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, UMH_WAIT_PROC);
 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         union drbd_state ns;
 276
 277         nps = drbd_try_outdate_peer(mdev);
 278
 279         /* Not using
 280            drbd_request_state(mdev, NS(pdsk, nps));
 281            here, because we might were able to re-establish the connection
 282            in the meantime. This can only partially be solved in the state's
 283            engine is_valid_state() and is_valid_state_transition()
 284            functions.
 285
 286            nps can be D_INCONSISTENT, D_OUTDATED or D_UNKNOWN.
 287            pdsk == D_INCONSISTENT while conn >= C_CONNECTED is valid,
 288            therefore we have to have the pre state change check here.
 289         */
 290         spin_lock_irq(&mdev->req_lock);
 291         ns = mdev->state;
 292         if (ns.conn < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &mdev->flags)) {
 293                 ns.pdsk = nps;
 294                 _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
 295         }
 296         spin_unlock_irq(&mdev->req_lock);
 297
 298         return 0;
 299 }
 300
 301 void drbd_try_outdate_peer_async(struct drbd_conf *mdev)
 302 {
 303         struct task_struct *opa;
 304
 305         opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev));
 306         if (IS_ERR(opa))
 307                 dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n");
 308 }
 309
 310 enum drbd_state_rv
 311 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 312 {
 313         const int max_tries = 4;
 314         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
 315         int try = 0;
 316         int forced = 0;
 317         union drbd_state mask, val;
 318         enum drbd_disk_state nps;
 319
 320         if (new_role == R_PRIMARY)
 321                 request_ping(mdev); /* Detect a dead peer ASAP */
 322
 323         mutex_lock(&mdev->state_mutex);
 324
 325         mask.i = 0; mask.role = R_MASK;
 326         val.i  = 0; val.role  = new_role;
 327
 328         while (try++ < max_tries) {
 329                 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 330
 331                 /* in case we first succeeded to outdate,
 332                  * but now suddenly could establish a connection */
 333                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 334                         val.pdsk = 0;
 335                         mask.pdsk = 0;
 336                         continue;
 337                 }
 338
 339                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
 340                     (mdev->state.disk < D_UP_TO_DATE &&
 341                      mdev->state.disk >= D_INCONSISTENT)) {
 342                         mask.disk = D_MASK;
 343                         val.disk  = D_UP_TO_DATE;
 344                         forced = 1;
 345                         continue;
 346                 }
 347
 348                 if (rv == SS_NO_UP_TO_DATE_DISK &&
 349                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 350                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 351                         nps = drbd_try_outdate_peer(mdev);
 352
 353                         if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
 354                                 val.disk = D_UP_TO_DATE;
 355                                 mask.disk = D_MASK;
 356                         }
 357
 358                         val.pdsk = nps;
 359                         mask.pdsk = D_MASK;
 360
 361                         continue;
 362                 }
 363
 364                 if (rv == SS_NOTHING_TO_DO)
 365                         goto fail;
 366                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
 367                         nps = drbd_try_outdate_peer(mdev);
 368
 369                         if (force && nps > D_OUTDATED) {
 370                                 dev_warn(DEV, "Forced into split brain situation!\n");
 371                                 nps = D_OUTDATED;
 372                         }
 373
 374                         mask.pdsk = D_MASK;
 375                         val.pdsk  = nps;
 376
 377                         continue;
 378                 }
 379                 if (rv == SS_TWO_PRIMARIES) {
 380                         /* Maybe the peer is detected as dead very soon...
 381                            retry at most once more in this case. */
 382                         schedule_timeout_interruptible((mdev->net_conf->ping_timeo+1)*HZ/10);
 383                         if (try < max_tries)
 384                                 try = max_tries - 1;
 385                         continue;
 386                 }
 387                 if (rv < SS_SUCCESS) {
 388                         rv = _drbd_request_state(mdev, mask, val,
 389                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 390                         if (rv < SS_SUCCESS)
 391                                 goto fail;
 392                 }
 393                 break;
 394         }
 395
 396         if (rv < SS_SUCCESS)
 397                 goto fail;
 398
 399         if (forced)
 400                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 401
 402         /* Wait until nothing is on the fly :) */
 403         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 404
 405         if (new_role == R_SECONDARY) {
 406                 set_disk_ro(mdev->vdisk, true);
 407                 if (get_ldev(mdev)) {
 408                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 409                         put_ldev(mdev);
 410                 }
 411         } else {
 412                 if (get_net_conf(mdev)) {
 413                         mdev->net_conf->want_lose = 0;
 414                         put_net_conf(mdev);
 415                 }
 416                 set_disk_ro(mdev->vdisk, false);
 417                 if (get_ldev(mdev)) {
 418                         if (((mdev->state.conn < C_CONNECTED ||
 419                                mdev->state.pdsk <= D_FAILED)
 420                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 421                                 drbd_uuid_new_current(mdev);
 422
 423                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 424                         put_ldev(mdev);
 425                 }
 426         }
 427
 428         /* writeout of activity log covered areas of the bitmap
 429          * to stable storage done in after state change already */
 430
 431         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 432                 /* if this was forced, we should consider sync */
 433                 if (forced)
 434                         drbd_send_uuids(mdev);
 435                 drbd_send_current_state(mdev);
 436         }
 437
 438         drbd_md_sync(mdev);
 439
 440         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 441  fail:
 442         mutex_unlock(&mdev->state_mutex);
 443         return rv;
 444 }
 445
 446 static struct drbd_conf *ensure_mdev(int minor, int create)
 447 {
 448         struct drbd_conf *mdev;
 449
 450         if (minor >= minor_count)
 451                 return NULL;
 452
 453         mdev = minor_to_mdev(minor);
 454
 455         if (!mdev && create) {
 456                 struct gendisk *disk = NULL;
 457                 mdev = drbd_new_device(minor);
 458
 459                 spin_lock_irq(&drbd_pp_lock);
 460                 if (minor_table[minor] == NULL) {
 461                         minor_table[minor] = mdev;
 462                         disk = mdev->vdisk;
 463                         mdev = NULL;
 464                 } /* else: we lost the race */
 465                 spin_unlock_irq(&drbd_pp_lock);
 466
 467                 if (disk) /* we won the race above */
 468                         /* in case we ever add a drbd_delete_device(),
 469                          * don't forget the del_gendisk! */
 470                         add_disk(disk);
 471                 else /* we lost the race above */
 472                         drbd_free_mdev(mdev);
 473
 474                 mdev = minor_to_mdev(minor);
 475         }
 476
 477         return mdev;
 478 }
 479
 480 static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 481                            struct drbd_nl_cfg_reply *reply)
 482 {
 483         struct primary primary_args;
 484
 485         memset(&primary_args, 0, sizeof(struct primary));
 486         if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) {
 487                 reply->ret_code = ERR_MANDATORY_TAG;
 488                 return 0;
 489         }
 490
 491         reply->ret_code =
 492                 drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force);
 493
 494         return 0;
 495 }
 496
 497 static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 498                              struct drbd_nl_cfg_reply *reply)
 499 {
 500         reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0);
 501
 502         return 0;
 503 }
 504
 505 /* initializes the md.*_offset members, so we are able to find
 506  * the on disk meta data */
 507 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 508                                        struct drbd_backing_dev *bdev)
 509 {
 510         sector_t md_size_sect = 0;
 511         switch (bdev->dc.meta_dev_idx) {
 512         default:
 513                 /* v07 style fixed size indexed meta data */
 514                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 515                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 516                 bdev->md.al_offset = MD_AL_OFFSET;
 517                 bdev->md.bm_offset = MD_BM_OFFSET;
 518                 break;
 519         case DRBD_MD_INDEX_FLEX_EXT:
 520                 /* just occupy the full device; unit: sectors */
 521                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 522                 bdev->md.md_offset = 0;
 523                 bdev->md.al_offset = MD_AL_OFFSET;
 524                 bdev->md.bm_offset = MD_BM_OFFSET;
 525                 break;
 526         case DRBD_MD_INDEX_INTERNAL:
 527         case DRBD_MD_INDEX_FLEX_INT:
 528                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 529                 /* al size is still fixed */
 530                 bdev->md.al_offset = -MD_AL_MAX_SIZE;
 531                 /* we need (slightly less than) ~ this much bitmap sectors: */
 532                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 533                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 534                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 535                 md_size_sect = ALIGN(md_size_sect, 8);
 536
 537                 /* plus the "drbd meta data super block",
 538                  * and the activity log; */
 539                 md_size_sect += MD_BM_OFFSET;
 540
 541                 bdev->md.md_size_sect = md_size_sect;
 542                 /* bitmap offset is adjusted by 'super' block size */
 543                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 544                 break;
 545         }
 546 }
 547
 548 /* input size is expected to be in KB */
 549 char *ppsize(char *buf, unsigned long long size)
 550 {
 551         /* Needs 9 bytes at max including trailing NUL:
 552          * -1ULL ==> "16384 EB" */
 553         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 554         int base = 0;
 555         while (size >= 10000 && base < sizeof(units)-1) {
 556                 /* shift + round */
 557                 size = (size >> 10) + !!(size & (1<<9));
 558                 base++;
 559         }
 560         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
 561
 562         return buf;
 563 }
 564
 565 /* there is still a theoretical deadlock when called from receiver
 566  * on an D_INCONSISTENT R_PRIMARY:
 567  *  remote READ does inc_ap_bio, receiver would need to receive answer
 568  *  packet from remote to dec_ap_bio again.
 569  *  receiver receive_sizes(), comes here,
 570  *  waits for ap_bio_cnt == 0. -> deadlock.
 571  * but this cannot happen, actually, because:
 572  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 573  *  (not connected, or bad/no disk on peer):
 574  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 575  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 576  *  peer may not initiate a resize.
 577  */
 578 void drbd_suspend_io(struct drbd_conf *mdev)
 579 {
 580         set_bit(SUSPEND_IO, &mdev->flags);
 581         if (is_susp(mdev->state))
 582                 return;
 583         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 584 }
 585
 586 void drbd_resume_io(struct drbd_conf *mdev)
 587 {
 588         clear_bit(SUSPEND_IO, &mdev->flags);
 589         wake_up(&mdev->misc_wait);
 590 }
 591
 592 /**
 593  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 594  * @mdev:       DRBD device.
 595  *
 596  * Returns 0 on success, negative return values indicate errors.
 597  * You should call drbd_md_sync() after calling this function.
 598  */
 599 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 600 {
 601         sector_t prev_first_sect, prev_size; /* previous meta location */
 602         sector_t la_size;
 603         sector_t size;
 604         char ppb[10];
 605
 606         int md_moved, la_size_changed;
 607         enum determine_dev_size rv = unchanged;
 608
 609         /* race:
 610          * application request passes inc_ap_bio,
 611          * but then cannot get an AL-reference.
 612          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 613          *
 614          * to avoid that:
 615          * Suspend IO right here.
 616          * still lock the act_log to not trigger ASSERTs there.
 617          */
 618         drbd_suspend_io(mdev);
 619
 620         /* no wait necessary anymore, actually we could assert that */
 621         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 622
 623         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 624         prev_size = mdev->ldev->md.md_size_sect;
 625         la_size = mdev->ldev->md.la_size_sect;
 626
 627         /* TODO: should only be some assert here, not (re)init... */
 628         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 629
 630         size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
 631
 632         if (drbd_get_capacity(mdev->this_bdev) != size ||
 633             drbd_bm_capacity(mdev) != size) {
 634                 int err;
 635                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 636                 if (unlikely(err)) {
 637                         /* currently there is only one error: ENOMEM! */
 638                         size = drbd_bm_capacity(mdev)>>1;
 639                         if (size == 0) {
 640                                 dev_err(DEV, "OUT OF MEMORY! "
 641                                     "Could not allocate bitmap!\n");
 642                         } else {
 643                                 dev_err(DEV, "BM resizing failed. "
 644                                     "Leaving size unchanged at size = %lu KB\n",
 645                                     (unsigned long)size);
 646                         }
 647                         rv = dev_size_error;
 648                 }
 649                 /* racy, see comments above. */
 650                 drbd_set_my_capacity(mdev, size);
 651                 mdev->ldev->md.la_size_sect = size;
 652                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 653                      (unsigned long long)size>>1);
 654         }
 655         if (rv == dev_size_error)
 656                 goto out;
 657
 658         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 659
 660         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 661                 || prev_size       != mdev->ldev->md.md_size_sect;
 662
 663         if (la_size_changed || md_moved) {
 664                 int err;
 665
 666                 drbd_al_shrink(mdev); /* All extents inactive. */
 667                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 668                          la_size_changed && md_moved ? "size changed and md moved" :
 669                          la_size_changed ? "size changed" : "md moved");
 670                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
 671                 err = drbd_bitmap_io(mdev, &drbd_bm_write,
 672                                 "size changed", BM_LOCKED_MASK);
 673                 if (err) {
 674                         rv = dev_size_error;
 675                         goto out;
 676                 }
 677                 drbd_md_mark_dirty(mdev);
 678         }
 679
 680         if (size > la_size)
 681                 rv = grew;
 682         if (size < la_size)
 683                 rv = shrunk;
 684 out:
 685         lc_unlock(mdev->act_log);
 686         wake_up(&mdev->al_wait);
 687         drbd_resume_io(mdev);
 688
 689         return rv;
 690 }
 691
 692 sector_t
 693 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
 694 {
 695         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 696         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 697         sector_t m_size; /* my size */
 698         sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
 699         sector_t size = 0;
 700
 701         m_size = drbd_get_max_capacity(bdev);
 702
 703         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 704                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 705                 p_size = m_size;
 706         }
 707
 708         if (p_size && m_size) {
 709                 size = min_t(sector_t, p_size, m_size);
 710         } else {
 711                 if (la_size) {
 712                         size = la_size;
 713                         if (m_size && m_size < size)
 714                                 size = m_size;
 715                         if (p_size && p_size < size)
 716                                 size = p_size;
 717                 } else {
 718                         if (m_size)
 719                                 size = m_size;
 720                         if (p_size)
 721                                 size = p_size;
 722                 }
 723         }
 724
 725         if (size == 0)
 726                 dev_err(DEV, "Both nodes diskless!\n");
 727
 728         if (u_size) {
 729                 if (u_size > size)
 730                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 731                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 732                 else
 733                         size = u_size;
 734         }
 735
 736         return size;
 737 }
 738
 739 /**
 740  * drbd_check_al_size() - Ensures that the AL is of the right size
 741  * @mdev:       DRBD device.
 742  *
 743  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 744  * failed, and 0 on success. You should call drbd_md_sync() after you called
 745  * this function.
 746  */
 747 static int drbd_check_al_size(struct drbd_conf *mdev)
 748 {
 749         struct lru_cache *n, *t;
 750         struct lc_element *e;
 751         unsigned int in_use;
 752         int i;
 753
 754         ERR_IF(mdev->sync_conf.al_extents < 7)
 755                 mdev->sync_conf.al_extents = 127;
 756
 757         if (mdev->act_log &&
 758             mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
 759                 return 0;
 760
 761         in_use = 0;
 762         t = mdev->act_log;
 763         n = lc_create("act_log", drbd_al_ext_cache,
 764                 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
 765
 766         if (n == NULL) {
 767                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 768                 return -ENOMEM;
 769         }
 770         spin_lock_irq(&mdev->al_lock);
 771         if (t) {
 772                 for (i = 0; i < t->nr_elements; i++) {
 773                         e = lc_element_by_index(t, i);
 774                         if (e->refcnt)
 775                                 dev_err(DEV, "refcnt(%d)==%d\n",
 776                                     e->lc_number, e->refcnt);
 777                         in_use += e->refcnt;
 778                 }
 779         }
 780         if (!in_use)
 781                 mdev->act_log = n;
 782         spin_unlock_irq(&mdev->al_lock);
 783         if (in_use) {
 784                 dev_err(DEV, "Activity log still in use!\n");
 785                 lc_destroy(n);
 786                 return -EBUSY;
 787         } else {
 788                 if (t)
 789                         lc_destroy(t);
 790         }
 791         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 792         return 0;
 793 }
 794
 795 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
 796 {
 797         struct request_queue * const q = mdev->rq_queue;
 798         int max_hw_sectors = max_bio_size >> 9;
 799         int max_segments = 0;
 800
 801         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 802                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 803
 804                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
 805                 max_segments = mdev->ldev->dc.max_bio_bvecs;
 806                 put_ldev(mdev);
 807         }
 808
 809         blk_queue_logical_block_size(q, 512);
 810         blk_queue_max_hw_sectors(q, max_hw_sectors);
 811         /* This is the workaround for "bio would need to, but cannot, be split" */
 812         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
 813         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
 814
 815         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 816                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 817
 818                 blk_queue_stack_limits(q, b);
 819
 820                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
 821                         dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
 822                                  q->backing_dev_info.ra_pages,
 823                                  b->backing_dev_info.ra_pages);
 824                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
 825                 }
 826                 put_ldev(mdev);
 827         }
 828 }
 829
 830 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
 831 {
 832         int now, new, local, peer;
 833
 834         now = queue_max_hw_sectors(mdev->rq_queue) << 9;
 835         local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
 836         peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
 837
 838         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 839                 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
 840                 mdev->local_max_bio_size = local;
 841                 put_ldev(mdev);
 842         }
 843
 844         /* We may ignore peer limits if the peer is modern enough.
 845            Because new from 8.3.8 onwards the peer can use multiple
 846            BIOs for a single peer_request */
 847         if (mdev->state.conn >= C_CONNECTED) {
 848                 if (mdev->agreed_pro_version < 94) {
 849                         peer = min_t(int, mdev->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
 850                         /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
 851                 } else if (mdev->agreed_pro_version == 94)
 852                         peer = DRBD_MAX_SIZE_H80_PACKET;
 853                 else /* drbd 8.3.8 onwards */
 854                         peer = DRBD_MAX_BIO_SIZE;
 855         }
 856
 857         new = min_t(int, local, peer);
 858
 859         if (mdev->state.role == R_PRIMARY && new < now)
 860                 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
 861
 862         if (new != now)
 863                 dev_info(DEV, "max BIO size = %u\n", new);
 864
 865         drbd_setup_queue_param(mdev, new);
 866 }
 867
 868 /* serialize deconfig (worker exiting, doing cleanup)
 869  * and reconfig (drbdsetup disk, drbdsetup net)
 870  *
 871  * Wait for a potentially exiting worker, then restart it,
 872  * or start a new one.  Flush any pending work, there may still be an
 873  * after_state_change queued.
 874  */
 875 static void drbd_reconfig_start(struct drbd_conf *mdev)
 876 {
 877         wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
 878         wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
 879         drbd_thread_start(&mdev->worker);
 880         drbd_flush_workqueue(mdev);
 881 }
 882
 883 /* if still unconfigured, stops worker again.
 884  * if configured now, clears CONFIG_PENDING.
 885  * wakes potential waiters */
 886 static void drbd_reconfig_done(struct drbd_conf *mdev)
 887 {
 888         spin_lock_irq(&mdev->req_lock);
 889         if (mdev->state.disk == D_DISKLESS &&
 890             mdev->state.conn == C_STANDALONE &&
 891             mdev->state.role == R_SECONDARY) {
 892                 set_bit(DEVICE_DYING, &mdev->flags);
 893                 drbd_thread_stop_nowait(&mdev->worker);
 894         } else
 895                 clear_bit(CONFIG_PENDING, &mdev->flags);
 896         spin_unlock_irq(&mdev->req_lock);
 897         wake_up(&mdev->state_wait);
 898 }
 899
 900 /* Make sure IO is suspended before calling this function(). */
 901 static void drbd_suspend_al(struct drbd_conf *mdev)
 902 {
 903         int s = 0;
 904
 905         if (lc_try_lock(mdev->act_log)) {
 906                 drbd_al_shrink(mdev);
 907                 lc_unlock(mdev->act_log);
 908         } else {
 909                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
 910                 return;
 911         }
 912
 913         spin_lock_irq(&mdev->req_lock);
 914         if (mdev->state.conn < C_CONNECTED)
 915                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
 916
 917         spin_unlock_irq(&mdev->req_lock);
 918
 919         if (s)
 920                 dev_info(DEV, "Suspended AL updates\n");
 921 }
 922
 923 /* does always return 0;
 924  * interesting return code is in reply->ret_code */
 925 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 926                              struct drbd_nl_cfg_reply *reply)
 927 {
 928         enum drbd_ret_code retcode;
 929         enum determine_dev_size dd;
 930         sector_t max_possible_sectors;
 931         sector_t min_md_device_sectors;
 932         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
 933         struct block_device *bdev;
 934         struct lru_cache *resync_lru = NULL;
 935         union drbd_state ns, os;
 936         enum drbd_state_rv rv;
 937         int cp_discovered = 0;
 938         int logical_block_size;
 939
 940         drbd_reconfig_start(mdev);
 941
 942         /* if you want to reconfigure, please tear down first */
 943         if (mdev->state.disk > D_DISKLESS) {
 944                 retcode = ERR_DISK_CONFIGURED;
 945                 goto fail;
 946         }
 947         /* It may just now have detached because of IO error.  Make sure
 948          * drbd_ldev_destroy is done already, we may end up here very fast,
 949          * e.g. if someone calls attach from the on-io-error handler,
 950          * to realize a "hot spare" feature (not that I'd recommend that) */
 951         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
 952
 953         /* make sure there is no leftover from previous force-detach attempts */
 954         clear_bit(FORCE_DETACH, &mdev->flags);
 955
 956         /* allocation not in the IO path, cqueue thread context */
 957         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
 958         if (!nbc) {
 959                 retcode = ERR_NOMEM;
 960                 goto fail;
 961         }
 962
 963         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
 964         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
 965         nbc->dc.fencing       = DRBD_FENCING_DEF;
 966         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
 967
 968         if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
 969                 retcode = ERR_MANDATORY_TAG;
 970                 goto fail;
 971         }
 972
 973         if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
 974                 retcode = ERR_MD_IDX_INVALID;
 975                 goto fail;
 976         }
 977
 978         if (get_net_conf(mdev)) {
 979                 int prot = mdev->net_conf->wire_protocol;
 980                 put_net_conf(mdev);
 981                 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
 982                         retcode = ERR_STONITH_AND_PROT_A;
 983                         goto fail;
 984                 }
 985         }
 986
 987         bdev = blkdev_get_by_path(nbc->dc.backing_dev,
 988                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
 989         if (IS_ERR(bdev)) {
 990                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
 991                         PTR_ERR(bdev));
 992                 retcode = ERR_OPEN_DISK;
 993                 goto fail;
 994         }
 995         nbc->backing_bdev = bdev;
 996
 997         /*
 998          * meta_dev_idx >= 0: external fixed size, possibly multiple
 999          * drbd sharing one meta device.  TODO in that case, paranoia
1000          * check that [md_bdev, meta_dev_idx] is not yet used by some
1001          * other drbd minor!  (if you use drbd.conf + drbdadm, that
1002          * should check it for you already; but if you don't, or
1003          * someone fooled it, we need to double check here)
1004          */
1005         bdev = blkdev_get_by_path(nbc->dc.meta_dev,
1006                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1007                                   (nbc->dc.meta_dev_idx < 0) ?
1008                                   (void *)mdev : (void *)drbd_m_holder);
1009         if (IS_ERR(bdev)) {
1010                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
1011                         PTR_ERR(bdev));
1012                 retcode = ERR_OPEN_MD_DISK;
1013                 goto fail;
1014         }
1015         nbc->md_bdev = bdev;
1016
1017         if ((nbc->backing_bdev == nbc->md_bdev) !=
1018             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1019              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1020                 retcode = ERR_MD_IDX_INVALID;
1021                 goto fail;
1022         }
1023
1024         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1025                         61, sizeof(struct bm_extent),
1026                         offsetof(struct bm_extent, lce));
1027         if (!resync_lru) {
1028                 retcode = ERR_NOMEM;
1029                 goto fail;
1030         }
1031
1032         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1033         drbd_md_set_sector_offsets(mdev, nbc);
1034
1035         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
1036                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1037                         (unsigned long long) drbd_get_max_capacity(nbc),
1038                         (unsigned long long) nbc->dc.disk_size);
1039                 retcode = ERR_DISK_TOO_SMALL;
1040                 goto fail;
1041         }
1042
1043         if (nbc->dc.meta_dev_idx < 0) {
1044                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1045                 /* at least one MB, otherwise it does not make sense */
1046                 min_md_device_sectors = (2<<10);
1047         } else {
1048                 max_possible_sectors = DRBD_MAX_SECTORS;
1049                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
1050         }
1051
1052         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1053                 retcode = ERR_MD_DISK_TOO_SMALL;
1054                 dev_warn(DEV, "refusing attach: md-device too small, "
1055                      "at least %llu sectors needed for this meta-disk type\n",
1056                      (unsigned long long) min_md_device_sectors);
1057                 goto fail;
1058         }
1059
1060         /* Make sure the new disk is big enough
1061          * (we may currently be R_PRIMARY with no local disk...) */
1062         if (drbd_get_max_capacity(nbc) <
1063             drbd_get_capacity(mdev->this_bdev)) {
1064                 retcode = ERR_DISK_TOO_SMALL;
1065                 goto fail;
1066         }
1067
1068         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1069
1070         if (nbc->known_size > max_possible_sectors) {
1071                 dev_warn(DEV, "==> truncating very big lower level device "
1072                         "to currently maximum possible %llu sectors <==\n",
1073                         (unsigned long long) max_possible_sectors);
1074                 if (nbc->dc.meta_dev_idx >= 0)
1075                         dev_warn(DEV, "==>> using internal or flexible "
1076                                       "meta data may help <<==\n");
1077         }
1078
1079         drbd_suspend_io(mdev);
1080         /* also wait for the last barrier ack. */
1081         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state));
1082         /* and for any other previously queued work */
1083         drbd_flush_workqueue(mdev);
1084
1085         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1086         retcode = rv;  /* FIXME: Type mismatch. */
1087         drbd_resume_io(mdev);
1088         if (rv < SS_SUCCESS)
1089                 goto fail;
1090
1091         if (!get_ldev_if_state(mdev, D_ATTACHING))
1092                 goto force_diskless;
1093
1094         drbd_md_set_sector_offsets(mdev, nbc);
1095
1096         /* allocate a second IO page if logical_block_size != 512 */
1097         logical_block_size = bdev_logical_block_size(nbc->md_bdev);
1098         if (logical_block_size == 0)
1099                 logical_block_size = MD_SECTOR_SIZE;
1100
1101         if (logical_block_size != MD_SECTOR_SIZE) {
1102                 if (!mdev->md_io_tmpp) {
1103                         struct page *page = alloc_page(GFP_NOIO);
1104                         if (!page)
1105                                 goto force_diskless_dec;
1106
1107                         dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
1108                              logical_block_size, MD_SECTOR_SIZE);
1109                         dev_warn(DEV, "Workaround engaged (has performance impact).\n");
1110
1111                         mdev->md_io_tmpp = page;
1112                 }
1113         }
1114
1115         if (!mdev->bitmap) {
1116                 if (drbd_bm_init(mdev)) {
1117                         retcode = ERR_NOMEM;
1118                         goto force_diskless_dec;
1119                 }
1120         }
1121
1122         retcode = drbd_md_read(mdev, nbc);
1123         if (retcode != NO_ERROR)
1124                 goto force_diskless_dec;
1125
1126         if (mdev->state.conn < C_CONNECTED &&
1127             mdev->state.role == R_PRIMARY &&
1128             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1129                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1130                     (unsigned long long)mdev->ed_uuid);
1131                 retcode = ERR_DATA_NOT_CURRENT;
1132                 goto force_diskless_dec;
1133         }
1134
1135         /* Since we are diskless, fix the activity log first... */
1136         if (drbd_check_al_size(mdev)) {
1137                 retcode = ERR_NOMEM;
1138                 goto force_diskless_dec;
1139         }
1140
1141         /* Prevent shrinking of consistent devices ! */
1142         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1143             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1144                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1145                 retcode = ERR_DISK_TOO_SMALL;
1146                 goto force_diskless_dec;
1147         }
1148
1149         if (!drbd_al_read_log(mdev, nbc)) {
1150                 retcode = ERR_IO_MD_DISK;
1151                 goto force_diskless_dec;
1152         }
1153
1154         /* Reset the "barriers don't work" bits here, then force meta data to
1155          * be written, to ensure we determine if barriers are supported. */
1156         if (nbc->dc.no_md_flush)
1157                 set_bit(MD_NO_FUA, &mdev->flags);
1158         else
1159                 clear_bit(MD_NO_FUA, &mdev->flags);
1160
1161         /* Point of no return reached.
1162          * Devices and memory are no longer released by error cleanup below.
1163          * now mdev takes over responsibility, and the state engine should
1164          * clean it up somewhere.  */
1165         D_ASSERT(mdev->ldev == NULL);
1166         mdev->ldev = nbc;
1167         mdev->resync = resync_lru;
1168         nbc = NULL;
1169         resync_lru = NULL;
1170
1171         mdev->write_ordering = WO_bdev_flush;
1172         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1173
1174         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1175                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1176         else
1177                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1178
1179         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1180             !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) {
1181                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1182                 cp_discovered = 1;
1183         }
1184
1185         mdev->send_cnt = 0;
1186         mdev->recv_cnt = 0;
1187         mdev->read_cnt = 0;
1188         mdev->writ_cnt = 0;
1189
1190         drbd_reconsider_max_bio_size(mdev);
1191
1192         /* If I am currently not R_PRIMARY,
1193          * but meta data primary indicator is set,
1194          * I just now recover from a hard crash,
1195          * and have been R_PRIMARY before that crash.
1196          *
1197          * Now, if I had no connection before that crash
1198          * (have been degraded R_PRIMARY), chances are that
1199          * I won't find my peer now either.
1200          *
1201          * In that case, and _only_ in that case,
1202          * we use the degr-wfc-timeout instead of the default,
1203          * so we can automatically recover from a crash of a
1204          * degraded but active "cluster" after a certain timeout.
1205          */
1206         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1207         if (mdev->state.role != R_PRIMARY &&
1208              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1209             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1210                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1211
1212         dd = drbd_determine_dev_size(mdev, 0);
1213         if (dd == dev_size_error) {
1214                 retcode = ERR_NOMEM_BITMAP;
1215                 goto force_diskless_dec;
1216         } else if (dd == grew)
1217                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1218
1219         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1220                 dev_info(DEV, "Assuming that all blocks are out of sync "
1221                      "(aka FullSync)\n");
1222                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1223                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1224                         retcode = ERR_IO_MD_DISK;
1225                         goto force_diskless_dec;
1226                 }
1227         } else {
1228                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1229                         "read from attaching", BM_LOCKED_MASK) < 0) {
1230                         retcode = ERR_IO_MD_DISK;
1231                         goto force_diskless_dec;
1232                 }
1233         }
1234
1235         if (cp_discovered) {
1236                 drbd_al_apply_to_bm(mdev);
1237                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1238                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1239                         retcode = ERR_IO_MD_DISK;
1240                         goto force_diskless_dec;
1241                 }
1242         }
1243
1244         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1245                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1246
1247         spin_lock_irq(&mdev->req_lock);
1248         os = mdev->state;
1249         ns.i = os.i;
1250         /* If MDF_CONSISTENT is not set go into inconsistent state,
1251            otherwise investigate MDF_WasUpToDate...
1252            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1253            otherwise into D_CONSISTENT state.
1254         */
1255         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1256                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1257                         ns.disk = D_CONSISTENT;
1258                 else
1259                         ns.disk = D_OUTDATED;
1260         } else {
1261                 ns.disk = D_INCONSISTENT;
1262         }
1263
1264         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1265                 ns.pdsk = D_OUTDATED;
1266
1267         if ( ns.disk == D_CONSISTENT &&
1268             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1269                 ns.disk = D_UP_TO_DATE;
1270
1271         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1272            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1273            this point, because drbd_request_state() modifies these
1274            flags. */
1275
1276         /* In case we are C_CONNECTED postpone any decision on the new disk
1277            state after the negotiation phase. */
1278         if (mdev->state.conn == C_CONNECTED) {
1279                 mdev->new_state_tmp.i = ns.i;
1280                 ns.i = os.i;
1281                 ns.disk = D_NEGOTIATING;
1282
1283                 /* We expect to receive up-to-date UUIDs soon.
1284                    To avoid a race in receive_state, free p_uuid while
1285                    holding req_lock. I.e. atomic with the state change */
1286                 kfree(mdev->p_uuid);
1287                 mdev->p_uuid = NULL;
1288         }
1289
1290         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1291         ns = mdev->state;
1292         spin_unlock_irq(&mdev->req_lock);
1293
1294         if (rv < SS_SUCCESS)
1295                 goto force_diskless_dec;
1296
1297         if (mdev->state.role == R_PRIMARY)
1298                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1299         else
1300                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1301
1302         drbd_md_mark_dirty(mdev);
1303         drbd_md_sync(mdev);
1304
1305         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1306         put_ldev(mdev);
1307         reply->ret_code = retcode;
1308         drbd_reconfig_done(mdev);
1309         return 0;
1310
1311  force_diskless_dec:
1312         put_ldev(mdev);
1313  force_diskless:
1314         drbd_force_state(mdev, NS(disk, D_FAILED));
1315         drbd_md_sync(mdev);
1316  fail:
1317         if (nbc) {
1318                 if (nbc->backing_bdev)
1319                         blkdev_put(nbc->backing_bdev,
1320                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1321                 if (nbc->md_bdev)
1322                         blkdev_put(nbc->md_bdev,
1323                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1324                 kfree(nbc);
1325         }
1326         lc_destroy(resync_lru);
1327
1328         reply->ret_code = retcode;
1329         drbd_reconfig_done(mdev);
1330         return 0;
1331 }
1332
1333 /* Detaching the disk is a process in multiple stages.  First we need to lock
1334  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1335  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1336  * internal references as well.
1337  * Only then we have finally detached. */
1338 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1339                           struct drbd_nl_cfg_reply *reply)
1340 {
1341         enum drbd_ret_code retcode;
1342         int ret;
1343         struct detach dt = {};
1344
1345         if (!detach_from_tags(mdev, nlp->tag_list, &dt)) {
1346                 reply->ret_code = ERR_MANDATORY_TAG;
1347                 goto out;
1348         }
1349
1350         if (dt.detach_force) {
1351                 set_bit(FORCE_DETACH, &mdev->flags);
1352                 drbd_force_state(mdev, NS(disk, D_FAILED));
1353                 reply->ret_code = SS_SUCCESS;
1354                 goto out;
1355         }
1356
1357         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1358         drbd_md_get_buffer(mdev); /* make sure there is no in-flight meta-data IO */
1359         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1360         drbd_md_put_buffer(mdev);
1361         /* D_FAILED will transition to DISKLESS. */
1362         ret = wait_event_interruptible(mdev->misc_wait,
1363                         mdev->state.disk != D_FAILED);
1364         drbd_resume_io(mdev);
1365
1366         if ((int)retcode == (int)SS_IS_DISKLESS)
1367                 retcode = SS_NOTHING_TO_DO;
1368         if (ret)
1369                 retcode = ERR_INTR;
1370         reply->ret_code = retcode;
1371 out:
1372         return 0;
1373 }
1374
1375 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1376                             struct drbd_nl_cfg_reply *reply)
1377 {
1378         int i, ns;
1379         enum drbd_ret_code retcode;
1380         struct net_conf *new_conf = NULL;
1381         struct crypto_hash *tfm = NULL;
1382         struct crypto_hash *integrity_w_tfm = NULL;
1383         struct crypto_hash *integrity_r_tfm = NULL;
1384         struct hlist_head *new_tl_hash = NULL;
1385         struct hlist_head *new_ee_hash = NULL;
1386         struct drbd_conf *odev;
1387         char hmac_name[CRYPTO_MAX_ALG_NAME];
1388         void *int_dig_out = NULL;
1389         void *int_dig_in = NULL;
1390         void *int_dig_vv = NULL;
1391         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1392
1393         drbd_reconfig_start(mdev);
1394
1395         if (mdev->state.conn > C_STANDALONE) {
1396                 retcode = ERR_NET_CONFIGURED;
1397                 goto fail;
1398         }
1399
1400         /* allocation not in the IO path, cqueue thread context */
1401         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1402         if (!new_conf) {
1403                 retcode = ERR_NOMEM;
1404                 goto fail;
1405         }
1406
1407         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1408         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1409         new_conf->ping_int         = DRBD_PING_INT_DEF;
1410         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1411         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1412         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1413         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1414         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1415         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1416         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1417         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1418         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1419         new_conf->want_lose        = 0;
1420         new_conf->two_primaries    = 0;
1421         new_conf->wire_protocol    = DRBD_PROT_C;
1422         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1423         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1424         new_conf->on_congestion    = DRBD_ON_CONGESTION_DEF;
1425         new_conf->cong_extents     = DRBD_CONG_EXTENTS_DEF;
1426
1427         if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1428                 retcode = ERR_MANDATORY_TAG;
1429                 goto fail;
1430         }
1431
1432         if (new_conf->two_primaries
1433             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1434                 retcode = ERR_NOT_PROTO_C;
1435                 goto fail;
1436         }
1437
1438         if (get_ldev(mdev)) {
1439                 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1440                 put_ldev(mdev);
1441                 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1442                         retcode = ERR_STONITH_AND_PROT_A;
1443                         goto fail;
1444                 }
1445         }
1446
1447         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1448                 retcode = ERR_CONG_NOT_PROTO_A;
1449                 goto fail;
1450         }
1451
1452         if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1453                 retcode = ERR_DISCARD;
1454                 goto fail;
1455         }
1456
1457         retcode = NO_ERROR;
1458
1459         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1460         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1461         for (i = 0; i < minor_count; i++) {
1462                 odev = minor_to_mdev(i);
1463                 if (!odev || odev == mdev)
1464                         continue;
1465                 if (get_net_conf(odev)) {
1466                         taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1467                         if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1468                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1469                                 retcode = ERR_LOCAL_ADDR;
1470
1471                         taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1472                         if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1473                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1474                                 retcode = ERR_PEER_ADDR;
1475
1476                         put_net_conf(odev);
1477                         if (retcode != NO_ERROR)
1478                                 goto fail;
1479                 }
1480         }
1481
1482         if (new_conf->cram_hmac_alg[0] != 0) {
1483                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1484                         new_conf->cram_hmac_alg);
1485                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1486                 if (IS_ERR(tfm)) {
1487                         tfm = NULL;
1488                         retcode = ERR_AUTH_ALG;
1489                         goto fail;
1490                 }
1491
1492                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1493                         retcode = ERR_AUTH_ALG_ND;
1494                         goto fail;
1495                 }
1496         }
1497
1498         if (new_conf->integrity_alg[0]) {
1499                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1500                 if (IS_ERR(integrity_w_tfm)) {
1501                         integrity_w_tfm = NULL;
1502                         retcode=ERR_INTEGRITY_ALG;
1503                         goto fail;
1504                 }
1505
1506                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1507                         retcode=ERR_INTEGRITY_ALG_ND;
1508                         goto fail;
1509                 }
1510
1511                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1512                 if (IS_ERR(integrity_r_tfm)) {
1513                         integrity_r_tfm = NULL;
1514                         retcode=ERR_INTEGRITY_ALG;
1515                         goto fail;
1516                 }
1517         }
1518
1519         ns = new_conf->max_epoch_size/8;
1520         if (mdev->tl_hash_s != ns) {
1521                 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1522                 if (!new_tl_hash) {
1523                         retcode = ERR_NOMEM;
1524                         goto fail;
1525                 }
1526         }
1527
1528         ns = new_conf->max_buffers/8;
1529         if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1530                 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1531                 if (!new_ee_hash) {
1532                         retcode = ERR_NOMEM;
1533                         goto fail;
1534                 }
1535         }
1536
1537         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1538
1539         if (integrity_w_tfm) {
1540                 i = crypto_hash_digestsize(integrity_w_tfm);
1541                 int_dig_out = kmalloc(i, GFP_KERNEL);
1542                 if (!int_dig_out) {
1543                         retcode = ERR_NOMEM;
1544                         goto fail;
1545                 }
1546                 int_dig_in = kmalloc(i, GFP_KERNEL);
1547                 if (!int_dig_in) {
1548                         retcode = ERR_NOMEM;
1549                         goto fail;
1550                 }
1551                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1552                 if (!int_dig_vv) {
1553                         retcode = ERR_NOMEM;
1554                         goto fail;
1555                 }
1556         }
1557
1558         if (!mdev->bitmap) {
1559                 if(drbd_bm_init(mdev)) {
1560                         retcode = ERR_NOMEM;
1561                         goto fail;
1562                 }
1563         }
1564
1565         drbd_flush_workqueue(mdev);
1566         spin_lock_irq(&mdev->req_lock);
1567         if (mdev->net_conf != NULL) {
1568                 retcode = ERR_NET_CONFIGURED;
1569                 spin_unlock_irq(&mdev->req_lock);
1570                 goto fail;
1571         }
1572         mdev->net_conf = new_conf;
1573
1574         mdev->send_cnt = 0;
1575         mdev->recv_cnt = 0;
1576
1577         if (new_tl_hash) {
1578                 kfree(mdev->tl_hash);
1579                 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1580                 mdev->tl_hash = new_tl_hash;
1581         }
1582
1583         if (new_ee_hash) {
1584                 kfree(mdev->ee_hash);
1585                 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1586                 mdev->ee_hash = new_ee_hash;
1587         }
1588
1589         crypto_free_hash(mdev->cram_hmac_tfm);
1590         mdev->cram_hmac_tfm = tfm;
1591
1592         crypto_free_hash(mdev->integrity_w_tfm);
1593         mdev->integrity_w_tfm = integrity_w_tfm;
1594
1595         crypto_free_hash(mdev->integrity_r_tfm);
1596         mdev->integrity_r_tfm = integrity_r_tfm;
1597
1598         kfree(mdev->int_dig_out);
1599         kfree(mdev->int_dig_in);
1600         kfree(mdev->int_dig_vv);
1601         mdev->int_dig_out=int_dig_out;
1602         mdev->int_dig_in=int_dig_in;
1603         mdev->int_dig_vv=int_dig_vv;
1604         retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL);
1605         spin_unlock_irq(&mdev->req_lock);
1606
1607         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1608         reply->ret_code = retcode;
1609         drbd_reconfig_done(mdev);
1610         return 0;
1611
1612 fail:
1613         kfree(int_dig_out);
1614         kfree(int_dig_in);
1615         kfree(int_dig_vv);
1616         crypto_free_hash(tfm);
1617         crypto_free_hash(integrity_w_tfm);
1618         crypto_free_hash(integrity_r_tfm);
1619         kfree(new_tl_hash);
1620         kfree(new_ee_hash);
1621         kfree(new_conf);
1622
1623         reply->ret_code = retcode;
1624         drbd_reconfig_done(mdev);
1625         return 0;
1626 }
1627
1628 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1629                               struct drbd_nl_cfg_reply *reply)
1630 {
1631         int retcode;
1632         struct disconnect dc;
1633
1634         memset(&dc, 0, sizeof(struct disconnect));
1635         if (!disconnect_from_tags(mdev, nlp->tag_list, &dc)) {
1636                 retcode = ERR_MANDATORY_TAG;
1637                 goto fail;
1638         }
1639
1640         if (dc.force) {
1641                 spin_lock_irq(&mdev->req_lock);
1642                 if (mdev->state.conn >= C_WF_CONNECTION)
1643                         _drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), CS_HARD, NULL);
1644                 spin_unlock_irq(&mdev->req_lock);
1645                 goto done;
1646         }
1647
1648         retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1649
1650         if (retcode == SS_NOTHING_TO_DO)
1651                 goto done;
1652         else if (retcode == SS_ALREADY_STANDALONE)
1653                 goto done;
1654         else if (retcode == SS_PRIMARY_NOP) {
1655                 /* Our statche checking code wants to see the peer outdated. */
1656                 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1657                                                       pdsk, D_OUTDATED));
1658         } else if (retcode == SS_CW_FAILED_BY_PEER) {
1659                 /* The peer probably wants to see us outdated. */
1660                 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1661                                                         disk, D_OUTDATED),
1662                                               CS_ORDERED);
1663                 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1664                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1665                         retcode = SS_SUCCESS;
1666                 }
1667         }
1668
1669         if (retcode < SS_SUCCESS)
1670                 goto fail;
1671
1672         if (wait_event_interruptible(mdev->state_wait,
1673                                      mdev->state.conn != C_DISCONNECTING)) {
1674                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1675                    someone else might connect us in the mean time! */
1676                 retcode = ERR_INTR;
1677                 goto fail;
1678         }
1679
1680  done:
1681         retcode = NO_ERROR;
1682  fail:
1683         drbd_md_sync(mdev);
1684         reply->ret_code = retcode;
1685         return 0;
1686 }
1687
1688 void resync_after_online_grow(struct drbd_conf *mdev)
1689 {
1690         int iass; /* I am sync source */
1691
1692         dev_info(DEV, "Resync of new storage after online grow\n");
1693         if (mdev->state.role != mdev->state.peer)
1694                 iass = (mdev->state.role == R_PRIMARY);
1695         else
1696                 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1697
1698         if (iass)
1699                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1700         else
1701                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1702 }
1703
1704 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1705                           struct drbd_nl_cfg_reply *reply)
1706 {
1707         struct resize rs;
1708         int retcode = NO_ERROR;
1709         enum determine_dev_size dd;
1710         enum dds_flags ddsf;
1711
1712         memset(&rs, 0, sizeof(struct resize));
1713         if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1714                 retcode = ERR_MANDATORY_TAG;
1715                 goto fail;
1716         }
1717
1718         if (mdev->state.conn > C_CONNECTED) {
1719                 retcode = ERR_RESIZE_RESYNC;
1720                 goto fail;
1721         }
1722
1723         if (mdev->state.role == R_SECONDARY &&
1724             mdev->state.peer == R_SECONDARY) {
1725                 retcode = ERR_NO_PRIMARY;
1726                 goto fail;
1727         }
1728
1729         if (!get_ldev(mdev)) {
1730                 retcode = ERR_NO_DISK;
1731                 goto fail;
1732         }
1733
1734         if (rs.no_resync && mdev->agreed_pro_version < 93) {
1735                 retcode = ERR_NEED_APV_93;
1736                 goto fail_ldev;
1737         }
1738
1739         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1740                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1741
1742         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1743         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1744         dd = drbd_determine_dev_size(mdev, ddsf);
1745         drbd_md_sync(mdev);
1746         put_ldev(mdev);
1747         if (dd == dev_size_error) {
1748                 retcode = ERR_NOMEM_BITMAP;
1749                 goto fail;
1750         }
1751
1752         if (mdev->state.conn == C_CONNECTED) {
1753                 if (dd == grew)
1754                         set_bit(RESIZE_PENDING, &mdev->flags);
1755
1756                 drbd_send_uuids(mdev);
1757                 drbd_send_sizes(mdev, 1, ddsf);
1758         }
1759
1760  fail:
1761         reply->ret_code = retcode;
1762         return 0;
1763
1764  fail_ldev:
1765         put_ldev(mdev);
1766         goto fail;
1767 }
1768
1769 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1770                                struct drbd_nl_cfg_reply *reply)
1771 {
1772         int retcode = NO_ERROR;
1773         int err;
1774         int ovr; /* online verify running */
1775         int rsr; /* re-sync running */
1776         struct crypto_hash *verify_tfm = NULL;
1777         struct crypto_hash *csums_tfm = NULL;
1778         struct syncer_conf sc;
1779         cpumask_var_t new_cpu_mask;
1780         int *rs_plan_s = NULL;
1781         int fifo_size;
1782
1783         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1784                 retcode = ERR_NOMEM;
1785                 goto fail;
1786         }
1787
1788         if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1789                 memset(&sc, 0, sizeof(struct syncer_conf));
1790                 sc.rate       = DRBD_RATE_DEF;
1791                 sc.after      = DRBD_AFTER_DEF;
1792                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1793                 sc.on_no_data  = DRBD_ON_NO_DATA_DEF;
1794                 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1795                 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1796                 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1797                 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1798                 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1799         } else
1800                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1801
1802         if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1803                 retcode = ERR_MANDATORY_TAG;
1804                 goto fail;
1805         }
1806
1807         /* re-sync running */
1808         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1809                 mdev->state.conn == C_SYNC_TARGET ||
1810                 mdev->state.conn == C_PAUSED_SYNC_S ||
1811                 mdev->state.conn == C_PAUSED_SYNC_T );
1812
1813         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1814                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1815                 goto fail;
1816         }
1817
1818         if (!rsr && sc.csums_alg[0]) {
1819                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1820                 if (IS_ERR(csums_tfm)) {
1821                         csums_tfm = NULL;
1822                         retcode = ERR_CSUMS_ALG;
1823                         goto fail;
1824                 }
1825
1826                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1827                         retcode = ERR_CSUMS_ALG_ND;
1828                         goto fail;
1829                 }
1830         }
1831
1832         /* online verify running */
1833         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1834
1835         if (ovr) {
1836                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1837                         retcode = ERR_VERIFY_RUNNING;
1838                         goto fail;
1839                 }
1840         }
1841
1842         if (!ovr && sc.verify_alg[0]) {
1843                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1844                 if (IS_ERR(verify_tfm)) {
1845                         verify_tfm = NULL;
1846                         retcode = ERR_VERIFY_ALG;
1847                         goto fail;
1848                 }
1849
1850                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1851                         retcode = ERR_VERIFY_ALG_ND;
1852                         goto fail;
1853                 }
1854         }
1855
1856         /* silently ignore cpu mask on UP kernel */
1857         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1858                 err = bitmap_parse(sc.cpu_mask, 32,
1859                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1860                 if (err) {
1861                         dev_warn(DEV, "bitmap_parse() failed with %d\n", err);
1862                         retcode = ERR_CPU_MASK_PARSE;
1863                         goto fail;
1864                 }
1865         }
1866
1867         ERR_IF (sc.rate < 1) sc.rate = 1;
1868         ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1869 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1870         if (sc.al_extents > AL_MAX) {
1871                 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1872                 sc.al_extents = AL_MAX;
1873         }
1874 #undef AL_MAX
1875
1876         /* to avoid spurious errors when configuring minors before configuring
1877          * the minors they depend on: if necessary, first create the minor we
1878          * depend on */
1879         if (sc.after >= 0)
1880                 ensure_mdev(sc.after, 1);
1881
1882         /* most sanity checks done, try to assign the new sync-after
1883          * dependency.  need to hold the global lock in there,
1884          * to avoid a race in the dependency loop check. */
1885         retcode = drbd_alter_sa(mdev, sc.after);
1886         if (retcode != NO_ERROR)
1887                 goto fail;
1888
1889         fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1890         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1891                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1892                 if (!rs_plan_s) {
1893                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1894                         retcode = ERR_NOMEM;
1895                         goto fail;
1896                 }
1897         }
1898
1899         /* ok, assign the rest of it as well.
1900          * lock against receive_SyncParam() */
1901         spin_lock(&mdev->peer_seq_lock);
1902         mdev->sync_conf = sc;
1903
1904         if (!rsr) {
1905                 crypto_free_hash(mdev->csums_tfm);
1906                 mdev->csums_tfm = csums_tfm;
1907                 csums_tfm = NULL;
1908         }
1909
1910         if (!ovr) {
1911                 crypto_free_hash(mdev->verify_tfm);
1912                 mdev->verify_tfm = verify_tfm;
1913                 verify_tfm = NULL;
1914         }
1915
1916         if (fifo_size != mdev->rs_plan_s.size) {
1917                 kfree(mdev->rs_plan_s.values);
1918                 mdev->rs_plan_s.values = rs_plan_s;
1919                 mdev->rs_plan_s.size   = fifo_size;
1920                 mdev->rs_planed = 0;
1921                 rs_plan_s = NULL;
1922         }
1923
1924         spin_unlock(&mdev->peer_seq_lock);
1925
1926         if (get_ldev(mdev)) {
1927                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1928                 drbd_al_shrink(mdev);
1929                 err = drbd_check_al_size(mdev);
1930                 lc_unlock(mdev->act_log);
1931                 wake_up(&mdev->al_wait);
1932
1933                 put_ldev(mdev);
1934                 drbd_md_sync(mdev);
1935
1936                 if (err) {
1937                         retcode = ERR_NOMEM;
1938                         goto fail;
1939                 }
1940         }
1941
1942         if (mdev->state.conn >= C_CONNECTED)
1943                 drbd_send_sync_param(mdev, &sc);
1944
1945         if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1946                 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1947                 drbd_calc_cpu_mask(mdev);
1948                 mdev->receiver.reset_cpu_mask = 1;
1949                 mdev->asender.reset_cpu_mask = 1;
1950                 mdev->worker.reset_cpu_mask = 1;
1951         }
1952
1953         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1954 fail:
1955         kfree(rs_plan_s);
1956         free_cpumask_var(new_cpu_mask);
1957         crypto_free_hash(csums_tfm);
1958         crypto_free_hash(verify_tfm);
1959         reply->ret_code = retcode;
1960         return 0;
1961 }
1962
1963 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1964                               struct drbd_nl_cfg_reply *reply)
1965 {
1966         int retcode;
1967
1968         /* If there is still bitmap IO pending, probably because of a previous
1969          * resync just being finished, wait for it before requesting a new resync. */
1970         drbd_suspend_io(mdev);
1971         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1972
1973         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1974
1975         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1976                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1977
1978         while (retcode == SS_NEED_CONNECTION) {
1979                 spin_lock_irq(&mdev->req_lock);
1980                 if (mdev->state.conn < C_CONNECTED)
1981                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1982                 spin_unlock_irq(&mdev->req_lock);
1983
1984                 if (retcode != SS_NEED_CONNECTION)
1985                         break;
1986
1987                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1988         }
1989         drbd_resume_io(mdev);
1990
1991         reply->ret_code = retcode;
1992         return 0;
1993 }
1994
1995 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
1996 {
1997         int rv;
1998
1999         rv = drbd_bmio_set_n_write(mdev);
2000         drbd_suspend_al(mdev);
2001         return rv;
2002 }
2003
2004 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2005                                    struct drbd_nl_cfg_reply *reply)
2006 {
2007         int retcode;
2008
2009         /* If there is still bitmap IO pending, probably because of a previous
2010          * resync just being finished, wait for it before requesting a new resync. */
2011         drbd_suspend_io(mdev);
2012         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2013
2014         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
2015
2016         if (retcode < SS_SUCCESS) {
2017                 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
2018                         /* The peer will get a resync upon connect anyways. Just make that
2019                            into a full resync. */
2020                         retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
2021                         if (retcode >= SS_SUCCESS) {
2022                                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
2023                                         "set_n_write from invalidate_peer",
2024                                         BM_LOCKED_SET_ALLOWED))
2025                                         retcode = ERR_IO_MD_DISK;
2026                         }
2027                 } else
2028                         retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
2029         }
2030         drbd_resume_io(mdev);
2031
2032         reply->ret_code = retcode;
2033         return 0;
2034 }
2035
2036 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2037                               struct drbd_nl_cfg_reply *reply)
2038 {
2039         int retcode = NO_ERROR;
2040
2041         if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2042                 retcode = ERR_PAUSE_IS_SET;
2043
2044         reply->ret_code = retcode;
2045         return 0;
2046 }
2047
2048 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2049                                struct drbd_nl_cfg_reply *reply)
2050 {
2051         int retcode = NO_ERROR;
2052         union drbd_state s;
2053
2054         if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2055                 s = mdev->state;
2056                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2057                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2058                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2059                 } else {
2060                         retcode = ERR_PAUSE_IS_CLEAR;
2061                 }
2062         }
2063
2064         reply->ret_code = retcode;
2065         return 0;
2066 }
2067
2068 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2069                               struct drbd_nl_cfg_reply *reply)
2070 {
2071         reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
2072
2073         return 0;
2074 }
2075
2076 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2077                              struct drbd_nl_cfg_reply *reply)
2078 {
2079         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2080                 drbd_uuid_new_current(mdev);
2081                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2082         }
2083         drbd_suspend_io(mdev);
2084         reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2085         if (reply->ret_code == SS_SUCCESS) {
2086                 if (mdev->state.conn < C_CONNECTED)
2087                         tl_clear(mdev);
2088                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2089                         tl_restart(mdev, fail_frozen_disk_io);
2090         }
2091         drbd_resume_io(mdev);
2092
2093         return 0;
2094 }
2095
2096 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2097                            struct drbd_nl_cfg_reply *reply)
2098 {
2099         reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
2100         return 0;
2101 }
2102
2103 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2104                            struct drbd_nl_cfg_reply *reply)
2105 {
2106         unsigned short *tl;
2107
2108         tl = reply->tag_list;
2109
2110         if (get_ldev(mdev)) {
2111                 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
2112                 put_ldev(mdev);
2113         }
2114
2115         if (get_net_conf(mdev)) {
2116                 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
2117                 put_net_conf(mdev);
2118         }
2119         tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
2120
2121         put_unaligned(TT_END, tl++); /* Close the tag list */
2122
2123         return (int)((char *)tl - (char *)reply->tag_list);
2124 }
2125
2126 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2127                              struct drbd_nl_cfg_reply *reply)
2128 {
2129         unsigned short *tl = reply->tag_list;
2130         union drbd_state s = mdev->state;
2131         unsigned long rs_left;
2132         unsigned int res;
2133
2134         tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
2135
2136         /* no local ref, no bitmap, no syncer progress. */
2137         if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
2138                 if (get_ldev(mdev)) {
2139                         drbd_get_syncer_progress(mdev, &rs_left, &res);
2140                         tl = tl_add_int(tl, T_sync_progress, &res);
2141                         put_ldev(mdev);
2142                 }
2143         }
2144         put_unaligned(TT_END, tl++); /* Close the tag list */
2145
2146         return (int)((char *)tl - (char *)reply->tag_list);
2147 }
2148
2149 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2150                              struct drbd_nl_cfg_reply *reply)
2151 {
2152         unsigned short *tl;
2153
2154         tl = reply->tag_list;
2155
2156         if (get_ldev(mdev)) {
2157                 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
2158                 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
2159                 put_ldev(mdev);
2160         }
2161         put_unaligned(TT_END, tl++); /* Close the tag list */
2162
2163         return (int)((char *)tl - (char *)reply->tag_list);
2164 }
2165
2166 /**
2167  * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
2168  * @mdev:       DRBD device.
2169  * @nlp:        Netlink/connector packet from drbdsetup
2170  * @reply:      Reply packet for drbdsetup
2171  */
2172 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2173                                     struct drbd_nl_cfg_reply *reply)
2174 {
2175         unsigned short *tl;
2176         char rv;
2177
2178         tl = reply->tag_list;
2179
2180         rv = mdev->state.pdsk == D_OUTDATED        ? UT_PEER_OUTDATED :
2181           test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
2182
2183         tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
2184         put_unaligned(TT_END, tl++); /* Close the tag list */
2185
2186         return (int)((char *)tl - (char *)reply->tag_list);
2187 }
2188
2189 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2190                                     struct drbd_nl_cfg_reply *reply)
2191 {
2192         /* default to resume from last known position, if possible */
2193         struct start_ov args =
2194                 { .start_sector = mdev->ov_start_sector };
2195
2196         if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2197                 reply->ret_code = ERR_MANDATORY_TAG;
2198                 return 0;
2199         }
2200
2201         /* If there is still bitmap IO pending, e.g. previous resync or verify
2202          * just being finished, wait for it before requesting a new resync. */
2203         drbd_suspend_io(mdev);
2204         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2205
2206         /* w_make_ov_request expects position to be aligned */
2207         mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
2208         reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2209         drbd_resume_io(mdev);
2210         return 0;
2211 }
2212
2213
2214 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2215                               struct drbd_nl_cfg_reply *reply)
2216 {
2217         int retcode = NO_ERROR;
2218         int skip_initial_sync = 0;
2219         int err;
2220
2221         struct new_c_uuid args;
2222
2223         memset(&args, 0, sizeof(struct new_c_uuid));
2224         if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2225                 reply->ret_code = ERR_MANDATORY_TAG;
2226                 return 0;
2227         }
2228
2229         mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
2230
2231         if (!get_ldev(mdev)) {
2232                 retcode = ERR_NO_DISK;
2233                 goto out;
2234         }
2235
2236         /* this is "skip initial sync", assume to be clean */
2237         if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
2238             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2239                 dev_info(DEV, "Preparing to skip initial sync\n");
2240                 skip_initial_sync = 1;
2241         } else if (mdev->state.conn != C_STANDALONE) {
2242                 retcode = ERR_CONNECTED;
2243                 goto out_dec;
2244         }
2245
2246         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2247         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2248
2249         if (args.clear_bm) {
2250                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2251                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2252                 if (err) {
2253                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2254                         retcode = ERR_IO_MD_DISK;
2255                 }
2256                 if (skip_initial_sync) {
2257                         drbd_send_uuids_skip_initial_sync(mdev);
2258                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2259                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2260                         spin_lock_irq(&mdev->req_lock);
2261                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2262                                         CS_VERBOSE, NULL);
2263                         spin_unlock_irq(&mdev->req_lock);
2264                 }
2265         }
2266
2267         drbd_md_sync(mdev);
2268 out_dec:
2269         put_ldev(mdev);
2270 out:
2271         mutex_unlock(&mdev->state_mutex);
2272
2273         reply->ret_code = retcode;
2274         return 0;
2275 }
2276
2277 struct cn_handler_struct {
2278         int (*function)(struct drbd_conf *,
2279                          struct drbd_nl_cfg_req *,
2280                          struct drbd_nl_cfg_reply *);
2281         int reply_body_size;
2282 };
2283
2284 static struct cn_handler_struct cnd_table[] = {
2285         [ P_primary ]           = { &drbd_nl_primary,           0 },
2286         [ P_secondary ]         = { &drbd_nl_secondary,         0 },
2287         [ P_disk_conf ]         = { &drbd_nl_disk_conf,         0 },
2288         [ P_detach ]            = { &drbd_nl_detach,            0 },
2289         [ P_net_conf ]          = { &drbd_nl_net_conf,          0 },
2290         [ P_disconnect ]        = { &drbd_nl_disconnect,        0 },
2291         [ P_resize ]            = { &drbd_nl_resize,            0 },
2292         [ P_syncer_conf ]       = { &drbd_nl_syncer_conf,       0 },
2293         [ P_invalidate ]        = { &drbd_nl_invalidate,        0 },
2294         [ P_invalidate_peer ]   = { &drbd_nl_invalidate_peer,   0 },
2295         [ P_pause_sync ]        = { &drbd_nl_pause_sync,        0 },
2296         [ P_resume_sync ]       = { &drbd_nl_resume_sync,       0 },
2297         [ P_suspend_io ]        = { &drbd_nl_suspend_io,        0 },
2298         [ P_resume_io ]         = { &drbd_nl_resume_io,         0 },
2299         [ P_outdate ]           = { &drbd_nl_outdate,           0 },
2300         [ P_get_config ]        = { &drbd_nl_get_config,
2301                                     sizeof(struct syncer_conf_tag_len_struct) +
2302                                     sizeof(struct disk_conf_tag_len_struct) +
2303                                     sizeof(struct net_conf_tag_len_struct) },
2304         [ P_get_state ]         = { &drbd_nl_get_state,
2305                                     sizeof(struct get_state_tag_len_struct) +
2306                                     sizeof(struct sync_progress_tag_len_struct) },
2307         [ P_get_uuids ]         = { &drbd_nl_get_uuids,
2308                                     sizeof(struct get_uuids_tag_len_struct) },
2309         [ P_get_timeout_flag ]  = { &drbd_nl_get_timeout_flag,
2310                                     sizeof(struct get_timeout_flag_tag_len_struct)},
2311         [ P_start_ov ]          = { &drbd_nl_start_ov,          0 },
2312         [ P_new_c_uuid ]        = { &drbd_nl_new_c_uuid,        0 },
2313 };
2314
2315 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2316 {
2317         struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2318         struct cn_handler_struct *cm;
2319         struct cn_msg *cn_reply;
2320         struct drbd_nl_cfg_reply *reply;
2321         struct drbd_conf *mdev;
2322         int retcode, rr;
2323         int reply_size = sizeof(struct cn_msg)
2324                 + sizeof(struct drbd_nl_cfg_reply)
2325                 + sizeof(short int);
2326
2327         if (!try_module_get(THIS_MODULE)) {
2328                 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2329                 return;
2330         }
2331
2332         if (!capable(CAP_SYS_ADMIN)) {
2333                 retcode = ERR_PERM;
2334                 goto fail;
2335         }
2336
2337         mdev = ensure_mdev(nlp->drbd_minor,
2338                         (nlp->flags & DRBD_NL_CREATE_DEVICE));
2339         if (!mdev) {
2340                 retcode = ERR_MINOR_INVALID;
2341                 goto fail;
2342         }
2343
2344         if (nlp->packet_type >= P_nl_after_last_packet ||
2345             nlp->packet_type == P_return_code_only) {
2346                 retcode = ERR_PACKET_NR;
2347                 goto fail;
2348         }
2349
2350         cm = cnd_table + nlp->packet_type;
2351
2352         /* This may happen if packet number is 0: */
2353         if (cm->function == NULL) {
2354                 retcode = ERR_PACKET_NR;
2355                 goto fail;
2356         }
2357
2358         reply_size += cm->reply_body_size;
2359
2360         /* allocation not in the IO path, cqueue thread context */
2361         cn_reply = kzalloc(reply_size, GFP_KERNEL);
2362         if (!cn_reply) {
2363                 retcode = ERR_NOMEM;
2364                 goto fail;
2365         }
2366         reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2367
2368         reply->packet_type =
2369                 cm->reply_body_size ? nlp->packet_type : P_return_code_only;
2370         reply->minor = nlp->drbd_minor;
2371         reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2372         /* reply->tag_list; might be modified by cm->function. */
2373
2374         rr = cm->function(mdev, nlp, reply);
2375
2376         cn_reply->id = req->id;
2377         cn_reply->seq = req->seq;
2378         cn_reply->ack = req->ack  + 1;
2379         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2380         cn_reply->flags = 0;
2381
2382         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2383         if (rr && rr != -ESRCH)
2384                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2385
2386         kfree(cn_reply);
2387         module_put(THIS_MODULE);
2388         return;
2389  fail:
2390         drbd_nl_send_reply(req, retcode);
2391         module_put(THIS_MODULE);
2392 }
2393
2394 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2395
2396 static unsigned short *
2397 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2398         unsigned short len, int nul_terminated)
2399 {
2400         unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2401         len = (len < l) ? len :  l;
2402         put_unaligned(tag, tl++);
2403         put_unaligned(len, tl++);
2404         memcpy(tl, data, len);
2405         tl = (unsigned short*)((char*)tl + len);
2406         if (nul_terminated)
2407                 *((char*)tl - 1) = 0;
2408         return tl;
2409 }
2410
2411 static unsigned short *
2412 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2413 {
2414         return __tl_add_blob(tl, tag, data, len, 0);
2415 }
2416
2417 static unsigned short *
2418 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2419 {
2420         return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2421 }
2422
2423 static unsigned short *
2424 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2425 {
2426         put_unaligned(tag, tl++);
2427         switch(tag_type(tag)) {
2428         case TT_INTEGER:
2429                 put_unaligned(sizeof(int), tl++);
2430                 put_unaligned(*(int *)val, (int *)tl);
2431                 tl = (unsigned short*)((char*)tl+sizeof(int));
2432                 break;
2433         case TT_INT64:
2434                 put_unaligned(sizeof(u64), tl++);
2435                 put_unaligned(*(u64 *)val, (u64 *)tl);
2436                 tl = (unsigned short*)((char*)tl+sizeof(u64));
2437                 break;
2438         default:
2439                 /* someone did something stupid. */
2440                 ;
2441         }
2442         return tl;
2443 }
2444
2445 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2446 {
2447         char buffer[sizeof(struct cn_msg)+
2448                     sizeof(struct drbd_nl_cfg_reply)+
2449                     sizeof(struct get_state_tag_len_struct)+
2450                     sizeof(short int)];
2451         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2452         struct drbd_nl_cfg_reply *reply =
2453                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2454         unsigned short *tl = reply->tag_list;
2455
2456         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2457
2458         tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2459
2460         put_unaligned(TT_END, tl++); /* Close the tag list */
2461
2462         cn_reply->id.idx = CN_IDX_DRBD;
2463         cn_reply->id.val = CN_VAL_DRBD;
2464
2465         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2466         cn_reply->ack = 0; /* not used here. */
2467         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2468                 (int)((char *)tl - (char *)reply->tag_list);
2469         cn_reply->flags = 0;
2470
2471         reply->packet_type = P_get_state;
2472         reply->minor = mdev_to_minor(mdev);
2473         reply->ret_code = NO_ERROR;
2474
2475         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2476 }
2477
2478 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2479 {
2480         char buffer[sizeof(struct cn_msg)+
2481                     sizeof(struct drbd_nl_cfg_reply)+
2482                     sizeof(struct call_helper_tag_len_struct)+
2483                     sizeof(short int)];
2484         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2485         struct drbd_nl_cfg_reply *reply =
2486                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2487         unsigned short *tl = reply->tag_list;
2488
2489         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2490
2491         tl = tl_add_str(tl, T_helper, helper_name);
2492         put_unaligned(TT_END, tl++); /* Close the tag list */
2493
2494         cn_reply->id.idx = CN_IDX_DRBD;
2495         cn_reply->id.val = CN_VAL_DRBD;
2496
2497         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2498         cn_reply->ack = 0; /* not used here. */
2499         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2500                 (int)((char *)tl - (char *)reply->tag_list);
2501         cn_reply->flags = 0;
2502
2503         reply->packet_type = P_call_helper;
2504         reply->minor = mdev_to_minor(mdev);
2505         reply->ret_code = NO_ERROR;
2506
2507         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2508 }
2509
2510 void drbd_bcast_ee(struct drbd_conf *mdev,
2511                 const char *reason, const int dgs,
2512                 const char* seen_hash, const char* calc_hash,
2513                 const struct drbd_epoch_entry* e)
2514 {
2515         struct cn_msg *cn_reply;
2516         struct drbd_nl_cfg_reply *reply;
2517         unsigned short *tl;
2518         struct page *page;
2519         unsigned len;
2520
2521         if (!e)
2522                 return;
2523         if (!reason || !reason[0])
2524                 return;
2525
2526         /* apparently we have to memcpy twice, first to prepare the data for the
2527          * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2528          * netlink skb. */
2529         /* receiver thread context, which is not in the writeout path (of this node),
2530          * but may be in the writeout path of the _other_ node.
2531          * GFP_NOIO to avoid potential "distributed deadlock". */
2532         cn_reply = kzalloc(
2533                 sizeof(struct cn_msg)+
2534                 sizeof(struct drbd_nl_cfg_reply)+
2535                 sizeof(struct dump_ee_tag_len_struct)+
2536                 sizeof(short int),
2537                 GFP_NOIO);
2538
2539         if (!cn_reply) {
2540                 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2541                                 (unsigned long long)e->sector, e->size);
2542                 return;
2543         }
2544
2545         reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2546         tl = reply->tag_list;
2547
2548         tl = tl_add_str(tl, T_dump_ee_reason, reason);
2549         tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2550         tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2551         tl = tl_add_int(tl, T_ee_sector, &e->sector);
2552         tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2553
2554         /* dump the first 32k */
2555         len = min_t(unsigned, e->size, 32 << 10);
2556         put_unaligned(T_ee_data, tl++);
2557         put_unaligned(len, tl++);
2558
2559         page = e->pages;
2560         page_chain_for_each(page) {
2561                 void *d = kmap_atomic(page);
2562                 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2563                 memcpy(tl, d, l);
2564                 kunmap_atomic(d);
2565                 tl = (unsigned short*)((char*)tl + l);
2566                 len -= l;
2567                 if (len == 0)
2568                         break;
2569         }
2570         put_unaligned(TT_END, tl++); /* Close the tag list */
2571
2572         cn_reply->id.idx = CN_IDX_DRBD;
2573         cn_reply->id.val = CN_VAL_DRBD;
2574
2575         cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2576         cn_reply->ack = 0; // not used here.
2577         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2578                 (int)((char*)tl - (char*)reply->tag_list);
2579         cn_reply->flags = 0;
2580
2581         reply->packet_type = P_dump_ee;
2582         reply->minor = mdev_to_minor(mdev);
2583         reply->ret_code = NO_ERROR;
2584
2585         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2586         kfree(cn_reply);
2587 }
2588
2589 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2590 {
2591         char buffer[sizeof(struct cn_msg)+
2592                     sizeof(struct drbd_nl_cfg_reply)+
2593                     sizeof(struct sync_progress_tag_len_struct)+
2594                     sizeof(short int)];
2595         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2596         struct drbd_nl_cfg_reply *reply =
2597                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2598         unsigned short *tl = reply->tag_list;
2599         unsigned long rs_left;
2600         unsigned int res;
2601
2602         /* no local ref, no bitmap, no syncer progress, no broadcast. */
2603         if (!get_ldev(mdev))
2604                 return;
2605         drbd_get_syncer_progress(mdev, &rs_left, &res);
2606         put_ldev(mdev);
2607
2608         tl = tl_add_int(tl, T_sync_progress, &res);
2609         put_unaligned(TT_END, tl++); /* Close the tag list */
2610
2611         cn_reply->id.idx = CN_IDX_DRBD;
2612         cn_reply->id.val = CN_VAL_DRBD;
2613
2614         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2615         cn_reply->ack = 0; /* not used here. */
2616         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2617                 (int)((char *)tl - (char *)reply->tag_list);
2618         cn_reply->flags = 0;
2619
2620         reply->packet_type = P_sync_progress;
2621         reply->minor = mdev_to_minor(mdev);
2622         reply->ret_code = NO_ERROR;
2623
2624         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2625 }
2626
2627 int __init drbd_nl_init(void)
2628 {
2629         static struct cb_id cn_id_drbd;
2630         int err, try=10;
2631
2632         cn_id_drbd.val = CN_VAL_DRBD;
2633         do {
2634                 cn_id_drbd.idx = cn_idx;
2635                 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2636                 if (!err)
2637                         break;
2638                 cn_idx = (cn_idx + CN_IDX_STEP);
2639         } while (try--);
2640
2641         if (err) {
2642                 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2643                 return err;
2644         }
2645
2646         return 0;
2647 }
2648
2649 void drbd_nl_cleanup(void)
2650 {
2651         static struct cb_id cn_id_drbd;
2652
2653         cn_id_drbd.idx = cn_idx;
2654         cn_id_drbd.val = CN_VAL_DRBD;
2655
2656         cn_del_callback(&cn_id_drbd);
2657 }
2658
2659 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2660 {
2661         char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2662         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2663         struct drbd_nl_cfg_reply *reply =
2664                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2665         int rr;
2666
2667         memset(buffer, 0, sizeof(buffer));
2668         cn_reply->id = req->id;
2669
2670         cn_reply->seq = req->seq;
2671         cn_reply->ack = req->ack  + 1;
2672         cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2673         cn_reply->flags = 0;
2674
2675         reply->packet_type = P_return_code_only;
2676         reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2677         reply->ret_code = ret_code;
2678
2679         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2680         if (rr && rr != -ESRCH)
2681                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2682 }
2683