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                 drbd_al_shrink(mdev);
 906                 lc_unlock(mdev->act_log);
 907         } else {
 908                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
 909                 return;
 910         }
 911
 912         spin_lock_irq(&mdev->tconn->req_lock);
 913         if (mdev->state.conn < C_CONNECTED)
 914                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
 915
 916         spin_unlock_irq(&mdev->tconn->req_lock);
 917
 918         if (s)
 919                 dev_info(DEV, "Suspended AL updates\n");
 920 }
 921
 922 /* does always return 0;
 923  * interesting return code is in reply->ret_code */
 924 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 925                              struct drbd_nl_cfg_reply *reply)
 926 {
 927         enum drbd_ret_code retcode;
 928         enum determine_dev_size dd;
 929         sector_t max_possible_sectors;
 930         sector_t min_md_device_sectors;
 931         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
 932         struct block_device *bdev;
 933         struct lru_cache *resync_lru = NULL;
 934         union drbd_state ns, os;
 935         enum drbd_state_rv rv;
 936         int cp_discovered = 0;
 937         int logical_block_size;
 938
 939         drbd_reconfig_start(mdev);
 940
 941         /* if you want to reconfigure, please tear down first */
 942         if (mdev->state.disk > D_DISKLESS) {
 943                 retcode = ERR_DISK_CONFIGURED;
 944                 goto fail;
 945         }
 946         /* It may just now have detached because of IO error.  Make sure
 947          * drbd_ldev_destroy is done already, we may end up here very fast,
 948          * e.g. if someone calls attach from the on-io-error handler,
 949          * to realize a "hot spare" feature (not that I'd recommend that) */
 950         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
 951
 952         /* allocation not in the IO path, cqueue thread context */
 953         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
 954         if (!nbc) {
 955                 retcode = ERR_NOMEM;
 956                 goto fail;
 957         }
 958
 959         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
 960         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
 961         nbc->dc.fencing       = DRBD_FENCING_DEF;
 962         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
 963
 964         if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
 965                 retcode = ERR_MANDATORY_TAG;
 966                 goto fail;
 967         }
 968
 969         if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
 970                 retcode = ERR_MD_IDX_INVALID;
 971                 goto fail;
 972         }
 973
 974         if (get_net_conf(mdev->tconn)) {
 975                 int prot = mdev->tconn->net_conf->wire_protocol;
 976                 put_net_conf(mdev->tconn);
 977                 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
 978                         retcode = ERR_STONITH_AND_PROT_A;
 979                         goto fail;
 980                 }
 981         }
 982
 983         bdev = blkdev_get_by_path(nbc->dc.backing_dev,
 984                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
 985         if (IS_ERR(bdev)) {
 986                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
 987                         PTR_ERR(bdev));
 988                 retcode = ERR_OPEN_DISK;
 989                 goto fail;
 990         }
 991         nbc->backing_bdev = bdev;
 992
 993         /*
 994          * meta_dev_idx >= 0: external fixed size, possibly multiple
 995          * drbd sharing one meta device.  TODO in that case, paranoia
 996          * check that [md_bdev, meta_dev_idx] is not yet used by some
 997          * other drbd minor!  (if you use drbd.conf + drbdadm, that
 998          * should check it for you already; but if you don't, or
 999          * someone fooled it, we need to double check here)
1000          */
1001         bdev = blkdev_get_by_path(nbc->dc.meta_dev,
1002                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1003                                   (nbc->dc.meta_dev_idx < 0) ?
1004                                   (void *)mdev : (void *)drbd_m_holder);
1005         if (IS_ERR(bdev)) {
1006                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
1007                         PTR_ERR(bdev));
1008                 retcode = ERR_OPEN_MD_DISK;
1009                 goto fail;
1010         }
1011         nbc->md_bdev = bdev;
1012
1013         if ((nbc->backing_bdev == nbc->md_bdev) !=
1014             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1015              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1016                 retcode = ERR_MD_IDX_INVALID;
1017                 goto fail;
1018         }
1019
1020         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1021                         61, sizeof(struct bm_extent),
1022                         offsetof(struct bm_extent, lce));
1023         if (!resync_lru) {
1024                 retcode = ERR_NOMEM;
1025                 goto fail;
1026         }
1027
1028         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1029         drbd_md_set_sector_offsets(mdev, nbc);
1030
1031         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
1032                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1033                         (unsigned long long) drbd_get_max_capacity(nbc),
1034                         (unsigned long long) nbc->dc.disk_size);
1035                 retcode = ERR_DISK_TO_SMALL;
1036                 goto fail;
1037         }
1038
1039         if (nbc->dc.meta_dev_idx < 0) {
1040                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1041                 /* at least one MB, otherwise it does not make sense */
1042                 min_md_device_sectors = (2<<10);
1043         } else {
1044                 max_possible_sectors = DRBD_MAX_SECTORS;
1045                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
1046         }
1047
1048         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1049                 retcode = ERR_MD_DISK_TO_SMALL;
1050                 dev_warn(DEV, "refusing attach: md-device too small, "
1051                      "at least %llu sectors needed for this meta-disk type\n",
1052                      (unsigned long long) min_md_device_sectors);
1053                 goto fail;
1054         }
1055
1056         /* Make sure the new disk is big enough
1057          * (we may currently be R_PRIMARY with no local disk...) */
1058         if (drbd_get_max_capacity(nbc) <
1059             drbd_get_capacity(mdev->this_bdev)) {
1060                 retcode = ERR_DISK_TO_SMALL;
1061                 goto fail;
1062         }
1063
1064         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1065
1066         if (nbc->known_size > max_possible_sectors) {
1067                 dev_warn(DEV, "==> truncating very big lower level device "
1068                         "to currently maximum possible %llu sectors <==\n",
1069                         (unsigned long long) max_possible_sectors);
1070                 if (nbc->dc.meta_dev_idx >= 0)
1071                         dev_warn(DEV, "==>> using internal or flexible "
1072                                       "meta data may help <<==\n");
1073         }
1074
1075         drbd_suspend_io(mdev);
1076         /* also wait for the last barrier ack. */
1077         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state));
1078         /* and for any other previously queued work */
1079         drbd_flush_workqueue(mdev);
1080
1081         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1082         retcode = rv;  /* FIXME: Type mismatch. */
1083         drbd_resume_io(mdev);
1084         if (rv < SS_SUCCESS)
1085                 goto fail;
1086
1087         if (!get_ldev_if_state(mdev, D_ATTACHING))
1088                 goto force_diskless;
1089
1090         drbd_md_set_sector_offsets(mdev, nbc);
1091
1092         /* allocate a second IO page if logical_block_size != 512 */
1093         logical_block_size = bdev_logical_block_size(nbc->md_bdev);
1094         if (logical_block_size == 0)
1095                 logical_block_size = MD_SECTOR_SIZE;
1096
1097         if (logical_block_size != MD_SECTOR_SIZE) {
1098                 if (!mdev->md_io_tmpp) {
1099                         struct page *page = alloc_page(GFP_NOIO);
1100                         if (!page)
1101                                 goto force_diskless_dec;
1102
1103                         dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
1104                              logical_block_size, MD_SECTOR_SIZE);
1105                         dev_warn(DEV, "Workaround engaged (has performance impact).\n");
1106
1107                         mdev->md_io_tmpp = page;
1108                 }
1109         }
1110
1111         if (!mdev->bitmap) {
1112                 if (drbd_bm_init(mdev)) {
1113                         retcode = ERR_NOMEM;
1114                         goto force_diskless_dec;
1115                 }
1116         }
1117
1118         retcode = drbd_md_read(mdev, nbc);
1119         if (retcode != NO_ERROR)
1120                 goto force_diskless_dec;
1121
1122         if (mdev->state.conn < C_CONNECTED &&
1123             mdev->state.role == R_PRIMARY &&
1124             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1125                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1126                     (unsigned long long)mdev->ed_uuid);
1127                 retcode = ERR_DATA_NOT_CURRENT;
1128                 goto force_diskless_dec;
1129         }
1130
1131         /* Since we are diskless, fix the activity log first... */
1132         if (drbd_check_al_size(mdev)) {
1133                 retcode = ERR_NOMEM;
1134                 goto force_diskless_dec;
1135         }
1136
1137         /* Prevent shrinking of consistent devices ! */
1138         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1139             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1140                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1141                 retcode = ERR_DISK_TO_SMALL;
1142                 goto force_diskless_dec;
1143         }
1144
1145         if (!drbd_al_read_log(mdev, nbc)) {
1146                 retcode = ERR_IO_MD_DISK;
1147                 goto force_diskless_dec;
1148         }
1149
1150         /* Reset the "barriers don't work" bits here, then force meta data to
1151          * be written, to ensure we determine if barriers are supported. */
1152         if (nbc->dc.no_md_flush)
1153                 set_bit(MD_NO_FUA, &mdev->flags);
1154         else
1155                 clear_bit(MD_NO_FUA, &mdev->flags);
1156
1157         /* Point of no return reached.
1158          * Devices and memory are no longer released by error cleanup below.
1159          * now mdev takes over responsibility, and the state engine should
1160          * clean it up somewhere.  */
1161         D_ASSERT(mdev->ldev == NULL);
1162         mdev->ldev = nbc;
1163         mdev->resync = resync_lru;
1164         nbc = NULL;
1165         resync_lru = NULL;
1166
1167         mdev->write_ordering = WO_bdev_flush;
1168         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1169
1170         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1171                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1172         else
1173                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1174
1175         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1176             !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) {
1177                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1178                 cp_discovered = 1;
1179         }
1180
1181         mdev->send_cnt = 0;
1182         mdev->recv_cnt = 0;
1183         mdev->read_cnt = 0;
1184         mdev->writ_cnt = 0;
1185
1186         drbd_reconsider_max_bio_size(mdev);
1187
1188         /* If I am currently not R_PRIMARY,
1189          * but meta data primary indicator is set,
1190          * I just now recover from a hard crash,
1191          * and have been R_PRIMARY before that crash.
1192          *
1193          * Now, if I had no connection before that crash
1194          * (have been degraded R_PRIMARY), chances are that
1195          * I won't find my peer now either.
1196          *
1197          * In that case, and _only_ in that case,
1198          * we use the degr-wfc-timeout instead of the default,
1199          * so we can automatically recover from a crash of a
1200          * degraded but active "cluster" after a certain timeout.
1201          */
1202         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1203         if (mdev->state.role != R_PRIMARY &&
1204              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1205             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1206                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1207
1208         dd = drbd_determine_dev_size(mdev, 0);
1209         if (dd == dev_size_error) {
1210                 retcode = ERR_NOMEM_BITMAP;
1211                 goto force_diskless_dec;
1212         } else if (dd == grew)
1213                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1214
1215         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1216                 dev_info(DEV, "Assuming that all blocks are out of sync "
1217                      "(aka FullSync)\n");
1218                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1219                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1220                         retcode = ERR_IO_MD_DISK;
1221                         goto force_diskless_dec;
1222                 }
1223         } else {
1224                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1225                         "read from attaching", BM_LOCKED_MASK) < 0) {
1226                         retcode = ERR_IO_MD_DISK;
1227                         goto force_diskless_dec;
1228                 }
1229         }
1230
1231         if (cp_discovered) {
1232                 drbd_al_apply_to_bm(mdev);
1233                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1234                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1235                         retcode = ERR_IO_MD_DISK;
1236                         goto force_diskless_dec;
1237                 }
1238         }
1239
1240         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1241                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1242
1243         spin_lock_irq(&mdev->tconn->req_lock);
1244         os = mdev->state;
1245         ns.i = os.i;
1246         /* If MDF_CONSISTENT is not set go into inconsistent state,
1247            otherwise investigate MDF_WasUpToDate...
1248            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1249            otherwise into D_CONSISTENT state.
1250         */
1251         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1252                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1253                         ns.disk = D_CONSISTENT;
1254                 else
1255                         ns.disk = D_OUTDATED;
1256         } else {
1257                 ns.disk = D_INCONSISTENT;
1258         }
1259
1260         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1261                 ns.pdsk = D_OUTDATED;
1262
1263         if ( ns.disk == D_CONSISTENT &&
1264             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1265                 ns.disk = D_UP_TO_DATE;
1266
1267         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1268            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1269            this point, because drbd_request_state() modifies these
1270            flags. */
1271
1272         /* In case we are C_CONNECTED postpone any decision on the new disk
1273            state after the negotiation phase. */
1274         if (mdev->state.conn == C_CONNECTED) {
1275                 mdev->new_state_tmp.i = ns.i;
1276                 ns.i = os.i;
1277                 ns.disk = D_NEGOTIATING;
1278
1279                 /* We expect to receive up-to-date UUIDs soon.
1280                    To avoid a race in receive_state, free p_uuid while
1281                    holding req_lock. I.e. atomic with the state change */
1282                 kfree(mdev->p_uuid);
1283                 mdev->p_uuid = NULL;
1284         }
1285
1286         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1287         ns = mdev->state;
1288         spin_unlock_irq(&mdev->tconn->req_lock);
1289
1290         if (rv < SS_SUCCESS)
1291                 goto force_diskless_dec;
1292
1293         if (mdev->state.role == R_PRIMARY)
1294                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1295         else
1296                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1297
1298         drbd_md_mark_dirty(mdev);
1299         drbd_md_sync(mdev);
1300
1301         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1302         put_ldev(mdev);
1303         reply->ret_code = retcode;
1304         drbd_reconfig_done(mdev);
1305         return 0;
1306
1307  force_diskless_dec:
1308         put_ldev(mdev);
1309  force_diskless:
1310         drbd_force_state(mdev, NS(disk, D_FAILED));
1311         drbd_md_sync(mdev);
1312  fail:
1313         if (nbc) {
1314                 if (nbc->backing_bdev)
1315                         blkdev_put(nbc->backing_bdev,
1316                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1317                 if (nbc->md_bdev)
1318                         blkdev_put(nbc->md_bdev,
1319                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1320                 kfree(nbc);
1321         }
1322         lc_destroy(resync_lru);
1323
1324         reply->ret_code = retcode;
1325         drbd_reconfig_done(mdev);
1326         return 0;
1327 }
1328
1329 /* Detaching the disk is a process in multiple stages.  First we need to lock
1330  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1331  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1332  * internal references as well.
1333  * Only then we have finally detached. */
1334 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1335                           struct drbd_nl_cfg_reply *reply)
1336 {
1337         enum drbd_ret_code retcode;
1338         int ret;
1339         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1340         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1341         /* D_FAILED will transition to DISKLESS. */
1342         ret = wait_event_interruptible(mdev->misc_wait,
1343                         mdev->state.disk != D_FAILED);
1344         drbd_resume_io(mdev);
1345         if ((int)retcode == (int)SS_IS_DISKLESS)
1346                 retcode = SS_NOTHING_TO_DO;
1347         if (ret)
1348                 retcode = ERR_INTR;
1349         reply->ret_code = retcode;
1350         return 0;
1351 }
1352
1353 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1354                             struct drbd_nl_cfg_reply *reply)
1355 {
1356         int i;
1357         enum drbd_ret_code retcode;
1358         struct net_conf *new_conf = NULL;
1359         struct crypto_hash *tfm = NULL;
1360         struct crypto_hash *integrity_w_tfm = NULL;
1361         struct crypto_hash *integrity_r_tfm = NULL;
1362         struct drbd_conf *odev;
1363         char hmac_name[CRYPTO_MAX_ALG_NAME];
1364         void *int_dig_out = NULL;
1365         void *int_dig_in = NULL;
1366         void *int_dig_vv = NULL;
1367         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1368
1369         drbd_reconfig_start(mdev);
1370
1371         if (mdev->state.conn > C_STANDALONE) {
1372                 retcode = ERR_NET_CONFIGURED;
1373                 goto fail;
1374         }
1375
1376         /* allocation not in the IO path, cqueue thread context */
1377         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1378         if (!new_conf) {
1379                 retcode = ERR_NOMEM;
1380                 goto fail;
1381         }
1382
1383         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1384         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1385         new_conf->ping_int         = DRBD_PING_INT_DEF;
1386         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1387         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1388         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1389         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1390         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1391         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1392         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1393         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1394         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1395         new_conf->want_lose        = 0;
1396         new_conf->two_primaries    = 0;
1397         new_conf->wire_protocol    = DRBD_PROT_C;
1398         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1399         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1400         new_conf->on_congestion    = DRBD_ON_CONGESTION_DEF;
1401         new_conf->cong_extents     = DRBD_CONG_EXTENTS_DEF;
1402
1403         if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1404                 retcode = ERR_MANDATORY_TAG;
1405                 goto fail;
1406         }
1407
1408         if (new_conf->two_primaries
1409             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1410                 retcode = ERR_NOT_PROTO_C;
1411                 goto fail;
1412         }
1413
1414         if (get_ldev(mdev)) {
1415                 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1416                 put_ldev(mdev);
1417                 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1418                         retcode = ERR_STONITH_AND_PROT_A;
1419                         goto fail;
1420                 }
1421         }
1422
1423         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1424                 retcode = ERR_CONG_NOT_PROTO_A;
1425                 goto fail;
1426         }
1427
1428         if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1429                 retcode = ERR_DISCARD;
1430                 goto fail;
1431         }
1432
1433         retcode = NO_ERROR;
1434
1435         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1436         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1437         for (i = 0; i < minor_count; i++) {
1438                 odev = minor_to_mdev(i);
1439                 if (!odev || odev == mdev)
1440                         continue;
1441                 if (get_net_conf(odev->tconn)) {
1442                         taken_addr = (struct sockaddr *)&odev->tconn->net_conf->my_addr;
1443                         if (new_conf->my_addr_len == odev->tconn->net_conf->my_addr_len &&
1444                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1445                                 retcode = ERR_LOCAL_ADDR;
1446
1447                         taken_addr = (struct sockaddr *)&odev->tconn->net_conf->peer_addr;
1448                         if (new_conf->peer_addr_len == odev->tconn->net_conf->peer_addr_len &&
1449                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1450                                 retcode = ERR_PEER_ADDR;
1451
1452                         put_net_conf(odev->tconn);
1453                         if (retcode != NO_ERROR)
1454                                 goto fail;
1455                 }
1456         }
1457
1458         if (new_conf->cram_hmac_alg[0] != 0) {
1459                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1460                         new_conf->cram_hmac_alg);
1461                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1462                 if (IS_ERR(tfm)) {
1463                         tfm = NULL;
1464                         retcode = ERR_AUTH_ALG;
1465                         goto fail;
1466                 }
1467
1468                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1469                         retcode = ERR_AUTH_ALG_ND;
1470                         goto fail;
1471                 }
1472         }
1473
1474         if (new_conf->integrity_alg[0]) {
1475                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1476                 if (IS_ERR(integrity_w_tfm)) {
1477                         integrity_w_tfm = NULL;
1478                         retcode=ERR_INTEGRITY_ALG;
1479                         goto fail;
1480                 }
1481
1482                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1483                         retcode=ERR_INTEGRITY_ALG_ND;
1484                         goto fail;
1485                 }
1486
1487                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1488                 if (IS_ERR(integrity_r_tfm)) {
1489                         integrity_r_tfm = NULL;
1490                         retcode=ERR_INTEGRITY_ALG;
1491                         goto fail;
1492                 }
1493         }
1494
1495         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1496
1497         if (integrity_w_tfm) {
1498                 i = crypto_hash_digestsize(integrity_w_tfm);
1499                 int_dig_out = kmalloc(i, GFP_KERNEL);
1500                 if (!int_dig_out) {
1501                         retcode = ERR_NOMEM;
1502                         goto fail;
1503                 }
1504                 int_dig_in = kmalloc(i, GFP_KERNEL);
1505                 if (!int_dig_in) {
1506                         retcode = ERR_NOMEM;
1507                         goto fail;
1508                 }
1509                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1510                 if (!int_dig_vv) {
1511                         retcode = ERR_NOMEM;
1512                         goto fail;
1513                 }
1514         }
1515
1516         if (!mdev->bitmap) {
1517                 if(drbd_bm_init(mdev)) {
1518                         retcode = ERR_NOMEM;
1519                         goto fail;
1520                 }
1521         }
1522
1523         drbd_flush_workqueue(mdev);
1524         spin_lock_irq(&mdev->tconn->req_lock);
1525         if (mdev->tconn->net_conf != NULL) {
1526                 retcode = ERR_NET_CONFIGURED;
1527                 spin_unlock_irq(&mdev->tconn->req_lock);
1528                 goto fail;
1529         }
1530         mdev->tconn->net_conf = new_conf;
1531
1532         mdev->send_cnt = 0;
1533         mdev->recv_cnt = 0;
1534
1535         crypto_free_hash(mdev->tconn->cram_hmac_tfm);
1536         mdev->tconn->cram_hmac_tfm = tfm;
1537
1538         crypto_free_hash(mdev->tconn->integrity_w_tfm);
1539         mdev->tconn->integrity_w_tfm = integrity_w_tfm;
1540
1541         crypto_free_hash(mdev->tconn->integrity_r_tfm);
1542         mdev->tconn->integrity_r_tfm = integrity_r_tfm;
1543
1544         kfree(mdev->tconn->int_dig_out);
1545         kfree(mdev->tconn->int_dig_in);
1546         kfree(mdev->tconn->int_dig_vv);
1547         mdev->tconn->int_dig_out=int_dig_out;
1548         mdev->tconn->int_dig_in=int_dig_in;
1549         mdev->tconn->int_dig_vv=int_dig_vv;
1550         retcode = _conn_request_state(mdev->tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
1551         spin_unlock_irq(&mdev->tconn->req_lock);
1552
1553         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1554         reply->ret_code = retcode;
1555         drbd_reconfig_done(mdev);
1556         return 0;
1557
1558 fail:
1559         kfree(int_dig_out);
1560         kfree(int_dig_in);
1561         kfree(int_dig_vv);
1562         crypto_free_hash(tfm);
1563         crypto_free_hash(integrity_w_tfm);
1564         crypto_free_hash(integrity_r_tfm);
1565         kfree(new_conf);
1566
1567         reply->ret_code = retcode;
1568         drbd_reconfig_done(mdev);
1569         return 0;
1570 }
1571
1572 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1573                               struct drbd_nl_cfg_reply *reply)
1574 {
1575         int retcode;
1576         struct disconnect dc;
1577
1578         memset(&dc, 0, sizeof(struct disconnect));
1579         if (!disconnect_from_tags(mdev, nlp->tag_list, &dc)) {
1580                 retcode = ERR_MANDATORY_TAG;
1581                 goto fail;
1582         }
1583
1584         if (dc.force) {
1585                 spin_lock_irq(&mdev->tconn->req_lock);
1586                 if (mdev->state.conn >= C_WF_CONNECTION)
1587                         _drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), CS_HARD, NULL);
1588                 spin_unlock_irq(&mdev->tconn->req_lock);
1589                 goto done;
1590         }
1591
1592         retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1593
1594         if (retcode == SS_NOTHING_TO_DO)
1595                 goto done;
1596         else if (retcode == SS_ALREADY_STANDALONE)
1597                 goto done;
1598         else if (retcode == SS_PRIMARY_NOP) {
1599                 /* Our statche checking code wants to see the peer outdated. */
1600                 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1601                                                       pdsk, D_OUTDATED));
1602         } else if (retcode == SS_CW_FAILED_BY_PEER) {
1603                 /* The peer probably wants to see us outdated. */
1604                 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1605                                                         disk, D_OUTDATED),
1606                                               CS_ORDERED);
1607                 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1608                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1609                         retcode = SS_SUCCESS;
1610                 }
1611         }
1612
1613         if (retcode < SS_SUCCESS)
1614                 goto fail;
1615
1616         if (wait_event_interruptible(mdev->state_wait,
1617                                      mdev->state.conn != C_DISCONNECTING)) {
1618                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1619                    someone else might connect us in the mean time! */
1620                 retcode = ERR_INTR;
1621                 goto fail;
1622         }
1623
1624  done:
1625         retcode = NO_ERROR;
1626  fail:
1627         drbd_md_sync(mdev);
1628         reply->ret_code = retcode;
1629         return 0;
1630 }
1631
1632 void resync_after_online_grow(struct drbd_conf *mdev)
1633 {
1634         int iass; /* I am sync source */
1635
1636         dev_info(DEV, "Resync of new storage after online grow\n");
1637         if (mdev->state.role != mdev->state.peer)
1638                 iass = (mdev->state.role == R_PRIMARY);
1639         else
1640                 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
1641
1642         if (iass)
1643                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1644         else
1645                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1646 }
1647
1648 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1649                           struct drbd_nl_cfg_reply *reply)
1650 {
1651         struct resize rs;
1652         int retcode = NO_ERROR;
1653         enum determine_dev_size dd;
1654         enum dds_flags ddsf;
1655
1656         memset(&rs, 0, sizeof(struct resize));
1657         if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1658                 retcode = ERR_MANDATORY_TAG;
1659                 goto fail;
1660         }
1661
1662         if (mdev->state.conn > C_CONNECTED) {
1663                 retcode = ERR_RESIZE_RESYNC;
1664                 goto fail;
1665         }
1666
1667         if (mdev->state.role == R_SECONDARY &&
1668             mdev->state.peer == R_SECONDARY) {
1669                 retcode = ERR_NO_PRIMARY;
1670                 goto fail;
1671         }
1672
1673         if (!get_ldev(mdev)) {
1674                 retcode = ERR_NO_DISK;
1675                 goto fail;
1676         }
1677
1678         if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
1679                 retcode = ERR_NEED_APV_93;
1680                 goto fail;
1681         }
1682
1683         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1684                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1685
1686         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1687         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1688         dd = drbd_determine_dev_size(mdev, ddsf);
1689         drbd_md_sync(mdev);
1690         put_ldev(mdev);
1691         if (dd == dev_size_error) {
1692                 retcode = ERR_NOMEM_BITMAP;
1693                 goto fail;
1694         }
1695
1696         if (mdev->state.conn == C_CONNECTED) {
1697                 if (dd == grew)
1698                         set_bit(RESIZE_PENDING, &mdev->flags);
1699
1700                 drbd_send_uuids(mdev);
1701                 drbd_send_sizes(mdev, 1, ddsf);
1702         }
1703
1704  fail:
1705         reply->ret_code = retcode;
1706         return 0;
1707 }
1708
1709 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1710                                struct drbd_nl_cfg_reply *reply)
1711 {
1712         int retcode = NO_ERROR;
1713         int err;
1714         int ovr; /* online verify running */
1715         int rsr; /* re-sync running */
1716         struct crypto_hash *verify_tfm = NULL;
1717         struct crypto_hash *csums_tfm = NULL;
1718         struct syncer_conf sc;
1719         cpumask_var_t new_cpu_mask;
1720         int *rs_plan_s = NULL;
1721         int fifo_size;
1722
1723         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1724                 retcode = ERR_NOMEM;
1725                 goto fail;
1726         }
1727
1728         if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1729                 memset(&sc, 0, sizeof(struct syncer_conf));
1730                 sc.rate       = DRBD_RATE_DEF;
1731                 sc.after      = DRBD_AFTER_DEF;
1732                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1733                 sc.on_no_data  = DRBD_ON_NO_DATA_DEF;
1734                 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1735                 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1736                 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1737                 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1738                 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1739         } else
1740                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1741
1742         if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1743                 retcode = ERR_MANDATORY_TAG;
1744                 goto fail;
1745         }
1746
1747         /* re-sync running */
1748         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1749                 mdev->state.conn == C_SYNC_TARGET ||
1750                 mdev->state.conn == C_PAUSED_SYNC_S ||
1751                 mdev->state.conn == C_PAUSED_SYNC_T );
1752
1753         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1754                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1755                 goto fail;
1756         }
1757
1758         if (!rsr && sc.csums_alg[0]) {
1759                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1760                 if (IS_ERR(csums_tfm)) {
1761                         csums_tfm = NULL;
1762                         retcode = ERR_CSUMS_ALG;
1763                         goto fail;
1764                 }
1765
1766                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1767                         retcode = ERR_CSUMS_ALG_ND;
1768                         goto fail;
1769                 }
1770         }
1771
1772         /* online verify running */
1773         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1774
1775         if (ovr) {
1776                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1777                         retcode = ERR_VERIFY_RUNNING;
1778                         goto fail;
1779                 }
1780         }
1781
1782         if (!ovr && sc.verify_alg[0]) {
1783                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1784                 if (IS_ERR(verify_tfm)) {
1785                         verify_tfm = NULL;
1786                         retcode = ERR_VERIFY_ALG;
1787                         goto fail;
1788                 }
1789
1790                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1791                         retcode = ERR_VERIFY_ALG_ND;
1792                         goto fail;
1793                 }
1794         }
1795
1796         /* silently ignore cpu mask on UP kernel */
1797         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1798                 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1799                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1800                 if (err) {
1801                         dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1802                         retcode = ERR_CPU_MASK_PARSE;
1803                         goto fail;
1804                 }
1805         }
1806
1807         if (!expect(sc.rate >= 1))
1808                 sc.rate = 1;
1809         if (!expect(sc.al_extents >= 7))
1810                 sc.al_extents = 127; /* arbitrary minimum */
1811 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1812         if (sc.al_extents > AL_MAX) {
1813                 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1814                 sc.al_extents = AL_MAX;
1815         }
1816 #undef AL_MAX
1817
1818         /* to avoid spurious errors when configuring minors before configuring
1819          * the minors they depend on: if necessary, first create the minor we
1820          * depend on */
1821         if (sc.after >= 0)
1822                 ensure_mdev(sc.after, 1);
1823
1824         /* most sanity checks done, try to assign the new sync-after
1825          * dependency.  need to hold the global lock in there,
1826          * to avoid a race in the dependency loop check. */
1827         retcode = drbd_alter_sa(mdev, sc.after);
1828         if (retcode != NO_ERROR)
1829                 goto fail;
1830
1831         fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1832         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1833                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1834                 if (!rs_plan_s) {
1835                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1836                         retcode = ERR_NOMEM;
1837                         goto fail;
1838                 }
1839         }
1840
1841         /* ok, assign the rest of it as well.
1842          * lock against receive_SyncParam() */
1843         spin_lock(&mdev->peer_seq_lock);
1844         mdev->sync_conf = sc;
1845
1846         if (!rsr) {
1847                 crypto_free_hash(mdev->csums_tfm);
1848                 mdev->csums_tfm = csums_tfm;
1849                 csums_tfm = NULL;
1850         }
1851
1852         if (!ovr) {
1853                 crypto_free_hash(mdev->verify_tfm);
1854                 mdev->verify_tfm = verify_tfm;
1855                 verify_tfm = NULL;
1856         }
1857
1858         if (fifo_size != mdev->rs_plan_s.size) {
1859                 kfree(mdev->rs_plan_s.values);
1860                 mdev->rs_plan_s.values = rs_plan_s;
1861                 mdev->rs_plan_s.size   = fifo_size;
1862                 mdev->rs_planed = 0;
1863                 rs_plan_s = NULL;
1864         }
1865
1866         spin_unlock(&mdev->peer_seq_lock);
1867
1868         if (get_ldev(mdev)) {
1869                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1870                 drbd_al_shrink(mdev);
1871                 err = drbd_check_al_size(mdev);
1872                 lc_unlock(mdev->act_log);
1873                 wake_up(&mdev->al_wait);
1874
1875                 put_ldev(mdev);
1876                 drbd_md_sync(mdev);
1877
1878                 if (err) {
1879                         retcode = ERR_NOMEM;
1880                         goto fail;
1881                 }
1882         }
1883
1884         if (mdev->state.conn >= C_CONNECTED)
1885                 drbd_send_sync_param(mdev, &sc);
1886
1887         if (!cpumask_equal(mdev->tconn->cpu_mask, new_cpu_mask)) {
1888                 cpumask_copy(mdev->tconn->cpu_mask, new_cpu_mask);
1889                 drbd_calc_cpu_mask(mdev->tconn);
1890                 mdev->tconn->receiver.reset_cpu_mask = 1;
1891                 mdev->tconn->asender.reset_cpu_mask = 1;
1892                 mdev->tconn->worker.reset_cpu_mask = 1;
1893         }
1894
1895         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1896 fail:
1897         kfree(rs_plan_s);
1898         free_cpumask_var(new_cpu_mask);
1899         crypto_free_hash(csums_tfm);
1900         crypto_free_hash(verify_tfm);
1901         reply->ret_code = retcode;
1902         return 0;
1903 }
1904
1905 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1906                               struct drbd_nl_cfg_reply *reply)
1907 {
1908         int retcode;
1909
1910         /* If there is still bitmap IO pending, probably because of a previous
1911          * resync just being finished, wait for it before requesting a new resync. */
1912         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1913
1914         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1915
1916         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1917                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1918
1919         while (retcode == SS_NEED_CONNECTION) {
1920                 spin_lock_irq(&mdev->tconn->req_lock);
1921                 if (mdev->state.conn < C_CONNECTED)
1922                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1923                 spin_unlock_irq(&mdev->tconn->req_lock);
1924
1925                 if (retcode != SS_NEED_CONNECTION)
1926                         break;
1927
1928                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1929         }
1930
1931         reply->ret_code = retcode;
1932         return 0;
1933 }
1934
1935 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
1936 {
1937         int rv;
1938
1939         rv = drbd_bmio_set_n_write(mdev);
1940         drbd_suspend_al(mdev);
1941         return rv;
1942 }
1943
1944 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1945                                    struct drbd_nl_cfg_reply *reply)
1946 {
1947         int retcode;
1948
1949         /* If there is still bitmap IO pending, probably because of a previous
1950          * resync just being finished, wait for it before requesting a new resync. */
1951         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1952
1953         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
1954
1955         if (retcode < SS_SUCCESS) {
1956                 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
1957                         /* The peer will get a resync upon connect anyways. Just make that
1958                            into a full resync. */
1959                         retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
1960                         if (retcode >= SS_SUCCESS) {
1961                                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
1962                                         "set_n_write from invalidate_peer",
1963                                         BM_LOCKED_SET_ALLOWED))
1964                                         retcode = ERR_IO_MD_DISK;
1965                         }
1966                 } else
1967                         retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
1968         }
1969
1970         reply->ret_code = retcode;
1971         return 0;
1972 }
1973
1974 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1975                               struct drbd_nl_cfg_reply *reply)
1976 {
1977         int retcode = NO_ERROR;
1978
1979         if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
1980                 retcode = ERR_PAUSE_IS_SET;
1981
1982         reply->ret_code = retcode;
1983         return 0;
1984 }
1985
1986 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1987                                struct drbd_nl_cfg_reply *reply)
1988 {
1989         int retcode = NO_ERROR;
1990         union drbd_state s;
1991
1992         if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
1993                 s = mdev->state;
1994                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
1995                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
1996                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
1997                 } else {
1998                         retcode = ERR_PAUSE_IS_CLEAR;
1999                 }
2000         }
2001
2002         reply->ret_code = retcode;
2003         return 0;
2004 }
2005
2006 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2007                               struct drbd_nl_cfg_reply *reply)
2008 {
2009         reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
2010
2011         return 0;
2012 }
2013
2014 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2015                              struct drbd_nl_cfg_reply *reply)
2016 {
2017         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2018                 drbd_uuid_new_current(mdev);
2019                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2020         }
2021         drbd_suspend_io(mdev);
2022         reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2023         if (reply->ret_code == SS_SUCCESS) {
2024                 if (mdev->state.conn < C_CONNECTED)
2025                         tl_clear(mdev);
2026                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2027                         tl_restart(mdev, FAIL_FROZEN_DISK_IO);
2028         }
2029         drbd_resume_io(mdev);
2030
2031         return 0;
2032 }
2033
2034 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2035                            struct drbd_nl_cfg_reply *reply)
2036 {
2037         reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
2038         return 0;
2039 }
2040
2041 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2042                            struct drbd_nl_cfg_reply *reply)
2043 {
2044         unsigned short *tl;
2045
2046         tl = reply->tag_list;
2047
2048         if (get_ldev(mdev)) {
2049                 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
2050                 put_ldev(mdev);
2051         }
2052
2053         if (get_net_conf(mdev->tconn)) {
2054                 tl = net_conf_to_tags(mdev, mdev->tconn->net_conf, tl);
2055                 put_net_conf(mdev->tconn);
2056         }
2057         tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
2058
2059         put_unaligned(TT_END, tl++); /* Close the tag list */
2060
2061         return (int)((char *)tl - (char *)reply->tag_list);
2062 }
2063
2064 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2065                              struct drbd_nl_cfg_reply *reply)
2066 {
2067         unsigned short *tl = reply->tag_list;
2068         union drbd_state s = mdev->state;
2069         unsigned long rs_left;
2070         unsigned int res;
2071
2072         tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
2073
2074         /* no local ref, no bitmap, no syncer progress. */
2075         if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
2076                 if (get_ldev(mdev)) {
2077                         drbd_get_syncer_progress(mdev, &rs_left, &res);
2078                         tl = tl_add_int(tl, T_sync_progress, &res);
2079                         put_ldev(mdev);
2080                 }
2081         }
2082         put_unaligned(TT_END, tl++); /* Close the tag list */
2083
2084         return (int)((char *)tl - (char *)reply->tag_list);
2085 }
2086
2087 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2088                              struct drbd_nl_cfg_reply *reply)
2089 {
2090         unsigned short *tl;
2091
2092         tl = reply->tag_list;
2093
2094         if (get_ldev(mdev)) {
2095                 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
2096                 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
2097                 put_ldev(mdev);
2098         }
2099         put_unaligned(TT_END, tl++); /* Close the tag list */
2100
2101         return (int)((char *)tl - (char *)reply->tag_list);
2102 }
2103
2104 /**
2105  * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
2106  * @mdev:       DRBD device.
2107  * @nlp:        Netlink/connector packet from drbdsetup
2108  * @reply:      Reply packet for drbdsetup
2109  */
2110 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2111                                     struct drbd_nl_cfg_reply *reply)
2112 {
2113         unsigned short *tl;
2114         char rv;
2115
2116         tl = reply->tag_list;
2117
2118         rv = mdev->state.pdsk == D_OUTDATED        ? UT_PEER_OUTDATED :
2119           test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
2120
2121         tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
2122         put_unaligned(TT_END, tl++); /* Close the tag list */
2123
2124         return (int)((char *)tl - (char *)reply->tag_list);
2125 }
2126
2127 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2128                                     struct drbd_nl_cfg_reply *reply)
2129 {
2130         /* default to resume from last known position, if possible */
2131         struct start_ov args =
2132                 { .start_sector = mdev->ov_start_sector };
2133
2134         if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2135                 reply->ret_code = ERR_MANDATORY_TAG;
2136                 return 0;
2137         }
2138
2139         /* If there is still bitmap IO pending, e.g. previous resync or verify
2140          * just being finished, wait for it before requesting a new resync. */
2141         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2142
2143         /* w_make_ov_request expects position to be aligned */
2144         mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
2145         reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2146         return 0;
2147 }
2148
2149
2150 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2151                               struct drbd_nl_cfg_reply *reply)
2152 {
2153         int retcode = NO_ERROR;
2154         int skip_initial_sync = 0;
2155         int err;
2156
2157         struct new_c_uuid args;
2158
2159         memset(&args, 0, sizeof(struct new_c_uuid));
2160         if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2161                 reply->ret_code = ERR_MANDATORY_TAG;
2162                 return 0;
2163         }
2164
2165         mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2166
2167         if (!get_ldev(mdev)) {
2168                 retcode = ERR_NO_DISK;
2169                 goto out;
2170         }
2171
2172         /* this is "skip initial sync", assume to be clean */
2173         if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2174             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2175                 dev_info(DEV, "Preparing to skip initial sync\n");
2176                 skip_initial_sync = 1;
2177         } else if (mdev->state.conn != C_STANDALONE) {
2178                 retcode = ERR_CONNECTED;
2179                 goto out_dec;
2180         }
2181
2182         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2183         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2184
2185         if (args.clear_bm) {
2186                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2187                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2188                 if (err) {
2189                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2190                         retcode = ERR_IO_MD_DISK;
2191                 }
2192                 if (skip_initial_sync) {
2193                         drbd_send_uuids_skip_initial_sync(mdev);
2194                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2195                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2196                         spin_lock_irq(&mdev->tconn->req_lock);
2197                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2198                                         CS_VERBOSE, NULL);
2199                         spin_unlock_irq(&mdev->tconn->req_lock);
2200                 }
2201         }
2202
2203         drbd_md_sync(mdev);
2204 out_dec:
2205         put_ldev(mdev);
2206 out:
2207         mutex_unlock(mdev->state_mutex);
2208
2209         reply->ret_code = retcode;
2210         return 0;
2211 }
2212
2213 struct cn_handler_struct {
2214         int (*function)(struct drbd_conf *,
2215                          struct drbd_nl_cfg_req *,
2216                          struct drbd_nl_cfg_reply *);
2217         int reply_body_size;
2218 };
2219
2220 static struct cn_handler_struct cnd_table[] = {
2221         [ P_primary ]           = { &drbd_nl_primary,           0 },
2222         [ P_secondary ]         = { &drbd_nl_secondary,         0 },
2223         [ P_disk_conf ]         = { &drbd_nl_disk_conf,         0 },
2224         [ P_detach ]            = { &drbd_nl_detach,            0 },
2225         [ P_net_conf ]          = { &drbd_nl_net_conf,          0 },
2226         [ P_disconnect ]        = { &drbd_nl_disconnect,        0 },
2227         [ P_resize ]            = { &drbd_nl_resize,            0 },
2228         [ P_syncer_conf ]       = { &drbd_nl_syncer_conf,       0 },
2229         [ P_invalidate ]        = { &drbd_nl_invalidate,        0 },
2230         [ P_invalidate_peer ]   = { &drbd_nl_invalidate_peer,   0 },
2231         [ P_pause_sync ]        = { &drbd_nl_pause_sync,        0 },
2232         [ P_resume_sync ]       = { &drbd_nl_resume_sync,       0 },
2233         [ P_suspend_io ]        = { &drbd_nl_suspend_io,        0 },
2234         [ P_resume_io ]         = { &drbd_nl_resume_io,         0 },
2235         [ P_outdate ]           = { &drbd_nl_outdate,           0 },
2236         [ P_get_config ]        = { &drbd_nl_get_config,
2237                                     sizeof(struct syncer_conf_tag_len_struct) +
2238                                     sizeof(struct disk_conf_tag_len_struct) +
2239                                     sizeof(struct net_conf_tag_len_struct) },
2240         [ P_get_state ]         = { &drbd_nl_get_state,
2241                                     sizeof(struct get_state_tag_len_struct) +
2242                                     sizeof(struct sync_progress_tag_len_struct) },
2243         [ P_get_uuids ]         = { &drbd_nl_get_uuids,
2244                                     sizeof(struct get_uuids_tag_len_struct) },
2245         [ P_get_timeout_flag ]  = { &drbd_nl_get_timeout_flag,
2246                                     sizeof(struct get_timeout_flag_tag_len_struct)},
2247         [ P_start_ov ]          = { &drbd_nl_start_ov,          0 },
2248         [ P_new_c_uuid ]        = { &drbd_nl_new_c_uuid,        0 },
2249 };
2250
2251 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2252 {
2253         struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2254         struct cn_handler_struct *cm;
2255         struct cn_msg *cn_reply;
2256         struct drbd_nl_cfg_reply *reply;
2257         struct drbd_conf *mdev;
2258         int retcode, rr;
2259         int reply_size = sizeof(struct cn_msg)
2260                 + sizeof(struct drbd_nl_cfg_reply)
2261                 + sizeof(short int);
2262
2263         if (!try_module_get(THIS_MODULE)) {
2264                 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2265                 return;
2266         }
2267
2268         if (!cap_raised(current_cap(), CAP_SYS_ADMIN)) {
2269                 retcode = ERR_PERM;
2270                 goto fail;
2271         }
2272
2273         mdev = ensure_mdev(nlp->drbd_minor,
2274                         (nlp->flags & DRBD_NL_CREATE_DEVICE));
2275         if (!mdev) {
2276                 retcode = ERR_MINOR_INVALID;
2277                 goto fail;
2278         }
2279
2280         if (nlp->packet_type >= P_nl_after_last_packet ||
2281             nlp->packet_type == P_return_code_only) {
2282                 retcode = ERR_PACKET_NR;
2283                 goto fail;
2284         }
2285
2286         cm = cnd_table + nlp->packet_type;
2287
2288         /* This may happen if packet number is 0: */
2289         if (cm->function == NULL) {
2290                 retcode = ERR_PACKET_NR;
2291                 goto fail;
2292         }
2293
2294         reply_size += cm->reply_body_size;
2295
2296         /* allocation not in the IO path, cqueue thread context */
2297         cn_reply = kzalloc(reply_size, GFP_KERNEL);
2298         if (!cn_reply) {
2299                 retcode = ERR_NOMEM;
2300                 goto fail;
2301         }
2302         reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2303
2304         reply->packet_type =
2305                 cm->reply_body_size ? nlp->packet_type : P_return_code_only;
2306         reply->minor = nlp->drbd_minor;
2307         reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2308         /* reply->tag_list; might be modified by cm->function. */
2309
2310         rr = cm->function(mdev, nlp, reply);
2311
2312         cn_reply->id = req->id;
2313         cn_reply->seq = req->seq;
2314         cn_reply->ack = req->ack  + 1;
2315         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2316         cn_reply->flags = 0;
2317
2318         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2319         if (rr && rr != -ESRCH)
2320                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2321
2322         kfree(cn_reply);
2323         module_put(THIS_MODULE);
2324         return;
2325  fail:
2326         drbd_nl_send_reply(req, retcode);
2327         module_put(THIS_MODULE);
2328 }
2329
2330 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2331
2332 static unsigned short *
2333 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2334         unsigned short len, int nul_terminated)
2335 {
2336         unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2337         len = (len < l) ? len :  l;
2338         put_unaligned(tag, tl++);
2339         put_unaligned(len, tl++);
2340         memcpy(tl, data, len);
2341         tl = (unsigned short*)((char*)tl + len);
2342         if (nul_terminated)
2343                 *((char*)tl - 1) = 0;
2344         return tl;
2345 }
2346
2347 static unsigned short *
2348 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2349 {
2350         return __tl_add_blob(tl, tag, data, len, 0);
2351 }
2352
2353 static unsigned short *
2354 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2355 {
2356         return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2357 }
2358
2359 static unsigned short *
2360 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2361 {
2362         put_unaligned(tag, tl++);
2363         switch(tag_type(tag)) {
2364         case TT_INTEGER:
2365                 put_unaligned(sizeof(int), tl++);
2366                 put_unaligned(*(int *)val, (int *)tl);
2367                 tl = (unsigned short*)((char*)tl+sizeof(int));
2368                 break;
2369         case TT_INT64:
2370                 put_unaligned(sizeof(u64), tl++);
2371                 put_unaligned(*(u64 *)val, (u64 *)tl);
2372                 tl = (unsigned short*)((char*)tl+sizeof(u64));
2373                 break;
2374         default:
2375                 /* someone did something stupid. */
2376                 ;
2377         }
2378         return tl;
2379 }
2380
2381 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2382 {
2383         char buffer[sizeof(struct cn_msg)+
2384                     sizeof(struct drbd_nl_cfg_reply)+
2385                     sizeof(struct get_state_tag_len_struct)+
2386                     sizeof(short int)];
2387         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2388         struct drbd_nl_cfg_reply *reply =
2389                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2390         unsigned short *tl = reply->tag_list;
2391
2392         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2393
2394         tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2395
2396         put_unaligned(TT_END, tl++); /* Close the tag list */
2397
2398         cn_reply->id.idx = CN_IDX_DRBD;
2399         cn_reply->id.val = CN_VAL_DRBD;
2400
2401         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2402         cn_reply->ack = 0; /* not used here. */
2403         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2404                 (int)((char *)tl - (char *)reply->tag_list);
2405         cn_reply->flags = 0;
2406
2407         reply->packet_type = P_get_state;
2408         reply->minor = mdev_to_minor(mdev);
2409         reply->ret_code = NO_ERROR;
2410
2411         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2412 }
2413
2414 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2415 {
2416         char buffer[sizeof(struct cn_msg)+
2417                     sizeof(struct drbd_nl_cfg_reply)+
2418                     sizeof(struct call_helper_tag_len_struct)+
2419                     sizeof(short int)];
2420         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2421         struct drbd_nl_cfg_reply *reply =
2422                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2423         unsigned short *tl = reply->tag_list;
2424
2425         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2426
2427         tl = tl_add_str(tl, T_helper, helper_name);
2428         put_unaligned(TT_END, tl++); /* Close the tag list */
2429
2430         cn_reply->id.idx = CN_IDX_DRBD;
2431         cn_reply->id.val = CN_VAL_DRBD;
2432
2433         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2434         cn_reply->ack = 0; /* not used here. */
2435         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2436                 (int)((char *)tl - (char *)reply->tag_list);
2437         cn_reply->flags = 0;
2438
2439         reply->packet_type = P_call_helper;
2440         reply->minor = mdev_to_minor(mdev);
2441         reply->ret_code = NO_ERROR;
2442
2443         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2444 }
2445
2446 void drbd_bcast_ee(struct drbd_conf *mdev, const char *reason, const int dgs,
2447                    const char *seen_hash, const char *calc_hash,
2448                            const struct drbd_peer_request *peer_req)
2449 {
2450         struct cn_msg *cn_reply;
2451         struct drbd_nl_cfg_reply *reply;
2452         unsigned short *tl;
2453         struct page *page;
2454         unsigned len;
2455
2456         if (!peer_req)
2457                 return;
2458         if (!reason || !reason[0])
2459                 return;
2460
2461         /* apparently we have to memcpy twice, first to prepare the data for the
2462          * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2463          * netlink skb. */
2464         /* receiver thread context, which is not in the writeout path (of this node),
2465          * but may be in the writeout path of the _other_ node.
2466          * GFP_NOIO to avoid potential "distributed deadlock". */
2467         cn_reply = kzalloc(
2468                 sizeof(struct cn_msg)+
2469                 sizeof(struct drbd_nl_cfg_reply)+
2470                 sizeof(struct dump_ee_tag_len_struct)+
2471                 sizeof(short int),
2472                 GFP_NOIO);
2473
2474         if (!cn_reply) {
2475                 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, "
2476                              "sector %llu, size %u\n",
2477                         (unsigned long long)peer_req->i.sector,
2478                         peer_req->i.size);
2479                 return;
2480         }
2481
2482         reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2483         tl = reply->tag_list;
2484
2485         tl = tl_add_str(tl, T_dump_ee_reason, reason);
2486         tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2487         tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2488         tl = tl_add_int(tl, T_ee_sector, &peer_req->i.sector);
2489         tl = tl_add_int(tl, T_ee_block_id, &peer_req->block_id);
2490
2491         /* dump the first 32k */
2492         len = min_t(unsigned, peer_req->i.size, 32 << 10);
2493         put_unaligned(T_ee_data, tl++);
2494         put_unaligned(len, tl++);
2495
2496         page = peer_req->pages;
2497         page_chain_for_each(page) {
2498                 void *d = kmap_atomic(page, KM_USER0);
2499                 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2500                 memcpy(tl, d, l);
2501                 kunmap_atomic(d, KM_USER0);
2502                 tl = (unsigned short*)((char*)tl + l);
2503                 len -= l;
2504                 if (len == 0)
2505                         break;
2506         }
2507         put_unaligned(TT_END, tl++); /* Close the tag list */
2508
2509         cn_reply->id.idx = CN_IDX_DRBD;
2510         cn_reply->id.val = CN_VAL_DRBD;
2511
2512         cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2513         cn_reply->ack = 0; // not used here.
2514         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2515                 (int)((char*)tl - (char*)reply->tag_list);
2516         cn_reply->flags = 0;
2517
2518         reply->packet_type = P_dump_ee;
2519         reply->minor = mdev_to_minor(mdev);
2520         reply->ret_code = NO_ERROR;
2521
2522         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2523         kfree(cn_reply);
2524 }
2525
2526 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2527 {
2528         char buffer[sizeof(struct cn_msg)+
2529                     sizeof(struct drbd_nl_cfg_reply)+
2530                     sizeof(struct sync_progress_tag_len_struct)+
2531                     sizeof(short int)];
2532         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2533         struct drbd_nl_cfg_reply *reply =
2534                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2535         unsigned short *tl = reply->tag_list;
2536         unsigned long rs_left;
2537         unsigned int res;
2538
2539         /* no local ref, no bitmap, no syncer progress, no broadcast. */
2540         if (!get_ldev(mdev))
2541                 return;
2542         drbd_get_syncer_progress(mdev, &rs_left, &res);
2543         put_ldev(mdev);
2544
2545         tl = tl_add_int(tl, T_sync_progress, &res);
2546         put_unaligned(TT_END, tl++); /* Close the tag list */
2547
2548         cn_reply->id.idx = CN_IDX_DRBD;
2549         cn_reply->id.val = CN_VAL_DRBD;
2550
2551         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2552         cn_reply->ack = 0; /* not used here. */
2553         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2554                 (int)((char *)tl - (char *)reply->tag_list);
2555         cn_reply->flags = 0;
2556
2557         reply->packet_type = P_sync_progress;
2558         reply->minor = mdev_to_minor(mdev);
2559         reply->ret_code = NO_ERROR;
2560
2561         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2562 }
2563
2564 int __init drbd_nl_init(void)
2565 {
2566         static struct cb_id cn_id_drbd;
2567         int err, try=10;
2568
2569         cn_id_drbd.val = CN_VAL_DRBD;
2570         do {
2571                 cn_id_drbd.idx = cn_idx;
2572                 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2573                 if (!err)
2574                         break;
2575                 cn_idx = (cn_idx + CN_IDX_STEP);
2576         } while (try--);
2577
2578         if (err) {
2579                 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2580                 return err;
2581         }
2582
2583         return 0;
2584 }
2585
2586 void drbd_nl_cleanup(void)
2587 {
2588         static struct cb_id cn_id_drbd;
2589
2590         cn_id_drbd.idx = cn_idx;
2591         cn_id_drbd.val = CN_VAL_DRBD;
2592
2593         cn_del_callback(&cn_id_drbd);
2594 }
2595
2596 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2597 {
2598         char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2599         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2600         struct drbd_nl_cfg_reply *reply =
2601                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2602         int rr;
2603
2604         memset(buffer, 0, sizeof(buffer));
2605         cn_reply->id = req->id;
2606
2607         cn_reply->seq = req->seq;
2608         cn_reply->ack = req->ack  + 1;
2609         cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2610         cn_reply->flags = 0;
2611
2612         reply->packet_type = P_return_code_only;
2613         reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2614         reply->ret_code = ret_code;
2615
2616         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2617         if (rr && rr != -ESRCH)
2618                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2619 }
2620