drivers/block/drbd/drbd_nl.c

   1 /*
   2    drbd_nl.c
   3
   4    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
   5
   6    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
   7    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
   8    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
   9
  10    drbd is free software; you can redistribute it and/or modify
  11    it under the terms of the GNU General Public License as published by
  12    the Free Software Foundation; either version 2, or (at your option)
  13    any later version.
  14
  15    drbd is distributed in the hope that it will be useful,
  16    but WITHOUT ANY WARRANTY; without even the implied warranty of
  17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18    GNU General Public License for more details.
  19
  20    You should have received a copy of the GNU General Public License
  21    along with drbd; see the file COPYING.  If not, write to
  22    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  23
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/drbd.h>
  28 #include <linux/in.h>
  29 #include <linux/fs.h>
  30 #include <linux/file.h>
  31 #include <linux/slab.h>
  32 #include <linux/connector.h>
  33 #include <linux/blkpg.h>
  34 #include <linux/cpumask.h>
  35 #include "drbd_int.h"
  36 #include "drbd_req.h"
  37 #include "drbd_wrappers.h"
  38 #include <asm/unaligned.h>
  39 #include <linux/drbd_tag_magic.h>
  40 #include <linux/drbd_limits.h>
  41 #include <linux/compiler.h>
  42 #include <linux/kthread.h>
  43
  44 static unsigned short *tl_add_blob(unsigned short *, enum drbd_tags, const void *, int);
  45 static unsigned short *tl_add_str(unsigned short *, enum drbd_tags, const char *);
  46 static unsigned short *tl_add_int(unsigned short *, enum drbd_tags, const void *);
  47
  48 /* see get_sb_bdev and bd_claim */
  49 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
  50
  51 /* Generate the tag_list to struct functions */
  52 #define NL_PACKET(name, number, fields) \
  53 static int name ## _from_tags(struct drbd_conf *mdev, \
  54         unsigned short *tags, struct name *arg) __attribute__ ((unused)); \
  55 static int name ## _from_tags(struct drbd_conf *mdev, \
  56         unsigned short *tags, struct name *arg) \
  57 { \
  58         int tag; \
  59         int dlen; \
  60         \
  61         while ((tag = get_unaligned(tags++)) != TT_END) {       \
  62                 dlen = get_unaligned(tags++);                   \
  63                 switch (tag_number(tag)) { \
  64                 fields \
  65                 default: \
  66                         if (tag & T_MANDATORY) { \
  67                                 dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \
  68                                 return 0; \
  69                         } \
  70                 } \
  71                 tags = (unsigned short *)((char *)tags + dlen); \
  72         } \
  73         return 1; \
  74 }
  75 #define NL_INTEGER(pn, pr, member) \
  76         case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \
  77                 arg->member = get_unaligned((int *)(tags));     \
  78                 break;
  79 #define NL_INT64(pn, pr, member) \
  80         case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \
  81                 arg->member = get_unaligned((u64 *)(tags));     \
  82                 break;
  83 #define NL_BIT(pn, pr, member) \
  84         case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \
  85                 arg->member = *(char *)(tags) ? 1 : 0; \
  86                 break;
  87 #define NL_STRING(pn, pr, member, len) \
  88         case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \
  89                 if (dlen > len) { \
  90                         dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \
  91                                 #member, dlen, (unsigned int)len); \
  92                         return 0; \
  93                 } \
  94                  arg->member ## _len = dlen; \
  95                  memcpy(arg->member, tags, min_t(size_t, dlen, len)); \
  96                  break;
  97 #include "linux/drbd_nl.h"
  98
  99 /* Generate the struct to tag_list functions */
 100 #define NL_PACKET(name, number, fields) \
 101 static unsigned short* \
 102 name ## _to_tags(struct drbd_conf *mdev, \
 103         struct name *arg, unsigned short *tags) __attribute__ ((unused)); \
 104 static unsigned short* \
 105 name ## _to_tags(struct drbd_conf *mdev, \
 106         struct name *arg, unsigned short *tags) \
 107 { \
 108         fields \
 109         return tags; \
 110 }
 111
 112 #define NL_INTEGER(pn, pr, member) \
 113         put_unaligned(pn | pr | TT_INTEGER, tags++);    \
 114         put_unaligned(sizeof(int), tags++);             \
 115         put_unaligned(arg->member, (int *)tags);        \
 116         tags = (unsigned short *)((char *)tags+sizeof(int));
 117 #define NL_INT64(pn, pr, member) \
 118         put_unaligned(pn | pr | TT_INT64, tags++);      \
 119         put_unaligned(sizeof(u64), tags++);             \
 120         put_unaligned(arg->member, (u64 *)tags);        \
 121         tags = (unsigned short *)((char *)tags+sizeof(u64));
 122 #define NL_BIT(pn, pr, member) \
 123         put_unaligned(pn | pr | TT_BIT, tags++);        \
 124         put_unaligned(sizeof(char), tags++);            \
 125         *(char *)tags = arg->member; \
 126         tags = (unsigned short *)((char *)tags+sizeof(char));
 127 #define NL_STRING(pn, pr, member, len) \
 128         put_unaligned(pn | pr | TT_STRING, tags++);     \
 129         put_unaligned(arg->member ## _len, tags++);     \
 130         memcpy(tags, arg->member, arg->member ## _len); \
 131         tags = (unsigned short *)((char *)tags + arg->member ## _len);
 132 #include "linux/drbd_nl.h"
 133
 134 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name);
 135 void drbd_nl_send_reply(struct cn_msg *, int);
 136
 137 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
 138 {
 139         char *envp[] = { "HOME=/",
 140                         "TERM=linux",
 141                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 142                         NULL, /* Will be set to address family */
 143                         NULL, /* Will be set to address */
 144                         NULL };
 145
 146         char mb[12], af[20], ad[60], *afs;
 147         char *argv[] = {usermode_helper, cmd, mb, NULL };
 148         int ret;
 149
 150         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 151
 152         if (get_net_conf(mdev->tconn)) {
 153                 switch (((struct sockaddr *)mdev->tconn->net_conf->peer_addr)->sa_family) {
 154                 case AF_INET6:
 155                         afs = "ipv6";
 156                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6",
 157                                  &((struct sockaddr_in6 *)mdev->tconn->net_conf->peer_addr)->sin6_addr);
 158                         break;
 159                 case AF_INET:
 160                         afs = "ipv4";
 161                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 162                                  &((struct sockaddr_in *)mdev->tconn->net_conf->peer_addr)->sin_addr);
 163                         break;
 164                 default:
 165                         afs = "ssocks";
 166                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 167                                  &((struct sockaddr_in *)mdev->tconn->net_conf->peer_addr)->sin_addr);
 168                 }
 169                 snprintf(af, 20, "DRBD_PEER_AF=%s", afs);
 170                 envp[3]=af;
 171                 envp[4]=ad;
 172                 put_net_conf(mdev->tconn);
 173         }
 174
 175         /* The helper may take some time.
 176          * write out any unsynced meta data changes now */
 177         drbd_md_sync(mdev);
 178
 179         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 180
 181         drbd_bcast_ev_helper(mdev, cmd);
 182         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 183         if (ret)
 184                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 185                                 usermode_helper, cmd, mb,
 186                                 (ret >> 8) & 0xff, ret);
 187         else
 188                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 189                                 usermode_helper, cmd, mb,
 190                                 (ret >> 8) & 0xff, ret);
 191
 192         if (ret < 0) /* Ignore any ERRNOs we got. */
 193                 ret = 0;
 194
 195         return ret;
 196 }
 197
 198 enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
 199 {
 200         char *ex_to_string;
 201         int r;
 202         enum drbd_disk_state nps;
 203         enum drbd_fencing_p fp;
 204
 205         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 206
 207         if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 208                 fp = mdev->ldev->dc.fencing;
 209                 put_ldev(mdev);
 210         } else {
 211                 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n");
 212                 nps = mdev->state.pdsk;
 213                 goto out;
 214         }
 215
 216         r = drbd_khelper(mdev, "fence-peer");
 217
 218         switch ((r>>8) & 0xff) {
 219         case 3: /* peer is inconsistent */
 220                 ex_to_string = "peer is inconsistent or worse";
 221                 nps = D_INCONSISTENT;
 222                 break;
 223         case 4: /* peer got outdated, or was already outdated */
 224                 ex_to_string = "peer was fenced";
 225                 nps = D_OUTDATED;
 226                 break;
 227         case 5: /* peer was down */
 228                 if (mdev->state.disk == D_UP_TO_DATE) {
 229                         /* we will(have) create(d) a new UUID anyways... */
 230                         ex_to_string = "peer is unreachable, assumed to be dead";
 231                         nps = D_OUTDATED;
 232                 } else {
 233                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 234                         nps = mdev->state.pdsk;
 235                 }
 236                 break;
 237         case 6: /* Peer is primary, voluntarily outdate myself.
 238                  * This is useful when an unconnected R_SECONDARY is asked to
 239                  * become R_PRIMARY, but finds the other peer being active. */
 240                 ex_to_string = "peer is active";
 241                 dev_warn(DEV, "Peer is primary, outdating myself.\n");
 242                 nps = D_UNKNOWN;
 243                 _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE);
 244                 break;
 245         case 7:
 246                 if (fp != FP_STONITH)
 247                         dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n");
 248                 ex_to_string = "peer was stonithed";
 249                 nps = D_OUTDATED;
 250                 break;
 251         default:
 252                 /* The script is broken ... */
 253                 nps = D_UNKNOWN;
 254                 dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 255                 return nps;
 256         }
 257
 258         dev_info(DEV, "fence-peer helper returned %d (%s)\n",
 259                         (r>>8) & 0xff, ex_to_string);
 260
 261 out:
 262         if (mdev->state.susp_fen && nps >= D_UNKNOWN) {
 263                 /* The handler was not successful... unfreeze here, the
 264                    state engine can not unfreeze... */
 265                 _drbd_request_state(mdev, NS(susp_fen, 0), CS_VERBOSE);
 266         }
 267
 268         return nps;
 269 }
 270
 271 static int _try_outdate_peer_async(void *data)
 272 {
 273         struct drbd_conf *mdev = (struct drbd_conf *)data;
 274         enum drbd_disk_state nps;
 275         union drbd_state ns;
 276
 277         nps = drbd_try_outdate_peer(mdev);
 278
 279         /* Not using
 280            drbd_request_state(mdev, NS(pdsk, nps));
 281            here, because we might were able to re-establish the connection
 282            in the meantime. This can only partially be solved in the state's
 283            engine is_valid_state() and is_valid_state_transition()
 284            functions.
 285
 286            nps can be D_INCONSISTENT, D_OUTDATED or D_UNKNOWN.
 287            pdsk == D_INCONSISTENT while conn >= C_CONNECTED is valid,
 288            therefore we have to have the pre state change check here.
 289         */
 290         spin_lock_irq(&mdev->tconn->req_lock);
 291         ns = mdev->state;
 292         if (ns.conn < C_WF_REPORT_PARAMS) {
 293                 ns.pdsk = nps;
 294                 _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
 295         }
 296         spin_unlock_irq(&mdev->tconn->req_lock);
 297
 298         return 0;
 299 }
 300
 301 void drbd_try_outdate_peer_async(struct drbd_conf *mdev)
 302 {
 303         struct task_struct *opa;
 304
 305         opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev));
 306         if (IS_ERR(opa))
 307                 dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n");
 308 }
 309
 310 enum drbd_state_rv
 311 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 312 {
 313         const int max_tries = 4;
 314         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
 315         int try = 0;
 316         int forced = 0;
 317         union drbd_state mask, val;
 318         enum drbd_disk_state nps;
 319
 320         if (new_role == R_PRIMARY)
 321                 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
 322
 323         mutex_lock(mdev->state_mutex);
 324
 325         mask.i = 0; mask.role = R_MASK;
 326         val.i  = 0; val.role  = new_role;
 327
 328         while (try++ < max_tries) {
 329                 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 330
 331                 /* in case we first succeeded to outdate,
 332                  * but now suddenly could establish a connection */
 333                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 334                         val.pdsk = 0;
 335                         mask.pdsk = 0;
 336                         continue;
 337                 }
 338
 339                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
 340                     (mdev->state.disk < D_UP_TO_DATE &&
 341                      mdev->state.disk >= D_INCONSISTENT)) {
 342                         mask.disk = D_MASK;
 343                         val.disk  = D_UP_TO_DATE;
 344                         forced = 1;
 345                         continue;
 346                 }
 347
 348                 if (rv == SS_NO_UP_TO_DATE_DISK &&
 349                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 350                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 351                         nps = drbd_try_outdate_peer(mdev);
 352
 353                         if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
 354                                 val.disk = D_UP_TO_DATE;
 355                                 mask.disk = D_MASK;
 356                         }
 357
 358                         val.pdsk = nps;
 359                         mask.pdsk = D_MASK;
 360
 361                         continue;
 362                 }
 363
 364                 if (rv == SS_NOTHING_TO_DO)
 365                         goto fail;
 366                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
 367                         nps = drbd_try_outdate_peer(mdev);
 368
 369                         if (force && nps > D_OUTDATED) {
 370                                 dev_warn(DEV, "Forced into split brain situation!\n");
 371                                 nps = D_OUTDATED;
 372                         }
 373
 374                         mask.pdsk = D_MASK;
 375                         val.pdsk  = nps;
 376
 377                         continue;
 378                 }
 379                 if (rv == SS_TWO_PRIMARIES) {
 380                         /* Maybe the peer is detected as dead very soon...
 381                            retry at most once more in this case. */
 382                         schedule_timeout_interruptible((mdev->tconn->net_conf->ping_timeo+1)*HZ/10);
 383                         if (try < max_tries)
 384                                 try = max_tries - 1;
 385                         continue;
 386                 }
 387                 if (rv < SS_SUCCESS) {
 388                         rv = _drbd_request_state(mdev, mask, val,
 389                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 390                         if (rv < SS_SUCCESS)
 391                                 goto fail;
 392                 }
 393                 break;
 394         }
 395
 396         if (rv < SS_SUCCESS)
 397                 goto fail;
 398
 399         if (forced)
 400                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 401
 402         /* Wait until nothing is on the fly :) */
 403         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 404
 405         if (new_role == R_SECONDARY) {
 406                 set_disk_ro(mdev->vdisk, true);
 407                 if (get_ldev(mdev)) {
 408                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 409                         put_ldev(mdev);
 410                 }
 411         } else {
 412                 if (get_net_conf(mdev->tconn)) {
 413                         mdev->tconn->net_conf->want_lose = 0;
 414                         put_net_conf(mdev->tconn);
 415                 }
 416                 set_disk_ro(mdev->vdisk, false);
 417                 if (get_ldev(mdev)) {
 418                         if (((mdev->state.conn < C_CONNECTED ||
 419                                mdev->state.pdsk <= D_FAILED)
 420                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 421                                 drbd_uuid_new_current(mdev);
 422
 423                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 424                         put_ldev(mdev);
 425                 }
 426         }
 427
 428         /* writeout of activity log covered areas of the bitmap
 429          * to stable storage done in after state change already */
 430
 431         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 432                 /* if this was forced, we should consider sync */
 433                 if (forced)
 434                         drbd_send_uuids(mdev);
 435                 drbd_send_state(mdev);
 436         }
 437
 438         drbd_md_sync(mdev);
 439
 440         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 441  fail:
 442         mutex_unlock(mdev->state_mutex);
 443         return rv;
 444 }
 445
 446 static struct drbd_conf *ensure_mdev(int minor, int create)
 447 {
 448         struct drbd_conf *mdev;
 449
 450         if (minor >= minor_count)
 451                 return NULL;
 452
 453         mdev = minor_to_mdev(minor);
 454
 455         if (!mdev && create) {
 456                 struct gendisk *disk = NULL;
 457                 mdev = drbd_new_device(minor);
 458
 459                 spin_lock_irq(&drbd_pp_lock);
 460                 if (minor_table[minor] == NULL) {
 461                         minor_table[minor] = mdev;
 462                         disk = mdev->vdisk;
 463                         mdev = NULL;
 464                 } /* else: we lost the race */
 465                 spin_unlock_irq(&drbd_pp_lock);
 466
 467                 if (disk) /* we won the race above */
 468                         /* in case we ever add a drbd_delete_device(),
 469                          * don't forget the del_gendisk! */
 470                         add_disk(disk);
 471                 else /* we lost the race above */
 472                         drbd_free_mdev(mdev);
 473
 474                 mdev = minor_to_mdev(minor);
 475         }
 476
 477         return mdev;
 478 }
 479
 480 static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 481                            struct drbd_nl_cfg_reply *reply)
 482 {
 483         struct primary primary_args;
 484
 485         memset(&primary_args, 0, sizeof(struct primary));
 486         if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) {
 487                 reply->ret_code = ERR_MANDATORY_TAG;
 488                 return 0;
 489         }
 490
 491         reply->ret_code =
 492                 drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force);
 493
 494         return 0;
 495 }
 496
 497 static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 498                              struct drbd_nl_cfg_reply *reply)
 499 {
 500         reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0);
 501
 502         return 0;
 503 }
 504
 505 /* initializes the md.*_offset members, so we are able to find
 506  * the on disk meta data */
 507 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 508                                        struct drbd_backing_dev *bdev)
 509 {
 510         sector_t md_size_sect = 0;
 511         switch (bdev->dc.meta_dev_idx) {
 512         default:
 513                 /* v07 style fixed size indexed meta data */
 514                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 515                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 516                 bdev->md.al_offset = MD_AL_OFFSET;
 517                 bdev->md.bm_offset = MD_BM_OFFSET;
 518                 break;
 519         case DRBD_MD_INDEX_FLEX_EXT:
 520                 /* just occupy the full device; unit: sectors */
 521                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 522                 bdev->md.md_offset = 0;
 523                 bdev->md.al_offset = MD_AL_OFFSET;
 524                 bdev->md.bm_offset = MD_BM_OFFSET;
 525                 break;
 526         case DRBD_MD_INDEX_INTERNAL:
 527         case DRBD_MD_INDEX_FLEX_INT:
 528                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 529                 /* al size is still fixed */
 530                 bdev->md.al_offset = -MD_AL_MAX_SIZE;
 531                 /* we need (slightly less than) ~ this much bitmap sectors: */
 532                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 533                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 534                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 535                 md_size_sect = ALIGN(md_size_sect, 8);
 536
 537                 /* plus the "drbd meta data super block",
 538                  * and the activity log; */
 539                 md_size_sect += MD_BM_OFFSET;
 540
 541                 bdev->md.md_size_sect = md_size_sect;
 542                 /* bitmap offset is adjusted by 'super' block size */
 543                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 544                 break;
 545         }
 546 }
 547
 548 /* input size is expected to be in KB */
 549 char *ppsize(char *buf, unsigned long long size)
 550 {
 551         /* Needs 9 bytes at max including trailing NUL:
 552          * -1ULL ==> "16384 EB" */
 553         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 554         int base = 0;
 555         while (size >= 10000 && base < sizeof(units)-1) {
 556                 /* shift + round */
 557                 size = (size >> 10) + !!(size & (1<<9));
 558                 base++;
 559         }
 560         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
 561
 562         return buf;
 563 }
 564
 565 /* there is still a theoretical deadlock when called from receiver
 566  * on an D_INCONSISTENT R_PRIMARY:
 567  *  remote READ does inc_ap_bio, receiver would need to receive answer
 568  *  packet from remote to dec_ap_bio again.
 569  *  receiver receive_sizes(), comes here,
 570  *  waits for ap_bio_cnt == 0. -> deadlock.
 571  * but this cannot happen, actually, because:
 572  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 573  *  (not connected, or bad/no disk on peer):
 574  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 575  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 576  *  peer may not initiate a resize.
 577  */
 578 void drbd_suspend_io(struct drbd_conf *mdev)
 579 {
 580         set_bit(SUSPEND_IO, &mdev->flags);
 581         if (is_susp(mdev->state))
 582                 return;
 583         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 584 }
 585
 586 void drbd_resume_io(struct drbd_conf *mdev)
 587 {
 588         clear_bit(SUSPEND_IO, &mdev->flags);
 589         wake_up(&mdev->misc_wait);
 590 }
 591
 592 /**
 593  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 594  * @mdev:       DRBD device.
 595  *
 596  * Returns 0 on success, negative return values indicate errors.
 597  * You should call drbd_md_sync() after calling this function.
 598  */
 599 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 600 {
 601         sector_t prev_first_sect, prev_size; /* previous meta location */
 602         sector_t la_size;
 603         sector_t size;
 604         char ppb[10];
 605
 606         int md_moved, la_size_changed;
 607         enum determine_dev_size rv = unchanged;
 608
 609         /* race:
 610          * application request passes inc_ap_bio,
 611          * but then cannot get an AL-reference.
 612          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 613          *
 614          * to avoid that:
 615          * Suspend IO right here.
 616          * still lock the act_log to not trigger ASSERTs there.
 617          */
 618         drbd_suspend_io(mdev);
 619
 620         /* no wait necessary anymore, actually we could assert that */
 621         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 622
 623         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 624         prev_size = mdev->ldev->md.md_size_sect;
 625         la_size = mdev->ldev->md.la_size_sect;
 626
 627         /* TODO: should only be some assert here, not (re)init... */
 628         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 629
 630         size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
 631
 632         if (drbd_get_capacity(mdev->this_bdev) != size ||
 633             drbd_bm_capacity(mdev) != size) {
 634                 int err;
 635                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 636                 if (unlikely(err)) {
 637                         /* currently there is only one error: ENOMEM! */
 638                         size = drbd_bm_capacity(mdev)>>1;
 639                         if (size == 0) {
 640                                 dev_err(DEV, "OUT OF MEMORY! "
 641                                     "Could not allocate bitmap!\n");
 642                         } else {
 643                                 dev_err(DEV, "BM resizing failed. "
 644                                     "Leaving size unchanged at size = %lu KB\n",
 645                                     (unsigned long)size);
 646                         }
 647                         rv = dev_size_error;
 648                 }
 649                 /* racy, see comments above. */
 650                 drbd_set_my_capacity(mdev, size);
 651                 mdev->ldev->md.la_size_sect = size;
 652                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 653                      (unsigned long long)size>>1);
 654         }
 655         if (rv == dev_size_error)
 656                 goto out;
 657
 658         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 659
 660         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 661                 || prev_size       != mdev->ldev->md.md_size_sect;
 662
 663         if (la_size_changed || md_moved) {
 664                 int err;
 665
 666                 drbd_al_shrink(mdev); /* All extents inactive. */
 667                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 668                          la_size_changed && md_moved ? "size changed and md moved" :
 669                          la_size_changed ? "size changed" : "md moved");
 670                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
 671                 err = drbd_bitmap_io(mdev, &drbd_bm_write,
 672                                 "size changed", BM_LOCKED_MASK);
 673                 if (err) {
 674                         rv = dev_size_error;
 675                         goto out;
 676                 }
 677                 drbd_md_mark_dirty(mdev);
 678         }
 679
 680         if (size > la_size)
 681                 rv = grew;
 682         if (size < la_size)
 683                 rv = shrunk;
 684 out:
 685         lc_unlock(mdev->act_log);
 686         wake_up(&mdev->al_wait);
 687         drbd_resume_io(mdev);
 688
 689         return rv;
 690 }
 691
 692 sector_t
 693 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
 694 {
 695         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 696         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 697         sector_t m_size; /* my size */
 698         sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
 699         sector_t size = 0;
 700
 701         m_size = drbd_get_max_capacity(bdev);
 702
 703         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 704                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 705                 p_size = m_size;
 706         }
 707
 708         if (p_size && m_size) {
 709                 size = min_t(sector_t, p_size, m_size);
 710         } else {
 711                 if (la_size) {
 712                         size = la_size;
 713                         if (m_size && m_size < size)
 714                                 size = m_size;
 715                         if (p_size && p_size < size)
 716                                 size = p_size;
 717                 } else {
 718                         if (m_size)
 719                                 size = m_size;
 720                         if (p_size)
 721                                 size = p_size;
 722                 }
 723         }
 724
 725         if (size == 0)
 726                 dev_err(DEV, "Both nodes diskless!\n");
 727
 728         if (u_size) {
 729                 if (u_size > size)
 730                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 731                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 732                 else
 733                         size = u_size;
 734         }
 735
 736         return size;
 737 }
 738
 739 /**
 740  * drbd_check_al_size() - Ensures that the AL is of the right size
 741  * @mdev:       DRBD device.
 742  *
 743  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 744  * failed, and 0 on success. You should call drbd_md_sync() after you called
 745  * this function.
 746  */
 747 static int drbd_check_al_size(struct drbd_conf *mdev)
 748 {
 749         struct lru_cache *n, *t;
 750         struct lc_element *e;
 751         unsigned int in_use;
 752         int i;
 753
 754         if (!expect(mdev->sync_conf.al_extents >= 7))
 755                 mdev->sync_conf.al_extents = 127;
 756
 757         if (mdev->act_log &&
 758             mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
 759                 return 0;
 760
 761         in_use = 0;
 762         t = mdev->act_log;
 763         n = lc_create("act_log", drbd_al_ext_cache,
 764                 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
 765
 766         if (n == NULL) {
 767                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 768                 return -ENOMEM;
 769         }
 770         spin_lock_irq(&mdev->al_lock);
 771         if (t) {
 772                 for (i = 0; i < t->nr_elements; i++) {
 773                         e = lc_element_by_index(t, i);
 774                         if (e->refcnt)
 775                                 dev_err(DEV, "refcnt(%d)==%d\n",
 776                                     e->lc_number, e->refcnt);
 777                         in_use += e->refcnt;
 778                 }
 779         }
 780         if (!in_use)
 781                 mdev->act_log = n;
 782         spin_unlock_irq(&mdev->al_lock);
 783         if (in_use) {
 784                 dev_err(DEV, "Activity log still in use!\n");
 785                 lc_destroy(n);
 786                 return -EBUSY;
 787         } else {
 788                 if (t)
 789                         lc_destroy(t);
 790         }
 791         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 792         return 0;
 793 }
 794
 795 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
 796 {
 797         struct request_queue * const q = mdev->rq_queue;
 798         int max_hw_sectors = max_bio_size >> 9;
 799         int max_segments = 0;
 800
 801         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 802                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 803
 804                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
 805                 max_segments = mdev->ldev->dc.max_bio_bvecs;
 806                 put_ldev(mdev);
 807         }
 808
 809         blk_queue_logical_block_size(q, 512);
 810         blk_queue_max_hw_sectors(q, max_hw_sectors);
 811         /* This is the workaround for "bio would need to, but cannot, be split" */
 812         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
 813         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
 814
 815         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 816                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 817
 818                 blk_queue_stack_limits(q, b);
 819
 820                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
 821                         dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
 822                                  q->backing_dev_info.ra_pages,
 823                                  b->backing_dev_info.ra_pages);
 824                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
 825                 }
 826                 put_ldev(mdev);
 827         }
 828 }
 829
 830 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
 831 {
 832         int now, new, local, peer;
 833
 834         now = queue_max_hw_sectors(mdev->rq_queue) << 9;
 835         local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
 836         peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
 837
 838         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 839                 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
 840                 mdev->local_max_bio_size = local;
 841                 put_ldev(mdev);
 842         }
 843
 844         /* We may ignore peer limits if the peer is modern enough.
 845            Because new from 8.3.8 onwards the peer can use multiple
 846            BIOs for a single peer_request */
 847         if (mdev->state.conn >= C_CONNECTED) {
 848                 if (mdev->tconn->agreed_pro_version < 94)
 849                         peer = mdev->peer_max_bio_size;
 850                 else if (mdev->tconn->agreed_pro_version == 94)
 851                         peer = DRBD_MAX_SIZE_H80_PACKET;
 852                 else /* drbd 8.3.8 onwards */
 853                         peer = DRBD_MAX_BIO_SIZE;
 854         }
 855
 856         new = min_t(int, local, peer);
 857
 858         if (mdev->state.role == R_PRIMARY && new < now)
 859                 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
 860
 861         if (new != now)
 862                 dev_info(DEV, "max BIO size = %u\n", new);
 863
 864         drbd_setup_queue_param(mdev, new);
 865 }
 866
 867 /* serialize deconfig (worker exiting, doing cleanup)
 868  * and reconfig (drbdsetup disk, drbdsetup net)
 869  *
 870  * Wait for a potentially exiting worker, then restart it,
 871  * or start a new one.  Flush any pending work, there may still be an
 872  * after_state_change queued.
 873  */
 874 static void drbd_reconfig_start(struct drbd_conf *mdev)
 875 {
 876         wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
 877         wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
 878         drbd_thread_start(&mdev->tconn->worker);
 879         drbd_flush_workqueue(mdev);
 880 }
 881
 882 /* if still unconfigured, stops worker again.
 883  * if configured now, clears CONFIG_PENDING.
 884  * wakes potential waiters */
 885 static void drbd_reconfig_done(struct drbd_conf *mdev)
 886 {
 887         spin_lock_irq(&mdev->tconn->req_lock);
 888         if (mdev->state.disk == D_DISKLESS &&
 889             mdev->state.conn == C_STANDALONE &&
 890             mdev->state.role == R_SECONDARY) {
 891                 set_bit(DEVICE_DYING, &mdev->flags);
 892                 drbd_thread_stop_nowait(&mdev->tconn->worker);
 893         } else
 894                 clear_bit(CONFIG_PENDING, &mdev->flags);
 895         spin_unlock_irq(&mdev->tconn->req_lock);
 896         wake_up(&mdev->state_wait);
 897 }
 898
 899 /* Make sure IO is suspended before calling this function(). */
 900 static void drbd_suspend_al(struct drbd_conf *mdev)
 901 {
 902         int s = 0;
 903
 904         if (!lc_try_lock(mdev->act_log)) {
 905                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
 906                 return;
 907         }
 908
 909         drbd_al_shrink(mdev);
 910         spin_lock_irq(&mdev->tconn->req_lock);
 911         if (mdev->state.conn < C_CONNECTED)
 912                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
 913         spin_unlock_irq(&mdev->tconn->req_lock);
 914         lc_unlock(mdev->act_log);
 915
 916         if (s)
 917                 dev_info(DEV, "Suspended AL updates\n");
 918 }
 919
 920 /* does always return 0;
 921  * interesting return code is in reply->ret_code */
 922 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 923                              struct drbd_nl_cfg_reply *reply)
 924 {
 925         enum drbd_ret_code retcode;
 926         enum determine_dev_size dd;
 927         sector_t max_possible_sectors;
 928         sector_t min_md_device_sectors;
 929         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
 930         struct block_device *bdev;
 931         struct lru_cache *resync_lru = NULL;
 932         union drbd_state ns, os;
 933         enum drbd_state_rv rv;
 934         int cp_discovered = 0;
 935         int logical_block_size;
 936
 937         drbd_reconfig_start(mdev);
 938
 939         /* if you want to reconfigure, please tear down first */
 940         if (mdev->state.disk > D_DISKLESS) {
 941                 retcode = ERR_DISK_CONFIGURED;
 942                 goto fail;
 943         }
 944         /* It may just now have detached because of IO error.  Make sure
 945          * drbd_ldev_destroy is done already, we may end up here very fast,
 946          * e.g. if someone calls attach from the on-io-error handler,
 947          * to realize a "hot spare" feature (not that I'd recommend that) */
 948         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
 949
 950         /* allocation not in the IO path, cqueue thread context */
 951         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
 952         if (!nbc) {
 953                 retcode = ERR_NOMEM;
 954                 goto fail;
 955         }
 956
 957         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
 958         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
 959         nbc->dc.fencing       = DRBD_FENCING_DEF;
 960         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
 961
 962         if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
 963                 retcode = ERR_MANDATORY_TAG;
 964                 goto fail;
 965         }
 966
 967         if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
 968                 retcode = ERR_MD_IDX_INVALID;
 969                 goto fail;
 970         }
 971
 972         if (get_net_conf(mdev->tconn)) {
 973                 int prot = mdev->tconn->net_conf->wire_protocol;
 974                 put_net_conf(mdev->tconn);
 975                 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
 976                         retcode = ERR_STONITH_AND_PROT_A;
 977                         goto fail;
 978                 }
 979         }
 980
 981         bdev = blkdev_get_by_path(nbc->dc.backing_dev,
 982                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
 983         if (IS_ERR(bdev)) {
 984                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
 985                         PTR_ERR(bdev));
 986                 retcode = ERR_OPEN_DISK;
 987                 goto fail;
 988         }
 989         nbc->backing_bdev = bdev;
 990
 991         /*
 992          * meta_dev_idx >= 0: external fixed size, possibly multiple
 993          * drbd sharing one meta device.  TODO in that case, paranoia
 994          * check that [md_bdev, meta_dev_idx] is not yet used by some
 995          * other drbd minor!  (if you use drbd.conf + drbdadm, that
 996          * should check it for you already; but if you don't, or
 997          * someone fooled it, we need to double check here)
 998          */
 999         bdev = blkdev_get_by_path(nbc->dc.meta_dev,
1000                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1001                                   (nbc->dc.meta_dev_idx < 0) ?
1002                                   (void *)mdev : (void *)drbd_m_holder);
1003         if (IS_ERR(bdev)) {
1004                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
1005                         PTR_ERR(bdev));
1006                 retcode = ERR_OPEN_MD_DISK;
1007                 goto fail;
1008         }
1009         nbc->md_bdev = bdev;
1010
1011         if ((nbc->backing_bdev == nbc->md_bdev) !=
1012             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1013              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1014                 retcode = ERR_MD_IDX_INVALID;
1015                 goto fail;
1016         }
1017
1018         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1019                         61, sizeof(struct bm_extent),
1020                         offsetof(struct bm_extent, lce));
1021         if (!resync_lru) {
1022                 retcode = ERR_NOMEM;
1023                 goto fail;
1024         }
1025
1026         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1027         drbd_md_set_sector_offsets(mdev, nbc);
1028
1029         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
1030                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1031                         (unsigned long long) drbd_get_max_capacity(nbc),
1032                         (unsigned long long) nbc->dc.disk_size);
1033                 retcode = ERR_DISK_TO_SMALL;
1034                 goto fail;
1035         }
1036
1037         if (nbc->dc.meta_dev_idx < 0) {
1038                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1039                 /* at least one MB, otherwise it does not make sense */
1040                 min_md_device_sectors = (2<<10);
1041         } else {
1042                 max_possible_sectors = DRBD_MAX_SECTORS;
1043                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
1044         }
1045
1046         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1047                 retcode = ERR_MD_DISK_TO_SMALL;
1048                 dev_warn(DEV, "refusing attach: md-device too small, "
1049                      "at least %llu sectors needed for this meta-disk type\n",
1050                      (unsigned long long) min_md_device_sectors);
1051                 goto fail;
1052         }
1053
1054         /* Make sure the new disk is big enough
1055          * (we may currently be R_PRIMARY with no local disk...) */
1056         if (drbd_get_max_capacity(nbc) <
1057             drbd_get_capacity(mdev->this_bdev)) {
1058                 retcode = ERR_DISK_TO_SMALL;
1059                 goto fail;
1060         }
1061
1062         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1063
1064         if (nbc->known_size > max_possible_sectors) {
1065                 dev_warn(DEV, "==> truncating very big lower level device "
1066                         "to currently maximum possible %llu sectors <==\n",
1067                         (unsigned long long) max_possible_sectors);
1068                 if (nbc->dc.meta_dev_idx >= 0)
1069                         dev_warn(DEV, "==>> using internal or flexible "
1070                                       "meta data may help <<==\n");
1071         }
1072
1073         drbd_suspend_io(mdev);
1074         /* also wait for the last barrier ack. */
1075         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state));
1076         /* and for any other previously queued work */
1077         drbd_flush_workqueue(mdev);
1078
1079         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1080         retcode = rv;  /* FIXME: Type mismatch. */
1081         drbd_resume_io(mdev);
1082         if (rv < SS_SUCCESS)
1083                 goto fail;
1084
1085         if (!get_ldev_if_state(mdev, D_ATTACHING))
1086                 goto force_diskless;
1087
1088         drbd_md_set_sector_offsets(mdev, nbc);
1089
1090         /* allocate a second IO page if logical_block_size != 512 */
1091         logical_block_size = bdev_logical_block_size(nbc->md_bdev);
1092         if (logical_block_size == 0)
1093                 logical_block_size = MD_SECTOR_SIZE;
1094
1095         if (logical_block_size != MD_SECTOR_SIZE) {
1096                 if (!mdev->md_io_tmpp) {
1097                         struct page *page = alloc_page(GFP_NOIO);
1098                         if (!page)
1099                                 goto force_diskless_dec;
1100
1101                         dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
1102                              logical_block_size, MD_SECTOR_SIZE);
1103                         dev_warn(DEV, "Workaround engaged (has performance impact).\n");
1104
1105                         mdev->md_io_tmpp = page;
1106                 }
1107         }
1108
1109         if (!mdev->bitmap) {
1110                 if (drbd_bm_init(mdev)) {
1111                         retcode = ERR_NOMEM;
1112                         goto force_diskless_dec;
1113                 }
1114         }
1115
1116         retcode = drbd_md_read(mdev, nbc);
1117         if (retcode != NO_ERROR)
1118                 goto force_diskless_dec;
1119
1120         if (mdev->state.conn < C_CONNECTED &&
1121             mdev->state.role == R_PRIMARY &&
1122             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1123                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1124                     (unsigned long long)mdev->ed_uuid);
1125                 retcode = ERR_DATA_NOT_CURRENT;
1126                 goto force_diskless_dec;
1127         }
1128
1129         /* Since we are diskless, fix the activity log first... */
1130         if (drbd_check_al_size(mdev)) {
1131                 retcode = ERR_NOMEM;
1132                 goto force_diskless_dec;
1133         }
1134
1135         /* Prevent shrinking of consistent devices ! */
1136         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1137             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1138                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1139                 retcode = ERR_DISK_TO_SMALL;
1140                 goto force_diskless_dec;
1141         }
1142
1143         if (!drbd_al_read_log(mdev, nbc)) {
1144                 retcode = ERR_IO_MD_DISK;
1145                 goto force_diskless_dec;
1146         }
1147
1148         /* Reset the "barriers don't work" bits here, then force meta data to
1149          * be written, to ensure we determine if barriers are supported. */
1150         if (nbc->dc.no_md_flush)
1151                 set_bit(MD_NO_FUA, &mdev->flags);
1152         else
1153                 clear_bit(MD_NO_FUA, &mdev->flags);
1154
1155         /* Point of no return reached.
1156          * Devices and memory are no longer released by error cleanup below.
1157          * now mdev takes over responsibility, and the state engine should
1158          * clean it up somewhere.  */
1159         D_ASSERT(mdev->ldev == NULL);
1160         mdev->ldev = nbc;
1161         mdev->resync = resync_lru;
1162         nbc = NULL;
1163         resync_lru = NULL;
1164
1165         mdev->write_ordering = WO_bdev_flush;
1166         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1167
1168         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1169                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1170         else
1171                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1172
1173         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1174             !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) {
1175                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1176                 cp_discovered = 1;
1177         }
1178
1179         mdev->send_cnt = 0;
1180         mdev->recv_cnt = 0;
1181         mdev->read_cnt = 0;
1182         mdev->writ_cnt = 0;
1183
1184         drbd_reconsider_max_bio_size(mdev);
1185
1186         /* If I am currently not R_PRIMARY,
1187          * but meta data primary indicator is set,
1188          * I just now recover from a hard crash,
1189          * and have been R_PRIMARY before that crash.
1190          *
1191          * Now, if I had no connection before that crash
1192          * (have been degraded R_PRIMARY), chances are that
1193          * I won't find my peer now either.
1194          *
1195          * In that case, and _only_ in that case,
1196          * we use the degr-wfc-timeout instead of the default,
1197          * so we can automatically recover from a crash of a
1198          * degraded but active "cluster" after a certain timeout.
1199          */
1200         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1201         if (mdev->state.role != R_PRIMARY &&
1202              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1203             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1204                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1205
1206         dd = drbd_determine_dev_size(mdev, 0);
1207         if (dd == dev_size_error) {
1208                 retcode = ERR_NOMEM_BITMAP;
1209                 goto force_diskless_dec;
1210         } else if (dd == grew)
1211                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1212
1213         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1214                 dev_info(DEV, "Assuming that all blocks are out of sync "
1215                      "(aka FullSync)\n");
1216                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1217                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1218                         retcode = ERR_IO_MD_DISK;
1219                         goto force_diskless_dec;
1220                 }
1221         } else {
1222                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1223                         "read from attaching", BM_LOCKED_MASK) < 0) {
1224                         retcode = ERR_IO_MD_DISK;
1225                         goto force_diskless_dec;
1226                 }
1227         }
1228
1229         if (cp_discovered) {
1230                 drbd_al_apply_to_bm(mdev);
1231                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1232                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1233                         retcode = ERR_IO_MD_DISK;
1234                         goto force_diskless_dec;
1235                 }
1236         }
1237
1238         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1239                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1240
1241         spin_lock_irq(&mdev->tconn->req_lock);
1242         os = mdev->state;
1243         ns.i = os.i;
1244         /* If MDF_CONSISTENT is not set go into inconsistent state,
1245            otherwise investigate MDF_WasUpToDate...
1246            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1247            otherwise into D_CONSISTENT state.
1248         */
1249         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1250                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1251                         ns.disk = D_CONSISTENT;
1252                 else
1253                         ns.disk = D_OUTDATED;
1254         } else {
1255                 ns.disk = D_INCONSISTENT;
1256         }
1257
1258         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1259                 ns.pdsk = D_OUTDATED;
1260
1261         if ( ns.disk == D_CONSISTENT &&
1262             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1263                 ns.disk = D_UP_TO_DATE;
1264
1265         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1266            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1267            this point, because drbd_request_state() modifies these
1268            flags. */
1269
1270         /* In case we are C_CONNECTED postpone any decision on the new disk
1271            state after the negotiation phase. */
1272         if (mdev->state.conn == C_CONNECTED) {
1273                 mdev->new_state_tmp.i = ns.i;
1274                 ns.i = os.i;
1275                 ns.disk = D_NEGOTIATING;
1276
1277                 /* We expect to receive up-to-date UUIDs soon.
1278                    To avoid a race in receive_state, free p_uuid while
1279                    holding req_lock. I.e. atomic with the state change */
1280                 kfree(mdev->p_uuid);
1281                 mdev->p_uuid = NULL;
1282         }
1283
1284         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1285         ns = mdev->state;
1286         spin_unlock_irq(&mdev->tconn->req_lock);
1287
1288         if (rv < SS_SUCCESS)
1289                 goto force_diskless_dec;
1290
1291         if (mdev->state.role == R_PRIMARY)
1292                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1293         else
1294                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1295
1296         drbd_md_mark_dirty(mdev);
1297         drbd_md_sync(mdev);
1298
1299         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1300         put_ldev(mdev);
1301         reply->ret_code = retcode;
1302         drbd_reconfig_done(mdev);
1303         return 0;
1304
1305  force_diskless_dec:
1306         put_ldev(mdev);
1307  force_diskless:
1308         drbd_force_state(mdev, NS(disk, D_FAILED));
1309         drbd_md_sync(mdev);
1310  fail:
1311         if (nbc) {
1312                 if (nbc->backing_bdev)
1313                         blkdev_put(nbc->backing_bdev,
1314                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1315                 if (nbc->md_bdev)
1316                         blkdev_put(nbc->md_bdev,
1317                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1318                 kfree(nbc);
1319         }
1320         lc_destroy(resync_lru);
1321
1322         reply->ret_code = retcode;
1323         drbd_reconfig_done(mdev);
1324         return 0;
1325 }
1326
1327 /* Detaching the disk is a process in multiple stages.  First we need to lock
1328  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1329  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1330  * internal references as well.
1331  * Only then we have finally detached. */
1332 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1333                           struct drbd_nl_cfg_reply *reply)
1334 {
1335         enum drbd_ret_code retcode;
1336         int ret;
1337         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1338         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1339         /* D_FAILED will transition to DISKLESS. */
1340         ret = wait_event_interruptible(mdev->misc_wait,
1341                         mdev->state.disk != D_FAILED);
1342         drbd_resume_io(mdev);
1343         if ((int)retcode == (int)SS_IS_DISKLESS)
1344                 retcode = SS_NOTHING_TO_DO;
1345         if (ret)
1346                 retcode = ERR_INTR;
1347         reply->ret_code = retcode;
1348         return 0;
1349 }
1350
1351 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1352                             struct drbd_nl_cfg_reply *reply)
1353 {
1354         int i;
1355         enum drbd_ret_code retcode;
1356         struct net_conf *new_conf = NULL;
1357         struct crypto_hash *tfm = NULL;
1358         struct crypto_hash *integrity_w_tfm = NULL;
1359         struct crypto_hash *integrity_r_tfm = NULL;
1360         struct drbd_conf *odev;
1361         char hmac_name[CRYPTO_MAX_ALG_NAME];
1362         void *int_dig_out = NULL;
1363         void *int_dig_in = NULL;
1364         void *int_dig_vv = NULL;
1365         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1366
1367         drbd_reconfig_start(mdev);
1368
1369         if (mdev->state.conn > C_STANDALONE) {
1370                 retcode = ERR_NET_CONFIGURED;
1371                 goto fail;
1372         }
1373
1374         /* allocation not in the IO path, cqueue thread context */
1375         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1376         if (!new_conf) {
1377                 retcode = ERR_NOMEM;
1378                 goto fail;
1379         }
1380
1381         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1382         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1383         new_conf->ping_int         = DRBD_PING_INT_DEF;
1384         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1385         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1386         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1387         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1388         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1389         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1390         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1391         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1392         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1393         new_conf->want_lose        = 0;
1394         new_conf->two_primaries    = 0;
1395         new_conf->wire_protocol    = DRBD_PROT_C;
1396         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1397         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1398         new_conf->on_congestion    = DRBD_ON_CONGESTION_DEF;
1399         new_conf->cong_extents     = DRBD_CONG_EXTENTS_DEF;
1400
1401         if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1402                 retcode = ERR_MANDATORY_TAG;
1403                 goto fail;
1404         }
1405
1406         if (new_conf->two_primaries
1407             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1408                 retcode = ERR_NOT_PROTO_C;
1409                 goto fail;
1410         }
1411
1412         if (get_ldev(mdev)) {
1413                 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1414                 put_ldev(mdev);
1415                 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1416                         retcode = ERR_STONITH_AND_PROT_A;
1417                         goto fail;
1418                 }
1419         }
1420
1421         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1422                 retcode = ERR_CONG_NOT_PROTO_A;
1423                 goto fail;
1424         }
1425
1426         if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1427                 retcode = ERR_DISCARD;
1428                 goto fail;
1429         }
1430
1431         retcode = NO_ERROR;
1432
1433         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1434         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1435         for (i = 0; i < minor_count; i++) {
1436                 odev = minor_to_mdev(i);
1437                 if (!odev || odev == mdev)
1438                         continue;
1439                 if (get_net_conf(odev->tconn)) {
1440                         taken_addr = (struct sockaddr *)&odev->tconn->net_conf->my_addr;
1441                         if (new_conf->my_addr_len == odev->tconn->net_conf->my_addr_len &&
1442                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1443                                 retcode = ERR_LOCAL_ADDR;
1444
1445                         taken_addr = (struct sockaddr *)&odev->tconn->net_conf->peer_addr;
1446                         if (new_conf->peer_addr_len == odev->tconn->net_conf->peer_addr_len &&
1447                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1448                                 retcode = ERR_PEER_ADDR;
1449
1450                         put_net_conf(odev->tconn);
1451                         if (retcode != NO_ERROR)
1452                                 goto fail;
1453                 }
1454         }
1455
1456         if (new_conf->cram_hmac_alg[0] != 0) {
1457                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1458                         new_conf->cram_hmac_alg);
1459                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1460                 if (IS_ERR(tfm)) {
1461                         tfm = NULL;
1462                         retcode = ERR_AUTH_ALG;
1463                         goto fail;
1464                 }
1465
1466                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1467                         retcode = ERR_AUTH_ALG_ND;
1468                         goto fail;
1469                 }
1470         }
1471
1472         if (new_conf->integrity_alg[0]) {
1473                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1474                 if (IS_ERR(integrity_w_tfm)) {
1475                         integrity_w_tfm = NULL;
1476                         retcode=ERR_INTEGRITY_ALG;
1477                         goto fail;
1478                 }
1479
1480                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1481                         retcode=ERR_INTEGRITY_ALG_ND;
1482                         goto fail;
1483                 }
1484
1485                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1486                 if (IS_ERR(integrity_r_tfm)) {
1487                         integrity_r_tfm = NULL;
1488                         retcode=ERR_INTEGRITY_ALG;
1489                         goto fail;
1490                 }
1491         }
1492
1493         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1494
1495         if (integrity_w_tfm) {
1496                 i = crypto_hash_digestsize(integrity_w_tfm);
1497                 int_dig_out = kmalloc(i, GFP_KERNEL);
1498                 if (!int_dig_out) {
1499                         retcode = ERR_NOMEM;
1500                         goto fail;
1501                 }
1502                 int_dig_in = kmalloc(i, GFP_KERNEL);
1503                 if (!int_dig_in) {
1504                         retcode = ERR_NOMEM;
1505                         goto fail;
1506                 }
1507                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1508                 if (!int_dig_vv) {
1509                         retcode = ERR_NOMEM;
1510                         goto fail;
1511                 }
1512         }
1513
1514         if (!mdev->bitmap) {
1515                 if(drbd_bm_init(mdev)) {
1516                         retcode = ERR_NOMEM;
1517                         goto fail;
1518                 }
1519         }
1520
1521         drbd_flush_workqueue(mdev);
1522         spin_lock_irq(&mdev->tconn->req_lock);
1523         if (mdev->tconn->net_conf != NULL) {
1524                 retcode = ERR_NET_CONFIGURED;
1525                 spin_unlock_irq(&mdev->tconn->req_lock);
1526                 goto fail;
1527         }
1528         mdev->tconn->net_conf = new_conf;
1529
1530         mdev->send_cnt = 0;
1531         mdev->recv_cnt = 0;
1532
1533         crypto_free_hash(mdev->tconn->cram_hmac_tfm);
1534         mdev->tconn->cram_hmac_tfm = tfm;
1535
1536         crypto_free_hash(mdev->tconn->integrity_w_tfm);
1537         mdev->tconn->integrity_w_tfm = integrity_w_tfm;
1538
1539         crypto_free_hash(mdev->tconn->integrity_r_tfm);
1540         mdev->tconn->integrity_r_tfm = integrity_r_tfm;
1541
1542         kfree(mdev->tconn->int_dig_out);
1543         kfree(mdev->tconn->int_dig_in);
1544         kfree(mdev->tconn->int_dig_vv);
1545         mdev->tconn->int_dig_out=int_dig_out;
1546         mdev->tconn->int_dig_in=int_dig_in;
1547         mdev->tconn->int_dig_vv=int_dig_vv;
1548         retcode = _conn_request_state(mdev->tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
1549         spin_unlock_irq(&mdev->tconn->req_lock);
1550
1551         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1552         reply->ret_code = retcode;
1553         drbd_reconfig_done(mdev);
1554         return 0;
1555
1556 fail:
1557         kfree(int_dig_out);
1558         kfree(int_dig_in);
1559         kfree(int_dig_vv);
1560         crypto_free_hash(tfm);
1561         crypto_free_hash(integrity_w_tfm);
1562         crypto_free_hash(integrity_r_tfm);
1563         kfree(new_conf);
1564
1565         reply->ret_code = retcode;
1566         drbd_reconfig_done(mdev);
1567         return 0;
1568 }
1569
1570 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1571                               struct drbd_nl_cfg_reply *reply)
1572 {
1573         struct drbd_tconn *tconn = mdev->tconn;
1574         int retcode;
1575         struct disconnect dc;
1576
1577         memset(&dc, 0, sizeof(struct disconnect));
1578         if (!disconnect_from_tags(mdev, nlp->tag_list, &dc)) {
1579                 retcode = ERR_MANDATORY_TAG;
1580                 goto fail;
1581         }
1582
1583         if (dc.force) {
1584                 spin_lock_irq(&tconn->req_lock);
1585                 if (tconn->cstate >= C_WF_CONNECTION)
1586                         _conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
1587                 spin_unlock_irq(&tconn->req_lock);
1588                 goto done;
1589         }
1590
1591         retcode = conn_request_state(tconn, NS(conn, C_DISCONNECTING), 0);
1592
1593         if (retcode == SS_NOTHING_TO_DO)
1594                 goto done;
1595         else if (retcode == SS_ALREADY_STANDALONE)
1596                 goto done;
1597         else if (retcode == SS_PRIMARY_NOP) {
1598                 /* Our state checking code wants to see the peer outdated. */
1599                 retcode = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
1600                                                         pdsk, D_OUTDATED), CS_VERBOSE);
1601         } else if (retcode == SS_CW_FAILED_BY_PEER) {
1602                 /* The peer probably wants to see us outdated. */
1603                 retcode = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
1604                                                         disk, D_OUTDATED), 0);
1605                 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1606                         conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
1607                         retcode = SS_SUCCESS;
1608                 }
1609         }
1610
1611         if (retcode < SS_SUCCESS)
1612                 goto fail;
1613
1614         if (wait_event_interruptible(tconn->ping_wait,
1615                                      tconn->cstate != C_DISCONNECTING)) {
1616                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1617                    someone else might connect us in the mean time! */
1618                 retcode = ERR_INTR;
1619                 goto fail;
1620         }
1621
1622  done:
1623         retcode = NO_ERROR;
1624  fail:
1625         drbd_md_sync(mdev);
1626         reply->ret_code = retcode;
1627         return 0;
1628 }
1629
1630 void resync_after_online_grow(struct drbd_conf *mdev)
1631 {
1632         int iass; /* I am sync source */
1633
1634         dev_info(DEV, "Resync of new storage after online grow\n");
1635         if (mdev->state.role != mdev->state.peer)
1636                 iass = (mdev->state.role == R_PRIMARY);
1637         else
1638                 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
1639
1640         if (iass)
1641                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1642         else
1643                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1644 }
1645
1646 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1647                           struct drbd_nl_cfg_reply *reply)
1648 {
1649         struct resize rs;
1650         int retcode = NO_ERROR;
1651         enum determine_dev_size dd;
1652         enum dds_flags ddsf;
1653
1654         memset(&rs, 0, sizeof(struct resize));
1655         if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1656                 retcode = ERR_MANDATORY_TAG;
1657                 goto fail;
1658         }
1659
1660         if (mdev->state.conn > C_CONNECTED) {
1661                 retcode = ERR_RESIZE_RESYNC;
1662                 goto fail;
1663         }
1664
1665         if (mdev->state.role == R_SECONDARY &&
1666             mdev->state.peer == R_SECONDARY) {
1667                 retcode = ERR_NO_PRIMARY;
1668                 goto fail;
1669         }
1670
1671         if (!get_ldev(mdev)) {
1672                 retcode = ERR_NO_DISK;
1673                 goto fail;
1674         }
1675
1676         if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
1677                 retcode = ERR_NEED_APV_93;
1678                 goto fail;
1679         }
1680
1681         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1682                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1683
1684         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1685         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1686         dd = drbd_determine_dev_size(mdev, ddsf);
1687         drbd_md_sync(mdev);
1688         put_ldev(mdev);
1689         if (dd == dev_size_error) {
1690                 retcode = ERR_NOMEM_BITMAP;
1691                 goto fail;
1692         }
1693
1694         if (mdev->state.conn == C_CONNECTED) {
1695                 if (dd == grew)
1696                         set_bit(RESIZE_PENDING, &mdev->flags);
1697
1698                 drbd_send_uuids(mdev);
1699                 drbd_send_sizes(mdev, 1, ddsf);
1700         }
1701
1702  fail:
1703         reply->ret_code = retcode;
1704         return 0;
1705 }
1706
1707 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1708                                struct drbd_nl_cfg_reply *reply)
1709 {
1710         int retcode = NO_ERROR;
1711         int err;
1712         int ovr; /* online verify running */
1713         int rsr; /* re-sync running */
1714         struct crypto_hash *verify_tfm = NULL;
1715         struct crypto_hash *csums_tfm = NULL;
1716         struct syncer_conf sc;
1717         cpumask_var_t new_cpu_mask;
1718         int *rs_plan_s = NULL;
1719         int fifo_size;
1720
1721         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1722                 retcode = ERR_NOMEM;
1723                 goto fail;
1724         }
1725
1726         if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1727                 memset(&sc, 0, sizeof(struct syncer_conf));
1728                 sc.rate       = DRBD_RATE_DEF;
1729                 sc.after      = DRBD_AFTER_DEF;
1730                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1731                 sc.on_no_data  = DRBD_ON_NO_DATA_DEF;
1732                 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1733                 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1734                 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1735                 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1736                 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1737         } else
1738                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1739
1740         if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1741                 retcode = ERR_MANDATORY_TAG;
1742                 goto fail;
1743         }
1744
1745         /* re-sync running */
1746         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1747                 mdev->state.conn == C_SYNC_TARGET ||
1748                 mdev->state.conn == C_PAUSED_SYNC_S ||
1749                 mdev->state.conn == C_PAUSED_SYNC_T );
1750
1751         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1752                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1753                 goto fail;
1754         }
1755
1756         if (!rsr && sc.csums_alg[0]) {
1757                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1758                 if (IS_ERR(csums_tfm)) {
1759                         csums_tfm = NULL;
1760                         retcode = ERR_CSUMS_ALG;
1761                         goto fail;
1762                 }
1763
1764                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1765                         retcode = ERR_CSUMS_ALG_ND;
1766                         goto fail;
1767                 }
1768         }
1769
1770         /* online verify running */
1771         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1772
1773         if (ovr) {
1774                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1775                         retcode = ERR_VERIFY_RUNNING;
1776                         goto fail;
1777                 }
1778         }
1779
1780         if (!ovr && sc.verify_alg[0]) {
1781                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1782                 if (IS_ERR(verify_tfm)) {
1783                         verify_tfm = NULL;
1784                         retcode = ERR_VERIFY_ALG;
1785                         goto fail;
1786                 }
1787
1788                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1789                         retcode = ERR_VERIFY_ALG_ND;
1790                         goto fail;
1791                 }
1792         }
1793
1794         /* silently ignore cpu mask on UP kernel */
1795         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1796                 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1797                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1798                 if (err) {
1799                         dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1800                         retcode = ERR_CPU_MASK_PARSE;
1801                         goto fail;
1802                 }
1803         }
1804
1805         if (!expect(sc.rate >= 1))
1806                 sc.rate = 1;
1807         if (!expect(sc.al_extents >= 7))
1808                 sc.al_extents = 127; /* arbitrary minimum */
1809 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1810         if (sc.al_extents > AL_MAX) {
1811                 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1812                 sc.al_extents = AL_MAX;
1813         }
1814 #undef AL_MAX
1815
1816         /* to avoid spurious errors when configuring minors before configuring
1817          * the minors they depend on: if necessary, first create the minor we
1818          * depend on */
1819         if (sc.after >= 0)
1820                 ensure_mdev(sc.after, 1);
1821
1822         /* most sanity checks done, try to assign the new sync-after
1823          * dependency.  need to hold the global lock in there,
1824          * to avoid a race in the dependency loop check. */
1825         retcode = drbd_alter_sa(mdev, sc.after);
1826         if (retcode != NO_ERROR)
1827                 goto fail;
1828
1829         fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1830         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1831                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1832                 if (!rs_plan_s) {
1833                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1834                         retcode = ERR_NOMEM;
1835                         goto fail;
1836                 }
1837         }
1838
1839         /* ok, assign the rest of it as well.
1840          * lock against receive_SyncParam() */
1841         spin_lock(&mdev->peer_seq_lock);
1842         mdev->sync_conf = sc;
1843
1844         if (!rsr) {
1845                 crypto_free_hash(mdev->csums_tfm);
1846                 mdev->csums_tfm = csums_tfm;
1847                 csums_tfm = NULL;
1848         }
1849
1850         if (!ovr) {
1851                 crypto_free_hash(mdev->verify_tfm);
1852                 mdev->verify_tfm = verify_tfm;
1853                 verify_tfm = NULL;
1854         }
1855
1856         if (fifo_size != mdev->rs_plan_s.size) {
1857                 kfree(mdev->rs_plan_s.values);
1858                 mdev->rs_plan_s.values = rs_plan_s;
1859                 mdev->rs_plan_s.size   = fifo_size;
1860                 mdev->rs_planed = 0;
1861                 rs_plan_s = NULL;
1862         }
1863
1864         spin_unlock(&mdev->peer_seq_lock);
1865
1866         if (get_ldev(mdev)) {
1867                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1868                 drbd_al_shrink(mdev);
1869                 err = drbd_check_al_size(mdev);
1870                 lc_unlock(mdev->act_log);
1871                 wake_up(&mdev->al_wait);
1872
1873                 put_ldev(mdev);
1874                 drbd_md_sync(mdev);
1875
1876                 if (err) {
1877                         retcode = ERR_NOMEM;
1878                         goto fail;
1879                 }
1880         }
1881
1882         if (mdev->state.conn >= C_CONNECTED)
1883                 drbd_send_sync_param(mdev, &sc);
1884
1885         if (!cpumask_equal(mdev->tconn->cpu_mask, new_cpu_mask)) {
1886                 cpumask_copy(mdev->tconn->cpu_mask, new_cpu_mask);
1887                 drbd_calc_cpu_mask(mdev->tconn);
1888                 mdev->tconn->receiver.reset_cpu_mask = 1;
1889                 mdev->tconn->asender.reset_cpu_mask = 1;
1890                 mdev->tconn->worker.reset_cpu_mask = 1;
1891         }
1892
1893         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1894 fail:
1895         kfree(rs_plan_s);
1896         free_cpumask_var(new_cpu_mask);
1897         crypto_free_hash(csums_tfm);
1898         crypto_free_hash(verify_tfm);
1899         reply->ret_code = retcode;
1900         return 0;
1901 }
1902
1903 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1904                               struct drbd_nl_cfg_reply *reply)
1905 {
1906         int retcode;
1907
1908         /* If there is still bitmap IO pending, probably because of a previous
1909          * resync just being finished, wait for it before requesting a new resync. */
1910         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1911
1912         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1913
1914         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1915                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1916
1917         while (retcode == SS_NEED_CONNECTION) {
1918                 spin_lock_irq(&mdev->tconn->req_lock);
1919                 if (mdev->state.conn < C_CONNECTED)
1920                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1921                 spin_unlock_irq(&mdev->tconn->req_lock);
1922
1923                 if (retcode != SS_NEED_CONNECTION)
1924                         break;
1925
1926                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1927         }
1928
1929         reply->ret_code = retcode;
1930         return 0;
1931 }
1932
1933 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
1934 {
1935         int rv;
1936
1937         rv = drbd_bmio_set_n_write(mdev);
1938         drbd_suspend_al(mdev);
1939         return rv;
1940 }
1941
1942 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1943                                    struct drbd_nl_cfg_reply *reply)
1944 {
1945         int retcode;
1946
1947         /* If there is still bitmap IO pending, probably because of a previous
1948          * resync just being finished, wait for it before requesting a new resync. */
1949         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1950
1951         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
1952
1953         if (retcode < SS_SUCCESS) {
1954                 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
1955                         /* The peer will get a resync upon connect anyways. Just make that
1956                            into a full resync. */
1957                         retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
1958                         if (retcode >= SS_SUCCESS) {
1959                                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
1960                                         "set_n_write from invalidate_peer",
1961                                         BM_LOCKED_SET_ALLOWED))
1962                                         retcode = ERR_IO_MD_DISK;
1963                         }
1964                 } else
1965                         retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
1966         }
1967
1968         reply->ret_code = retcode;
1969         return 0;
1970 }
1971
1972 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1973                               struct drbd_nl_cfg_reply *reply)
1974 {
1975         int retcode = NO_ERROR;
1976
1977         if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
1978                 retcode = ERR_PAUSE_IS_SET;
1979
1980         reply->ret_code = retcode;
1981         return 0;
1982 }
1983
1984 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1985                                struct drbd_nl_cfg_reply *reply)
1986 {
1987         int retcode = NO_ERROR;
1988         union drbd_state s;
1989
1990         if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
1991                 s = mdev->state;
1992                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
1993                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
1994                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
1995                 } else {
1996                         retcode = ERR_PAUSE_IS_CLEAR;
1997                 }
1998         }
1999
2000         reply->ret_code = retcode;
2001         return 0;
2002 }
2003
2004 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2005                               struct drbd_nl_cfg_reply *reply)
2006 {
2007         reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
2008
2009         return 0;
2010 }
2011
2012 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2013                              struct drbd_nl_cfg_reply *reply)
2014 {
2015         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2016                 drbd_uuid_new_current(mdev);
2017                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2018         }
2019         drbd_suspend_io(mdev);
2020         reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2021         if (reply->ret_code == SS_SUCCESS) {
2022                 if (mdev->state.conn < C_CONNECTED)
2023                         tl_clear(mdev);
2024                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2025                         tl_restart(mdev, FAIL_FROZEN_DISK_IO);
2026         }
2027         drbd_resume_io(mdev);
2028
2029         return 0;
2030 }
2031
2032 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2033                            struct drbd_nl_cfg_reply *reply)
2034 {
2035         reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
2036         return 0;
2037 }
2038
2039 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2040                            struct drbd_nl_cfg_reply *reply)
2041 {
2042         unsigned short *tl;
2043
2044         tl = reply->tag_list;
2045
2046         if (get_ldev(mdev)) {
2047                 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
2048                 put_ldev(mdev);
2049         }
2050
2051         if (get_net_conf(mdev->tconn)) {
2052                 tl = net_conf_to_tags(mdev, mdev->tconn->net_conf, tl);
2053                 put_net_conf(mdev->tconn);
2054         }
2055         tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
2056
2057         put_unaligned(TT_END, tl++); /* Close the tag list */
2058
2059         return (int)((char *)tl - (char *)reply->tag_list);
2060 }
2061
2062 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2063                              struct drbd_nl_cfg_reply *reply)
2064 {
2065         unsigned short *tl = reply->tag_list;
2066         union drbd_state s = mdev->state;
2067         unsigned long rs_left;
2068         unsigned int res;
2069
2070         tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
2071
2072         /* no local ref, no bitmap, no syncer progress. */
2073         if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
2074                 if (get_ldev(mdev)) {
2075                         drbd_get_syncer_progress(mdev, &rs_left, &res);
2076                         tl = tl_add_int(tl, T_sync_progress, &res);
2077                         put_ldev(mdev);
2078                 }
2079         }
2080         put_unaligned(TT_END, tl++); /* Close the tag list */
2081
2082         return (int)((char *)tl - (char *)reply->tag_list);
2083 }
2084
2085 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2086                              struct drbd_nl_cfg_reply *reply)
2087 {
2088         unsigned short *tl;
2089
2090         tl = reply->tag_list;
2091
2092         if (get_ldev(mdev)) {
2093                 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
2094                 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
2095                 put_ldev(mdev);
2096         }
2097         put_unaligned(TT_END, tl++); /* Close the tag list */
2098
2099         return (int)((char *)tl - (char *)reply->tag_list);
2100 }
2101
2102 /**
2103  * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
2104  * @mdev:       DRBD device.
2105  * @nlp:        Netlink/connector packet from drbdsetup
2106  * @reply:      Reply packet for drbdsetup
2107  */
2108 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2109                                     struct drbd_nl_cfg_reply *reply)
2110 {
2111         unsigned short *tl;
2112         char rv;
2113
2114         tl = reply->tag_list;
2115
2116         rv = mdev->state.pdsk == D_OUTDATED        ? UT_PEER_OUTDATED :
2117           test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
2118
2119         tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
2120         put_unaligned(TT_END, tl++); /* Close the tag list */
2121
2122         return (int)((char *)tl - (char *)reply->tag_list);
2123 }
2124
2125 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2126                                     struct drbd_nl_cfg_reply *reply)
2127 {
2128         /* default to resume from last known position, if possible */
2129         struct start_ov args =
2130                 { .start_sector = mdev->ov_start_sector };
2131
2132         if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2133                 reply->ret_code = ERR_MANDATORY_TAG;
2134                 return 0;
2135         }
2136
2137         /* If there is still bitmap IO pending, e.g. previous resync or verify
2138          * just being finished, wait for it before requesting a new resync. */
2139         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2140
2141         /* w_make_ov_request expects position to be aligned */
2142         mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
2143         reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2144         return 0;
2145 }
2146
2147
2148 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2149                               struct drbd_nl_cfg_reply *reply)
2150 {
2151         int retcode = NO_ERROR;
2152         int skip_initial_sync = 0;
2153         int err;
2154
2155         struct new_c_uuid args;
2156
2157         memset(&args, 0, sizeof(struct new_c_uuid));
2158         if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2159                 reply->ret_code = ERR_MANDATORY_TAG;
2160                 return 0;
2161         }
2162
2163         mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2164
2165         if (!get_ldev(mdev)) {
2166                 retcode = ERR_NO_DISK;
2167                 goto out;
2168         }
2169
2170         /* this is "skip initial sync", assume to be clean */
2171         if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2172             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2173                 dev_info(DEV, "Preparing to skip initial sync\n");
2174                 skip_initial_sync = 1;
2175         } else if (mdev->state.conn != C_STANDALONE) {
2176                 retcode = ERR_CONNECTED;
2177                 goto out_dec;
2178         }
2179
2180         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2181         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2182
2183         if (args.clear_bm) {
2184                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2185                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2186                 if (err) {
2187                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2188                         retcode = ERR_IO_MD_DISK;
2189                 }
2190                 if (skip_initial_sync) {
2191                         drbd_send_uuids_skip_initial_sync(mdev);
2192                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2193                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2194                         spin_lock_irq(&mdev->tconn->req_lock);
2195                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2196                                         CS_VERBOSE, NULL);
2197                         spin_unlock_irq(&mdev->tconn->req_lock);
2198                 }
2199         }
2200
2201         drbd_md_sync(mdev);
2202 out_dec:
2203         put_ldev(mdev);
2204 out:
2205         mutex_unlock(mdev->state_mutex);
2206
2207         reply->ret_code = retcode;
2208         return 0;
2209 }
2210
2211 struct cn_handler_struct {
2212         int (*function)(struct drbd_conf *,
2213                          struct drbd_nl_cfg_req *,
2214                          struct drbd_nl_cfg_reply *);
2215         int reply_body_size;
2216 };
2217
2218 static struct cn_handler_struct cnd_table[] = {
2219         [ P_primary ]           = { &drbd_nl_primary,           0 },
2220         [ P_secondary ]         = { &drbd_nl_secondary,         0 },
2221         [ P_disk_conf ]         = { &drbd_nl_disk_conf,         0 },
2222         [ P_detach ]            = { &drbd_nl_detach,            0 },
2223         [ P_net_conf ]          = { &drbd_nl_net_conf,          0 },
2224         [ P_disconnect ]        = { &drbd_nl_disconnect,        0 },
2225         [ P_resize ]            = { &drbd_nl_resize,            0 },
2226         [ P_syncer_conf ]       = { &drbd_nl_syncer_conf,       0 },
2227         [ P_invalidate ]        = { &drbd_nl_invalidate,        0 },
2228         [ P_invalidate_peer ]   = { &drbd_nl_invalidate_peer,   0 },
2229         [ P_pause_sync ]        = { &drbd_nl_pause_sync,        0 },
2230         [ P_resume_sync ]       = { &drbd_nl_resume_sync,       0 },
2231         [ P_suspend_io ]        = { &drbd_nl_suspend_io,        0 },
2232         [ P_resume_io ]         = { &drbd_nl_resume_io,         0 },
2233         [ P_outdate ]           = { &drbd_nl_outdate,           0 },
2234         [ P_get_config ]        = { &drbd_nl_get_config,
2235                                     sizeof(struct syncer_conf_tag_len_struct) +
2236                                     sizeof(struct disk_conf_tag_len_struct) +
2237                                     sizeof(struct net_conf_tag_len_struct) },
2238         [ P_get_state ]         = { &drbd_nl_get_state,
2239                                     sizeof(struct get_state_tag_len_struct) +
2240                                     sizeof(struct sync_progress_tag_len_struct) },
2241         [ P_get_uuids ]         = { &drbd_nl_get_uuids,
2242                                     sizeof(struct get_uuids_tag_len_struct) },
2243         [ P_get_timeout_flag ]  = { &drbd_nl_get_timeout_flag,
2244                                     sizeof(struct get_timeout_flag_tag_len_struct)},
2245         [ P_start_ov ]          = { &drbd_nl_start_ov,          0 },
2246         [ P_new_c_uuid ]        = { &drbd_nl_new_c_uuid,        0 },
2247 };
2248
2249 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2250 {
2251         struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2252         struct cn_handler_struct *cm;
2253         struct cn_msg *cn_reply;
2254         struct drbd_nl_cfg_reply *reply;
2255         struct drbd_conf *mdev;
2256         int retcode, rr;
2257         int reply_size = sizeof(struct cn_msg)
2258                 + sizeof(struct drbd_nl_cfg_reply)
2259                 + sizeof(short int);
2260
2261         if (!try_module_get(THIS_MODULE)) {
2262                 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2263                 return;
2264         }
2265
2266         if (!cap_raised(current_cap(), CAP_SYS_ADMIN)) {
2267                 retcode = ERR_PERM;
2268                 goto fail;
2269         }
2270
2271         mdev = ensure_mdev(nlp->drbd_minor,
2272                         (nlp->flags & DRBD_NL_CREATE_DEVICE));
2273         if (!mdev) {
2274                 retcode = ERR_MINOR_INVALID;
2275                 goto fail;
2276         }
2277
2278         if (nlp->packet_type >= P_nl_after_last_packet ||
2279             nlp->packet_type == P_return_code_only) {
2280                 retcode = ERR_PACKET_NR;
2281                 goto fail;
2282         }
2283
2284         cm = cnd_table + nlp->packet_type;
2285
2286         /* This may happen if packet number is 0: */
2287         if (cm->function == NULL) {
2288                 retcode = ERR_PACKET_NR;
2289                 goto fail;
2290         }
2291
2292         reply_size += cm->reply_body_size;
2293
2294         /* allocation not in the IO path, cqueue thread context */
2295         cn_reply = kzalloc(reply_size, GFP_KERNEL);
2296         if (!cn_reply) {
2297                 retcode = ERR_NOMEM;
2298                 goto fail;
2299         }
2300         reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2301
2302         reply->packet_type =
2303                 cm->reply_body_size ? nlp->packet_type : P_return_code_only;
2304         reply->minor = nlp->drbd_minor;
2305         reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2306         /* reply->tag_list; might be modified by cm->function. */
2307
2308         rr = cm->function(mdev, nlp, reply);
2309
2310         cn_reply->id = req->id;
2311         cn_reply->seq = req->seq;
2312         cn_reply->ack = req->ack  + 1;
2313         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2314         cn_reply->flags = 0;
2315
2316         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2317         if (rr && rr != -ESRCH)
2318                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2319
2320         kfree(cn_reply);
2321         module_put(THIS_MODULE);
2322         return;
2323  fail:
2324         drbd_nl_send_reply(req, retcode);
2325         module_put(THIS_MODULE);
2326 }
2327
2328 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2329
2330 static unsigned short *
2331 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2332         unsigned short len, int nul_terminated)
2333 {
2334         unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2335         len = (len < l) ? len :  l;
2336         put_unaligned(tag, tl++);
2337         put_unaligned(len, tl++);
2338         memcpy(tl, data, len);
2339         tl = (unsigned short*)((char*)tl + len);
2340         if (nul_terminated)
2341                 *((char*)tl - 1) = 0;
2342         return tl;
2343 }
2344
2345 static unsigned short *
2346 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2347 {
2348         return __tl_add_blob(tl, tag, data, len, 0);
2349 }
2350
2351 static unsigned short *
2352 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2353 {
2354         return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2355 }
2356
2357 static unsigned short *
2358 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2359 {
2360         put_unaligned(tag, tl++);
2361         switch(tag_type(tag)) {
2362         case TT_INTEGER:
2363                 put_unaligned(sizeof(int), tl++);
2364                 put_unaligned(*(int *)val, (int *)tl);
2365                 tl = (unsigned short*)((char*)tl+sizeof(int));
2366                 break;
2367         case TT_INT64:
2368                 put_unaligned(sizeof(u64), tl++);
2369                 put_unaligned(*(u64 *)val, (u64 *)tl);
2370                 tl = (unsigned short*)((char*)tl+sizeof(u64));
2371                 break;
2372         default:
2373                 /* someone did something stupid. */
2374                 ;
2375         }
2376         return tl;
2377 }
2378
2379 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2380 {
2381         char buffer[sizeof(struct cn_msg)+
2382                     sizeof(struct drbd_nl_cfg_reply)+
2383                     sizeof(struct get_state_tag_len_struct)+
2384                     sizeof(short int)];
2385         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2386         struct drbd_nl_cfg_reply *reply =
2387                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2388         unsigned short *tl = reply->tag_list;
2389
2390         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2391
2392         tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2393
2394         put_unaligned(TT_END, tl++); /* Close the tag list */
2395
2396         cn_reply->id.idx = CN_IDX_DRBD;
2397         cn_reply->id.val = CN_VAL_DRBD;
2398
2399         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2400         cn_reply->ack = 0; /* not used here. */
2401         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2402                 (int)((char *)tl - (char *)reply->tag_list);
2403         cn_reply->flags = 0;
2404
2405         reply->packet_type = P_get_state;
2406         reply->minor = mdev_to_minor(mdev);
2407         reply->ret_code = NO_ERROR;
2408
2409         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2410 }
2411
2412 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2413 {
2414         char buffer[sizeof(struct cn_msg)+
2415                     sizeof(struct drbd_nl_cfg_reply)+
2416                     sizeof(struct call_helper_tag_len_struct)+
2417                     sizeof(short int)];
2418         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2419         struct drbd_nl_cfg_reply *reply =
2420                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2421         unsigned short *tl = reply->tag_list;
2422
2423         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2424
2425         tl = tl_add_str(tl, T_helper, helper_name);
2426         put_unaligned(TT_END, tl++); /* Close the tag list */
2427
2428         cn_reply->id.idx = CN_IDX_DRBD;
2429         cn_reply->id.val = CN_VAL_DRBD;
2430
2431         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2432         cn_reply->ack = 0; /* not used here. */
2433         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2434                 (int)((char *)tl - (char *)reply->tag_list);
2435         cn_reply->flags = 0;
2436
2437         reply->packet_type = P_call_helper;
2438         reply->minor = mdev_to_minor(mdev);
2439         reply->ret_code = NO_ERROR;
2440
2441         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2442 }
2443
2444 void drbd_bcast_ee(struct drbd_conf *mdev, const char *reason, const int dgs,
2445                    const char *seen_hash, const char *calc_hash,
2446                            const struct drbd_peer_request *peer_req)
2447 {
2448         struct cn_msg *cn_reply;
2449         struct drbd_nl_cfg_reply *reply;
2450         unsigned short *tl;
2451         struct page *page;
2452         unsigned len;
2453
2454         if (!peer_req)
2455                 return;
2456         if (!reason || !reason[0])
2457                 return;
2458
2459         /* apparently we have to memcpy twice, first to prepare the data for the
2460          * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2461          * netlink skb. */
2462         /* receiver thread context, which is not in the writeout path (of this node),
2463          * but may be in the writeout path of the _other_ node.
2464          * GFP_NOIO to avoid potential "distributed deadlock". */
2465         cn_reply = kzalloc(
2466                 sizeof(struct cn_msg)+
2467                 sizeof(struct drbd_nl_cfg_reply)+
2468                 sizeof(struct dump_ee_tag_len_struct)+
2469                 sizeof(short int),
2470                 GFP_NOIO);
2471
2472         if (!cn_reply) {
2473                 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, "
2474                              "sector %llu, size %u\n",
2475                         (unsigned long long)peer_req->i.sector,
2476                         peer_req->i.size);
2477                 return;
2478         }
2479
2480         reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2481         tl = reply->tag_list;
2482
2483         tl = tl_add_str(tl, T_dump_ee_reason, reason);
2484         tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2485         tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2486         tl = tl_add_int(tl, T_ee_sector, &peer_req->i.sector);
2487         tl = tl_add_int(tl, T_ee_block_id, &peer_req->block_id);
2488
2489         /* dump the first 32k */
2490         len = min_t(unsigned, peer_req->i.size, 32 << 10);
2491         put_unaligned(T_ee_data, tl++);
2492         put_unaligned(len, tl++);
2493
2494         page = peer_req->pages;
2495         page_chain_for_each(page) {
2496                 void *d = kmap_atomic(page, KM_USER0);
2497                 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2498                 memcpy(tl, d, l);
2499                 kunmap_atomic(d, KM_USER0);
2500                 tl = (unsigned short*)((char*)tl + l);
2501                 len -= l;
2502                 if (len == 0)
2503                         break;
2504         }
2505         put_unaligned(TT_END, tl++); /* Close the tag list */
2506
2507         cn_reply->id.idx = CN_IDX_DRBD;
2508         cn_reply->id.val = CN_VAL_DRBD;
2509
2510         cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2511         cn_reply->ack = 0; // not used here.
2512         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2513                 (int)((char*)tl - (char*)reply->tag_list);
2514         cn_reply->flags = 0;
2515
2516         reply->packet_type = P_dump_ee;
2517         reply->minor = mdev_to_minor(mdev);
2518         reply->ret_code = NO_ERROR;
2519
2520         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2521         kfree(cn_reply);
2522 }
2523
2524 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2525 {
2526         char buffer[sizeof(struct cn_msg)+
2527                     sizeof(struct drbd_nl_cfg_reply)+
2528                     sizeof(struct sync_progress_tag_len_struct)+
2529                     sizeof(short int)];
2530         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2531         struct drbd_nl_cfg_reply *reply =
2532                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2533         unsigned short *tl = reply->tag_list;
2534         unsigned long rs_left;
2535         unsigned int res;
2536
2537         /* no local ref, no bitmap, no syncer progress, no broadcast. */
2538         if (!get_ldev(mdev))
2539                 return;
2540         drbd_get_syncer_progress(mdev, &rs_left, &res);
2541         put_ldev(mdev);
2542
2543         tl = tl_add_int(tl, T_sync_progress, &res);
2544         put_unaligned(TT_END, tl++); /* Close the tag list */
2545
2546         cn_reply->id.idx = CN_IDX_DRBD;
2547         cn_reply->id.val = CN_VAL_DRBD;
2548
2549         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2550         cn_reply->ack = 0; /* not used here. */
2551         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2552                 (int)((char *)tl - (char *)reply->tag_list);
2553         cn_reply->flags = 0;
2554
2555         reply->packet_type = P_sync_progress;
2556         reply->minor = mdev_to_minor(mdev);
2557         reply->ret_code = NO_ERROR;
2558
2559         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2560 }
2561
2562 int __init drbd_nl_init(void)
2563 {
2564         static struct cb_id cn_id_drbd;
2565         int err, try=10;
2566
2567         cn_id_drbd.val = CN_VAL_DRBD;
2568         do {
2569                 cn_id_drbd.idx = cn_idx;
2570                 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2571                 if (!err)
2572                         break;
2573                 cn_idx = (cn_idx + CN_IDX_STEP);
2574         } while (try--);
2575
2576         if (err) {
2577                 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2578                 return err;
2579         }
2580
2581         return 0;
2582 }
2583
2584 void drbd_nl_cleanup(void)
2585 {
2586         static struct cb_id cn_id_drbd;
2587
2588         cn_id_drbd.idx = cn_idx;
2589         cn_id_drbd.val = CN_VAL_DRBD;
2590
2591         cn_del_callback(&cn_id_drbd);
2592 }
2593
2594 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2595 {
2596         char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2597         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2598         struct drbd_nl_cfg_reply *reply =
2599                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2600         int rr;
2601
2602         memset(buffer, 0, sizeof(buffer));
2603         cn_reply->id = req->id;
2604
2605         cn_reply->seq = req->seq;
2606         cn_reply->ack = req->ack  + 1;
2607         cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2608         cn_reply->flags = 0;
2609
2610         reply->packet_type = P_return_code_only;
2611         reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2612         reply->ret_code = ret_code;
2613
2614         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2615         if (rr && rr != -ESRCH)
2616                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2617 }
2618