mddev *mddev, struct cluster_msg *msg)
 404 {
 405         struct md_cluster_info *cinfo = mddev->cluster_info;
 406
 407         md_reload_sb(mddev);
 408         dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR);
 409 }
 410
 411 static void process_remove_disk(struct mddev *mddev, struct cluster_msg *msg)
 412 {
 413         struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, msg->raid_slot);
 414
 415         if (rdev)
 416                 md_kick_rdev_from_array(rdev);
 417         else
 418                 pr_warn("%s: %d Could not find disk(%d) to REMOVE\n", __func__, __LINE__, msg->raid_slot);
 419 }
 420
 421 static void process_readd_disk(struct mddev *mddev, struct cluster_msg *msg)
 422 {
 423         struct md_rdev *rdev = md_find_rdev_nr_rcu(mddev, msg->raid_slot);
 424
 425         if (rdev && test_bit(Faulty, &rdev->flags))
 426                 clear_bit(Faulty, &rdev->flags);
 427         else
 428                 pr_warn("%s: %d Could not find disk(%d) which is faulty", __func__, __LINE__, msg->raid_slot);
 429 }
 430
 431 static void process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg)
 432 {
 433         switch (msg->type) {
 434         case METADATA_UPDATED:
 435                 pr_info("%s: %d Received message: METADATA_UPDATE from %d\n",
 436                         __func__, __LINE__, msg->slot);
 437                 process_metadata_update(mddev, msg);
 438                 break;
 439         case RESYNCING:
 440                 pr_info("%s: %d Received message: RESYNCING from %d\n",
 441                         __func__, __LINE__, msg->slot);
 442                 process_suspend_info(mddev->cluster_info, msg->slot,
 443                                 msg->low, msg->high);
 444                 break;
 445         case NEWDISK:
 446                 pr_info("%s: %d Received message: NEWDISK from %d\n",
 447                         __func__, __LINE__, msg->slot);
 448                 process_add_new_disk(mddev, msg);
 449                 break;
 450         case REMOVE:
 451                 pr_info("%s: %d Received REMOVE from %d\n",
 452                         __func__, __LINE__, msg->slot);
 453                 process_remove_disk(mddev, msg);
 454                 break;
 455         case RE_ADD:
 456                 pr_info("%s: %d Received RE_ADD from %d\n",
 457                         __func__, __LINE__, msg->slot);
 458                 process_readd_disk(mddev, msg);
 459                 break;
 460         default:
 461                 pr_warn("%s:%d Received unknown message from %d\n",
 462                         __func__, __LINE__, msg->slot);
 463         }
 464 }
 465
 466 /*
 467  * thread for receiving message
 468  */
 469 static void recv_daemon(struct md_thread *thread)
 470 {
 471         struct md_cluster_info *cinfo = thread->mddev->cluster_info;
 472         struct dlm_lock_resource *ack_lockres = cinfo->ack_lockres;
 473         struct dlm_lock_resource *message_lockres = cinfo->message_lockres;
 474         struct cluster_msg msg;
 475
 476         /*get CR on Message*/
 477         if (dlm_lock_sync(message_lockres, DLM_LOCK_CR)) {
 478                 pr_err("md/raid1:failed to get CR on MESSAGE\n");
 479                 return;
 480         }
 481
 482         /* read lvb and wake up thread to process this message_lockres */
 483         memcpy(&msg, message_lockres->lksb.sb_lvbptr, sizeof(struct cluster_msg));
 484         process_recvd_msg(thread->mddev, &msg);
 485
 486         /*release CR on ack_lockres*/
 487         dlm_unlock_sync(ack_lockres);
 488         /*up-convert to EX on message_lockres*/
 489         dlm_lock_sync(message_lockres, DLM_LOCK_EX);
 490         /*get CR on ack_lockres again*/
 491         dlm_lock_sync(ack_lockres, DLM_LOCK_CR);
 492         /*release CR on message_lockres*/
 493         dlm_unlock_sync(message_lockres);
 494 }
 495
 496 /* lock_comm()
 497  * Takes the lock on the TOKEN lock resource so no other
 498  * node can communicate while the operation is underway.
 499  */
 500 static int lock_comm(struct md_cluster_info *cinfo)
 501 {
 502         int error;
 503
 504         error = dlm_lock_sync(cinfo->token_lockres, DLM_LOCK_EX);
 505         if (error)
 506                 pr_err("md-cluster(%s:%d): failed to get EX on TOKEN (%d)\n",
 507                                 __func__, __LINE__, error);
 508         return error;
 509 }
 510
 511 static void unlock_comm(struct md_cluster_info *cinfo)
 512 {
 513         dlm_unlock_sync(cinfo->token_lockres);
 514 }
 515
 516 /* __sendmsg()
 517  * This function performs the actual sending of the message. This function is
 518  * usually called after performing the encompassing operation
 519  * The function:
 520  * 1. Grabs the message lockresource in EX mode
 521  * 2. Copies the message to the message LVB
 522  * 3. Downconverts message lockresource to CR
 523  * 4. Upconverts ack lock resource from CR to EX. This forces the BAST on other nodes
 524  *    and the other nodes read the message. The thread will wait here until all other
 525  *    nodes have released ack lock resource.
 526  * 5. Downconvert ack lockresource to CR
 527  */
 528 static int __sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg)
 529 {
 530         int error;
 531         int slot = cinfo->slot_number - 1;
 532
 533         cmsg->slot = cpu_to_le32(slot);
 534         /*get EX on Message*/
 535         error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_EX);
 536         if (error) {
 537                 pr_err("md-cluster: failed to get EX on MESSAGE (%d)\n", error);
 538                 goto failed_message;
 539         }
 540
 541         memcpy(cinfo->message_lockres->lksb.sb_lvbptr, (void *)cmsg,
 542                         sizeof(struct cluster_msg));
 543         /*down-convert EX to CR on Message*/
 544         error = dlm_lock_sync(cinfo->message_lockres, DLM_LOCK_CR);
 545         if (error) {
 546                 pr_err("md-cluster: failed to convert EX to CR on MESSAGE(%d)\n",
 547                                 error);
 548                 goto failed_message;
 549         }
 550
 551         /*up-convert CR to EX on Ack*/
 552         error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_EX);
 553         if (error) {
 554                 pr_err("md-cluster: failed to convert CR to EX on ACK(%d)\n",
 555                                 error);
 556                 goto failed_ack;
 557         }
 558
 559         /*down-convert EX to CR on Ack*/
 560         error = dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR);
 561         if (error) {
 562                 pr_err("md-cluster: failed to convert EX to CR on ACK(%d)\n",
 563                                 error);
 564                 goto failed_ack;
 565         }
 566
 567 failed_ack:
 568         dlm_unlock_sync(cinfo->message_lockres);
 569 failed_message:
 570         return error;
 571 }
 572
 573 static int sendmsg(struct md_cluster_info *cinfo, struct cluster_msg *cmsg)
 574 {
 575         int ret;
 576
 577         lock_comm(cinfo);
 578         ret = __sendmsg(cinfo, cmsg);
 579         unlock_comm(cinfo);
 580         return ret;
 581 }
 582
 583 static int gather_all_resync_info(struct mddev *mddev, int total_slots)
 584 {
 585         struct md_cluster_info *cinfo = mddev->cluster_info;
 586         int i, ret = 0;
 587         struct dlm_lock_resource *bm_lockres;
 588         struct suspend_info *s;
 589         char str[64];
 590
 591
 592         for (i = 0; i < total_slots; i++) {
 593                 memset(str, '\0', 64);
 594                 snprintf(str, 64, "bitmap%04d", i);
 595                 bm_lockres = lockres_init(mddev, str, NULL, 1);
 596                 if (!bm_lockres)
 597                         return -ENOMEM;
 598                 if (i == (cinfo->slot_number - 1))
 599                         continue;
 600
 601                 bm_lockres->flags |= DLM_LKF_NOQUEUE;
 602                 ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
 603                 if (ret == -EAGAIN) {
 604                         memset(bm_lockres->lksb.sb_lvbptr, '\0', LVB_SIZE);
 605                         s = read_resync_info(mddev, bm_lockres);
 606                         if (s) {
 607                                 pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n",
 608                                                 __func__, __LINE__,
 609                                                 (unsigned long long) s->lo,
 610                                                 (unsigned long long) s->hi, i);
 611                                 spin_lock_irq(&cinfo->suspend_lock);
 612                                 s->slot = i;
 613                                 list_add(&s->list, &cinfo->suspend_list);
 614                                 spin_unlock_irq(&cinfo->suspend_lock);
 615                         }
 616                         ret = 0;
 617                         lockres_free(bm_lockres);
 618                         continue;
 619                 }
 620                 if (ret)
 621                         goto out;
 622                 /* TODO: Read the disk bitmap sb and check if it needs recovery */
 623                 dlm_unlock_sync(bm_lockres);
 624                 lockres_free(bm_lockres);
 625         }
 626 out:
 627         return ret;
 628 }
 629
 630 static int join(struct mddev *mddev, int nodes)
 631 {
 632         struct md_cluster_info *cinfo;
 633         int ret, ops_rv;
 634         char str[64];
 635
 636         if (!try_module_get(THIS_MODULE))
 637                 return -ENOENT;
 638
 639         cinfo = kzalloc(sizeof(struct md_cluster_info), GFP_KERNEL);
 640         if (!cinfo)
 641                 return -ENOMEM;
 642
 643         init_completion(&cinfo->completion);
 644
 645         mutex_init(&cinfo->sb_mutex);
 646         mddev->cluster_info = cinfo;
 647
 648         memset(str, 0, 64);
 649         pretty_uuid(str, mddev->uuid);
 650         ret = dlm_new_lockspace(str, mddev->bitmap_info.cluster_name,
 651                                 DLM_LSFL_FS, LVB_SIZE,
 652                                 &md_ls_ops, mddev, &ops_rv, &cinfo->lockspace);
 653         if (ret)
 654                 goto err;
 655         wait_for_completion(&cinfo->completion);
 656         if (nodes < cinfo->slot_number) {
 657                 pr_err("md-cluster: Slot allotted(%d) is greater than available slots(%d).",
 658                         cinfo->slot_number, nodes);
 659                 ret = -ERANGE;
 660                 goto err;
 661         }
 662         cinfo->sb_lock = lockres_init(mddev, "cmd-super",
 663                                         NULL, 0);
 664         if (!cinfo->sb_lock) {
 665                 ret = -ENOMEM;
 666                 goto err;
 667         }
 668         /* Initiate the communication resources */
 669         ret = -ENOMEM;
 670         cinfo->recv_thread = md_register_thread(recv_daemon, mddev, "cluster_recv");
 671         if (!cinfo->recv_thread) {
 672                 pr_err("md-cluster: cannot allocate memory for recv_thread!\n");
 673                 goto err;
 674         }
 675         cinfo->message_lockres = lockres_init(mddev, "message", NULL, 1);
 676         if (!cinfo->message_lockres)
 677                 goto err;
 678         cinfo->token_lockres = lockres_init(mddev, "token", NULL, 0);
 679         if (!cinfo->token_lockres)
 680                 goto err;
 681         cinfo->ack_lockres = lockres_init(mddev, "ack", ack_bast, 0);
 682         if (!cinfo->ack_lockres)
 683                 goto err;
 684         cinfo->no_new_dev_lockres = lockres_init(mddev, "no-new-dev", NULL, 0);
 685         if (!cinfo->no_new_dev_lockres)
 686                 goto err;
 687
 688         /* get sync CR lock on ACK. */
 689         if (dlm_lock_sync(cinfo->ack_lockres, DLM_LOCK_CR))
 690                 pr_err("md-cluster: failed to get a sync CR lock on ACK!(%d)\n",
 691                                 ret);
 692         /* get sync CR lock on no-new-dev. */
 693         if (dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR))
 694                 pr_err("md-cluster: failed to get a sync CR lock on no-new-dev!(%d)\n", ret);
 695
 696
 697         pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number);
 698         snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1);
 699         cinfo->bitmap_lockres = lockres_init(mddev, str, NULL, 1);
 700         if (!cinfo->bitmap_lockres)
 701                 goto err;
 702         if (dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW)) {
 703                 pr_err("Failed to get bitmap lock\n");
 704                 ret = -EINVAL;
 705                 goto err;
 706         }
 707
 708         INIT_LIST_HEAD(&cinfo->suspend_list);
 709         spin_lock_init(&cinfo->suspend_lock);
 710
 711         ret = gather_all_resync_info(mddev, nodes);
 712         if (ret)
 713                 goto err;
 714
 715         return 0;
 716 err:
 717         lockres_free(cinfo->message_lockres);
 718         lockres_free(cinfo->token_lockres);
 719         lockres_free(cinfo->ack_lockres);
 720         lockres_free(cinfo->no_new_dev_lockres);
 721         lockres_free(cinfo->bitmap_lockres);
 722         lockres_free(cinfo->sb_lock);
 723         if (cinfo->lockspace)
 724                 dlm_release_lockspace(cinfo->lockspace, 2);
 725         mddev->cluster_info = NULL;
 726         kfree(cinfo);
 727         module_put(THIS_MODULE);
 728         return ret;
 729 }
 730
 731 static int leave(struct mddev *mddev)
 732 {
 733         struct md_cluster_info *cinfo = mddev->cluster_info;
 734
 735         if (!cinfo)
 736                 return 0;
 737         md_unregister_thread(&cinfo->recovery_thread);
 738         md_unregister_thread(&cinfo->recv_thread);
 739         lockres_free(cinfo->message_lockres);
 740         lockres_free(cinfo->token_lockres);
 741         lockres_free(cinfo->ack_lockres);
 742         lockres_free(cinfo->no_new_dev_lockres);
 743         lockres_free(cinfo->sb_lock);
 744         lockres_free(cinfo->bitmap_lockres);
 745         dlm_release_lockspace(cinfo->lockspace, 2);
 746         return 0;
 747 }
 748
 749 /* slot_number(): Returns the MD slot number to use
 750  * DLM starts the slot numbers from 1, wheras cluster-md
 751  * wants the number to be from zero, so we deduct one
 752  */
 753 static int slot_number(struct mddev *mddev)
 754 {
 755         struct md_cluster_info *cinfo = mddev->cluster_info;
 756
 757         return cinfo->slot_number - 1;
 758 }
 759
 760 static void resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi)
 761 {
 762         struct md_cluster_info *cinfo = mddev->cluster_info;
 763
 764         add_resync_info(mddev, cinfo->bitmap_lockres, lo, hi);
 765         /* Re-acquire the lock to refresh LVB */
 766         dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW);
 767 }
 768
 769 static int metadata_update_start(struct mddev *mddev)
 770 {
 771         return lock_comm(mddev->cluster_info);
 772 }
 773
 774 static int metadata_update_finish(struct mddev *mddev)
 775 {
 776         struct md_cluster_info *cinfo = mddev->cluster_info;
 777         struct cluster_msg cmsg;
 778         int ret;
 779
 780         memset(&cmsg, 0, sizeof(cmsg));
 781         cmsg.type = cpu_to_le32(METADATA_UPDATED);
 782         ret = __sendmsg(cinfo, &cmsg);
 783         unlock_comm(cinfo);
 784         return ret;
 785 }
 786
 787 static int metadata_update_cancel(struct mddev *mddev)
 788 {
 789         struct md_cluster_info *cinfo = mddev->cluster_info;
 790
 791         return dlm_unlock_sync(cinfo->token_lockres);
 792 }
 793
 794 static int resync_send(struct mddev *mddev, enum msg_type type,
 795                 sector_t lo, sector_t hi)
 796 {
 797         struct md_cluster_info *cinfo = mddev->cluster_info;
 798         struct cluster_msg cmsg;
 799         int slot = cinfo->slot_number - 1;
 800
 801         pr_info("%s:%d lo: %llu hi: %llu\n", __func__, __LINE__,
 802                         (unsigned long long)lo,
 803                         (unsigned long long)hi);
 804         resync_info_update(mddev, lo, hi);
 805         cmsg.type = cpu_to_le32(type);
 806         cmsg.slot = cpu_to_le32(slot);
 807         cmsg.low = cpu_to_le64(lo);
 808         cmsg.high = cpu_to_le64(hi);
 809         return sendmsg(cinfo, &cmsg);
 810 }
 811
 812 static int resync_start(struct mddev *mddev, sector_t lo, sector_t hi)
 813 {
 814         pr_info("%s:%d\n", __func__, __LINE__);
 815         return resync_send(mddev, RESYNCING, lo, hi);
 816 }
 817
 818 static void resync_finish(struct mddev *mddev)
 819 {
 820         pr_info("%s:%d\n", __func__, __LINE__);
 821         resync_send(mddev, RESYNCING, 0, 0);
 822 }
 823
 824 static int area_resyncing(struct mddev *mddev, int direction,
 825                 sector_t lo, sector_t hi)
 826 {
 827         struct md_cluster_info *cinfo = mddev->cluster_info;
 828         int ret = 0;
 829         struct suspend_info *s;
 830
 831         if ((direction == READ) &&
 832                 test_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state))
 833                 return 1;
 834
 835         spin_lock_irq(&cinfo->suspend_lock);
 836         if (list_empty(&cinfo->suspend_list))
 837                 goto out;
 838         list_for_each_entry(s, &cinfo->suspend_list, list)
 839                 if (hi > s->lo && lo < s->hi) {
 840                         ret = 1;
 841                         break;
 842                 }
 843 out:
 844         spin_unlock_irq(&cinfo->suspend_lock);
 845         return ret;
 846 }
 847
 848 static int add_new_disk_start(struct mddev *mddev, struct md_rdev *rdev)
 849 {
 850         struct md_cluster_info *cinfo = mddev->cluster_info;
 851         struct cluster_msg cmsg;
 852         int ret = 0;
 853         struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
 854         char *uuid = sb->device_uuid;
 855
 856         memset(&cmsg, 0, sizeof(cmsg));
 857         cmsg.type = cpu_to_le32(NEWDISK);
 858         memcpy(cmsg.uuid, uuid, 16);
 859         cmsg.raid_slot = rdev->desc_nr;
 860         lock_comm(cinfo);
 861         ret = __sendmsg(cinfo, &cmsg);
 862         if (ret)
 863                 return ret;
 864         cinfo->no_new_dev_lockres->flags |= DLM_LKF_NOQUEUE;
 865         ret = dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_EX);
 866         cinfo->no_new_dev_lockres->flags &= ~DLM_LKF_NOQUEUE;
 867         /* Some node does not "see" the device */
 868         if (ret == -EAGAIN)
 869                 ret = -ENOENT;
 870         else
 871                 dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR);
 872         return ret;
 873 }
 874
 875 static int add_new_disk_finish(struct mddev *mddev)
 876 {
 877         struct cluster_msg cmsg;
 878         struct md_cluster_info *cinfo = mddev->cluster_info;
 879         int ret;
 880         /* Write sb and inform others */
 881         md_update_sb(mddev, 1);
 882         cmsg.type = METADATA_UPDATED;
 883         ret = __sendmsg(cinfo, &cmsg);
 884         unlock_comm(cinfo);
 885         return ret;
 886 }
 887
 888 static int new_disk_ack(struct mddev *mddev, bool ack)
 889 {
 890         struct md_cluster_info *cinfo = mddev->cluster_info;
 891
 892         if (!test_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state)) {
 893                 pr_warn("md-cluster(%s): Spurious cluster confirmation\n", mdname(mddev));
 894                 return -EINVAL;
 895         }
 896
 897         if (ack)
 898                 dlm_unlock_sync(cinfo->no_new_dev_lockres);
 899         complete(&cinfo->newdisk_completion);
 900         return 0;
 901 }
 902
 903 static int remove_disk(struct mddev *mddev, struct md_rdev *rdev)
 904 {
 905         struct cluster_msg cmsg;
 906         struct md_cluster_info *cinfo = mddev->cluster_info;
 907         cmsg.type = REMOVE;
 908         cmsg.raid_slot = rdev->desc_nr;
 909         return __sendmsg(cinfo, &cmsg);
 910 }
 911
 912 static int gather_bitmaps(struct md_rdev *rdev)
 913 {
 914         int sn, err;
 915         sector_t lo, hi;
 916         struct cluster_msg cmsg;
 917         struct mddev *mddev = rdev->mddev;
 918         struct md_cluster_info *cinfo = mddev->cluster_info;
 919
 920         cmsg.type = RE_ADD;
 921         cmsg.raid_slot = rdev->desc_nr;
 922         err = sendmsg(cinfo, &cmsg);
 923         if (err)
 924                 goto out;
 925
 926         for (sn = 0; sn < mddev->bitmap_info.nodes; sn++) {
 927                 if (sn == (cinfo->slot_number - 1))
 928                         continue;
 929                 err = bitmap_copy_from_slot(mddev, sn, &lo, &hi, false);
 930                 if (err) {
 931                         pr_warn("md-cluster: Could not gather bitmaps from slot %d", sn);
 932                         goto out;
 933                 }
 934                 if ((hi > 0) && (lo < mddev->recovery_cp))
 935                         mddev->recovery_cp = lo;
 936         }
 937 out:
 938         return err;
 939 }
 940
 941 static struct md_cluster_operations cluster_ops = {
 942         .join   = join,
 943         .leave  = leave,
 944         .slot_number = slot_number,
 945         .resync_info_update = resync_info_update,
 946         .resync_start = resync_start,
 947         .resync_finish = resync_finish,
 948         .metadata_update_start = metadata_update_start,
 949         .metadata_update_finish = metadata_update_finish,
 950         .metadata_update_cancel = metadata_update_cancel,
 951         .area_resyncing = area_resyncing,
 952         .add_new_disk_start = add_new_disk_start,
 953         .add_new_disk_finish = add_new_disk_finish,
 954         .new_disk_ack = new_disk_ack,
 955         .remove_disk = remove_disk,
 956         .gather_bitmaps = gather_bitmaps,
 957 };
 958
 959 static int __init cluster_init(void)
 960 {
 961         pr_warn("md-cluster: EXPERIMENTAL. Use with caution\n");
 962         pr_info("Registering Cluster MD functions\n");
 963         register_md_cluster_operations(&cluster_ops, THIS_MODULE);
 964         return 0;
 965 }
 966
 967 static void cluster_exit(void)
 968 {
 969         unregister_md_cluster_operations();
 970 }
 971
 972 module_init(cluster_init);
 973 module_exit(cluster_exit);
 974 MODULE_LICENSE("GPL");
 975 MODULE_DESCRIPTION("Clustering support for MD");