drivers/block/drbd/drbd_nl.c

   1 /*
   2    drbd_nl.c
   3
   4    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
   5
   6    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
   7    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
   8    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
   9
  10    drbd is free software; you can redistribute it and/or modify
  11    it under the terms of the GNU General Public License as published by
  12    the Free Software Foundation; either version 2, or (at your option)
  13    any later version.
  14
  15    drbd is distributed in the hope that it will be useful,
  16    but WITHOUT ANY WARRANTY; without even the implied warranty of
  17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18    GNU General Public License for more details.
  19
  20    You should have received a copy of the GNU General Public License
  21    along with drbd; see the file COPYING.  If not, write to
  22    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  23
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/drbd.h>
  28 #include <linux/in.h>
  29 #include <linux/fs.h>
  30 #include <linux/file.h>
  31 #include <linux/slab.h>
  32 #include <linux/connector.h>
  33 #include <linux/blkpg.h>
  34 #include <linux/cpumask.h>
  35 #include "drbd_int.h"
  36 #include "drbd_req.h"
  37 #include "drbd_wrappers.h"
  38 #include <asm/unaligned.h>
  39 #include <linux/drbd_tag_magic.h>
  40 #include <linux/drbd_limits.h>
  41 #include <linux/compiler.h>
  42 #include <linux/kthread.h>
  43
  44 static unsigned short *tl_add_blob(unsigned short *, enum drbd_tags, const void *, int);
  45 static unsigned short *tl_add_str(unsigned short *, enum drbd_tags, const char *);
  46 static unsigned short *tl_add_int(unsigned short *, enum drbd_tags, const void *);
  47
  48 /* see get_sb_bdev and bd_claim */
  49 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
  50
  51 /* Generate the tag_list to struct functions */
  52 #define NL_PACKET(name, number, fields) \
  53 static int name ## _from_tags(struct drbd_conf *mdev, \
  54         unsigned short *tags, struct name *arg) __attribute__ ((unused)); \
  55 static int name ## _from_tags(struct drbd_conf *mdev, \
  56         unsigned short *tags, struct name *arg) \
  57 { \
  58         int tag; \
  59         int dlen; \
  60         \
  61         while ((tag = get_unaligned(tags++)) != TT_END) {       \
  62                 dlen = get_unaligned(tags++);                   \
  63                 switch (tag_number(tag)) { \
  64                 fields \
  65                 default: \
  66                         if (tag & T_MANDATORY) { \
  67                                 dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \
  68                                 return 0; \
  69                         } \
  70                 } \
  71                 tags = (unsigned short *)((char *)tags + dlen); \
  72         } \
  73         return 1; \
  74 }
  75 #define NL_INTEGER(pn, pr, member) \
  76         case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \
  77                 arg->member = get_unaligned((int *)(tags));     \
  78                 break;
  79 #define NL_INT64(pn, pr, member) \
  80         case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \
  81                 arg->member = get_unaligned((u64 *)(tags));     \
  82                 break;
  83 #define NL_BIT(pn, pr, member) \
  84         case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \
  85                 arg->member = *(char *)(tags) ? 1 : 0; \
  86                 break;
  87 #define NL_STRING(pn, pr, member, len) \
  88         case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \
  89                 if (dlen > len) { \
  90                         dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \
  91                                 #member, dlen, (unsigned int)len); \
  92                         return 0; \
  93                 } \
  94                  arg->member ## _len = dlen; \
  95                  memcpy(arg->member, tags, min_t(size_t, dlen, len)); \
  96                  break;
  97 #include "linux/drbd_nl.h"
  98
  99 /* Generate the struct to tag_list functions */
 100 #define NL_PACKET(name, number, fields) \
 101 static unsigned short* \
 102 name ## _to_tags(struct drbd_conf *mdev, \
 103         struct name *arg, unsigned short *tags) __attribute__ ((unused)); \
 104 static unsigned short* \
 105 name ## _to_tags(struct drbd_conf *mdev, \
 106         struct name *arg, unsigned short *tags) \
 107 { \
 108         fields \
 109         return tags; \
 110 }
 111
 112 #define NL_INTEGER(pn, pr, member) \
 113         put_unaligned(pn | pr | TT_INTEGER, tags++);    \
 114         put_unaligned(sizeof(int), tags++);             \
 115         put_unaligned(arg->member, (int *)tags);        \
 116         tags = (unsigned short *)((char *)tags+sizeof(int));
 117 #define NL_INT64(pn, pr, member) \
 118         put_unaligned(pn | pr | TT_INT64, tags++);      \
 119         put_unaligned(sizeof(u64), tags++);             \
 120         put_unaligned(arg->member, (u64 *)tags);        \
 121         tags = (unsigned short *)((char *)tags+sizeof(u64));
 122 #define NL_BIT(pn, pr, member) \
 123         put_unaligned(pn | pr | TT_BIT, tags++);        \
 124         put_unaligned(sizeof(char), tags++);            \
 125         *(char *)tags = arg->member; \
 126         tags = (unsigned short *)((char *)tags+sizeof(char));
 127 #define NL_STRING(pn, pr, member, len) \
 128         put_unaligned(pn | pr | TT_STRING, tags++);     \
 129         put_unaligned(arg->member ## _len, tags++);     \
 130         memcpy(tags, arg->member, arg->member ## _len); \
 131         tags = (unsigned short *)((char *)tags + arg->member ## _len);
 132 #include "linux/drbd_nl.h"
 133
 134 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name);
 135 void drbd_nl_send_reply(struct cn_msg *, int);
 136
 137 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
 138 {
 139         char *envp[] = { "HOME=/",
 140                         "TERM=linux",
 141                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 142                         NULL, /* Will be set to address family */
 143                         NULL, /* Will be set to address */
 144                         NULL };
 145
 146         char mb[12], af[20], ad[60], *afs;
 147         char *argv[] = {usermode_helper, cmd, mb, NULL };
 148         int ret;
 149
 150         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 151
 152         if (get_net_conf(mdev)) {
 153                 switch (((struct sockaddr *)mdev->net_conf->peer_addr)->sa_family) {
 154                 case AF_INET6:
 155                         afs = "ipv6";
 156                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6",
 157                                  &((struct sockaddr_in6 *)mdev->net_conf->peer_addr)->sin6_addr);
 158                         break;
 159                 case AF_INET:
 160                         afs = "ipv4";
 161                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 162                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
 163                         break;
 164                 default:
 165                         afs = "ssocks";
 166                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 167                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
 168                 }
 169                 snprintf(af, 20, "DRBD_PEER_AF=%s", afs);
 170                 envp[3]=af;
 171                 envp[4]=ad;
 172                 put_net_conf(mdev);
 173         }
 174
 175         /* The helper may take some time.
 176          * write out any unsynced meta data changes now */
 177         drbd_md_sync(mdev);
 178
 179         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 180
 181         drbd_bcast_ev_helper(mdev, cmd);
 182         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 183         if (ret)
 184                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 185                                 usermode_helper, cmd, mb,
 186                                 (ret >> 8) & 0xff, ret);
 187         else
 188                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 189                                 usermode_helper, cmd, mb,
 190                                 (ret >> 8) & 0xff, ret);
 191
 192         if (ret < 0) /* Ignore any ERRNOs we got. */
 193                 ret = 0;
 194
 195         return ret;
 196 }
 197
 198 enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
 199 {
 200         char *ex_to_string;
 201         int r;
 202         enum drbd_disk_state nps;
 203         enum drbd_fencing_p fp;
 204
 205         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 206
 207         if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 208                 fp = mdev->ldev->dc.fencing;
 209                 put_ldev(mdev);
 210         } else {
 211                 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n");
 212                 nps = mdev->state.pdsk;
 213                 goto out;
 214         }
 215
 216         r = drbd_khelper(mdev, "fence-peer");
 217
 218         switch ((r>>8) & 0xff) {
 219         case 3: /* peer is inconsistent */
 220                 ex_to_string = "peer is inconsistent or worse";
 221                 nps = D_INCONSISTENT;
 222                 break;
 223         case 4: /* peer got outdated, or was already outdated */
 224                 ex_to_string = "peer was fenced";
 225                 nps = D_OUTDATED;
 226                 break;
 227         case 5: /* peer was down */
 228                 if (mdev->state.disk == D_UP_TO_DATE) {
 229                         /* we will(have) create(d) a new UUID anyways... */
 230                         ex_to_string = "peer is unreachable, assumed to be dead";
 231                         nps = D_OUTDATED;
 232                 } else {
 233                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 234                         nps = mdev->state.pdsk;
 235                 }
 236                 break;
 237         case 6: /* Peer is primary, voluntarily outdate myself.
 238                  * This is useful when an unconnected R_SECONDARY is asked to
 239                  * become R_PRIMARY, but finds the other peer being active. */
 240                 ex_to_string = "peer is active";
 241                 dev_warn(DEV, "Peer is primary, outdating myself.\n");
 242                 nps = D_UNKNOWN;
 243                 _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE);
 244                 break;
 245         case 7:
 246                 if (fp != FP_STONITH)
 247                         dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n");
 248                 ex_to_string = "peer was stonithed";
 249                 nps = D_OUTDATED;
 250                 break;
 251         default:
 252                 /* The script is broken ... */
 253                 nps = D_UNKNOWN;
 254                 dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 255                 return nps;
 256         }
 257
 258         dev_info(DEV, "fence-peer helper returned %d (%s)\n",
 259                         (r>>8) & 0xff, ex_to_string);
 260
 261 out:
 262         if (mdev->state.susp_fen && nps >= D_UNKNOWN) {
 263                 /* The handler was not successful... unfreeze here, the
 264                    state engine can not unfreeze... */
 265                 _drbd_request_state(mdev, NS(susp_fen, 0), CS_VERBOSE);
 266         }
 267
 268         return nps;
 269 }
 270
 271 static int _try_outdate_peer_async(void *data)
 272 {
 273         struct drbd_conf *mdev = (struct drbd_conf *)data;
 274         enum drbd_disk_state nps;
 275
 276         nps = drbd_try_outdate_peer(mdev);
 277         drbd_request_state(mdev, NS(pdsk, nps));
 278
 279         return 0;
 280 }
 281
 282 void drbd_try_outdate_peer_async(struct drbd_conf *mdev)
 283 {
 284         struct task_struct *opa;
 285
 286         opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev));
 287         if (IS_ERR(opa))
 288                 dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n");
 289 }
 290
 291 int drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 292 {
 293         const int max_tries = 4;
 294         int r = 0;
 295         int try = 0;
 296         int forced = 0;
 297         union drbd_state mask, val;
 298         enum drbd_disk_state nps;
 299
 300         if (new_role == R_PRIMARY)
 301                 request_ping(mdev); /* Detect a dead peer ASAP */
 302
 303         mutex_lock(&mdev->state_mutex);
 304
 305         mask.i = 0; mask.role = R_MASK;
 306         val.i  = 0; val.role  = new_role;
 307
 308         while (try++ < max_tries) {
 309                 r = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 310
 311                 /* in case we first succeeded to outdate,
 312                  * but now suddenly could establish a connection */
 313                 if (r == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 314                         val.pdsk = 0;
 315                         mask.pdsk = 0;
 316                         continue;
 317                 }
 318
 319                 if (r == SS_NO_UP_TO_DATE_DISK && force &&
 320                     (mdev->state.disk < D_UP_TO_DATE &&
 321                      mdev->state.disk >= D_INCONSISTENT)) {
 322                         mask.disk = D_MASK;
 323                         val.disk  = D_UP_TO_DATE;
 324                         forced = 1;
 325                         continue;
 326                 }
 327
 328                 if (r == SS_NO_UP_TO_DATE_DISK &&
 329                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 330                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 331                         nps = drbd_try_outdate_peer(mdev);
 332
 333                         if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
 334                                 val.disk = D_UP_TO_DATE;
 335                                 mask.disk = D_MASK;
 336                         }
 337
 338                         val.pdsk = nps;
 339                         mask.pdsk = D_MASK;
 340
 341                         continue;
 342                 }
 343
 344                 if (r == SS_NOTHING_TO_DO)
 345                         goto fail;
 346                 if (r == SS_PRIMARY_NOP && mask.pdsk == 0) {
 347                         nps = drbd_try_outdate_peer(mdev);
 348
 349                         if (force && nps > D_OUTDATED) {
 350                                 dev_warn(DEV, "Forced into split brain situation!\n");
 351                                 nps = D_OUTDATED;
 352                         }
 353
 354                         mask.pdsk = D_MASK;
 355                         val.pdsk  = nps;
 356
 357                         continue;
 358                 }
 359                 if (r == SS_TWO_PRIMARIES) {
 360                         /* Maybe the peer is detected as dead very soon...
 361                            retry at most once more in this case. */
 362                         __set_current_state(TASK_INTERRUPTIBLE);
 363                         schedule_timeout((mdev->net_conf->ping_timeo+1)*HZ/10);
 364                         if (try < max_tries)
 365                                 try = max_tries - 1;
 366                         continue;
 367                 }
 368                 if (r < SS_SUCCESS) {
 369                         r = _drbd_request_state(mdev, mask, val,
 370                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 371                         if (r < SS_SUCCESS)
 372                                 goto fail;
 373                 }
 374                 break;
 375         }
 376
 377         if (r < SS_SUCCESS)
 378                 goto fail;
 379
 380         if (forced)
 381                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 382
 383         /* Wait until nothing is on the fly :) */
 384         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 385
 386         if (new_role == R_SECONDARY) {
 387                 set_disk_ro(mdev->vdisk, TRUE);
 388                 if (get_ldev(mdev)) {
 389                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 390                         put_ldev(mdev);
 391                 }
 392         } else {
 393                 if (get_net_conf(mdev)) {
 394                         mdev->net_conf->want_lose = 0;
 395                         put_net_conf(mdev);
 396                 }
 397                 set_disk_ro(mdev->vdisk, FALSE);
 398                 if (get_ldev(mdev)) {
 399                         if (((mdev->state.conn < C_CONNECTED ||
 400                                mdev->state.pdsk <= D_FAILED)
 401                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 402                                 drbd_uuid_new_current(mdev);
 403
 404                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 405                         put_ldev(mdev);
 406                 }
 407         }
 408
 409         if ((new_role == R_SECONDARY) && get_ldev(mdev)) {
 410                 drbd_al_to_on_disk_bm(mdev);
 411                 put_ldev(mdev);
 412         }
 413
 414         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 415                 /* if this was forced, we should consider sync */
 416                 if (forced)
 417                         drbd_send_uuids(mdev);
 418                 drbd_send_state(mdev);
 419         }
 420
 421         drbd_md_sync(mdev);
 422
 423         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 424  fail:
 425         mutex_unlock(&mdev->state_mutex);
 426         return r;
 427 }
 428
 429 static struct drbd_conf *ensure_mdev(int minor, int create)
 430 {
 431         struct drbd_conf *mdev;
 432
 433         if (minor >= minor_count)
 434                 return NULL;
 435
 436         mdev = minor_to_mdev(minor);
 437
 438         if (!mdev && create) {
 439                 struct gendisk *disk = NULL;
 440                 mdev = drbd_new_device(minor);
 441
 442                 spin_lock_irq(&drbd_pp_lock);
 443                 if (minor_table[minor] == NULL) {
 444                         minor_table[minor] = mdev;
 445                         disk = mdev->vdisk;
 446                         mdev = NULL;
 447                 } /* else: we lost the race */
 448                 spin_unlock_irq(&drbd_pp_lock);
 449
 450                 if (disk) /* we won the race above */
 451                         /* in case we ever add a drbd_delete_device(),
 452                          * don't forget the del_gendisk! */
 453                         add_disk(disk);
 454                 else /* we lost the race above */
 455                         drbd_free_mdev(mdev);
 456
 457                 mdev = minor_to_mdev(minor);
 458         }
 459
 460         return mdev;
 461 }
 462
 463 static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 464                            struct drbd_nl_cfg_reply *reply)
 465 {
 466         struct primary primary_args;
 467
 468         memset(&primary_args, 0, sizeof(struct primary));
 469         if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) {
 470                 reply->ret_code = ERR_MANDATORY_TAG;
 471                 return 0;
 472         }
 473
 474         reply->ret_code =
 475                 drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force);
 476
 477         return 0;
 478 }
 479
 480 static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 481                              struct drbd_nl_cfg_reply *reply)
 482 {
 483         reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0);
 484
 485         return 0;
 486 }
 487
 488 /* initializes the md.*_offset members, so we are able to find
 489  * the on disk meta data */
 490 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 491                                        struct drbd_backing_dev *bdev)
 492 {
 493         sector_t md_size_sect = 0;
 494         switch (bdev->dc.meta_dev_idx) {
 495         default:
 496                 /* v07 style fixed size indexed meta data */
 497                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 498                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 499                 bdev->md.al_offset = MD_AL_OFFSET;
 500                 bdev->md.bm_offset = MD_BM_OFFSET;
 501                 break;
 502         case DRBD_MD_INDEX_FLEX_EXT:
 503                 /* just occupy the full device; unit: sectors */
 504                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 505                 bdev->md.md_offset = 0;
 506                 bdev->md.al_offset = MD_AL_OFFSET;
 507                 bdev->md.bm_offset = MD_BM_OFFSET;
 508                 break;
 509         case DRBD_MD_INDEX_INTERNAL:
 510         case DRBD_MD_INDEX_FLEX_INT:
 511                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 512                 /* al size is still fixed */
 513                 bdev->md.al_offset = -MD_AL_MAX_SIZE;
 514                 /* we need (slightly less than) ~ this much bitmap sectors: */
 515                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 516                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 517                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 518                 md_size_sect = ALIGN(md_size_sect, 8);
 519
 520                 /* plus the "drbd meta data super block",
 521                  * and the activity log; */
 522                 md_size_sect += MD_BM_OFFSET;
 523
 524                 bdev->md.md_size_sect = md_size_sect;
 525                 /* bitmap offset is adjusted by 'super' block size */
 526                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 527                 break;
 528         }
 529 }
 530
 531 char *ppsize(char *buf, unsigned long long size)
 532 {
 533         /* Needs 9 bytes at max. */
 534         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 535         int base = 0;
 536         while (size >= 10000) {
 537                 /* shift + round */
 538                 size = (size >> 10) + !!(size & (1<<9));
 539                 base++;
 540         }
 541         sprintf(buf, "%lu %cB", (long)size, units[base]);
 542
 543         return buf;
 544 }
 545
 546 /* there is still a theoretical deadlock when called from receiver
 547  * on an D_INCONSISTENT R_PRIMARY:
 548  *  remote READ does inc_ap_bio, receiver would need to receive answer
 549  *  packet from remote to dec_ap_bio again.
 550  *  receiver receive_sizes(), comes here,
 551  *  waits for ap_bio_cnt == 0. -> deadlock.
 552  * but this cannot happen, actually, because:
 553  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 554  *  (not connected, or bad/no disk on peer):
 555  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 556  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 557  *  peer may not initiate a resize.
 558  */
 559 void drbd_suspend_io(struct drbd_conf *mdev)
 560 {
 561         set_bit(SUSPEND_IO, &mdev->flags);
 562         if (is_susp(mdev->state))
 563                 return;
 564         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 565 }
 566
 567 void drbd_resume_io(struct drbd_conf *mdev)
 568 {
 569         clear_bit(SUSPEND_IO, &mdev->flags);
 570         wake_up(&mdev->misc_wait);
 571 }
 572
 573 /**
 574  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 575  * @mdev:       DRBD device.
 576  *
 577  * Returns 0 on success, negative return values indicate errors.
 578  * You should call drbd_md_sync() after calling this function.
 579  */
 580 enum determine_dev_size drbd_determin_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 581 {
 582         sector_t prev_first_sect, prev_size; /* previous meta location */
 583         sector_t la_size;
 584         sector_t size;
 585         char ppb[10];
 586
 587         int md_moved, la_size_changed;
 588         enum determine_dev_size rv = unchanged;
 589
 590         /* race:
 591          * application request passes inc_ap_bio,
 592          * but then cannot get an AL-reference.
 593          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 594          *
 595          * to avoid that:
 596          * Suspend IO right here.
 597          * still lock the act_log to not trigger ASSERTs there.
 598          */
 599         drbd_suspend_io(mdev);
 600
 601         /* no wait necessary anymore, actually we could assert that */
 602         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 603
 604         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 605         prev_size = mdev->ldev->md.md_size_sect;
 606         la_size = mdev->ldev->md.la_size_sect;
 607
 608         /* TODO: should only be some assert here, not (re)init... */
 609         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 610
 611         size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
 612
 613         if (drbd_get_capacity(mdev->this_bdev) != size ||
 614             drbd_bm_capacity(mdev) != size) {
 615                 int err;
 616                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 617                 if (unlikely(err)) {
 618                         /* currently there is only one error: ENOMEM! */
 619                         size = drbd_bm_capacity(mdev)>>1;
 620                         if (size == 0) {
 621                                 dev_err(DEV, "OUT OF MEMORY! "
 622                                     "Could not allocate bitmap!\n");
 623                         } else {
 624                                 dev_err(DEV, "BM resizing failed. "
 625                                     "Leaving size unchanged at size = %lu KB\n",
 626                                     (unsigned long)size);
 627                         }
 628                         rv = dev_size_error;
 629                 }
 630                 /* racy, see comments above. */
 631                 drbd_set_my_capacity(mdev, size);
 632                 mdev->ldev->md.la_size_sect = size;
 633                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 634                      (unsigned long long)size>>1);
 635         }
 636         if (rv == dev_size_error)
 637                 goto out;
 638
 639         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 640
 641         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 642                 || prev_size       != mdev->ldev->md.md_size_sect;
 643
 644         if (la_size_changed || md_moved) {
 645                 drbd_al_shrink(mdev); /* All extents inactive. */
 646                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 647                          la_size_changed && md_moved ? "size changed and md moved" :
 648                          la_size_changed ? "size changed" : "md moved");
 649                 rv = drbd_bitmap_io(mdev, &drbd_bm_write, "size changed"); /* does drbd_resume_io() ! */
 650                 drbd_md_mark_dirty(mdev);
 651         }
 652
 653         if (size > la_size)
 654                 rv = grew;
 655         if (size < la_size)
 656                 rv = shrunk;
 657 out:
 658         lc_unlock(mdev->act_log);
 659         wake_up(&mdev->al_wait);
 660         drbd_resume_io(mdev);
 661
 662         return rv;
 663 }
 664
 665 sector_t
 666 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
 667 {
 668         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 669         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 670         sector_t m_size; /* my size */
 671         sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
 672         sector_t size = 0;
 673
 674         m_size = drbd_get_max_capacity(bdev);
 675
 676         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 677                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 678                 p_size = m_size;
 679         }
 680
 681         if (p_size && m_size) {
 682                 size = min_t(sector_t, p_size, m_size);
 683         } else {
 684                 if (la_size) {
 685                         size = la_size;
 686                         if (m_size && m_size < size)
 687                                 size = m_size;
 688                         if (p_size && p_size < size)
 689                                 size = p_size;
 690                 } else {
 691                         if (m_size)
 692                                 size = m_size;
 693                         if (p_size)
 694                                 size = p_size;
 695                 }
 696         }
 697
 698         if (size == 0)
 699                 dev_err(DEV, "Both nodes diskless!\n");
 700
 701         if (u_size) {
 702                 if (u_size > size)
 703                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 704                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 705                 else
 706                         size = u_size;
 707         }
 708
 709         return size;
 710 }
 711
 712 /**
 713  * drbd_check_al_size() - Ensures that the AL is of the right size
 714  * @mdev:       DRBD device.
 715  *
 716  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 717  * failed, and 0 on success. You should call drbd_md_sync() after you called
 718  * this function.
 719  */
 720 static int drbd_check_al_size(struct drbd_conf *mdev)
 721 {
 722         struct lru_cache *n, *t;
 723         struct lc_element *e;
 724         unsigned int in_use;
 725         int i;
 726
 727         ERR_IF(mdev->sync_conf.al_extents < 7)
 728                 mdev->sync_conf.al_extents = 127;
 729
 730         if (mdev->act_log &&
 731             mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
 732                 return 0;
 733
 734         in_use = 0;
 735         t = mdev->act_log;
 736         n = lc_create("act_log", drbd_al_ext_cache,
 737                 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
 738
 739         if (n == NULL) {
 740                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 741                 return -ENOMEM;
 742         }
 743         spin_lock_irq(&mdev->al_lock);
 744         if (t) {
 745                 for (i = 0; i < t->nr_elements; i++) {
 746                         e = lc_element_by_index(t, i);
 747                         if (e->refcnt)
 748                                 dev_err(DEV, "refcnt(%d)==%d\n",
 749                                     e->lc_number, e->refcnt);
 750                         in_use += e->refcnt;
 751                 }
 752         }
 753         if (!in_use)
 754                 mdev->act_log = n;
 755         spin_unlock_irq(&mdev->al_lock);
 756         if (in_use) {
 757                 dev_err(DEV, "Activity log still in use!\n");
 758                 lc_destroy(n);
 759                 return -EBUSY;
 760         } else {
 761                 if (t)
 762                         lc_destroy(t);
 763         }
 764         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 765         return 0;
 766 }
 767
 768 void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size) __must_hold(local)
 769 {
 770         struct request_queue * const q = mdev->rq_queue;
 771         struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 772         int max_segments = mdev->ldev->dc.max_bio_bvecs;
 773         int max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
 774
 775         blk_queue_logical_block_size(q, 512);
 776         blk_queue_max_hw_sectors(q, max_hw_sectors);
 777         /* This is the workaround for "bio would need to, but cannot, be split" */
 778         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
 779         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
 780         blk_queue_stack_limits(q, b);
 781
 782         dev_info(DEV, "max BIO size = %u\n", queue_max_hw_sectors(q) << 9);
 783
 784         if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
 785                 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
 786                      q->backing_dev_info.ra_pages,
 787                      b->backing_dev_info.ra_pages);
 788                 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
 789         }
 790 }
 791
 792 /* serialize deconfig (worker exiting, doing cleanup)
 793  * and reconfig (drbdsetup disk, drbdsetup net)
 794  *
 795  * Wait for a potentially exiting worker, then restart it,
 796  * or start a new one.  Flush any pending work, there may still be an
 797  * after_state_change queued.
 798  */
 799 static void drbd_reconfig_start(struct drbd_conf *mdev)
 800 {
 801         wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
 802         wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
 803         drbd_thread_start(&mdev->worker);
 804         drbd_flush_workqueue(mdev);
 805 }
 806
 807 /* if still unconfigured, stops worker again.
 808  * if configured now, clears CONFIG_PENDING.
 809  * wakes potential waiters */
 810 static void drbd_reconfig_done(struct drbd_conf *mdev)
 811 {
 812         spin_lock_irq(&mdev->req_lock);
 813         if (mdev->state.disk == D_DISKLESS &&
 814             mdev->state.conn == C_STANDALONE &&
 815             mdev->state.role == R_SECONDARY) {
 816                 set_bit(DEVICE_DYING, &mdev->flags);
 817                 drbd_thread_stop_nowait(&mdev->worker);
 818         } else
 819                 clear_bit(CONFIG_PENDING, &mdev->flags);
 820         spin_unlock_irq(&mdev->req_lock);
 821         wake_up(&mdev->state_wait);
 822 }
 823
 824 /* Make sure IO is suspended before calling this function(). */
 825 static void drbd_suspend_al(struct drbd_conf *mdev)
 826 {
 827         int s = 0;
 828
 829         if (lc_try_lock(mdev->act_log)) {
 830                 drbd_al_shrink(mdev);
 831                 lc_unlock(mdev->act_log);
 832         } else {
 833                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
 834                 return;
 835         }
 836
 837         spin_lock_irq(&mdev->req_lock);
 838         if (mdev->state.conn < C_CONNECTED)
 839                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
 840
 841         spin_unlock_irq(&mdev->req_lock);
 842
 843         if (s)
 844                 dev_info(DEV, "Suspended AL updates\n");
 845 }
 846
 847 /* does always return 0;
 848  * interesting return code is in reply->ret_code */
 849 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 850                              struct drbd_nl_cfg_reply *reply)
 851 {
 852         enum drbd_ret_codes retcode;
 853         enum determine_dev_size dd;
 854         sector_t max_possible_sectors;
 855         sector_t min_md_device_sectors;
 856         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
 857         struct block_device *bdev;
 858         struct lru_cache *resync_lru = NULL;
 859         union drbd_state ns, os;
 860         unsigned int max_bio_size;
 861         int rv;
 862         int cp_discovered = 0;
 863         int logical_block_size;
 864
 865         drbd_reconfig_start(mdev);
 866
 867         /* if you want to reconfigure, please tear down first */
 868         if (mdev->state.disk > D_DISKLESS) {
 869                 retcode = ERR_DISK_CONFIGURED;
 870                 goto fail;
 871         }
 872         /* It may just now have detached because of IO error.  Make sure
 873          * drbd_ldev_destroy is done already, we may end up here very fast,
 874          * e.g. if someone calls attach from the on-io-error handler,
 875          * to realize a "hot spare" feature (not that I'd recommend that) */
 876         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
 877
 878         /* allocation not in the IO path, cqueue thread context */
 879         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
 880         if (!nbc) {
 881                 retcode = ERR_NOMEM;
 882                 goto fail;
 883         }
 884
 885         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
 886         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
 887         nbc->dc.fencing       = DRBD_FENCING_DEF;
 888         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
 889
 890         if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
 891                 retcode = ERR_MANDATORY_TAG;
 892                 goto fail;
 893         }
 894
 895         if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
 896                 retcode = ERR_MD_IDX_INVALID;
 897                 goto fail;
 898         }
 899
 900         if (get_net_conf(mdev)) {
 901                 int prot = mdev->net_conf->wire_protocol;
 902                 put_net_conf(mdev);
 903                 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
 904                         retcode = ERR_STONITH_AND_PROT_A;
 905                         goto fail;
 906                 }
 907         }
 908
 909         bdev = blkdev_get_by_path(nbc->dc.backing_dev,
 910                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
 911         if (IS_ERR(bdev)) {
 912                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
 913                         PTR_ERR(bdev));
 914                 retcode = ERR_OPEN_DISK;
 915                 goto fail;
 916         }
 917         nbc->backing_bdev = bdev;
 918
 919         /*
 920          * meta_dev_idx >= 0: external fixed size, possibly multiple
 921          * drbd sharing one meta device.  TODO in that case, paranoia
 922          * check that [md_bdev, meta_dev_idx] is not yet used by some
 923          * other drbd minor!  (if you use drbd.conf + drbdadm, that
 924          * should check it for you already; but if you don't, or
 925          * someone fooled it, we need to double check here)
 926          */
 927         bdev = blkdev_get_by_path(nbc->dc.meta_dev,
 928                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
 929                                   (nbc->dc.meta_dev_idx < 0) ?
 930                                   (void *)mdev : (void *)drbd_m_holder);
 931         if (IS_ERR(bdev)) {
 932                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
 933                         PTR_ERR(bdev));
 934                 retcode = ERR_OPEN_MD_DISK;
 935                 goto fail;
 936         }
 937         nbc->md_bdev = bdev;
 938
 939         if ((nbc->backing_bdev == nbc->md_bdev) !=
 940             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
 941              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
 942                 retcode = ERR_MD_IDX_INVALID;
 943                 goto fail;
 944         }
 945
 946         resync_lru = lc_create("resync", drbd_bm_ext_cache,
 947                         61, sizeof(struct bm_extent),
 948                         offsetof(struct bm_extent, lce));
 949         if (!resync_lru) {
 950                 retcode = ERR_NOMEM;
 951                 goto fail;
 952         }
 953
 954         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
 955         drbd_md_set_sector_offsets(mdev, nbc);
 956
 957         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
 958                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
 959                         (unsigned long long) drbd_get_max_capacity(nbc),
 960                         (unsigned long long) nbc->dc.disk_size);
 961                 retcode = ERR_DISK_TO_SMALL;
 962                 goto fail;
 963         }
 964
 965         if (nbc->dc.meta_dev_idx < 0) {
 966                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
 967                 /* at least one MB, otherwise it does not make sense */
 968                 min_md_device_sectors = (2<<10);
 969         } else {
 970                 max_possible_sectors = DRBD_MAX_SECTORS;
 971                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
 972         }
 973
 974         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
 975                 retcode = ERR_MD_DISK_TO_SMALL;
 976                 dev_warn(DEV, "refusing attach: md-device too small, "
 977                      "at least %llu sectors needed for this meta-disk type\n",
 978                      (unsigned long long) min_md_device_sectors);
 979                 goto fail;
 980         }
 981
 982         /* Make sure the new disk is big enough
 983          * (we may currently be R_PRIMARY with no local disk...) */
 984         if (drbd_get_max_capacity(nbc) <
 985             drbd_get_capacity(mdev->this_bdev)) {
 986                 retcode = ERR_DISK_TO_SMALL;
 987                 goto fail;
 988         }
 989
 990         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
 991
 992         if (nbc->known_size > max_possible_sectors) {
 993                 dev_warn(DEV, "==> truncating very big lower level device "
 994                         "to currently maximum possible %llu sectors <==\n",
 995                         (unsigned long long) max_possible_sectors);
 996                 if (nbc->dc.meta_dev_idx >= 0)
 997                         dev_warn(DEV, "==>> using internal or flexible "
 998                                       "meta data may help <<==\n");
 999         }
1000
1001         drbd_suspend_io(mdev);
1002         /* also wait for the last barrier ack. */
1003         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state));
1004         /* and for any other previously queued work */
1005         drbd_flush_workqueue(mdev);
1006
1007         retcode = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1008         drbd_resume_io(mdev);
1009         if (retcode < SS_SUCCESS)
1010                 goto fail;
1011
1012         if (!get_ldev_if_state(mdev, D_ATTACHING))
1013                 goto force_diskless;
1014
1015         drbd_md_set_sector_offsets(mdev, nbc);
1016
1017         /* allocate a second IO page if logical_block_size != 512 */
1018         logical_block_size = bdev_logical_block_size(nbc->md_bdev);
1019         if (logical_block_size == 0)
1020                 logical_block_size = MD_SECTOR_SIZE;
1021
1022         if (logical_block_size != MD_SECTOR_SIZE) {
1023                 if (!mdev->md_io_tmpp) {
1024                         struct page *page = alloc_page(GFP_NOIO);
1025                         if (!page)
1026                                 goto force_diskless_dec;
1027
1028                         dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
1029                              logical_block_size, MD_SECTOR_SIZE);
1030                         dev_warn(DEV, "Workaround engaged (has performance impact).\n");
1031
1032                         mdev->md_io_tmpp = page;
1033                 }
1034         }
1035
1036         if (!mdev->bitmap) {
1037                 if (drbd_bm_init(mdev)) {
1038                         retcode = ERR_NOMEM;
1039                         goto force_diskless_dec;
1040                 }
1041         }
1042
1043         retcode = drbd_md_read(mdev, nbc);
1044         if (retcode != NO_ERROR)
1045                 goto force_diskless_dec;
1046
1047         if (mdev->state.conn < C_CONNECTED &&
1048             mdev->state.role == R_PRIMARY &&
1049             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1050                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1051                     (unsigned long long)mdev->ed_uuid);
1052                 retcode = ERR_DATA_NOT_CURRENT;
1053                 goto force_diskless_dec;
1054         }
1055
1056         /* Since we are diskless, fix the activity log first... */
1057         if (drbd_check_al_size(mdev)) {
1058                 retcode = ERR_NOMEM;
1059                 goto force_diskless_dec;
1060         }
1061
1062         /* Prevent shrinking of consistent devices ! */
1063         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1064             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1065                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1066                 retcode = ERR_DISK_TO_SMALL;
1067                 goto force_diskless_dec;
1068         }
1069
1070         if (!drbd_al_read_log(mdev, nbc)) {
1071                 retcode = ERR_IO_MD_DISK;
1072                 goto force_diskless_dec;
1073         }
1074
1075         /* Reset the "barriers don't work" bits here, then force meta data to
1076          * be written, to ensure we determine if barriers are supported. */
1077         if (nbc->dc.no_md_flush)
1078                 set_bit(MD_NO_FUA, &mdev->flags);
1079         else
1080                 clear_bit(MD_NO_FUA, &mdev->flags);
1081
1082         /* Point of no return reached.
1083          * Devices and memory are no longer released by error cleanup below.
1084          * now mdev takes over responsibility, and the state engine should
1085          * clean it up somewhere.  */
1086         D_ASSERT(mdev->ldev == NULL);
1087         mdev->ldev = nbc;
1088         mdev->resync = resync_lru;
1089         nbc = NULL;
1090         resync_lru = NULL;
1091
1092         mdev->write_ordering = WO_bdev_flush;
1093         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1094
1095         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1096                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1097         else
1098                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1099
1100         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1101             !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) {
1102                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1103                 cp_discovered = 1;
1104         }
1105
1106         mdev->send_cnt = 0;
1107         mdev->recv_cnt = 0;
1108         mdev->read_cnt = 0;
1109         mdev->writ_cnt = 0;
1110
1111         max_bio_size = DRBD_MAX_BIO_SIZE;
1112         if (mdev->state.conn == C_CONNECTED) {
1113                 /* We are Primary, Connected, and now attach a new local
1114                  * backing store. We must not increase the user visible maximum
1115                  * bio size on this device to something the peer may not be
1116                  * able to handle. */
1117                 if (mdev->agreed_pro_version < 94)
1118                         max_bio_size = queue_max_hw_sectors(mdev->rq_queue) << 9;
1119                 else if (mdev->agreed_pro_version == 94)
1120                         max_bio_size = DRBD_MAX_SIZE_H80_PACKET;
1121                 /* else: drbd 8.3.9 and later, stay with default */
1122         }
1123
1124         drbd_setup_queue_param(mdev, max_bio_size);
1125
1126         /* If I am currently not R_PRIMARY,
1127          * but meta data primary indicator is set,
1128          * I just now recover from a hard crash,
1129          * and have been R_PRIMARY before that crash.
1130          *
1131          * Now, if I had no connection before that crash
1132          * (have been degraded R_PRIMARY), chances are that
1133          * I won't find my peer now either.
1134          *
1135          * In that case, and _only_ in that case,
1136          * we use the degr-wfc-timeout instead of the default,
1137          * so we can automatically recover from a crash of a
1138          * degraded but active "cluster" after a certain timeout.
1139          */
1140         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1141         if (mdev->state.role != R_PRIMARY &&
1142              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1143             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1144                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1145
1146         dd = drbd_determin_dev_size(mdev, 0);
1147         if (dd == dev_size_error) {
1148                 retcode = ERR_NOMEM_BITMAP;
1149                 goto force_diskless_dec;
1150         } else if (dd == grew)
1151                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1152
1153         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1154                 dev_info(DEV, "Assuming that all blocks are out of sync "
1155                      "(aka FullSync)\n");
1156                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from attaching")) {
1157                         retcode = ERR_IO_MD_DISK;
1158                         goto force_diskless_dec;
1159                 }
1160         } else {
1161                 if (drbd_bitmap_io(mdev, &drbd_bm_read, "read from attaching") < 0) {
1162                         retcode = ERR_IO_MD_DISK;
1163                         goto force_diskless_dec;
1164                 }
1165         }
1166
1167         if (cp_discovered) {
1168                 drbd_al_apply_to_bm(mdev);
1169                 drbd_al_to_on_disk_bm(mdev);
1170         }
1171
1172         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1173                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1174
1175         spin_lock_irq(&mdev->req_lock);
1176         os = mdev->state;
1177         ns.i = os.i;
1178         /* If MDF_CONSISTENT is not set go into inconsistent state,
1179            otherwise investigate MDF_WasUpToDate...
1180            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1181            otherwise into D_CONSISTENT state.
1182         */
1183         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1184                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1185                         ns.disk = D_CONSISTENT;
1186                 else
1187                         ns.disk = D_OUTDATED;
1188         } else {
1189                 ns.disk = D_INCONSISTENT;
1190         }
1191
1192         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1193                 ns.pdsk = D_OUTDATED;
1194
1195         if ( ns.disk == D_CONSISTENT &&
1196             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1197                 ns.disk = D_UP_TO_DATE;
1198
1199         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1200            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1201            this point, because drbd_request_state() modifies these
1202            flags. */
1203
1204         /* In case we are C_CONNECTED postpone any decision on the new disk
1205            state after the negotiation phase. */
1206         if (mdev->state.conn == C_CONNECTED) {
1207                 mdev->new_state_tmp.i = ns.i;
1208                 ns.i = os.i;
1209                 ns.disk = D_NEGOTIATING;
1210
1211                 /* We expect to receive up-to-date UUIDs soon.
1212                    To avoid a race in receive_state, free p_uuid while
1213                    holding req_lock. I.e. atomic with the state change */
1214                 kfree(mdev->p_uuid);
1215                 mdev->p_uuid = NULL;
1216         }
1217
1218         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1219         ns = mdev->state;
1220         spin_unlock_irq(&mdev->req_lock);
1221
1222         if (rv < SS_SUCCESS)
1223                 goto force_diskless_dec;
1224
1225         if (mdev->state.role == R_PRIMARY)
1226                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1227         else
1228                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1229
1230         drbd_md_mark_dirty(mdev);
1231         drbd_md_sync(mdev);
1232
1233         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1234         put_ldev(mdev);
1235         reply->ret_code = retcode;
1236         drbd_reconfig_done(mdev);
1237         return 0;
1238
1239  force_diskless_dec:
1240         put_ldev(mdev);
1241  force_diskless:
1242         drbd_force_state(mdev, NS(disk, D_FAILED));
1243         drbd_md_sync(mdev);
1244  fail:
1245         if (nbc) {
1246                 if (nbc->backing_bdev)
1247                         blkdev_put(nbc->backing_bdev,
1248                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1249                 if (nbc->md_bdev)
1250                         blkdev_put(nbc->md_bdev,
1251                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1252                 kfree(nbc);
1253         }
1254         lc_destroy(resync_lru);
1255
1256         reply->ret_code = retcode;
1257         drbd_reconfig_done(mdev);
1258         return 0;
1259 }
1260
1261 /* Detaching the disk is a process in multiple stages.  First we need to lock
1262  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1263  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1264  * internal references as well.
1265  * Only then we have finally detached. */
1266 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1267                           struct drbd_nl_cfg_reply *reply)
1268 {
1269         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1270         reply->ret_code = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1271         if (mdev->state.disk == D_DISKLESS)
1272                 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1273         drbd_resume_io(mdev);
1274         return 0;
1275 }
1276
1277 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1278                             struct drbd_nl_cfg_reply *reply)
1279 {
1280         int i, ns;
1281         enum drbd_ret_codes retcode;
1282         struct net_conf *new_conf = NULL;
1283         struct crypto_hash *tfm = NULL;
1284         struct crypto_hash *integrity_w_tfm = NULL;
1285         struct crypto_hash *integrity_r_tfm = NULL;
1286         struct hlist_head *new_tl_hash = NULL;
1287         struct hlist_head *new_ee_hash = NULL;
1288         struct drbd_conf *odev;
1289         char hmac_name[CRYPTO_MAX_ALG_NAME];
1290         void *int_dig_out = NULL;
1291         void *int_dig_in = NULL;
1292         void *int_dig_vv = NULL;
1293         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1294
1295         drbd_reconfig_start(mdev);
1296
1297         if (mdev->state.conn > C_STANDALONE) {
1298                 retcode = ERR_NET_CONFIGURED;
1299                 goto fail;
1300         }
1301
1302         /* allocation not in the IO path, cqueue thread context */
1303         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1304         if (!new_conf) {
1305                 retcode = ERR_NOMEM;
1306                 goto fail;
1307         }
1308
1309         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1310         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1311         new_conf->ping_int         = DRBD_PING_INT_DEF;
1312         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1313         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1314         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1315         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1316         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1317         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1318         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1319         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1320         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1321         new_conf->want_lose        = 0;
1322         new_conf->two_primaries    = 0;
1323         new_conf->wire_protocol    = DRBD_PROT_C;
1324         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1325         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1326         new_conf->on_congestion    = DRBD_ON_CONGESTION_DEF;
1327         new_conf->cong_extents     = DRBD_CONG_EXTENTS_DEF;
1328
1329         if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1330                 retcode = ERR_MANDATORY_TAG;
1331                 goto fail;
1332         }
1333
1334         if (new_conf->two_primaries
1335             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1336                 retcode = ERR_NOT_PROTO_C;
1337                 goto fail;
1338         }
1339
1340         if (get_ldev(mdev)) {
1341                 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1342                 put_ldev(mdev);
1343                 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1344                         retcode = ERR_STONITH_AND_PROT_A;
1345                         goto fail;
1346                 }
1347         }
1348
1349         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1350                 retcode = ERR_CONG_NOT_PROTO_A;
1351                 goto fail;
1352         }
1353
1354         if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1355                 retcode = ERR_DISCARD;
1356                 goto fail;
1357         }
1358
1359         retcode = NO_ERROR;
1360
1361         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1362         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1363         for (i = 0; i < minor_count; i++) {
1364                 odev = minor_to_mdev(i);
1365                 if (!odev || odev == mdev)
1366                         continue;
1367                 if (get_net_conf(odev)) {
1368                         taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1369                         if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1370                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1371                                 retcode = ERR_LOCAL_ADDR;
1372
1373                         taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1374                         if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1375                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1376                                 retcode = ERR_PEER_ADDR;
1377
1378                         put_net_conf(odev);
1379                         if (retcode != NO_ERROR)
1380                                 goto fail;
1381                 }
1382         }
1383
1384         if (new_conf->cram_hmac_alg[0] != 0) {
1385                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1386                         new_conf->cram_hmac_alg);
1387                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1388                 if (IS_ERR(tfm)) {
1389                         tfm = NULL;
1390                         retcode = ERR_AUTH_ALG;
1391                         goto fail;
1392                 }
1393
1394                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1395                         retcode = ERR_AUTH_ALG_ND;
1396                         goto fail;
1397                 }
1398         }
1399
1400         if (new_conf->integrity_alg[0]) {
1401                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1402                 if (IS_ERR(integrity_w_tfm)) {
1403                         integrity_w_tfm = NULL;
1404                         retcode=ERR_INTEGRITY_ALG;
1405                         goto fail;
1406                 }
1407
1408                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1409                         retcode=ERR_INTEGRITY_ALG_ND;
1410                         goto fail;
1411                 }
1412
1413                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1414                 if (IS_ERR(integrity_r_tfm)) {
1415                         integrity_r_tfm = NULL;
1416                         retcode=ERR_INTEGRITY_ALG;
1417                         goto fail;
1418                 }
1419         }
1420
1421         ns = new_conf->max_epoch_size/8;
1422         if (mdev->tl_hash_s != ns) {
1423                 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1424                 if (!new_tl_hash) {
1425                         retcode = ERR_NOMEM;
1426                         goto fail;
1427                 }
1428         }
1429
1430         ns = new_conf->max_buffers/8;
1431         if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1432                 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1433                 if (!new_ee_hash) {
1434                         retcode = ERR_NOMEM;
1435                         goto fail;
1436                 }
1437         }
1438
1439         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1440
1441         if (integrity_w_tfm) {
1442                 i = crypto_hash_digestsize(integrity_w_tfm);
1443                 int_dig_out = kmalloc(i, GFP_KERNEL);
1444                 if (!int_dig_out) {
1445                         retcode = ERR_NOMEM;
1446                         goto fail;
1447                 }
1448                 int_dig_in = kmalloc(i, GFP_KERNEL);
1449                 if (!int_dig_in) {
1450                         retcode = ERR_NOMEM;
1451                         goto fail;
1452                 }
1453                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1454                 if (!int_dig_vv) {
1455                         retcode = ERR_NOMEM;
1456                         goto fail;
1457                 }
1458         }
1459
1460         if (!mdev->bitmap) {
1461                 if(drbd_bm_init(mdev)) {
1462                         retcode = ERR_NOMEM;
1463                         goto fail;
1464                 }
1465         }
1466
1467         drbd_flush_workqueue(mdev);
1468         spin_lock_irq(&mdev->req_lock);
1469         if (mdev->net_conf != NULL) {
1470                 retcode = ERR_NET_CONFIGURED;
1471                 spin_unlock_irq(&mdev->req_lock);
1472                 goto fail;
1473         }
1474         mdev->net_conf = new_conf;
1475
1476         mdev->send_cnt = 0;
1477         mdev->recv_cnt = 0;
1478
1479         if (new_tl_hash) {
1480                 kfree(mdev->tl_hash);
1481                 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1482                 mdev->tl_hash = new_tl_hash;
1483         }
1484
1485         if (new_ee_hash) {
1486                 kfree(mdev->ee_hash);
1487                 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1488                 mdev->ee_hash = new_ee_hash;
1489         }
1490
1491         crypto_free_hash(mdev->cram_hmac_tfm);
1492         mdev->cram_hmac_tfm = tfm;
1493
1494         crypto_free_hash(mdev->integrity_w_tfm);
1495         mdev->integrity_w_tfm = integrity_w_tfm;
1496
1497         crypto_free_hash(mdev->integrity_r_tfm);
1498         mdev->integrity_r_tfm = integrity_r_tfm;
1499
1500         kfree(mdev->int_dig_out);
1501         kfree(mdev->int_dig_in);
1502         kfree(mdev->int_dig_vv);
1503         mdev->int_dig_out=int_dig_out;
1504         mdev->int_dig_in=int_dig_in;
1505         mdev->int_dig_vv=int_dig_vv;
1506         retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL);
1507         spin_unlock_irq(&mdev->req_lock);
1508
1509         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1510         reply->ret_code = retcode;
1511         drbd_reconfig_done(mdev);
1512         return 0;
1513
1514 fail:
1515         kfree(int_dig_out);
1516         kfree(int_dig_in);
1517         kfree(int_dig_vv);
1518         crypto_free_hash(tfm);
1519         crypto_free_hash(integrity_w_tfm);
1520         crypto_free_hash(integrity_r_tfm);
1521         kfree(new_tl_hash);
1522         kfree(new_ee_hash);
1523         kfree(new_conf);
1524
1525         reply->ret_code = retcode;
1526         drbd_reconfig_done(mdev);
1527         return 0;
1528 }
1529
1530 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1531                               struct drbd_nl_cfg_reply *reply)
1532 {
1533         int retcode;
1534
1535         retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1536
1537         if (retcode == SS_NOTHING_TO_DO)
1538                 goto done;
1539         else if (retcode == SS_ALREADY_STANDALONE)
1540                 goto done;
1541         else if (retcode == SS_PRIMARY_NOP) {
1542                 /* Our statche checking code wants to see the peer outdated. */
1543                 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1544                                                       pdsk, D_OUTDATED));
1545         } else if (retcode == SS_CW_FAILED_BY_PEER) {
1546                 /* The peer probably wants to see us outdated. */
1547                 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1548                                                         disk, D_OUTDATED),
1549                                               CS_ORDERED);
1550                 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1551                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1552                         retcode = SS_SUCCESS;
1553                 }
1554         }
1555
1556         if (retcode < SS_SUCCESS)
1557                 goto fail;
1558
1559         if (wait_event_interruptible(mdev->state_wait,
1560                                      mdev->state.conn != C_DISCONNECTING)) {
1561                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1562                    someone else might connect us in the mean time! */
1563                 retcode = ERR_INTR;
1564                 goto fail;
1565         }
1566
1567  done:
1568         retcode = NO_ERROR;
1569  fail:
1570         drbd_md_sync(mdev);
1571         reply->ret_code = retcode;
1572         return 0;
1573 }
1574
1575 void resync_after_online_grow(struct drbd_conf *mdev)
1576 {
1577         int iass; /* I am sync source */
1578
1579         dev_info(DEV, "Resync of new storage after online grow\n");
1580         if (mdev->state.role != mdev->state.peer)
1581                 iass = (mdev->state.role == R_PRIMARY);
1582         else
1583                 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1584
1585         if (iass)
1586                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1587         else
1588                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1589 }
1590
1591 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1592                           struct drbd_nl_cfg_reply *reply)
1593 {
1594         struct resize rs;
1595         int retcode = NO_ERROR;
1596         enum determine_dev_size dd;
1597         enum dds_flags ddsf;
1598
1599         memset(&rs, 0, sizeof(struct resize));
1600         if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1601                 retcode = ERR_MANDATORY_TAG;
1602                 goto fail;
1603         }
1604
1605         if (mdev->state.conn > C_CONNECTED) {
1606                 retcode = ERR_RESIZE_RESYNC;
1607                 goto fail;
1608         }
1609
1610         if (mdev->state.role == R_SECONDARY &&
1611             mdev->state.peer == R_SECONDARY) {
1612                 retcode = ERR_NO_PRIMARY;
1613                 goto fail;
1614         }
1615
1616         if (!get_ldev(mdev)) {
1617                 retcode = ERR_NO_DISK;
1618                 goto fail;
1619         }
1620
1621         if (rs.no_resync && mdev->agreed_pro_version < 93) {
1622                 retcode = ERR_NEED_APV_93;
1623                 goto fail;
1624         }
1625
1626         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1627                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1628
1629         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1630         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1631         dd = drbd_determin_dev_size(mdev, ddsf);
1632         drbd_md_sync(mdev);
1633         put_ldev(mdev);
1634         if (dd == dev_size_error) {
1635                 retcode = ERR_NOMEM_BITMAP;
1636                 goto fail;
1637         }
1638
1639         if (mdev->state.conn == C_CONNECTED) {
1640                 if (dd == grew)
1641                         set_bit(RESIZE_PENDING, &mdev->flags);
1642
1643                 drbd_send_uuids(mdev);
1644                 drbd_send_sizes(mdev, 1, ddsf);
1645         }
1646
1647  fail:
1648         reply->ret_code = retcode;
1649         return 0;
1650 }
1651
1652 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1653                                struct drbd_nl_cfg_reply *reply)
1654 {
1655         int retcode = NO_ERROR;
1656         int err;
1657         int ovr; /* online verify running */
1658         int rsr; /* re-sync running */
1659         struct crypto_hash *verify_tfm = NULL;
1660         struct crypto_hash *csums_tfm = NULL;
1661         struct syncer_conf sc;
1662         cpumask_var_t new_cpu_mask;
1663         int *rs_plan_s = NULL;
1664         int fifo_size;
1665
1666         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1667                 retcode = ERR_NOMEM;
1668                 goto fail;
1669         }
1670
1671         if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1672                 memset(&sc, 0, sizeof(struct syncer_conf));
1673                 sc.rate       = DRBD_RATE_DEF;
1674                 sc.after      = DRBD_AFTER_DEF;
1675                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1676                 sc.on_no_data  = DRBD_ON_NO_DATA_DEF;
1677                 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1678                 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1679                 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1680                 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1681                 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1682         } else
1683                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1684
1685         if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1686                 retcode = ERR_MANDATORY_TAG;
1687                 goto fail;
1688         }
1689
1690         /* re-sync running */
1691         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1692                 mdev->state.conn == C_SYNC_TARGET ||
1693                 mdev->state.conn == C_PAUSED_SYNC_S ||
1694                 mdev->state.conn == C_PAUSED_SYNC_T );
1695
1696         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1697                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1698                 goto fail;
1699         }
1700
1701         if (!rsr && sc.csums_alg[0]) {
1702                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1703                 if (IS_ERR(csums_tfm)) {
1704                         csums_tfm = NULL;
1705                         retcode = ERR_CSUMS_ALG;
1706                         goto fail;
1707                 }
1708
1709                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1710                         retcode = ERR_CSUMS_ALG_ND;
1711                         goto fail;
1712                 }
1713         }
1714
1715         /* online verify running */
1716         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1717
1718         if (ovr) {
1719                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1720                         retcode = ERR_VERIFY_RUNNING;
1721                         goto fail;
1722                 }
1723         }
1724
1725         if (!ovr && sc.verify_alg[0]) {
1726                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1727                 if (IS_ERR(verify_tfm)) {
1728                         verify_tfm = NULL;
1729                         retcode = ERR_VERIFY_ALG;
1730                         goto fail;
1731                 }
1732
1733                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1734                         retcode = ERR_VERIFY_ALG_ND;
1735                         goto fail;
1736                 }
1737         }
1738
1739         /* silently ignore cpu mask on UP kernel */
1740         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1741                 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1742                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1743                 if (err) {
1744                         dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1745                         retcode = ERR_CPU_MASK_PARSE;
1746                         goto fail;
1747                 }
1748         }
1749
1750         ERR_IF (sc.rate < 1) sc.rate = 1;
1751         ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1752 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1753         if (sc.al_extents > AL_MAX) {
1754                 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1755                 sc.al_extents = AL_MAX;
1756         }
1757 #undef AL_MAX
1758
1759         /* to avoid spurious errors when configuring minors before configuring
1760          * the minors they depend on: if necessary, first create the minor we
1761          * depend on */
1762         if (sc.after >= 0)
1763                 ensure_mdev(sc.after, 1);
1764
1765         /* most sanity checks done, try to assign the new sync-after
1766          * dependency.  need to hold the global lock in there,
1767          * to avoid a race in the dependency loop check. */
1768         retcode = drbd_alter_sa(mdev, sc.after);
1769         if (retcode != NO_ERROR)
1770                 goto fail;
1771
1772         fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1773         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1774                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1775                 if (!rs_plan_s) {
1776                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1777                         retcode = ERR_NOMEM;
1778                         goto fail;
1779                 }
1780         }
1781
1782         /* ok, assign the rest of it as well.
1783          * lock against receive_SyncParam() */
1784         spin_lock(&mdev->peer_seq_lock);
1785         mdev->sync_conf = sc;
1786
1787         if (!rsr) {
1788                 crypto_free_hash(mdev->csums_tfm);
1789                 mdev->csums_tfm = csums_tfm;
1790                 csums_tfm = NULL;
1791         }
1792
1793         if (!ovr) {
1794                 crypto_free_hash(mdev->verify_tfm);
1795                 mdev->verify_tfm = verify_tfm;
1796                 verify_tfm = NULL;
1797         }
1798
1799         if (fifo_size != mdev->rs_plan_s.size) {
1800                 kfree(mdev->rs_plan_s.values);
1801                 mdev->rs_plan_s.values = rs_plan_s;
1802                 mdev->rs_plan_s.size   = fifo_size;
1803                 mdev->rs_planed = 0;
1804                 rs_plan_s = NULL;
1805         }
1806
1807         spin_unlock(&mdev->peer_seq_lock);
1808
1809         if (get_ldev(mdev)) {
1810                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1811                 drbd_al_shrink(mdev);
1812                 err = drbd_check_al_size(mdev);
1813                 lc_unlock(mdev->act_log);
1814                 wake_up(&mdev->al_wait);
1815
1816                 put_ldev(mdev);
1817                 drbd_md_sync(mdev);
1818
1819                 if (err) {
1820                         retcode = ERR_NOMEM;
1821                         goto fail;
1822                 }
1823         }
1824
1825         if (mdev->state.conn >= C_CONNECTED)
1826                 drbd_send_sync_param(mdev, &sc);
1827
1828         if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1829                 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1830                 drbd_calc_cpu_mask(mdev);
1831                 mdev->receiver.reset_cpu_mask = 1;
1832                 mdev->asender.reset_cpu_mask = 1;
1833                 mdev->worker.reset_cpu_mask = 1;
1834         }
1835
1836         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1837 fail:
1838         kfree(rs_plan_s);
1839         free_cpumask_var(new_cpu_mask);
1840         crypto_free_hash(csums_tfm);
1841         crypto_free_hash(verify_tfm);
1842         reply->ret_code = retcode;
1843         return 0;
1844 }
1845
1846 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1847                               struct drbd_nl_cfg_reply *reply)
1848 {
1849         int retcode;
1850
1851         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1852
1853         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1854                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1855
1856         while (retcode == SS_NEED_CONNECTION) {
1857                 spin_lock_irq(&mdev->req_lock);
1858                 if (mdev->state.conn < C_CONNECTED)
1859                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1860                 spin_unlock_irq(&mdev->req_lock);
1861
1862                 if (retcode != SS_NEED_CONNECTION)
1863                         break;
1864
1865                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1866         }
1867
1868         reply->ret_code = retcode;
1869         return 0;
1870 }
1871
1872 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
1873 {
1874         int rv;
1875
1876         rv = drbd_bmio_set_n_write(mdev);
1877         drbd_suspend_al(mdev);
1878         return rv;
1879 }
1880
1881 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1882                                    struct drbd_nl_cfg_reply *reply)
1883 {
1884         int retcode;
1885
1886         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
1887
1888         if (retcode < SS_SUCCESS) {
1889                 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
1890                         /* The peer will get a resync upon connect anyways. Just make that
1891                            into a full resync. */
1892                         retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
1893                         if (retcode >= SS_SUCCESS) {
1894                                 /* open coded drbd_bitmap_io() */
1895                                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
1896                                                    "set_n_write from invalidate_peer"))
1897                                         retcode = ERR_IO_MD_DISK;
1898                         }
1899                 } else
1900                         retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
1901         }
1902
1903         reply->ret_code = retcode;
1904         return 0;
1905 }
1906
1907 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1908                               struct drbd_nl_cfg_reply *reply)
1909 {
1910         int retcode = NO_ERROR;
1911
1912         if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
1913                 retcode = ERR_PAUSE_IS_SET;
1914
1915         reply->ret_code = retcode;
1916         return 0;
1917 }
1918
1919 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1920                                struct drbd_nl_cfg_reply *reply)
1921 {
1922         int retcode = NO_ERROR;
1923
1924         if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO)
1925                 retcode = ERR_PAUSE_IS_CLEAR;
1926
1927         reply->ret_code = retcode;
1928         return 0;
1929 }
1930
1931 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1932                               struct drbd_nl_cfg_reply *reply)
1933 {
1934         reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
1935
1936         return 0;
1937 }
1938
1939 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1940                              struct drbd_nl_cfg_reply *reply)
1941 {
1942         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
1943                 drbd_uuid_new_current(mdev);
1944                 clear_bit(NEW_CUR_UUID, &mdev->flags);
1945         }
1946         drbd_suspend_io(mdev);
1947         reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
1948         if (reply->ret_code == SS_SUCCESS) {
1949                 if (mdev->state.conn < C_CONNECTED)
1950                         tl_clear(mdev);
1951                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
1952                         tl_restart(mdev, fail_frozen_disk_io);
1953         }
1954         drbd_resume_io(mdev);
1955
1956         return 0;
1957 }
1958
1959 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1960                            struct drbd_nl_cfg_reply *reply)
1961 {
1962         reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
1963         return 0;
1964 }
1965
1966 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1967                            struct drbd_nl_cfg_reply *reply)
1968 {
1969         unsigned short *tl;
1970
1971         tl = reply->tag_list;
1972
1973         if (get_ldev(mdev)) {
1974                 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
1975                 put_ldev(mdev);
1976         }
1977
1978         if (get_net_conf(mdev)) {
1979                 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
1980                 put_net_conf(mdev);
1981         }
1982         tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
1983
1984         put_unaligned(TT_END, tl++); /* Close the tag list */
1985
1986         return (int)((char *)tl - (char *)reply->tag_list);
1987 }
1988
1989 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1990                              struct drbd_nl_cfg_reply *reply)
1991 {
1992         unsigned short *tl = reply->tag_list;
1993         union drbd_state s = mdev->state;
1994         unsigned long rs_left;
1995         unsigned int res;
1996
1997         tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
1998
1999         /* no local ref, no bitmap, no syncer progress. */
2000         if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
2001                 if (get_ldev(mdev)) {
2002                         drbd_get_syncer_progress(mdev, &rs_left, &res);
2003                         tl = tl_add_int(tl, T_sync_progress, &res);
2004                         put_ldev(mdev);
2005                 }
2006         }
2007         put_unaligned(TT_END, tl++); /* Close the tag list */
2008
2009         return (int)((char *)tl - (char *)reply->tag_list);
2010 }
2011
2012 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2013                              struct drbd_nl_cfg_reply *reply)
2014 {
2015         unsigned short *tl;
2016
2017         tl = reply->tag_list;
2018
2019         if (get_ldev(mdev)) {
2020                 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
2021                 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
2022                 put_ldev(mdev);
2023         }
2024         put_unaligned(TT_END, tl++); /* Close the tag list */
2025
2026         return (int)((char *)tl - (char *)reply->tag_list);
2027 }
2028
2029 /**
2030  * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
2031  * @mdev:       DRBD device.
2032  * @nlp:        Netlink/connector packet from drbdsetup
2033  * @reply:      Reply packet for drbdsetup
2034  */
2035 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2036                                     struct drbd_nl_cfg_reply *reply)
2037 {
2038         unsigned short *tl;
2039         char rv;
2040
2041         tl = reply->tag_list;
2042
2043         rv = mdev->state.pdsk == D_OUTDATED        ? UT_PEER_OUTDATED :
2044           test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
2045
2046         tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
2047         put_unaligned(TT_END, tl++); /* Close the tag list */
2048
2049         return (int)((char *)tl - (char *)reply->tag_list);
2050 }
2051
2052 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2053                                     struct drbd_nl_cfg_reply *reply)
2054 {
2055         /* default to resume from last known position, if possible */
2056         struct start_ov args =
2057                 { .start_sector = mdev->ov_start_sector };
2058
2059         if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2060                 reply->ret_code = ERR_MANDATORY_TAG;
2061                 return 0;
2062         }
2063         /* w_make_ov_request expects position to be aligned */
2064         mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
2065         reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2066         return 0;
2067 }
2068
2069
2070 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2071                               struct drbd_nl_cfg_reply *reply)
2072 {
2073         int retcode = NO_ERROR;
2074         int skip_initial_sync = 0;
2075         int err;
2076
2077         struct new_c_uuid args;
2078
2079         memset(&args, 0, sizeof(struct new_c_uuid));
2080         if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2081                 reply->ret_code = ERR_MANDATORY_TAG;
2082                 return 0;
2083         }
2084
2085         mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
2086
2087         if (!get_ldev(mdev)) {
2088                 retcode = ERR_NO_DISK;
2089                 goto out;
2090         }
2091
2092         /* this is "skip initial sync", assume to be clean */
2093         if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
2094             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2095                 dev_info(DEV, "Preparing to skip initial sync\n");
2096                 skip_initial_sync = 1;
2097         } else if (mdev->state.conn != C_STANDALONE) {
2098                 retcode = ERR_CONNECTED;
2099                 goto out_dec;
2100         }
2101
2102         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2103         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2104
2105         if (args.clear_bm) {
2106                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, "clear_n_write from new_c_uuid");
2107                 if (err) {
2108                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2109                         retcode = ERR_IO_MD_DISK;
2110                 }
2111                 if (skip_initial_sync) {
2112                         drbd_send_uuids_skip_initial_sync(mdev);
2113                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2114                         spin_lock_irq(&mdev->req_lock);
2115                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2116                                         CS_VERBOSE, NULL);
2117                         spin_unlock_irq(&mdev->req_lock);
2118                 }
2119         }
2120
2121         drbd_md_sync(mdev);
2122 out_dec:
2123         put_ldev(mdev);
2124 out:
2125         mutex_unlock(&mdev->state_mutex);
2126
2127         reply->ret_code = retcode;
2128         return 0;
2129 }
2130
2131 struct cn_handler_struct {
2132         int (*function)(struct drbd_conf *,
2133                          struct drbd_nl_cfg_req *,
2134                          struct drbd_nl_cfg_reply *);
2135         int reply_body_size;
2136 };
2137
2138 static struct cn_handler_struct cnd_table[] = {
2139         [ P_primary ]           = { &drbd_nl_primary,           0 },
2140         [ P_secondary ]         = { &drbd_nl_secondary,         0 },
2141         [ P_disk_conf ]         = { &drbd_nl_disk_conf,         0 },
2142         [ P_detach ]            = { &drbd_nl_detach,            0 },
2143         [ P_net_conf ]          = { &drbd_nl_net_conf,          0 },
2144         [ P_disconnect ]        = { &drbd_nl_disconnect,        0 },
2145         [ P_resize ]            = { &drbd_nl_resize,            0 },
2146         [ P_syncer_conf ]       = { &drbd_nl_syncer_conf,       0 },
2147         [ P_invalidate ]        = { &drbd_nl_invalidate,        0 },
2148         [ P_invalidate_peer ]   = { &drbd_nl_invalidate_peer,   0 },
2149         [ P_pause_sync ]        = { &drbd_nl_pause_sync,        0 },
2150         [ P_resume_sync ]       = { &drbd_nl_resume_sync,       0 },
2151         [ P_suspend_io ]        = { &drbd_nl_suspend_io,        0 },
2152         [ P_resume_io ]         = { &drbd_nl_resume_io,         0 },
2153         [ P_outdate ]           = { &drbd_nl_outdate,           0 },
2154         [ P_get_config ]        = { &drbd_nl_get_config,
2155                                     sizeof(struct syncer_conf_tag_len_struct) +
2156                                     sizeof(struct disk_conf_tag_len_struct) +
2157                                     sizeof(struct net_conf_tag_len_struct) },
2158         [ P_get_state ]         = { &drbd_nl_get_state,
2159                                     sizeof(struct get_state_tag_len_struct) +
2160                                     sizeof(struct sync_progress_tag_len_struct) },
2161         [ P_get_uuids ]         = { &drbd_nl_get_uuids,
2162                                     sizeof(struct get_uuids_tag_len_struct) },
2163         [ P_get_timeout_flag ]  = { &drbd_nl_get_timeout_flag,
2164                                     sizeof(struct get_timeout_flag_tag_len_struct)},
2165         [ P_start_ov ]          = { &drbd_nl_start_ov,          0 },
2166         [ P_new_c_uuid ]        = { &drbd_nl_new_c_uuid,        0 },
2167 };
2168
2169 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2170 {
2171         struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2172         struct cn_handler_struct *cm;
2173         struct cn_msg *cn_reply;
2174         struct drbd_nl_cfg_reply *reply;
2175         struct drbd_conf *mdev;
2176         int retcode, rr;
2177         int reply_size = sizeof(struct cn_msg)
2178                 + sizeof(struct drbd_nl_cfg_reply)
2179                 + sizeof(short int);
2180
2181         if (!try_module_get(THIS_MODULE)) {
2182                 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2183                 return;
2184         }
2185
2186         if (!cap_raised(nsp->eff_cap, CAP_SYS_ADMIN)) {
2187                 retcode = ERR_PERM;
2188                 goto fail;
2189         }
2190
2191         mdev = ensure_mdev(nlp->drbd_minor,
2192                         (nlp->flags & DRBD_NL_CREATE_DEVICE));
2193         if (!mdev) {
2194                 retcode = ERR_MINOR_INVALID;
2195                 goto fail;
2196         }
2197
2198         if (nlp->packet_type >= P_nl_after_last_packet ||
2199             nlp->packet_type == P_return_code_only) {
2200                 retcode = ERR_PACKET_NR;
2201                 goto fail;
2202         }
2203
2204         cm = cnd_table + nlp->packet_type;
2205
2206         /* This may happen if packet number is 0: */
2207         if (cm->function == NULL) {
2208                 retcode = ERR_PACKET_NR;
2209                 goto fail;
2210         }
2211
2212         reply_size += cm->reply_body_size;
2213
2214         /* allocation not in the IO path, cqueue thread context */
2215         cn_reply = kzalloc(reply_size, GFP_KERNEL);
2216         if (!cn_reply) {
2217                 retcode = ERR_NOMEM;
2218                 goto fail;
2219         }
2220         reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2221
2222         reply->packet_type =
2223                 cm->reply_body_size ? nlp->packet_type : P_return_code_only;
2224         reply->minor = nlp->drbd_minor;
2225         reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2226         /* reply->tag_list; might be modified by cm->function. */
2227
2228         rr = cm->function(mdev, nlp, reply);
2229
2230         cn_reply->id = req->id;
2231         cn_reply->seq = req->seq;
2232         cn_reply->ack = req->ack  + 1;
2233         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2234         cn_reply->flags = 0;
2235
2236         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2237         if (rr && rr != -ESRCH)
2238                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2239
2240         kfree(cn_reply);
2241         module_put(THIS_MODULE);
2242         return;
2243  fail:
2244         drbd_nl_send_reply(req, retcode);
2245         module_put(THIS_MODULE);
2246 }
2247
2248 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2249
2250 static unsigned short *
2251 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2252         unsigned short len, int nul_terminated)
2253 {
2254         unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2255         len = (len < l) ? len :  l;
2256         put_unaligned(tag, tl++);
2257         put_unaligned(len, tl++);
2258         memcpy(tl, data, len);
2259         tl = (unsigned short*)((char*)tl + len);
2260         if (nul_terminated)
2261                 *((char*)tl - 1) = 0;
2262         return tl;
2263 }
2264
2265 static unsigned short *
2266 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2267 {
2268         return __tl_add_blob(tl, tag, data, len, 0);
2269 }
2270
2271 static unsigned short *
2272 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2273 {
2274         return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2275 }
2276
2277 static unsigned short *
2278 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2279 {
2280         put_unaligned(tag, tl++);
2281         switch(tag_type(tag)) {
2282         case TT_INTEGER:
2283                 put_unaligned(sizeof(int), tl++);
2284                 put_unaligned(*(int *)val, (int *)tl);
2285                 tl = (unsigned short*)((char*)tl+sizeof(int));
2286                 break;
2287         case TT_INT64:
2288                 put_unaligned(sizeof(u64), tl++);
2289                 put_unaligned(*(u64 *)val, (u64 *)tl);
2290                 tl = (unsigned short*)((char*)tl+sizeof(u64));
2291                 break;
2292         default:
2293                 /* someone did something stupid. */
2294                 ;
2295         }
2296         return tl;
2297 }
2298
2299 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2300 {
2301         char buffer[sizeof(struct cn_msg)+
2302                     sizeof(struct drbd_nl_cfg_reply)+
2303                     sizeof(struct get_state_tag_len_struct)+
2304                     sizeof(short int)];
2305         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2306         struct drbd_nl_cfg_reply *reply =
2307                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2308         unsigned short *tl = reply->tag_list;
2309
2310         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2311
2312         tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2313
2314         put_unaligned(TT_END, tl++); /* Close the tag list */
2315
2316         cn_reply->id.idx = CN_IDX_DRBD;
2317         cn_reply->id.val = CN_VAL_DRBD;
2318
2319         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2320         cn_reply->ack = 0; /* not used here. */
2321         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2322                 (int)((char *)tl - (char *)reply->tag_list);
2323         cn_reply->flags = 0;
2324
2325         reply->packet_type = P_get_state;
2326         reply->minor = mdev_to_minor(mdev);
2327         reply->ret_code = NO_ERROR;
2328
2329         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2330 }
2331
2332 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2333 {
2334         char buffer[sizeof(struct cn_msg)+
2335                     sizeof(struct drbd_nl_cfg_reply)+
2336                     sizeof(struct call_helper_tag_len_struct)+
2337                     sizeof(short int)];
2338         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2339         struct drbd_nl_cfg_reply *reply =
2340                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2341         unsigned short *tl = reply->tag_list;
2342
2343         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2344
2345         tl = tl_add_str(tl, T_helper, helper_name);
2346         put_unaligned(TT_END, tl++); /* Close the tag list */
2347
2348         cn_reply->id.idx = CN_IDX_DRBD;
2349         cn_reply->id.val = CN_VAL_DRBD;
2350
2351         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2352         cn_reply->ack = 0; /* not used here. */
2353         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2354                 (int)((char *)tl - (char *)reply->tag_list);
2355         cn_reply->flags = 0;
2356
2357         reply->packet_type = P_call_helper;
2358         reply->minor = mdev_to_minor(mdev);
2359         reply->ret_code = NO_ERROR;
2360
2361         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2362 }
2363
2364 void drbd_bcast_ee(struct drbd_conf *mdev,
2365                 const char *reason, const int dgs,
2366                 const char* seen_hash, const char* calc_hash,
2367                 const struct drbd_epoch_entry* e)
2368 {
2369         struct cn_msg *cn_reply;
2370         struct drbd_nl_cfg_reply *reply;
2371         unsigned short *tl;
2372         struct page *page;
2373         unsigned len;
2374
2375         if (!e)
2376                 return;
2377         if (!reason || !reason[0])
2378                 return;
2379
2380         /* apparently we have to memcpy twice, first to prepare the data for the
2381          * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2382          * netlink skb. */
2383         /* receiver thread context, which is not in the writeout path (of this node),
2384          * but may be in the writeout path of the _other_ node.
2385          * GFP_NOIO to avoid potential "distributed deadlock". */
2386         cn_reply = kzalloc(
2387                 sizeof(struct cn_msg)+
2388                 sizeof(struct drbd_nl_cfg_reply)+
2389                 sizeof(struct dump_ee_tag_len_struct)+
2390                 sizeof(short int),
2391                 GFP_NOIO);
2392
2393         if (!cn_reply) {
2394                 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2395                                 (unsigned long long)e->sector, e->size);
2396                 return;
2397         }
2398
2399         reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2400         tl = reply->tag_list;
2401
2402         tl = tl_add_str(tl, T_dump_ee_reason, reason);
2403         tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2404         tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2405         tl = tl_add_int(tl, T_ee_sector, &e->sector);
2406         tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2407
2408         /* dump the first 32k */
2409         len = min_t(unsigned, e->size, 32 << 10);
2410         put_unaligned(T_ee_data, tl++);
2411         put_unaligned(len, tl++);
2412
2413         page = e->pages;
2414         page_chain_for_each(page) {
2415                 void *d = kmap_atomic(page, KM_USER0);
2416                 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2417                 memcpy(tl, d, l);
2418                 kunmap_atomic(d, KM_USER0);
2419                 tl = (unsigned short*)((char*)tl + l);
2420                 len -= l;
2421                 if (len == 0)
2422                         break;
2423         }
2424         put_unaligned(TT_END, tl++); /* Close the tag list */
2425
2426         cn_reply->id.idx = CN_IDX_DRBD;
2427         cn_reply->id.val = CN_VAL_DRBD;
2428
2429         cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2430         cn_reply->ack = 0; // not used here.
2431         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2432                 (int)((char*)tl - (char*)reply->tag_list);
2433         cn_reply->flags = 0;
2434
2435         reply->packet_type = P_dump_ee;
2436         reply->minor = mdev_to_minor(mdev);
2437         reply->ret_code = NO_ERROR;
2438
2439         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2440         kfree(cn_reply);
2441 }
2442
2443 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2444 {
2445         char buffer[sizeof(struct cn_msg)+
2446                     sizeof(struct drbd_nl_cfg_reply)+
2447                     sizeof(struct sync_progress_tag_len_struct)+
2448                     sizeof(short int)];
2449         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2450         struct drbd_nl_cfg_reply *reply =
2451                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2452         unsigned short *tl = reply->tag_list;
2453         unsigned long rs_left;
2454         unsigned int res;
2455
2456         /* no local ref, no bitmap, no syncer progress, no broadcast. */
2457         if (!get_ldev(mdev))
2458                 return;
2459         drbd_get_syncer_progress(mdev, &rs_left, &res);
2460         put_ldev(mdev);
2461
2462         tl = tl_add_int(tl, T_sync_progress, &res);
2463         put_unaligned(TT_END, tl++); /* Close the tag list */
2464
2465         cn_reply->id.idx = CN_IDX_DRBD;
2466         cn_reply->id.val = CN_VAL_DRBD;
2467
2468         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2469         cn_reply->ack = 0; /* not used here. */
2470         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2471                 (int)((char *)tl - (char *)reply->tag_list);
2472         cn_reply->flags = 0;
2473
2474         reply->packet_type = P_sync_progress;
2475         reply->minor = mdev_to_minor(mdev);
2476         reply->ret_code = NO_ERROR;
2477
2478         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2479 }
2480
2481 int __init drbd_nl_init(void)
2482 {
2483         static struct cb_id cn_id_drbd;
2484         int err, try=10;
2485
2486         cn_id_drbd.val = CN_VAL_DRBD;
2487         do {
2488                 cn_id_drbd.idx = cn_idx;
2489                 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2490                 if (!err)
2491                         break;
2492                 cn_idx = (cn_idx + CN_IDX_STEP);
2493         } while (try--);
2494
2495         if (err) {
2496                 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2497                 return err;
2498         }
2499
2500         return 0;
2501 }
2502
2503 void drbd_nl_cleanup(void)
2504 {
2505         static struct cb_id cn_id_drbd;
2506
2507         cn_id_drbd.idx = cn_idx;
2508         cn_id_drbd.val = CN_VAL_DRBD;
2509
2510         cn_del_callback(&cn_id_drbd);
2511 }
2512
2513 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2514 {
2515         char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2516         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2517         struct drbd_nl_cfg_reply *reply =
2518                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2519         int rr;
2520
2521         memset(buffer, 0, sizeof(buffer));
2522         cn_reply->id = req->id;
2523
2524         cn_reply->seq = req->seq;
2525         cn_reply->ack = req->ack  + 1;
2526         cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2527         cn_reply->flags = 0;
2528
2529         reply->packet_type = P_return_code_only;
2530         reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2531         reply->ret_code = ret_code;
2532
2533         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2534         if (rr && rr != -ESRCH)
2535                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2536 }
2537