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_SECTORS;
 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 >= DRBD_AL_EXTENTS_MIN))
 755                 mdev->sync_conf.al_extents = DRBD_AL_EXTENTS_MIN;
 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, AL_UPDATES_PER_TRANSACTION,
 764                 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
 765
 766         if (n == NULL) {
 767                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 768                 return -ENOMEM;
 769         }
 770         spin_lock_irq(&mdev->al_lock);
 771         if (t) {
 772                 for (i = 0; i < t->nr_elements; i++) {
 773                         e = lc_element_by_index(t, i);
 774                         if (e->refcnt)
 775                                 dev_err(DEV, "refcnt(%d)==%d\n",
 776                                     e->lc_number, e->refcnt);
 777                         in_use += e->refcnt;
 778                 }
 779         }
 780         if (!in_use)
 781                 mdev->act_log = n;
 782         spin_unlock_irq(&mdev->al_lock);
 783         if (in_use) {
 784                 dev_err(DEV, "Activity log still in use!\n");
 785                 lc_destroy(n);
 786                 return -EBUSY;
 787         } else {
 788                 if (t)
 789                         lc_destroy(t);
 790         }
 791         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 792         return 0;
 793 }
 794
 795 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
 796 {
 797         struct request_queue * const q = mdev->rq_queue;
 798         int max_hw_sectors = max_bio_size >> 9;
 799         int max_segments = 0;
 800
 801         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 802                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 803
 804                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
 805                 max_segments = mdev->ldev->dc.max_bio_bvecs;
 806                 put_ldev(mdev);
 807         }
 808
 809         blk_queue_logical_block_size(q, 512);
 810         blk_queue_max_hw_sectors(q, max_hw_sectors);
 811         /* This is the workaround for "bio would need to, but cannot, be split" */
 812         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
 813         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
 814
 815         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 816                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 817
 818                 blk_queue_stack_limits(q, b);
 819
 820                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
 821                         dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
 822                                  q->backing_dev_info.ra_pages,
 823                                  b->backing_dev_info.ra_pages);
 824                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
 825                 }
 826                 put_ldev(mdev);
 827         }
 828 }
 829
 830 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
 831 {
 832         int now, new, local, peer;
 833
 834         now = queue_max_hw_sectors(mdev->rq_queue) << 9;
 835         local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
 836         peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
 837
 838         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 839                 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
 840                 mdev->local_max_bio_size = local;
 841                 put_ldev(mdev);
 842         }
 843
 844         /* We may ignore peer limits if the peer is modern enough.
 845            Because new from 8.3.8 onwards the peer can use multiple
 846            BIOs for a single peer_request */
 847         if (mdev->state.conn >= C_CONNECTED) {
 848                 if (mdev->tconn->agreed_pro_version < 94)
 849                         peer = mdev->peer_max_bio_size;
 850                 else if (mdev->tconn->agreed_pro_version == 94)
 851                         peer = DRBD_MAX_SIZE_H80_PACKET;
 852                 else /* drbd 8.3.8 onwards */
 853                         peer = DRBD_MAX_BIO_SIZE;
 854         }
 855
 856         new = min_t(int, local, peer);
 857
 858         if (mdev->state.role == R_PRIMARY && new < now)
 859                 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
 860
 861         if (new != now)
 862                 dev_info(DEV, "max BIO size = %u\n", new);
 863
 864         drbd_setup_queue_param(mdev, new);
 865 }
 866
 867 /* serialize deconfig (worker exiting, doing cleanup)
 868  * and reconfig (drbdsetup disk, drbdsetup net)
 869  *
 870  * Wait for a potentially exiting worker, then restart it,
 871  * or start a new one.  Flush any pending work, there may still be an
 872  * after_state_change queued.
 873  */
 874 static void drbd_reconfig_start(struct drbd_conf *mdev)
 875 {
 876         wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
 877         wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
 878         drbd_thread_start(&mdev->tconn->worker);
 879         drbd_flush_workqueue(mdev);
 880 }
 881
 882 /* if still unconfigured, stops worker again.
 883  * if configured now, clears CONFIG_PENDING.
 884  * wakes potential waiters */
 885 static void drbd_reconfig_done(struct drbd_conf *mdev)
 886 {
 887         spin_lock_irq(&mdev->tconn->req_lock);
 888         if (mdev->state.disk == D_DISKLESS &&
 889             mdev->state.conn == C_STANDALONE &&
 890             mdev->state.role == R_SECONDARY) {
 891                 set_bit(DEVICE_DYING, &mdev->flags);
 892                 drbd_thread_stop_nowait(&mdev->tconn->worker);
 893         } else
 894                 clear_bit(CONFIG_PENDING, &mdev->flags);
 895         spin_unlock_irq(&mdev->tconn->req_lock);
 896         wake_up(&mdev->state_wait);
 897 }
 898
 899 /* Make sure IO is suspended before calling this function(). */
 900 static void drbd_suspend_al(struct drbd_conf *mdev)
 901 {
 902         int s = 0;
 903
 904         if (!lc_try_lock(mdev->act_log)) {
 905                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
 906                 return;
 907         }
 908
 909         drbd_al_shrink(mdev);
 910         spin_lock_irq(&mdev->tconn->req_lock);
 911         if (mdev->state.conn < C_CONNECTED)
 912                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
 913         spin_unlock_irq(&mdev->tconn->req_lock);
 914         lc_unlock(mdev->act_log);
 915
 916         if (s)
 917                 dev_info(DEV, "Suspended AL updates\n");
 918 }
 919
 920 /* does always return 0;
 921  * interesting return code is in reply->ret_code */
 922 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 923                              struct drbd_nl_cfg_reply *reply)
 924 {
 925         enum drbd_ret_code retcode;
 926         enum determine_dev_size dd;
 927         sector_t max_possible_sectors;
 928         sector_t min_md_device_sectors;
 929         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
 930         struct block_device *bdev;
 931         struct lru_cache *resync_lru = NULL;
 932         union drbd_state ns, os;
 933         enum drbd_state_rv rv;
 934         int cp_discovered = 0;
 935
 936         drbd_reconfig_start(mdev);
 937
 938         /* if you want to reconfigure, please tear down first */
 939         if (mdev->state.disk > D_DISKLESS) {
 940                 retcode = ERR_DISK_CONFIGURED;
 941                 goto fail;
 942         }
 943         /* It may just now have detached because of IO error.  Make sure
 944          * drbd_ldev_destroy is done already, we may end up here very fast,
 945          * e.g. if someone calls attach from the on-io-error handler,
 946          * to realize a "hot spare" feature (not that I'd recommend that) */
 947         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
 948
 949         /* allocation not in the IO path, cqueue thread context */
 950         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
 951         if (!nbc) {
 952                 retcode = ERR_NOMEM;
 953                 goto fail;
 954         }
 955
 956         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
 957         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
 958         nbc->dc.fencing       = DRBD_FENCING_DEF;
 959         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
 960
 961         if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
 962                 retcode = ERR_MANDATORY_TAG;
 963                 goto fail;
 964         }
 965
 966         if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
 967                 retcode = ERR_MD_IDX_INVALID;
 968                 goto fail;
 969         }
 970
 971         if (get_net_conf(mdev->tconn)) {
 972                 int prot = mdev->tconn->net_conf->wire_protocol;
 973                 put_net_conf(mdev->tconn);
 974                 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
 975                         retcode = ERR_STONITH_AND_PROT_A;
 976                         goto fail;
 977                 }
 978         }
 979
 980         bdev = blkdev_get_by_path(nbc->dc.backing_dev,
 981                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
 982         if (IS_ERR(bdev)) {
 983                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
 984                         PTR_ERR(bdev));
 985                 retcode = ERR_OPEN_DISK;
 986                 goto fail;
 987         }
 988         nbc->backing_bdev = bdev;
 989
 990         /*
 991          * meta_dev_idx >= 0: external fixed size, possibly multiple
 992          * drbd sharing one meta device.  TODO in that case, paranoia
 993          * check that [md_bdev, meta_dev_idx] is not yet used by some
 994          * other drbd minor!  (if you use drbd.conf + drbdadm, that
 995          * should check it for you already; but if you don't, or
 996          * someone fooled it, we need to double check here)
 997          */
 998         bdev = blkdev_get_by_path(nbc->dc.meta_dev,
 999                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1000                                   (nbc->dc.meta_dev_idx < 0) ?
1001                                   (void *)mdev : (void *)drbd_m_holder);
1002         if (IS_ERR(bdev)) {
1003                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
1004                         PTR_ERR(bdev));
1005                 retcode = ERR_OPEN_MD_DISK;
1006                 goto fail;
1007         }
1008         nbc->md_bdev = bdev;
1009
1010         if ((nbc->backing_bdev == nbc->md_bdev) !=
1011             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1012              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1013                 retcode = ERR_MD_IDX_INVALID;
1014                 goto fail;
1015         }
1016
1017         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1018                         1, 61, sizeof(struct bm_extent),
1019                         offsetof(struct bm_extent, lce));
1020         if (!resync_lru) {
1021                 retcode = ERR_NOMEM;
1022                 goto fail;
1023         }
1024
1025         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1026         drbd_md_set_sector_offsets(mdev, nbc);
1027
1028         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
1029                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1030                         (unsigned long long) drbd_get_max_capacity(nbc),
1031                         (unsigned long long) nbc->dc.disk_size);
1032                 retcode = ERR_DISK_TO_SMALL;
1033                 goto fail;
1034         }
1035
1036         if (nbc->dc.meta_dev_idx < 0) {
1037                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1038                 /* at least one MB, otherwise it does not make sense */
1039                 min_md_device_sectors = (2<<10);
1040         } else {
1041                 max_possible_sectors = DRBD_MAX_SECTORS;
1042                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
1043         }
1044
1045         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1046                 retcode = ERR_MD_DISK_TO_SMALL;
1047                 dev_warn(DEV, "refusing attach: md-device too small, "
1048                      "at least %llu sectors needed for this meta-disk type\n",
1049                      (unsigned long long) min_md_device_sectors);
1050                 goto fail;
1051         }
1052
1053         /* Make sure the new disk is big enough
1054          * (we may currently be R_PRIMARY with no local disk...) */
1055         if (drbd_get_max_capacity(nbc) <
1056             drbd_get_capacity(mdev->this_bdev)) {
1057                 retcode = ERR_DISK_TO_SMALL;
1058                 goto fail;
1059         }
1060
1061         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1062
1063         if (nbc->known_size > max_possible_sectors) {
1064                 dev_warn(DEV, "==> truncating very big lower level device "
1065                         "to currently maximum possible %llu sectors <==\n",
1066                         (unsigned long long) max_possible_sectors);
1067                 if (nbc->dc.meta_dev_idx >= 0)
1068                         dev_warn(DEV, "==>> using internal or flexible "
1069                                       "meta data may help <<==\n");
1070         }
1071
1072         drbd_suspend_io(mdev);
1073         /* also wait for the last barrier ack. */
1074         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state));
1075         /* and for any other previously queued work */
1076         drbd_flush_workqueue(mdev);
1077
1078         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1079         retcode = rv;  /* FIXME: Type mismatch. */
1080         drbd_resume_io(mdev);
1081         if (rv < SS_SUCCESS)
1082                 goto fail;
1083
1084         if (!get_ldev_if_state(mdev, D_ATTACHING))
1085                 goto force_diskless;
1086
1087         drbd_md_set_sector_offsets(mdev, nbc);
1088
1089         if (!mdev->bitmap) {
1090                 if (drbd_bm_init(mdev)) {
1091                         retcode = ERR_NOMEM;
1092                         goto force_diskless_dec;
1093                 }
1094         }
1095
1096         retcode = drbd_md_read(mdev, nbc);
1097         if (retcode != NO_ERROR)
1098                 goto force_diskless_dec;
1099
1100         if (mdev->state.conn < C_CONNECTED &&
1101             mdev->state.role == R_PRIMARY &&
1102             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1103                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1104                     (unsigned long long)mdev->ed_uuid);
1105                 retcode = ERR_DATA_NOT_CURRENT;
1106                 goto force_diskless_dec;
1107         }
1108
1109         /* Since we are diskless, fix the activity log first... */
1110         if (drbd_check_al_size(mdev)) {
1111                 retcode = ERR_NOMEM;
1112                 goto force_diskless_dec;
1113         }
1114
1115         /* Prevent shrinking of consistent devices ! */
1116         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1117             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1118                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1119                 retcode = ERR_DISK_TO_SMALL;
1120                 goto force_diskless_dec;
1121         }
1122
1123         if (!drbd_al_read_log(mdev, nbc)) {
1124                 retcode = ERR_IO_MD_DISK;
1125                 goto force_diskless_dec;
1126         }
1127
1128         /* Reset the "barriers don't work" bits here, then force meta data to
1129          * be written, to ensure we determine if barriers are supported. */
1130         if (nbc->dc.no_md_flush)
1131                 set_bit(MD_NO_FUA, &mdev->flags);
1132         else
1133                 clear_bit(MD_NO_FUA, &mdev->flags);
1134
1135         /* Point of no return reached.
1136          * Devices and memory are no longer released by error cleanup below.
1137          * now mdev takes over responsibility, and the state engine should
1138          * clean it up somewhere.  */
1139         D_ASSERT(mdev->ldev == NULL);
1140         mdev->ldev = nbc;
1141         mdev->resync = resync_lru;
1142         nbc = NULL;
1143         resync_lru = NULL;
1144
1145         mdev->write_ordering = WO_bdev_flush;
1146         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1147
1148         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1149                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1150         else
1151                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1152
1153         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1154             !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) {
1155                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1156                 cp_discovered = 1;
1157         }
1158
1159         mdev->send_cnt = 0;
1160         mdev->recv_cnt = 0;
1161         mdev->read_cnt = 0;
1162         mdev->writ_cnt = 0;
1163
1164         drbd_reconsider_max_bio_size(mdev);
1165
1166         /* If I am currently not R_PRIMARY,
1167          * but meta data primary indicator is set,
1168          * I just now recover from a hard crash,
1169          * and have been R_PRIMARY before that crash.
1170          *
1171          * Now, if I had no connection before that crash
1172          * (have been degraded R_PRIMARY), chances are that
1173          * I won't find my peer now either.
1174          *
1175          * In that case, and _only_ in that case,
1176          * we use the degr-wfc-timeout instead of the default,
1177          * so we can automatically recover from a crash of a
1178          * degraded but active "cluster" after a certain timeout.
1179          */
1180         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1181         if (mdev->state.role != R_PRIMARY &&
1182              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1183             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1184                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1185
1186         dd = drbd_determine_dev_size(mdev, 0);
1187         if (dd == dev_size_error) {
1188                 retcode = ERR_NOMEM_BITMAP;
1189                 goto force_diskless_dec;
1190         } else if (dd == grew)
1191                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1192
1193         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1194                 dev_info(DEV, "Assuming that all blocks are out of sync "
1195                      "(aka FullSync)\n");
1196                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1197                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1198                         retcode = ERR_IO_MD_DISK;
1199                         goto force_diskless_dec;
1200                 }
1201         } else {
1202                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1203                         "read from attaching", BM_LOCKED_MASK) < 0) {
1204                         retcode = ERR_IO_MD_DISK;
1205                         goto force_diskless_dec;
1206                 }
1207         }
1208
1209         if (cp_discovered) {
1210                 drbd_al_apply_to_bm(mdev);
1211                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1212                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1213                         retcode = ERR_IO_MD_DISK;
1214                         goto force_diskless_dec;
1215                 }
1216         }
1217
1218         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1219                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1220
1221         spin_lock_irq(&mdev->tconn->req_lock);
1222         os = mdev->state;
1223         ns.i = os.i;
1224         /* If MDF_CONSISTENT is not set go into inconsistent state,
1225            otherwise investigate MDF_WasUpToDate...
1226            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1227            otherwise into D_CONSISTENT state.
1228         */
1229         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1230                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1231                         ns.disk = D_CONSISTENT;
1232                 else
1233                         ns.disk = D_OUTDATED;
1234         } else {
1235                 ns.disk = D_INCONSISTENT;
1236         }
1237
1238         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1239                 ns.pdsk = D_OUTDATED;
1240
1241         if ( ns.disk == D_CONSISTENT &&
1242             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1243                 ns.disk = D_UP_TO_DATE;
1244
1245         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1246            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1247            this point, because drbd_request_state() modifies these
1248            flags. */
1249
1250         /* In case we are C_CONNECTED postpone any decision on the new disk
1251            state after the negotiation phase. */
1252         if (mdev->state.conn == C_CONNECTED) {
1253                 mdev->new_state_tmp.i = ns.i;
1254                 ns.i = os.i;
1255                 ns.disk = D_NEGOTIATING;
1256
1257                 /* We expect to receive up-to-date UUIDs soon.
1258                    To avoid a race in receive_state, free p_uuid while
1259                    holding req_lock. I.e. atomic with the state change */
1260                 kfree(mdev->p_uuid);
1261                 mdev->p_uuid = NULL;
1262         }
1263
1264         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1265         ns = mdev->state;
1266         spin_unlock_irq(&mdev->tconn->req_lock);
1267
1268         if (rv < SS_SUCCESS)
1269                 goto force_diskless_dec;
1270
1271         if (mdev->state.role == R_PRIMARY)
1272                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1273         else
1274                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1275
1276         drbd_md_mark_dirty(mdev);
1277         drbd_md_sync(mdev);
1278
1279         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1280         put_ldev(mdev);
1281         reply->ret_code = retcode;
1282         drbd_reconfig_done(mdev);
1283         return 0;
1284
1285  force_diskless_dec:
1286         put_ldev(mdev);
1287  force_diskless:
1288         drbd_force_state(mdev, NS(disk, D_FAILED));
1289         drbd_md_sync(mdev);
1290  fail:
1291         if (nbc) {
1292                 if (nbc->backing_bdev)
1293                         blkdev_put(nbc->backing_bdev,
1294                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1295                 if (nbc->md_bdev)
1296                         blkdev_put(nbc->md_bdev,
1297                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1298                 kfree(nbc);
1299         }
1300         lc_destroy(resync_lru);
1301
1302         reply->ret_code = retcode;
1303         drbd_reconfig_done(mdev);
1304         return 0;
1305 }
1306
1307 /* Detaching the disk is a process in multiple stages.  First we need to lock
1308  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1309  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1310  * internal references as well.
1311  * Only then we have finally detached. */
1312 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1313                           struct drbd_nl_cfg_reply *reply)
1314 {
1315         enum drbd_ret_code retcode;
1316         int ret;
1317         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1318         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1319         /* D_FAILED will transition to DISKLESS. */
1320         ret = wait_event_interruptible(mdev->misc_wait,
1321                         mdev->state.disk != D_FAILED);
1322         drbd_resume_io(mdev);
1323         if ((int)retcode == (int)SS_IS_DISKLESS)
1324                 retcode = SS_NOTHING_TO_DO;
1325         if (ret)
1326                 retcode = ERR_INTR;
1327         reply->ret_code = retcode;
1328         return 0;
1329 }
1330
1331 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1332                             struct drbd_nl_cfg_reply *reply)
1333 {
1334         int i;
1335         enum drbd_ret_code retcode;
1336         struct net_conf *new_conf = NULL;
1337         struct crypto_hash *tfm = NULL;
1338         struct crypto_hash *integrity_w_tfm = NULL;
1339         struct crypto_hash *integrity_r_tfm = NULL;
1340         struct drbd_conf *odev;
1341         char hmac_name[CRYPTO_MAX_ALG_NAME];
1342         void *int_dig_out = NULL;
1343         void *int_dig_in = NULL;
1344         void *int_dig_vv = NULL;
1345         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1346
1347         drbd_reconfig_start(mdev);
1348
1349         if (mdev->state.conn > C_STANDALONE) {
1350                 retcode = ERR_NET_CONFIGURED;
1351                 goto fail;
1352         }
1353
1354         /* allocation not in the IO path, cqueue thread context */
1355         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1356         if (!new_conf) {
1357                 retcode = ERR_NOMEM;
1358                 goto fail;
1359         }
1360
1361         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1362         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1363         new_conf->ping_int         = DRBD_PING_INT_DEF;
1364         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1365         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1366         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1367         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1368         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1369         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1370         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1371         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1372         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1373         new_conf->want_lose        = 0;
1374         new_conf->two_primaries    = 0;
1375         new_conf->wire_protocol    = DRBD_PROT_C;
1376         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1377         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1378         new_conf->on_congestion    = DRBD_ON_CONGESTION_DEF;
1379         new_conf->cong_extents     = DRBD_CONG_EXTENTS_DEF;
1380
1381         if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1382                 retcode = ERR_MANDATORY_TAG;
1383                 goto fail;
1384         }
1385
1386         if (new_conf->two_primaries
1387             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1388                 retcode = ERR_NOT_PROTO_C;
1389                 goto fail;
1390         }
1391
1392         if (get_ldev(mdev)) {
1393                 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1394                 put_ldev(mdev);
1395                 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1396                         retcode = ERR_STONITH_AND_PROT_A;
1397                         goto fail;
1398                 }
1399         }
1400
1401         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1402                 retcode = ERR_CONG_NOT_PROTO_A;
1403                 goto fail;
1404         }
1405
1406         if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1407                 retcode = ERR_DISCARD;
1408                 goto fail;
1409         }
1410
1411         retcode = NO_ERROR;
1412
1413         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1414         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1415         for (i = 0; i < minor_count; i++) {
1416                 odev = minor_to_mdev(i);
1417                 if (!odev || odev == mdev)
1418                         continue;
1419                 if (get_net_conf(odev->tconn)) {
1420                         taken_addr = (struct sockaddr *)&odev->tconn->net_conf->my_addr;
1421                         if (new_conf->my_addr_len == odev->tconn->net_conf->my_addr_len &&
1422                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1423                                 retcode = ERR_LOCAL_ADDR;
1424
1425                         taken_addr = (struct sockaddr *)&odev->tconn->net_conf->peer_addr;
1426                         if (new_conf->peer_addr_len == odev->tconn->net_conf->peer_addr_len &&
1427                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1428                                 retcode = ERR_PEER_ADDR;
1429
1430                         put_net_conf(odev->tconn);
1431                         if (retcode != NO_ERROR)
1432                                 goto fail;
1433                 }
1434         }
1435
1436         if (new_conf->cram_hmac_alg[0] != 0) {
1437                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1438                         new_conf->cram_hmac_alg);
1439                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1440                 if (IS_ERR(tfm)) {
1441                         tfm = NULL;
1442                         retcode = ERR_AUTH_ALG;
1443                         goto fail;
1444                 }
1445
1446                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1447                         retcode = ERR_AUTH_ALG_ND;
1448                         goto fail;
1449                 }
1450         }
1451
1452         if (new_conf->integrity_alg[0]) {
1453                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1454                 if (IS_ERR(integrity_w_tfm)) {
1455                         integrity_w_tfm = NULL;
1456                         retcode=ERR_INTEGRITY_ALG;
1457                         goto fail;
1458                 }
1459
1460                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1461                         retcode=ERR_INTEGRITY_ALG_ND;
1462                         goto fail;
1463                 }
1464
1465                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1466                 if (IS_ERR(integrity_r_tfm)) {
1467                         integrity_r_tfm = NULL;
1468                         retcode=ERR_INTEGRITY_ALG;
1469                         goto fail;
1470                 }
1471         }
1472
1473         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1474
1475         if (integrity_w_tfm) {
1476                 i = crypto_hash_digestsize(integrity_w_tfm);
1477                 int_dig_out = kmalloc(i, GFP_KERNEL);
1478                 if (!int_dig_out) {
1479                         retcode = ERR_NOMEM;
1480                         goto fail;
1481                 }
1482                 int_dig_in = kmalloc(i, GFP_KERNEL);
1483                 if (!int_dig_in) {
1484                         retcode = ERR_NOMEM;
1485                         goto fail;
1486                 }
1487                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1488                 if (!int_dig_vv) {
1489                         retcode = ERR_NOMEM;
1490                         goto fail;
1491                 }
1492         }
1493
1494         if (!mdev->bitmap) {
1495                 if(drbd_bm_init(mdev)) {
1496                         retcode = ERR_NOMEM;
1497                         goto fail;
1498                 }
1499         }
1500
1501         drbd_flush_workqueue(mdev);
1502         spin_lock_irq(&mdev->tconn->req_lock);
1503         if (mdev->tconn->net_conf != NULL) {
1504                 retcode = ERR_NET_CONFIGURED;
1505                 spin_unlock_irq(&mdev->tconn->req_lock);
1506                 goto fail;
1507         }
1508         mdev->tconn->net_conf = new_conf;
1509
1510         mdev->send_cnt = 0;
1511         mdev->recv_cnt = 0;
1512
1513         crypto_free_hash(mdev->tconn->cram_hmac_tfm);
1514         mdev->tconn->cram_hmac_tfm = tfm;
1515
1516         crypto_free_hash(mdev->tconn->integrity_w_tfm);
1517         mdev->tconn->integrity_w_tfm = integrity_w_tfm;
1518
1519         crypto_free_hash(mdev->tconn->integrity_r_tfm);
1520         mdev->tconn->integrity_r_tfm = integrity_r_tfm;
1521
1522         kfree(mdev->tconn->int_dig_out);
1523         kfree(mdev->tconn->int_dig_in);
1524         kfree(mdev->tconn->int_dig_vv);
1525         mdev->tconn->int_dig_out=int_dig_out;
1526         mdev->tconn->int_dig_in=int_dig_in;
1527         mdev->tconn->int_dig_vv=int_dig_vv;
1528         retcode = _conn_request_state(mdev->tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
1529         spin_unlock_irq(&mdev->tconn->req_lock);
1530
1531         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1532         reply->ret_code = retcode;
1533         drbd_reconfig_done(mdev);
1534         return 0;
1535
1536 fail:
1537         kfree(int_dig_out);
1538         kfree(int_dig_in);
1539         kfree(int_dig_vv);
1540         crypto_free_hash(tfm);
1541         crypto_free_hash(integrity_w_tfm);
1542         crypto_free_hash(integrity_r_tfm);
1543         kfree(new_conf);
1544
1545         reply->ret_code = retcode;
1546         drbd_reconfig_done(mdev);
1547         return 0;
1548 }
1549
1550 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1551                               struct drbd_nl_cfg_reply *reply)
1552 {
1553         struct drbd_tconn *tconn = mdev->tconn;
1554         int retcode;
1555         struct disconnect dc;
1556
1557         memset(&dc, 0, sizeof(struct disconnect));
1558         if (!disconnect_from_tags(mdev, nlp->tag_list, &dc)) {
1559                 retcode = ERR_MANDATORY_TAG;
1560                 goto fail;
1561         }
1562
1563         if (dc.force) {
1564                 spin_lock_irq(&tconn->req_lock);
1565                 if (tconn->cstate >= C_WF_CONNECTION)
1566                         _conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
1567                 spin_unlock_irq(&tconn->req_lock);
1568                 goto done;
1569         }
1570
1571         retcode = conn_request_state(tconn, NS(conn, C_DISCONNECTING), 0);
1572
1573         if (retcode == SS_NOTHING_TO_DO)
1574                 goto done;
1575         else if (retcode == SS_ALREADY_STANDALONE)
1576                 goto done;
1577         else if (retcode == SS_PRIMARY_NOP) {
1578                 /* Our state checking code wants to see the peer outdated. */
1579                 retcode = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
1580                                                         pdsk, D_OUTDATED), CS_VERBOSE);
1581         } else if (retcode == SS_CW_FAILED_BY_PEER) {
1582                 /* The peer probably wants to see us outdated. */
1583                 retcode = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
1584                                                         disk, D_OUTDATED), 0);
1585                 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1586                         conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
1587                         retcode = SS_SUCCESS;
1588                 }
1589         }
1590
1591         if (retcode < SS_SUCCESS)
1592                 goto fail;
1593
1594         if (wait_event_interruptible(tconn->ping_wait,
1595                                      tconn->cstate != C_DISCONNECTING)) {
1596                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1597                    someone else might connect us in the mean time! */
1598                 retcode = ERR_INTR;
1599                 goto fail;
1600         }
1601
1602  done:
1603         retcode = NO_ERROR;
1604  fail:
1605         drbd_md_sync(mdev);
1606         reply->ret_code = retcode;
1607         return 0;
1608 }
1609
1610 void resync_after_online_grow(struct drbd_conf *mdev)
1611 {
1612         int iass; /* I am sync source */
1613
1614         dev_info(DEV, "Resync of new storage after online grow\n");
1615         if (mdev->state.role != mdev->state.peer)
1616                 iass = (mdev->state.role == R_PRIMARY);
1617         else
1618                 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
1619
1620         if (iass)
1621                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1622         else
1623                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1624 }
1625
1626 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1627                           struct drbd_nl_cfg_reply *reply)
1628 {
1629         struct resize rs;
1630         int retcode = NO_ERROR;
1631         enum determine_dev_size dd;
1632         enum dds_flags ddsf;
1633
1634         memset(&rs, 0, sizeof(struct resize));
1635         if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1636                 retcode = ERR_MANDATORY_TAG;
1637                 goto fail;
1638         }
1639
1640         if (mdev->state.conn > C_CONNECTED) {
1641                 retcode = ERR_RESIZE_RESYNC;
1642                 goto fail;
1643         }
1644
1645         if (mdev->state.role == R_SECONDARY &&
1646             mdev->state.peer == R_SECONDARY) {
1647                 retcode = ERR_NO_PRIMARY;
1648                 goto fail;
1649         }
1650
1651         if (!get_ldev(mdev)) {
1652                 retcode = ERR_NO_DISK;
1653                 goto fail;
1654         }
1655
1656         if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
1657                 retcode = ERR_NEED_APV_93;
1658                 goto fail;
1659         }
1660
1661         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1662                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1663
1664         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1665         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1666         dd = drbd_determine_dev_size(mdev, ddsf);
1667         drbd_md_sync(mdev);
1668         put_ldev(mdev);
1669         if (dd == dev_size_error) {
1670                 retcode = ERR_NOMEM_BITMAP;
1671                 goto fail;
1672         }
1673
1674         if (mdev->state.conn == C_CONNECTED) {
1675                 if (dd == grew)
1676                         set_bit(RESIZE_PENDING, &mdev->flags);
1677
1678                 drbd_send_uuids(mdev);
1679                 drbd_send_sizes(mdev, 1, ddsf);
1680         }
1681
1682  fail:
1683         reply->ret_code = retcode;
1684         return 0;
1685 }
1686
1687 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1688                                struct drbd_nl_cfg_reply *reply)
1689 {
1690         int retcode = NO_ERROR;
1691         int err;
1692         int ovr; /* online verify running */
1693         int rsr; /* re-sync running */
1694         struct crypto_hash *verify_tfm = NULL;
1695         struct crypto_hash *csums_tfm = NULL;
1696         struct syncer_conf sc;
1697         cpumask_var_t new_cpu_mask;
1698         int *rs_plan_s = NULL;
1699         int fifo_size;
1700
1701         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1702                 retcode = ERR_NOMEM;
1703                 goto fail;
1704         }
1705
1706         if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1707                 memset(&sc, 0, sizeof(struct syncer_conf));
1708                 sc.rate       = DRBD_RATE_DEF;
1709                 sc.after      = DRBD_AFTER_DEF;
1710                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1711                 sc.on_no_data  = DRBD_ON_NO_DATA_DEF;
1712                 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1713                 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1714                 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1715                 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1716                 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1717         } else
1718                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1719
1720         if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1721                 retcode = ERR_MANDATORY_TAG;
1722                 goto fail;
1723         }
1724
1725         /* re-sync running */
1726         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1727                 mdev->state.conn == C_SYNC_TARGET ||
1728                 mdev->state.conn == C_PAUSED_SYNC_S ||
1729                 mdev->state.conn == C_PAUSED_SYNC_T );
1730
1731         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1732                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1733                 goto fail;
1734         }
1735
1736         if (!rsr && sc.csums_alg[0]) {
1737                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1738                 if (IS_ERR(csums_tfm)) {
1739                         csums_tfm = NULL;
1740                         retcode = ERR_CSUMS_ALG;
1741                         goto fail;
1742                 }
1743
1744                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1745                         retcode = ERR_CSUMS_ALG_ND;
1746                         goto fail;
1747                 }
1748         }
1749
1750         /* online verify running */
1751         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1752
1753         if (ovr) {
1754                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1755                         retcode = ERR_VERIFY_RUNNING;
1756                         goto fail;
1757                 }
1758         }
1759
1760         if (!ovr && sc.verify_alg[0]) {
1761                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1762                 if (IS_ERR(verify_tfm)) {
1763                         verify_tfm = NULL;
1764                         retcode = ERR_VERIFY_ALG;
1765                         goto fail;
1766                 }
1767
1768                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1769                         retcode = ERR_VERIFY_ALG_ND;
1770                         goto fail;
1771                 }
1772         }
1773
1774         /* silently ignore cpu mask on UP kernel */
1775         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1776                 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1777                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1778                 if (err) {
1779                         dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1780                         retcode = ERR_CPU_MASK_PARSE;
1781                         goto fail;
1782                 }
1783         }
1784
1785         if (!expect(sc.rate >= 1))
1786                 sc.rate = 1;
1787
1788         /* clip to allowed range */
1789         if (!expect(sc.al_extents >= DRBD_AL_EXTENTS_MIN))
1790                 sc.al_extents = DRBD_AL_EXTENTS_MIN;
1791         if (!expect(sc.al_extents <= DRBD_AL_EXTENTS_MAX))
1792                 sc.al_extents = DRBD_AL_EXTENTS_MAX;
1793
1794         /* to avoid spurious errors when configuring minors before configuring
1795          * the minors they depend on: if necessary, first create the minor we
1796          * depend on */
1797         if (sc.after >= 0)
1798                 ensure_mdev(sc.after, 1);
1799
1800         /* most sanity checks done, try to assign the new sync-after
1801          * dependency.  need to hold the global lock in there,
1802          * to avoid a race in the dependency loop check. */
1803         retcode = drbd_alter_sa(mdev, sc.after);
1804         if (retcode != NO_ERROR)
1805                 goto fail;
1806
1807         fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1808         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1809                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1810                 if (!rs_plan_s) {
1811                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1812                         retcode = ERR_NOMEM;
1813                         goto fail;
1814                 }
1815         }
1816
1817         /* ok, assign the rest of it as well.
1818          * lock against receive_SyncParam() */
1819         spin_lock(&mdev->peer_seq_lock);
1820         mdev->sync_conf = sc;
1821
1822         if (!rsr) {
1823                 crypto_free_hash(mdev->csums_tfm);
1824                 mdev->csums_tfm = csums_tfm;
1825                 csums_tfm = NULL;
1826         }
1827
1828         if (!ovr) {
1829                 crypto_free_hash(mdev->verify_tfm);
1830                 mdev->verify_tfm = verify_tfm;
1831                 verify_tfm = NULL;
1832         }
1833
1834         if (fifo_size != mdev->rs_plan_s.size) {
1835                 kfree(mdev->rs_plan_s.values);
1836                 mdev->rs_plan_s.values = rs_plan_s;
1837                 mdev->rs_plan_s.size   = fifo_size;
1838                 mdev->rs_planed = 0;
1839                 rs_plan_s = NULL;
1840         }
1841
1842         spin_unlock(&mdev->peer_seq_lock);
1843
1844         if (get_ldev(mdev)) {
1845                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1846                 drbd_al_shrink(mdev);
1847                 err = drbd_check_al_size(mdev);
1848                 lc_unlock(mdev->act_log);
1849                 wake_up(&mdev->al_wait);
1850
1851                 put_ldev(mdev);
1852                 drbd_md_sync(mdev);
1853
1854                 if (err) {
1855                         retcode = ERR_NOMEM;
1856                         goto fail;
1857                 }
1858         }
1859
1860         if (mdev->state.conn >= C_CONNECTED)
1861                 drbd_send_sync_param(mdev, &sc);
1862
1863         if (!cpumask_equal(mdev->tconn->cpu_mask, new_cpu_mask)) {
1864                 cpumask_copy(mdev->tconn->cpu_mask, new_cpu_mask);
1865                 drbd_calc_cpu_mask(mdev->tconn);
1866                 mdev->tconn->receiver.reset_cpu_mask = 1;
1867                 mdev->tconn->asender.reset_cpu_mask = 1;
1868                 mdev->tconn->worker.reset_cpu_mask = 1;
1869         }
1870
1871         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1872 fail:
1873         kfree(rs_plan_s);
1874         free_cpumask_var(new_cpu_mask);
1875         crypto_free_hash(csums_tfm);
1876         crypto_free_hash(verify_tfm);
1877         reply->ret_code = retcode;
1878         return 0;
1879 }
1880
1881 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1882                               struct drbd_nl_cfg_reply *reply)
1883 {
1884         int retcode;
1885
1886         /* If there is still bitmap IO pending, probably because of a previous
1887          * resync just being finished, wait for it before requesting a new resync. */
1888         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1889
1890         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1891
1892         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1893                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1894
1895         while (retcode == SS_NEED_CONNECTION) {
1896                 spin_lock_irq(&mdev->tconn->req_lock);
1897                 if (mdev->state.conn < C_CONNECTED)
1898                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1899                 spin_unlock_irq(&mdev->tconn->req_lock);
1900
1901                 if (retcode != SS_NEED_CONNECTION)
1902                         break;
1903
1904                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1905         }
1906
1907         reply->ret_code = retcode;
1908         return 0;
1909 }
1910
1911 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
1912 {
1913         int rv;
1914
1915         rv = drbd_bmio_set_n_write(mdev);
1916         drbd_suspend_al(mdev);
1917         return rv;
1918 }
1919
1920 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1921                                    struct drbd_nl_cfg_reply *reply)
1922 {
1923         int retcode;
1924
1925         /* If there is still bitmap IO pending, probably because of a previous
1926          * resync just being finished, wait for it before requesting a new resync. */
1927         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1928
1929         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
1930
1931         if (retcode < SS_SUCCESS) {
1932                 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
1933                         /* The peer will get a resync upon connect anyways. Just make that
1934                            into a full resync. */
1935                         retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
1936                         if (retcode >= SS_SUCCESS) {
1937                                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
1938                                         "set_n_write from invalidate_peer",
1939                                         BM_LOCKED_SET_ALLOWED))
1940                                         retcode = ERR_IO_MD_DISK;
1941                         }
1942                 } else
1943                         retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
1944         }
1945
1946         reply->ret_code = retcode;
1947         return 0;
1948 }
1949
1950 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1951                               struct drbd_nl_cfg_reply *reply)
1952 {
1953         int retcode = NO_ERROR;
1954
1955         if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
1956                 retcode = ERR_PAUSE_IS_SET;
1957
1958         reply->ret_code = retcode;
1959         return 0;
1960 }
1961
1962 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1963                                struct drbd_nl_cfg_reply *reply)
1964 {
1965         int retcode = NO_ERROR;
1966         union drbd_state s;
1967
1968         if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
1969                 s = mdev->state;
1970                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
1971                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
1972                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
1973                 } else {
1974                         retcode = ERR_PAUSE_IS_CLEAR;
1975                 }
1976         }
1977
1978         reply->ret_code = retcode;
1979         return 0;
1980 }
1981
1982 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1983                               struct drbd_nl_cfg_reply *reply)
1984 {
1985         reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
1986
1987         return 0;
1988 }
1989
1990 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1991                              struct drbd_nl_cfg_reply *reply)
1992 {
1993         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
1994                 drbd_uuid_new_current(mdev);
1995                 clear_bit(NEW_CUR_UUID, &mdev->flags);
1996         }
1997         drbd_suspend_io(mdev);
1998         reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
1999         if (reply->ret_code == SS_SUCCESS) {
2000                 if (mdev->state.conn < C_CONNECTED)
2001                         tl_clear(mdev);
2002                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2003                         tl_restart(mdev, FAIL_FROZEN_DISK_IO);
2004         }
2005         drbd_resume_io(mdev);
2006
2007         return 0;
2008 }
2009
2010 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2011                            struct drbd_nl_cfg_reply *reply)
2012 {
2013         reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
2014         return 0;
2015 }
2016
2017 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2018                            struct drbd_nl_cfg_reply *reply)
2019 {
2020         unsigned short *tl;
2021
2022         tl = reply->tag_list;
2023
2024         if (get_ldev(mdev)) {
2025                 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
2026                 put_ldev(mdev);
2027         }
2028
2029         if (get_net_conf(mdev->tconn)) {
2030                 tl = net_conf_to_tags(mdev, mdev->tconn->net_conf, tl);
2031                 put_net_conf(mdev->tconn);
2032         }
2033         tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
2034
2035         put_unaligned(TT_END, tl++); /* Close the tag list */
2036
2037         return (int)((char *)tl - (char *)reply->tag_list);
2038 }
2039
2040 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2041                              struct drbd_nl_cfg_reply *reply)
2042 {
2043         unsigned short *tl = reply->tag_list;
2044         union drbd_state s = mdev->state;
2045         unsigned long rs_left;
2046         unsigned int res;
2047
2048         tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
2049
2050         /* no local ref, no bitmap, no syncer progress. */
2051         if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
2052                 if (get_ldev(mdev)) {
2053                         drbd_get_syncer_progress(mdev, &rs_left, &res);
2054                         tl = tl_add_int(tl, T_sync_progress, &res);
2055                         put_ldev(mdev);
2056                 }
2057         }
2058         put_unaligned(TT_END, tl++); /* Close the tag list */
2059
2060         return (int)((char *)tl - (char *)reply->tag_list);
2061 }
2062
2063 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2064                              struct drbd_nl_cfg_reply *reply)
2065 {
2066         unsigned short *tl;
2067
2068         tl = reply->tag_list;
2069
2070         if (get_ldev(mdev)) {
2071                 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
2072                 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
2073                 put_ldev(mdev);
2074         }
2075         put_unaligned(TT_END, tl++); /* Close the tag list */
2076
2077         return (int)((char *)tl - (char *)reply->tag_list);
2078 }
2079
2080 /**
2081  * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
2082  * @mdev:       DRBD device.
2083  * @nlp:        Netlink/connector packet from drbdsetup
2084  * @reply:      Reply packet for drbdsetup
2085  */
2086 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2087                                     struct drbd_nl_cfg_reply *reply)
2088 {
2089         unsigned short *tl;
2090         char rv;
2091
2092         tl = reply->tag_list;
2093
2094         rv = mdev->state.pdsk == D_OUTDATED        ? UT_PEER_OUTDATED :
2095           test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
2096
2097         tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
2098         put_unaligned(TT_END, tl++); /* Close the tag list */
2099
2100         return (int)((char *)tl - (char *)reply->tag_list);
2101 }
2102
2103 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2104                                     struct drbd_nl_cfg_reply *reply)
2105 {
2106         /* default to resume from last known position, if possible */
2107         struct start_ov args =
2108                 { .start_sector = mdev->ov_start_sector };
2109
2110         if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2111                 reply->ret_code = ERR_MANDATORY_TAG;
2112                 return 0;
2113         }
2114
2115         /* If there is still bitmap IO pending, e.g. previous resync or verify
2116          * just being finished, wait for it before requesting a new resync. */
2117         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2118
2119         /* w_make_ov_request expects position to be aligned */
2120         mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
2121         reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2122         return 0;
2123 }
2124
2125
2126 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2127                               struct drbd_nl_cfg_reply *reply)
2128 {
2129         int retcode = NO_ERROR;
2130         int skip_initial_sync = 0;
2131         int err;
2132
2133         struct new_c_uuid args;
2134
2135         memset(&args, 0, sizeof(struct new_c_uuid));
2136         if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2137                 reply->ret_code = ERR_MANDATORY_TAG;
2138                 return 0;
2139         }
2140
2141         mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2142
2143         if (!get_ldev(mdev)) {
2144                 retcode = ERR_NO_DISK;
2145                 goto out;
2146         }
2147
2148         /* this is "skip initial sync", assume to be clean */
2149         if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2150             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2151                 dev_info(DEV, "Preparing to skip initial sync\n");
2152                 skip_initial_sync = 1;
2153         } else if (mdev->state.conn != C_STANDALONE) {
2154                 retcode = ERR_CONNECTED;
2155                 goto out_dec;
2156         }
2157
2158         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2159         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2160
2161         if (args.clear_bm) {
2162                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2163                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2164                 if (err) {
2165                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2166                         retcode = ERR_IO_MD_DISK;
2167                 }
2168                 if (skip_initial_sync) {
2169                         drbd_send_uuids_skip_initial_sync(mdev);
2170                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2171                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2172                         spin_lock_irq(&mdev->tconn->req_lock);
2173                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2174                                         CS_VERBOSE, NULL);
2175                         spin_unlock_irq(&mdev->tconn->req_lock);
2176                 }
2177         }
2178
2179         drbd_md_sync(mdev);
2180 out_dec:
2181         put_ldev(mdev);
2182 out:
2183         mutex_unlock(mdev->state_mutex);
2184
2185         reply->ret_code = retcode;
2186         return 0;
2187 }
2188
2189 struct cn_handler_struct {
2190         int (*function)(struct drbd_conf *,
2191                          struct drbd_nl_cfg_req *,
2192                          struct drbd_nl_cfg_reply *);
2193         int reply_body_size;
2194 };
2195
2196 static struct cn_handler_struct cnd_table[] = {
2197         [ P_primary ]           = { &drbd_nl_primary,           0 },
2198         [ P_secondary ]         = { &drbd_nl_secondary,         0 },
2199         [ P_disk_conf ]         = { &drbd_nl_disk_conf,         0 },
2200         [ P_detach ]            = { &drbd_nl_detach,            0 },
2201         [ P_net_conf ]          = { &drbd_nl_net_conf,          0 },
2202         [ P_disconnect ]        = { &drbd_nl_disconnect,        0 },
2203         [ P_resize ]            = { &drbd_nl_resize,            0 },
2204         [ P_syncer_conf ]       = { &drbd_nl_syncer_conf,       0 },
2205         [ P_invalidate ]        = { &drbd_nl_invalidate,        0 },
2206         [ P_invalidate_peer ]   = { &drbd_nl_invalidate_peer,   0 },
2207         [ P_pause_sync ]        = { &drbd_nl_pause_sync,        0 },
2208         [ P_resume_sync ]       = { &drbd_nl_resume_sync,       0 },
2209         [ P_suspend_io ]        = { &drbd_nl_suspend_io,        0 },
2210         [ P_resume_io ]         = { &drbd_nl_resume_io,         0 },
2211         [ P_outdate ]           = { &drbd_nl_outdate,           0 },
2212         [ P_get_config ]        = { &drbd_nl_get_config,
2213                                     sizeof(struct syncer_conf_tag_len_struct) +
2214                                     sizeof(struct disk_conf_tag_len_struct) +
2215                                     sizeof(struct net_conf_tag_len_struct) },
2216         [ P_get_state ]         = { &drbd_nl_get_state,
2217                                     sizeof(struct get_state_tag_len_struct) +
2218                                     sizeof(struct sync_progress_tag_len_struct) },
2219         [ P_get_uuids ]         = { &drbd_nl_get_uuids,
2220                                     sizeof(struct get_uuids_tag_len_struct) },
2221         [ P_get_timeout_flag ]  = { &drbd_nl_get_timeout_flag,
2222                                     sizeof(struct get_timeout_flag_tag_len_struct)},
2223         [ P_start_ov ]          = { &drbd_nl_start_ov,          0 },
2224         [ P_new_c_uuid ]        = { &drbd_nl_new_c_uuid,        0 },
2225 };
2226
2227 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2228 {
2229         struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2230         struct cn_handler_struct *cm;
2231         struct cn_msg *cn_reply;
2232         struct drbd_nl_cfg_reply *reply;
2233         struct drbd_conf *mdev;
2234         int retcode, rr;
2235         int reply_size = sizeof(struct cn_msg)
2236                 + sizeof(struct drbd_nl_cfg_reply)
2237                 + sizeof(short int);
2238
2239         if (!try_module_get(THIS_MODULE)) {
2240                 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2241                 return;
2242         }
2243
2244         if (!cap_raised(current_cap(), CAP_SYS_ADMIN)) {
2245                 retcode = ERR_PERM;
2246                 goto fail;
2247         }
2248
2249         mdev = ensure_mdev(nlp->drbd_minor,
2250                         (nlp->flags & DRBD_NL_CREATE_DEVICE));
2251         if (!mdev) {
2252                 retcode = ERR_MINOR_INVALID;
2253                 goto fail;
2254         }
2255
2256         if (nlp->packet_type >= P_nl_after_last_packet ||
2257             nlp->packet_type == P_return_code_only) {
2258                 retcode = ERR_PACKET_NR;
2259                 goto fail;
2260         }
2261
2262         cm = cnd_table + nlp->packet_type;
2263
2264         /* This may happen if packet number is 0: */
2265         if (cm->function == NULL) {
2266                 retcode = ERR_PACKET_NR;
2267                 goto fail;
2268         }
2269
2270         reply_size += cm->reply_body_size;
2271
2272         /* allocation not in the IO path, cqueue thread context */
2273         cn_reply = kzalloc(reply_size, GFP_KERNEL);
2274         if (!cn_reply) {
2275                 retcode = ERR_NOMEM;
2276                 goto fail;
2277         }
2278         reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2279
2280         reply->packet_type =
2281                 cm->reply_body_size ? nlp->packet_type : P_return_code_only;
2282         reply->minor = nlp->drbd_minor;
2283         reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2284         /* reply->tag_list; might be modified by cm->function. */
2285
2286         rr = cm->function(mdev, nlp, reply);
2287
2288         cn_reply->id = req->id;
2289         cn_reply->seq = req->seq;
2290         cn_reply->ack = req->ack  + 1;
2291         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2292         cn_reply->flags = 0;
2293
2294         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2295         if (rr && rr != -ESRCH)
2296                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2297
2298         kfree(cn_reply);
2299         module_put(THIS_MODULE);
2300         return;
2301  fail:
2302         drbd_nl_send_reply(req, retcode);
2303         module_put(THIS_MODULE);
2304 }
2305
2306 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2307
2308 static unsigned short *
2309 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2310         unsigned short len, int nul_terminated)
2311 {
2312         unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2313         len = (len < l) ? len :  l;
2314         put_unaligned(tag, tl++);
2315         put_unaligned(len, tl++);
2316         memcpy(tl, data, len);
2317         tl = (unsigned short*)((char*)tl + len);
2318         if (nul_terminated)
2319                 *((char*)tl - 1) = 0;
2320         return tl;
2321 }
2322
2323 static unsigned short *
2324 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2325 {
2326         return __tl_add_blob(tl, tag, data, len, 0);
2327 }
2328
2329 static unsigned short *
2330 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2331 {
2332         return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2333 }
2334
2335 static unsigned short *
2336 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2337 {
2338         put_unaligned(tag, tl++);
2339         switch(tag_type(tag)) {
2340         case TT_INTEGER:
2341                 put_unaligned(sizeof(int), tl++);
2342                 put_unaligned(*(int *)val, (int *)tl);
2343                 tl = (unsigned short*)((char*)tl+sizeof(int));
2344                 break;
2345         case TT_INT64:
2346                 put_unaligned(sizeof(u64), tl++);
2347                 put_unaligned(*(u64 *)val, (u64 *)tl);
2348                 tl = (unsigned short*)((char*)tl+sizeof(u64));
2349                 break;
2350         default:
2351                 /* someone did something stupid. */
2352                 ;
2353         }
2354         return tl;
2355 }
2356
2357 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2358 {
2359         char buffer[sizeof(struct cn_msg)+
2360                     sizeof(struct drbd_nl_cfg_reply)+
2361                     sizeof(struct get_state_tag_len_struct)+
2362                     sizeof(short int)];
2363         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2364         struct drbd_nl_cfg_reply *reply =
2365                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2366         unsigned short *tl = reply->tag_list;
2367
2368         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2369
2370         tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2371
2372         put_unaligned(TT_END, tl++); /* Close the tag list */
2373
2374         cn_reply->id.idx = CN_IDX_DRBD;
2375         cn_reply->id.val = CN_VAL_DRBD;
2376
2377         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2378         cn_reply->ack = 0; /* not used here. */
2379         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2380                 (int)((char *)tl - (char *)reply->tag_list);
2381         cn_reply->flags = 0;
2382
2383         reply->packet_type = P_get_state;
2384         reply->minor = mdev_to_minor(mdev);
2385         reply->ret_code = NO_ERROR;
2386
2387         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2388 }
2389
2390 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2391 {
2392         char buffer[sizeof(struct cn_msg)+
2393                     sizeof(struct drbd_nl_cfg_reply)+
2394                     sizeof(struct call_helper_tag_len_struct)+
2395                     sizeof(short int)];
2396         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2397         struct drbd_nl_cfg_reply *reply =
2398                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2399         unsigned short *tl = reply->tag_list;
2400
2401         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2402
2403         tl = tl_add_str(tl, T_helper, helper_name);
2404         put_unaligned(TT_END, tl++); /* Close the tag list */
2405
2406         cn_reply->id.idx = CN_IDX_DRBD;
2407         cn_reply->id.val = CN_VAL_DRBD;
2408
2409         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2410         cn_reply->ack = 0; /* not used here. */
2411         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2412                 (int)((char *)tl - (char *)reply->tag_list);
2413         cn_reply->flags = 0;
2414
2415         reply->packet_type = P_call_helper;
2416         reply->minor = mdev_to_minor(mdev);
2417         reply->ret_code = NO_ERROR;
2418
2419         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2420 }
2421
2422 void drbd_bcast_ee(struct drbd_conf *mdev, const char *reason, const int dgs,
2423                    const char *seen_hash, const char *calc_hash,
2424                            const struct drbd_peer_request *peer_req)
2425 {
2426         struct cn_msg *cn_reply;
2427         struct drbd_nl_cfg_reply *reply;
2428         unsigned short *tl;
2429         struct page *page;
2430         unsigned len;
2431
2432         if (!peer_req)
2433                 return;
2434         if (!reason || !reason[0])
2435                 return;
2436
2437         /* apparently we have to memcpy twice, first to prepare the data for the
2438          * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2439          * netlink skb. */
2440         /* receiver thread context, which is not in the writeout path (of this node),
2441          * but may be in the writeout path of the _other_ node.
2442          * GFP_NOIO to avoid potential "distributed deadlock". */
2443         cn_reply = kzalloc(
2444                 sizeof(struct cn_msg)+
2445                 sizeof(struct drbd_nl_cfg_reply)+
2446                 sizeof(struct dump_ee_tag_len_struct)+
2447                 sizeof(short int),
2448                 GFP_NOIO);
2449
2450         if (!cn_reply) {
2451                 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, "
2452                              "sector %llu, size %u\n",
2453                         (unsigned long long)peer_req->i.sector,
2454                         peer_req->i.size);
2455                 return;
2456         }
2457
2458         reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2459         tl = reply->tag_list;
2460
2461         tl = tl_add_str(tl, T_dump_ee_reason, reason);
2462         tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2463         tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2464         tl = tl_add_int(tl, T_ee_sector, &peer_req->i.sector);
2465         tl = tl_add_int(tl, T_ee_block_id, &peer_req->block_id);
2466
2467         /* dump the first 32k */
2468         len = min_t(unsigned, peer_req->i.size, 32 << 10);
2469         put_unaligned(T_ee_data, tl++);
2470         put_unaligned(len, tl++);
2471
2472         page = peer_req->pages;
2473         page_chain_for_each(page) {
2474                 void *d = kmap_atomic(page, KM_USER0);
2475                 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2476                 memcpy(tl, d, l);
2477                 kunmap_atomic(d, KM_USER0);
2478                 tl = (unsigned short*)((char*)tl + l);
2479                 len -= l;
2480                 if (len == 0)
2481                         break;
2482         }
2483         put_unaligned(TT_END, tl++); /* Close the tag list */
2484
2485         cn_reply->id.idx = CN_IDX_DRBD;
2486         cn_reply->id.val = CN_VAL_DRBD;
2487
2488         cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2489         cn_reply->ack = 0; // not used here.
2490         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2491                 (int)((char*)tl - (char*)reply->tag_list);
2492         cn_reply->flags = 0;
2493
2494         reply->packet_type = P_dump_ee;
2495         reply->minor = mdev_to_minor(mdev);
2496         reply->ret_code = NO_ERROR;
2497
2498         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2499         kfree(cn_reply);
2500 }
2501
2502 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2503 {
2504         char buffer[sizeof(struct cn_msg)+
2505                     sizeof(struct drbd_nl_cfg_reply)+
2506                     sizeof(struct sync_progress_tag_len_struct)+
2507                     sizeof(short int)];
2508         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2509         struct drbd_nl_cfg_reply *reply =
2510                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2511         unsigned short *tl = reply->tag_list;
2512         unsigned long rs_left;
2513         unsigned int res;
2514
2515         /* no local ref, no bitmap, no syncer progress, no broadcast. */
2516         if (!get_ldev(mdev))
2517                 return;
2518         drbd_get_syncer_progress(mdev, &rs_left, &res);
2519         put_ldev(mdev);
2520
2521         tl = tl_add_int(tl, T_sync_progress, &res);
2522         put_unaligned(TT_END, tl++); /* Close the tag list */
2523
2524         cn_reply->id.idx = CN_IDX_DRBD;
2525         cn_reply->id.val = CN_VAL_DRBD;
2526
2527         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2528         cn_reply->ack = 0; /* not used here. */
2529         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2530                 (int)((char *)tl - (char *)reply->tag_list);
2531         cn_reply->flags = 0;
2532
2533         reply->packet_type = P_sync_progress;
2534         reply->minor = mdev_to_minor(mdev);
2535         reply->ret_code = NO_ERROR;
2536
2537         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2538 }
2539
2540 int __init drbd_nl_init(void)
2541 {
2542         static struct cb_id cn_id_drbd;
2543         int err, try=10;
2544
2545         cn_id_drbd.val = CN_VAL_DRBD;
2546         do {
2547                 cn_id_drbd.idx = cn_idx;
2548                 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2549                 if (!err)
2550                         break;
2551                 cn_idx = (cn_idx + CN_IDX_STEP);
2552         } while (try--);
2553
2554         if (err) {
2555                 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2556                 return err;
2557         }
2558
2559         return 0;
2560 }
2561
2562 void drbd_nl_cleanup(void)
2563 {
2564         static struct cb_id cn_id_drbd;
2565
2566         cn_id_drbd.idx = cn_idx;
2567         cn_id_drbd.val = CN_VAL_DRBD;
2568
2569         cn_del_callback(&cn_id_drbd);
2570 }
2571
2572 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2573 {
2574         char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2575         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2576         struct drbd_nl_cfg_reply *reply =
2577                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2578         int rr;
2579
2580         memset(buffer, 0, sizeof(buffer));
2581         cn_reply->id = req->id;
2582
2583         cn_reply->seq = req->seq;
2584         cn_reply->ack = req->ack  + 1;
2585         cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2586         cn_reply->flags = 0;
2587
2588         reply->packet_type = P_return_code_only;
2589         reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2590         reply->ret_code = ret_code;
2591
2592         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2593         if (rr && rr != -ESRCH)
2594                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2595 }
2596