drivers/block/drbd/drbd_nl.c

   1 /*
   2    drbd_nl.c
   3
   4    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
   5
   6    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
   7    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
   8    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
   9
  10    drbd is free software; you can redistribute it and/or modify
  11    it under the terms of the GNU General Public License as published by
  12    the Free Software Foundation; either version 2, or (at your option)
  13    any later version.
  14
  15    drbd is distributed in the hope that it will be useful,
  16    but WITHOUT ANY WARRANTY; without even the implied warranty of
  17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18    GNU General Public License for more details.
  19
  20    You should have received a copy of the GNU General Public License
  21    along with drbd; see the file COPYING.  If not, write to
  22    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  23
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/drbd.h>
  28 #include <linux/in.h>
  29 #include <linux/fs.h>
  30 #include <linux/file.h>
  31 #include <linux/slab.h>
  32 #include <linux/connector.h>
  33 #include <linux/blkpg.h>
  34 #include <linux/cpumask.h>
  35 #include "drbd_int.h"
  36 #include "drbd_req.h"
  37 #include "drbd_wrappers.h"
  38 #include <asm/unaligned.h>
  39 #include <linux/drbd_tag_magic.h>
  40 #include <linux/drbd_limits.h>
  41 #include <linux/compiler.h>
  42 #include <linux/kthread.h>
  43
  44 static unsigned short *tl_add_blob(unsigned short *, enum drbd_tags, const void *, int);
  45 static unsigned short *tl_add_str(unsigned short *, enum drbd_tags, const char *);
  46 static unsigned short *tl_add_int(unsigned short *, enum drbd_tags, const void *);
  47
  48 /* see get_sb_bdev and bd_claim */
  49 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
  50
  51 /* Generate the tag_list to struct functions */
  52 #define NL_PACKET(name, number, fields) \
  53 static int name ## _from_tags(struct drbd_conf *mdev, \
  54         unsigned short *tags, struct name *arg) __attribute__ ((unused)); \
  55 static int name ## _from_tags(struct drbd_conf *mdev, \
  56         unsigned short *tags, struct name *arg) \
  57 { \
  58         int tag; \
  59         int dlen; \
  60         \
  61         while ((tag = get_unaligned(tags++)) != TT_END) {       \
  62                 dlen = get_unaligned(tags++);                   \
  63                 switch (tag_number(tag)) { \
  64                 fields \
  65                 default: \
  66                         if (tag & T_MANDATORY) { \
  67                                 dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \
  68                                 return 0; \
  69                         } \
  70                 } \
  71                 tags = (unsigned short *)((char *)tags + dlen); \
  72         } \
  73         return 1; \
  74 }
  75 #define NL_INTEGER(pn, pr, member) \
  76         case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \
  77                 arg->member = get_unaligned((int *)(tags));     \
  78                 break;
  79 #define NL_INT64(pn, pr, member) \
  80         case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \
  81                 arg->member = get_unaligned((u64 *)(tags));     \
  82                 break;
  83 #define NL_BIT(pn, pr, member) \
  84         case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \
  85                 arg->member = *(char *)(tags) ? 1 : 0; \
  86                 break;
  87 #define NL_STRING(pn, pr, member, len) \
  88         case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \
  89                 if (dlen > len) { \
  90                         dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \
  91                                 #member, dlen, (unsigned int)len); \
  92                         return 0; \
  93                 } \
  94                  arg->member ## _len = dlen; \
  95                  memcpy(arg->member, tags, min_t(size_t, dlen, len)); \
  96                  break;
  97 #include "linux/drbd_nl.h"
  98
  99 /* Generate the struct to tag_list functions */
 100 #define NL_PACKET(name, number, fields) \
 101 static unsigned short* \
 102 name ## _to_tags(struct drbd_conf *mdev, \
 103         struct name *arg, unsigned short *tags) __attribute__ ((unused)); \
 104 static unsigned short* \
 105 name ## _to_tags(struct drbd_conf *mdev, \
 106         struct name *arg, unsigned short *tags) \
 107 { \
 108         fields \
 109         return tags; \
 110 }
 111
 112 #define NL_INTEGER(pn, pr, member) \
 113         put_unaligned(pn | pr | TT_INTEGER, tags++);    \
 114         put_unaligned(sizeof(int), tags++);             \
 115         put_unaligned(arg->member, (int *)tags);        \
 116         tags = (unsigned short *)((char *)tags+sizeof(int));
 117 #define NL_INT64(pn, pr, member) \
 118         put_unaligned(pn | pr | TT_INT64, tags++);      \
 119         put_unaligned(sizeof(u64), tags++);             \
 120         put_unaligned(arg->member, (u64 *)tags);        \
 121         tags = (unsigned short *)((char *)tags+sizeof(u64));
 122 #define NL_BIT(pn, pr, member) \
 123         put_unaligned(pn | pr | TT_BIT, tags++);        \
 124         put_unaligned(sizeof(char), tags++);            \
 125         *(char *)tags = arg->member; \
 126         tags = (unsigned short *)((char *)tags+sizeof(char));
 127 #define NL_STRING(pn, pr, member, len) \
 128         put_unaligned(pn | pr | TT_STRING, tags++);     \
 129         put_unaligned(arg->member ## _len, tags++);     \
 130         memcpy(tags, arg->member, arg->member ## _len); \
 131         tags = (unsigned short *)((char *)tags + arg->member ## _len);
 132 #include "linux/drbd_nl.h"
 133
 134 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name);
 135 void drbd_nl_send_reply(struct cn_msg *, int);
 136
 137 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
 138 {
 139         char *envp[] = { "HOME=/",
 140                         "TERM=linux",
 141                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 142                         NULL, /* Will be set to address family */
 143                         NULL, /* Will be set to address */
 144                         NULL };
 145
 146         char mb[12], af[20], ad[60], *afs;
 147         char *argv[] = {usermode_helper, cmd, mb, NULL };
 148         int ret;
 149
 150         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 151
 152         if (get_net_conf(mdev)) {
 153                 switch (((struct sockaddr *)mdev->net_conf->peer_addr)->sa_family) {
 154                 case AF_INET6:
 155                         afs = "ipv6";
 156                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6",
 157                                  &((struct sockaddr_in6 *)mdev->net_conf->peer_addr)->sin6_addr);
 158                         break;
 159                 case AF_INET:
 160                         afs = "ipv4";
 161                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 162                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
 163                         break;
 164                 default:
 165                         afs = "ssocks";
 166                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 167                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
 168                 }
 169                 snprintf(af, 20, "DRBD_PEER_AF=%s", afs);
 170                 envp[3]=af;
 171                 envp[4]=ad;
 172                 put_net_conf(mdev);
 173         }
 174
 175         /* The helper may take some time.
 176          * write out any unsynced meta data changes now */
 177         drbd_md_sync(mdev);
 178
 179         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 180
 181         drbd_bcast_ev_helper(mdev, cmd);
 182         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 183         if (ret)
 184                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 185                                 usermode_helper, cmd, mb,
 186                                 (ret >> 8) & 0xff, ret);
 187         else
 188                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 189                                 usermode_helper, cmd, mb,
 190                                 (ret >> 8) & 0xff, ret);
 191
 192         if (ret < 0) /* Ignore any ERRNOs we got. */
 193                 ret = 0;
 194
 195         return ret;
 196 }
 197
 198 enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
 199 {
 200         char *ex_to_string;
 201         int r;
 202         enum drbd_disk_state nps;
 203         enum drbd_fencing_p fp;
 204
 205         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 206
 207         if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 208                 fp = mdev->ldev->dc.fencing;
 209                 put_ldev(mdev);
 210         } else {
 211                 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n");
 212                 nps = mdev->state.pdsk;
 213                 goto out;
 214         }
 215
 216         r = drbd_khelper(mdev, "fence-peer");
 217
 218         switch ((r>>8) & 0xff) {
 219         case 3: /* peer is inconsistent */
 220                 ex_to_string = "peer is inconsistent or worse";
 221                 nps = D_INCONSISTENT;
 222                 break;
 223         case 4: /* peer got outdated, or was already outdated */
 224                 ex_to_string = "peer was fenced";
 225                 nps = D_OUTDATED;
 226                 break;
 227         case 5: /* peer was down */
 228                 if (mdev->state.disk == D_UP_TO_DATE) {
 229                         /* we will(have) create(d) a new UUID anyways... */
 230                         ex_to_string = "peer is unreachable, assumed to be dead";
 231                         nps = D_OUTDATED;
 232                 } else {
 233                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 234                         nps = mdev->state.pdsk;
 235                 }
 236                 break;
 237         case 6: /* Peer is primary, voluntarily outdate myself.
 238                  * This is useful when an unconnected R_SECONDARY is asked to
 239                  * become R_PRIMARY, but finds the other peer being active. */
 240                 ex_to_string = "peer is active";
 241                 dev_warn(DEV, "Peer is primary, outdating myself.\n");
 242                 nps = D_UNKNOWN;
 243                 _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE);
 244                 break;
 245         case 7:
 246                 if (fp != FP_STONITH)
 247                         dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n");
 248                 ex_to_string = "peer was stonithed";
 249                 nps = D_OUTDATED;
 250                 break;
 251         default:
 252                 /* The script is broken ... */
 253                 nps = D_UNKNOWN;
 254                 dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 255                 return nps;
 256         }
 257
 258         dev_info(DEV, "fence-peer helper returned %d (%s)\n",
 259                         (r>>8) & 0xff, ex_to_string);
 260
 261 out:
 262         if (mdev->state.susp_fen && nps >= D_UNKNOWN) {
 263                 /* The handler was not successful... unfreeze here, the
 264                    state engine can not unfreeze... */
 265                 _drbd_request_state(mdev, NS(susp_fen, 0), CS_VERBOSE);
 266         }
 267
 268         return nps;
 269 }
 270
 271 static int _try_outdate_peer_async(void *data)
 272 {
 273         struct drbd_conf *mdev = (struct drbd_conf *)data;
 274         enum drbd_disk_state nps;
 275
 276         nps = drbd_try_outdate_peer(mdev);
 277         drbd_request_state(mdev, NS(pdsk, nps));
 278
 279         return 0;
 280 }
 281
 282 void drbd_try_outdate_peer_async(struct drbd_conf *mdev)
 283 {
 284         struct task_struct *opa;
 285
 286         opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev));
 287         if (IS_ERR(opa))
 288                 dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n");
 289 }
 290
 291 enum drbd_state_rv
 292 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 293 {
 294         const int max_tries = 4;
 295         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
 296         int try = 0;
 297         int forced = 0;
 298         union drbd_state mask, val;
 299         enum drbd_disk_state nps;
 300
 301         if (new_role == R_PRIMARY)
 302                 request_ping(mdev); /* Detect a dead peer ASAP */
 303
 304         mutex_lock(&mdev->state_mutex);
 305
 306         mask.i = 0; mask.role = R_MASK;
 307         val.i  = 0; val.role  = new_role;
 308
 309         while (try++ < max_tries) {
 310                 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 311
 312                 /* in case we first succeeded to outdate,
 313                  * but now suddenly could establish a connection */
 314                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 315                         val.pdsk = 0;
 316                         mask.pdsk = 0;
 317                         continue;
 318                 }
 319
 320                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
 321                     (mdev->state.disk < D_UP_TO_DATE &&
 322                      mdev->state.disk >= D_INCONSISTENT)) {
 323                         mask.disk = D_MASK;
 324                         val.disk  = D_UP_TO_DATE;
 325                         forced = 1;
 326                         continue;
 327                 }
 328
 329                 if (rv == SS_NO_UP_TO_DATE_DISK &&
 330                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 331                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 332                         nps = drbd_try_outdate_peer(mdev);
 333
 334                         if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
 335                                 val.disk = D_UP_TO_DATE;
 336                                 mask.disk = D_MASK;
 337                         }
 338
 339                         val.pdsk = nps;
 340                         mask.pdsk = D_MASK;
 341
 342                         continue;
 343                 }
 344
 345                 if (rv == SS_NOTHING_TO_DO)
 346                         goto fail;
 347                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
 348                         nps = drbd_try_outdate_peer(mdev);
 349
 350                         if (force && nps > D_OUTDATED) {
 351                                 dev_warn(DEV, "Forced into split brain situation!\n");
 352                                 nps = D_OUTDATED;
 353                         }
 354
 355                         mask.pdsk = D_MASK;
 356                         val.pdsk  = nps;
 357
 358                         continue;
 359                 }
 360                 if (rv == SS_TWO_PRIMARIES) {
 361                         /* Maybe the peer is detected as dead very soon...
 362                            retry at most once more in this case. */
 363                         schedule_timeout_interruptible((mdev->net_conf->ping_timeo+1)*HZ/10);
 364                         if (try < max_tries)
 365                                 try = max_tries - 1;
 366                         continue;
 367                 }
 368                 if (rv < SS_SUCCESS) {
 369                         rv = _drbd_request_state(mdev, mask, val,
 370                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 371                         if (rv < SS_SUCCESS)
 372                                 goto fail;
 373                 }
 374                 break;
 375         }
 376
 377         if (rv < SS_SUCCESS)
 378                 goto fail;
 379
 380         if (forced)
 381                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 382
 383         /* Wait until nothing is on the fly :) */
 384         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 385
 386         if (new_role == R_SECONDARY) {
 387                 set_disk_ro(mdev->vdisk, true);
 388                 if (get_ldev(mdev)) {
 389                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 390                         put_ldev(mdev);
 391                 }
 392         } else {
 393                 if (get_net_conf(mdev)) {
 394                         mdev->net_conf->want_lose = 0;
 395                         put_net_conf(mdev);
 396                 }
 397                 set_disk_ro(mdev->vdisk, false);
 398                 if (get_ldev(mdev)) {
 399                         if (((mdev->state.conn < C_CONNECTED ||
 400                                mdev->state.pdsk <= D_FAILED)
 401                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 402                                 drbd_uuid_new_current(mdev);
 403
 404                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 405                         put_ldev(mdev);
 406                 }
 407         }
 408
 409         /* writeout of activity log covered areas of the bitmap
 410          * to stable storage done in after state change already */
 411
 412         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 413                 /* if this was forced, we should consider sync */
 414                 if (forced)
 415                         drbd_send_uuids(mdev);
 416                 drbd_send_state(mdev);
 417         }
 418
 419         drbd_md_sync(mdev);
 420
 421         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 422  fail:
 423         mutex_unlock(&mdev->state_mutex);
 424         return rv;
 425 }
 426
 427 static struct drbd_conf *ensure_mdev(int minor, int create)
 428 {
 429         struct drbd_conf *mdev;
 430
 431         if (minor >= minor_count)
 432                 return NULL;
 433
 434         mdev = minor_to_mdev(minor);
 435
 436         if (!mdev && create) {
 437                 struct gendisk *disk = NULL;
 438                 mdev = drbd_new_device(minor);
 439
 440                 spin_lock_irq(&drbd_pp_lock);
 441                 if (minor_table[minor] == NULL) {
 442                         minor_table[minor] = mdev;
 443                         disk = mdev->vdisk;
 444                         mdev = NULL;
 445                 } /* else: we lost the race */
 446                 spin_unlock_irq(&drbd_pp_lock);
 447
 448                 if (disk) /* we won the race above */
 449                         /* in case we ever add a drbd_delete_device(),
 450                          * don't forget the del_gendisk! */
 451                         add_disk(disk);
 452                 else /* we lost the race above */
 453                         drbd_free_mdev(mdev);
 454
 455                 mdev = minor_to_mdev(minor);
 456         }
 457
 458         return mdev;
 459 }
 460
 461 static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 462                            struct drbd_nl_cfg_reply *reply)
 463 {
 464         struct primary primary_args;
 465
 466         memset(&primary_args, 0, sizeof(struct primary));
 467         if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) {
 468                 reply->ret_code = ERR_MANDATORY_TAG;
 469                 return 0;
 470         }
 471
 472         reply->ret_code =
 473                 drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force);
 474
 475         return 0;
 476 }
 477
 478 static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 479                              struct drbd_nl_cfg_reply *reply)
 480 {
 481         reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0);
 482
 483         return 0;
 484 }
 485
 486 /* initializes the md.*_offset members, so we are able to find
 487  * the on disk meta data */
 488 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 489                                        struct drbd_backing_dev *bdev)
 490 {
 491         sector_t md_size_sect = 0;
 492         switch (bdev->dc.meta_dev_idx) {
 493         default:
 494                 /* v07 style fixed size indexed meta data */
 495                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 496                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 497                 bdev->md.al_offset = MD_AL_OFFSET;
 498                 bdev->md.bm_offset = MD_BM_OFFSET;
 499                 break;
 500         case DRBD_MD_INDEX_FLEX_EXT:
 501                 /* just occupy the full device; unit: sectors */
 502                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 503                 bdev->md.md_offset = 0;
 504                 bdev->md.al_offset = MD_AL_OFFSET;
 505                 bdev->md.bm_offset = MD_BM_OFFSET;
 506                 break;
 507         case DRBD_MD_INDEX_INTERNAL:
 508         case DRBD_MD_INDEX_FLEX_INT:
 509                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 510                 /* al size is still fixed */
 511                 bdev->md.al_offset = -MD_AL_MAX_SIZE;
 512                 /* we need (slightly less than) ~ this much bitmap sectors: */
 513                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 514                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 515                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 516                 md_size_sect = ALIGN(md_size_sect, 8);
 517
 518                 /* plus the "drbd meta data super block",
 519                  * and the activity log; */
 520                 md_size_sect += MD_BM_OFFSET;
 521
 522                 bdev->md.md_size_sect = md_size_sect;
 523                 /* bitmap offset is adjusted by 'super' block size */
 524                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 525                 break;
 526         }
 527 }
 528
 529 /* input size is expected to be in KB */
 530 char *ppsize(char *buf, unsigned long long size)
 531 {
 532         /* Needs 9 bytes at max including trailing NUL:
 533          * -1ULL ==> "16384 EB" */
 534         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 535         int base = 0;
 536         while (size >= 10000 && base < sizeof(units)-1) {
 537                 /* shift + round */
 538                 size = (size >> 10) + !!(size & (1<<9));
 539                 base++;
 540         }
 541         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
 542
 543         return buf;
 544 }
 545
 546 /* there is still a theoretical deadlock when called from receiver
 547  * on an D_INCONSISTENT R_PRIMARY:
 548  *  remote READ does inc_ap_bio, receiver would need to receive answer
 549  *  packet from remote to dec_ap_bio again.
 550  *  receiver receive_sizes(), comes here,
 551  *  waits for ap_bio_cnt == 0. -> deadlock.
 552  * but this cannot happen, actually, because:
 553  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 554  *  (not connected, or bad/no disk on peer):
 555  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 556  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 557  *  peer may not initiate a resize.
 558  */
 559 void drbd_suspend_io(struct drbd_conf *mdev)
 560 {
 561         set_bit(SUSPEND_IO, &mdev->flags);
 562         if (is_susp(mdev->state))
 563                 return;
 564         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 565 }
 566
 567 void drbd_resume_io(struct drbd_conf *mdev)
 568 {
 569         clear_bit(SUSPEND_IO, &mdev->flags);
 570         wake_up(&mdev->misc_wait);
 571 }
 572
 573 /**
 574  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 575  * @mdev:       DRBD device.
 576  *
 577  * Returns 0 on success, negative return values indicate errors.
 578  * You should call drbd_md_sync() after calling this function.
 579  */
 580 enum determine_dev_size drbd_determin_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 581 {
 582         sector_t prev_first_sect, prev_size; /* previous meta location */
 583         sector_t la_size;
 584         sector_t size;
 585         char ppb[10];
 586
 587         int md_moved, la_size_changed;
 588         enum determine_dev_size rv = unchanged;
 589
 590         /* race:
 591          * application request passes inc_ap_bio,
 592          * but then cannot get an AL-reference.
 593          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 594          *
 595          * to avoid that:
 596          * Suspend IO right here.
 597          * still lock the act_log to not trigger ASSERTs there.
 598          */
 599         drbd_suspend_io(mdev);
 600
 601         /* no wait necessary anymore, actually we could assert that */
 602         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 603
 604         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 605         prev_size = mdev->ldev->md.md_size_sect;
 606         la_size = mdev->ldev->md.la_size_sect;
 607
 608         /* TODO: should only be some assert here, not (re)init... */
 609         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 610
 611         size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
 612
 613         if (drbd_get_capacity(mdev->this_bdev) != size ||
 614             drbd_bm_capacity(mdev) != size) {
 615                 int err;
 616                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 617                 if (unlikely(err)) {
 618                         /* currently there is only one error: ENOMEM! */
 619                         size = drbd_bm_capacity(mdev)>>1;
 620                         if (size == 0) {
 621                                 dev_err(DEV, "OUT OF MEMORY! "
 622                                     "Could not allocate bitmap!\n");
 623                         } else {
 624                                 dev_err(DEV, "BM resizing failed. "
 625                                     "Leaving size unchanged at size = %lu KB\n",
 626                                     (unsigned long)size);
 627                         }
 628                         rv = dev_size_error;
 629                 }
 630                 /* racy, see comments above. */
 631                 drbd_set_my_capacity(mdev, size);
 632                 mdev->ldev->md.la_size_sect = size;
 633                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 634                      (unsigned long long)size>>1);
 635         }
 636         if (rv == dev_size_error)
 637                 goto out;
 638
 639         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 640
 641         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 642                 || prev_size       != mdev->ldev->md.md_size_sect;
 643
 644         if (la_size_changed || md_moved) {
 645                 int err;
 646
 647                 drbd_al_shrink(mdev); /* All extents inactive. */
 648                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 649                          la_size_changed && md_moved ? "size changed and md moved" :
 650                          la_size_changed ? "size changed" : "md moved");
 651                 err = drbd_bitmap_io(mdev, &drbd_bm_write, "size changed"); /* does drbd_resume_io() ! */
 652                 if (err) {
 653                         rv = dev_size_error;
 654                         goto out;
 655                 }
 656                 drbd_md_mark_dirty(mdev);
 657         }
 658
 659         if (size > la_size)
 660                 rv = grew;
 661         if (size < la_size)
 662                 rv = shrunk;
 663 out:
 664         lc_unlock(mdev->act_log);
 665         wake_up(&mdev->al_wait);
 666         drbd_resume_io(mdev);
 667
 668         return rv;
 669 }
 670
 671 sector_t
 672 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
 673 {
 674         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 675         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 676         sector_t m_size; /* my size */
 677         sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
 678         sector_t size = 0;
 679
 680         m_size = drbd_get_max_capacity(bdev);
 681
 682         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 683                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 684                 p_size = m_size;
 685         }
 686
 687         if (p_size && m_size) {
 688                 size = min_t(sector_t, p_size, m_size);
 689         } else {
 690                 if (la_size) {
 691                         size = la_size;
 692                         if (m_size && m_size < size)
 693                                 size = m_size;
 694                         if (p_size && p_size < size)
 695                                 size = p_size;
 696                 } else {
 697                         if (m_size)
 698                                 size = m_size;
 699                         if (p_size)
 700                                 size = p_size;
 701                 }
 702         }
 703
 704         if (size == 0)
 705                 dev_err(DEV, "Both nodes diskless!\n");
 706
 707         if (u_size) {
 708                 if (u_size > size)
 709                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 710                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 711                 else
 712                         size = u_size;
 713         }
 714
 715         return size;
 716 }
 717
 718 /**
 719  * drbd_check_al_size() - Ensures that the AL is of the right size
 720  * @mdev:       DRBD device.
 721  *
 722  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 723  * failed, and 0 on success. You should call drbd_md_sync() after you called
 724  * this function.
 725  */
 726 static int drbd_check_al_size(struct drbd_conf *mdev)
 727 {
 728         struct lru_cache *n, *t;
 729         struct lc_element *e;
 730         unsigned int in_use;
 731         int i;
 732
 733         ERR_IF(mdev->sync_conf.al_extents < 7)
 734                 mdev->sync_conf.al_extents = 127;
 735
 736         if (mdev->act_log &&
 737             mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
 738                 return 0;
 739
 740         in_use = 0;
 741         t = mdev->act_log;
 742         n = lc_create("act_log", drbd_al_ext_cache,
 743                 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
 744
 745         if (n == NULL) {
 746                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 747                 return -ENOMEM;
 748         }
 749         spin_lock_irq(&mdev->al_lock);
 750         if (t) {
 751                 for (i = 0; i < t->nr_elements; i++) {
 752                         e = lc_element_by_index(t, i);
 753                         if (e->refcnt)
 754                                 dev_err(DEV, "refcnt(%d)==%d\n",
 755                                     e->lc_number, e->refcnt);
 756                         in_use += e->refcnt;
 757                 }
 758         }
 759         if (!in_use)
 760                 mdev->act_log = n;
 761         spin_unlock_irq(&mdev->al_lock);
 762         if (in_use) {
 763                 dev_err(DEV, "Activity log still in use!\n");
 764                 lc_destroy(n);
 765                 return -EBUSY;
 766         } else {
 767                 if (t)
 768                         lc_destroy(t);
 769         }
 770         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 771         return 0;
 772 }
 773
 774 void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size) __must_hold(local)
 775 {
 776         struct request_queue * const q = mdev->rq_queue;
 777         struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 778         int max_segments = mdev->ldev->dc.max_bio_bvecs;
 779         int max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
 780
 781         blk_queue_logical_block_size(q, 512);
 782         blk_queue_max_hw_sectors(q, max_hw_sectors);
 783         /* This is the workaround for "bio would need to, but cannot, be split" */
 784         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
 785         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
 786         blk_queue_stack_limits(q, b);
 787
 788         dev_info(DEV, "max BIO size = %u\n", queue_max_hw_sectors(q) << 9);
 789
 790         if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
 791                 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
 792                      q->backing_dev_info.ra_pages,
 793                      b->backing_dev_info.ra_pages);
 794                 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
 795         }
 796 }
 797
 798 /* serialize deconfig (worker exiting, doing cleanup)
 799  * and reconfig (drbdsetup disk, drbdsetup net)
 800  *
 801  * Wait for a potentially exiting worker, then restart it,
 802  * or start a new one.  Flush any pending work, there may still be an
 803  * after_state_change queued.
 804  */
 805 static void drbd_reconfig_start(struct drbd_conf *mdev)
 806 {
 807         wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
 808         wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
 809         drbd_thread_start(&mdev->worker);
 810         drbd_flush_workqueue(mdev);
 811 }
 812
 813 /* if still unconfigured, stops worker again.
 814  * if configured now, clears CONFIG_PENDING.
 815  * wakes potential waiters */
 816 static void drbd_reconfig_done(struct drbd_conf *mdev)
 817 {
 818         spin_lock_irq(&mdev->req_lock);
 819         if (mdev->state.disk == D_DISKLESS &&
 820             mdev->state.conn == C_STANDALONE &&
 821             mdev->state.role == R_SECONDARY) {
 822                 set_bit(DEVICE_DYING, &mdev->flags);
 823                 drbd_thread_stop_nowait(&mdev->worker);
 824         } else
 825                 clear_bit(CONFIG_PENDING, &mdev->flags);
 826         spin_unlock_irq(&mdev->req_lock);
 827         wake_up(&mdev->state_wait);
 828 }
 829
 830 /* Make sure IO is suspended before calling this function(). */
 831 static void drbd_suspend_al(struct drbd_conf *mdev)
 832 {
 833         int s = 0;
 834
 835         if (lc_try_lock(mdev->act_log)) {
 836                 drbd_al_shrink(mdev);
 837                 lc_unlock(mdev->act_log);
 838         } else {
 839                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
 840                 return;
 841         }
 842
 843         spin_lock_irq(&mdev->req_lock);
 844         if (mdev->state.conn < C_CONNECTED)
 845                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
 846
 847         spin_unlock_irq(&mdev->req_lock);
 848
 849         if (s)
 850                 dev_info(DEV, "Suspended AL updates\n");
 851 }
 852
 853 /* does always return 0;
 854  * interesting return code is in reply->ret_code */
 855 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 856                              struct drbd_nl_cfg_reply *reply)
 857 {
 858         enum drbd_ret_code retcode;
 859         enum determine_dev_size dd;
 860         sector_t max_possible_sectors;
 861         sector_t min_md_device_sectors;
 862         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
 863         struct block_device *bdev;
 864         struct lru_cache *resync_lru = NULL;
 865         union drbd_state ns, os;
 866         unsigned int max_bio_size;
 867         enum drbd_state_rv rv;
 868         int cp_discovered = 0;
 869         int logical_block_size;
 870
 871         drbd_reconfig_start(mdev);
 872
 873         /* if you want to reconfigure, please tear down first */
 874         if (mdev->state.disk > D_DISKLESS) {
 875                 retcode = ERR_DISK_CONFIGURED;
 876                 goto fail;
 877         }
 878         /* It may just now have detached because of IO error.  Make sure
 879          * drbd_ldev_destroy is done already, we may end up here very fast,
 880          * e.g. if someone calls attach from the on-io-error handler,
 881          * to realize a "hot spare" feature (not that I'd recommend that) */
 882         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
 883
 884         /* allocation not in the IO path, cqueue thread context */
 885         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
 886         if (!nbc) {
 887                 retcode = ERR_NOMEM;
 888                 goto fail;
 889         }
 890
 891         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
 892         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
 893         nbc->dc.fencing       = DRBD_FENCING_DEF;
 894         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
 895
 896         if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
 897                 retcode = ERR_MANDATORY_TAG;
 898                 goto fail;
 899         }
 900
 901         if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
 902                 retcode = ERR_MD_IDX_INVALID;
 903                 goto fail;
 904         }
 905
 906         if (get_net_conf(mdev)) {
 907                 int prot = mdev->net_conf->wire_protocol;
 908                 put_net_conf(mdev);
 909                 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
 910                         retcode = ERR_STONITH_AND_PROT_A;
 911                         goto fail;
 912                 }
 913         }
 914
 915         bdev = blkdev_get_by_path(nbc->dc.backing_dev,
 916                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
 917         if (IS_ERR(bdev)) {
 918                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
 919                         PTR_ERR(bdev));
 920                 retcode = ERR_OPEN_DISK;
 921                 goto fail;
 922         }
 923         nbc->backing_bdev = bdev;
 924
 925         /*
 926          * meta_dev_idx >= 0: external fixed size, possibly multiple
 927          * drbd sharing one meta device.  TODO in that case, paranoia
 928          * check that [md_bdev, meta_dev_idx] is not yet used by some
 929          * other drbd minor!  (if you use drbd.conf + drbdadm, that
 930          * should check it for you already; but if you don't, or
 931          * someone fooled it, we need to double check here)
 932          */
 933         bdev = blkdev_get_by_path(nbc->dc.meta_dev,
 934                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
 935                                   (nbc->dc.meta_dev_idx < 0) ?
 936                                   (void *)mdev : (void *)drbd_m_holder);
 937         if (IS_ERR(bdev)) {
 938                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
 939                         PTR_ERR(bdev));
 940                 retcode = ERR_OPEN_MD_DISK;
 941                 goto fail;
 942         }
 943         nbc->md_bdev = bdev;
 944
 945         if ((nbc->backing_bdev == nbc->md_bdev) !=
 946             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
 947              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
 948                 retcode = ERR_MD_IDX_INVALID;
 949                 goto fail;
 950         }
 951
 952         resync_lru = lc_create("resync", drbd_bm_ext_cache,
 953                         61, sizeof(struct bm_extent),
 954                         offsetof(struct bm_extent, lce));
 955         if (!resync_lru) {
 956                 retcode = ERR_NOMEM;
 957                 goto fail;
 958         }
 959
 960         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
 961         drbd_md_set_sector_offsets(mdev, nbc);
 962
 963         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
 964                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
 965                         (unsigned long long) drbd_get_max_capacity(nbc),
 966                         (unsigned long long) nbc->dc.disk_size);
 967                 retcode = ERR_DISK_TO_SMALL;
 968                 goto fail;
 969         }
 970
 971         if (nbc->dc.meta_dev_idx < 0) {
 972                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
 973                 /* at least one MB, otherwise it does not make sense */
 974                 min_md_device_sectors = (2<<10);
 975         } else {
 976                 max_possible_sectors = DRBD_MAX_SECTORS;
 977                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
 978         }
 979
 980         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
 981                 retcode = ERR_MD_DISK_TO_SMALL;
 982                 dev_warn(DEV, "refusing attach: md-device too small, "
 983                      "at least %llu sectors needed for this meta-disk type\n",
 984                      (unsigned long long) min_md_device_sectors);
 985                 goto fail;
 986         }
 987
 988         /* Make sure the new disk is big enough
 989          * (we may currently be R_PRIMARY with no local disk...) */
 990         if (drbd_get_max_capacity(nbc) <
 991             drbd_get_capacity(mdev->this_bdev)) {
 992                 retcode = ERR_DISK_TO_SMALL;
 993                 goto fail;
 994         }
 995
 996         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
 997
 998         if (nbc->known_size > max_possible_sectors) {
 999                 dev_warn(DEV, "==> truncating very big lower level device "
1000                         "to currently maximum possible %llu sectors <==\n",
1001                         (unsigned long long) max_possible_sectors);
1002                 if (nbc->dc.meta_dev_idx >= 0)
1003                         dev_warn(DEV, "==>> using internal or flexible "
1004                                       "meta data may help <<==\n");
1005         }
1006
1007         drbd_suspend_io(mdev);
1008         /* also wait for the last barrier ack. */
1009         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state));
1010         /* and for any other previously queued work */
1011         drbd_flush_workqueue(mdev);
1012
1013         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1014         retcode = rv;  /* FIXME: Type mismatch. */
1015         drbd_resume_io(mdev);
1016         if (rv < SS_SUCCESS)
1017                 goto fail;
1018
1019         if (!get_ldev_if_state(mdev, D_ATTACHING))
1020                 goto force_diskless;
1021
1022         drbd_md_set_sector_offsets(mdev, nbc);
1023
1024         /* allocate a second IO page if logical_block_size != 512 */
1025         logical_block_size = bdev_logical_block_size(nbc->md_bdev);
1026         if (logical_block_size == 0)
1027                 logical_block_size = MD_SECTOR_SIZE;
1028
1029         if (logical_block_size != MD_SECTOR_SIZE) {
1030                 if (!mdev->md_io_tmpp) {
1031                         struct page *page = alloc_page(GFP_NOIO);
1032                         if (!page)
1033                                 goto force_diskless_dec;
1034
1035                         dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
1036                              logical_block_size, MD_SECTOR_SIZE);
1037                         dev_warn(DEV, "Workaround engaged (has performance impact).\n");
1038
1039                         mdev->md_io_tmpp = page;
1040                 }
1041         }
1042
1043         if (!mdev->bitmap) {
1044                 if (drbd_bm_init(mdev)) {
1045                         retcode = ERR_NOMEM;
1046                         goto force_diskless_dec;
1047                 }
1048         }
1049
1050         retcode = drbd_md_read(mdev, nbc);
1051         if (retcode != NO_ERROR)
1052                 goto force_diskless_dec;
1053
1054         if (mdev->state.conn < C_CONNECTED &&
1055             mdev->state.role == R_PRIMARY &&
1056             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1057                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1058                     (unsigned long long)mdev->ed_uuid);
1059                 retcode = ERR_DATA_NOT_CURRENT;
1060                 goto force_diskless_dec;
1061         }
1062
1063         /* Since we are diskless, fix the activity log first... */
1064         if (drbd_check_al_size(mdev)) {
1065                 retcode = ERR_NOMEM;
1066                 goto force_diskless_dec;
1067         }
1068
1069         /* Prevent shrinking of consistent devices ! */
1070         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1071             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1072                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1073                 retcode = ERR_DISK_TO_SMALL;
1074                 goto force_diskless_dec;
1075         }
1076
1077         if (!drbd_al_read_log(mdev, nbc)) {
1078                 retcode = ERR_IO_MD_DISK;
1079                 goto force_diskless_dec;
1080         }
1081
1082         /* Reset the "barriers don't work" bits here, then force meta data to
1083          * be written, to ensure we determine if barriers are supported. */
1084         if (nbc->dc.no_md_flush)
1085                 set_bit(MD_NO_FUA, &mdev->flags);
1086         else
1087                 clear_bit(MD_NO_FUA, &mdev->flags);
1088
1089         /* Point of no return reached.
1090          * Devices and memory are no longer released by error cleanup below.
1091          * now mdev takes over responsibility, and the state engine should
1092          * clean it up somewhere.  */
1093         D_ASSERT(mdev->ldev == NULL);
1094         mdev->ldev = nbc;
1095         mdev->resync = resync_lru;
1096         nbc = NULL;
1097         resync_lru = NULL;
1098
1099         mdev->write_ordering = WO_bdev_flush;
1100         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1101
1102         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1103                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1104         else
1105                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1106
1107         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1108             !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) {
1109                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1110                 cp_discovered = 1;
1111         }
1112
1113         mdev->send_cnt = 0;
1114         mdev->recv_cnt = 0;
1115         mdev->read_cnt = 0;
1116         mdev->writ_cnt = 0;
1117
1118         max_bio_size = DRBD_MAX_BIO_SIZE;
1119         if (mdev->state.conn == C_CONNECTED) {
1120                 /* We are Primary, Connected, and now attach a new local
1121                  * backing store. We must not increase the user visible maximum
1122                  * bio size on this device to something the peer may not be
1123                  * able to handle. */
1124                 if (mdev->agreed_pro_version < 94)
1125                         max_bio_size = queue_max_hw_sectors(mdev->rq_queue) << 9;
1126                 else if (mdev->agreed_pro_version == 94)
1127                         max_bio_size = DRBD_MAX_SIZE_H80_PACKET;
1128                 /* else: drbd 8.3.9 and later, stay with default */
1129         }
1130
1131         drbd_setup_queue_param(mdev, max_bio_size);
1132
1133         /* If I am currently not R_PRIMARY,
1134          * but meta data primary indicator is set,
1135          * I just now recover from a hard crash,
1136          * and have been R_PRIMARY before that crash.
1137          *
1138          * Now, if I had no connection before that crash
1139          * (have been degraded R_PRIMARY), chances are that
1140          * I won't find my peer now either.
1141          *
1142          * In that case, and _only_ in that case,
1143          * we use the degr-wfc-timeout instead of the default,
1144          * so we can automatically recover from a crash of a
1145          * degraded but active "cluster" after a certain timeout.
1146          */
1147         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1148         if (mdev->state.role != R_PRIMARY &&
1149              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1150             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1151                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1152
1153         dd = drbd_determin_dev_size(mdev, 0);
1154         if (dd == dev_size_error) {
1155                 retcode = ERR_NOMEM_BITMAP;
1156                 goto force_diskless_dec;
1157         } else if (dd == grew)
1158                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1159
1160         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1161                 dev_info(DEV, "Assuming that all blocks are out of sync "
1162                      "(aka FullSync)\n");
1163                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from attaching")) {
1164                         retcode = ERR_IO_MD_DISK;
1165                         goto force_diskless_dec;
1166                 }
1167         } else {
1168                 if (drbd_bitmap_io(mdev, &drbd_bm_read, "read from attaching") < 0) {
1169                         retcode = ERR_IO_MD_DISK;
1170                         goto force_diskless_dec;
1171                 }
1172         }
1173
1174         if (cp_discovered) {
1175                 drbd_al_apply_to_bm(mdev);
1176                 if (drbd_bitmap_io(mdev, &drbd_bm_write, "crashed primary apply AL")) {
1177                         retcode = ERR_IO_MD_DISK;
1178                         goto force_diskless_dec;
1179                 }
1180         }
1181
1182         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1183                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1184
1185         spin_lock_irq(&mdev->req_lock);
1186         os = mdev->state;
1187         ns.i = os.i;
1188         /* If MDF_CONSISTENT is not set go into inconsistent state,
1189            otherwise investigate MDF_WasUpToDate...
1190            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1191            otherwise into D_CONSISTENT state.
1192         */
1193         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1194                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1195                         ns.disk = D_CONSISTENT;
1196                 else
1197                         ns.disk = D_OUTDATED;
1198         } else {
1199                 ns.disk = D_INCONSISTENT;
1200         }
1201
1202         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1203                 ns.pdsk = D_OUTDATED;
1204
1205         if ( ns.disk == D_CONSISTENT &&
1206             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1207                 ns.disk = D_UP_TO_DATE;
1208
1209         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1210            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1211            this point, because drbd_request_state() modifies these
1212            flags. */
1213
1214         /* In case we are C_CONNECTED postpone any decision on the new disk
1215            state after the negotiation phase. */
1216         if (mdev->state.conn == C_CONNECTED) {
1217                 mdev->new_state_tmp.i = ns.i;
1218                 ns.i = os.i;
1219                 ns.disk = D_NEGOTIATING;
1220
1221                 /* We expect to receive up-to-date UUIDs soon.
1222                    To avoid a race in receive_state, free p_uuid while
1223                    holding req_lock. I.e. atomic with the state change */
1224                 kfree(mdev->p_uuid);
1225                 mdev->p_uuid = NULL;
1226         }
1227
1228         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1229         ns = mdev->state;
1230         spin_unlock_irq(&mdev->req_lock);
1231
1232         if (rv < SS_SUCCESS)
1233                 goto force_diskless_dec;
1234
1235         if (mdev->state.role == R_PRIMARY)
1236                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1237         else
1238                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1239
1240         drbd_md_mark_dirty(mdev);
1241         drbd_md_sync(mdev);
1242
1243         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1244         put_ldev(mdev);
1245         reply->ret_code = retcode;
1246         drbd_reconfig_done(mdev);
1247         return 0;
1248
1249  force_diskless_dec:
1250         put_ldev(mdev);
1251  force_diskless:
1252         drbd_force_state(mdev, NS(disk, D_FAILED));
1253         drbd_md_sync(mdev);
1254  fail:
1255         if (nbc) {
1256                 if (nbc->backing_bdev)
1257                         blkdev_put(nbc->backing_bdev,
1258                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1259                 if (nbc->md_bdev)
1260                         blkdev_put(nbc->md_bdev,
1261                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1262                 kfree(nbc);
1263         }
1264         lc_destroy(resync_lru);
1265
1266         reply->ret_code = retcode;
1267         drbd_reconfig_done(mdev);
1268         return 0;
1269 }
1270
1271 /* Detaching the disk is a process in multiple stages.  First we need to lock
1272  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1273  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1274  * internal references as well.
1275  * Only then we have finally detached. */
1276 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1277                           struct drbd_nl_cfg_reply *reply)
1278 {
1279         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1280         reply->ret_code = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1281         if (mdev->state.disk == D_DISKLESS)
1282                 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1283         drbd_resume_io(mdev);
1284         return 0;
1285 }
1286
1287 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1288                             struct drbd_nl_cfg_reply *reply)
1289 {
1290         int i, ns;
1291         enum drbd_ret_code retcode;
1292         struct net_conf *new_conf = NULL;
1293         struct crypto_hash *tfm = NULL;
1294         struct crypto_hash *integrity_w_tfm = NULL;
1295         struct crypto_hash *integrity_r_tfm = NULL;
1296         struct hlist_head *new_tl_hash = NULL;
1297         struct hlist_head *new_ee_hash = NULL;
1298         struct drbd_conf *odev;
1299         char hmac_name[CRYPTO_MAX_ALG_NAME];
1300         void *int_dig_out = NULL;
1301         void *int_dig_in = NULL;
1302         void *int_dig_vv = NULL;
1303         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1304
1305         drbd_reconfig_start(mdev);
1306
1307         if (mdev->state.conn > C_STANDALONE) {
1308                 retcode = ERR_NET_CONFIGURED;
1309                 goto fail;
1310         }
1311
1312         /* allocation not in the IO path, cqueue thread context */
1313         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1314         if (!new_conf) {
1315                 retcode = ERR_NOMEM;
1316                 goto fail;
1317         }
1318
1319         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1320         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1321         new_conf->ping_int         = DRBD_PING_INT_DEF;
1322         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1323         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1324         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1325         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1326         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1327         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1328         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1329         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1330         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1331         new_conf->want_lose        = 0;
1332         new_conf->two_primaries    = 0;
1333         new_conf->wire_protocol    = DRBD_PROT_C;
1334         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1335         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1336         new_conf->on_congestion    = DRBD_ON_CONGESTION_DEF;
1337         new_conf->cong_extents     = DRBD_CONG_EXTENTS_DEF;
1338
1339         if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1340                 retcode = ERR_MANDATORY_TAG;
1341                 goto fail;
1342         }
1343
1344         if (new_conf->two_primaries
1345             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1346                 retcode = ERR_NOT_PROTO_C;
1347                 goto fail;
1348         }
1349
1350         if (get_ldev(mdev)) {
1351                 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1352                 put_ldev(mdev);
1353                 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1354                         retcode = ERR_STONITH_AND_PROT_A;
1355                         goto fail;
1356                 }
1357         }
1358
1359         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1360                 retcode = ERR_CONG_NOT_PROTO_A;
1361                 goto fail;
1362         }
1363
1364         if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1365                 retcode = ERR_DISCARD;
1366                 goto fail;
1367         }
1368
1369         retcode = NO_ERROR;
1370
1371         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1372         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1373         for (i = 0; i < minor_count; i++) {
1374                 odev = minor_to_mdev(i);
1375                 if (!odev || odev == mdev)
1376                         continue;
1377                 if (get_net_conf(odev)) {
1378                         taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1379                         if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1380                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1381                                 retcode = ERR_LOCAL_ADDR;
1382
1383                         taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1384                         if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1385                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1386                                 retcode = ERR_PEER_ADDR;
1387
1388                         put_net_conf(odev);
1389                         if (retcode != NO_ERROR)
1390                                 goto fail;
1391                 }
1392         }
1393
1394         if (new_conf->cram_hmac_alg[0] != 0) {
1395                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1396                         new_conf->cram_hmac_alg);
1397                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1398                 if (IS_ERR(tfm)) {
1399                         tfm = NULL;
1400                         retcode = ERR_AUTH_ALG;
1401                         goto fail;
1402                 }
1403
1404                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1405                         retcode = ERR_AUTH_ALG_ND;
1406                         goto fail;
1407                 }
1408         }
1409
1410         if (new_conf->integrity_alg[0]) {
1411                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1412                 if (IS_ERR(integrity_w_tfm)) {
1413                         integrity_w_tfm = NULL;
1414                         retcode=ERR_INTEGRITY_ALG;
1415                         goto fail;
1416                 }
1417
1418                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1419                         retcode=ERR_INTEGRITY_ALG_ND;
1420                         goto fail;
1421                 }
1422
1423                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1424                 if (IS_ERR(integrity_r_tfm)) {
1425                         integrity_r_tfm = NULL;
1426                         retcode=ERR_INTEGRITY_ALG;
1427                         goto fail;
1428                 }
1429         }
1430
1431         ns = new_conf->max_epoch_size/8;
1432         if (mdev->tl_hash_s != ns) {
1433                 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1434                 if (!new_tl_hash) {
1435                         retcode = ERR_NOMEM;
1436                         goto fail;
1437                 }
1438         }
1439
1440         ns = new_conf->max_buffers/8;
1441         if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1442                 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1443                 if (!new_ee_hash) {
1444                         retcode = ERR_NOMEM;
1445                         goto fail;
1446                 }
1447         }
1448
1449         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1450
1451         if (integrity_w_tfm) {
1452                 i = crypto_hash_digestsize(integrity_w_tfm);
1453                 int_dig_out = kmalloc(i, GFP_KERNEL);
1454                 if (!int_dig_out) {
1455                         retcode = ERR_NOMEM;
1456                         goto fail;
1457                 }
1458                 int_dig_in = kmalloc(i, GFP_KERNEL);
1459                 if (!int_dig_in) {
1460                         retcode = ERR_NOMEM;
1461                         goto fail;
1462                 }
1463                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1464                 if (!int_dig_vv) {
1465                         retcode = ERR_NOMEM;
1466                         goto fail;
1467                 }
1468         }
1469
1470         if (!mdev->bitmap) {
1471                 if(drbd_bm_init(mdev)) {
1472                         retcode = ERR_NOMEM;
1473                         goto fail;
1474                 }
1475         }
1476
1477         drbd_flush_workqueue(mdev);
1478         spin_lock_irq(&mdev->req_lock);
1479         if (mdev->net_conf != NULL) {
1480                 retcode = ERR_NET_CONFIGURED;
1481                 spin_unlock_irq(&mdev->req_lock);
1482                 goto fail;
1483         }
1484         mdev->net_conf = new_conf;
1485
1486         mdev->send_cnt = 0;
1487         mdev->recv_cnt = 0;
1488
1489         if (new_tl_hash) {
1490                 kfree(mdev->tl_hash);
1491                 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1492                 mdev->tl_hash = new_tl_hash;
1493         }
1494
1495         if (new_ee_hash) {
1496                 kfree(mdev->ee_hash);
1497                 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1498                 mdev->ee_hash = new_ee_hash;
1499         }
1500
1501         crypto_free_hash(mdev->cram_hmac_tfm);
1502         mdev->cram_hmac_tfm = tfm;
1503
1504         crypto_free_hash(mdev->integrity_w_tfm);
1505         mdev->integrity_w_tfm = integrity_w_tfm;
1506
1507         crypto_free_hash(mdev->integrity_r_tfm);
1508         mdev->integrity_r_tfm = integrity_r_tfm;
1509
1510         kfree(mdev->int_dig_out);
1511         kfree(mdev->int_dig_in);
1512         kfree(mdev->int_dig_vv);
1513         mdev->int_dig_out=int_dig_out;
1514         mdev->int_dig_in=int_dig_in;
1515         mdev->int_dig_vv=int_dig_vv;
1516         retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL);
1517         spin_unlock_irq(&mdev->req_lock);
1518
1519         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1520         reply->ret_code = retcode;
1521         drbd_reconfig_done(mdev);
1522         return 0;
1523
1524 fail:
1525         kfree(int_dig_out);
1526         kfree(int_dig_in);
1527         kfree(int_dig_vv);
1528         crypto_free_hash(tfm);
1529         crypto_free_hash(integrity_w_tfm);
1530         crypto_free_hash(integrity_r_tfm);
1531         kfree(new_tl_hash);
1532         kfree(new_ee_hash);
1533         kfree(new_conf);
1534
1535         reply->ret_code = retcode;
1536         drbd_reconfig_done(mdev);
1537         return 0;
1538 }
1539
1540 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1541                               struct drbd_nl_cfg_reply *reply)
1542 {
1543         int retcode;
1544         struct disconnect dc;
1545
1546         memset(&dc, 0, sizeof(struct disconnect));
1547         if (!disconnect_from_tags(mdev, nlp->tag_list, &dc)) {
1548                 retcode = ERR_MANDATORY_TAG;
1549                 goto fail;
1550         }
1551
1552         if (dc.force) {
1553                 spin_lock_irq(&mdev->req_lock);
1554                 if (mdev->state.conn >= C_WF_CONNECTION)
1555                         _drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), CS_HARD, NULL);
1556                 spin_unlock_irq(&mdev->req_lock);
1557                 goto done;
1558         }
1559
1560         retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1561
1562         if (retcode == SS_NOTHING_TO_DO)
1563                 goto done;
1564         else if (retcode == SS_ALREADY_STANDALONE)
1565                 goto done;
1566         else if (retcode == SS_PRIMARY_NOP) {
1567                 /* Our statche checking code wants to see the peer outdated. */
1568                 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1569                                                       pdsk, D_OUTDATED));
1570         } else if (retcode == SS_CW_FAILED_BY_PEER) {
1571                 /* The peer probably wants to see us outdated. */
1572                 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1573                                                         disk, D_OUTDATED),
1574                                               CS_ORDERED);
1575                 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1576                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1577                         retcode = SS_SUCCESS;
1578                 }
1579         }
1580
1581         if (retcode < SS_SUCCESS)
1582                 goto fail;
1583
1584         if (wait_event_interruptible(mdev->state_wait,
1585                                      mdev->state.conn != C_DISCONNECTING)) {
1586                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1587                    someone else might connect us in the mean time! */
1588                 retcode = ERR_INTR;
1589                 goto fail;
1590         }
1591
1592  done:
1593         retcode = NO_ERROR;
1594  fail:
1595         drbd_md_sync(mdev);
1596         reply->ret_code = retcode;
1597         return 0;
1598 }
1599
1600 void resync_after_online_grow(struct drbd_conf *mdev)
1601 {
1602         int iass; /* I am sync source */
1603
1604         dev_info(DEV, "Resync of new storage after online grow\n");
1605         if (mdev->state.role != mdev->state.peer)
1606                 iass = (mdev->state.role == R_PRIMARY);
1607         else
1608                 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1609
1610         if (iass)
1611                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1612         else
1613                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1614 }
1615
1616 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1617                           struct drbd_nl_cfg_reply *reply)
1618 {
1619         struct resize rs;
1620         int retcode = NO_ERROR;
1621         enum determine_dev_size dd;
1622         enum dds_flags ddsf;
1623
1624         memset(&rs, 0, sizeof(struct resize));
1625         if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1626                 retcode = ERR_MANDATORY_TAG;
1627                 goto fail;
1628         }
1629
1630         if (mdev->state.conn > C_CONNECTED) {
1631                 retcode = ERR_RESIZE_RESYNC;
1632                 goto fail;
1633         }
1634
1635         if (mdev->state.role == R_SECONDARY &&
1636             mdev->state.peer == R_SECONDARY) {
1637                 retcode = ERR_NO_PRIMARY;
1638                 goto fail;
1639         }
1640
1641         if (!get_ldev(mdev)) {
1642                 retcode = ERR_NO_DISK;
1643                 goto fail;
1644         }
1645
1646         if (rs.no_resync && mdev->agreed_pro_version < 93) {
1647                 retcode = ERR_NEED_APV_93;
1648                 goto fail;
1649         }
1650
1651         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1652                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1653
1654         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1655         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1656         dd = drbd_determin_dev_size(mdev, ddsf);
1657         drbd_md_sync(mdev);
1658         put_ldev(mdev);
1659         if (dd == dev_size_error) {
1660                 retcode = ERR_NOMEM_BITMAP;
1661                 goto fail;
1662         }
1663
1664         if (mdev->state.conn == C_CONNECTED) {
1665                 if (dd == grew)
1666                         set_bit(RESIZE_PENDING, &mdev->flags);
1667
1668                 drbd_send_uuids(mdev);
1669                 drbd_send_sizes(mdev, 1, ddsf);
1670         }
1671
1672  fail:
1673         reply->ret_code = retcode;
1674         return 0;
1675 }
1676
1677 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1678                                struct drbd_nl_cfg_reply *reply)
1679 {
1680         int retcode = NO_ERROR;
1681         int err;
1682         int ovr; /* online verify running */
1683         int rsr; /* re-sync running */
1684         struct crypto_hash *verify_tfm = NULL;
1685         struct crypto_hash *csums_tfm = NULL;
1686         struct syncer_conf sc;
1687         cpumask_var_t new_cpu_mask;
1688         int *rs_plan_s = NULL;
1689         int fifo_size;
1690
1691         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1692                 retcode = ERR_NOMEM;
1693                 goto fail;
1694         }
1695
1696         if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1697                 memset(&sc, 0, sizeof(struct syncer_conf));
1698                 sc.rate       = DRBD_RATE_DEF;
1699                 sc.after      = DRBD_AFTER_DEF;
1700                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1701                 sc.on_no_data  = DRBD_ON_NO_DATA_DEF;
1702                 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1703                 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1704                 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1705                 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1706                 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1707         } else
1708                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1709
1710         if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1711                 retcode = ERR_MANDATORY_TAG;
1712                 goto fail;
1713         }
1714
1715         /* re-sync running */
1716         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1717                 mdev->state.conn == C_SYNC_TARGET ||
1718                 mdev->state.conn == C_PAUSED_SYNC_S ||
1719                 mdev->state.conn == C_PAUSED_SYNC_T );
1720
1721         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1722                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1723                 goto fail;
1724         }
1725
1726         if (!rsr && sc.csums_alg[0]) {
1727                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1728                 if (IS_ERR(csums_tfm)) {
1729                         csums_tfm = NULL;
1730                         retcode = ERR_CSUMS_ALG;
1731                         goto fail;
1732                 }
1733
1734                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1735                         retcode = ERR_CSUMS_ALG_ND;
1736                         goto fail;
1737                 }
1738         }
1739
1740         /* online verify running */
1741         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1742
1743         if (ovr) {
1744                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1745                         retcode = ERR_VERIFY_RUNNING;
1746                         goto fail;
1747                 }
1748         }
1749
1750         if (!ovr && sc.verify_alg[0]) {
1751                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1752                 if (IS_ERR(verify_tfm)) {
1753                         verify_tfm = NULL;
1754                         retcode = ERR_VERIFY_ALG;
1755                         goto fail;
1756                 }
1757
1758                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1759                         retcode = ERR_VERIFY_ALG_ND;
1760                         goto fail;
1761                 }
1762         }
1763
1764         /* silently ignore cpu mask on UP kernel */
1765         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1766                 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1767                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1768                 if (err) {
1769                         dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1770                         retcode = ERR_CPU_MASK_PARSE;
1771                         goto fail;
1772                 }
1773         }
1774
1775         ERR_IF (sc.rate < 1) sc.rate = 1;
1776         ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1777 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1778         if (sc.al_extents > AL_MAX) {
1779                 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1780                 sc.al_extents = AL_MAX;
1781         }
1782 #undef AL_MAX
1783
1784         /* to avoid spurious errors when configuring minors before configuring
1785          * the minors they depend on: if necessary, first create the minor we
1786          * depend on */
1787         if (sc.after >= 0)
1788                 ensure_mdev(sc.after, 1);
1789
1790         /* most sanity checks done, try to assign the new sync-after
1791          * dependency.  need to hold the global lock in there,
1792          * to avoid a race in the dependency loop check. */
1793         retcode = drbd_alter_sa(mdev, sc.after);
1794         if (retcode != NO_ERROR)
1795                 goto fail;
1796
1797         fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1798         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1799                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1800                 if (!rs_plan_s) {
1801                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1802                         retcode = ERR_NOMEM;
1803                         goto fail;
1804                 }
1805         }
1806
1807         /* ok, assign the rest of it as well.
1808          * lock against receive_SyncParam() */
1809         spin_lock(&mdev->peer_seq_lock);
1810         mdev->sync_conf = sc;
1811
1812         if (!rsr) {
1813                 crypto_free_hash(mdev->csums_tfm);
1814                 mdev->csums_tfm = csums_tfm;
1815                 csums_tfm = NULL;
1816         }
1817
1818         if (!ovr) {
1819                 crypto_free_hash(mdev->verify_tfm);
1820                 mdev->verify_tfm = verify_tfm;
1821                 verify_tfm = NULL;
1822         }
1823
1824         if (fifo_size != mdev->rs_plan_s.size) {
1825                 kfree(mdev->rs_plan_s.values);
1826                 mdev->rs_plan_s.values = rs_plan_s;
1827                 mdev->rs_plan_s.size   = fifo_size;
1828                 mdev->rs_planed = 0;
1829                 rs_plan_s = NULL;
1830         }
1831
1832         spin_unlock(&mdev->peer_seq_lock);
1833
1834         if (get_ldev(mdev)) {
1835                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1836                 drbd_al_shrink(mdev);
1837                 err = drbd_check_al_size(mdev);
1838                 lc_unlock(mdev->act_log);
1839                 wake_up(&mdev->al_wait);
1840
1841                 put_ldev(mdev);
1842                 drbd_md_sync(mdev);
1843
1844                 if (err) {
1845                         retcode = ERR_NOMEM;
1846                         goto fail;
1847                 }
1848         }
1849
1850         if (mdev->state.conn >= C_CONNECTED)
1851                 drbd_send_sync_param(mdev, &sc);
1852
1853         if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1854                 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1855                 drbd_calc_cpu_mask(mdev);
1856                 mdev->receiver.reset_cpu_mask = 1;
1857                 mdev->asender.reset_cpu_mask = 1;
1858                 mdev->worker.reset_cpu_mask = 1;
1859         }
1860
1861         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1862 fail:
1863         kfree(rs_plan_s);
1864         free_cpumask_var(new_cpu_mask);
1865         crypto_free_hash(csums_tfm);
1866         crypto_free_hash(verify_tfm);
1867         reply->ret_code = retcode;
1868         return 0;
1869 }
1870
1871 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1872                               struct drbd_nl_cfg_reply *reply)
1873 {
1874         int retcode;
1875
1876         /* If there is still bitmap IO pending, probably because of a previous
1877          * resync just being finished, wait for it before requesting a new resync. */
1878         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1879
1880         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1881
1882         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1883                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1884
1885         while (retcode == SS_NEED_CONNECTION) {
1886                 spin_lock_irq(&mdev->req_lock);
1887                 if (mdev->state.conn < C_CONNECTED)
1888                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1889                 spin_unlock_irq(&mdev->req_lock);
1890
1891                 if (retcode != SS_NEED_CONNECTION)
1892                         break;
1893
1894                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1895         }
1896
1897         reply->ret_code = retcode;
1898         return 0;
1899 }
1900
1901 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
1902 {
1903         int rv;
1904
1905         rv = drbd_bmio_set_n_write(mdev);
1906         drbd_suspend_al(mdev);
1907         return rv;
1908 }
1909
1910 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1911                                    struct drbd_nl_cfg_reply *reply)
1912 {
1913         int retcode;
1914
1915         /* If there is still bitmap IO pending, probably because of a previous
1916          * resync just being finished, wait for it before requesting a new resync. */
1917         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1918
1919         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
1920
1921         if (retcode < SS_SUCCESS) {
1922                 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
1923                         /* The peer will get a resync upon connect anyways. Just make that
1924                            into a full resync. */
1925                         retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
1926                         if (retcode >= SS_SUCCESS) {
1927                                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
1928                                                    "set_n_write from invalidate_peer"))
1929                                         retcode = ERR_IO_MD_DISK;
1930                         }
1931                 } else
1932                         retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
1933         }
1934
1935         reply->ret_code = retcode;
1936         return 0;
1937 }
1938
1939 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1940                               struct drbd_nl_cfg_reply *reply)
1941 {
1942         int retcode = NO_ERROR;
1943
1944         if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
1945                 retcode = ERR_PAUSE_IS_SET;
1946
1947         reply->ret_code = retcode;
1948         return 0;
1949 }
1950
1951 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1952                                struct drbd_nl_cfg_reply *reply)
1953 {
1954         int retcode = NO_ERROR;
1955         union drbd_state s;
1956
1957         if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
1958                 s = mdev->state;
1959                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
1960                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
1961                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
1962                 } else {
1963                         retcode = ERR_PAUSE_IS_CLEAR;
1964                 }
1965         }
1966
1967         reply->ret_code = retcode;
1968         return 0;
1969 }
1970
1971 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1972                               struct drbd_nl_cfg_reply *reply)
1973 {
1974         reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
1975
1976         return 0;
1977 }
1978
1979 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1980                              struct drbd_nl_cfg_reply *reply)
1981 {
1982         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
1983                 drbd_uuid_new_current(mdev);
1984                 clear_bit(NEW_CUR_UUID, &mdev->flags);
1985         }
1986         drbd_suspend_io(mdev);
1987         reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
1988         if (reply->ret_code == SS_SUCCESS) {
1989                 if (mdev->state.conn < C_CONNECTED)
1990                         tl_clear(mdev);
1991                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
1992                         tl_restart(mdev, fail_frozen_disk_io);
1993         }
1994         drbd_resume_io(mdev);
1995
1996         return 0;
1997 }
1998
1999 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2000                            struct drbd_nl_cfg_reply *reply)
2001 {
2002         reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
2003         return 0;
2004 }
2005
2006 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2007                            struct drbd_nl_cfg_reply *reply)
2008 {
2009         unsigned short *tl;
2010
2011         tl = reply->tag_list;
2012
2013         if (get_ldev(mdev)) {
2014                 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
2015                 put_ldev(mdev);
2016         }
2017
2018         if (get_net_conf(mdev)) {
2019                 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
2020                 put_net_conf(mdev);
2021         }
2022         tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
2023
2024         put_unaligned(TT_END, tl++); /* Close the tag list */
2025
2026         return (int)((char *)tl - (char *)reply->tag_list);
2027 }
2028
2029 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2030                              struct drbd_nl_cfg_reply *reply)
2031 {
2032         unsigned short *tl = reply->tag_list;
2033         union drbd_state s = mdev->state;
2034         unsigned long rs_left;
2035         unsigned int res;
2036
2037         tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
2038
2039         /* no local ref, no bitmap, no syncer progress. */
2040         if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
2041                 if (get_ldev(mdev)) {
2042                         drbd_get_syncer_progress(mdev, &rs_left, &res);
2043                         tl = tl_add_int(tl, T_sync_progress, &res);
2044                         put_ldev(mdev);
2045                 }
2046         }
2047         put_unaligned(TT_END, tl++); /* Close the tag list */
2048
2049         return (int)((char *)tl - (char *)reply->tag_list);
2050 }
2051
2052 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2053                              struct drbd_nl_cfg_reply *reply)
2054 {
2055         unsigned short *tl;
2056
2057         tl = reply->tag_list;
2058
2059         if (get_ldev(mdev)) {
2060                 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
2061                 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
2062                 put_ldev(mdev);
2063         }
2064         put_unaligned(TT_END, tl++); /* Close the tag list */
2065
2066         return (int)((char *)tl - (char *)reply->tag_list);
2067 }
2068
2069 /**
2070  * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
2071  * @mdev:       DRBD device.
2072  * @nlp:        Netlink/connector packet from drbdsetup
2073  * @reply:      Reply packet for drbdsetup
2074  */
2075 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2076                                     struct drbd_nl_cfg_reply *reply)
2077 {
2078         unsigned short *tl;
2079         char rv;
2080
2081         tl = reply->tag_list;
2082
2083         rv = mdev->state.pdsk == D_OUTDATED        ? UT_PEER_OUTDATED :
2084           test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
2085
2086         tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
2087         put_unaligned(TT_END, tl++); /* Close the tag list */
2088
2089         return (int)((char *)tl - (char *)reply->tag_list);
2090 }
2091
2092 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2093                                     struct drbd_nl_cfg_reply *reply)
2094 {
2095         /* default to resume from last known position, if possible */
2096         struct start_ov args =
2097                 { .start_sector = mdev->ov_start_sector };
2098
2099         if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2100                 reply->ret_code = ERR_MANDATORY_TAG;
2101                 return 0;
2102         }
2103         /* w_make_ov_request expects position to be aligned */
2104         mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
2105         reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2106         return 0;
2107 }
2108
2109
2110 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2111                               struct drbd_nl_cfg_reply *reply)
2112 {
2113         int retcode = NO_ERROR;
2114         int skip_initial_sync = 0;
2115         int err;
2116
2117         struct new_c_uuid args;
2118
2119         memset(&args, 0, sizeof(struct new_c_uuid));
2120         if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2121                 reply->ret_code = ERR_MANDATORY_TAG;
2122                 return 0;
2123         }
2124
2125         mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
2126
2127         if (!get_ldev(mdev)) {
2128                 retcode = ERR_NO_DISK;
2129                 goto out;
2130         }
2131
2132         /* this is "skip initial sync", assume to be clean */
2133         if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
2134             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2135                 dev_info(DEV, "Preparing to skip initial sync\n");
2136                 skip_initial_sync = 1;
2137         } else if (mdev->state.conn != C_STANDALONE) {
2138                 retcode = ERR_CONNECTED;
2139                 goto out_dec;
2140         }
2141
2142         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2143         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2144
2145         if (args.clear_bm) {
2146                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, "clear_n_write from new_c_uuid");
2147                 if (err) {
2148                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2149                         retcode = ERR_IO_MD_DISK;
2150                 }
2151                 if (skip_initial_sync) {
2152                         drbd_send_uuids_skip_initial_sync(mdev);
2153                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2154                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2155                         spin_lock_irq(&mdev->req_lock);
2156                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2157                                         CS_VERBOSE, NULL);
2158                         spin_unlock_irq(&mdev->req_lock);
2159                 }
2160         }
2161
2162         drbd_md_sync(mdev);
2163 out_dec:
2164         put_ldev(mdev);
2165 out:
2166         mutex_unlock(&mdev->state_mutex);
2167
2168         reply->ret_code = retcode;
2169         return 0;
2170 }
2171
2172 struct cn_handler_struct {
2173         int (*function)(struct drbd_conf *,
2174                          struct drbd_nl_cfg_req *,
2175                          struct drbd_nl_cfg_reply *);
2176         int reply_body_size;
2177 };
2178
2179 static struct cn_handler_struct cnd_table[] = {
2180         [ P_primary ]           = { &drbd_nl_primary,           0 },
2181         [ P_secondary ]         = { &drbd_nl_secondary,         0 },
2182         [ P_disk_conf ]         = { &drbd_nl_disk_conf,         0 },
2183         [ P_detach ]            = { &drbd_nl_detach,            0 },
2184         [ P_net_conf ]          = { &drbd_nl_net_conf,          0 },
2185         [ P_disconnect ]        = { &drbd_nl_disconnect,        0 },
2186         [ P_resize ]            = { &drbd_nl_resize,            0 },
2187         [ P_syncer_conf ]       = { &drbd_nl_syncer_conf,       0 },
2188         [ P_invalidate ]        = { &drbd_nl_invalidate,        0 },
2189         [ P_invalidate_peer ]   = { &drbd_nl_invalidate_peer,   0 },
2190         [ P_pause_sync ]        = { &drbd_nl_pause_sync,        0 },
2191         [ P_resume_sync ]       = { &drbd_nl_resume_sync,       0 },
2192         [ P_suspend_io ]        = { &drbd_nl_suspend_io,        0 },
2193         [ P_resume_io ]         = { &drbd_nl_resume_io,         0 },
2194         [ P_outdate ]           = { &drbd_nl_outdate,           0 },
2195         [ P_get_config ]        = { &drbd_nl_get_config,
2196                                     sizeof(struct syncer_conf_tag_len_struct) +
2197                                     sizeof(struct disk_conf_tag_len_struct) +
2198                                     sizeof(struct net_conf_tag_len_struct) },
2199         [ P_get_state ]         = { &drbd_nl_get_state,
2200                                     sizeof(struct get_state_tag_len_struct) +
2201                                     sizeof(struct sync_progress_tag_len_struct) },
2202         [ P_get_uuids ]         = { &drbd_nl_get_uuids,
2203                                     sizeof(struct get_uuids_tag_len_struct) },
2204         [ P_get_timeout_flag ]  = { &drbd_nl_get_timeout_flag,
2205                                     sizeof(struct get_timeout_flag_tag_len_struct)},
2206         [ P_start_ov ]          = { &drbd_nl_start_ov,          0 },
2207         [ P_new_c_uuid ]        = { &drbd_nl_new_c_uuid,        0 },
2208 };
2209
2210 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2211 {
2212         struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2213         struct cn_handler_struct *cm;
2214         struct cn_msg *cn_reply;
2215         struct drbd_nl_cfg_reply *reply;
2216         struct drbd_conf *mdev;
2217         int retcode, rr;
2218         int reply_size = sizeof(struct cn_msg)
2219                 + sizeof(struct drbd_nl_cfg_reply)
2220                 + sizeof(short int);
2221
2222         if (!try_module_get(THIS_MODULE)) {
2223                 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2224                 return;
2225         }
2226
2227         if (!cap_raised(nsp->eff_cap, CAP_SYS_ADMIN)) {
2228                 retcode = ERR_PERM;
2229                 goto fail;
2230         }
2231
2232         mdev = ensure_mdev(nlp->drbd_minor,
2233                         (nlp->flags & DRBD_NL_CREATE_DEVICE));
2234         if (!mdev) {
2235                 retcode = ERR_MINOR_INVALID;
2236                 goto fail;
2237         }
2238
2239         if (nlp->packet_type >= P_nl_after_last_packet ||
2240             nlp->packet_type == P_return_code_only) {
2241                 retcode = ERR_PACKET_NR;
2242                 goto fail;
2243         }
2244
2245         cm = cnd_table + nlp->packet_type;
2246
2247         /* This may happen if packet number is 0: */
2248         if (cm->function == NULL) {
2249                 retcode = ERR_PACKET_NR;
2250                 goto fail;
2251         }
2252
2253         reply_size += cm->reply_body_size;
2254
2255         /* allocation not in the IO path, cqueue thread context */
2256         cn_reply = kzalloc(reply_size, GFP_KERNEL);
2257         if (!cn_reply) {
2258                 retcode = ERR_NOMEM;
2259                 goto fail;
2260         }
2261         reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2262
2263         reply->packet_type =
2264                 cm->reply_body_size ? nlp->packet_type : P_return_code_only;
2265         reply->minor = nlp->drbd_minor;
2266         reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2267         /* reply->tag_list; might be modified by cm->function. */
2268
2269         rr = cm->function(mdev, nlp, reply);
2270
2271         cn_reply->id = req->id;
2272         cn_reply->seq = req->seq;
2273         cn_reply->ack = req->ack  + 1;
2274         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2275         cn_reply->flags = 0;
2276
2277         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2278         if (rr && rr != -ESRCH)
2279                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2280
2281         kfree(cn_reply);
2282         module_put(THIS_MODULE);
2283         return;
2284  fail:
2285         drbd_nl_send_reply(req, retcode);
2286         module_put(THIS_MODULE);
2287 }
2288
2289 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2290
2291 static unsigned short *
2292 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2293         unsigned short len, int nul_terminated)
2294 {
2295         unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2296         len = (len < l) ? len :  l;
2297         put_unaligned(tag, tl++);
2298         put_unaligned(len, tl++);
2299         memcpy(tl, data, len);
2300         tl = (unsigned short*)((char*)tl + len);
2301         if (nul_terminated)
2302                 *((char*)tl - 1) = 0;
2303         return tl;
2304 }
2305
2306 static unsigned short *
2307 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2308 {
2309         return __tl_add_blob(tl, tag, data, len, 0);
2310 }
2311
2312 static unsigned short *
2313 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2314 {
2315         return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2316 }
2317
2318 static unsigned short *
2319 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2320 {
2321         put_unaligned(tag, tl++);
2322         switch(tag_type(tag)) {
2323         case TT_INTEGER:
2324                 put_unaligned(sizeof(int), tl++);
2325                 put_unaligned(*(int *)val, (int *)tl);
2326                 tl = (unsigned short*)((char*)tl+sizeof(int));
2327                 break;
2328         case TT_INT64:
2329                 put_unaligned(sizeof(u64), tl++);
2330                 put_unaligned(*(u64 *)val, (u64 *)tl);
2331                 tl = (unsigned short*)((char*)tl+sizeof(u64));
2332                 break;
2333         default:
2334                 /* someone did something stupid. */
2335                 ;
2336         }
2337         return tl;
2338 }
2339
2340 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2341 {
2342         char buffer[sizeof(struct cn_msg)+
2343                     sizeof(struct drbd_nl_cfg_reply)+
2344                     sizeof(struct get_state_tag_len_struct)+
2345                     sizeof(short int)];
2346         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2347         struct drbd_nl_cfg_reply *reply =
2348                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2349         unsigned short *tl = reply->tag_list;
2350
2351         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2352
2353         tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2354
2355         put_unaligned(TT_END, tl++); /* Close the tag list */
2356
2357         cn_reply->id.idx = CN_IDX_DRBD;
2358         cn_reply->id.val = CN_VAL_DRBD;
2359
2360         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2361         cn_reply->ack = 0; /* not used here. */
2362         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2363                 (int)((char *)tl - (char *)reply->tag_list);
2364         cn_reply->flags = 0;
2365
2366         reply->packet_type = P_get_state;
2367         reply->minor = mdev_to_minor(mdev);
2368         reply->ret_code = NO_ERROR;
2369
2370         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2371 }
2372
2373 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2374 {
2375         char buffer[sizeof(struct cn_msg)+
2376                     sizeof(struct drbd_nl_cfg_reply)+
2377                     sizeof(struct call_helper_tag_len_struct)+
2378                     sizeof(short int)];
2379         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2380         struct drbd_nl_cfg_reply *reply =
2381                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2382         unsigned short *tl = reply->tag_list;
2383
2384         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2385
2386         tl = tl_add_str(tl, T_helper, helper_name);
2387         put_unaligned(TT_END, tl++); /* Close the tag list */
2388
2389         cn_reply->id.idx = CN_IDX_DRBD;
2390         cn_reply->id.val = CN_VAL_DRBD;
2391
2392         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2393         cn_reply->ack = 0; /* not used here. */
2394         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2395                 (int)((char *)tl - (char *)reply->tag_list);
2396         cn_reply->flags = 0;
2397
2398         reply->packet_type = P_call_helper;
2399         reply->minor = mdev_to_minor(mdev);
2400         reply->ret_code = NO_ERROR;
2401
2402         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2403 }
2404
2405 void drbd_bcast_ee(struct drbd_conf *mdev,
2406                 const char *reason, const int dgs,
2407                 const char* seen_hash, const char* calc_hash,
2408                 const struct drbd_epoch_entry* e)
2409 {
2410         struct cn_msg *cn_reply;
2411         struct drbd_nl_cfg_reply *reply;
2412         unsigned short *tl;
2413         struct page *page;
2414         unsigned len;
2415
2416         if (!e)
2417                 return;
2418         if (!reason || !reason[0])
2419                 return;
2420
2421         /* apparently we have to memcpy twice, first to prepare the data for the
2422          * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2423          * netlink skb. */
2424         /* receiver thread context, which is not in the writeout path (of this node),
2425          * but may be in the writeout path of the _other_ node.
2426          * GFP_NOIO to avoid potential "distributed deadlock". */
2427         cn_reply = kzalloc(
2428                 sizeof(struct cn_msg)+
2429                 sizeof(struct drbd_nl_cfg_reply)+
2430                 sizeof(struct dump_ee_tag_len_struct)+
2431                 sizeof(short int),
2432                 GFP_NOIO);
2433
2434         if (!cn_reply) {
2435                 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2436                                 (unsigned long long)e->sector, e->size);
2437                 return;
2438         }
2439
2440         reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2441         tl = reply->tag_list;
2442
2443         tl = tl_add_str(tl, T_dump_ee_reason, reason);
2444         tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2445         tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2446         tl = tl_add_int(tl, T_ee_sector, &e->sector);
2447         tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2448
2449         /* dump the first 32k */
2450         len = min_t(unsigned, e->size, 32 << 10);
2451         put_unaligned(T_ee_data, tl++);
2452         put_unaligned(len, tl++);
2453
2454         page = e->pages;
2455         page_chain_for_each(page) {
2456                 void *d = kmap_atomic(page, KM_USER0);
2457                 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2458                 memcpy(tl, d, l);
2459                 kunmap_atomic(d, KM_USER0);
2460                 tl = (unsigned short*)((char*)tl + l);
2461                 len -= l;
2462                 if (len == 0)
2463                         break;
2464         }
2465         put_unaligned(TT_END, tl++); /* Close the tag list */
2466
2467         cn_reply->id.idx = CN_IDX_DRBD;
2468         cn_reply->id.val = CN_VAL_DRBD;
2469
2470         cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2471         cn_reply->ack = 0; // not used here.
2472         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2473                 (int)((char*)tl - (char*)reply->tag_list);
2474         cn_reply->flags = 0;
2475
2476         reply->packet_type = P_dump_ee;
2477         reply->minor = mdev_to_minor(mdev);
2478         reply->ret_code = NO_ERROR;
2479
2480         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2481         kfree(cn_reply);
2482 }
2483
2484 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2485 {
2486         char buffer[sizeof(struct cn_msg)+
2487                     sizeof(struct drbd_nl_cfg_reply)+
2488                     sizeof(struct sync_progress_tag_len_struct)+
2489                     sizeof(short int)];
2490         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2491         struct drbd_nl_cfg_reply *reply =
2492                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2493         unsigned short *tl = reply->tag_list;
2494         unsigned long rs_left;
2495         unsigned int res;
2496
2497         /* no local ref, no bitmap, no syncer progress, no broadcast. */
2498         if (!get_ldev(mdev))
2499                 return;
2500         drbd_get_syncer_progress(mdev, &rs_left, &res);
2501         put_ldev(mdev);
2502
2503         tl = tl_add_int(tl, T_sync_progress, &res);
2504         put_unaligned(TT_END, tl++); /* Close the tag list */
2505
2506         cn_reply->id.idx = CN_IDX_DRBD;
2507         cn_reply->id.val = CN_VAL_DRBD;
2508
2509         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2510         cn_reply->ack = 0; /* not used here. */
2511         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2512                 (int)((char *)tl - (char *)reply->tag_list);
2513         cn_reply->flags = 0;
2514
2515         reply->packet_type = P_sync_progress;
2516         reply->minor = mdev_to_minor(mdev);
2517         reply->ret_code = NO_ERROR;
2518
2519         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2520 }
2521
2522 int __init drbd_nl_init(void)
2523 {
2524         static struct cb_id cn_id_drbd;
2525         int err, try=10;
2526
2527         cn_id_drbd.val = CN_VAL_DRBD;
2528         do {
2529                 cn_id_drbd.idx = cn_idx;
2530                 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2531                 if (!err)
2532                         break;
2533                 cn_idx = (cn_idx + CN_IDX_STEP);
2534         } while (try--);
2535
2536         if (err) {
2537                 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2538                 return err;
2539         }
2540
2541         return 0;
2542 }
2543
2544 void drbd_nl_cleanup(void)
2545 {
2546         static struct cb_id cn_id_drbd;
2547
2548         cn_id_drbd.idx = cn_idx;
2549         cn_id_drbd.val = CN_VAL_DRBD;
2550
2551         cn_del_callback(&cn_id_drbd);
2552 }
2553
2554 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2555 {
2556         char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2557         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2558         struct drbd_nl_cfg_reply *reply =
2559                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2560         int rr;
2561
2562         memset(buffer, 0, sizeof(buffer));
2563         cn_reply->id = req->id;
2564
2565         cn_reply->seq = req->seq;
2566         cn_reply->ack = req->ack  + 1;
2567         cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2568         cn_reply->flags = 0;
2569
2570         reply->packet_type = P_return_code_only;
2571         reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2572         reply->ret_code = ret_code;
2573
2574         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2575         if (rr && rr != -ESRCH)
2576                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2577 }
2578