SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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_wrappers.h"
  37 #include <asm/unaligned.h>
  38 #include <linux/drbd_tag_magic.h>
  39 #include <linux/drbd_limits.h>
  40
  41 static unsigned short *tl_add_blob(unsigned short *, enum drbd_tags, const void *, int);
  42 static unsigned short *tl_add_str(unsigned short *, enum drbd_tags, const char *);
  43 static unsigned short *tl_add_int(unsigned short *, enum drbd_tags, const void *);
  44
  45 /* see get_sb_bdev and bd_claim */
  46 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
  47
  48 /* Generate the tag_list to struct functions */
  49 #define NL_PACKET(name, number, fields) \
  50 static int name ## _from_tags(struct drbd_conf *mdev, \
  51         unsigned short *tags, struct name *arg) __attribute__ ((unused)); \
  52 static int name ## _from_tags(struct drbd_conf *mdev, \
  53         unsigned short *tags, struct name *arg) \
  54 { \
  55         int tag; \
  56         int dlen; \
  57         \
  58         while ((tag = get_unaligned(tags++)) != TT_END) {       \
  59                 dlen = get_unaligned(tags++);                   \
  60                 switch (tag_number(tag)) { \
  61                 fields \
  62                 default: \
  63                         if (tag & T_MANDATORY) { \
  64                                 dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \
  65                                 return 0; \
  66                         } \
  67                 } \
  68                 tags = (unsigned short *)((char *)tags + dlen); \
  69         } \
  70         return 1; \
  71 }
  72 #define NL_INTEGER(pn, pr, member) \
  73         case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \
  74                 arg->member = get_unaligned((int *)(tags));     \
  75                 break;
  76 #define NL_INT64(pn, pr, member) \
  77         case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \
  78                 arg->member = get_unaligned((u64 *)(tags));     \
  79                 break;
  80 #define NL_BIT(pn, pr, member) \
  81         case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \
  82                 arg->member = *(char *)(tags) ? 1 : 0; \
  83                 break;
  84 #define NL_STRING(pn, pr, member, len) \
  85         case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \
  86                 if (dlen > len) { \
  87                         dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \
  88                                 #member, dlen, (unsigned int)len); \
  89                         return 0; \
  90                 } \
  91                  arg->member ## _len = dlen; \
  92                  memcpy(arg->member, tags, min_t(size_t, dlen, len)); \
  93                  break;
  94 #include "linux/drbd_nl.h"
  95
  96 /* Generate the struct to tag_list functions */
  97 #define NL_PACKET(name, number, fields) \
  98 static unsigned short* \
  99 name ## _to_tags(struct drbd_conf *mdev, \
 100         struct name *arg, unsigned short *tags) __attribute__ ((unused)); \
 101 static unsigned short* \
 102 name ## _to_tags(struct drbd_conf *mdev, \
 103         struct name *arg, unsigned short *tags) \
 104 { \
 105         fields \
 106         return tags; \
 107 }
 108
 109 #define NL_INTEGER(pn, pr, member) \
 110         put_unaligned(pn | pr | TT_INTEGER, tags++);    \
 111         put_unaligned(sizeof(int), tags++);             \
 112         put_unaligned(arg->member, (int *)tags);        \
 113         tags = (unsigned short *)((char *)tags+sizeof(int));
 114 #define NL_INT64(pn, pr, member) \
 115         put_unaligned(pn | pr | TT_INT64, tags++);      \
 116         put_unaligned(sizeof(u64), tags++);             \
 117         put_unaligned(arg->member, (u64 *)tags);        \
 118         tags = (unsigned short *)((char *)tags+sizeof(u64));
 119 #define NL_BIT(pn, pr, member) \
 120         put_unaligned(pn | pr | TT_BIT, tags++);        \
 121         put_unaligned(sizeof(char), tags++);            \
 122         *(char *)tags = arg->member; \
 123         tags = (unsigned short *)((char *)tags+sizeof(char));
 124 #define NL_STRING(pn, pr, member, len) \
 125         put_unaligned(pn | pr | TT_STRING, tags++);     \
 126         put_unaligned(arg->member ## _len, tags++);     \
 127         memcpy(tags, arg->member, arg->member ## _len); \
 128         tags = (unsigned short *)((char *)tags + arg->member ## _len);
 129 #include "linux/drbd_nl.h"
 130
 131 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name);
 132 void drbd_nl_send_reply(struct cn_msg *, int);
 133
 134 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
 135 {
 136         char *envp[] = { "HOME=/",
 137                         "TERM=linux",
 138                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 139                         NULL, /* Will be set to address family */
 140                         NULL, /* Will be set to address */
 141                         NULL };
 142
 143         char mb[12], af[20], ad[60], *afs;
 144         char *argv[] = {usermode_helper, cmd, mb, NULL };
 145         int ret;
 146
 147         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 148
 149         if (get_net_conf(mdev)) {
 150                 switch (((struct sockaddr *)mdev->net_conf->peer_addr)->sa_family) {
 151                 case AF_INET6:
 152                         afs = "ipv6";
 153                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6",
 154                                  &((struct sockaddr_in6 *)mdev->net_conf->peer_addr)->sin6_addr);
 155                         break;
 156                 case AF_INET:
 157                         afs = "ipv4";
 158                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 159                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
 160                         break;
 161                 default:
 162                         afs = "ssocks";
 163                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 164                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
 165                 }
 166                 snprintf(af, 20, "DRBD_PEER_AF=%s", afs);
 167                 envp[3]=af;
 168                 envp[4]=ad;
 169                 put_net_conf(mdev);
 170         }
 171
 172         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 173
 174         drbd_bcast_ev_helper(mdev, cmd);
 175         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 176         if (ret)
 177                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 178                                 usermode_helper, cmd, mb,
 179                                 (ret >> 8) & 0xff, ret);
 180         else
 181                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 182                                 usermode_helper, cmd, mb,
 183                                 (ret >> 8) & 0xff, ret);
 184
 185         if (ret < 0) /* Ignore any ERRNOs we got. */
 186                 ret = 0;
 187
 188         return ret;
 189 }
 190
 191 enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
 192 {
 193         char *ex_to_string;
 194         int r;
 195         enum drbd_disk_state nps;
 196         enum drbd_fencing_p fp;
 197
 198         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 199
 200         if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 201                 fp = mdev->ldev->dc.fencing;
 202                 put_ldev(mdev);
 203         } else {
 204                 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n");
 205                 return mdev->state.pdsk;
 206         }
 207
 208         if (fp == FP_STONITH)
 209                 _drbd_request_state(mdev, NS(susp, 1), CS_WAIT_COMPLETE);
 210
 211         r = drbd_khelper(mdev, "fence-peer");
 212
 213         switch ((r>>8) & 0xff) {
 214         case 3: /* peer is inconsistent */
 215                 ex_to_string = "peer is inconsistent or worse";
 216                 nps = D_INCONSISTENT;
 217                 break;
 218         case 4: /* peer got outdated, or was already outdated */
 219                 ex_to_string = "peer was fenced";
 220                 nps = D_OUTDATED;
 221                 break;
 222         case 5: /* peer was down */
 223                 if (mdev->state.disk == D_UP_TO_DATE) {
 224                         /* we will(have) create(d) a new UUID anyways... */
 225                         ex_to_string = "peer is unreachable, assumed to be dead";
 226                         nps = D_OUTDATED;
 227                 } else {
 228                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 229                         nps = mdev->state.pdsk;
 230                 }
 231                 break;
 232         case 6: /* Peer is primary, voluntarily outdate myself.
 233                  * This is useful when an unconnected R_SECONDARY is asked to
 234                  * become R_PRIMARY, but finds the other peer being active. */
 235                 ex_to_string = "peer is active";
 236                 dev_warn(DEV, "Peer is primary, outdating myself.\n");
 237                 nps = D_UNKNOWN;
 238                 _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE);
 239                 break;
 240         case 7:
 241                 if (fp != FP_STONITH)
 242                         dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n");
 243                 ex_to_string = "peer was stonithed";
 244                 nps = D_OUTDATED;
 245                 break;
 246         default:
 247                 /* The script is broken ... */
 248                 nps = D_UNKNOWN;
 249                 dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 250                 return nps;
 251         }
 252
 253         dev_info(DEV, "fence-peer helper returned %d (%s)\n",
 254                         (r>>8) & 0xff, ex_to_string);
 255         return nps;
 256 }
 257
 258
 259 int drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 260 {
 261         const int max_tries = 4;
 262         int r = 0;
 263         int try = 0;
 264         int forced = 0;
 265         union drbd_state mask, val;
 266         enum drbd_disk_state nps;
 267
 268         if (new_role == R_PRIMARY)
 269                 request_ping(mdev); /* Detect a dead peer ASAP */
 270
 271         mutex_lock(&mdev->state_mutex);
 272
 273         mask.i = 0; mask.role = R_MASK;
 274         val.i  = 0; val.role  = new_role;
 275
 276         while (try++ < max_tries) {
 277                 r = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 278
 279                 /* in case we first succeeded to outdate,
 280                  * but now suddenly could establish a connection */
 281                 if (r == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 282                         val.pdsk = 0;
 283                         mask.pdsk = 0;
 284                         continue;
 285                 }
 286
 287                 if (r == SS_NO_UP_TO_DATE_DISK && force &&
 288                     (mdev->state.disk == D_INCONSISTENT ||
 289                      mdev->state.disk == D_OUTDATED)) {
 290                         mask.disk = D_MASK;
 291                         val.disk  = D_UP_TO_DATE;
 292                         forced = 1;
 293                         continue;
 294                 }
 295
 296                 if (r == SS_NO_UP_TO_DATE_DISK &&
 297                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 298                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 299                         nps = drbd_try_outdate_peer(mdev);
 300
 301                         if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
 302                                 val.disk = D_UP_TO_DATE;
 303                                 mask.disk = D_MASK;
 304                         }
 305
 306                         val.pdsk = nps;
 307                         mask.pdsk = D_MASK;
 308
 309                         continue;
 310                 }
 311
 312                 if (r == SS_NOTHING_TO_DO)
 313                         goto fail;
 314                 if (r == SS_PRIMARY_NOP && mask.pdsk == 0) {
 315                         nps = drbd_try_outdate_peer(mdev);
 316
 317                         if (force && nps > D_OUTDATED) {
 318                                 dev_warn(DEV, "Forced into split brain situation!\n");
 319                                 nps = D_OUTDATED;
 320                         }
 321
 322                         mask.pdsk = D_MASK;
 323                         val.pdsk  = nps;
 324
 325                         continue;
 326                 }
 327                 if (r == SS_TWO_PRIMARIES) {
 328                         /* Maybe the peer is detected as dead very soon...
 329                            retry at most once more in this case. */
 330                         __set_current_state(TASK_INTERRUPTIBLE);
 331                         schedule_timeout((mdev->net_conf->ping_timeo+1)*HZ/10);
 332                         if (try < max_tries)
 333                                 try = max_tries - 1;
 334                         continue;
 335                 }
 336                 if (r < SS_SUCCESS) {
 337                         r = _drbd_request_state(mdev, mask, val,
 338                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 339                         if (r < SS_SUCCESS)
 340                                 goto fail;
 341                 }
 342                 break;
 343         }
 344
 345         if (r < SS_SUCCESS)
 346                 goto fail;
 347
 348         if (forced)
 349                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 350
 351         /* Wait until nothing is on the fly :) */
 352         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 353
 354         if (new_role == R_SECONDARY) {
 355                 set_disk_ro(mdev->vdisk, TRUE);
 356                 if (get_ldev(mdev)) {
 357                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 358                         put_ldev(mdev);
 359                 }
 360         } else {
 361                 if (get_net_conf(mdev)) {
 362                         mdev->net_conf->want_lose = 0;
 363                         put_net_conf(mdev);
 364                 }
 365                 set_disk_ro(mdev->vdisk, FALSE);
 366                 if (get_ldev(mdev)) {
 367                         if (((mdev->state.conn < C_CONNECTED ||
 368                                mdev->state.pdsk <= D_FAILED)
 369                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 370                                 drbd_uuid_new_current(mdev);
 371
 372                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 373                         put_ldev(mdev);
 374                 }
 375         }
 376
 377         if ((new_role == R_SECONDARY) && get_ldev(mdev)) {
 378                 drbd_al_to_on_disk_bm(mdev);
 379                 put_ldev(mdev);
 380         }
 381
 382         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 383                 /* if this was forced, we should consider sync */
 384                 if (forced)
 385                         drbd_send_uuids(mdev);
 386                 drbd_send_state(mdev);
 387         }
 388
 389         drbd_md_sync(mdev);
 390
 391         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 392  fail:
 393         mutex_unlock(&mdev->state_mutex);
 394         return r;
 395 }
 396
 397
 398 static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 399                            struct drbd_nl_cfg_reply *reply)
 400 {
 401         struct primary primary_args;
 402
 403         memset(&primary_args, 0, sizeof(struct primary));
 404         if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) {
 405                 reply->ret_code = ERR_MANDATORY_TAG;
 406                 return 0;
 407         }
 408
 409         reply->ret_code =
 410                 drbd_set_role(mdev, R_PRIMARY, primary_args.overwrite_peer);
 411
 412         return 0;
 413 }
 414
 415 static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 416                              struct drbd_nl_cfg_reply *reply)
 417 {
 418         reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0);
 419
 420         return 0;
 421 }
 422
 423 /* initializes the md.*_offset members, so we are able to find
 424  * the on disk meta data */
 425 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 426                                        struct drbd_backing_dev *bdev)
 427 {
 428         sector_t md_size_sect = 0;
 429         switch (bdev->dc.meta_dev_idx) {
 430         default:
 431                 /* v07 style fixed size indexed meta data */
 432                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 433                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 434                 bdev->md.al_offset = MD_AL_OFFSET;
 435                 bdev->md.bm_offset = MD_BM_OFFSET;
 436                 break;
 437         case DRBD_MD_INDEX_FLEX_EXT:
 438                 /* just occupy the full device; unit: sectors */
 439                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 440                 bdev->md.md_offset = 0;
 441                 bdev->md.al_offset = MD_AL_OFFSET;
 442                 bdev->md.bm_offset = MD_BM_OFFSET;
 443                 break;
 444         case DRBD_MD_INDEX_INTERNAL:
 445         case DRBD_MD_INDEX_FLEX_INT:
 446                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 447                 /* al size is still fixed */
 448                 bdev->md.al_offset = -MD_AL_MAX_SIZE;
 449                 /* we need (slightly less than) ~ this much bitmap sectors: */
 450                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 451                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 452                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 453                 md_size_sect = ALIGN(md_size_sect, 8);
 454
 455                 /* plus the "drbd meta data super block",
 456                  * and the activity log; */
 457                 md_size_sect += MD_BM_OFFSET;
 458
 459                 bdev->md.md_size_sect = md_size_sect;
 460                 /* bitmap offset is adjusted by 'super' block size */
 461                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 462                 break;
 463         }
 464 }
 465
 466 char *ppsize(char *buf, unsigned long long size)
 467 {
 468         /* Needs 9 bytes at max. */
 469         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 470         int base = 0;
 471         while (size >= 10000) {
 472                 /* shift + round */
 473                 size = (size >> 10) + !!(size & (1<<9));
 474                 base++;
 475         }
 476         sprintf(buf, "%lu %cB", (long)size, units[base]);
 477
 478         return buf;
 479 }
 480
 481 /* there is still a theoretical deadlock when called from receiver
 482  * on an D_INCONSISTENT R_PRIMARY:
 483  *  remote READ does inc_ap_bio, receiver would need to receive answer
 484  *  packet from remote to dec_ap_bio again.
 485  *  receiver receive_sizes(), comes here,
 486  *  waits for ap_bio_cnt == 0. -> deadlock.
 487  * but this cannot happen, actually, because:
 488  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 489  *  (not connected, or bad/no disk on peer):
 490  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 491  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 492  *  peer may not initiate a resize.
 493  */
 494 void drbd_suspend_io(struct drbd_conf *mdev)
 495 {
 496         set_bit(SUSPEND_IO, &mdev->flags);
 497         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 498 }
 499
 500 void drbd_resume_io(struct drbd_conf *mdev)
 501 {
 502         clear_bit(SUSPEND_IO, &mdev->flags);
 503         wake_up(&mdev->misc_wait);
 504 }
 505
 506 /**
 507  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 508  * @mdev:       DRBD device.
 509  *
 510  * Returns 0 on success, negative return values indicate errors.
 511  * You should call drbd_md_sync() after calling this function.
 512  */
 513 enum determine_dev_size drbd_determin_dev_size(struct drbd_conf *mdev, int force) __must_hold(local)
 514 {
 515         sector_t prev_first_sect, prev_size; /* previous meta location */
 516         sector_t la_size;
 517         sector_t size;
 518         char ppb[10];
 519
 520         int md_moved, la_size_changed;
 521         enum determine_dev_size rv = unchanged;
 522
 523         /* race:
 524          * application request passes inc_ap_bio,
 525          * but then cannot get an AL-reference.
 526          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 527          *
 528          * to avoid that:
 529          * Suspend IO right here.
 530          * still lock the act_log to not trigger ASSERTs there.
 531          */
 532         drbd_suspend_io(mdev);
 533
 534         /* no wait necessary anymore, actually we could assert that */
 535         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 536
 537         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 538         prev_size = mdev->ldev->md.md_size_sect;
 539         la_size = mdev->ldev->md.la_size_sect;
 540
 541         /* TODO: should only be some assert here, not (re)init... */
 542         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 543
 544         size = drbd_new_dev_size(mdev, mdev->ldev, force);
 545
 546         if (drbd_get_capacity(mdev->this_bdev) != size ||
 547             drbd_bm_capacity(mdev) != size) {
 548                 int err;
 549                 err = drbd_bm_resize(mdev, size);
 550                 if (unlikely(err)) {
 551                         /* currently there is only one error: ENOMEM! */
 552                         size = drbd_bm_capacity(mdev)>>1;
 553                         if (size == 0) {
 554                                 dev_err(DEV, "OUT OF MEMORY! "
 555                                     "Could not allocate bitmap!\n");
 556                         } else {
 557                                 dev_err(DEV, "BM resizing failed. "
 558                                     "Leaving size unchanged at size = %lu KB\n",
 559                                     (unsigned long)size);
 560                         }
 561                         rv = dev_size_error;
 562                 }
 563                 /* racy, see comments above. */
 564                 drbd_set_my_capacity(mdev, size);
 565                 mdev->ldev->md.la_size_sect = size;
 566                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 567                      (unsigned long long)size>>1);
 568         }
 569         if (rv == dev_size_error)
 570                 goto out;
 571
 572         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 573
 574         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 575                 || prev_size       != mdev->ldev->md.md_size_sect;
 576
 577         if (la_size_changed || md_moved) {
 578                 drbd_al_shrink(mdev); /* All extents inactive. */
 579                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 580                          la_size_changed && md_moved ? "size changed and md moved" :
 581                          la_size_changed ? "size changed" : "md moved");
 582                 rv = drbd_bitmap_io(mdev, &drbd_bm_write, "size changed"); /* does drbd_resume_io() ! */
 583                 drbd_md_mark_dirty(mdev);
 584         }
 585
 586         if (size > la_size)
 587                 rv = grew;
 588         if (size < la_size)
 589                 rv = shrunk;
 590 out:
 591         lc_unlock(mdev->act_log);
 592         wake_up(&mdev->al_wait);
 593         drbd_resume_io(mdev);
 594
 595         return rv;
 596 }
 597
 598 sector_t
 599 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
 600 {
 601         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 602         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 603         sector_t m_size; /* my size */
 604         sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
 605         sector_t size = 0;
 606
 607         m_size = drbd_get_max_capacity(bdev);
 608
 609         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 610                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 611                 p_size = m_size;
 612         }
 613
 614         if (p_size && m_size) {
 615                 size = min_t(sector_t, p_size, m_size);
 616         } else {
 617                 if (la_size) {
 618                         size = la_size;
 619                         if (m_size && m_size < size)
 620                                 size = m_size;
 621                         if (p_size && p_size < size)
 622                                 size = p_size;
 623                 } else {
 624                         if (m_size)
 625                                 size = m_size;
 626                         if (p_size)
 627                                 size = p_size;
 628                 }
 629         }
 630
 631         if (size == 0)
 632                 dev_err(DEV, "Both nodes diskless!\n");
 633
 634         if (u_size) {
 635                 if (u_size > size)
 636                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 637                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 638                 else
 639                         size = u_size;
 640         }
 641
 642         return size;
 643 }
 644
 645 /**
 646  * drbd_check_al_size() - Ensures that the AL is of the right size
 647  * @mdev:       DRBD device.
 648  *
 649  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 650  * failed, and 0 on success. You should call drbd_md_sync() after you called
 651  * this function.
 652  */
 653 static int drbd_check_al_size(struct drbd_conf *mdev)
 654 {
 655         struct lru_cache *n, *t;
 656         struct lc_element *e;
 657         unsigned int in_use;
 658         int i;
 659
 660         ERR_IF(mdev->sync_conf.al_extents < 7)
 661                 mdev->sync_conf.al_extents = 127;
 662
 663         if (mdev->act_log &&
 664             mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
 665                 return 0;
 666
 667         in_use = 0;
 668         t = mdev->act_log;
 669         n = lc_create("act_log", drbd_al_ext_cache,
 670                 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
 671
 672         if (n == NULL) {
 673                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 674                 return -ENOMEM;
 675         }
 676         spin_lock_irq(&mdev->al_lock);
 677         if (t) {
 678                 for (i = 0; i < t->nr_elements; i++) {
 679                         e = lc_element_by_index(t, i);
 680                         if (e->refcnt)
 681                                 dev_err(DEV, "refcnt(%d)==%d\n",
 682                                     e->lc_number, e->refcnt);
 683                         in_use += e->refcnt;
 684                 }
 685         }
 686         if (!in_use)
 687                 mdev->act_log = n;
 688         spin_unlock_irq(&mdev->al_lock);
 689         if (in_use) {
 690                 dev_err(DEV, "Activity log still in use!\n");
 691                 lc_destroy(n);
 692                 return -EBUSY;
 693         } else {
 694                 if (t)
 695                         lc_destroy(t);
 696         }
 697         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 698         return 0;
 699 }
 700
 701 void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_seg_s) __must_hold(local)
 702 {
 703         struct request_queue * const q = mdev->rq_queue;
 704         struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 705         int max_segments = mdev->ldev->dc.max_bio_bvecs;
 706
 707         if (b->merge_bvec_fn && !mdev->ldev->dc.use_bmbv)
 708                 max_seg_s = PAGE_SIZE;
 709
 710         max_seg_s = min(queue_max_sectors(b) * queue_logical_block_size(b), max_seg_s);
 711
 712         blk_queue_max_sectors(q, max_seg_s >> 9);
 713         blk_queue_max_phys_segments(q, max_segments ? max_segments : MAX_PHYS_SEGMENTS);
 714         blk_queue_max_hw_segments(q, max_segments ? max_segments : MAX_HW_SEGMENTS);
 715         blk_queue_max_segment_size(q, max_seg_s);
 716         blk_queue_logical_block_size(q, 512);
 717         blk_queue_segment_boundary(q, PAGE_SIZE-1);
 718         blk_stack_limits(&q->limits, &b->limits, 0);
 719
 720         if (b->merge_bvec_fn)
 721                 dev_warn(DEV, "Backing device's merge_bvec_fn() = %p\n",
 722                      b->merge_bvec_fn);
 723         dev_info(DEV, "max_segment_size ( = BIO size ) = %u\n", queue_max_segment_size(q));
 724
 725         if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
 726                 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
 727                      q->backing_dev_info.ra_pages,
 728                      b->backing_dev_info.ra_pages);
 729                 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
 730         }
 731 }
 732
 733 /* serialize deconfig (worker exiting, doing cleanup)
 734  * and reconfig (drbdsetup disk, drbdsetup net)
 735  *
 736  * wait for a potentially exiting worker, then restart it,
 737  * or start a new one.
 738  */
 739 static void drbd_reconfig_start(struct drbd_conf *mdev)
 740 {
 741         wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
 742         wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
 743         drbd_thread_start(&mdev->worker);
 744 }
 745
 746 /* if still unconfigured, stops worker again.
 747  * if configured now, clears CONFIG_PENDING.
 748  * wakes potential waiters */
 749 static void drbd_reconfig_done(struct drbd_conf *mdev)
 750 {
 751         spin_lock_irq(&mdev->req_lock);
 752         if (mdev->state.disk == D_DISKLESS &&
 753             mdev->state.conn == C_STANDALONE &&
 754             mdev->state.role == R_SECONDARY) {
 755                 set_bit(DEVICE_DYING, &mdev->flags);
 756                 drbd_thread_stop_nowait(&mdev->worker);
 757         } else
 758                 clear_bit(CONFIG_PENDING, &mdev->flags);
 759         spin_unlock_irq(&mdev->req_lock);
 760         wake_up(&mdev->state_wait);
 761 }
 762
 763 /* does always return 0;
 764  * interesting return code is in reply->ret_code */
 765 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 766                              struct drbd_nl_cfg_reply *reply)
 767 {
 768         enum drbd_ret_codes retcode;
 769         enum determine_dev_size dd;
 770         sector_t max_possible_sectors;
 771         sector_t min_md_device_sectors;
 772         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
 773         struct inode *inode, *inode2;
 774         struct lru_cache *resync_lru = NULL;
 775         union drbd_state ns, os;
 776         int rv;
 777         int cp_discovered = 0;
 778         int logical_block_size;
 779
 780         drbd_reconfig_start(mdev);
 781
 782         /* if you want to reconfigure, please tear down first */
 783         if (mdev->state.disk > D_DISKLESS) {
 784                 retcode = ERR_DISK_CONFIGURED;
 785                 goto fail;
 786         }
 787
 788         /* allocation not in the IO path, cqueue thread context */
 789         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
 790         if (!nbc) {
 791                 retcode = ERR_NOMEM;
 792                 goto fail;
 793         }
 794
 795         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
 796         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
 797         nbc->dc.fencing       = DRBD_FENCING_DEF;
 798         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
 799
 800         if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
 801                 retcode = ERR_MANDATORY_TAG;
 802                 goto fail;
 803         }
 804
 805         if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
 806                 retcode = ERR_MD_IDX_INVALID;
 807                 goto fail;
 808         }
 809
 810         nbc->lo_file = filp_open(nbc->dc.backing_dev, O_RDWR, 0);
 811         if (IS_ERR(nbc->lo_file)) {
 812                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
 813                     PTR_ERR(nbc->lo_file));
 814                 nbc->lo_file = NULL;
 815                 retcode = ERR_OPEN_DISK;
 816                 goto fail;
 817         }
 818
 819         inode = nbc->lo_file->f_dentry->d_inode;
 820
 821         if (!S_ISBLK(inode->i_mode)) {
 822                 retcode = ERR_DISK_NOT_BDEV;
 823                 goto fail;
 824         }
 825
 826         nbc->md_file = filp_open(nbc->dc.meta_dev, O_RDWR, 0);
 827         if (IS_ERR(nbc->md_file)) {
 828                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
 829                     PTR_ERR(nbc->md_file));
 830                 nbc->md_file = NULL;
 831                 retcode = ERR_OPEN_MD_DISK;
 832                 goto fail;
 833         }
 834
 835         inode2 = nbc->md_file->f_dentry->d_inode;
 836
 837         if (!S_ISBLK(inode2->i_mode)) {
 838                 retcode = ERR_MD_NOT_BDEV;
 839                 goto fail;
 840         }
 841
 842         nbc->backing_bdev = inode->i_bdev;
 843         if (bd_claim(nbc->backing_bdev, mdev)) {
 844                 printk(KERN_ERR "drbd: bd_claim(%p,%p); failed [%p;%p;%u]\n",
 845                        nbc->backing_bdev, mdev,
 846                        nbc->backing_bdev->bd_holder,
 847                        nbc->backing_bdev->bd_contains->bd_holder,
 848                        nbc->backing_bdev->bd_holders);
 849                 retcode = ERR_BDCLAIM_DISK;
 850                 goto fail;
 851         }
 852
 853         resync_lru = lc_create("resync", drbd_bm_ext_cache,
 854                         61, sizeof(struct bm_extent),
 855                         offsetof(struct bm_extent, lce));
 856         if (!resync_lru) {
 857                 retcode = ERR_NOMEM;
 858                 goto release_bdev_fail;
 859         }
 860
 861         /* meta_dev_idx >= 0: external fixed size,
 862          * possibly multiple drbd sharing one meta device.
 863          * TODO in that case, paranoia check that [md_bdev, meta_dev_idx] is
 864          * not yet used by some other drbd minor!
 865          * (if you use drbd.conf + drbdadm,
 866          * that should check it for you already; but if you don't, or someone
 867          * fooled it, we need to double check here) */
 868         nbc->md_bdev = inode2->i_bdev;
 869         if (bd_claim(nbc->md_bdev, (nbc->dc.meta_dev_idx < 0) ? (void *)mdev
 870                                 : (void *) drbd_m_holder)) {
 871                 retcode = ERR_BDCLAIM_MD_DISK;
 872                 goto release_bdev_fail;
 873         }
 874
 875         if ((nbc->backing_bdev == nbc->md_bdev) !=
 876             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
 877              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
 878                 retcode = ERR_MD_IDX_INVALID;
 879                 goto release_bdev2_fail;
 880         }
 881
 882         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
 883         drbd_md_set_sector_offsets(mdev, nbc);
 884
 885         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
 886                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
 887                         (unsigned long long) drbd_get_max_capacity(nbc),
 888                         (unsigned long long) nbc->dc.disk_size);
 889                 retcode = ERR_DISK_TO_SMALL;
 890                 goto release_bdev2_fail;
 891         }
 892
 893         if (nbc->dc.meta_dev_idx < 0) {
 894                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
 895                 /* at least one MB, otherwise it does not make sense */
 896                 min_md_device_sectors = (2<<10);
 897         } else {
 898                 max_possible_sectors = DRBD_MAX_SECTORS;
 899                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
 900         }
 901
 902         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
 903                 retcode = ERR_MD_DISK_TO_SMALL;
 904                 dev_warn(DEV, "refusing attach: md-device too small, "
 905                      "at least %llu sectors needed for this meta-disk type\n",
 906                      (unsigned long long) min_md_device_sectors);
 907                 goto release_bdev2_fail;
 908         }
 909
 910         /* Make sure the new disk is big enough
 911          * (we may currently be R_PRIMARY with no local disk...) */
 912         if (drbd_get_max_capacity(nbc) <
 913             drbd_get_capacity(mdev->this_bdev)) {
 914                 retcode = ERR_DISK_TO_SMALL;
 915                 goto release_bdev2_fail;
 916         }
 917
 918         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
 919
 920         if (nbc->known_size > max_possible_sectors) {
 921                 dev_warn(DEV, "==> truncating very big lower level device "
 922                         "to currently maximum possible %llu sectors <==\n",
 923                         (unsigned long long) max_possible_sectors);
 924                 if (nbc->dc.meta_dev_idx >= 0)
 925                         dev_warn(DEV, "==>> using internal or flexible "
 926                                       "meta data may help <<==\n");
 927         }
 928
 929         drbd_suspend_io(mdev);
 930         /* also wait for the last barrier ack. */
 931         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt));
 932         /* and for any other previously queued work */
 933         drbd_flush_workqueue(mdev);
 934
 935         retcode = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
 936         drbd_resume_io(mdev);
 937         if (retcode < SS_SUCCESS)
 938                 goto release_bdev2_fail;
 939
 940         if (!get_ldev_if_state(mdev, D_ATTACHING))
 941                 goto force_diskless;
 942
 943         drbd_md_set_sector_offsets(mdev, nbc);
 944
 945         if (!mdev->bitmap) {
 946                 if (drbd_bm_init(mdev)) {
 947                         retcode = ERR_NOMEM;
 948                         goto force_diskless_dec;
 949                 }
 950         }
 951
 952         retcode = drbd_md_read(mdev, nbc);
 953         if (retcode != NO_ERROR)
 954                 goto force_diskless_dec;
 955
 956         if (mdev->state.conn < C_CONNECTED &&
 957             mdev->state.role == R_PRIMARY &&
 958             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
 959                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
 960                     (unsigned long long)mdev->ed_uuid);
 961                 retcode = ERR_DATA_NOT_CURRENT;
 962                 goto force_diskless_dec;
 963         }
 964
 965         /* Since we are diskless, fix the activity log first... */
 966         if (drbd_check_al_size(mdev)) {
 967                 retcode = ERR_NOMEM;
 968                 goto force_diskless_dec;
 969         }
 970
 971         /* Prevent shrinking of consistent devices ! */
 972         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
 973             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
 974                 dev_warn(DEV, "refusing to truncate a consistent device\n");
 975                 retcode = ERR_DISK_TO_SMALL;
 976                 goto force_diskless_dec;
 977         }
 978
 979         if (!drbd_al_read_log(mdev, nbc)) {
 980                 retcode = ERR_IO_MD_DISK;
 981                 goto force_diskless_dec;
 982         }
 983
 984         /* allocate a second IO page if logical_block_size != 512 */
 985         logical_block_size = bdev_logical_block_size(nbc->md_bdev);
 986         if (logical_block_size == 0)
 987                 logical_block_size = MD_SECTOR_SIZE;
 988
 989         if (logical_block_size != MD_SECTOR_SIZE) {
 990                 if (!mdev->md_io_tmpp) {
 991                         struct page *page = alloc_page(GFP_NOIO);
 992                         if (!page)
 993                                 goto force_diskless_dec;
 994
 995                         dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
 996                              logical_block_size, MD_SECTOR_SIZE);
 997                         dev_warn(DEV, "Workaround engaged (has performance impact).\n");
 998
 999                         mdev->md_io_tmpp = page;
1000                 }
1001         }
1002
1003         /* Reset the "barriers don't work" bits here, then force meta data to
1004          * be written, to ensure we determine if barriers are supported. */
1005         if (nbc->dc.no_md_flush)
1006                 set_bit(MD_NO_BARRIER, &mdev->flags);
1007         else
1008                 clear_bit(MD_NO_BARRIER, &mdev->flags);
1009
1010         /* Point of no return reached.
1011          * Devices and memory are no longer released by error cleanup below.
1012          * now mdev takes over responsibility, and the state engine should
1013          * clean it up somewhere.  */
1014         D_ASSERT(mdev->ldev == NULL);
1015         mdev->ldev = nbc;
1016         mdev->resync = resync_lru;
1017         nbc = NULL;
1018         resync_lru = NULL;
1019
1020         mdev->write_ordering = WO_bio_barrier;
1021         drbd_bump_write_ordering(mdev, WO_bio_barrier);
1022
1023         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1024                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1025         else
1026                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1027
1028         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND)) {
1029                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1030                 cp_discovered = 1;
1031         }
1032
1033         mdev->send_cnt = 0;
1034         mdev->recv_cnt = 0;
1035         mdev->read_cnt = 0;
1036         mdev->writ_cnt = 0;
1037
1038         drbd_setup_queue_param(mdev, DRBD_MAX_SEGMENT_SIZE);
1039
1040         /* If I am currently not R_PRIMARY,
1041          * but meta data primary indicator is set,
1042          * I just now recover from a hard crash,
1043          * and have been R_PRIMARY before that crash.
1044          *
1045          * Now, if I had no connection before that crash
1046          * (have been degraded R_PRIMARY), chances are that
1047          * I won't find my peer now either.
1048          *
1049          * In that case, and _only_ in that case,
1050          * we use the degr-wfc-timeout instead of the default,
1051          * so we can automatically recover from a crash of a
1052          * degraded but active "cluster" after a certain timeout.
1053          */
1054         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1055         if (mdev->state.role != R_PRIMARY &&
1056              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1057             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1058                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1059
1060         dd = drbd_determin_dev_size(mdev, 0);
1061         if (dd == dev_size_error) {
1062                 retcode = ERR_NOMEM_BITMAP;
1063                 goto force_diskless_dec;
1064         } else if (dd == grew)
1065                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1066
1067         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1068                 dev_info(DEV, "Assuming that all blocks are out of sync "
1069                      "(aka FullSync)\n");
1070                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from attaching")) {
1071                         retcode = ERR_IO_MD_DISK;
1072                         goto force_diskless_dec;
1073                 }
1074         } else {
1075                 if (drbd_bitmap_io(mdev, &drbd_bm_read, "read from attaching") < 0) {
1076                         retcode = ERR_IO_MD_DISK;
1077                         goto force_diskless_dec;
1078                 }
1079         }
1080
1081         if (cp_discovered) {
1082                 drbd_al_apply_to_bm(mdev);
1083                 drbd_al_to_on_disk_bm(mdev);
1084         }
1085
1086         spin_lock_irq(&mdev->req_lock);
1087         os = mdev->state;
1088         ns.i = os.i;
1089         /* If MDF_CONSISTENT is not set go into inconsistent state,
1090            otherwise investigate MDF_WasUpToDate...
1091            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1092            otherwise into D_CONSISTENT state.
1093         */
1094         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1095                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1096                         ns.disk = D_CONSISTENT;
1097                 else
1098                         ns.disk = D_OUTDATED;
1099         } else {
1100                 ns.disk = D_INCONSISTENT;
1101         }
1102
1103         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1104                 ns.pdsk = D_OUTDATED;
1105
1106         if ( ns.disk == D_CONSISTENT &&
1107             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1108                 ns.disk = D_UP_TO_DATE;
1109
1110         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1111            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1112            this point, because drbd_request_state() modifies these
1113            flags. */
1114
1115         /* In case we are C_CONNECTED postpone any decision on the new disk
1116            state after the negotiation phase. */
1117         if (mdev->state.conn == C_CONNECTED) {
1118                 mdev->new_state_tmp.i = ns.i;
1119                 ns.i = os.i;
1120                 ns.disk = D_NEGOTIATING;
1121         }
1122
1123         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1124         ns = mdev->state;
1125         spin_unlock_irq(&mdev->req_lock);
1126
1127         if (rv < SS_SUCCESS)
1128                 goto force_diskless_dec;
1129
1130         if (mdev->state.role == R_PRIMARY)
1131                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1132         else
1133                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1134
1135         drbd_md_mark_dirty(mdev);
1136         drbd_md_sync(mdev);
1137
1138         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1139         put_ldev(mdev);
1140         reply->ret_code = retcode;
1141         drbd_reconfig_done(mdev);
1142         return 0;
1143
1144  force_diskless_dec:
1145         put_ldev(mdev);
1146  force_diskless:
1147         drbd_force_state(mdev, NS(disk, D_DISKLESS));
1148         drbd_md_sync(mdev);
1149  release_bdev2_fail:
1150         if (nbc)
1151                 bd_release(nbc->md_bdev);
1152  release_bdev_fail:
1153         if (nbc)
1154                 bd_release(nbc->backing_bdev);
1155  fail:
1156         if (nbc) {
1157                 if (nbc->lo_file)
1158                         fput(nbc->lo_file);
1159                 if (nbc->md_file)
1160                         fput(nbc->md_file);
1161                 kfree(nbc);
1162         }
1163         lc_destroy(resync_lru);
1164
1165         reply->ret_code = retcode;
1166         drbd_reconfig_done(mdev);
1167         return 0;
1168 }
1169
1170 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1171                           struct drbd_nl_cfg_reply *reply)
1172 {
1173         reply->ret_code = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1174         return 0;
1175 }
1176
1177 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1178                             struct drbd_nl_cfg_reply *reply)
1179 {
1180         int i, ns;
1181         enum drbd_ret_codes retcode;
1182         struct net_conf *new_conf = NULL;
1183         struct crypto_hash *tfm = NULL;
1184         struct crypto_hash *integrity_w_tfm = NULL;
1185         struct crypto_hash *integrity_r_tfm = NULL;
1186         struct hlist_head *new_tl_hash = NULL;
1187         struct hlist_head *new_ee_hash = NULL;
1188         struct drbd_conf *odev;
1189         char hmac_name[CRYPTO_MAX_ALG_NAME];
1190         void *int_dig_out = NULL;
1191         void *int_dig_in = NULL;
1192         void *int_dig_vv = NULL;
1193         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1194
1195         drbd_reconfig_start(mdev);
1196
1197         if (mdev->state.conn > C_STANDALONE) {
1198                 retcode = ERR_NET_CONFIGURED;
1199                 goto fail;
1200         }
1201
1202         /* allocation not in the IO path, cqueue thread context */
1203         new_conf = kmalloc(sizeof(struct net_conf), GFP_KERNEL);
1204         if (!new_conf) {
1205                 retcode = ERR_NOMEM;
1206                 goto fail;
1207         }
1208
1209         memset(new_conf, 0, sizeof(struct net_conf));
1210         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1211         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1212         new_conf->ping_int         = DRBD_PING_INT_DEF;
1213         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1214         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1215         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1216         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1217         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1218         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1219         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1220         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1221         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1222         new_conf->want_lose        = 0;
1223         new_conf->two_primaries    = 0;
1224         new_conf->wire_protocol    = DRBD_PROT_C;
1225         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1226         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1227
1228         if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1229                 retcode = ERR_MANDATORY_TAG;
1230                 goto fail;
1231         }
1232
1233         if (new_conf->two_primaries
1234             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1235                 retcode = ERR_NOT_PROTO_C;
1236                 goto fail;
1237         };
1238
1239         if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1240                 retcode = ERR_DISCARD;
1241                 goto fail;
1242         }
1243
1244         retcode = NO_ERROR;
1245
1246         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1247         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1248         for (i = 0; i < minor_count; i++) {
1249                 odev = minor_to_mdev(i);
1250                 if (!odev || odev == mdev)
1251                         continue;
1252                 if (get_net_conf(odev)) {
1253                         taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1254                         if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1255                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1256                                 retcode = ERR_LOCAL_ADDR;
1257
1258                         taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1259                         if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1260                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1261                                 retcode = ERR_PEER_ADDR;
1262
1263                         put_net_conf(odev);
1264                         if (retcode != NO_ERROR)
1265                                 goto fail;
1266                 }
1267         }
1268
1269         if (new_conf->cram_hmac_alg[0] != 0) {
1270                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1271                         new_conf->cram_hmac_alg);
1272                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1273                 if (IS_ERR(tfm)) {
1274                         tfm = NULL;
1275                         retcode = ERR_AUTH_ALG;
1276                         goto fail;
1277                 }
1278
1279                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1280                         retcode = ERR_AUTH_ALG_ND;
1281                         goto fail;
1282                 }
1283         }
1284
1285         if (new_conf->integrity_alg[0]) {
1286                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1287                 if (IS_ERR(integrity_w_tfm)) {
1288                         integrity_w_tfm = NULL;
1289                         retcode=ERR_INTEGRITY_ALG;
1290                         goto fail;
1291                 }
1292
1293                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1294                         retcode=ERR_INTEGRITY_ALG_ND;
1295                         goto fail;
1296                 }
1297
1298                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1299                 if (IS_ERR(integrity_r_tfm)) {
1300                         integrity_r_tfm = NULL;
1301                         retcode=ERR_INTEGRITY_ALG;
1302                         goto fail;
1303                 }
1304         }
1305
1306         ns = new_conf->max_epoch_size/8;
1307         if (mdev->tl_hash_s != ns) {
1308                 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1309                 if (!new_tl_hash) {
1310                         retcode = ERR_NOMEM;
1311                         goto fail;
1312                 }
1313         }
1314
1315         ns = new_conf->max_buffers/8;
1316         if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1317                 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1318                 if (!new_ee_hash) {
1319                         retcode = ERR_NOMEM;
1320                         goto fail;
1321                 }
1322         }
1323
1324         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1325
1326         if (integrity_w_tfm) {
1327                 i = crypto_hash_digestsize(integrity_w_tfm);
1328                 int_dig_out = kmalloc(i, GFP_KERNEL);
1329                 if (!int_dig_out) {
1330                         retcode = ERR_NOMEM;
1331                         goto fail;
1332                 }
1333                 int_dig_in = kmalloc(i, GFP_KERNEL);
1334                 if (!int_dig_in) {
1335                         retcode = ERR_NOMEM;
1336                         goto fail;
1337                 }
1338                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1339                 if (!int_dig_vv) {
1340                         retcode = ERR_NOMEM;
1341                         goto fail;
1342                 }
1343         }
1344
1345         if (!mdev->bitmap) {
1346                 if(drbd_bm_init(mdev)) {
1347                         retcode = ERR_NOMEM;
1348                         goto fail;
1349                 }
1350         }
1351
1352         spin_lock_irq(&mdev->req_lock);
1353         if (mdev->net_conf != NULL) {
1354                 retcode = ERR_NET_CONFIGURED;
1355                 spin_unlock_irq(&mdev->req_lock);
1356                 goto fail;
1357         }
1358         mdev->net_conf = new_conf;
1359
1360         mdev->send_cnt = 0;
1361         mdev->recv_cnt = 0;
1362
1363         if (new_tl_hash) {
1364                 kfree(mdev->tl_hash);
1365                 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1366                 mdev->tl_hash = new_tl_hash;
1367         }
1368
1369         if (new_ee_hash) {
1370                 kfree(mdev->ee_hash);
1371                 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1372                 mdev->ee_hash = new_ee_hash;
1373         }
1374
1375         crypto_free_hash(mdev->cram_hmac_tfm);
1376         mdev->cram_hmac_tfm = tfm;
1377
1378         crypto_free_hash(mdev->integrity_w_tfm);
1379         mdev->integrity_w_tfm = integrity_w_tfm;
1380
1381         crypto_free_hash(mdev->integrity_r_tfm);
1382         mdev->integrity_r_tfm = integrity_r_tfm;
1383
1384         kfree(mdev->int_dig_out);
1385         kfree(mdev->int_dig_in);
1386         kfree(mdev->int_dig_vv);
1387         mdev->int_dig_out=int_dig_out;
1388         mdev->int_dig_in=int_dig_in;
1389         mdev->int_dig_vv=int_dig_vv;
1390         spin_unlock_irq(&mdev->req_lock);
1391
1392         retcode = _drbd_request_state(mdev, NS(conn, C_UNCONNECTED), CS_VERBOSE);
1393
1394         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1395         reply->ret_code = retcode;
1396         drbd_reconfig_done(mdev);
1397         return 0;
1398
1399 fail:
1400         kfree(int_dig_out);
1401         kfree(int_dig_in);
1402         kfree(int_dig_vv);
1403         crypto_free_hash(tfm);
1404         crypto_free_hash(integrity_w_tfm);
1405         crypto_free_hash(integrity_r_tfm);
1406         kfree(new_tl_hash);
1407         kfree(new_ee_hash);
1408         kfree(new_conf);
1409
1410         reply->ret_code = retcode;
1411         drbd_reconfig_done(mdev);
1412         return 0;
1413 }
1414
1415 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1416                               struct drbd_nl_cfg_reply *reply)
1417 {
1418         int retcode;
1419
1420         retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1421
1422         if (retcode == SS_NOTHING_TO_DO)
1423                 goto done;
1424         else if (retcode == SS_ALREADY_STANDALONE)
1425                 goto done;
1426         else if (retcode == SS_PRIMARY_NOP) {
1427                 /* Our statche checking code wants to see the peer outdated. */
1428                 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1429                                                       pdsk, D_OUTDATED));
1430         } else if (retcode == SS_CW_FAILED_BY_PEER) {
1431                 /* The peer probably wants to see us outdated. */
1432                 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1433                                                         disk, D_OUTDATED),
1434                                               CS_ORDERED);
1435                 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1436                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1437                         retcode = SS_SUCCESS;
1438                 }
1439         }
1440
1441         if (retcode < SS_SUCCESS)
1442                 goto fail;
1443
1444         if (wait_event_interruptible(mdev->state_wait,
1445                                      mdev->state.conn != C_DISCONNECTING)) {
1446                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1447                    someone else might connect us in the mean time! */
1448                 retcode = ERR_INTR;
1449                 goto fail;
1450         }
1451
1452  done:
1453         retcode = NO_ERROR;
1454  fail:
1455         drbd_md_sync(mdev);
1456         reply->ret_code = retcode;
1457         return 0;
1458 }
1459
1460 void resync_after_online_grow(struct drbd_conf *mdev)
1461 {
1462         int iass; /* I am sync source */
1463
1464         dev_info(DEV, "Resync of new storage after online grow\n");
1465         if (mdev->state.role != mdev->state.peer)
1466                 iass = (mdev->state.role == R_PRIMARY);
1467         else
1468                 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1469
1470         if (iass)
1471                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1472         else
1473                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1474 }
1475
1476 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1477                           struct drbd_nl_cfg_reply *reply)
1478 {
1479         struct resize rs;
1480         int retcode = NO_ERROR;
1481         int ldsc = 0; /* local disk size changed */
1482         enum determine_dev_size dd;
1483
1484         memset(&rs, 0, sizeof(struct resize));
1485         if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1486                 retcode = ERR_MANDATORY_TAG;
1487                 goto fail;
1488         }
1489
1490         if (mdev->state.conn > C_CONNECTED) {
1491                 retcode = ERR_RESIZE_RESYNC;
1492                 goto fail;
1493         }
1494
1495         if (mdev->state.role == R_SECONDARY &&
1496             mdev->state.peer == R_SECONDARY) {
1497                 retcode = ERR_NO_PRIMARY;
1498                 goto fail;
1499         }
1500
1501         if (!get_ldev(mdev)) {
1502                 retcode = ERR_NO_DISK;
1503                 goto fail;
1504         }
1505
1506         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) {
1507                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1508                 ldsc = 1;
1509         }
1510
1511         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1512         dd = drbd_determin_dev_size(mdev, rs.resize_force);
1513         drbd_md_sync(mdev);
1514         put_ldev(mdev);
1515         if (dd == dev_size_error) {
1516                 retcode = ERR_NOMEM_BITMAP;
1517                 goto fail;
1518         }
1519
1520         if (mdev->state.conn == C_CONNECTED && (dd != unchanged || ldsc)) {
1521                 if (dd == grew)
1522                         set_bit(RESIZE_PENDING, &mdev->flags);
1523
1524                 drbd_send_uuids(mdev);
1525                 drbd_send_sizes(mdev, 1);
1526         }
1527
1528  fail:
1529         reply->ret_code = retcode;
1530         return 0;
1531 }
1532
1533 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1534                                struct drbd_nl_cfg_reply *reply)
1535 {
1536         int retcode = NO_ERROR;
1537         int err;
1538         int ovr; /* online verify running */
1539         int rsr; /* re-sync running */
1540         struct crypto_hash *verify_tfm = NULL;
1541         struct crypto_hash *csums_tfm = NULL;
1542         struct syncer_conf sc;
1543         cpumask_var_t new_cpu_mask;
1544
1545         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1546                 retcode = ERR_NOMEM;
1547                 goto fail;
1548         }
1549
1550         if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1551                 memset(&sc, 0, sizeof(struct syncer_conf));
1552                 sc.rate       = DRBD_RATE_DEF;
1553                 sc.after      = DRBD_AFTER_DEF;
1554                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1555         } else
1556                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1557
1558         if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1559                 retcode = ERR_MANDATORY_TAG;
1560                 goto fail;
1561         }
1562
1563         /* re-sync running */
1564         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1565                 mdev->state.conn == C_SYNC_TARGET ||
1566                 mdev->state.conn == C_PAUSED_SYNC_S ||
1567                 mdev->state.conn == C_PAUSED_SYNC_T );
1568
1569         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1570                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1571                 goto fail;
1572         }
1573
1574         if (!rsr && sc.csums_alg[0]) {
1575                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1576                 if (IS_ERR(csums_tfm)) {
1577                         csums_tfm = NULL;
1578                         retcode = ERR_CSUMS_ALG;
1579                         goto fail;
1580                 }
1581
1582                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1583                         retcode = ERR_CSUMS_ALG_ND;
1584                         goto fail;
1585                 }
1586         }
1587
1588         /* online verify running */
1589         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1590
1591         if (ovr) {
1592                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1593                         retcode = ERR_VERIFY_RUNNING;
1594                         goto fail;
1595                 }
1596         }
1597
1598         if (!ovr && sc.verify_alg[0]) {
1599                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1600                 if (IS_ERR(verify_tfm)) {
1601                         verify_tfm = NULL;
1602                         retcode = ERR_VERIFY_ALG;
1603                         goto fail;
1604                 }
1605
1606                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1607                         retcode = ERR_VERIFY_ALG_ND;
1608                         goto fail;
1609                 }
1610         }
1611
1612         /* silently ignore cpu mask on UP kernel */
1613         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1614                 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1615                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1616                 if (err) {
1617                         dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1618                         retcode = ERR_CPU_MASK_PARSE;
1619                         goto fail;
1620                 }
1621         }
1622
1623         ERR_IF (sc.rate < 1) sc.rate = 1;
1624         ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1625 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1626         if (sc.al_extents > AL_MAX) {
1627                 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1628                 sc.al_extents = AL_MAX;
1629         }
1630 #undef AL_MAX
1631
1632         /* most sanity checks done, try to assign the new sync-after
1633          * dependency.  need to hold the global lock in there,
1634          * to avoid a race in the dependency loop check. */
1635         retcode = drbd_alter_sa(mdev, sc.after);
1636         if (retcode != NO_ERROR)
1637                 goto fail;
1638
1639         /* ok, assign the rest of it as well.
1640          * lock against receive_SyncParam() */
1641         spin_lock(&mdev->peer_seq_lock);
1642         mdev->sync_conf = sc;
1643
1644         if (!rsr) {
1645                 crypto_free_hash(mdev->csums_tfm);
1646                 mdev->csums_tfm = csums_tfm;
1647                 csums_tfm = NULL;
1648         }
1649
1650         if (!ovr) {
1651                 crypto_free_hash(mdev->verify_tfm);
1652                 mdev->verify_tfm = verify_tfm;
1653                 verify_tfm = NULL;
1654         }
1655         spin_unlock(&mdev->peer_seq_lock);
1656
1657         if (get_ldev(mdev)) {
1658                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1659                 drbd_al_shrink(mdev);
1660                 err = drbd_check_al_size(mdev);
1661                 lc_unlock(mdev->act_log);
1662                 wake_up(&mdev->al_wait);
1663
1664                 put_ldev(mdev);
1665                 drbd_md_sync(mdev);
1666
1667                 if (err) {
1668                         retcode = ERR_NOMEM;
1669                         goto fail;
1670                 }
1671         }
1672
1673         if (mdev->state.conn >= C_CONNECTED)
1674                 drbd_send_sync_param(mdev, &sc);
1675
1676         if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1677                 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1678                 drbd_calc_cpu_mask(mdev);
1679                 mdev->receiver.reset_cpu_mask = 1;
1680                 mdev->asender.reset_cpu_mask = 1;
1681                 mdev->worker.reset_cpu_mask = 1;
1682         }
1683
1684         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1685 fail:
1686         free_cpumask_var(new_cpu_mask);
1687         crypto_free_hash(csums_tfm);
1688         crypto_free_hash(verify_tfm);
1689         reply->ret_code = retcode;
1690         return 0;
1691 }
1692
1693 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1694                               struct drbd_nl_cfg_reply *reply)
1695 {
1696         int retcode;
1697
1698         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1699
1700         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1701                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1702
1703         while (retcode == SS_NEED_CONNECTION) {
1704                 spin_lock_irq(&mdev->req_lock);
1705                 if (mdev->state.conn < C_CONNECTED)
1706                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1707                 spin_unlock_irq(&mdev->req_lock);
1708
1709                 if (retcode != SS_NEED_CONNECTION)
1710                         break;
1711
1712                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1713         }
1714
1715         reply->ret_code = retcode;
1716         return 0;
1717 }
1718
1719 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1720                                    struct drbd_nl_cfg_reply *reply)
1721 {
1722
1723         reply->ret_code = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
1724
1725         return 0;
1726 }
1727
1728 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1729                               struct drbd_nl_cfg_reply *reply)
1730 {
1731         int retcode = NO_ERROR;
1732
1733         if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
1734                 retcode = ERR_PAUSE_IS_SET;
1735
1736         reply->ret_code = retcode;
1737         return 0;
1738 }
1739
1740 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1741                                struct drbd_nl_cfg_reply *reply)
1742 {
1743         int retcode = NO_ERROR;
1744
1745         if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO)
1746                 retcode = ERR_PAUSE_IS_CLEAR;
1747
1748         reply->ret_code = retcode;
1749         return 0;
1750 }
1751
1752 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1753                               struct drbd_nl_cfg_reply *reply)
1754 {
1755         reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
1756
1757         return 0;
1758 }
1759
1760 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1761                              struct drbd_nl_cfg_reply *reply)
1762 {
1763         reply->ret_code = drbd_request_state(mdev, NS(susp, 0));
1764         return 0;
1765 }
1766
1767 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1768                            struct drbd_nl_cfg_reply *reply)
1769 {
1770         reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
1771         return 0;
1772 }
1773
1774 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1775                            struct drbd_nl_cfg_reply *reply)
1776 {
1777         unsigned short *tl;
1778
1779         tl = reply->tag_list;
1780
1781         if (get_ldev(mdev)) {
1782                 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
1783                 put_ldev(mdev);
1784         }
1785
1786         if (get_net_conf(mdev)) {
1787                 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
1788                 put_net_conf(mdev);
1789         }
1790         tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
1791
1792         put_unaligned(TT_END, tl++); /* Close the tag list */
1793
1794         return (int)((char *)tl - (char *)reply->tag_list);
1795 }
1796
1797 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1798                              struct drbd_nl_cfg_reply *reply)
1799 {
1800         unsigned short *tl = reply->tag_list;
1801         union drbd_state s = mdev->state;
1802         unsigned long rs_left;
1803         unsigned int res;
1804
1805         tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
1806
1807         /* no local ref, no bitmap, no syncer progress. */
1808         if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
1809                 if (get_ldev(mdev)) {
1810                         drbd_get_syncer_progress(mdev, &rs_left, &res);
1811                         tl = tl_add_int(tl, T_sync_progress, &res);
1812                         put_ldev(mdev);
1813                 }
1814         }
1815         put_unaligned(TT_END, tl++); /* Close the tag list */
1816
1817         return (int)((char *)tl - (char *)reply->tag_list);
1818 }
1819
1820 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1821                              struct drbd_nl_cfg_reply *reply)
1822 {
1823         unsigned short *tl;
1824
1825         tl = reply->tag_list;
1826
1827         if (get_ldev(mdev)) {
1828                 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
1829                 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
1830                 put_ldev(mdev);
1831         }
1832         put_unaligned(TT_END, tl++); /* Close the tag list */
1833
1834         return (int)((char *)tl - (char *)reply->tag_list);
1835 }
1836
1837 /**
1838  * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
1839  * @mdev:       DRBD device.
1840  * @nlp:        Netlink/connector packet from drbdsetup
1841  * @reply:      Reply packet for drbdsetup
1842  */
1843 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1844                                     struct drbd_nl_cfg_reply *reply)
1845 {
1846         unsigned short *tl;
1847         char rv;
1848
1849         tl = reply->tag_list;
1850
1851         rv = mdev->state.pdsk == D_OUTDATED        ? UT_PEER_OUTDATED :
1852           test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
1853
1854         tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
1855         put_unaligned(TT_END, tl++); /* Close the tag list */
1856
1857         return (int)((char *)tl - (char *)reply->tag_list);
1858 }
1859
1860 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1861                                     struct drbd_nl_cfg_reply *reply)
1862 {
1863         /* default to resume from last known position, if possible */
1864         struct start_ov args =
1865                 { .start_sector = mdev->ov_start_sector };
1866
1867         if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
1868                 reply->ret_code = ERR_MANDATORY_TAG;
1869                 return 0;
1870         }
1871         /* w_make_ov_request expects position to be aligned */
1872         mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
1873         reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
1874         return 0;
1875 }
1876
1877
1878 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1879                               struct drbd_nl_cfg_reply *reply)
1880 {
1881         int retcode = NO_ERROR;
1882         int skip_initial_sync = 0;
1883         int err;
1884
1885         struct new_c_uuid args;
1886
1887         memset(&args, 0, sizeof(struct new_c_uuid));
1888         if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
1889                 reply->ret_code = ERR_MANDATORY_TAG;
1890                 return 0;
1891         }
1892
1893         mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
1894
1895         if (!get_ldev(mdev)) {
1896                 retcode = ERR_NO_DISK;
1897                 goto out;
1898         }
1899
1900         /* this is "skip initial sync", assume to be clean */
1901         if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
1902             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
1903                 dev_info(DEV, "Preparing to skip initial sync\n");
1904                 skip_initial_sync = 1;
1905         } else if (mdev->state.conn != C_STANDALONE) {
1906                 retcode = ERR_CONNECTED;
1907                 goto out_dec;
1908         }
1909
1910         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
1911         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
1912
1913         if (args.clear_bm) {
1914                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, "clear_n_write from new_c_uuid");
1915                 if (err) {
1916                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
1917                         retcode = ERR_IO_MD_DISK;
1918                 }
1919                 if (skip_initial_sync) {
1920                         drbd_send_uuids_skip_initial_sync(mdev);
1921                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
1922                         spin_lock_irq(&mdev->req_lock);
1923                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
1924                                         CS_VERBOSE, NULL);
1925                         spin_unlock_irq(&mdev->req_lock);
1926                 }
1927         }
1928
1929         drbd_md_sync(mdev);
1930 out_dec:
1931         put_ldev(mdev);
1932 out:
1933         mutex_unlock(&mdev->state_mutex);
1934
1935         reply->ret_code = retcode;
1936         return 0;
1937 }
1938
1939 static struct drbd_conf *ensure_mdev(struct drbd_nl_cfg_req *nlp)
1940 {
1941         struct drbd_conf *mdev;
1942
1943         if (nlp->drbd_minor >= minor_count)
1944                 return NULL;
1945
1946         mdev = minor_to_mdev(nlp->drbd_minor);
1947
1948         if (!mdev && (nlp->flags & DRBD_NL_CREATE_DEVICE)) {
1949                 struct gendisk *disk = NULL;
1950                 mdev = drbd_new_device(nlp->drbd_minor);
1951
1952                 spin_lock_irq(&drbd_pp_lock);
1953                 if (minor_table[nlp->drbd_minor] == NULL) {
1954                         minor_table[nlp->drbd_minor] = mdev;
1955                         disk = mdev->vdisk;
1956                         mdev = NULL;
1957                 } /* else: we lost the race */
1958                 spin_unlock_irq(&drbd_pp_lock);
1959
1960                 if (disk) /* we won the race above */
1961                         /* in case we ever add a drbd_delete_device(),
1962                          * don't forget the del_gendisk! */
1963                         add_disk(disk);
1964                 else /* we lost the race above */
1965                         drbd_free_mdev(mdev);
1966
1967                 mdev = minor_to_mdev(nlp->drbd_minor);
1968         }
1969
1970         return mdev;
1971 }
1972
1973 struct cn_handler_struct {
1974         int (*function)(struct drbd_conf *,
1975                          struct drbd_nl_cfg_req *,
1976                          struct drbd_nl_cfg_reply *);
1977         int reply_body_size;
1978 };
1979
1980 static struct cn_handler_struct cnd_table[] = {
1981         [ P_primary ]           = { &drbd_nl_primary,           0 },
1982         [ P_secondary ]         = { &drbd_nl_secondary,         0 },
1983         [ P_disk_conf ]         = { &drbd_nl_disk_conf,         0 },
1984         [ P_detach ]            = { &drbd_nl_detach,            0 },
1985         [ P_net_conf ]          = { &drbd_nl_net_conf,          0 },
1986         [ P_disconnect ]        = { &drbd_nl_disconnect,        0 },
1987         [ P_resize ]            = { &drbd_nl_resize,            0 },
1988         [ P_syncer_conf ]       = { &drbd_nl_syncer_conf,       0 },
1989         [ P_invalidate ]        = { &drbd_nl_invalidate,        0 },
1990         [ P_invalidate_peer ]   = { &drbd_nl_invalidate_peer,   0 },
1991         [ P_pause_sync ]        = { &drbd_nl_pause_sync,        0 },
1992         [ P_resume_sync ]       = { &drbd_nl_resume_sync,       0 },
1993         [ P_suspend_io ]        = { &drbd_nl_suspend_io,        0 },
1994         [ P_resume_io ]         = { &drbd_nl_resume_io,         0 },
1995         [ P_outdate ]           = { &drbd_nl_outdate,           0 },
1996         [ P_get_config ]        = { &drbd_nl_get_config,
1997                                     sizeof(struct syncer_conf_tag_len_struct) +
1998                                     sizeof(struct disk_conf_tag_len_struct) +
1999                                     sizeof(struct net_conf_tag_len_struct) },
2000         [ P_get_state ]         = { &drbd_nl_get_state,
2001                                     sizeof(struct get_state_tag_len_struct) +
2002                                     sizeof(struct sync_progress_tag_len_struct) },
2003         [ P_get_uuids ]         = { &drbd_nl_get_uuids,
2004                                     sizeof(struct get_uuids_tag_len_struct) },
2005         [ P_get_timeout_flag ]  = { &drbd_nl_get_timeout_flag,
2006                                     sizeof(struct get_timeout_flag_tag_len_struct)},
2007         [ P_start_ov ]          = { &drbd_nl_start_ov,          0 },
2008         [ P_new_c_uuid ]        = { &drbd_nl_new_c_uuid,        0 },
2009 };
2010
2011 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2012 {
2013         struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2014         struct cn_handler_struct *cm;
2015         struct cn_msg *cn_reply;
2016         struct drbd_nl_cfg_reply *reply;
2017         struct drbd_conf *mdev;
2018         int retcode, rr;
2019         int reply_size = sizeof(struct cn_msg)
2020                 + sizeof(struct drbd_nl_cfg_reply)
2021                 + sizeof(short int);
2022
2023         if (!try_module_get(THIS_MODULE)) {
2024                 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2025                 return;
2026         }
2027
2028         if (!cap_raised(nsp->eff_cap, CAP_SYS_ADMIN)) {
2029                 retcode = ERR_PERM;
2030                 goto fail;
2031         }
2032
2033         mdev = ensure_mdev(nlp);
2034         if (!mdev) {
2035                 retcode = ERR_MINOR_INVALID;
2036                 goto fail;
2037         }
2038
2039         if (nlp->packet_type >= P_nl_after_last_packet) {
2040                 retcode = ERR_PACKET_NR;
2041                 goto fail;
2042         }
2043
2044         cm = cnd_table + nlp->packet_type;
2045
2046         /* This may happen if packet number is 0: */
2047         if (cm->function == NULL) {
2048                 retcode = ERR_PACKET_NR;
2049                 goto fail;
2050         }
2051
2052         reply_size += cm->reply_body_size;
2053
2054         /* allocation not in the IO path, cqueue thread context */
2055         cn_reply = kmalloc(reply_size, GFP_KERNEL);
2056         if (!cn_reply) {
2057                 retcode = ERR_NOMEM;
2058                 goto fail;
2059         }
2060         reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2061
2062         reply->packet_type =
2063                 cm->reply_body_size ? nlp->packet_type : P_nl_after_last_packet;
2064         reply->minor = nlp->drbd_minor;
2065         reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2066         /* reply->tag_list; might be modified by cm->function. */
2067
2068         rr = cm->function(mdev, nlp, reply);
2069
2070         cn_reply->id = req->id;
2071         cn_reply->seq = req->seq;
2072         cn_reply->ack = req->ack  + 1;
2073         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2074         cn_reply->flags = 0;
2075
2076         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2077         if (rr && rr != -ESRCH)
2078                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2079
2080         kfree(cn_reply);
2081         module_put(THIS_MODULE);
2082         return;
2083  fail:
2084         drbd_nl_send_reply(req, retcode);
2085         module_put(THIS_MODULE);
2086 }
2087
2088 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2089
2090 static unsigned short *
2091 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2092         unsigned short len, int nul_terminated)
2093 {
2094         unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2095         len = (len < l) ? len :  l;
2096         put_unaligned(tag, tl++);
2097         put_unaligned(len, tl++);
2098         memcpy(tl, data, len);
2099         tl = (unsigned short*)((char*)tl + len);
2100         if (nul_terminated)
2101                 *((char*)tl - 1) = 0;
2102         return tl;
2103 }
2104
2105 static unsigned short *
2106 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2107 {
2108         return __tl_add_blob(tl, tag, data, len, 0);
2109 }
2110
2111 static unsigned short *
2112 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2113 {
2114         return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2115 }
2116
2117 static unsigned short *
2118 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2119 {
2120         put_unaligned(tag, tl++);
2121         switch(tag_type(tag)) {
2122         case TT_INTEGER:
2123                 put_unaligned(sizeof(int), tl++);
2124                 put_unaligned(*(int *)val, (int *)tl);
2125                 tl = (unsigned short*)((char*)tl+sizeof(int));
2126                 break;
2127         case TT_INT64:
2128                 put_unaligned(sizeof(u64), tl++);
2129                 put_unaligned(*(u64 *)val, (u64 *)tl);
2130                 tl = (unsigned short*)((char*)tl+sizeof(u64));
2131                 break;
2132         default:
2133                 /* someone did something stupid. */
2134                 ;
2135         }
2136         return tl;
2137 }
2138
2139 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2140 {
2141         char buffer[sizeof(struct cn_msg)+
2142                     sizeof(struct drbd_nl_cfg_reply)+
2143                     sizeof(struct get_state_tag_len_struct)+
2144                     sizeof(short int)];
2145         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2146         struct drbd_nl_cfg_reply *reply =
2147                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2148         unsigned short *tl = reply->tag_list;
2149
2150         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2151
2152         tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2153
2154         put_unaligned(TT_END, tl++); /* Close the tag list */
2155
2156         cn_reply->id.idx = CN_IDX_DRBD;
2157         cn_reply->id.val = CN_VAL_DRBD;
2158
2159         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2160         cn_reply->ack = 0; /* not used here. */
2161         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2162                 (int)((char *)tl - (char *)reply->tag_list);
2163         cn_reply->flags = 0;
2164
2165         reply->packet_type = P_get_state;
2166         reply->minor = mdev_to_minor(mdev);
2167         reply->ret_code = NO_ERROR;
2168
2169         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2170 }
2171
2172 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2173 {
2174         char buffer[sizeof(struct cn_msg)+
2175                     sizeof(struct drbd_nl_cfg_reply)+
2176                     sizeof(struct call_helper_tag_len_struct)+
2177                     sizeof(short int)];
2178         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2179         struct drbd_nl_cfg_reply *reply =
2180                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2181         unsigned short *tl = reply->tag_list;
2182
2183         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2184
2185         tl = tl_add_str(tl, T_helper, helper_name);
2186         put_unaligned(TT_END, tl++); /* Close the tag list */
2187
2188         cn_reply->id.idx = CN_IDX_DRBD;
2189         cn_reply->id.val = CN_VAL_DRBD;
2190
2191         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2192         cn_reply->ack = 0; /* not used here. */
2193         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2194                 (int)((char *)tl - (char *)reply->tag_list);
2195         cn_reply->flags = 0;
2196
2197         reply->packet_type = P_call_helper;
2198         reply->minor = mdev_to_minor(mdev);
2199         reply->ret_code = NO_ERROR;
2200
2201         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2202 }
2203
2204 void drbd_bcast_ee(struct drbd_conf *mdev,
2205                 const char *reason, const int dgs,
2206                 const char* seen_hash, const char* calc_hash,
2207                 const struct drbd_epoch_entry* e)
2208 {
2209         struct cn_msg *cn_reply;
2210         struct drbd_nl_cfg_reply *reply;
2211         struct bio_vec *bvec;
2212         unsigned short *tl;
2213         int i;
2214
2215         if (!e)
2216                 return;
2217         if (!reason || !reason[0])
2218                 return;
2219
2220         /* apparently we have to memcpy twice, first to prepare the data for the
2221          * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2222          * netlink skb. */
2223         /* receiver thread context, which is not in the writeout path (of this node),
2224          * but may be in the writeout path of the _other_ node.
2225          * GFP_NOIO to avoid potential "distributed deadlock". */
2226         cn_reply = kmalloc(
2227                 sizeof(struct cn_msg)+
2228                 sizeof(struct drbd_nl_cfg_reply)+
2229                 sizeof(struct dump_ee_tag_len_struct)+
2230                 sizeof(short int),
2231                 GFP_NOIO);
2232
2233         if (!cn_reply) {
2234                 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2235                                 (unsigned long long)e->sector, e->size);
2236                 return;
2237         }
2238
2239         reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2240         tl = reply->tag_list;
2241
2242         tl = tl_add_str(tl, T_dump_ee_reason, reason);
2243         tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2244         tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2245         tl = tl_add_int(tl, T_ee_sector, &e->sector);
2246         tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2247
2248         put_unaligned(T_ee_data, tl++);
2249         put_unaligned(e->size, tl++);
2250
2251         __bio_for_each_segment(bvec, e->private_bio, i, 0) {
2252                 void *d = kmap(bvec->bv_page);
2253                 memcpy(tl, d + bvec->bv_offset, bvec->bv_len);
2254                 kunmap(bvec->bv_page);
2255                 tl=(unsigned short*)((char*)tl + bvec->bv_len);
2256         }
2257         put_unaligned(TT_END, tl++); /* Close the tag list */
2258
2259         cn_reply->id.idx = CN_IDX_DRBD;
2260         cn_reply->id.val = CN_VAL_DRBD;
2261
2262         cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2263         cn_reply->ack = 0; // not used here.
2264         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2265                 (int)((char*)tl - (char*)reply->tag_list);
2266         cn_reply->flags = 0;
2267
2268         reply->packet_type = P_dump_ee;
2269         reply->minor = mdev_to_minor(mdev);
2270         reply->ret_code = NO_ERROR;
2271
2272         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2273         kfree(cn_reply);
2274 }
2275
2276 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2277 {
2278         char buffer[sizeof(struct cn_msg)+
2279                     sizeof(struct drbd_nl_cfg_reply)+
2280                     sizeof(struct sync_progress_tag_len_struct)+
2281                     sizeof(short int)];
2282         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2283         struct drbd_nl_cfg_reply *reply =
2284                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2285         unsigned short *tl = reply->tag_list;
2286         unsigned long rs_left;
2287         unsigned int res;
2288
2289         /* no local ref, no bitmap, no syncer progress, no broadcast. */
2290         if (!get_ldev(mdev))
2291                 return;
2292         drbd_get_syncer_progress(mdev, &rs_left, &res);
2293         put_ldev(mdev);
2294
2295         tl = tl_add_int(tl, T_sync_progress, &res);
2296         put_unaligned(TT_END, tl++); /* Close the tag list */
2297
2298         cn_reply->id.idx = CN_IDX_DRBD;
2299         cn_reply->id.val = CN_VAL_DRBD;
2300
2301         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2302         cn_reply->ack = 0; /* not used here. */
2303         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2304                 (int)((char *)tl - (char *)reply->tag_list);
2305         cn_reply->flags = 0;
2306
2307         reply->packet_type = P_sync_progress;
2308         reply->minor = mdev_to_minor(mdev);
2309         reply->ret_code = NO_ERROR;
2310
2311         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2312 }
2313
2314 int __init drbd_nl_init(void)
2315 {
2316         static struct cb_id cn_id_drbd;
2317         int err, try=10;
2318
2319         cn_id_drbd.val = CN_VAL_DRBD;
2320         do {
2321                 cn_id_drbd.idx = cn_idx;
2322                 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2323                 if (!err)
2324                         break;
2325                 cn_idx = (cn_idx + CN_IDX_STEP);
2326         } while (try--);
2327
2328         if (err) {
2329                 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2330                 return err;
2331         }
2332
2333         return 0;
2334 }
2335
2336 void drbd_nl_cleanup(void)
2337 {
2338         static struct cb_id cn_id_drbd;
2339
2340         cn_id_drbd.idx = cn_idx;
2341         cn_id_drbd.val = CN_VAL_DRBD;
2342
2343         cn_del_callback(&cn_id_drbd);
2344 }
2345
2346 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2347 {
2348         char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2349         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2350         struct drbd_nl_cfg_reply *reply =
2351                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2352         int rr;
2353
2354         cn_reply->id = req->id;
2355
2356         cn_reply->seq = req->seq;
2357         cn_reply->ack = req->ack  + 1;
2358         cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2359         cn_reply->flags = 0;
2360
2361         reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2362         reply->ret_code = ret_code;
2363
2364         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2365         if (rr && rr != -ESRCH)
2366                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2367 }
2368