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                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
 652                 err = drbd_bitmap_io(mdev, &drbd_bm_write,
 653                                 "size changed", BM_LOCKED_MASK);
 654                 if (err) {
 655                         rv = dev_size_error;
 656                         goto out;
 657                 }
 658                 drbd_md_mark_dirty(mdev);
 659         }
 660
 661         if (size > la_size)
 662                 rv = grew;
 663         if (size < la_size)
 664                 rv = shrunk;
 665 out:
 666         lc_unlock(mdev->act_log);
 667         wake_up(&mdev->al_wait);
 668         drbd_resume_io(mdev);
 669
 670         return rv;
 671 }
 672
 673 sector_t
 674 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
 675 {
 676         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 677         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 678         sector_t m_size; /* my size */
 679         sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
 680         sector_t size = 0;
 681
 682         m_size = drbd_get_max_capacity(bdev);
 683
 684         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 685                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 686                 p_size = m_size;
 687         }
 688
 689         if (p_size && m_size) {
 690                 size = min_t(sector_t, p_size, m_size);
 691         } else {
 692                 if (la_size) {
 693                         size = la_size;
 694                         if (m_size && m_size < size)
 695                                 size = m_size;
 696                         if (p_size && p_size < size)
 697                                 size = p_size;
 698                 } else {
 699                         if (m_size)
 700                                 size = m_size;
 701                         if (p_size)
 702                                 size = p_size;
 703                 }
 704         }
 705
 706         if (size == 0)
 707                 dev_err(DEV, "Both nodes diskless!\n");
 708
 709         if (u_size) {
 710                 if (u_size > size)
 711                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 712                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 713                 else
 714                         size = u_size;
 715         }
 716
 717         return size;
 718 }
 719
 720 /**
 721  * drbd_check_al_size() - Ensures that the AL is of the right size
 722  * @mdev:       DRBD device.
 723  *
 724  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 725  * failed, and 0 on success. You should call drbd_md_sync() after you called
 726  * this function.
 727  */
 728 static int drbd_check_al_size(struct drbd_conf *mdev)
 729 {
 730         struct lru_cache *n, *t;
 731         struct lc_element *e;
 732         unsigned int in_use;
 733         int i;
 734
 735         ERR_IF(mdev->sync_conf.al_extents < 7)
 736                 mdev->sync_conf.al_extents = 127;
 737
 738         if (mdev->act_log &&
 739             mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
 740                 return 0;
 741
 742         in_use = 0;
 743         t = mdev->act_log;
 744         n = lc_create("act_log", drbd_al_ext_cache,
 745                 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
 746
 747         if (n == NULL) {
 748                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 749                 return -ENOMEM;
 750         }
 751         spin_lock_irq(&mdev->al_lock);
 752         if (t) {
 753                 for (i = 0; i < t->nr_elements; i++) {
 754                         e = lc_element_by_index(t, i);
 755                         if (e->refcnt)
 756                                 dev_err(DEV, "refcnt(%d)==%d\n",
 757                                     e->lc_number, e->refcnt);
 758                         in_use += e->refcnt;
 759                 }
 760         }
 761         if (!in_use)
 762                 mdev->act_log = n;
 763         spin_unlock_irq(&mdev->al_lock);
 764         if (in_use) {
 765                 dev_err(DEV, "Activity log still in use!\n");
 766                 lc_destroy(n);
 767                 return -EBUSY;
 768         } else {
 769                 if (t)
 770                         lc_destroy(t);
 771         }
 772         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 773         return 0;
 774 }
 775
 776 void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size) __must_hold(local)
 777 {
 778         struct request_queue * const q = mdev->rq_queue;
 779         struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 780         int max_segments = mdev->ldev->dc.max_bio_bvecs;
 781         int max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
 782
 783         blk_queue_logical_block_size(q, 512);
 784         blk_queue_max_hw_sectors(q, max_hw_sectors);
 785         /* This is the workaround for "bio would need to, but cannot, be split" */
 786         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
 787         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
 788         blk_queue_stack_limits(q, b);
 789
 790         dev_info(DEV, "max BIO size = %u\n", queue_max_hw_sectors(q) << 9);
 791
 792         if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
 793                 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
 794                      q->backing_dev_info.ra_pages,
 795                      b->backing_dev_info.ra_pages);
 796                 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
 797         }
 798 }
 799
 800 /* serialize deconfig (worker exiting, doing cleanup)
 801  * and reconfig (drbdsetup disk, drbdsetup net)
 802  *
 803  * Wait for a potentially exiting worker, then restart it,
 804  * or start a new one.  Flush any pending work, there may still be an
 805  * after_state_change queued.
 806  */
 807 static void drbd_reconfig_start(struct drbd_conf *mdev)
 808 {
 809         wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
 810         wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
 811         drbd_thread_start(&mdev->worker);
 812         drbd_flush_workqueue(mdev);
 813 }
 814
 815 /* if still unconfigured, stops worker again.
 816  * if configured now, clears CONFIG_PENDING.
 817  * wakes potential waiters */
 818 static void drbd_reconfig_done(struct drbd_conf *mdev)
 819 {
 820         spin_lock_irq(&mdev->req_lock);
 821         if (mdev->state.disk == D_DISKLESS &&
 822             mdev->state.conn == C_STANDALONE &&
 823             mdev->state.role == R_SECONDARY) {
 824                 set_bit(DEVICE_DYING, &mdev->flags);
 825                 drbd_thread_stop_nowait(&mdev->worker);
 826         } else
 827                 clear_bit(CONFIG_PENDING, &mdev->flags);
 828         spin_unlock_irq(&mdev->req_lock);
 829         wake_up(&mdev->state_wait);
 830 }
 831
 832 /* Make sure IO is suspended before calling this function(). */
 833 static void drbd_suspend_al(struct drbd_conf *mdev)
 834 {
 835         int s = 0;
 836
 837         if (lc_try_lock(mdev->act_log)) {
 838                 drbd_al_shrink(mdev);
 839                 lc_unlock(mdev->act_log);
 840         } else {
 841                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
 842                 return;
 843         }
 844
 845         spin_lock_irq(&mdev->req_lock);
 846         if (mdev->state.conn < C_CONNECTED)
 847                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
 848
 849         spin_unlock_irq(&mdev->req_lock);
 850
 851         if (s)
 852                 dev_info(DEV, "Suspended AL updates\n");
 853 }
 854
 855 /* does always return 0;
 856  * interesting return code is in reply->ret_code */
 857 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 858                              struct drbd_nl_cfg_reply *reply)
 859 {
 860         enum drbd_ret_code retcode;
 861         enum determine_dev_size dd;
 862         sector_t max_possible_sectors;
 863         sector_t min_md_device_sectors;
 864         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
 865         struct block_device *bdev;
 866         struct lru_cache *resync_lru = NULL;
 867         union drbd_state ns, os;
 868         unsigned int max_bio_size;
 869         enum drbd_state_rv rv;
 870         int cp_discovered = 0;
 871         int logical_block_size;
 872
 873         drbd_reconfig_start(mdev);
 874
 875         /* if you want to reconfigure, please tear down first */
 876         if (mdev->state.disk > D_DISKLESS) {
 877                 retcode = ERR_DISK_CONFIGURED;
 878                 goto fail;
 879         }
 880         /* It may just now have detached because of IO error.  Make sure
 881          * drbd_ldev_destroy is done already, we may end up here very fast,
 882          * e.g. if someone calls attach from the on-io-error handler,
 883          * to realize a "hot spare" feature (not that I'd recommend that) */
 884         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
 885
 886         /* allocation not in the IO path, cqueue thread context */
 887         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
 888         if (!nbc) {
 889                 retcode = ERR_NOMEM;
 890                 goto fail;
 891         }
 892
 893         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
 894         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
 895         nbc->dc.fencing       = DRBD_FENCING_DEF;
 896         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
 897
 898         if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
 899                 retcode = ERR_MANDATORY_TAG;
 900                 goto fail;
 901         }
 902
 903         if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
 904                 retcode = ERR_MD_IDX_INVALID;
 905                 goto fail;
 906         }
 907
 908         if (get_net_conf(mdev)) {
 909                 int prot = mdev->net_conf->wire_protocol;
 910                 put_net_conf(mdev);
 911                 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
 912                         retcode = ERR_STONITH_AND_PROT_A;
 913                         goto fail;
 914                 }
 915         }
 916
 917         bdev = blkdev_get_by_path(nbc->dc.backing_dev,
 918                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
 919         if (IS_ERR(bdev)) {
 920                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
 921                         PTR_ERR(bdev));
 922                 retcode = ERR_OPEN_DISK;
 923                 goto fail;
 924         }
 925         nbc->backing_bdev = bdev;
 926
 927         /*
 928          * meta_dev_idx >= 0: external fixed size, possibly multiple
 929          * drbd sharing one meta device.  TODO in that case, paranoia
 930          * check that [md_bdev, meta_dev_idx] is not yet used by some
 931          * other drbd minor!  (if you use drbd.conf + drbdadm, that
 932          * should check it for you already; but if you don't, or
 933          * someone fooled it, we need to double check here)
 934          */
 935         bdev = blkdev_get_by_path(nbc->dc.meta_dev,
 936                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
 937                                   (nbc->dc.meta_dev_idx < 0) ?
 938                                   (void *)mdev : (void *)drbd_m_holder);
 939         if (IS_ERR(bdev)) {
 940                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
 941                         PTR_ERR(bdev));
 942                 retcode = ERR_OPEN_MD_DISK;
 943                 goto fail;
 944         }
 945         nbc->md_bdev = bdev;
 946
 947         if ((nbc->backing_bdev == nbc->md_bdev) !=
 948             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
 949              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
 950                 retcode = ERR_MD_IDX_INVALID;
 951                 goto fail;
 952         }
 953
 954         resync_lru = lc_create("resync", drbd_bm_ext_cache,
 955                         61, sizeof(struct bm_extent),
 956                         offsetof(struct bm_extent, lce));
 957         if (!resync_lru) {
 958                 retcode = ERR_NOMEM;
 959                 goto fail;
 960         }
 961
 962         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
 963         drbd_md_set_sector_offsets(mdev, nbc);
 964
 965         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
 966                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
 967                         (unsigned long long) drbd_get_max_capacity(nbc),
 968                         (unsigned long long) nbc->dc.disk_size);
 969                 retcode = ERR_DISK_TO_SMALL;
 970                 goto fail;
 971         }
 972
 973         if (nbc->dc.meta_dev_idx < 0) {
 974                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
 975                 /* at least one MB, otherwise it does not make sense */
 976                 min_md_device_sectors = (2<<10);
 977         } else {
 978                 max_possible_sectors = DRBD_MAX_SECTORS;
 979                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
 980         }
 981
 982         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
 983                 retcode = ERR_MD_DISK_TO_SMALL;
 984                 dev_warn(DEV, "refusing attach: md-device too small, "
 985                      "at least %llu sectors needed for this meta-disk type\n",
 986                      (unsigned long long) min_md_device_sectors);
 987                 goto fail;
 988         }
 989
 990         /* Make sure the new disk is big enough
 991          * (we may currently be R_PRIMARY with no local disk...) */
 992         if (drbd_get_max_capacity(nbc) <
 993             drbd_get_capacity(mdev->this_bdev)) {
 994                 retcode = ERR_DISK_TO_SMALL;
 995                 goto fail;
 996         }
 997
 998         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
 999
1000         if (nbc->known_size > max_possible_sectors) {
1001                 dev_warn(DEV, "==> truncating very big lower level device "
1002                         "to currently maximum possible %llu sectors <==\n",
1003                         (unsigned long long) max_possible_sectors);
1004                 if (nbc->dc.meta_dev_idx >= 0)
1005                         dev_warn(DEV, "==>> using internal or flexible "
1006                                       "meta data may help <<==\n");
1007         }
1008
1009         drbd_suspend_io(mdev);
1010         /* also wait for the last barrier ack. */
1011         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state));
1012         /* and for any other previously queued work */
1013         drbd_flush_workqueue(mdev);
1014
1015         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1016         retcode = rv;  /* FIXME: Type mismatch. */
1017         drbd_resume_io(mdev);
1018         if (rv < SS_SUCCESS)
1019                 goto fail;
1020
1021         if (!get_ldev_if_state(mdev, D_ATTACHING))
1022                 goto force_diskless;
1023
1024         drbd_md_set_sector_offsets(mdev, nbc);
1025
1026         /* allocate a second IO page if logical_block_size != 512 */
1027         logical_block_size = bdev_logical_block_size(nbc->md_bdev);
1028         if (logical_block_size == 0)
1029                 logical_block_size = MD_SECTOR_SIZE;
1030
1031         if (logical_block_size != MD_SECTOR_SIZE) {
1032                 if (!mdev->md_io_tmpp) {
1033                         struct page *page = alloc_page(GFP_NOIO);
1034                         if (!page)
1035                                 goto force_diskless_dec;
1036
1037                         dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
1038                              logical_block_size, MD_SECTOR_SIZE);
1039                         dev_warn(DEV, "Workaround engaged (has performance impact).\n");
1040
1041                         mdev->md_io_tmpp = page;
1042                 }
1043         }
1044
1045         if (!mdev->bitmap) {
1046                 if (drbd_bm_init(mdev)) {
1047                         retcode = ERR_NOMEM;
1048                         goto force_diskless_dec;
1049                 }
1050         }
1051
1052         retcode = drbd_md_read(mdev, nbc);
1053         if (retcode != NO_ERROR)
1054                 goto force_diskless_dec;
1055
1056         if (mdev->state.conn < C_CONNECTED &&
1057             mdev->state.role == R_PRIMARY &&
1058             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1059                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1060                     (unsigned long long)mdev->ed_uuid);
1061                 retcode = ERR_DATA_NOT_CURRENT;
1062                 goto force_diskless_dec;
1063         }
1064
1065         /* Since we are diskless, fix the activity log first... */
1066         if (drbd_check_al_size(mdev)) {
1067                 retcode = ERR_NOMEM;
1068                 goto force_diskless_dec;
1069         }
1070
1071         /* Prevent shrinking of consistent devices ! */
1072         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1073             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1074                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1075                 retcode = ERR_DISK_TO_SMALL;
1076                 goto force_diskless_dec;
1077         }
1078
1079         if (!drbd_al_read_log(mdev, nbc)) {
1080                 retcode = ERR_IO_MD_DISK;
1081                 goto force_diskless_dec;
1082         }
1083
1084         /* Reset the "barriers don't work" bits here, then force meta data to
1085          * be written, to ensure we determine if barriers are supported. */
1086         if (nbc->dc.no_md_flush)
1087                 set_bit(MD_NO_FUA, &mdev->flags);
1088         else
1089                 clear_bit(MD_NO_FUA, &mdev->flags);
1090
1091         /* Point of no return reached.
1092          * Devices and memory are no longer released by error cleanup below.
1093          * now mdev takes over responsibility, and the state engine should
1094          * clean it up somewhere.  */
1095         D_ASSERT(mdev->ldev == NULL);
1096         mdev->ldev = nbc;
1097         mdev->resync = resync_lru;
1098         nbc = NULL;
1099         resync_lru = NULL;
1100
1101         mdev->write_ordering = WO_bdev_flush;
1102         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1103
1104         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1105                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1106         else
1107                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1108
1109         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1110             !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) {
1111                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1112                 cp_discovered = 1;
1113         }
1114
1115         mdev->send_cnt = 0;
1116         mdev->recv_cnt = 0;
1117         mdev->read_cnt = 0;
1118         mdev->writ_cnt = 0;
1119
1120         max_bio_size = DRBD_MAX_BIO_SIZE;
1121         if (mdev->state.conn == C_CONNECTED) {
1122                 /* We are Primary, Connected, and now attach a new local
1123                  * backing store. We must not increase the user visible maximum
1124                  * bio size on this device to something the peer may not be
1125                  * able to handle. */
1126                 if (mdev->agreed_pro_version < 94)
1127                         max_bio_size = queue_max_hw_sectors(mdev->rq_queue) << 9;
1128                 else if (mdev->agreed_pro_version == 94)
1129                         max_bio_size = DRBD_MAX_SIZE_H80_PACKET;
1130                 /* else: drbd 8.3.9 and later, stay with default */
1131         }
1132
1133         drbd_setup_queue_param(mdev, max_bio_size);
1134
1135         /* If I am currently not R_PRIMARY,
1136          * but meta data primary indicator is set,
1137          * I just now recover from a hard crash,
1138          * and have been R_PRIMARY before that crash.
1139          *
1140          * Now, if I had no connection before that crash
1141          * (have been degraded R_PRIMARY), chances are that
1142          * I won't find my peer now either.
1143          *
1144          * In that case, and _only_ in that case,
1145          * we use the degr-wfc-timeout instead of the default,
1146          * so we can automatically recover from a crash of a
1147          * degraded but active "cluster" after a certain timeout.
1148          */
1149         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1150         if (mdev->state.role != R_PRIMARY &&
1151              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1152             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1153                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1154
1155         dd = drbd_determin_dev_size(mdev, 0);
1156         if (dd == dev_size_error) {
1157                 retcode = ERR_NOMEM_BITMAP;
1158                 goto force_diskless_dec;
1159         } else if (dd == grew)
1160                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1161
1162         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1163                 dev_info(DEV, "Assuming that all blocks are out of sync "
1164                      "(aka FullSync)\n");
1165                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1166                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1167                         retcode = ERR_IO_MD_DISK;
1168                         goto force_diskless_dec;
1169                 }
1170         } else {
1171                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1172                         "read from attaching", BM_LOCKED_MASK) < 0) {
1173                         retcode = ERR_IO_MD_DISK;
1174                         goto force_diskless_dec;
1175                 }
1176         }
1177
1178         if (cp_discovered) {
1179                 drbd_al_apply_to_bm(mdev);
1180                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1181                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1182                         retcode = ERR_IO_MD_DISK;
1183                         goto force_diskless_dec;
1184                 }
1185         }
1186
1187         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1188                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1189
1190         spin_lock_irq(&mdev->req_lock);
1191         os = mdev->state;
1192         ns.i = os.i;
1193         /* If MDF_CONSISTENT is not set go into inconsistent state,
1194            otherwise investigate MDF_WasUpToDate...
1195            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1196            otherwise into D_CONSISTENT state.
1197         */
1198         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1199                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1200                         ns.disk = D_CONSISTENT;
1201                 else
1202                         ns.disk = D_OUTDATED;
1203         } else {
1204                 ns.disk = D_INCONSISTENT;
1205         }
1206
1207         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1208                 ns.pdsk = D_OUTDATED;
1209
1210         if ( ns.disk == D_CONSISTENT &&
1211             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1212                 ns.disk = D_UP_TO_DATE;
1213
1214         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1215            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1216            this point, because drbd_request_state() modifies these
1217            flags. */
1218
1219         /* In case we are C_CONNECTED postpone any decision on the new disk
1220            state after the negotiation phase. */
1221         if (mdev->state.conn == C_CONNECTED) {
1222                 mdev->new_state_tmp.i = ns.i;
1223                 ns.i = os.i;
1224                 ns.disk = D_NEGOTIATING;
1225
1226                 /* We expect to receive up-to-date UUIDs soon.
1227                    To avoid a race in receive_state, free p_uuid while
1228                    holding req_lock. I.e. atomic with the state change */
1229                 kfree(mdev->p_uuid);
1230                 mdev->p_uuid = NULL;
1231         }
1232
1233         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1234         ns = mdev->state;
1235         spin_unlock_irq(&mdev->req_lock);
1236
1237         if (rv < SS_SUCCESS)
1238                 goto force_diskless_dec;
1239
1240         if (mdev->state.role == R_PRIMARY)
1241                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1242         else
1243                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1244
1245         drbd_md_mark_dirty(mdev);
1246         drbd_md_sync(mdev);
1247
1248         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1249         put_ldev(mdev);
1250         reply->ret_code = retcode;
1251         drbd_reconfig_done(mdev);
1252         return 0;
1253
1254  force_diskless_dec:
1255         put_ldev(mdev);
1256  force_diskless:
1257         drbd_force_state(mdev, NS(disk, D_FAILED));
1258         drbd_md_sync(mdev);
1259  fail:
1260         if (nbc) {
1261                 if (nbc->backing_bdev)
1262                         blkdev_put(nbc->backing_bdev,
1263                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1264                 if (nbc->md_bdev)
1265                         blkdev_put(nbc->md_bdev,
1266                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1267                 kfree(nbc);
1268         }
1269         lc_destroy(resync_lru);
1270
1271         reply->ret_code = retcode;
1272         drbd_reconfig_done(mdev);
1273         return 0;
1274 }
1275
1276 /* Detaching the disk is a process in multiple stages.  First we need to lock
1277  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1278  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1279  * internal references as well.
1280  * Only then we have finally detached. */
1281 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1282                           struct drbd_nl_cfg_reply *reply)
1283 {
1284         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1285         reply->ret_code = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1286         if (mdev->state.disk == D_DISKLESS)
1287                 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1288         drbd_resume_io(mdev);
1289         return 0;
1290 }
1291
1292 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1293                             struct drbd_nl_cfg_reply *reply)
1294 {
1295         int i, ns;
1296         enum drbd_ret_code retcode;
1297         struct net_conf *new_conf = NULL;
1298         struct crypto_hash *tfm = NULL;
1299         struct crypto_hash *integrity_w_tfm = NULL;
1300         struct crypto_hash *integrity_r_tfm = NULL;
1301         struct hlist_head *new_tl_hash = NULL;
1302         struct hlist_head *new_ee_hash = NULL;
1303         struct drbd_conf *odev;
1304         char hmac_name[CRYPTO_MAX_ALG_NAME];
1305         void *int_dig_out = NULL;
1306         void *int_dig_in = NULL;
1307         void *int_dig_vv = NULL;
1308         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1309
1310         drbd_reconfig_start(mdev);
1311
1312         if (mdev->state.conn > C_STANDALONE) {
1313                 retcode = ERR_NET_CONFIGURED;
1314                 goto fail;
1315         }
1316
1317         /* allocation not in the IO path, cqueue thread context */
1318         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1319         if (!new_conf) {
1320                 retcode = ERR_NOMEM;
1321                 goto fail;
1322         }
1323
1324         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1325         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1326         new_conf->ping_int         = DRBD_PING_INT_DEF;
1327         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1328         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1329         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1330         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1331         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1332         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1333         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1334         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1335         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1336         new_conf->want_lose        = 0;
1337         new_conf->two_primaries    = 0;
1338         new_conf->wire_protocol    = DRBD_PROT_C;
1339         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1340         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1341         new_conf->on_congestion    = DRBD_ON_CONGESTION_DEF;
1342         new_conf->cong_extents     = DRBD_CONG_EXTENTS_DEF;
1343
1344         if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1345                 retcode = ERR_MANDATORY_TAG;
1346                 goto fail;
1347         }
1348
1349         if (new_conf->two_primaries
1350             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1351                 retcode = ERR_NOT_PROTO_C;
1352                 goto fail;
1353         }
1354
1355         if (get_ldev(mdev)) {
1356                 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1357                 put_ldev(mdev);
1358                 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1359                         retcode = ERR_STONITH_AND_PROT_A;
1360                         goto fail;
1361                 }
1362         }
1363
1364         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1365                 retcode = ERR_CONG_NOT_PROTO_A;
1366                 goto fail;
1367         }
1368
1369         if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1370                 retcode = ERR_DISCARD;
1371                 goto fail;
1372         }
1373
1374         retcode = NO_ERROR;
1375
1376         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1377         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1378         for (i = 0; i < minor_count; i++) {
1379                 odev = minor_to_mdev(i);
1380                 if (!odev || odev == mdev)
1381                         continue;
1382                 if (get_net_conf(odev)) {
1383                         taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1384                         if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1385                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1386                                 retcode = ERR_LOCAL_ADDR;
1387
1388                         taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1389                         if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1390                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1391                                 retcode = ERR_PEER_ADDR;
1392
1393                         put_net_conf(odev);
1394                         if (retcode != NO_ERROR)
1395                                 goto fail;
1396                 }
1397         }
1398
1399         if (new_conf->cram_hmac_alg[0] != 0) {
1400                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1401                         new_conf->cram_hmac_alg);
1402                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1403                 if (IS_ERR(tfm)) {
1404                         tfm = NULL;
1405                         retcode = ERR_AUTH_ALG;
1406                         goto fail;
1407                 }
1408
1409                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1410                         retcode = ERR_AUTH_ALG_ND;
1411                         goto fail;
1412                 }
1413         }
1414
1415         if (new_conf->integrity_alg[0]) {
1416                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1417                 if (IS_ERR(integrity_w_tfm)) {
1418                         integrity_w_tfm = NULL;
1419                         retcode=ERR_INTEGRITY_ALG;
1420                         goto fail;
1421                 }
1422
1423                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1424                         retcode=ERR_INTEGRITY_ALG_ND;
1425                         goto fail;
1426                 }
1427
1428                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1429                 if (IS_ERR(integrity_r_tfm)) {
1430                         integrity_r_tfm = NULL;
1431                         retcode=ERR_INTEGRITY_ALG;
1432                         goto fail;
1433                 }
1434         }
1435
1436         ns = new_conf->max_epoch_size/8;
1437         if (mdev->tl_hash_s != ns) {
1438                 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1439                 if (!new_tl_hash) {
1440                         retcode = ERR_NOMEM;
1441                         goto fail;
1442                 }
1443         }
1444
1445         ns = new_conf->max_buffers/8;
1446         if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1447                 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1448                 if (!new_ee_hash) {
1449                         retcode = ERR_NOMEM;
1450                         goto fail;
1451                 }
1452         }
1453
1454         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1455
1456         if (integrity_w_tfm) {
1457                 i = crypto_hash_digestsize(integrity_w_tfm);
1458                 int_dig_out = kmalloc(i, GFP_KERNEL);
1459                 if (!int_dig_out) {
1460                         retcode = ERR_NOMEM;
1461                         goto fail;
1462                 }
1463                 int_dig_in = kmalloc(i, GFP_KERNEL);
1464                 if (!int_dig_in) {
1465                         retcode = ERR_NOMEM;
1466                         goto fail;
1467                 }
1468                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1469                 if (!int_dig_vv) {
1470                         retcode = ERR_NOMEM;
1471                         goto fail;
1472                 }
1473         }
1474
1475         if (!mdev->bitmap) {
1476                 if(drbd_bm_init(mdev)) {
1477                         retcode = ERR_NOMEM;
1478                         goto fail;
1479                 }
1480         }
1481
1482         drbd_flush_workqueue(mdev);
1483         spin_lock_irq(&mdev->req_lock);
1484         if (mdev->net_conf != NULL) {
1485                 retcode = ERR_NET_CONFIGURED;
1486                 spin_unlock_irq(&mdev->req_lock);
1487                 goto fail;
1488         }
1489         mdev->net_conf = new_conf;
1490
1491         mdev->send_cnt = 0;
1492         mdev->recv_cnt = 0;
1493
1494         if (new_tl_hash) {
1495                 kfree(mdev->tl_hash);
1496                 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1497                 mdev->tl_hash = new_tl_hash;
1498         }
1499
1500         if (new_ee_hash) {
1501                 kfree(mdev->ee_hash);
1502                 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1503                 mdev->ee_hash = new_ee_hash;
1504         }
1505
1506         crypto_free_hash(mdev->cram_hmac_tfm);
1507         mdev->cram_hmac_tfm = tfm;
1508
1509         crypto_free_hash(mdev->integrity_w_tfm);
1510         mdev->integrity_w_tfm = integrity_w_tfm;
1511
1512         crypto_free_hash(mdev->integrity_r_tfm);
1513         mdev->integrity_r_tfm = integrity_r_tfm;
1514
1515         kfree(mdev->int_dig_out);
1516         kfree(mdev->int_dig_in);
1517         kfree(mdev->int_dig_vv);
1518         mdev->int_dig_out=int_dig_out;
1519         mdev->int_dig_in=int_dig_in;
1520         mdev->int_dig_vv=int_dig_vv;
1521         retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL);
1522         spin_unlock_irq(&mdev->req_lock);
1523
1524         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1525         reply->ret_code = retcode;
1526         drbd_reconfig_done(mdev);
1527         return 0;
1528
1529 fail:
1530         kfree(int_dig_out);
1531         kfree(int_dig_in);
1532         kfree(int_dig_vv);
1533         crypto_free_hash(tfm);
1534         crypto_free_hash(integrity_w_tfm);
1535         crypto_free_hash(integrity_r_tfm);
1536         kfree(new_tl_hash);
1537         kfree(new_ee_hash);
1538         kfree(new_conf);
1539
1540         reply->ret_code = retcode;
1541         drbd_reconfig_done(mdev);
1542         return 0;
1543 }
1544
1545 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1546                               struct drbd_nl_cfg_reply *reply)
1547 {
1548         int retcode;
1549         struct disconnect dc;
1550
1551         memset(&dc, 0, sizeof(struct disconnect));
1552         if (!disconnect_from_tags(mdev, nlp->tag_list, &dc)) {
1553                 retcode = ERR_MANDATORY_TAG;
1554                 goto fail;
1555         }
1556
1557         if (dc.force) {
1558                 spin_lock_irq(&mdev->req_lock);
1559                 if (mdev->state.conn >= C_WF_CONNECTION)
1560                         _drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), CS_HARD, NULL);
1561                 spin_unlock_irq(&mdev->req_lock);
1562                 goto done;
1563         }
1564
1565         retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1566
1567         if (retcode == SS_NOTHING_TO_DO)
1568                 goto done;
1569         else if (retcode == SS_ALREADY_STANDALONE)
1570                 goto done;
1571         else if (retcode == SS_PRIMARY_NOP) {
1572                 /* Our statche checking code wants to see the peer outdated. */
1573                 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1574                                                       pdsk, D_OUTDATED));
1575         } else if (retcode == SS_CW_FAILED_BY_PEER) {
1576                 /* The peer probably wants to see us outdated. */
1577                 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1578                                                         disk, D_OUTDATED),
1579                                               CS_ORDERED);
1580                 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1581                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1582                         retcode = SS_SUCCESS;
1583                 }
1584         }
1585
1586         if (retcode < SS_SUCCESS)
1587                 goto fail;
1588
1589         if (wait_event_interruptible(mdev->state_wait,
1590                                      mdev->state.conn != C_DISCONNECTING)) {
1591                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1592                    someone else might connect us in the mean time! */
1593                 retcode = ERR_INTR;
1594                 goto fail;
1595         }
1596
1597  done:
1598         retcode = NO_ERROR;
1599  fail:
1600         drbd_md_sync(mdev);
1601         reply->ret_code = retcode;
1602         return 0;
1603 }
1604
1605 void resync_after_online_grow(struct drbd_conf *mdev)
1606 {
1607         int iass; /* I am sync source */
1608
1609         dev_info(DEV, "Resync of new storage after online grow\n");
1610         if (mdev->state.role != mdev->state.peer)
1611                 iass = (mdev->state.role == R_PRIMARY);
1612         else
1613                 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1614
1615         if (iass)
1616                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1617         else
1618                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1619 }
1620
1621 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1622                           struct drbd_nl_cfg_reply *reply)
1623 {
1624         struct resize rs;
1625         int retcode = NO_ERROR;
1626         enum determine_dev_size dd;
1627         enum dds_flags ddsf;
1628
1629         memset(&rs, 0, sizeof(struct resize));
1630         if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1631                 retcode = ERR_MANDATORY_TAG;
1632                 goto fail;
1633         }
1634
1635         if (mdev->state.conn > C_CONNECTED) {
1636                 retcode = ERR_RESIZE_RESYNC;
1637                 goto fail;
1638         }
1639
1640         if (mdev->state.role == R_SECONDARY &&
1641             mdev->state.peer == R_SECONDARY) {
1642                 retcode = ERR_NO_PRIMARY;
1643                 goto fail;
1644         }
1645
1646         if (!get_ldev(mdev)) {
1647                 retcode = ERR_NO_DISK;
1648                 goto fail;
1649         }
1650
1651         if (rs.no_resync && mdev->agreed_pro_version < 93) {
1652                 retcode = ERR_NEED_APV_93;
1653                 goto fail;
1654         }
1655
1656         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1657                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1658
1659         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1660         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1661         dd = drbd_determin_dev_size(mdev, ddsf);
1662         drbd_md_sync(mdev);
1663         put_ldev(mdev);
1664         if (dd == dev_size_error) {
1665                 retcode = ERR_NOMEM_BITMAP;
1666                 goto fail;
1667         }
1668
1669         if (mdev->state.conn == C_CONNECTED) {
1670                 if (dd == grew)
1671                         set_bit(RESIZE_PENDING, &mdev->flags);
1672
1673                 drbd_send_uuids(mdev);
1674                 drbd_send_sizes(mdev, 1, ddsf);
1675         }
1676
1677  fail:
1678         reply->ret_code = retcode;
1679         return 0;
1680 }
1681
1682 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1683                                struct drbd_nl_cfg_reply *reply)
1684 {
1685         int retcode = NO_ERROR;
1686         int err;
1687         int ovr; /* online verify running */
1688         int rsr; /* re-sync running */
1689         struct crypto_hash *verify_tfm = NULL;
1690         struct crypto_hash *csums_tfm = NULL;
1691         struct syncer_conf sc;
1692         cpumask_var_t new_cpu_mask;
1693         int *rs_plan_s = NULL;
1694         int fifo_size;
1695
1696         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1697                 retcode = ERR_NOMEM;
1698                 goto fail;
1699         }
1700
1701         if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1702                 memset(&sc, 0, sizeof(struct syncer_conf));
1703                 sc.rate       = DRBD_RATE_DEF;
1704                 sc.after      = DRBD_AFTER_DEF;
1705                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1706                 sc.on_no_data  = DRBD_ON_NO_DATA_DEF;
1707                 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1708                 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1709                 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1710                 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1711                 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1712         } else
1713                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1714
1715         if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1716                 retcode = ERR_MANDATORY_TAG;
1717                 goto fail;
1718         }
1719
1720         /* re-sync running */
1721         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1722                 mdev->state.conn == C_SYNC_TARGET ||
1723                 mdev->state.conn == C_PAUSED_SYNC_S ||
1724                 mdev->state.conn == C_PAUSED_SYNC_T );
1725
1726         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1727                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1728                 goto fail;
1729         }
1730
1731         if (!rsr && sc.csums_alg[0]) {
1732                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1733                 if (IS_ERR(csums_tfm)) {
1734                         csums_tfm = NULL;
1735                         retcode = ERR_CSUMS_ALG;
1736                         goto fail;
1737                 }
1738
1739                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1740                         retcode = ERR_CSUMS_ALG_ND;
1741                         goto fail;
1742                 }
1743         }
1744
1745         /* online verify running */
1746         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1747
1748         if (ovr) {
1749                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1750                         retcode = ERR_VERIFY_RUNNING;
1751                         goto fail;
1752                 }
1753         }
1754
1755         if (!ovr && sc.verify_alg[0]) {
1756                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1757                 if (IS_ERR(verify_tfm)) {
1758                         verify_tfm = NULL;
1759                         retcode = ERR_VERIFY_ALG;
1760                         goto fail;
1761                 }
1762
1763                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1764                         retcode = ERR_VERIFY_ALG_ND;
1765                         goto fail;
1766                 }
1767         }
1768
1769         /* silently ignore cpu mask on UP kernel */
1770         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1771                 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1772                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1773                 if (err) {
1774                         dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1775                         retcode = ERR_CPU_MASK_PARSE;
1776                         goto fail;
1777                 }
1778         }
1779
1780         ERR_IF (sc.rate < 1) sc.rate = 1;
1781         ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1782 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1783         if (sc.al_extents > AL_MAX) {
1784                 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1785                 sc.al_extents = AL_MAX;
1786         }
1787 #undef AL_MAX
1788
1789         /* to avoid spurious errors when configuring minors before configuring
1790          * the minors they depend on: if necessary, first create the minor we
1791          * depend on */
1792         if (sc.after >= 0)
1793                 ensure_mdev(sc.after, 1);
1794
1795         /* most sanity checks done, try to assign the new sync-after
1796          * dependency.  need to hold the global lock in there,
1797          * to avoid a race in the dependency loop check. */
1798         retcode = drbd_alter_sa(mdev, sc.after);
1799         if (retcode != NO_ERROR)
1800                 goto fail;
1801
1802         fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1803         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1804                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1805                 if (!rs_plan_s) {
1806                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1807                         retcode = ERR_NOMEM;
1808                         goto fail;
1809                 }
1810         }
1811
1812         /* ok, assign the rest of it as well.
1813          * lock against receive_SyncParam() */
1814         spin_lock(&mdev->peer_seq_lock);
1815         mdev->sync_conf = sc;
1816
1817         if (!rsr) {
1818                 crypto_free_hash(mdev->csums_tfm);
1819                 mdev->csums_tfm = csums_tfm;
1820                 csums_tfm = NULL;
1821         }
1822
1823         if (!ovr) {
1824                 crypto_free_hash(mdev->verify_tfm);
1825                 mdev->verify_tfm = verify_tfm;
1826                 verify_tfm = NULL;
1827         }
1828
1829         if (fifo_size != mdev->rs_plan_s.size) {
1830                 kfree(mdev->rs_plan_s.values);
1831                 mdev->rs_plan_s.values = rs_plan_s;
1832                 mdev->rs_plan_s.size   = fifo_size;
1833                 mdev->rs_planed = 0;
1834                 rs_plan_s = NULL;
1835         }
1836
1837         spin_unlock(&mdev->peer_seq_lock);
1838
1839         if (get_ldev(mdev)) {
1840                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1841                 drbd_al_shrink(mdev);
1842                 err = drbd_check_al_size(mdev);
1843                 lc_unlock(mdev->act_log);
1844                 wake_up(&mdev->al_wait);
1845
1846                 put_ldev(mdev);
1847                 drbd_md_sync(mdev);
1848
1849                 if (err) {
1850                         retcode = ERR_NOMEM;
1851                         goto fail;
1852                 }
1853         }
1854
1855         if (mdev->state.conn >= C_CONNECTED)
1856                 drbd_send_sync_param(mdev, &sc);
1857
1858         if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1859                 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1860                 drbd_calc_cpu_mask(mdev);
1861                 mdev->receiver.reset_cpu_mask = 1;
1862                 mdev->asender.reset_cpu_mask = 1;
1863                 mdev->worker.reset_cpu_mask = 1;
1864         }
1865
1866         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1867 fail:
1868         kfree(rs_plan_s);
1869         free_cpumask_var(new_cpu_mask);
1870         crypto_free_hash(csums_tfm);
1871         crypto_free_hash(verify_tfm);
1872         reply->ret_code = retcode;
1873         return 0;
1874 }
1875
1876 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1877                               struct drbd_nl_cfg_reply *reply)
1878 {
1879         int retcode;
1880
1881         /* If there is still bitmap IO pending, probably because of a previous
1882          * resync just being finished, wait for it before requesting a new resync. */
1883         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1884
1885         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1886
1887         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1888                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1889
1890         while (retcode == SS_NEED_CONNECTION) {
1891                 spin_lock_irq(&mdev->req_lock);
1892                 if (mdev->state.conn < C_CONNECTED)
1893                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1894                 spin_unlock_irq(&mdev->req_lock);
1895
1896                 if (retcode != SS_NEED_CONNECTION)
1897                         break;
1898
1899                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1900         }
1901
1902         reply->ret_code = retcode;
1903         return 0;
1904 }
1905
1906 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
1907 {
1908         int rv;
1909
1910         rv = drbd_bmio_set_n_write(mdev);
1911         drbd_suspend_al(mdev);
1912         return rv;
1913 }
1914
1915 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1916                                    struct drbd_nl_cfg_reply *reply)
1917 {
1918         int retcode;
1919
1920         /* If there is still bitmap IO pending, probably because of a previous
1921          * resync just being finished, wait for it before requesting a new resync. */
1922         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1923
1924         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
1925
1926         if (retcode < SS_SUCCESS) {
1927                 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
1928                         /* The peer will get a resync upon connect anyways. Just make that
1929                            into a full resync. */
1930                         retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
1931                         if (retcode >= SS_SUCCESS) {
1932                                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
1933                                         "set_n_write from invalidate_peer",
1934                                         BM_LOCKED_SET_ALLOWED))
1935                                         retcode = ERR_IO_MD_DISK;
1936                         }
1937                 } else
1938                         retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
1939         }
1940
1941         reply->ret_code = retcode;
1942         return 0;
1943 }
1944
1945 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1946                               struct drbd_nl_cfg_reply *reply)
1947 {
1948         int retcode = NO_ERROR;
1949
1950         if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
1951                 retcode = ERR_PAUSE_IS_SET;
1952
1953         reply->ret_code = retcode;
1954         return 0;
1955 }
1956
1957 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1958                                struct drbd_nl_cfg_reply *reply)
1959 {
1960         int retcode = NO_ERROR;
1961         union drbd_state s;
1962
1963         if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
1964                 s = mdev->state;
1965                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
1966                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
1967                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
1968                 } else {
1969                         retcode = ERR_PAUSE_IS_CLEAR;
1970                 }
1971         }
1972
1973         reply->ret_code = retcode;
1974         return 0;
1975 }
1976
1977 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1978                               struct drbd_nl_cfg_reply *reply)
1979 {
1980         reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
1981
1982         return 0;
1983 }
1984
1985 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1986                              struct drbd_nl_cfg_reply *reply)
1987 {
1988         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
1989                 drbd_uuid_new_current(mdev);
1990                 clear_bit(NEW_CUR_UUID, &mdev->flags);
1991         }
1992         drbd_suspend_io(mdev);
1993         reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
1994         if (reply->ret_code == SS_SUCCESS) {
1995                 if (mdev->state.conn < C_CONNECTED)
1996                         tl_clear(mdev);
1997                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
1998                         tl_restart(mdev, fail_frozen_disk_io);
1999         }
2000         drbd_resume_io(mdev);
2001
2002         return 0;
2003 }
2004
2005 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2006                            struct drbd_nl_cfg_reply *reply)
2007 {
2008         reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
2009         return 0;
2010 }
2011
2012 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2013                            struct drbd_nl_cfg_reply *reply)
2014 {
2015         unsigned short *tl;
2016
2017         tl = reply->tag_list;
2018
2019         if (get_ldev(mdev)) {
2020                 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
2021                 put_ldev(mdev);
2022         }
2023
2024         if (get_net_conf(mdev)) {
2025                 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
2026                 put_net_conf(mdev);
2027         }
2028         tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
2029
2030         put_unaligned(TT_END, tl++); /* Close the tag list */
2031
2032         return (int)((char *)tl - (char *)reply->tag_list);
2033 }
2034
2035 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2036                              struct drbd_nl_cfg_reply *reply)
2037 {
2038         unsigned short *tl = reply->tag_list;
2039         union drbd_state s = mdev->state;
2040         unsigned long rs_left;
2041         unsigned int res;
2042
2043         tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
2044
2045         /* no local ref, no bitmap, no syncer progress. */
2046         if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
2047                 if (get_ldev(mdev)) {
2048                         drbd_get_syncer_progress(mdev, &rs_left, &res);
2049                         tl = tl_add_int(tl, T_sync_progress, &res);
2050                         put_ldev(mdev);
2051                 }
2052         }
2053         put_unaligned(TT_END, tl++); /* Close the tag list */
2054
2055         return (int)((char *)tl - (char *)reply->tag_list);
2056 }
2057
2058 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2059                              struct drbd_nl_cfg_reply *reply)
2060 {
2061         unsigned short *tl;
2062
2063         tl = reply->tag_list;
2064
2065         if (get_ldev(mdev)) {
2066                 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
2067                 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
2068                 put_ldev(mdev);
2069         }
2070         put_unaligned(TT_END, tl++); /* Close the tag list */
2071
2072         return (int)((char *)tl - (char *)reply->tag_list);
2073 }
2074
2075 /**
2076  * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
2077  * @mdev:       DRBD device.
2078  * @nlp:        Netlink/connector packet from drbdsetup
2079  * @reply:      Reply packet for drbdsetup
2080  */
2081 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2082                                     struct drbd_nl_cfg_reply *reply)
2083 {
2084         unsigned short *tl;
2085         char rv;
2086
2087         tl = reply->tag_list;
2088
2089         rv = mdev->state.pdsk == D_OUTDATED        ? UT_PEER_OUTDATED :
2090           test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
2091
2092         tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
2093         put_unaligned(TT_END, tl++); /* Close the tag list */
2094
2095         return (int)((char *)tl - (char *)reply->tag_list);
2096 }
2097
2098 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2099                                     struct drbd_nl_cfg_reply *reply)
2100 {
2101         /* default to resume from last known position, if possible */
2102         struct start_ov args =
2103                 { .start_sector = mdev->ov_start_sector };
2104
2105         if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2106                 reply->ret_code = ERR_MANDATORY_TAG;
2107                 return 0;
2108         }
2109
2110         /* If there is still bitmap IO pending, e.g. previous resync or verify
2111          * just being finished, wait for it before requesting a new resync. */
2112         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2113
2114         /* w_make_ov_request expects position to be aligned */
2115         mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
2116         reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2117         return 0;
2118 }
2119
2120
2121 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2122                               struct drbd_nl_cfg_reply *reply)
2123 {
2124         int retcode = NO_ERROR;
2125         int skip_initial_sync = 0;
2126         int err;
2127
2128         struct new_c_uuid args;
2129
2130         memset(&args, 0, sizeof(struct new_c_uuid));
2131         if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2132                 reply->ret_code = ERR_MANDATORY_TAG;
2133                 return 0;
2134         }
2135
2136         mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
2137
2138         if (!get_ldev(mdev)) {
2139                 retcode = ERR_NO_DISK;
2140                 goto out;
2141         }
2142
2143         /* this is "skip initial sync", assume to be clean */
2144         if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
2145             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2146                 dev_info(DEV, "Preparing to skip initial sync\n");
2147                 skip_initial_sync = 1;
2148         } else if (mdev->state.conn != C_STANDALONE) {
2149                 retcode = ERR_CONNECTED;
2150                 goto out_dec;
2151         }
2152
2153         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2154         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2155
2156         if (args.clear_bm) {
2157                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2158                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2159                 if (err) {
2160                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2161                         retcode = ERR_IO_MD_DISK;
2162                 }
2163                 if (skip_initial_sync) {
2164                         drbd_send_uuids_skip_initial_sync(mdev);
2165                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2166                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2167                         spin_lock_irq(&mdev->req_lock);
2168                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2169                                         CS_VERBOSE, NULL);
2170                         spin_unlock_irq(&mdev->req_lock);
2171                 }
2172         }
2173
2174         drbd_md_sync(mdev);
2175 out_dec:
2176         put_ldev(mdev);
2177 out:
2178         mutex_unlock(&mdev->state_mutex);
2179
2180         reply->ret_code = retcode;
2181         return 0;
2182 }
2183
2184 struct cn_handler_struct {
2185         int (*function)(struct drbd_conf *,
2186                          struct drbd_nl_cfg_req *,
2187                          struct drbd_nl_cfg_reply *);
2188         int reply_body_size;
2189 };
2190
2191 static struct cn_handler_struct cnd_table[] = {
2192         [ P_primary ]           = { &drbd_nl_primary,           0 },
2193         [ P_secondary ]         = { &drbd_nl_secondary,         0 },
2194         [ P_disk_conf ]         = { &drbd_nl_disk_conf,         0 },
2195         [ P_detach ]            = { &drbd_nl_detach,            0 },
2196         [ P_net_conf ]          = { &drbd_nl_net_conf,          0 },
2197         [ P_disconnect ]        = { &drbd_nl_disconnect,        0 },
2198         [ P_resize ]            = { &drbd_nl_resize,            0 },
2199         [ P_syncer_conf ]       = { &drbd_nl_syncer_conf,       0 },
2200         [ P_invalidate ]        = { &drbd_nl_invalidate,        0 },
2201         [ P_invalidate_peer ]   = { &drbd_nl_invalidate_peer,   0 },
2202         [ P_pause_sync ]        = { &drbd_nl_pause_sync,        0 },
2203         [ P_resume_sync ]       = { &drbd_nl_resume_sync,       0 },
2204         [ P_suspend_io ]        = { &drbd_nl_suspend_io,        0 },
2205         [ P_resume_io ]         = { &drbd_nl_resume_io,         0 },
2206         [ P_outdate ]           = { &drbd_nl_outdate,           0 },
2207         [ P_get_config ]        = { &drbd_nl_get_config,
2208                                     sizeof(struct syncer_conf_tag_len_struct) +
2209                                     sizeof(struct disk_conf_tag_len_struct) +
2210                                     sizeof(struct net_conf_tag_len_struct) },
2211         [ P_get_state ]         = { &drbd_nl_get_state,
2212                                     sizeof(struct get_state_tag_len_struct) +
2213                                     sizeof(struct sync_progress_tag_len_struct) },
2214         [ P_get_uuids ]         = { &drbd_nl_get_uuids,
2215                                     sizeof(struct get_uuids_tag_len_struct) },
2216         [ P_get_timeout_flag ]  = { &drbd_nl_get_timeout_flag,
2217                                     sizeof(struct get_timeout_flag_tag_len_struct)},
2218         [ P_start_ov ]          = { &drbd_nl_start_ov,          0 },
2219         [ P_new_c_uuid ]        = { &drbd_nl_new_c_uuid,        0 },
2220 };
2221
2222 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2223 {
2224         struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2225         struct cn_handler_struct *cm;
2226         struct cn_msg *cn_reply;
2227         struct drbd_nl_cfg_reply *reply;
2228         struct drbd_conf *mdev;
2229         int retcode, rr;
2230         int reply_size = sizeof(struct cn_msg)
2231                 + sizeof(struct drbd_nl_cfg_reply)
2232                 + sizeof(short int);
2233
2234         if (!try_module_get(THIS_MODULE)) {
2235                 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2236                 return;
2237         }
2238
2239         if (!cap_raised(current_cap(), CAP_SYS_ADMIN)) {
2240                 retcode = ERR_PERM;
2241                 goto fail;
2242         }
2243
2244         mdev = ensure_mdev(nlp->drbd_minor,
2245                         (nlp->flags & DRBD_NL_CREATE_DEVICE));
2246         if (!mdev) {
2247                 retcode = ERR_MINOR_INVALID;
2248                 goto fail;
2249         }
2250
2251         if (nlp->packet_type >= P_nl_after_last_packet ||
2252             nlp->packet_type == P_return_code_only) {
2253                 retcode = ERR_PACKET_NR;
2254                 goto fail;
2255         }
2256
2257         cm = cnd_table + nlp->packet_type;
2258
2259         /* This may happen if packet number is 0: */
2260         if (cm->function == NULL) {
2261                 retcode = ERR_PACKET_NR;
2262                 goto fail;
2263         }
2264
2265         reply_size += cm->reply_body_size;
2266
2267         /* allocation not in the IO path, cqueue thread context */
2268         cn_reply = kzalloc(reply_size, GFP_KERNEL);
2269         if (!cn_reply) {
2270                 retcode = ERR_NOMEM;
2271                 goto fail;
2272         }
2273         reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2274
2275         reply->packet_type =
2276                 cm->reply_body_size ? nlp->packet_type : P_return_code_only;
2277         reply->minor = nlp->drbd_minor;
2278         reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2279         /* reply->tag_list; might be modified by cm->function. */
2280
2281         rr = cm->function(mdev, nlp, reply);
2282
2283         cn_reply->id = req->id;
2284         cn_reply->seq = req->seq;
2285         cn_reply->ack = req->ack  + 1;
2286         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2287         cn_reply->flags = 0;
2288
2289         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2290         if (rr && rr != -ESRCH)
2291                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2292
2293         kfree(cn_reply);
2294         module_put(THIS_MODULE);
2295         return;
2296  fail:
2297         drbd_nl_send_reply(req, retcode);
2298         module_put(THIS_MODULE);
2299 }
2300
2301 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2302
2303 static unsigned short *
2304 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2305         unsigned short len, int nul_terminated)
2306 {
2307         unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2308         len = (len < l) ? len :  l;
2309         put_unaligned(tag, tl++);
2310         put_unaligned(len, tl++);
2311         memcpy(tl, data, len);
2312         tl = (unsigned short*)((char*)tl + len);
2313         if (nul_terminated)
2314                 *((char*)tl - 1) = 0;
2315         return tl;
2316 }
2317
2318 static unsigned short *
2319 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2320 {
2321         return __tl_add_blob(tl, tag, data, len, 0);
2322 }
2323
2324 static unsigned short *
2325 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2326 {
2327         return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2328 }
2329
2330 static unsigned short *
2331 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2332 {
2333         put_unaligned(tag, tl++);
2334         switch(tag_type(tag)) {
2335         case TT_INTEGER:
2336                 put_unaligned(sizeof(int), tl++);
2337                 put_unaligned(*(int *)val, (int *)tl);
2338                 tl = (unsigned short*)((char*)tl+sizeof(int));
2339                 break;
2340         case TT_INT64:
2341                 put_unaligned(sizeof(u64), tl++);
2342                 put_unaligned(*(u64 *)val, (u64 *)tl);
2343                 tl = (unsigned short*)((char*)tl+sizeof(u64));
2344                 break;
2345         default:
2346                 /* someone did something stupid. */
2347                 ;
2348         }
2349         return tl;
2350 }
2351
2352 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2353 {
2354         char buffer[sizeof(struct cn_msg)+
2355                     sizeof(struct drbd_nl_cfg_reply)+
2356                     sizeof(struct get_state_tag_len_struct)+
2357                     sizeof(short int)];
2358         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2359         struct drbd_nl_cfg_reply *reply =
2360                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2361         unsigned short *tl = reply->tag_list;
2362
2363         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2364
2365         tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2366
2367         put_unaligned(TT_END, tl++); /* Close the tag list */
2368
2369         cn_reply->id.idx = CN_IDX_DRBD;
2370         cn_reply->id.val = CN_VAL_DRBD;
2371
2372         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2373         cn_reply->ack = 0; /* not used here. */
2374         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2375                 (int)((char *)tl - (char *)reply->tag_list);
2376         cn_reply->flags = 0;
2377
2378         reply->packet_type = P_get_state;
2379         reply->minor = mdev_to_minor(mdev);
2380         reply->ret_code = NO_ERROR;
2381
2382         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2383 }
2384
2385 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2386 {
2387         char buffer[sizeof(struct cn_msg)+
2388                     sizeof(struct drbd_nl_cfg_reply)+
2389                     sizeof(struct call_helper_tag_len_struct)+
2390                     sizeof(short int)];
2391         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2392         struct drbd_nl_cfg_reply *reply =
2393                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2394         unsigned short *tl = reply->tag_list;
2395
2396         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2397
2398         tl = tl_add_str(tl, T_helper, helper_name);
2399         put_unaligned(TT_END, tl++); /* Close the tag list */
2400
2401         cn_reply->id.idx = CN_IDX_DRBD;
2402         cn_reply->id.val = CN_VAL_DRBD;
2403
2404         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2405         cn_reply->ack = 0; /* not used here. */
2406         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2407                 (int)((char *)tl - (char *)reply->tag_list);
2408         cn_reply->flags = 0;
2409
2410         reply->packet_type = P_call_helper;
2411         reply->minor = mdev_to_minor(mdev);
2412         reply->ret_code = NO_ERROR;
2413
2414         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2415 }
2416
2417 void drbd_bcast_ee(struct drbd_conf *mdev,
2418                 const char *reason, const int dgs,
2419                 const char* seen_hash, const char* calc_hash,
2420                 const struct drbd_epoch_entry* e)
2421 {
2422         struct cn_msg *cn_reply;
2423         struct drbd_nl_cfg_reply *reply;
2424         unsigned short *tl;
2425         struct page *page;
2426         unsigned len;
2427
2428         if (!e)
2429                 return;
2430         if (!reason || !reason[0])
2431                 return;
2432
2433         /* apparently we have to memcpy twice, first to prepare the data for the
2434          * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2435          * netlink skb. */
2436         /* receiver thread context, which is not in the writeout path (of this node),
2437          * but may be in the writeout path of the _other_ node.
2438          * GFP_NOIO to avoid potential "distributed deadlock". */
2439         cn_reply = kzalloc(
2440                 sizeof(struct cn_msg)+
2441                 sizeof(struct drbd_nl_cfg_reply)+
2442                 sizeof(struct dump_ee_tag_len_struct)+
2443                 sizeof(short int),
2444                 GFP_NOIO);
2445
2446         if (!cn_reply) {
2447                 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2448                                 (unsigned long long)e->sector, e->size);
2449                 return;
2450         }
2451
2452         reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2453         tl = reply->tag_list;
2454
2455         tl = tl_add_str(tl, T_dump_ee_reason, reason);
2456         tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2457         tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2458         tl = tl_add_int(tl, T_ee_sector, &e->sector);
2459         tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2460
2461         /* dump the first 32k */
2462         len = min_t(unsigned, e->size, 32 << 10);
2463         put_unaligned(T_ee_data, tl++);
2464         put_unaligned(len, tl++);
2465
2466         page = e->pages;
2467         page_chain_for_each(page) {
2468                 void *d = kmap_atomic(page, KM_USER0);
2469                 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2470                 memcpy(tl, d, l);
2471                 kunmap_atomic(d, KM_USER0);
2472                 tl = (unsigned short*)((char*)tl + l);
2473                 len -= l;
2474                 if (len == 0)
2475                         break;
2476         }
2477         put_unaligned(TT_END, tl++); /* Close the tag list */
2478
2479         cn_reply->id.idx = CN_IDX_DRBD;
2480         cn_reply->id.val = CN_VAL_DRBD;
2481
2482         cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2483         cn_reply->ack = 0; // not used here.
2484         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2485                 (int)((char*)tl - (char*)reply->tag_list);
2486         cn_reply->flags = 0;
2487
2488         reply->packet_type = P_dump_ee;
2489         reply->minor = mdev_to_minor(mdev);
2490         reply->ret_code = NO_ERROR;
2491
2492         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2493         kfree(cn_reply);
2494 }
2495
2496 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2497 {
2498         char buffer[sizeof(struct cn_msg)+
2499                     sizeof(struct drbd_nl_cfg_reply)+
2500                     sizeof(struct sync_progress_tag_len_struct)+
2501                     sizeof(short int)];
2502         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2503         struct drbd_nl_cfg_reply *reply =
2504                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2505         unsigned short *tl = reply->tag_list;
2506         unsigned long rs_left;
2507         unsigned int res;
2508
2509         /* no local ref, no bitmap, no syncer progress, no broadcast. */
2510         if (!get_ldev(mdev))
2511                 return;
2512         drbd_get_syncer_progress(mdev, &rs_left, &res);
2513         put_ldev(mdev);
2514
2515         tl = tl_add_int(tl, T_sync_progress, &res);
2516         put_unaligned(TT_END, tl++); /* Close the tag list */
2517
2518         cn_reply->id.idx = CN_IDX_DRBD;
2519         cn_reply->id.val = CN_VAL_DRBD;
2520
2521         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2522         cn_reply->ack = 0; /* not used here. */
2523         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2524                 (int)((char *)tl - (char *)reply->tag_list);
2525         cn_reply->flags = 0;
2526
2527         reply->packet_type = P_sync_progress;
2528         reply->minor = mdev_to_minor(mdev);
2529         reply->ret_code = NO_ERROR;
2530
2531         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2532 }
2533
2534 int __init drbd_nl_init(void)
2535 {
2536         static struct cb_id cn_id_drbd;
2537         int err, try=10;
2538
2539         cn_id_drbd.val = CN_VAL_DRBD;
2540         do {
2541                 cn_id_drbd.idx = cn_idx;
2542                 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2543                 if (!err)
2544                         break;
2545                 cn_idx = (cn_idx + CN_IDX_STEP);
2546         } while (try--);
2547
2548         if (err) {
2549                 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2550                 return err;
2551         }
2552
2553         return 0;
2554 }
2555
2556 void drbd_nl_cleanup(void)
2557 {
2558         static struct cb_id cn_id_drbd;
2559
2560         cn_id_drbd.idx = cn_idx;
2561         cn_id_drbd.val = CN_VAL_DRBD;
2562
2563         cn_del_callback(&cn_id_drbd);
2564 }
2565
2566 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2567 {
2568         char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2569         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2570         struct drbd_nl_cfg_reply *reply =
2571                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2572         int rr;
2573
2574         memset(buffer, 0, sizeof(buffer));
2575         cn_reply->id = req->id;
2576
2577         cn_reply->seq = req->seq;
2578         cn_reply->ack = req->ack  + 1;
2579         cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2580         cn_reply->flags = 0;
2581
2582         reply->packet_type = P_return_code_only;
2583         reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2584         reply->ret_code = ret_code;
2585
2586         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2587         if (rr && rr != -ESRCH)
2588                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2589 }
2590