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