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         if (current == mdev->worker.task)
 151                 drbd_set_flag(mdev, CALLBACK_PENDING);
 152
 153         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 154
 155         if (get_net_conf(mdev)) {
 156                 switch (((struct sockaddr *)mdev->net_conf->peer_addr)->sa_family) {
 157                 case AF_INET6:
 158                         afs = "ipv6";
 159                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6",
 160                                  &((struct sockaddr_in6 *)mdev->net_conf->peer_addr)->sin6_addr);
 161                         break;
 162                 case AF_INET:
 163                         afs = "ipv4";
 164                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 165                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
 166                         break;
 167                 default:
 168                         afs = "ssocks";
 169                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 170                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
 171                 }
 172                 snprintf(af, 20, "DRBD_PEER_AF=%s", afs);
 173                 envp[3]=af;
 174                 envp[4]=ad;
 175                 put_net_conf(mdev);
 176         }
 177
 178         /* The helper may take some time.
 179          * write out any unsynced meta data changes now */
 180         drbd_md_sync(mdev);
 181
 182         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 183
 184         drbd_bcast_ev_helper(mdev, cmd);
 185         ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC);
 186         if (ret)
 187                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 188                                 usermode_helper, cmd, mb,
 189                                 (ret >> 8) & 0xff, ret);
 190         else
 191                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 192                                 usermode_helper, cmd, mb,
 193                                 (ret >> 8) & 0xff, ret);
 194
 195         if (current == mdev->worker.task)
 196                 drbd_clear_flag(mdev, CALLBACK_PENDING);
 197
 198         if (ret < 0) /* Ignore any ERRNOs we got. */
 199                 ret = 0;
 200
 201         return ret;
 202 }
 203
 204 enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
 205 {
 206         char *ex_to_string;
 207         int r;
 208         enum drbd_disk_state nps;
 209         enum drbd_fencing_p fp;
 210
 211         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 212
 213         if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 214                 fp = mdev->ldev->dc.fencing;
 215                 put_ldev(mdev);
 216         } else {
 217                 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n");
 218                 nps = mdev->state.pdsk;
 219                 goto out;
 220         }
 221
 222         r = drbd_khelper(mdev, "fence-peer");
 223
 224         switch ((r>>8) & 0xff) {
 225         case 3: /* peer is inconsistent */
 226                 ex_to_string = "peer is inconsistent or worse";
 227                 nps = D_INCONSISTENT;
 228                 break;
 229         case 4: /* peer got outdated, or was already outdated */
 230                 ex_to_string = "peer was fenced";
 231                 nps = D_OUTDATED;
 232                 break;
 233         case 5: /* peer was down */
 234                 if (mdev->state.disk == D_UP_TO_DATE) {
 235                         /* we will(have) create(d) a new UUID anyways... */
 236                         ex_to_string = "peer is unreachable, assumed to be dead";
 237                         nps = D_OUTDATED;
 238                 } else {
 239                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 240                         nps = mdev->state.pdsk;
 241                 }
 242                 break;
 243         case 6: /* Peer is primary, voluntarily outdate myself.
 244                  * This is useful when an unconnected R_SECONDARY is asked to
 245                  * become R_PRIMARY, but finds the other peer being active. */
 246                 ex_to_string = "peer is active";
 247                 dev_warn(DEV, "Peer is primary, outdating myself.\n");
 248                 nps = D_UNKNOWN;
 249                 _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE);
 250                 break;
 251         case 7:
 252                 if (fp != FP_STONITH)
 253                         dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n");
 254                 ex_to_string = "peer was stonithed";
 255                 nps = D_OUTDATED;
 256                 break;
 257         default:
 258                 /* The script is broken ... */
 259                 nps = D_UNKNOWN;
 260                 dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 261                 return nps;
 262         }
 263
 264         dev_info(DEV, "fence-peer helper returned %d (%s)\n",
 265                         (r>>8) & 0xff, ex_to_string);
 266
 267 out:
 268         if (mdev->state.susp_fen && nps >= D_UNKNOWN) {
 269                 /* The handler was not successful... unfreeze here, the
 270                    state engine can not unfreeze... */
 271                 _drbd_request_state(mdev, NS(susp_fen, 0), CS_VERBOSE);
 272         }
 273
 274         return nps;
 275 }
 276
 277 static int _try_outdate_peer_async(void *data)
 278 {
 279         struct drbd_conf *mdev = (struct drbd_conf *)data;
 280         enum drbd_disk_state nps;
 281         union drbd_state ns;
 282
 283         nps = drbd_try_outdate_peer(mdev);
 284
 285         /* Not using
 286            drbd_request_state(mdev, NS(pdsk, nps));
 287            here, because we might were able to re-establish the connection
 288            in the meantime. This can only partially be solved in the state's
 289            engine is_valid_state() and is_valid_state_transition()
 290            functions.
 291
 292            nps can be D_INCONSISTENT, D_OUTDATED or D_UNKNOWN.
 293            pdsk == D_INCONSISTENT while conn >= C_CONNECTED is valid,
 294            therefore we have to have the pre state change check here.
 295         */
 296         spin_lock_irq(&mdev->req_lock);
 297         ns = mdev->state;
 298         if (ns.conn < C_WF_REPORT_PARAMS && !drbd_test_flag(mdev, STATE_SENT)) {
 299                 ns.pdsk = nps;
 300                 _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
 301         }
 302         spin_unlock_irq(&mdev->req_lock);
 303
 304         return 0;
 305 }
 306
 307 void drbd_try_outdate_peer_async(struct drbd_conf *mdev)
 308 {
 309         struct task_struct *opa;
 310
 311         opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev));
 312         if (IS_ERR(opa))
 313                 dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n");
 314 }
 315
 316 enum drbd_state_rv
 317 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 318 {
 319         const int max_tries = 4;
 320         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
 321         int try = 0;
 322         int forced = 0;
 323         union drbd_state mask, val;
 324         enum drbd_disk_state nps;
 325
 326         if (new_role == R_PRIMARY)
 327                 request_ping(mdev); /* Detect a dead peer ASAP */
 328
 329         mutex_lock(&mdev->state_mutex);
 330
 331         mask.i = 0; mask.role = R_MASK;
 332         val.i  = 0; val.role  = new_role;
 333
 334         while (try++ < max_tries) {
 335                 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 336
 337                 /* in case we first succeeded to outdate,
 338                  * but now suddenly could establish a connection */
 339                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 340                         val.pdsk = 0;
 341                         mask.pdsk = 0;
 342                         continue;
 343                 }
 344
 345                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
 346                     (mdev->state.disk < D_UP_TO_DATE &&
 347                      mdev->state.disk >= D_INCONSISTENT)) {
 348                         mask.disk = D_MASK;
 349                         val.disk  = D_UP_TO_DATE;
 350                         forced = 1;
 351                         continue;
 352                 }
 353
 354                 if (rv == SS_NO_UP_TO_DATE_DISK &&
 355                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 356                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 357                         nps = drbd_try_outdate_peer(mdev);
 358
 359                         if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
 360                                 val.disk = D_UP_TO_DATE;
 361                                 mask.disk = D_MASK;
 362                         }
 363
 364                         val.pdsk = nps;
 365                         mask.pdsk = D_MASK;
 366
 367                         continue;
 368                 }
 369
 370                 if (rv == SS_NOTHING_TO_DO)
 371                         goto fail;
 372                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
 373                         nps = drbd_try_outdate_peer(mdev);
 374
 375                         if (force && nps > D_OUTDATED) {
 376                                 dev_warn(DEV, "Forced into split brain situation!\n");
 377                                 nps = D_OUTDATED;
 378                         }
 379
 380                         mask.pdsk = D_MASK;
 381                         val.pdsk  = nps;
 382
 383                         continue;
 384                 }
 385                 if (rv == SS_TWO_PRIMARIES) {
 386                         /* Maybe the peer is detected as dead very soon...
 387                            retry at most once more in this case. */
 388                         schedule_timeout_interruptible((mdev->net_conf->ping_timeo+1)*HZ/10);
 389                         if (try < max_tries)
 390                                 try = max_tries - 1;
 391                         continue;
 392                 }
 393                 if (rv < SS_SUCCESS) {
 394                         rv = _drbd_request_state(mdev, mask, val,
 395                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 396                         if (rv < SS_SUCCESS)
 397                                 goto fail;
 398                 }
 399                 break;
 400         }
 401
 402         if (rv < SS_SUCCESS)
 403                 goto fail;
 404
 405         if (forced)
 406                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 407
 408         /* Wait until nothing is on the fly :) */
 409         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 410
 411         if (new_role == R_SECONDARY) {
 412                 set_disk_ro(mdev->vdisk, true);
 413                 if (get_ldev(mdev)) {
 414                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 415                         put_ldev(mdev);
 416                 }
 417         } else {
 418                 if (get_net_conf(mdev)) {
 419                         mdev->net_conf->want_lose = 0;
 420                         put_net_conf(mdev);
 421                 }
 422                 set_disk_ro(mdev->vdisk, false);
 423                 if (get_ldev(mdev)) {
 424                         if (((mdev->state.conn < C_CONNECTED ||
 425                                mdev->state.pdsk <= D_FAILED)
 426                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 427                                 drbd_uuid_new_current(mdev);
 428
 429                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 430                         put_ldev(mdev);
 431                 }
 432         }
 433
 434         /* writeout of activity log covered areas of the bitmap
 435          * to stable storage done in after state change already */
 436
 437         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 438                 /* if this was forced, we should consider sync */
 439                 if (forced)
 440                         drbd_send_uuids(mdev);
 441                 drbd_send_current_state(mdev);
 442         }
 443
 444         drbd_md_sync(mdev);
 445
 446         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 447  fail:
 448         mutex_unlock(&mdev->state_mutex);
 449         return rv;
 450 }
 451
 452 static struct drbd_conf *ensure_mdev(int minor, int create)
 453 {
 454         struct drbd_conf *mdev;
 455
 456         if (minor >= minor_count)
 457                 return NULL;
 458
 459         mdev = minor_to_mdev(minor);
 460
 461         if (!mdev && create) {
 462                 struct gendisk *disk = NULL;
 463                 mdev = drbd_new_device(minor);
 464
 465                 spin_lock_irq(&drbd_pp_lock);
 466                 if (minor_table[minor] == NULL) {
 467                         minor_table[minor] = mdev;
 468                         disk = mdev->vdisk;
 469                         mdev = NULL;
 470                 } /* else: we lost the race */
 471                 spin_unlock_irq(&drbd_pp_lock);
 472
 473                 if (disk) /* we won the race above */
 474                         /* in case we ever add a drbd_delete_device(),
 475                          * don't forget the del_gendisk! */
 476                         add_disk(disk);
 477                 else /* we lost the race above */
 478                         drbd_free_mdev(mdev);
 479
 480                 mdev = minor_to_mdev(minor);
 481         }
 482
 483         return mdev;
 484 }
 485
 486 static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 487                            struct drbd_nl_cfg_reply *reply)
 488 {
 489         struct primary primary_args;
 490
 491         memset(&primary_args, 0, sizeof(struct primary));
 492         if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) {
 493                 reply->ret_code = ERR_MANDATORY_TAG;
 494                 return 0;
 495         }
 496
 497         reply->ret_code =
 498                 drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force);
 499
 500         return 0;
 501 }
 502
 503 static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 504                              struct drbd_nl_cfg_reply *reply)
 505 {
 506         reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0);
 507
 508         return 0;
 509 }
 510
 511 /* initializes the md.*_offset members, so we are able to find
 512  * the on disk meta data */
 513 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 514                                        struct drbd_backing_dev *bdev)
 515 {
 516         sector_t md_size_sect = 0;
 517         switch (bdev->dc.meta_dev_idx) {
 518         default:
 519                 /* v07 style fixed size indexed meta data */
 520                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 521                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 522                 bdev->md.al_offset = MD_AL_OFFSET;
 523                 bdev->md.bm_offset = MD_BM_OFFSET;
 524                 break;
 525         case DRBD_MD_INDEX_FLEX_EXT:
 526                 /* just occupy the full device; unit: sectors */
 527                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 528                 bdev->md.md_offset = 0;
 529                 bdev->md.al_offset = MD_AL_OFFSET;
 530                 bdev->md.bm_offset = MD_BM_OFFSET;
 531                 break;
 532         case DRBD_MD_INDEX_INTERNAL:
 533         case DRBD_MD_INDEX_FLEX_INT:
 534                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 535                 /* al size is still fixed */
 536                 bdev->md.al_offset = -MD_AL_MAX_SIZE;
 537                 /* we need (slightly less than) ~ this much bitmap sectors: */
 538                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 539                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 540                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 541                 md_size_sect = ALIGN(md_size_sect, 8);
 542
 543                 /* plus the "drbd meta data super block",
 544                  * and the activity log; */
 545                 md_size_sect += MD_BM_OFFSET;
 546
 547                 bdev->md.md_size_sect = md_size_sect;
 548                 /* bitmap offset is adjusted by 'super' block size */
 549                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 550                 break;
 551         }
 552 }
 553
 554 /* input size is expected to be in KB */
 555 char *ppsize(char *buf, unsigned long long size)
 556 {
 557         /* Needs 9 bytes at max including trailing NUL:
 558          * -1ULL ==> "16384 EB" */
 559         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 560         int base = 0;
 561         while (size >= 10000 && base < sizeof(units)-1) {
 562                 /* shift + round */
 563                 size = (size >> 10) + !!(size & (1<<9));
 564                 base++;
 565         }
 566         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
 567
 568         return buf;
 569 }
 570
 571 /* there is still a theoretical deadlock when called from receiver
 572  * on an D_INCONSISTENT R_PRIMARY:
 573  *  remote READ does inc_ap_bio, receiver would need to receive answer
 574  *  packet from remote to dec_ap_bio again.
 575  *  receiver receive_sizes(), comes here,
 576  *  waits for ap_bio_cnt == 0. -> deadlock.
 577  * but this cannot happen, actually, because:
 578  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 579  *  (not connected, or bad/no disk on peer):
 580  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 581  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 582  *  peer may not initiate a resize.
 583  */
 584 void drbd_suspend_io(struct drbd_conf *mdev)
 585 {
 586         drbd_set_flag(mdev, SUSPEND_IO);
 587         if (is_susp(mdev->state))
 588                 return;
 589         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 590 }
 591
 592 void drbd_resume_io(struct drbd_conf *mdev)
 593 {
 594         drbd_clear_flag(mdev, SUSPEND_IO);
 595         wake_up(&mdev->misc_wait);
 596 }
 597
 598 /**
 599  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 600  * @mdev:       DRBD device.
 601  *
 602  * Returns 0 on success, negative return values indicate errors.
 603  * You should call drbd_md_sync() after calling this function.
 604  */
 605 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 606 {
 607         sector_t prev_first_sect, prev_size; /* previous meta location */
 608         sector_t la_size;
 609         sector_t size;
 610         char ppb[10];
 611
 612         int md_moved, la_size_changed;
 613         enum determine_dev_size rv = unchanged;
 614
 615         /* race:
 616          * application request passes inc_ap_bio,
 617          * but then cannot get an AL-reference.
 618          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 619          *
 620          * to avoid that:
 621          * Suspend IO right here.
 622          * still lock the act_log to not trigger ASSERTs there.
 623          */
 624         drbd_suspend_io(mdev);
 625
 626         /* no wait necessary anymore, actually we could assert that */
 627         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 628
 629         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 630         prev_size = mdev->ldev->md.md_size_sect;
 631         la_size = mdev->ldev->md.la_size_sect;
 632
 633         /* TODO: should only be some assert here, not (re)init... */
 634         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 635
 636         size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
 637
 638         if (drbd_get_capacity(mdev->this_bdev) != size ||
 639             drbd_bm_capacity(mdev) != size) {
 640                 int err;
 641                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 642                 if (unlikely(err)) {
 643                         /* currently there is only one error: ENOMEM! */
 644                         size = drbd_bm_capacity(mdev)>>1;
 645                         if (size == 0) {
 646                                 dev_err(DEV, "OUT OF MEMORY! "
 647                                     "Could not allocate bitmap!\n");
 648                         } else {
 649                                 dev_err(DEV, "BM resizing failed. "
 650                                     "Leaving size unchanged at size = %lu KB\n",
 651                                     (unsigned long)size);
 652                         }
 653                         rv = dev_size_error;
 654                 }
 655                 /* racy, see comments above. */
 656                 drbd_set_my_capacity(mdev, size);
 657                 mdev->ldev->md.la_size_sect = size;
 658                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 659                      (unsigned long long)size>>1);
 660         }
 661         if (rv == dev_size_error)
 662                 goto out;
 663
 664         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 665
 666         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 667                 || prev_size       != mdev->ldev->md.md_size_sect;
 668
 669         if (la_size_changed || md_moved) {
 670                 int err;
 671
 672                 drbd_al_shrink(mdev); /* All extents inactive. */
 673                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 674                          la_size_changed && md_moved ? "size changed and md moved" :
 675                          la_size_changed ? "size changed" : "md moved");
 676                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
 677                 err = drbd_bitmap_io(mdev, md_moved ? &drbd_bm_write_all : &drbd_bm_write,
 678                                      "size changed", BM_LOCKED_MASK);
 679                 if (err) {
 680                         rv = dev_size_error;
 681                         goto out;
 682                 }
 683                 drbd_md_mark_dirty(mdev);
 684         }
 685
 686         if (size > la_size)
 687                 rv = grew;
 688         if (size < la_size)
 689                 rv = shrunk;
 690 out:
 691         lc_unlock(mdev->act_log);
 692         wake_up(&mdev->al_wait);
 693         drbd_resume_io(mdev);
 694
 695         return rv;
 696 }
 697
 698 sector_t
 699 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
 700 {
 701         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 702         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 703         sector_t m_size; /* my size */
 704         sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
 705         sector_t size = 0;
 706
 707         m_size = drbd_get_max_capacity(bdev);
 708
 709         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 710                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 711                 p_size = m_size;
 712         }
 713
 714         if (p_size && m_size) {
 715                 size = min_t(sector_t, p_size, m_size);
 716         } else {
 717                 if (la_size) {
 718                         size = la_size;
 719                         if (m_size && m_size < size)
 720                                 size = m_size;
 721                         if (p_size && p_size < size)
 722                                 size = p_size;
 723                 } else {
 724                         if (m_size)
 725                                 size = m_size;
 726                         if (p_size)
 727                                 size = p_size;
 728                 }
 729         }
 730
 731         if (size == 0)
 732                 dev_err(DEV, "Both nodes diskless!\n");
 733
 734         if (u_size) {
 735                 if (u_size > size)
 736                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 737                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 738                 else
 739                         size = u_size;
 740         }
 741
 742         return size;
 743 }
 744
 745 /**
 746  * drbd_check_al_size() - Ensures that the AL is of the right size
 747  * @mdev:       DRBD device.
 748  *
 749  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 750  * failed, and 0 on success. You should call drbd_md_sync() after you called
 751  * this function.
 752  */
 753 static int drbd_check_al_size(struct drbd_conf *mdev)
 754 {
 755         struct lru_cache *n, *t;
 756         struct lc_element *e;
 757         unsigned int in_use;
 758         int i;
 759
 760         ERR_IF(mdev->sync_conf.al_extents < 7)
 761                 mdev->sync_conf.al_extents = 127;
 762
 763         if (mdev->act_log &&
 764             mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
 765                 return 0;
 766
 767         in_use = 0;
 768         t = mdev->act_log;
 769         n = lc_create("act_log", drbd_al_ext_cache,
 770                 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
 771
 772         if (n == NULL) {
 773                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 774                 return -ENOMEM;
 775         }
 776         spin_lock_irq(&mdev->al_lock);
 777         if (t) {
 778                 for (i = 0; i < t->nr_elements; i++) {
 779                         e = lc_element_by_index(t, i);
 780                         if (e->refcnt)
 781                                 dev_err(DEV, "refcnt(%d)==%d\n",
 782                                     e->lc_number, e->refcnt);
 783                         in_use += e->refcnt;
 784                 }
 785         }
 786         if (!in_use)
 787                 mdev->act_log = n;
 788         spin_unlock_irq(&mdev->al_lock);
 789         if (in_use) {
 790                 dev_err(DEV, "Activity log still in use!\n");
 791                 lc_destroy(n);
 792                 return -EBUSY;
 793         } else {
 794                 if (t)
 795                         lc_destroy(t);
 796         }
 797         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 798         return 0;
 799 }
 800
 801 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
 802 {
 803         struct request_queue * const q = mdev->rq_queue;
 804         unsigned int max_hw_sectors = max_bio_size >> 9;
 805         unsigned int max_segments = 0;
 806
 807         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 808                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 809
 810                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
 811                 max_segments = mdev->ldev->dc.max_bio_bvecs;
 812                 put_ldev(mdev);
 813         }
 814
 815         blk_queue_logical_block_size(q, 512);
 816         blk_queue_max_hw_sectors(q, max_hw_sectors);
 817         /* This is the workaround for "bio would need to, but cannot, be split" */
 818         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
 819         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
 820
 821         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 822                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 823
 824                 blk_queue_stack_limits(q, b);
 825
 826                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
 827                         dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
 828                                  q->backing_dev_info.ra_pages,
 829                                  b->backing_dev_info.ra_pages);
 830                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
 831                 }
 832                 put_ldev(mdev);
 833         }
 834 }
 835
 836 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
 837 {
 838         unsigned int now, new, local, peer;
 839
 840         now = queue_max_hw_sectors(mdev->rq_queue) << 9;
 841         local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
 842         peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
 843
 844         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 845                 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
 846                 mdev->local_max_bio_size = local;
 847                 put_ldev(mdev);
 848         }
 849         local = min(local, DRBD_MAX_BIO_SIZE);
 850
 851         /* We may ignore peer limits if the peer is modern enough.
 852            Because new from 8.3.8 onwards the peer can use multiple
 853            BIOs for a single peer_request */
 854         if (mdev->state.conn >= C_CONNECTED) {
 855                 if (mdev->agreed_pro_version < 94) {
 856                         peer = min(mdev->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
 857                         /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
 858                 } else if (mdev->agreed_pro_version == 94)
 859                         peer = DRBD_MAX_SIZE_H80_PACKET;
 860                 else /* drbd 8.3.8 onwards */
 861                         peer = DRBD_MAX_BIO_SIZE;
 862         }
 863
 864         new = min(local, peer);
 865
 866         if (mdev->state.role == R_PRIMARY && new < now)
 867                 dev_err(DEV, "ASSERT FAILED new < now; (%u < %u)\n", new, now);
 868
 869         if (new != now)
 870                 dev_info(DEV, "max BIO size = %u\n", new);
 871
 872         drbd_setup_queue_param(mdev, new);
 873 }
 874
 875 /* serialize deconfig (worker exiting, doing cleanup)
 876  * and reconfig (drbdsetup disk, drbdsetup net)
 877  *
 878  * Wait for a potentially exiting worker, then restart it,
 879  * or start a new one.  Flush any pending work, there may still be an
 880  * after_state_change queued.
 881  */
 882 static void drbd_reconfig_start(struct drbd_conf *mdev)
 883 {
 884         wait_event(mdev->state_wait, !drbd_test_and_set_flag(mdev, CONFIG_PENDING));
 885         wait_event(mdev->state_wait, !drbd_test_flag(mdev, DEVICE_DYING));
 886         drbd_thread_start(&mdev->worker);
 887         drbd_flush_workqueue(mdev);
 888 }
 889
 890 /* if still unconfigured, stops worker again.
 891  * if configured now, clears CONFIG_PENDING.
 892  * wakes potential waiters */
 893 static void drbd_reconfig_done(struct drbd_conf *mdev)
 894 {
 895         spin_lock_irq(&mdev->req_lock);
 896         if (mdev->state.disk == D_DISKLESS &&
 897             mdev->state.conn == C_STANDALONE &&
 898             mdev->state.role == R_SECONDARY) {
 899                 drbd_set_flag(mdev, DEVICE_DYING);
 900                 drbd_thread_stop_nowait(&mdev->worker);
 901         } else
 902                 drbd_clear_flag(mdev, CONFIG_PENDING);
 903         spin_unlock_irq(&mdev->req_lock);
 904         wake_up(&mdev->state_wait);
 905 }
 906
 907 /* Make sure IO is suspended before calling this function(). */
 908 static void drbd_suspend_al(struct drbd_conf *mdev)
 909 {
 910         int s = 0;
 911
 912         if (lc_try_lock(mdev->act_log)) {
 913                 drbd_al_shrink(mdev);
 914                 lc_unlock(mdev->act_log);
 915         } else {
 916                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
 917                 return;
 918         }
 919
 920         spin_lock_irq(&mdev->req_lock);
 921         if (mdev->state.conn < C_CONNECTED)
 922                 s = !drbd_test_and_set_flag(mdev, AL_SUSPENDED);
 923
 924         spin_unlock_irq(&mdev->req_lock);
 925
 926         if (s)
 927                 dev_info(DEV, "Suspended AL updates\n");
 928 }
 929
 930 /* does always return 0;
 931  * interesting return code is in reply->ret_code */
 932 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 933                              struct drbd_nl_cfg_reply *reply)
 934 {
 935         enum drbd_ret_code retcode;
 936         enum determine_dev_size dd;
 937         sector_t max_possible_sectors;
 938         sector_t min_md_device_sectors;
 939         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
 940         struct block_device *bdev;
 941         struct lru_cache *resync_lru = NULL;
 942         union drbd_state ns, os;
 943         enum drbd_state_rv rv;
 944         int cp_discovered = 0;
 945         int logical_block_size;
 946
 947         drbd_reconfig_start(mdev);
 948
 949         /* if you want to reconfigure, please tear down first */
 950         if (mdev->state.disk > D_DISKLESS) {
 951                 retcode = ERR_DISK_CONFIGURED;
 952                 goto fail;
 953         }
 954         /* It may just now have detached because of IO error.  Make sure
 955          * drbd_ldev_destroy is done already, we may end up here very fast,
 956          * e.g. if someone calls attach from the on-io-error handler,
 957          * to realize a "hot spare" feature (not that I'd recommend that) */
 958         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
 959
 960         /* make sure there is no leftover from previous force-detach attempts */
 961         drbd_clear_flag(mdev, FORCE_DETACH);
 962         drbd_clear_flag(mdev, WAS_IO_ERROR);
 963         drbd_clear_flag(mdev, WAS_READ_ERROR);
 964
 965         /* and no leftover from previously aborted resync or verify, either */
 966         mdev->rs_total = 0;
 967         mdev->rs_failed = 0;
 968         atomic_set(&mdev->rs_pending_cnt, 0);
 969
 970         /* allocation not in the IO path, cqueue thread context */
 971         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
 972         if (!nbc) {
 973                 retcode = ERR_NOMEM;
 974                 goto fail;
 975         }
 976
 977         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
 978         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
 979         nbc->dc.fencing       = DRBD_FENCING_DEF;
 980         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
 981
 982         spin_lock_init(&nbc->md.uuid_lock);
 983
 984         if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
 985                 retcode = ERR_MANDATORY_TAG;
 986                 goto fail;
 987         }
 988
 989         if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
 990                 retcode = ERR_MD_IDX_INVALID;
 991                 goto fail;
 992         }
 993
 994         if (get_net_conf(mdev)) {
 995                 int prot = mdev->net_conf->wire_protocol;
 996                 put_net_conf(mdev);
 997                 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
 998                         retcode = ERR_STONITH_AND_PROT_A;
 999                         goto fail;
1000                 }
1001         }
1002
1003         bdev = blkdev_get_by_path(nbc->dc.backing_dev,
1004                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1005         if (IS_ERR(bdev)) {
1006                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
1007                         PTR_ERR(bdev));
1008                 retcode = ERR_OPEN_DISK;
1009                 goto fail;
1010         }
1011         nbc->backing_bdev = bdev;
1012
1013         /*
1014          * meta_dev_idx >= 0: external fixed size, possibly multiple
1015          * drbd sharing one meta device.  TODO in that case, paranoia
1016          * check that [md_bdev, meta_dev_idx] is not yet used by some
1017          * other drbd minor!  (if you use drbd.conf + drbdadm, that
1018          * should check it for you already; but if you don't, or
1019          * someone fooled it, we need to double check here)
1020          */
1021         bdev = blkdev_get_by_path(nbc->dc.meta_dev,
1022                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1023                                   (nbc->dc.meta_dev_idx < 0) ?
1024                                   (void *)mdev : (void *)drbd_m_holder);
1025         if (IS_ERR(bdev)) {
1026                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
1027                         PTR_ERR(bdev));
1028                 retcode = ERR_OPEN_MD_DISK;
1029                 goto fail;
1030         }
1031         nbc->md_bdev = bdev;
1032
1033         if ((nbc->backing_bdev == nbc->md_bdev) !=
1034             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1035              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1036                 retcode = ERR_MD_IDX_INVALID;
1037                 goto fail;
1038         }
1039
1040         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1041                         61, sizeof(struct bm_extent),
1042                         offsetof(struct bm_extent, lce));
1043         if (!resync_lru) {
1044                 retcode = ERR_NOMEM;
1045                 goto fail;
1046         }
1047
1048         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1049         drbd_md_set_sector_offsets(mdev, nbc);
1050
1051         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
1052                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1053                         (unsigned long long) drbd_get_max_capacity(nbc),
1054                         (unsigned long long) nbc->dc.disk_size);
1055                 retcode = ERR_DISK_TOO_SMALL;
1056                 goto fail;
1057         }
1058
1059         if (nbc->dc.meta_dev_idx < 0) {
1060                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1061                 /* at least one MB, otherwise it does not make sense */
1062                 min_md_device_sectors = (2<<10);
1063         } else {
1064                 max_possible_sectors = DRBD_MAX_SECTORS;
1065                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
1066         }
1067
1068         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1069                 retcode = ERR_MD_DISK_TOO_SMALL;
1070                 dev_warn(DEV, "refusing attach: md-device too small, "
1071                      "at least %llu sectors needed for this meta-disk type\n",
1072                      (unsigned long long) min_md_device_sectors);
1073                 goto fail;
1074         }
1075
1076         /* Make sure the new disk is big enough
1077          * (we may currently be R_PRIMARY with no local disk...) */
1078         if (drbd_get_max_capacity(nbc) <
1079             drbd_get_capacity(mdev->this_bdev)) {
1080                 retcode = ERR_DISK_TOO_SMALL;
1081                 goto fail;
1082         }
1083
1084         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1085
1086         if (nbc->known_size > max_possible_sectors) {
1087                 dev_warn(DEV, "==> truncating very big lower level device "
1088                         "to currently maximum possible %llu sectors <==\n",
1089                         (unsigned long long) max_possible_sectors);
1090                 if (nbc->dc.meta_dev_idx >= 0)
1091                         dev_warn(DEV, "==>> using internal or flexible "
1092                                       "meta data may help <<==\n");
1093         }
1094
1095         drbd_suspend_io(mdev);
1096         /* also wait for the last barrier ack. */
1097         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state));
1098         /* and for any other previously queued work */
1099         drbd_flush_workqueue(mdev);
1100
1101         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1102         retcode = rv;  /* FIXME: Type mismatch. */
1103         drbd_resume_io(mdev);
1104         if (rv < SS_SUCCESS)
1105                 goto fail;
1106
1107         if (!get_ldev_if_state(mdev, D_ATTACHING))
1108                 goto force_diskless;
1109
1110         drbd_md_set_sector_offsets(mdev, nbc);
1111
1112         /* allocate a second IO page if logical_block_size != 512 */
1113         logical_block_size = bdev_logical_block_size(nbc->md_bdev);
1114         if (logical_block_size == 0)
1115                 logical_block_size = MD_SECTOR_SIZE;
1116
1117         if (logical_block_size != MD_SECTOR_SIZE) {
1118                 if (!mdev->md_io_tmpp) {
1119                         struct page *page = alloc_page(GFP_NOIO);
1120                         if (!page)
1121                                 goto force_diskless_dec;
1122
1123                         dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
1124                              logical_block_size, MD_SECTOR_SIZE);
1125                         dev_warn(DEV, "Workaround engaged (has performance impact).\n");
1126
1127                         mdev->md_io_tmpp = page;
1128                 }
1129         }
1130
1131         if (!mdev->bitmap) {
1132                 if (drbd_bm_init(mdev)) {
1133                         retcode = ERR_NOMEM;
1134                         goto force_diskless_dec;
1135                 }
1136         }
1137
1138         retcode = drbd_md_read(mdev, nbc);
1139         if (retcode != NO_ERROR)
1140                 goto force_diskless_dec;
1141
1142         if (mdev->state.conn < C_CONNECTED &&
1143             mdev->state.role == R_PRIMARY &&
1144             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1145                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1146                     (unsigned long long)mdev->ed_uuid);
1147                 retcode = ERR_DATA_NOT_CURRENT;
1148                 goto force_diskless_dec;
1149         }
1150
1151         /* Since we are diskless, fix the activity log first... */
1152         if (drbd_check_al_size(mdev)) {
1153                 retcode = ERR_NOMEM;
1154                 goto force_diskless_dec;
1155         }
1156
1157         /* Prevent shrinking of consistent devices ! */
1158         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1159             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1160                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1161                 retcode = ERR_DISK_TOO_SMALL;
1162                 goto force_diskless_dec;
1163         }
1164
1165         if (!drbd_al_read_log(mdev, nbc)) {
1166                 retcode = ERR_IO_MD_DISK;
1167                 goto force_diskless_dec;
1168         }
1169
1170         /* Reset the "barriers don't work" bits here, then force meta data to
1171          * be written, to ensure we determine if barriers are supported. */
1172         if (nbc->dc.no_md_flush)
1173                 drbd_set_flag(mdev, MD_NO_FUA);
1174         else
1175                 drbd_clear_flag(mdev, MD_NO_FUA);
1176
1177         /* Point of no return reached.
1178          * Devices and memory are no longer released by error cleanup below.
1179          * now mdev takes over responsibility, and the state engine should
1180          * clean it up somewhere.  */
1181         D_ASSERT(mdev->ldev == NULL);
1182         mdev->ldev = nbc;
1183         mdev->resync = resync_lru;
1184         nbc = NULL;
1185         resync_lru = NULL;
1186
1187         mdev->write_ordering = WO_bdev_flush;
1188         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1189
1190         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1191                 drbd_set_flag(mdev, CRASHED_PRIMARY);
1192         else
1193                 drbd_clear_flag(mdev, CRASHED_PRIMARY);
1194
1195         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1196             !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) {
1197                 drbd_set_flag(mdev, CRASHED_PRIMARY);
1198                 cp_discovered = 1;
1199         }
1200
1201         mdev->send_cnt = 0;
1202         mdev->recv_cnt = 0;
1203         mdev->read_cnt = 0;
1204         mdev->writ_cnt = 0;
1205
1206         drbd_reconsider_max_bio_size(mdev);
1207
1208         /* If I am currently not R_PRIMARY,
1209          * but meta data primary indicator is set,
1210          * I just now recover from a hard crash,
1211          * and have been R_PRIMARY before that crash.
1212          *
1213          * Now, if I had no connection before that crash
1214          * (have been degraded R_PRIMARY), chances are that
1215          * I won't find my peer now either.
1216          *
1217          * In that case, and _only_ in that case,
1218          * we use the degr-wfc-timeout instead of the default,
1219          * so we can automatically recover from a crash of a
1220          * degraded but active "cluster" after a certain timeout.
1221          */
1222         drbd_clear_flag(mdev, USE_DEGR_WFC_T);
1223         if (mdev->state.role != R_PRIMARY &&
1224              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1225             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1226                 drbd_set_flag(mdev, USE_DEGR_WFC_T);
1227
1228         dd = drbd_determine_dev_size(mdev, 0);
1229         if (dd == dev_size_error) {
1230                 retcode = ERR_NOMEM_BITMAP;
1231                 goto force_diskless_dec;
1232         } else if (dd == grew)
1233                 drbd_set_flag(mdev, RESYNC_AFTER_NEG);
1234
1235         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1236                 dev_info(DEV, "Assuming that all blocks are out of sync "
1237                      "(aka FullSync)\n");
1238                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1239                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1240                         retcode = ERR_IO_MD_DISK;
1241                         goto force_diskless_dec;
1242                 }
1243         } else {
1244                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1245                         "read from attaching", BM_LOCKED_MASK) < 0) {
1246                         retcode = ERR_IO_MD_DISK;
1247                         goto force_diskless_dec;
1248                 }
1249         }
1250
1251         if (cp_discovered) {
1252                 drbd_al_apply_to_bm(mdev);
1253                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1254                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1255                         retcode = ERR_IO_MD_DISK;
1256                         goto force_diskless_dec;
1257                 }
1258         }
1259
1260         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1261                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1262
1263         spin_lock_irq(&mdev->req_lock);
1264         os = mdev->state;
1265         ns.i = os.i;
1266         /* If MDF_CONSISTENT is not set go into inconsistent state,
1267            otherwise investigate MDF_WasUpToDate...
1268            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1269            otherwise into D_CONSISTENT state.
1270         */
1271         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1272                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1273                         ns.disk = D_CONSISTENT;
1274                 else
1275                         ns.disk = D_OUTDATED;
1276         } else {
1277                 ns.disk = D_INCONSISTENT;
1278         }
1279
1280         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1281                 ns.pdsk = D_OUTDATED;
1282
1283         if ( ns.disk == D_CONSISTENT &&
1284             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1285                 ns.disk = D_UP_TO_DATE;
1286
1287         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1288            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1289            this point, because drbd_request_state() modifies these
1290            flags. */
1291
1292         /* In case we are C_CONNECTED postpone any decision on the new disk
1293            state after the negotiation phase. */
1294         if (mdev->state.conn == C_CONNECTED) {
1295                 mdev->new_state_tmp.i = ns.i;
1296                 ns.i = os.i;
1297                 ns.disk = D_NEGOTIATING;
1298
1299                 /* We expect to receive up-to-date UUIDs soon.
1300                    To avoid a race in receive_state, free p_uuid while
1301                    holding req_lock. I.e. atomic with the state change */
1302                 kfree(mdev->p_uuid);
1303                 mdev->p_uuid = NULL;
1304         }
1305
1306         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1307         ns = mdev->state;
1308         spin_unlock_irq(&mdev->req_lock);
1309
1310         if (rv < SS_SUCCESS)
1311                 goto force_diskless_dec;
1312
1313         if (mdev->state.role == R_PRIMARY)
1314                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1315         else
1316                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1317
1318         drbd_md_mark_dirty(mdev);
1319         drbd_md_sync(mdev);
1320
1321         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1322         put_ldev(mdev);
1323         reply->ret_code = retcode;
1324         drbd_reconfig_done(mdev);
1325         return 0;
1326
1327  force_diskless_dec:
1328         put_ldev(mdev);
1329  force_diskless:
1330         drbd_force_state(mdev, NS(disk, D_FAILED));
1331         drbd_md_sync(mdev);
1332  fail:
1333         if (nbc) {
1334                 if (nbc->backing_bdev)
1335                         blkdev_put(nbc->backing_bdev,
1336                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1337                 if (nbc->md_bdev)
1338                         blkdev_put(nbc->md_bdev,
1339                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1340                 kfree(nbc);
1341         }
1342         lc_destroy(resync_lru);
1343
1344         reply->ret_code = retcode;
1345         drbd_reconfig_done(mdev);
1346         return 0;
1347 }
1348
1349 /* Detaching the disk is a process in multiple stages.  First we need to lock
1350  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1351  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1352  * internal references as well.
1353  * Only then we have finally detached. */
1354 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1355                           struct drbd_nl_cfg_reply *reply)
1356 {
1357         enum drbd_ret_code retcode;
1358         int ret;
1359         struct detach dt = {};
1360
1361         if (!detach_from_tags(mdev, nlp->tag_list, &dt)) {
1362                 reply->ret_code = ERR_MANDATORY_TAG;
1363                 goto out;
1364         }
1365
1366         if (dt.detach_force) {
1367                 drbd_set_flag(mdev, FORCE_DETACH);
1368                 drbd_force_state(mdev, NS(disk, D_FAILED));
1369                 reply->ret_code = SS_SUCCESS;
1370                 goto out;
1371         }
1372
1373         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1374         drbd_md_get_buffer(mdev); /* make sure there is no in-flight meta-data IO */
1375         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1376         drbd_md_put_buffer(mdev);
1377         /* D_FAILED will transition to DISKLESS. */
1378         ret = wait_event_interruptible(mdev->misc_wait,
1379                         mdev->state.disk != D_FAILED);
1380         drbd_resume_io(mdev);
1381
1382         if ((int)retcode == (int)SS_IS_DISKLESS)
1383                 retcode = SS_NOTHING_TO_DO;
1384         if (ret)
1385                 retcode = ERR_INTR;
1386         reply->ret_code = retcode;
1387 out:
1388         return 0;
1389 }
1390
1391 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1392                             struct drbd_nl_cfg_reply *reply)
1393 {
1394         int i, ns;
1395         enum drbd_ret_code retcode;
1396         struct net_conf *new_conf = NULL;
1397         struct crypto_hash *tfm = NULL;
1398         struct crypto_hash *integrity_w_tfm = NULL;
1399         struct crypto_hash *integrity_r_tfm = NULL;
1400         struct hlist_head *new_tl_hash = NULL;
1401         struct hlist_head *new_ee_hash = NULL;
1402         struct drbd_conf *odev;
1403         char hmac_name[CRYPTO_MAX_ALG_NAME];
1404         void *int_dig_out = NULL;
1405         void *int_dig_in = NULL;
1406         void *int_dig_vv = NULL;
1407         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1408
1409         drbd_reconfig_start(mdev);
1410
1411         if (mdev->state.conn > C_STANDALONE) {
1412                 retcode = ERR_NET_CONFIGURED;
1413                 goto fail;
1414         }
1415
1416         /* allocation not in the IO path, cqueue thread context */
1417         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1418         if (!new_conf) {
1419                 retcode = ERR_NOMEM;
1420                 goto fail;
1421         }
1422
1423         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1424         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1425         new_conf->ping_int         = DRBD_PING_INT_DEF;
1426         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1427         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1428         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1429         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1430         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1431         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1432         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1433         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1434         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1435         new_conf->want_lose        = 0;
1436         new_conf->two_primaries    = 0;
1437         new_conf->wire_protocol    = DRBD_PROT_C;
1438         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1439         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1440         new_conf->on_congestion    = DRBD_ON_CONGESTION_DEF;
1441         new_conf->cong_extents     = DRBD_CONG_EXTENTS_DEF;
1442
1443         if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1444                 retcode = ERR_MANDATORY_TAG;
1445                 goto fail;
1446         }
1447
1448         if (new_conf->two_primaries
1449             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1450                 retcode = ERR_NOT_PROTO_C;
1451                 goto fail;
1452         }
1453
1454         if (get_ldev(mdev)) {
1455                 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1456                 put_ldev(mdev);
1457                 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1458                         retcode = ERR_STONITH_AND_PROT_A;
1459                         goto fail;
1460                 }
1461         }
1462
1463         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1464                 retcode = ERR_CONG_NOT_PROTO_A;
1465                 goto fail;
1466         }
1467
1468         if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1469                 retcode = ERR_DISCARD;
1470                 goto fail;
1471         }
1472
1473         retcode = NO_ERROR;
1474
1475         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1476         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1477         for (i = 0; i < minor_count; i++) {
1478                 odev = minor_to_mdev(i);
1479                 if (!odev || odev == mdev)
1480                         continue;
1481                 if (get_net_conf(odev)) {
1482                         taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1483                         if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1484                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1485                                 retcode = ERR_LOCAL_ADDR;
1486
1487                         taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1488                         if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1489                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1490                                 retcode = ERR_PEER_ADDR;
1491
1492                         put_net_conf(odev);
1493                         if (retcode != NO_ERROR)
1494                                 goto fail;
1495                 }
1496         }
1497
1498         if (new_conf->cram_hmac_alg[0] != 0) {
1499                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1500                         new_conf->cram_hmac_alg);
1501                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1502                 if (IS_ERR(tfm)) {
1503                         tfm = NULL;
1504                         retcode = ERR_AUTH_ALG;
1505                         goto fail;
1506                 }
1507
1508                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1509                         retcode = ERR_AUTH_ALG_ND;
1510                         goto fail;
1511                 }
1512         }
1513
1514         if (new_conf->integrity_alg[0]) {
1515                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1516                 if (IS_ERR(integrity_w_tfm)) {
1517                         integrity_w_tfm = NULL;
1518                         retcode=ERR_INTEGRITY_ALG;
1519                         goto fail;
1520                 }
1521
1522                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1523                         retcode=ERR_INTEGRITY_ALG_ND;
1524                         goto fail;
1525                 }
1526
1527                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1528                 if (IS_ERR(integrity_r_tfm)) {
1529                         integrity_r_tfm = NULL;
1530                         retcode=ERR_INTEGRITY_ALG;
1531                         goto fail;
1532                 }
1533         }
1534
1535         ns = new_conf->max_epoch_size/8;
1536         if (mdev->tl_hash_s != ns) {
1537                 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1538                 if (!new_tl_hash) {
1539                         retcode = ERR_NOMEM;
1540                         goto fail;
1541                 }
1542         }
1543
1544         ns = new_conf->max_buffers/8;
1545         if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1546                 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1547                 if (!new_ee_hash) {
1548                         retcode = ERR_NOMEM;
1549                         goto fail;
1550                 }
1551         }
1552
1553         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1554
1555         if (integrity_w_tfm) {
1556                 i = crypto_hash_digestsize(integrity_w_tfm);
1557                 int_dig_out = kmalloc(i, GFP_KERNEL);
1558                 if (!int_dig_out) {
1559                         retcode = ERR_NOMEM;
1560                         goto fail;
1561                 }
1562                 int_dig_in = kmalloc(i, GFP_KERNEL);
1563                 if (!int_dig_in) {
1564                         retcode = ERR_NOMEM;
1565                         goto fail;
1566                 }
1567                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1568                 if (!int_dig_vv) {
1569                         retcode = ERR_NOMEM;
1570                         goto fail;
1571                 }
1572         }
1573
1574         if (!mdev->bitmap) {
1575                 if(drbd_bm_init(mdev)) {
1576                         retcode = ERR_NOMEM;
1577                         goto fail;
1578                 }
1579         }
1580
1581         drbd_flush_workqueue(mdev);
1582         spin_lock_irq(&mdev->req_lock);
1583         if (mdev->net_conf != NULL) {
1584                 retcode = ERR_NET_CONFIGURED;
1585                 spin_unlock_irq(&mdev->req_lock);
1586                 goto fail;
1587         }
1588         mdev->net_conf = new_conf;
1589
1590         mdev->send_cnt = 0;
1591         mdev->recv_cnt = 0;
1592
1593         if (new_tl_hash) {
1594                 kfree(mdev->tl_hash);
1595                 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1596                 mdev->tl_hash = new_tl_hash;
1597         }
1598
1599         if (new_ee_hash) {
1600                 kfree(mdev->ee_hash);
1601                 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1602                 mdev->ee_hash = new_ee_hash;
1603         }
1604
1605         crypto_free_hash(mdev->cram_hmac_tfm);
1606         mdev->cram_hmac_tfm = tfm;
1607
1608         crypto_free_hash(mdev->integrity_w_tfm);
1609         mdev->integrity_w_tfm = integrity_w_tfm;
1610
1611         crypto_free_hash(mdev->integrity_r_tfm);
1612         mdev->integrity_r_tfm = integrity_r_tfm;
1613
1614         kfree(mdev->int_dig_out);
1615         kfree(mdev->int_dig_in);
1616         kfree(mdev->int_dig_vv);
1617         mdev->int_dig_out=int_dig_out;
1618         mdev->int_dig_in=int_dig_in;
1619         mdev->int_dig_vv=int_dig_vv;
1620         retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL);
1621         spin_unlock_irq(&mdev->req_lock);
1622
1623         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1624         reply->ret_code = retcode;
1625         drbd_reconfig_done(mdev);
1626         return 0;
1627
1628 fail:
1629         kfree(int_dig_out);
1630         kfree(int_dig_in);
1631         kfree(int_dig_vv);
1632         crypto_free_hash(tfm);
1633         crypto_free_hash(integrity_w_tfm);
1634         crypto_free_hash(integrity_r_tfm);
1635         kfree(new_tl_hash);
1636         kfree(new_ee_hash);
1637         kfree(new_conf);
1638
1639         reply->ret_code = retcode;
1640         drbd_reconfig_done(mdev);
1641         return 0;
1642 }
1643
1644 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1645                               struct drbd_nl_cfg_reply *reply)
1646 {
1647         int retcode;
1648         struct disconnect dc;
1649
1650         memset(&dc, 0, sizeof(struct disconnect));
1651         if (!disconnect_from_tags(mdev, nlp->tag_list, &dc)) {
1652                 retcode = ERR_MANDATORY_TAG;
1653                 goto fail;
1654         }
1655
1656         if (dc.force) {
1657                 spin_lock_irq(&mdev->req_lock);
1658                 if (mdev->state.conn >= C_WF_CONNECTION)
1659                         _drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), CS_HARD, NULL);
1660                 spin_unlock_irq(&mdev->req_lock);
1661                 goto done;
1662         }
1663
1664         retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1665
1666         if (retcode == SS_NOTHING_TO_DO)
1667                 goto done;
1668         else if (retcode == SS_ALREADY_STANDALONE)
1669                 goto done;
1670         else if (retcode == SS_PRIMARY_NOP) {
1671                 /* Our statche checking code wants to see the peer outdated. */
1672                 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1673                                                       pdsk, D_OUTDATED));
1674         } else if (retcode == SS_CW_FAILED_BY_PEER) {
1675                 /* The peer probably wants to see us outdated. */
1676                 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1677                                                         disk, D_OUTDATED),
1678                                               CS_ORDERED);
1679                 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1680                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1681                         retcode = SS_SUCCESS;
1682                 }
1683         }
1684
1685         if (retcode < SS_SUCCESS)
1686                 goto fail;
1687
1688         if (wait_event_interruptible(mdev->state_wait,
1689                                      mdev->state.conn != C_DISCONNECTING)) {
1690                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1691                    someone else might connect us in the mean time! */
1692                 retcode = ERR_INTR;
1693                 goto fail;
1694         }
1695
1696  done:
1697         retcode = NO_ERROR;
1698  fail:
1699         drbd_md_sync(mdev);
1700         reply->ret_code = retcode;
1701         return 0;
1702 }
1703
1704 void resync_after_online_grow(struct drbd_conf *mdev)
1705 {
1706         int iass; /* I am sync source */
1707
1708         dev_info(DEV, "Resync of new storage after online grow\n");
1709         if (mdev->state.role != mdev->state.peer)
1710                 iass = (mdev->state.role == R_PRIMARY);
1711         else
1712                 iass = drbd_test_flag(mdev, DISCARD_CONCURRENT);
1713
1714         if (iass)
1715                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1716         else
1717                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1718 }
1719
1720 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1721                           struct drbd_nl_cfg_reply *reply)
1722 {
1723         struct resize rs;
1724         int retcode = NO_ERROR;
1725         enum determine_dev_size dd;
1726         enum dds_flags ddsf;
1727
1728         memset(&rs, 0, sizeof(struct resize));
1729         if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1730                 retcode = ERR_MANDATORY_TAG;
1731                 goto fail;
1732         }
1733
1734         if (mdev->state.conn > C_CONNECTED) {
1735                 retcode = ERR_RESIZE_RESYNC;
1736                 goto fail;
1737         }
1738
1739         if (mdev->state.role == R_SECONDARY &&
1740             mdev->state.peer == R_SECONDARY) {
1741                 retcode = ERR_NO_PRIMARY;
1742                 goto fail;
1743         }
1744
1745         if (!get_ldev(mdev)) {
1746                 retcode = ERR_NO_DISK;
1747                 goto fail;
1748         }
1749
1750         if (rs.no_resync && mdev->agreed_pro_version < 93) {
1751                 retcode = ERR_NEED_APV_93;
1752                 goto fail_ldev;
1753         }
1754
1755         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1756                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1757
1758         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1759         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1760         dd = drbd_determine_dev_size(mdev, ddsf);
1761         drbd_md_sync(mdev);
1762         put_ldev(mdev);
1763         if (dd == dev_size_error) {
1764                 retcode = ERR_NOMEM_BITMAP;
1765                 goto fail;
1766         }
1767
1768         if (mdev->state.conn == C_CONNECTED) {
1769                 if (dd == grew)
1770                         drbd_set_flag(mdev, RESIZE_PENDING);
1771
1772                 drbd_send_uuids(mdev);
1773                 drbd_send_sizes(mdev, 1, ddsf);
1774         }
1775
1776  fail:
1777         reply->ret_code = retcode;
1778         return 0;
1779
1780  fail_ldev:
1781         put_ldev(mdev);
1782         goto fail;
1783 }
1784
1785 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1786                                struct drbd_nl_cfg_reply *reply)
1787 {
1788         int retcode = NO_ERROR;
1789         int err;
1790         int ovr; /* online verify running */
1791         int rsr; /* re-sync running */
1792         struct crypto_hash *verify_tfm = NULL;
1793         struct crypto_hash *csums_tfm = NULL;
1794         struct syncer_conf sc;
1795         cpumask_var_t new_cpu_mask;
1796         int *rs_plan_s = NULL;
1797         int fifo_size;
1798
1799         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1800                 retcode = ERR_NOMEM;
1801                 goto fail;
1802         }
1803
1804         if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1805                 memset(&sc, 0, sizeof(struct syncer_conf));
1806                 sc.rate       = DRBD_RATE_DEF;
1807                 sc.after      = DRBD_AFTER_DEF;
1808                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1809                 sc.on_no_data  = DRBD_ON_NO_DATA_DEF;
1810                 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1811                 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1812                 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1813                 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1814                 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1815         } else
1816                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1817
1818         if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1819                 retcode = ERR_MANDATORY_TAG;
1820                 goto fail;
1821         }
1822
1823         /* re-sync running */
1824         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1825                 mdev->state.conn == C_SYNC_TARGET ||
1826                 mdev->state.conn == C_PAUSED_SYNC_S ||
1827                 mdev->state.conn == C_PAUSED_SYNC_T );
1828
1829         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1830                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1831                 goto fail;
1832         }
1833
1834         if (!rsr && sc.csums_alg[0]) {
1835                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1836                 if (IS_ERR(csums_tfm)) {
1837                         csums_tfm = NULL;
1838                         retcode = ERR_CSUMS_ALG;
1839                         goto fail;
1840                 }
1841
1842                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1843                         retcode = ERR_CSUMS_ALG_ND;
1844                         goto fail;
1845                 }
1846         }
1847
1848         /* online verify running */
1849         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1850
1851         if (ovr) {
1852                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1853                         retcode = ERR_VERIFY_RUNNING;
1854                         goto fail;
1855                 }
1856         }
1857
1858         if (!ovr && sc.verify_alg[0]) {
1859                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1860                 if (IS_ERR(verify_tfm)) {
1861                         verify_tfm = NULL;
1862                         retcode = ERR_VERIFY_ALG;
1863                         goto fail;
1864                 }
1865
1866                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1867                         retcode = ERR_VERIFY_ALG_ND;
1868                         goto fail;
1869                 }
1870         }
1871
1872         /* silently ignore cpu mask on UP kernel */
1873         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1874                 err = bitmap_parse(sc.cpu_mask, 32,
1875                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1876                 if (err) {
1877                         dev_warn(DEV, "bitmap_parse() failed with %d\n", err);
1878                         retcode = ERR_CPU_MASK_PARSE;
1879                         goto fail;
1880                 }
1881         }
1882
1883         ERR_IF (sc.rate < 1) sc.rate = 1;
1884         ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1885 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1886         if (sc.al_extents > AL_MAX) {
1887                 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1888                 sc.al_extents = AL_MAX;
1889         }
1890 #undef AL_MAX
1891
1892         /* to avoid spurious errors when configuring minors before configuring
1893          * the minors they depend on: if necessary, first create the minor we
1894          * depend on */
1895         if (sc.after >= 0)
1896                 ensure_mdev(sc.after, 1);
1897
1898         /* most sanity checks done, try to assign the new sync-after
1899          * dependency.  need to hold the global lock in there,
1900          * to avoid a race in the dependency loop check. */
1901         retcode = drbd_alter_sa(mdev, sc.after);
1902         if (retcode != NO_ERROR)
1903                 goto fail;
1904
1905         fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1906         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1907                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1908                 if (!rs_plan_s) {
1909                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1910                         retcode = ERR_NOMEM;
1911                         goto fail;
1912                 }
1913         }
1914
1915         /* ok, assign the rest of it as well.
1916          * lock against receive_SyncParam() */
1917         spin_lock(&mdev->peer_seq_lock);
1918         mdev->sync_conf = sc;
1919
1920         if (!rsr) {
1921                 crypto_free_hash(mdev->csums_tfm);
1922                 mdev->csums_tfm = csums_tfm;
1923                 csums_tfm = NULL;
1924         }
1925
1926         if (!ovr) {
1927                 crypto_free_hash(mdev->verify_tfm);
1928                 mdev->verify_tfm = verify_tfm;
1929                 verify_tfm = NULL;
1930         }
1931
1932         if (fifo_size != mdev->rs_plan_s.size) {
1933                 kfree(mdev->rs_plan_s.values);
1934                 mdev->rs_plan_s.values = rs_plan_s;
1935                 mdev->rs_plan_s.size   = fifo_size;
1936                 mdev->rs_planed = 0;
1937                 rs_plan_s = NULL;
1938         }
1939
1940         spin_unlock(&mdev->peer_seq_lock);
1941
1942         if (get_ldev(mdev)) {
1943                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1944                 drbd_al_shrink(mdev);
1945                 err = drbd_check_al_size(mdev);
1946                 lc_unlock(mdev->act_log);
1947                 wake_up(&mdev->al_wait);
1948
1949                 put_ldev(mdev);
1950                 drbd_md_sync(mdev);
1951
1952                 if (err) {
1953                         retcode = ERR_NOMEM;
1954                         goto fail;
1955                 }
1956         }
1957
1958         if (mdev->state.conn >= C_CONNECTED)
1959                 drbd_send_sync_param(mdev, &sc);
1960
1961         if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1962                 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1963                 drbd_calc_cpu_mask(mdev);
1964                 mdev->receiver.reset_cpu_mask = 1;
1965                 mdev->asender.reset_cpu_mask = 1;
1966                 mdev->worker.reset_cpu_mask = 1;
1967         }
1968
1969         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1970 fail:
1971         kfree(rs_plan_s);
1972         free_cpumask_var(new_cpu_mask);
1973         crypto_free_hash(csums_tfm);
1974         crypto_free_hash(verify_tfm);
1975         reply->ret_code = retcode;
1976         return 0;
1977 }
1978
1979 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1980                               struct drbd_nl_cfg_reply *reply)
1981 {
1982         int retcode;
1983
1984         /* If there is still bitmap IO pending, probably because of a previous
1985          * resync just being finished, wait for it before requesting a new resync.
1986          * Also wait for it's after_state_ch(). */
1987         drbd_suspend_io(mdev);
1988         wait_event(mdev->misc_wait, !drbd_test_flag(mdev, BITMAP_IO));
1989         drbd_flush_workqueue(mdev);
1990
1991         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1992
1993         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1994                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1995
1996         while (retcode == SS_NEED_CONNECTION) {
1997                 spin_lock_irq(&mdev->req_lock);
1998                 if (mdev->state.conn < C_CONNECTED)
1999                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2000                 spin_unlock_irq(&mdev->req_lock);
2001
2002                 if (retcode != SS_NEED_CONNECTION)
2003                         break;
2004
2005                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2006         }
2007         drbd_resume_io(mdev);
2008
2009         reply->ret_code = retcode;
2010         return 0;
2011 }
2012
2013 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2014 {
2015         int rv;
2016
2017         rv = drbd_bmio_set_n_write(mdev);
2018         drbd_suspend_al(mdev);
2019         return rv;
2020 }
2021
2022 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2023                                    struct drbd_nl_cfg_reply *reply)
2024 {
2025         int retcode;
2026
2027         /* If there is still bitmap IO pending, probably because of a previous
2028          * resync just being finished, wait for it before requesting a new resync.
2029          * Also wait for it's after_state_ch(). */
2030         drbd_suspend_io(mdev);
2031         wait_event(mdev->misc_wait, !drbd_test_flag(mdev, BITMAP_IO));
2032         drbd_flush_workqueue(mdev);
2033
2034         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
2035
2036         if (retcode < SS_SUCCESS) {
2037                 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
2038                         /* The peer will get a resync upon connect anyways. Just make that
2039                            into a full resync. */
2040                         retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
2041                         if (retcode >= SS_SUCCESS) {
2042                                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
2043                                         "set_n_write from invalidate_peer",
2044                                         BM_LOCKED_SET_ALLOWED))
2045                                         retcode = ERR_IO_MD_DISK;
2046                         }
2047                 } else
2048                         retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
2049         }
2050         drbd_resume_io(mdev);
2051
2052         reply->ret_code = retcode;
2053         return 0;
2054 }
2055
2056 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2057                               struct drbd_nl_cfg_reply *reply)
2058 {
2059         int retcode = NO_ERROR;
2060
2061         if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2062                 retcode = ERR_PAUSE_IS_SET;
2063
2064         reply->ret_code = retcode;
2065         return 0;
2066 }
2067
2068 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2069                                struct drbd_nl_cfg_reply *reply)
2070 {
2071         int retcode = NO_ERROR;
2072         union drbd_state s;
2073
2074         if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2075                 s = mdev->state;
2076                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2077                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2078                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2079                 } else {
2080                         retcode = ERR_PAUSE_IS_CLEAR;
2081                 }
2082         }
2083
2084         reply->ret_code = retcode;
2085         return 0;
2086 }
2087
2088 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2089                               struct drbd_nl_cfg_reply *reply)
2090 {
2091         reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
2092
2093         return 0;
2094 }
2095
2096 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2097                              struct drbd_nl_cfg_reply *reply)
2098 {
2099         if (drbd_test_flag(mdev, NEW_CUR_UUID)) {
2100                 drbd_uuid_new_current(mdev);
2101                 drbd_clear_flag(mdev, NEW_CUR_UUID);
2102         }
2103         drbd_suspend_io(mdev);
2104         reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2105         if (reply->ret_code == SS_SUCCESS) {
2106                 if (mdev->state.conn < C_CONNECTED)
2107                         tl_clear(mdev);
2108                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2109                         tl_restart(mdev, fail_frozen_disk_io);
2110         }
2111         drbd_resume_io(mdev);
2112
2113         return 0;
2114 }
2115
2116 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2117                            struct drbd_nl_cfg_reply *reply)
2118 {
2119         reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
2120         return 0;
2121 }
2122
2123 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2124                            struct drbd_nl_cfg_reply *reply)
2125 {
2126         unsigned short *tl;
2127
2128         tl = reply->tag_list;
2129
2130         if (get_ldev(mdev)) {
2131                 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
2132                 put_ldev(mdev);
2133         }
2134
2135         if (get_net_conf(mdev)) {
2136                 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
2137                 put_net_conf(mdev);
2138         }
2139         tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
2140
2141         put_unaligned(TT_END, tl++); /* Close the tag list */
2142
2143         return (int)((char *)tl - (char *)reply->tag_list);
2144 }
2145
2146 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2147                              struct drbd_nl_cfg_reply *reply)
2148 {
2149         unsigned short *tl = reply->tag_list;
2150         union drbd_state s = mdev->state;
2151         unsigned long rs_left;
2152         unsigned int res;
2153
2154         tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
2155
2156         /* no local ref, no bitmap, no syncer progress. */
2157         if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
2158                 if (get_ldev(mdev)) {
2159                         drbd_get_syncer_progress(mdev, &rs_left, &res);
2160                         tl = tl_add_int(tl, T_sync_progress, &res);
2161                         put_ldev(mdev);
2162                 }
2163         }
2164         put_unaligned(TT_END, tl++); /* Close the tag list */
2165
2166         return (int)((char *)tl - (char *)reply->tag_list);
2167 }
2168
2169 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2170                              struct drbd_nl_cfg_reply *reply)
2171 {
2172         unsigned short *tl;
2173
2174         tl = reply->tag_list;
2175
2176         if (get_ldev(mdev)) {
2177                 unsigned long flags;
2178                 spin_lock_irqsave(&mdev->ldev->md.uuid_lock, flags);
2179                 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
2180                 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
2181                 spin_unlock_irqrestore(&mdev->ldev->md.uuid_lock, flags);
2182                 put_ldev(mdev);
2183         }
2184         put_unaligned(TT_END, tl++); /* Close the tag list */
2185
2186         return (int)((char *)tl - (char *)reply->tag_list);
2187 }
2188
2189 /**
2190  * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
2191  * @mdev:       DRBD device.
2192  * @nlp:        Netlink/connector packet from drbdsetup
2193  * @reply:      Reply packet for drbdsetup
2194  */
2195 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2196                                     struct drbd_nl_cfg_reply *reply)
2197 {
2198         unsigned short *tl;
2199         char rv;
2200
2201         tl = reply->tag_list;
2202
2203         rv = mdev->state.pdsk == D_OUTDATED        ? UT_PEER_OUTDATED :
2204           drbd_test_flag(mdev, USE_DEGR_WFC_T) ? UT_DEGRADED : UT_DEFAULT;
2205
2206         tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
2207         put_unaligned(TT_END, tl++); /* Close the tag list */
2208
2209         return (int)((char *)tl - (char *)reply->tag_list);
2210 }
2211
2212 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2213                                     struct drbd_nl_cfg_reply *reply)
2214 {
2215         /* default to resume from last known position, if possible */
2216         struct start_ov args = {
2217                 .start_sector = mdev->ov_start_sector,
2218                 .stop_sector = ULLONG_MAX,
2219         };
2220
2221         if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2222                 reply->ret_code = ERR_MANDATORY_TAG;
2223                 return 0;
2224         }
2225
2226         /* If there is still bitmap IO pending, e.g. previous resync or verify
2227          * just being finished, wait for it before requesting a new resync. */
2228         drbd_suspend_io(mdev);
2229         wait_event(mdev->misc_wait, !drbd_test_flag(mdev, BITMAP_IO));
2230
2231         /* w_make_ov_request expects start position to be aligned */
2232         mdev->ov_start_sector = args.start_sector & ~(BM_SECT_PER_BIT-1);
2233         mdev->ov_stop_sector = args.stop_sector;
2234         reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2235         drbd_resume_io(mdev);
2236         return 0;
2237 }
2238
2239
2240 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2241                               struct drbd_nl_cfg_reply *reply)
2242 {
2243         int retcode = NO_ERROR;
2244         int skip_initial_sync = 0;
2245         int err;
2246
2247         struct new_c_uuid args;
2248
2249         memset(&args, 0, sizeof(struct new_c_uuid));
2250         if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2251                 reply->ret_code = ERR_MANDATORY_TAG;
2252                 return 0;
2253         }
2254
2255         mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
2256
2257         if (!get_ldev(mdev)) {
2258                 retcode = ERR_NO_DISK;
2259                 goto out;
2260         }
2261
2262         /* this is "skip initial sync", assume to be clean */
2263         if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
2264             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2265                 dev_info(DEV, "Preparing to skip initial sync\n");
2266                 skip_initial_sync = 1;
2267         } else if (mdev->state.conn != C_STANDALONE) {
2268                 retcode = ERR_CONNECTED;
2269                 goto out_dec;
2270         }
2271
2272         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2273         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2274
2275         if (args.clear_bm) {
2276                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2277                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2278                 if (err) {
2279                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2280                         retcode = ERR_IO_MD_DISK;
2281                 }
2282                 if (skip_initial_sync) {
2283                         drbd_send_uuids_skip_initial_sync(mdev);
2284                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2285                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2286                         spin_lock_irq(&mdev->req_lock);
2287                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2288                                         CS_VERBOSE, NULL);
2289                         spin_unlock_irq(&mdev->req_lock);
2290                 }
2291         }
2292
2293         drbd_md_sync(mdev);
2294 out_dec:
2295         put_ldev(mdev);
2296 out:
2297         mutex_unlock(&mdev->state_mutex);
2298
2299         reply->ret_code = retcode;
2300         return 0;
2301 }
2302
2303 struct cn_handler_struct {
2304         int (*function)(struct drbd_conf *,
2305                          struct drbd_nl_cfg_req *,
2306                          struct drbd_nl_cfg_reply *);
2307         int reply_body_size;
2308 };
2309
2310 static struct cn_handler_struct cnd_table[] = {
2311         [ P_primary ]           = { &drbd_nl_primary,           0 },
2312         [ P_secondary ]         = { &drbd_nl_secondary,         0 },
2313         [ P_disk_conf ]         = { &drbd_nl_disk_conf,         0 },
2314         [ P_detach ]            = { &drbd_nl_detach,            0 },
2315         [ P_net_conf ]          = { &drbd_nl_net_conf,          0 },
2316         [ P_disconnect ]        = { &drbd_nl_disconnect,        0 },
2317         [ P_resize ]            = { &drbd_nl_resize,            0 },
2318         [ P_syncer_conf ]       = { &drbd_nl_syncer_conf,       0 },
2319         [ P_invalidate ]        = { &drbd_nl_invalidate,        0 },
2320         [ P_invalidate_peer ]   = { &drbd_nl_invalidate_peer,   0 },
2321         [ P_pause_sync ]        = { &drbd_nl_pause_sync,        0 },
2322         [ P_resume_sync ]       = { &drbd_nl_resume_sync,       0 },
2323         [ P_suspend_io ]        = { &drbd_nl_suspend_io,        0 },
2324         [ P_resume_io ]         = { &drbd_nl_resume_io,         0 },
2325         [ P_outdate ]           = { &drbd_nl_outdate,           0 },
2326         [ P_get_config ]        = { &drbd_nl_get_config,
2327                                     sizeof(struct syncer_conf_tag_len_struct) +
2328                                     sizeof(struct disk_conf_tag_len_struct) +
2329                                     sizeof(struct net_conf_tag_len_struct) },
2330         [ P_get_state ]         = { &drbd_nl_get_state,
2331                                     sizeof(struct get_state_tag_len_struct) +
2332                                     sizeof(struct sync_progress_tag_len_struct) },
2333         [ P_get_uuids ]         = { &drbd_nl_get_uuids,
2334                                     sizeof(struct get_uuids_tag_len_struct) },
2335         [ P_get_timeout_flag ]  = { &drbd_nl_get_timeout_flag,
2336                                     sizeof(struct get_timeout_flag_tag_len_struct)},
2337         [ P_start_ov ]          = { &drbd_nl_start_ov,          0 },
2338         [ P_new_c_uuid ]        = { &drbd_nl_new_c_uuid,        0 },
2339 };
2340
2341 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2342 {
2343         struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2344         struct cn_handler_struct *cm;
2345         struct cn_msg *cn_reply;
2346         struct drbd_nl_cfg_reply *reply;
2347         struct drbd_conf *mdev;
2348         int retcode, rr;
2349         int reply_size = sizeof(struct cn_msg)
2350                 + sizeof(struct drbd_nl_cfg_reply)
2351                 + sizeof(short int);
2352
2353         if (!try_module_get(THIS_MODULE)) {
2354                 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2355                 return;
2356         }
2357
2358         if (!capable(CAP_SYS_ADMIN)) {
2359                 retcode = ERR_PERM;
2360                 goto fail;
2361         }
2362
2363         mdev = ensure_mdev(nlp->drbd_minor,
2364                         (nlp->flags & DRBD_NL_CREATE_DEVICE));
2365         if (!mdev) {
2366                 retcode = ERR_MINOR_INVALID;
2367                 goto fail;
2368         }
2369
2370         if (nlp->packet_type >= P_nl_after_last_packet ||
2371             nlp->packet_type == P_return_code_only) {
2372                 retcode = ERR_PACKET_NR;
2373                 goto fail;
2374         }
2375
2376         cm = cnd_table + nlp->packet_type;
2377
2378         /* This may happen if packet number is 0: */
2379         if (cm->function == NULL) {
2380                 retcode = ERR_PACKET_NR;
2381                 goto fail;
2382         }
2383
2384         reply_size += cm->reply_body_size;
2385
2386         /* allocation not in the IO path, cqueue thread context */
2387         cn_reply = kzalloc(reply_size, GFP_KERNEL);
2388         if (!cn_reply) {
2389                 retcode = ERR_NOMEM;
2390                 goto fail;
2391         }
2392         reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2393
2394         reply->packet_type =
2395                 cm->reply_body_size ? nlp->packet_type : P_return_code_only;
2396         reply->minor = nlp->drbd_minor;
2397         reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2398         /* reply->tag_list; might be modified by cm->function. */
2399
2400         rr = cm->function(mdev, nlp, reply);
2401
2402         cn_reply->id = req->id;
2403         cn_reply->seq = req->seq;
2404         cn_reply->ack = req->ack  + 1;
2405         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2406         cn_reply->flags = 0;
2407
2408         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2409         if (rr && rr != -ESRCH)
2410                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2411
2412         kfree(cn_reply);
2413         module_put(THIS_MODULE);
2414         return;
2415  fail:
2416         drbd_nl_send_reply(req, retcode);
2417         module_put(THIS_MODULE);
2418 }
2419
2420 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2421
2422 static unsigned short *
2423 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2424         unsigned short len, int nul_terminated)
2425 {
2426         unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2427         len = (len < l) ? len :  l;
2428         put_unaligned(tag, tl++);
2429         put_unaligned(len, tl++);
2430         memcpy(tl, data, len);
2431         tl = (unsigned short*)((char*)tl + len);
2432         if (nul_terminated)
2433                 *((char*)tl - 1) = 0;
2434         return tl;
2435 }
2436
2437 static unsigned short *
2438 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2439 {
2440         return __tl_add_blob(tl, tag, data, len, 0);
2441 }
2442
2443 static unsigned short *
2444 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2445 {
2446         return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2447 }
2448
2449 static unsigned short *
2450 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2451 {
2452         put_unaligned(tag, tl++);
2453         switch(tag_type(tag)) {
2454         case TT_INTEGER:
2455                 put_unaligned(sizeof(int), tl++);
2456                 put_unaligned(*(int *)val, (int *)tl);
2457                 tl = (unsigned short*)((char*)tl+sizeof(int));
2458                 break;
2459         case TT_INT64:
2460                 put_unaligned(sizeof(u64), tl++);
2461                 put_unaligned(*(u64 *)val, (u64 *)tl);
2462                 tl = (unsigned short*)((char*)tl+sizeof(u64));
2463                 break;
2464         default:
2465                 /* someone did something stupid. */
2466                 ;
2467         }
2468         return tl;
2469 }
2470
2471 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2472 {
2473         char buffer[sizeof(struct cn_msg)+
2474                     sizeof(struct drbd_nl_cfg_reply)+
2475                     sizeof(struct get_state_tag_len_struct)+
2476                     sizeof(short int)];
2477         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2478         struct drbd_nl_cfg_reply *reply =
2479                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2480         unsigned short *tl = reply->tag_list;
2481
2482         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2483
2484         tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2485
2486         put_unaligned(TT_END, tl++); /* Close the tag list */
2487
2488         cn_reply->id.idx = CN_IDX_DRBD;
2489         cn_reply->id.val = CN_VAL_DRBD;
2490
2491         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2492         cn_reply->ack = 0; /* not used here. */
2493         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2494                 (int)((char *)tl - (char *)reply->tag_list);
2495         cn_reply->flags = 0;
2496
2497         reply->packet_type = P_get_state;
2498         reply->minor = mdev_to_minor(mdev);
2499         reply->ret_code = NO_ERROR;
2500
2501         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2502 }
2503
2504 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2505 {
2506         char buffer[sizeof(struct cn_msg)+
2507                     sizeof(struct drbd_nl_cfg_reply)+
2508                     sizeof(struct call_helper_tag_len_struct)+
2509                     sizeof(short int)];
2510         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2511         struct drbd_nl_cfg_reply *reply =
2512                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2513         unsigned short *tl = reply->tag_list;
2514
2515         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2516
2517         tl = tl_add_str(tl, T_helper, helper_name);
2518         put_unaligned(TT_END, tl++); /* Close the tag list */
2519
2520         cn_reply->id.idx = CN_IDX_DRBD;
2521         cn_reply->id.val = CN_VAL_DRBD;
2522
2523         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2524         cn_reply->ack = 0; /* not used here. */
2525         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2526                 (int)((char *)tl - (char *)reply->tag_list);
2527         cn_reply->flags = 0;
2528
2529         reply->packet_type = P_call_helper;
2530         reply->minor = mdev_to_minor(mdev);
2531         reply->ret_code = NO_ERROR;
2532
2533         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2534 }
2535
2536 void drbd_bcast_ee(struct drbd_conf *mdev,
2537                 const char *reason, const int dgs,
2538                 const char* seen_hash, const char* calc_hash,
2539                 const struct drbd_epoch_entry* e)
2540 {
2541         struct cn_msg *cn_reply;
2542         struct drbd_nl_cfg_reply *reply;
2543         unsigned short *tl;
2544         struct page *page;
2545         unsigned len;
2546
2547         if (!e)
2548                 return;
2549         if (!reason || !reason[0])
2550                 return;
2551
2552         /* apparently we have to memcpy twice, first to prepare the data for the
2553          * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2554          * netlink skb. */
2555         /* receiver thread context, which is not in the writeout path (of this node),
2556          * but may be in the writeout path of the _other_ node.
2557          * GFP_NOIO to avoid potential "distributed deadlock". */
2558         cn_reply = kzalloc(
2559                 sizeof(struct cn_msg)+
2560                 sizeof(struct drbd_nl_cfg_reply)+
2561                 sizeof(struct dump_ee_tag_len_struct)+
2562                 sizeof(short int),
2563                 GFP_NOIO);
2564
2565         if (!cn_reply) {
2566                 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2567                                 (unsigned long long)e->sector, e->size);
2568                 return;
2569         }
2570
2571         reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2572         tl = reply->tag_list;
2573
2574         tl = tl_add_str(tl, T_dump_ee_reason, reason);
2575         tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2576         tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2577         tl = tl_add_int(tl, T_ee_sector, &e->sector);
2578         tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2579
2580         /* dump the first 32k */
2581         len = min_t(unsigned, e->size, 32 << 10);
2582         put_unaligned(T_ee_data, tl++);
2583         put_unaligned(len, tl++);
2584
2585         page = e->pages;
2586         page_chain_for_each(page) {
2587                 void *d = kmap_atomic(page);
2588                 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2589                 memcpy(tl, d, l);
2590                 kunmap_atomic(d);
2591                 tl = (unsigned short*)((char*)tl + l);
2592                 len -= l;
2593                 if (len == 0)
2594                         break;
2595         }
2596         put_unaligned(TT_END, tl++); /* Close the tag list */
2597
2598         cn_reply->id.idx = CN_IDX_DRBD;
2599         cn_reply->id.val = CN_VAL_DRBD;
2600
2601         cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2602         cn_reply->ack = 0; // not used here.
2603         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2604                 (int)((char*)tl - (char*)reply->tag_list);
2605         cn_reply->flags = 0;
2606
2607         reply->packet_type = P_dump_ee;
2608         reply->minor = mdev_to_minor(mdev);
2609         reply->ret_code = NO_ERROR;
2610
2611         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2612         kfree(cn_reply);
2613 }
2614
2615 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2616 {
2617         char buffer[sizeof(struct cn_msg)+
2618                     sizeof(struct drbd_nl_cfg_reply)+
2619                     sizeof(struct sync_progress_tag_len_struct)+
2620                     sizeof(short int)];
2621         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2622         struct drbd_nl_cfg_reply *reply =
2623                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2624         unsigned short *tl = reply->tag_list;
2625         unsigned long rs_left;
2626         unsigned int res;
2627
2628         /* no local ref, no bitmap, no syncer progress, no broadcast. */
2629         if (!get_ldev(mdev))
2630                 return;
2631         drbd_get_syncer_progress(mdev, &rs_left, &res);
2632         put_ldev(mdev);
2633
2634         tl = tl_add_int(tl, T_sync_progress, &res);
2635         put_unaligned(TT_END, tl++); /* Close the tag list */
2636
2637         cn_reply->id.idx = CN_IDX_DRBD;
2638         cn_reply->id.val = CN_VAL_DRBD;
2639
2640         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2641         cn_reply->ack = 0; /* not used here. */
2642         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2643                 (int)((char *)tl - (char *)reply->tag_list);
2644         cn_reply->flags = 0;
2645
2646         reply->packet_type = P_sync_progress;
2647         reply->minor = mdev_to_minor(mdev);
2648         reply->ret_code = NO_ERROR;
2649
2650         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2651 }
2652
2653 int __init drbd_nl_init(void)
2654 {
2655         static struct cb_id cn_id_drbd;
2656         int err, try=10;
2657
2658         cn_id_drbd.val = CN_VAL_DRBD;
2659         do {
2660                 cn_id_drbd.idx = cn_idx;
2661                 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2662                 if (!err)
2663                         break;
2664                 cn_idx = (cn_idx + CN_IDX_STEP);
2665         } while (try--);
2666
2667         if (err) {
2668                 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2669                 return err;
2670         }
2671
2672         return 0;
2673 }
2674
2675 void drbd_nl_cleanup(void)
2676 {
2677         static struct cb_id cn_id_drbd;
2678
2679         cn_id_drbd.idx = cn_idx;
2680         cn_id_drbd.val = CN_VAL_DRBD;
2681
2682         cn_del_callback(&cn_id_drbd);
2683 }
2684
2685 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2686 {
2687         char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2688         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2689         struct drbd_nl_cfg_reply *reply =
2690                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2691         int rr;
2692
2693         memset(buffer, 0, sizeof(buffer));
2694         cn_reply->id = req->id;
2695
2696         cn_reply->seq = req->seq;
2697         cn_reply->ack = req->ack  + 1;
2698         cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2699         cn_reply->flags = 0;
2700
2701         reply->packet_type = P_return_code_only;
2702         reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2703         reply->ret_code = ret_code;
2704
2705         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2706         if (rr && rr != -ESRCH)
2707                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2708 }
2709