drivers/block/drbd/drbd_nl.c

   1 /*
   2    drbd_nl.c
   3
   4    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
   5
   6    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
   7    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
   8    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
   9
  10    drbd is free software; you can redistribute it and/or modify
  11    it under the terms of the GNU General Public License as published by
  12    the Free Software Foundation; either version 2, or (at your option)
  13    any later version.
  14
  15    drbd is distributed in the hope that it will be useful,
  16    but WITHOUT ANY WARRANTY; without even the implied warranty of
  17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18    GNU General Public License for more details.
  19
  20    You should have received a copy of the GNU General Public License
  21    along with drbd; see the file COPYING.  If not, write to
  22    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  23
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/drbd.h>
  28 #include <linux/in.h>
  29 #include <linux/fs.h>
  30 #include <linux/file.h>
  31 #include <linux/slab.h>
  32 #include <linux/connector.h>
  33 #include <linux/blkpg.h>
  34 #include <linux/cpumask.h>
  35 #include "drbd_int.h"
  36 #include "drbd_req.h"
  37 #include "drbd_wrappers.h"
  38 #include <asm/unaligned.h>
  39 #include <linux/drbd_tag_magic.h>
  40 #include <linux/drbd_limits.h>
  41 #include <linux/compiler.h>
  42 #include <linux/kthread.h>
  43
  44 static unsigned short *tl_add_blob(unsigned short *, enum drbd_tags, const void *, int);
  45 static unsigned short *tl_add_str(unsigned short *, enum drbd_tags, const char *);
  46 static unsigned short *tl_add_int(unsigned short *, enum drbd_tags, const void *);
  47
  48 /* see get_sb_bdev and bd_claim */
  49 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
  50
  51 /* Generate the tag_list to struct functions */
  52 #define NL_PACKET(name, number, fields) \
  53 static int name ## _from_tags(struct drbd_conf *mdev, \
  54         unsigned short *tags, struct name *arg) __attribute__ ((unused)); \
  55 static int name ## _from_tags(struct drbd_conf *mdev, \
  56         unsigned short *tags, struct name *arg) \
  57 { \
  58         int tag; \
  59         int dlen; \
  60         \
  61         while ((tag = get_unaligned(tags++)) != TT_END) {       \
  62                 dlen = get_unaligned(tags++);                   \
  63                 switch (tag_number(tag)) { \
  64                 fields \
  65                 default: \
  66                         if (tag & T_MANDATORY) { \
  67                                 dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \
  68                                 return 0; \
  69                         } \
  70                 } \
  71                 tags = (unsigned short *)((char *)tags + dlen); \
  72         } \
  73         return 1; \
  74 }
  75 #define NL_INTEGER(pn, pr, member) \
  76         case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \
  77                 arg->member = get_unaligned((int *)(tags));     \
  78                 break;
  79 #define NL_INT64(pn, pr, member) \
  80         case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \
  81                 arg->member = get_unaligned((u64 *)(tags));     \
  82                 break;
  83 #define NL_BIT(pn, pr, member) \
  84         case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \
  85                 arg->member = *(char *)(tags) ? 1 : 0; \
  86                 break;
  87 #define NL_STRING(pn, pr, member, len) \
  88         case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \
  89                 if (dlen > len) { \
  90                         dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \
  91                                 #member, dlen, (unsigned int)len); \
  92                         return 0; \
  93                 } \
  94                  arg->member ## _len = dlen; \
  95                  memcpy(arg->member, tags, min_t(size_t, dlen, len)); \
  96                  break;
  97 #include "linux/drbd_nl.h"
  98
  99 /* Generate the struct to tag_list functions */
 100 #define NL_PACKET(name, number, fields) \
 101 static unsigned short* \
 102 name ## _to_tags(struct drbd_conf *mdev, \
 103         struct name *arg, unsigned short *tags) __attribute__ ((unused)); \
 104 static unsigned short* \
 105 name ## _to_tags(struct drbd_conf *mdev, \
 106         struct name *arg, unsigned short *tags) \
 107 { \
 108         fields \
 109         return tags; \
 110 }
 111
 112 #define NL_INTEGER(pn, pr, member) \
 113         put_unaligned(pn | pr | TT_INTEGER, tags++);    \
 114         put_unaligned(sizeof(int), tags++);             \
 115         put_unaligned(arg->member, (int *)tags);        \
 116         tags = (unsigned short *)((char *)tags+sizeof(int));
 117 #define NL_INT64(pn, pr, member) \
 118         put_unaligned(pn | pr | TT_INT64, tags++);      \
 119         put_unaligned(sizeof(u64), tags++);             \
 120         put_unaligned(arg->member, (u64 *)tags);        \
 121         tags = (unsigned short *)((char *)tags+sizeof(u64));
 122 #define NL_BIT(pn, pr, member) \
 123         put_unaligned(pn | pr | TT_BIT, tags++);        \
 124         put_unaligned(sizeof(char), tags++);            \
 125         *(char *)tags = arg->member; \
 126         tags = (unsigned short *)((char *)tags+sizeof(char));
 127 #define NL_STRING(pn, pr, member, len) \
 128         put_unaligned(pn | pr | TT_STRING, tags++);     \
 129         put_unaligned(arg->member ## _len, tags++);     \
 130         memcpy(tags, arg->member, arg->member ## _len); \
 131         tags = (unsigned short *)((char *)tags + arg->member ## _len);
 132 #include "linux/drbd_nl.h"
 133
 134 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name);
 135 void drbd_nl_send_reply(struct cn_msg *, int);
 136
 137 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
 138 {
 139         char *envp[] = { "HOME=/",
 140                         "TERM=linux",
 141                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 142                         NULL, /* Will be set to address family */
 143                         NULL, /* Will be set to address */
 144                         NULL };
 145
 146         char mb[12], af[20], ad[60], *afs;
 147         char *argv[] = {usermode_helper, cmd, mb, NULL };
 148         int ret;
 149
 150         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 151
 152         if (get_net_conf(mdev)) {
 153                 switch (((struct sockaddr *)mdev->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->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->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->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);
 173         }
 174
 175         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 176
 177         drbd_bcast_ev_helper(mdev, cmd);
 178         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 179         if (ret)
 180                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 181                                 usermode_helper, cmd, mb,
 182                                 (ret >> 8) & 0xff, ret);
 183         else
 184                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 185                                 usermode_helper, cmd, mb,
 186                                 (ret >> 8) & 0xff, ret);
 187
 188         if (ret < 0) /* Ignore any ERRNOs we got. */
 189                 ret = 0;
 190
 191         return ret;
 192 }
 193
 194 enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
 195 {
 196         char *ex_to_string;
 197         int r;
 198         enum drbd_disk_state nps;
 199         enum drbd_fencing_p fp;
 200
 201         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 202
 203         if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 204                 fp = mdev->ldev->dc.fencing;
 205                 put_ldev(mdev);
 206         } else {
 207                 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n");
 208                 return mdev->state.pdsk;
 209         }
 210
 211         r = drbd_khelper(mdev, "fence-peer");
 212
 213         switch ((r>>8) & 0xff) {
 214         case 3: /* peer is inconsistent */
 215                 ex_to_string = "peer is inconsistent or worse";
 216                 nps = D_INCONSISTENT;
 217                 break;
 218         case 4: /* peer got outdated, or was already outdated */
 219                 ex_to_string = "peer was fenced";
 220                 nps = D_OUTDATED;
 221                 break;
 222         case 5: /* peer was down */
 223                 if (mdev->state.disk == D_UP_TO_DATE) {
 224                         /* we will(have) create(d) a new UUID anyways... */
 225                         ex_to_string = "peer is unreachable, assumed to be dead";
 226                         nps = D_OUTDATED;
 227                 } else {
 228                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 229                         nps = mdev->state.pdsk;
 230                 }
 231                 break;
 232         case 6: /* Peer is primary, voluntarily outdate myself.
 233                  * This is useful when an unconnected R_SECONDARY is asked to
 234                  * become R_PRIMARY, but finds the other peer being active. */
 235                 ex_to_string = "peer is active";
 236                 dev_warn(DEV, "Peer is primary, outdating myself.\n");
 237                 nps = D_UNKNOWN;
 238                 _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE);
 239                 break;
 240         case 7:
 241                 if (fp != FP_STONITH)
 242                         dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n");
 243                 ex_to_string = "peer was stonithed";
 244                 nps = D_OUTDATED;
 245                 break;
 246         default:
 247                 /* The script is broken ... */
 248                 nps = D_UNKNOWN;
 249                 dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 250                 return nps;
 251         }
 252
 253         dev_info(DEV, "fence-peer helper returned %d (%s)\n",
 254                         (r>>8) & 0xff, ex_to_string);
 255         return nps;
 256 }
 257
 258 static int _try_outdate_peer_async(void *data)
 259 {
 260         struct drbd_conf *mdev = (struct drbd_conf *)data;
 261         enum drbd_disk_state nps;
 262
 263         nps = drbd_try_outdate_peer(mdev);
 264         drbd_request_state(mdev, NS(pdsk, nps));
 265
 266         return 0;
 267 }
 268
 269 void drbd_try_outdate_peer_async(struct drbd_conf *mdev)
 270 {
 271         struct task_struct *opa;
 272
 273         opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev));
 274         if (IS_ERR(opa))
 275                 dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n");
 276 }
 277
 278 int drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 279 {
 280         const int max_tries = 4;
 281         int r = 0;
 282         int try = 0;
 283         int forced = 0;
 284         union drbd_state mask, val;
 285         enum drbd_disk_state nps;
 286
 287         if (new_role == R_PRIMARY)
 288                 request_ping(mdev); /* Detect a dead peer ASAP */
 289
 290         mutex_lock(&mdev->state_mutex);
 291
 292         mask.i = 0; mask.role = R_MASK;
 293         val.i  = 0; val.role  = new_role;
 294
 295         while (try++ < max_tries) {
 296                 r = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 297
 298                 /* in case we first succeeded to outdate,
 299                  * but now suddenly could establish a connection */
 300                 if (r == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 301                         val.pdsk = 0;
 302                         mask.pdsk = 0;
 303                         continue;
 304                 }
 305
 306                 if (r == SS_NO_UP_TO_DATE_DISK && force &&
 307                     (mdev->state.disk < D_UP_TO_DATE &&
 308                      mdev->state.disk >= D_INCONSISTENT)) {
 309                         mask.disk = D_MASK;
 310                         val.disk  = D_UP_TO_DATE;
 311                         forced = 1;
 312                         continue;
 313                 }
 314
 315                 if (r == SS_NO_UP_TO_DATE_DISK &&
 316                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 317                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 318                         nps = drbd_try_outdate_peer(mdev);
 319
 320                         if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
 321                                 val.disk = D_UP_TO_DATE;
 322                                 mask.disk = D_MASK;
 323                         }
 324
 325                         val.pdsk = nps;
 326                         mask.pdsk = D_MASK;
 327
 328                         continue;
 329                 }
 330
 331                 if (r == SS_NOTHING_TO_DO)
 332                         goto fail;
 333                 if (r == SS_PRIMARY_NOP && mask.pdsk == 0) {
 334                         nps = drbd_try_outdate_peer(mdev);
 335
 336                         if (force && nps > D_OUTDATED) {
 337                                 dev_warn(DEV, "Forced into split brain situation!\n");
 338                                 nps = D_OUTDATED;
 339                         }
 340
 341                         mask.pdsk = D_MASK;
 342                         val.pdsk  = nps;
 343
 344                         continue;
 345                 }
 346                 if (r == SS_TWO_PRIMARIES) {
 347                         /* Maybe the peer is detected as dead very soon...
 348                            retry at most once more in this case. */
 349                         __set_current_state(TASK_INTERRUPTIBLE);
 350                         schedule_timeout((mdev->net_conf->ping_timeo+1)*HZ/10);
 351                         if (try < max_tries)
 352                                 try = max_tries - 1;
 353                         continue;
 354                 }
 355                 if (r < SS_SUCCESS) {
 356                         r = _drbd_request_state(mdev, mask, val,
 357                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 358                         if (r < SS_SUCCESS)
 359                                 goto fail;
 360                 }
 361                 break;
 362         }
 363
 364         if (r < SS_SUCCESS)
 365                 goto fail;
 366
 367         if (forced)
 368                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 369
 370         /* Wait until nothing is on the fly :) */
 371         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 372
 373         if (new_role == R_SECONDARY) {
 374                 set_disk_ro(mdev->vdisk, TRUE);
 375                 if (get_ldev(mdev)) {
 376                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 377                         put_ldev(mdev);
 378                 }
 379         } else {
 380                 if (get_net_conf(mdev)) {
 381                         mdev->net_conf->want_lose = 0;
 382                         put_net_conf(mdev);
 383                 }
 384                 set_disk_ro(mdev->vdisk, FALSE);
 385                 if (get_ldev(mdev)) {
 386                         if (((mdev->state.conn < C_CONNECTED ||
 387                                mdev->state.pdsk <= D_FAILED)
 388                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 389                                 drbd_uuid_new_current(mdev);
 390
 391                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 392                         put_ldev(mdev);
 393                 }
 394         }
 395
 396         if ((new_role == R_SECONDARY) && get_ldev(mdev)) {
 397                 drbd_al_to_on_disk_bm(mdev);
 398                 put_ldev(mdev);
 399         }
 400
 401         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 402                 /* if this was forced, we should consider sync */
 403                 if (forced)
 404                         drbd_send_uuids(mdev);
 405                 drbd_send_state(mdev);
 406         }
 407
 408         drbd_md_sync(mdev);
 409
 410         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 411  fail:
 412         mutex_unlock(&mdev->state_mutex);
 413         return r;
 414 }
 415
 416 static struct drbd_conf *ensure_mdev(int minor, int create)
 417 {
 418         struct drbd_conf *mdev;
 419
 420         if (minor >= minor_count)
 421                 return NULL;
 422
 423         mdev = minor_to_mdev(minor);
 424
 425         if (!mdev && create) {
 426                 struct gendisk *disk = NULL;
 427                 mdev = drbd_new_device(minor);
 428
 429                 spin_lock_irq(&drbd_pp_lock);
 430                 if (minor_table[minor] == NULL) {
 431                         minor_table[minor] = mdev;
 432                         disk = mdev->vdisk;
 433                         mdev = NULL;
 434                 } /* else: we lost the race */
 435                 spin_unlock_irq(&drbd_pp_lock);
 436
 437                 if (disk) /* we won the race above */
 438                         /* in case we ever add a drbd_delete_device(),
 439                          * don't forget the del_gendisk! */
 440                         add_disk(disk);
 441                 else /* we lost the race above */
 442                         drbd_free_mdev(mdev);
 443
 444                 mdev = minor_to_mdev(minor);
 445         }
 446
 447         return mdev;
 448 }
 449
 450 static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 451                            struct drbd_nl_cfg_reply *reply)
 452 {
 453         struct primary primary_args;
 454
 455         memset(&primary_args, 0, sizeof(struct primary));
 456         if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) {
 457                 reply->ret_code = ERR_MANDATORY_TAG;
 458                 return 0;
 459         }
 460
 461         reply->ret_code =
 462                 drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force);
 463
 464         return 0;
 465 }
 466
 467 static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 468                              struct drbd_nl_cfg_reply *reply)
 469 {
 470         reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0);
 471
 472         return 0;
 473 }
 474
 475 /* initializes the md.*_offset members, so we are able to find
 476  * the on disk meta data */
 477 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 478                                        struct drbd_backing_dev *bdev)
 479 {
 480         sector_t md_size_sect = 0;
 481         switch (bdev->dc.meta_dev_idx) {
 482         default:
 483                 /* v07 style fixed size indexed meta data */
 484                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 485                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 486                 bdev->md.al_offset = MD_AL_OFFSET;
 487                 bdev->md.bm_offset = MD_BM_OFFSET;
 488                 break;
 489         case DRBD_MD_INDEX_FLEX_EXT:
 490                 /* just occupy the full device; unit: sectors */
 491                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 492                 bdev->md.md_offset = 0;
 493                 bdev->md.al_offset = MD_AL_OFFSET;
 494                 bdev->md.bm_offset = MD_BM_OFFSET;
 495                 break;
 496         case DRBD_MD_INDEX_INTERNAL:
 497         case DRBD_MD_INDEX_FLEX_INT:
 498                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 499                 /* al size is still fixed */
 500                 bdev->md.al_offset = -MD_AL_MAX_SIZE;
 501                 /* we need (slightly less than) ~ this much bitmap sectors: */
 502                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 503                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 504                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 505                 md_size_sect = ALIGN(md_size_sect, 8);
 506
 507                 /* plus the "drbd meta data super block",
 508                  * and the activity log; */
 509                 md_size_sect += MD_BM_OFFSET;
 510
 511                 bdev->md.md_size_sect = md_size_sect;
 512                 /* bitmap offset is adjusted by 'super' block size */
 513                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 514                 break;
 515         }
 516 }
 517
 518 char *ppsize(char *buf, unsigned long long size)
 519 {
 520         /* Needs 9 bytes at max. */
 521         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 522         int base = 0;
 523         while (size >= 10000) {
 524                 /* shift + round */
 525                 size = (size >> 10) + !!(size & (1<<9));
 526                 base++;
 527         }
 528         sprintf(buf, "%lu %cB", (long)size, units[base]);
 529
 530         return buf;
 531 }
 532
 533 /* there is still a theoretical deadlock when called from receiver
 534  * on an D_INCONSISTENT R_PRIMARY:
 535  *  remote READ does inc_ap_bio, receiver would need to receive answer
 536  *  packet from remote to dec_ap_bio again.
 537  *  receiver receive_sizes(), comes here,
 538  *  waits for ap_bio_cnt == 0. -> deadlock.
 539  * but this cannot happen, actually, because:
 540  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 541  *  (not connected, or bad/no disk on peer):
 542  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 543  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 544  *  peer may not initiate a resize.
 545  */
 546 void drbd_suspend_io(struct drbd_conf *mdev)
 547 {
 548         set_bit(SUSPEND_IO, &mdev->flags);
 549         if (mdev->state.susp)
 550                 return;
 551         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 552 }
 553
 554 void drbd_resume_io(struct drbd_conf *mdev)
 555 {
 556         clear_bit(SUSPEND_IO, &mdev->flags);
 557         wake_up(&mdev->misc_wait);
 558 }
 559
 560 /**
 561  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 562  * @mdev:       DRBD device.
 563  *
 564  * Returns 0 on success, negative return values indicate errors.
 565  * You should call drbd_md_sync() after calling this function.
 566  */
 567 enum determine_dev_size drbd_determin_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 568 {
 569         sector_t prev_first_sect, prev_size; /* previous meta location */
 570         sector_t la_size;
 571         sector_t size;
 572         char ppb[10];
 573
 574         int md_moved, la_size_changed;
 575         enum determine_dev_size rv = unchanged;
 576
 577         /* race:
 578          * application request passes inc_ap_bio,
 579          * but then cannot get an AL-reference.
 580          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 581          *
 582          * to avoid that:
 583          * Suspend IO right here.
 584          * still lock the act_log to not trigger ASSERTs there.
 585          */
 586         drbd_suspend_io(mdev);
 587
 588         /* no wait necessary anymore, actually we could assert that */
 589         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 590
 591         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 592         prev_size = mdev->ldev->md.md_size_sect;
 593         la_size = mdev->ldev->md.la_size_sect;
 594
 595         /* TODO: should only be some assert here, not (re)init... */
 596         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 597
 598         size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
 599
 600         if (drbd_get_capacity(mdev->this_bdev) != size ||
 601             drbd_bm_capacity(mdev) != size) {
 602                 int err;
 603                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 604                 if (unlikely(err)) {
 605                         /* currently there is only one error: ENOMEM! */
 606                         size = drbd_bm_capacity(mdev)>>1;
 607                         if (size == 0) {
 608                                 dev_err(DEV, "OUT OF MEMORY! "
 609                                     "Could not allocate bitmap!\n");
 610                         } else {
 611                                 dev_err(DEV, "BM resizing failed. "
 612                                     "Leaving size unchanged at size = %lu KB\n",
 613                                     (unsigned long)size);
 614                         }
 615                         rv = dev_size_error;
 616                 }
 617                 /* racy, see comments above. */
 618                 drbd_set_my_capacity(mdev, size);
 619                 mdev->ldev->md.la_size_sect = size;
 620                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 621                      (unsigned long long)size>>1);
 622         }
 623         if (rv == dev_size_error)
 624                 goto out;
 625
 626         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 627
 628         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 629                 || prev_size       != mdev->ldev->md.md_size_sect;
 630
 631         if (la_size_changed || md_moved) {
 632                 drbd_al_shrink(mdev); /* All extents inactive. */
 633                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 634                          la_size_changed && md_moved ? "size changed and md moved" :
 635                          la_size_changed ? "size changed" : "md moved");
 636                 rv = drbd_bitmap_io(mdev, &drbd_bm_write, "size changed"); /* does drbd_resume_io() ! */
 637                 drbd_md_mark_dirty(mdev);
 638         }
 639
 640         if (size > la_size)
 641                 rv = grew;
 642         if (size < la_size)
 643                 rv = shrunk;
 644 out:
 645         lc_unlock(mdev->act_log);
 646         wake_up(&mdev->al_wait);
 647         drbd_resume_io(mdev);
 648
 649         return rv;
 650 }
 651
 652 sector_t
 653 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
 654 {
 655         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 656         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 657         sector_t m_size; /* my size */
 658         sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
 659         sector_t size = 0;
 660
 661         m_size = drbd_get_max_capacity(bdev);
 662
 663         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 664                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 665                 p_size = m_size;
 666         }
 667
 668         if (p_size && m_size) {
 669                 size = min_t(sector_t, p_size, m_size);
 670         } else {
 671                 if (la_size) {
 672                         size = la_size;
 673                         if (m_size && m_size < size)
 674                                 size = m_size;
 675                         if (p_size && p_size < size)
 676                                 size = p_size;
 677                 } else {
 678                         if (m_size)
 679                                 size = m_size;
 680                         if (p_size)
 681                                 size = p_size;
 682                 }
 683         }
 684
 685         if (size == 0)
 686                 dev_err(DEV, "Both nodes diskless!\n");
 687
 688         if (u_size) {
 689                 if (u_size > size)
 690                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 691                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 692                 else
 693                         size = u_size;
 694         }
 695
 696         return size;
 697 }
 698
 699 /**
 700  * drbd_check_al_size() - Ensures that the AL is of the right size
 701  * @mdev:       DRBD device.
 702  *
 703  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 704  * failed, and 0 on success. You should call drbd_md_sync() after you called
 705  * this function.
 706  */
 707 static int drbd_check_al_size(struct drbd_conf *mdev)
 708 {
 709         struct lru_cache *n, *t;
 710         struct lc_element *e;
 711         unsigned int in_use;
 712         int i;
 713
 714         ERR_IF(mdev->sync_conf.al_extents < 7)
 715                 mdev->sync_conf.al_extents = 127;
 716
 717         if (mdev->act_log &&
 718             mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
 719                 return 0;
 720
 721         in_use = 0;
 722         t = mdev->act_log;
 723         n = lc_create("act_log", drbd_al_ext_cache,
 724                 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
 725
 726         if (n == NULL) {
 727                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 728                 return -ENOMEM;
 729         }
 730         spin_lock_irq(&mdev->al_lock);
 731         if (t) {
 732                 for (i = 0; i < t->nr_elements; i++) {
 733                         e = lc_element_by_index(t, i);
 734                         if (e->refcnt)
 735                                 dev_err(DEV, "refcnt(%d)==%d\n",
 736                                     e->lc_number, e->refcnt);
 737                         in_use += e->refcnt;
 738                 }
 739         }
 740         if (!in_use)
 741                 mdev->act_log = n;
 742         spin_unlock_irq(&mdev->al_lock);
 743         if (in_use) {
 744                 dev_err(DEV, "Activity log still in use!\n");
 745                 lc_destroy(n);
 746                 return -EBUSY;
 747         } else {
 748                 if (t)
 749                         lc_destroy(t);
 750         }
 751         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 752         return 0;
 753 }
 754
 755 void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_seg_s) __must_hold(local)
 756 {
 757         struct request_queue * const q = mdev->rq_queue;
 758         struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 759         int max_segments = mdev->ldev->dc.max_bio_bvecs;
 760
 761         max_seg_s = min(queue_max_sectors(b) * queue_logical_block_size(b), max_seg_s);
 762
 763         blk_queue_max_hw_sectors(q, max_seg_s >> 9);
 764         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
 765         blk_queue_max_segment_size(q, max_seg_s);
 766         blk_queue_logical_block_size(q, 512);
 767         blk_queue_segment_boundary(q, PAGE_SIZE-1);
 768         blk_stack_limits(&q->limits, &b->limits, 0);
 769
 770         if (b->merge_bvec_fn)
 771                 dev_warn(DEV, "Backing device's merge_bvec_fn() = %p\n",
 772                      b->merge_bvec_fn);
 773         dev_info(DEV, "max_segment_size ( = BIO size ) = %u\n", queue_max_segment_size(q));
 774
 775         if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
 776                 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
 777                      q->backing_dev_info.ra_pages,
 778                      b->backing_dev_info.ra_pages);
 779                 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
 780         }
 781 }
 782
 783 /* serialize deconfig (worker exiting, doing cleanup)
 784  * and reconfig (drbdsetup disk, drbdsetup net)
 785  *
 786  * Wait for a potentially exiting worker, then restart it,
 787  * or start a new one.  Flush any pending work, there may still be an
 788  * after_state_change queued.
 789  */
 790 static void drbd_reconfig_start(struct drbd_conf *mdev)
 791 {
 792         wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
 793         wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
 794         drbd_thread_start(&mdev->worker);
 795         drbd_flush_workqueue(mdev);
 796 }
 797
 798 /* if still unconfigured, stops worker again.
 799  * if configured now, clears CONFIG_PENDING.
 800  * wakes potential waiters */
 801 static void drbd_reconfig_done(struct drbd_conf *mdev)
 802 {
 803         spin_lock_irq(&mdev->req_lock);
 804         if (mdev->state.disk == D_DISKLESS &&
 805             mdev->state.conn == C_STANDALONE &&
 806             mdev->state.role == R_SECONDARY) {
 807                 set_bit(DEVICE_DYING, &mdev->flags);
 808                 drbd_thread_stop_nowait(&mdev->worker);
 809         } else
 810                 clear_bit(CONFIG_PENDING, &mdev->flags);
 811         spin_unlock_irq(&mdev->req_lock);
 812         wake_up(&mdev->state_wait);
 813 }
 814
 815 /* Make sure IO is suspended before calling this function(). */
 816 static void drbd_suspend_al(struct drbd_conf *mdev)
 817 {
 818         int s = 0;
 819
 820         if (lc_try_lock(mdev->act_log)) {
 821                 drbd_al_shrink(mdev);
 822                 lc_unlock(mdev->act_log);
 823         } else {
 824                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
 825                 return;
 826         }
 827
 828         spin_lock_irq(&mdev->req_lock);
 829         if (mdev->state.conn < C_CONNECTED)
 830                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
 831
 832         spin_unlock_irq(&mdev->req_lock);
 833
 834         if (s)
 835                 dev_info(DEV, "Suspended AL updates\n");
 836 }
 837
 838 /* does always return 0;
 839  * interesting return code is in reply->ret_code */
 840 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 841                              struct drbd_nl_cfg_reply *reply)
 842 {
 843         enum drbd_ret_codes retcode;
 844         enum determine_dev_size dd;
 845         sector_t max_possible_sectors;
 846         sector_t min_md_device_sectors;
 847         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
 848         struct inode *inode, *inode2;
 849         struct lru_cache *resync_lru = NULL;
 850         union drbd_state ns, os;
 851         int rv;
 852         int cp_discovered = 0;
 853         int logical_block_size;
 854
 855         drbd_reconfig_start(mdev);
 856
 857         /* if you want to reconfigure, please tear down first */
 858         if (mdev->state.disk > D_DISKLESS) {
 859                 retcode = ERR_DISK_CONFIGURED;
 860                 goto fail;
 861         }
 862
 863         /* allocation not in the IO path, cqueue thread context */
 864         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
 865         if (!nbc) {
 866                 retcode = ERR_NOMEM;
 867                 goto fail;
 868         }
 869
 870         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
 871         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
 872         nbc->dc.fencing       = DRBD_FENCING_DEF;
 873         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
 874
 875         if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
 876                 retcode = ERR_MANDATORY_TAG;
 877                 goto fail;
 878         }
 879
 880         if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
 881                 retcode = ERR_MD_IDX_INVALID;
 882                 goto fail;
 883         }
 884
 885         if (get_net_conf(mdev)) {
 886                 int prot = mdev->net_conf->wire_protocol;
 887                 put_net_conf(mdev);
 888                 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
 889                         retcode = ERR_STONITH_AND_PROT_A;
 890                         goto fail;
 891                 }
 892         }
 893
 894         nbc->lo_file = filp_open(nbc->dc.backing_dev, O_RDWR, 0);
 895         if (IS_ERR(nbc->lo_file)) {
 896                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
 897                     PTR_ERR(nbc->lo_file));
 898                 nbc->lo_file = NULL;
 899                 retcode = ERR_OPEN_DISK;
 900                 goto fail;
 901         }
 902
 903         inode = nbc->lo_file->f_dentry->d_inode;
 904
 905         if (!S_ISBLK(inode->i_mode)) {
 906                 retcode = ERR_DISK_NOT_BDEV;
 907                 goto fail;
 908         }
 909
 910         nbc->md_file = filp_open(nbc->dc.meta_dev, O_RDWR, 0);
 911         if (IS_ERR(nbc->md_file)) {
 912                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
 913                     PTR_ERR(nbc->md_file));
 914                 nbc->md_file = NULL;
 915                 retcode = ERR_OPEN_MD_DISK;
 916                 goto fail;
 917         }
 918
 919         inode2 = nbc->md_file->f_dentry->d_inode;
 920
 921         if (!S_ISBLK(inode2->i_mode)) {
 922                 retcode = ERR_MD_NOT_BDEV;
 923                 goto fail;
 924         }
 925
 926         nbc->backing_bdev = inode->i_bdev;
 927         if (bd_claim(nbc->backing_bdev, mdev)) {
 928                 printk(KERN_ERR "drbd: bd_claim(%p,%p); failed [%p;%p;%u]\n",
 929                        nbc->backing_bdev, mdev,
 930                        nbc->backing_bdev->bd_holder,
 931                        nbc->backing_bdev->bd_contains->bd_holder,
 932                        nbc->backing_bdev->bd_holders);
 933                 retcode = ERR_BDCLAIM_DISK;
 934                 goto fail;
 935         }
 936
 937         resync_lru = lc_create("resync", drbd_bm_ext_cache,
 938                         61, sizeof(struct bm_extent),
 939                         offsetof(struct bm_extent, lce));
 940         if (!resync_lru) {
 941                 retcode = ERR_NOMEM;
 942                 goto release_bdev_fail;
 943         }
 944
 945         /* meta_dev_idx >= 0: external fixed size,
 946          * possibly multiple drbd sharing one meta device.
 947          * TODO in that case, paranoia check that [md_bdev, meta_dev_idx] is
 948          * not yet used by some other drbd minor!
 949          * (if you use drbd.conf + drbdadm,
 950          * that should check it for you already; but if you don't, or someone
 951          * fooled it, we need to double check here) */
 952         nbc->md_bdev = inode2->i_bdev;
 953         if (bd_claim(nbc->md_bdev, (nbc->dc.meta_dev_idx < 0) ? (void *)mdev
 954                                 : (void *) drbd_m_holder)) {
 955                 retcode = ERR_BDCLAIM_MD_DISK;
 956                 goto release_bdev_fail;
 957         }
 958
 959         if ((nbc->backing_bdev == nbc->md_bdev) !=
 960             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
 961              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
 962                 retcode = ERR_MD_IDX_INVALID;
 963                 goto release_bdev2_fail;
 964         }
 965
 966         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
 967         drbd_md_set_sector_offsets(mdev, nbc);
 968
 969         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
 970                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
 971                         (unsigned long long) drbd_get_max_capacity(nbc),
 972                         (unsigned long long) nbc->dc.disk_size);
 973                 retcode = ERR_DISK_TO_SMALL;
 974                 goto release_bdev2_fail;
 975         }
 976
 977         if (nbc->dc.meta_dev_idx < 0) {
 978                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
 979                 /* at least one MB, otherwise it does not make sense */
 980                 min_md_device_sectors = (2<<10);
 981         } else {
 982                 max_possible_sectors = DRBD_MAX_SECTORS;
 983                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
 984         }
 985
 986         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
 987                 retcode = ERR_MD_DISK_TO_SMALL;
 988                 dev_warn(DEV, "refusing attach: md-device too small, "
 989                      "at least %llu sectors needed for this meta-disk type\n",
 990                      (unsigned long long) min_md_device_sectors);
 991                 goto release_bdev2_fail;
 992         }
 993
 994         /* Make sure the new disk is big enough
 995          * (we may currently be R_PRIMARY with no local disk...) */
 996         if (drbd_get_max_capacity(nbc) <
 997             drbd_get_capacity(mdev->this_bdev)) {
 998                 retcode = ERR_DISK_TO_SMALL;
 999                 goto release_bdev2_fail;
1000         }
1001
1002         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1003
1004         if (nbc->known_size > max_possible_sectors) {
1005                 dev_warn(DEV, "==> truncating very big lower level device "
1006                         "to currently maximum possible %llu sectors <==\n",
1007                         (unsigned long long) max_possible_sectors);
1008                 if (nbc->dc.meta_dev_idx >= 0)
1009                         dev_warn(DEV, "==>> using internal or flexible "
1010                                       "meta data may help <<==\n");
1011         }
1012
1013         drbd_suspend_io(mdev);
1014         /* also wait for the last barrier ack. */
1015         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || mdev->state.susp);
1016         /* and for any other previously queued work */
1017         drbd_flush_workqueue(mdev);
1018
1019         retcode = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1020         drbd_resume_io(mdev);
1021         if (retcode < SS_SUCCESS)
1022                 goto release_bdev2_fail;
1023
1024         if (!get_ldev_if_state(mdev, D_ATTACHING))
1025                 goto force_diskless;
1026
1027         drbd_md_set_sector_offsets(mdev, nbc);
1028
1029         /* allocate a second IO page if logical_block_size != 512 */
1030         logical_block_size = bdev_logical_block_size(nbc->md_bdev);
1031         if (logical_block_size == 0)
1032                 logical_block_size = MD_SECTOR_SIZE;
1033
1034         if (logical_block_size != MD_SECTOR_SIZE) {
1035                 if (!mdev->md_io_tmpp) {
1036                         struct page *page = alloc_page(GFP_NOIO);
1037                         if (!page)
1038                                 goto force_diskless_dec;
1039
1040                         dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
1041                              logical_block_size, MD_SECTOR_SIZE);
1042                         dev_warn(DEV, "Workaround engaged (has performance impact).\n");
1043
1044                         mdev->md_io_tmpp = page;
1045                 }
1046         }
1047
1048         if (!mdev->bitmap) {
1049                 if (drbd_bm_init(mdev)) {
1050                         retcode = ERR_NOMEM;
1051                         goto force_diskless_dec;
1052                 }
1053         }
1054
1055         retcode = drbd_md_read(mdev, nbc);
1056         if (retcode != NO_ERROR)
1057                 goto force_diskless_dec;
1058
1059         if (mdev->state.conn < C_CONNECTED &&
1060             mdev->state.role == R_PRIMARY &&
1061             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1062                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1063                     (unsigned long long)mdev->ed_uuid);
1064                 retcode = ERR_DATA_NOT_CURRENT;
1065                 goto force_diskless_dec;
1066         }
1067
1068         /* Since we are diskless, fix the activity log first... */
1069         if (drbd_check_al_size(mdev)) {
1070                 retcode = ERR_NOMEM;
1071                 goto force_diskless_dec;
1072         }
1073
1074         /* Prevent shrinking of consistent devices ! */
1075         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1076             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1077                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1078                 retcode = ERR_DISK_TO_SMALL;
1079                 goto force_diskless_dec;
1080         }
1081
1082         if (!drbd_al_read_log(mdev, nbc)) {
1083                 retcode = ERR_IO_MD_DISK;
1084                 goto force_diskless_dec;
1085         }
1086
1087         /* Reset the "barriers don't work" bits here, then force meta data to
1088          * be written, to ensure we determine if barriers are supported. */
1089         if (nbc->dc.no_md_flush)
1090                 set_bit(MD_NO_BARRIER, &mdev->flags);
1091         else
1092                 clear_bit(MD_NO_BARRIER, &mdev->flags);
1093
1094         /* Point of no return reached.
1095          * Devices and memory are no longer released by error cleanup below.
1096          * now mdev takes over responsibility, and the state engine should
1097          * clean it up somewhere.  */
1098         D_ASSERT(mdev->ldev == NULL);
1099         mdev->ldev = nbc;
1100         mdev->resync = resync_lru;
1101         nbc = NULL;
1102         resync_lru = NULL;
1103
1104         mdev->write_ordering = WO_bio_barrier;
1105         drbd_bump_write_ordering(mdev, WO_bio_barrier);
1106
1107         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1108                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1109         else
1110                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1111
1112         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1113             !(mdev->state.role == R_PRIMARY && mdev->state.susp &&
1114               mdev->sync_conf.on_no_data == OND_SUSPEND_IO)) {
1115                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1116                 cp_discovered = 1;
1117         }
1118
1119         mdev->send_cnt = 0;
1120         mdev->recv_cnt = 0;
1121         mdev->read_cnt = 0;
1122         mdev->writ_cnt = 0;
1123
1124         drbd_setup_queue_param(mdev, mdev->state.conn == C_CONNECTED &&
1125                                mdev->agreed_pro_version < 95 ?
1126                                DRBD_MAX_SIZE_H80_PACKET : DRBD_MAX_SEGMENT_SIZE);
1127
1128         /* If I am currently not R_PRIMARY,
1129          * but meta data primary indicator is set,
1130          * I just now recover from a hard crash,
1131          * and have been R_PRIMARY before that crash.
1132          *
1133          * Now, if I had no connection before that crash
1134          * (have been degraded R_PRIMARY), chances are that
1135          * I won't find my peer now either.
1136          *
1137          * In that case, and _only_ in that case,
1138          * we use the degr-wfc-timeout instead of the default,
1139          * so we can automatically recover from a crash of a
1140          * degraded but active "cluster" after a certain timeout.
1141          */
1142         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1143         if (mdev->state.role != R_PRIMARY &&
1144              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1145             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1146                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1147
1148         dd = drbd_determin_dev_size(mdev, 0);
1149         if (dd == dev_size_error) {
1150                 retcode = ERR_NOMEM_BITMAP;
1151                 goto force_diskless_dec;
1152         } else if (dd == grew)
1153                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1154
1155         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1156                 dev_info(DEV, "Assuming that all blocks are out of sync "
1157                      "(aka FullSync)\n");
1158                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from attaching")) {
1159                         retcode = ERR_IO_MD_DISK;
1160                         goto force_diskless_dec;
1161                 }
1162         } else {
1163                 if (drbd_bitmap_io(mdev, &drbd_bm_read, "read from attaching") < 0) {
1164                         retcode = ERR_IO_MD_DISK;
1165                         goto force_diskless_dec;
1166                 }
1167         }
1168
1169         if (cp_discovered) {
1170                 drbd_al_apply_to_bm(mdev);
1171                 drbd_al_to_on_disk_bm(mdev);
1172         }
1173
1174         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1175                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1176
1177         spin_lock_irq(&mdev->req_lock);
1178         os = mdev->state;
1179         ns.i = os.i;
1180         /* If MDF_CONSISTENT is not set go into inconsistent state,
1181            otherwise investigate MDF_WasUpToDate...
1182            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1183            otherwise into D_CONSISTENT state.
1184         */
1185         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1186                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1187                         ns.disk = D_CONSISTENT;
1188                 else
1189                         ns.disk = D_OUTDATED;
1190         } else {
1191                 ns.disk = D_INCONSISTENT;
1192         }
1193
1194         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1195                 ns.pdsk = D_OUTDATED;
1196
1197         if ( ns.disk == D_CONSISTENT &&
1198             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1199                 ns.disk = D_UP_TO_DATE;
1200
1201         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1202            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1203            this point, because drbd_request_state() modifies these
1204            flags. */
1205
1206         /* In case we are C_CONNECTED postpone any decision on the new disk
1207            state after the negotiation phase. */
1208         if (mdev->state.conn == C_CONNECTED) {
1209                 mdev->new_state_tmp.i = ns.i;
1210                 ns.i = os.i;
1211                 ns.disk = D_NEGOTIATING;
1212
1213                 /* We expect to receive up-to-date UUIDs soon.
1214                    To avoid a race in receive_state, free p_uuid while
1215                    holding req_lock. I.e. atomic with the state change */
1216                 kfree(mdev->p_uuid);
1217                 mdev->p_uuid = NULL;
1218         }
1219
1220         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1221         ns = mdev->state;
1222         spin_unlock_irq(&mdev->req_lock);
1223
1224         if (rv < SS_SUCCESS)
1225                 goto force_diskless_dec;
1226
1227         if (mdev->state.role == R_PRIMARY)
1228                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1229         else
1230                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1231
1232         drbd_md_mark_dirty(mdev);
1233         drbd_md_sync(mdev);
1234
1235         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1236         put_ldev(mdev);
1237         reply->ret_code = retcode;
1238         drbd_reconfig_done(mdev);
1239         return 0;
1240
1241  force_diskless_dec:
1242         put_ldev(mdev);
1243  force_diskless:
1244         drbd_force_state(mdev, NS(disk, D_DISKLESS));
1245         drbd_md_sync(mdev);
1246  release_bdev2_fail:
1247         if (nbc)
1248                 bd_release(nbc->md_bdev);
1249  release_bdev_fail:
1250         if (nbc)
1251                 bd_release(nbc->backing_bdev);
1252  fail:
1253         if (nbc) {
1254                 if (nbc->lo_file)
1255                         fput(nbc->lo_file);
1256                 if (nbc->md_file)
1257                         fput(nbc->md_file);
1258                 kfree(nbc);
1259         }
1260         lc_destroy(resync_lru);
1261
1262         reply->ret_code = retcode;
1263         drbd_reconfig_done(mdev);
1264         return 0;
1265 }
1266
1267 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1268                           struct drbd_nl_cfg_reply *reply)
1269 {
1270         reply->ret_code = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1271         return 0;
1272 }
1273
1274 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1275                             struct drbd_nl_cfg_reply *reply)
1276 {
1277         int i, ns;
1278         enum drbd_ret_codes retcode;
1279         struct net_conf *new_conf = NULL;
1280         struct crypto_hash *tfm = NULL;
1281         struct crypto_hash *integrity_w_tfm = NULL;
1282         struct crypto_hash *integrity_r_tfm = NULL;
1283         struct hlist_head *new_tl_hash = NULL;
1284         struct hlist_head *new_ee_hash = NULL;
1285         struct drbd_conf *odev;
1286         char hmac_name[CRYPTO_MAX_ALG_NAME];
1287         void *int_dig_out = NULL;
1288         void *int_dig_in = NULL;
1289         void *int_dig_vv = NULL;
1290         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1291
1292         drbd_reconfig_start(mdev);
1293
1294         if (mdev->state.conn > C_STANDALONE) {
1295                 retcode = ERR_NET_CONFIGURED;
1296                 goto fail;
1297         }
1298
1299         /* allocation not in the IO path, cqueue thread context */
1300         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1301         if (!new_conf) {
1302                 retcode = ERR_NOMEM;
1303                 goto fail;
1304         }
1305
1306         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1307         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1308         new_conf->ping_int         = DRBD_PING_INT_DEF;
1309         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1310         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1311         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1312         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1313         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1314         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1315         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1316         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1317         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1318         new_conf->want_lose        = 0;
1319         new_conf->two_primaries    = 0;
1320         new_conf->wire_protocol    = DRBD_PROT_C;
1321         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1322         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1323
1324         if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1325                 retcode = ERR_MANDATORY_TAG;
1326                 goto fail;
1327         }
1328
1329         if (new_conf->two_primaries
1330             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1331                 retcode = ERR_NOT_PROTO_C;
1332                 goto fail;
1333         }
1334
1335         if (get_ldev(mdev)) {
1336                 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1337                 put_ldev(mdev);
1338                 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1339                         retcode = ERR_STONITH_AND_PROT_A;
1340                         goto fail;
1341                 }
1342         }
1343
1344         if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1345                 retcode = ERR_DISCARD;
1346                 goto fail;
1347         }
1348
1349         retcode = NO_ERROR;
1350
1351         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1352         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1353         for (i = 0; i < minor_count; i++) {
1354                 odev = minor_to_mdev(i);
1355                 if (!odev || odev == mdev)
1356                         continue;
1357                 if (get_net_conf(odev)) {
1358                         taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1359                         if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1360                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1361                                 retcode = ERR_LOCAL_ADDR;
1362
1363                         taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1364                         if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1365                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1366                                 retcode = ERR_PEER_ADDR;
1367
1368                         put_net_conf(odev);
1369                         if (retcode != NO_ERROR)
1370                                 goto fail;
1371                 }
1372         }
1373
1374         if (new_conf->cram_hmac_alg[0] != 0) {
1375                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1376                         new_conf->cram_hmac_alg);
1377                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1378                 if (IS_ERR(tfm)) {
1379                         tfm = NULL;
1380                         retcode = ERR_AUTH_ALG;
1381                         goto fail;
1382                 }
1383
1384                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1385                         retcode = ERR_AUTH_ALG_ND;
1386                         goto fail;
1387                 }
1388         }
1389
1390         if (new_conf->integrity_alg[0]) {
1391                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1392                 if (IS_ERR(integrity_w_tfm)) {
1393                         integrity_w_tfm = NULL;
1394                         retcode=ERR_INTEGRITY_ALG;
1395                         goto fail;
1396                 }
1397
1398                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1399                         retcode=ERR_INTEGRITY_ALG_ND;
1400                         goto fail;
1401                 }
1402
1403                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1404                 if (IS_ERR(integrity_r_tfm)) {
1405                         integrity_r_tfm = NULL;
1406                         retcode=ERR_INTEGRITY_ALG;
1407                         goto fail;
1408                 }
1409         }
1410
1411         ns = new_conf->max_epoch_size/8;
1412         if (mdev->tl_hash_s != ns) {
1413                 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1414                 if (!new_tl_hash) {
1415                         retcode = ERR_NOMEM;
1416                         goto fail;
1417                 }
1418         }
1419
1420         ns = new_conf->max_buffers/8;
1421         if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1422                 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1423                 if (!new_ee_hash) {
1424                         retcode = ERR_NOMEM;
1425                         goto fail;
1426                 }
1427         }
1428
1429         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1430
1431         if (integrity_w_tfm) {
1432                 i = crypto_hash_digestsize(integrity_w_tfm);
1433                 int_dig_out = kmalloc(i, GFP_KERNEL);
1434                 if (!int_dig_out) {
1435                         retcode = ERR_NOMEM;
1436                         goto fail;
1437                 }
1438                 int_dig_in = kmalloc(i, GFP_KERNEL);
1439                 if (!int_dig_in) {
1440                         retcode = ERR_NOMEM;
1441                         goto fail;
1442                 }
1443                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1444                 if (!int_dig_vv) {
1445                         retcode = ERR_NOMEM;
1446                         goto fail;
1447                 }
1448         }
1449
1450         if (!mdev->bitmap) {
1451                 if(drbd_bm_init(mdev)) {
1452                         retcode = ERR_NOMEM;
1453                         goto fail;
1454                 }
1455         }
1456
1457         drbd_flush_workqueue(mdev);
1458         spin_lock_irq(&mdev->req_lock);
1459         if (mdev->net_conf != NULL) {
1460                 retcode = ERR_NET_CONFIGURED;
1461                 spin_unlock_irq(&mdev->req_lock);
1462                 goto fail;
1463         }
1464         mdev->net_conf = new_conf;
1465
1466         mdev->send_cnt = 0;
1467         mdev->recv_cnt = 0;
1468
1469         if (new_tl_hash) {
1470                 kfree(mdev->tl_hash);
1471                 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1472                 mdev->tl_hash = new_tl_hash;
1473         }
1474
1475         if (new_ee_hash) {
1476                 kfree(mdev->ee_hash);
1477                 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1478                 mdev->ee_hash = new_ee_hash;
1479         }
1480
1481         crypto_free_hash(mdev->cram_hmac_tfm);
1482         mdev->cram_hmac_tfm = tfm;
1483
1484         crypto_free_hash(mdev->integrity_w_tfm);
1485         mdev->integrity_w_tfm = integrity_w_tfm;
1486
1487         crypto_free_hash(mdev->integrity_r_tfm);
1488         mdev->integrity_r_tfm = integrity_r_tfm;
1489
1490         kfree(mdev->int_dig_out);
1491         kfree(mdev->int_dig_in);
1492         kfree(mdev->int_dig_vv);
1493         mdev->int_dig_out=int_dig_out;
1494         mdev->int_dig_in=int_dig_in;
1495         mdev->int_dig_vv=int_dig_vv;
1496         retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL);
1497         spin_unlock_irq(&mdev->req_lock);
1498
1499         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1500         reply->ret_code = retcode;
1501         drbd_reconfig_done(mdev);
1502         return 0;
1503
1504 fail:
1505         kfree(int_dig_out);
1506         kfree(int_dig_in);
1507         kfree(int_dig_vv);
1508         crypto_free_hash(tfm);
1509         crypto_free_hash(integrity_w_tfm);
1510         crypto_free_hash(integrity_r_tfm);
1511         kfree(new_tl_hash);
1512         kfree(new_ee_hash);
1513         kfree(new_conf);
1514
1515         reply->ret_code = retcode;
1516         drbd_reconfig_done(mdev);
1517         return 0;
1518 }
1519
1520 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1521                               struct drbd_nl_cfg_reply *reply)
1522 {
1523         int retcode;
1524
1525         retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1526
1527         if (retcode == SS_NOTHING_TO_DO)
1528                 goto done;
1529         else if (retcode == SS_ALREADY_STANDALONE)
1530                 goto done;
1531         else if (retcode == SS_PRIMARY_NOP) {
1532                 /* Our statche checking code wants to see the peer outdated. */
1533                 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1534                                                       pdsk, D_OUTDATED));
1535         } else if (retcode == SS_CW_FAILED_BY_PEER) {
1536                 /* The peer probably wants to see us outdated. */
1537                 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1538                                                         disk, D_OUTDATED),
1539                                               CS_ORDERED);
1540                 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1541                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1542                         retcode = SS_SUCCESS;
1543                 }
1544         }
1545
1546         if (retcode < SS_SUCCESS)
1547                 goto fail;
1548
1549         if (wait_event_interruptible(mdev->state_wait,
1550                                      mdev->state.conn != C_DISCONNECTING)) {
1551                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1552                    someone else might connect us in the mean time! */
1553                 retcode = ERR_INTR;
1554                 goto fail;
1555         }
1556
1557  done:
1558         retcode = NO_ERROR;
1559  fail:
1560         drbd_md_sync(mdev);
1561         reply->ret_code = retcode;
1562         return 0;
1563 }
1564
1565 void resync_after_online_grow(struct drbd_conf *mdev)
1566 {
1567         int iass; /* I am sync source */
1568
1569         dev_info(DEV, "Resync of new storage after online grow\n");
1570         if (mdev->state.role != mdev->state.peer)
1571                 iass = (mdev->state.role == R_PRIMARY);
1572         else
1573                 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1574
1575         if (iass)
1576                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1577         else
1578                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1579 }
1580
1581 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1582                           struct drbd_nl_cfg_reply *reply)
1583 {
1584         struct resize rs;
1585         int retcode = NO_ERROR;
1586         enum determine_dev_size dd;
1587         enum dds_flags ddsf;
1588
1589         memset(&rs, 0, sizeof(struct resize));
1590         if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1591                 retcode = ERR_MANDATORY_TAG;
1592                 goto fail;
1593         }
1594
1595         if (mdev->state.conn > C_CONNECTED) {
1596                 retcode = ERR_RESIZE_RESYNC;
1597                 goto fail;
1598         }
1599
1600         if (mdev->state.role == R_SECONDARY &&
1601             mdev->state.peer == R_SECONDARY) {
1602                 retcode = ERR_NO_PRIMARY;
1603                 goto fail;
1604         }
1605
1606         if (!get_ldev(mdev)) {
1607                 retcode = ERR_NO_DISK;
1608                 goto fail;
1609         }
1610
1611         if (rs.no_resync && mdev->agreed_pro_version < 93) {
1612                 retcode = ERR_NEED_APV_93;
1613                 goto fail;
1614         }
1615
1616         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1617                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1618
1619         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1620         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1621         dd = drbd_determin_dev_size(mdev, ddsf);
1622         drbd_md_sync(mdev);
1623         put_ldev(mdev);
1624         if (dd == dev_size_error) {
1625                 retcode = ERR_NOMEM_BITMAP;
1626                 goto fail;
1627         }
1628
1629         if (mdev->state.conn == C_CONNECTED) {
1630                 if (dd == grew)
1631                         set_bit(RESIZE_PENDING, &mdev->flags);
1632
1633                 drbd_send_uuids(mdev);
1634                 drbd_send_sizes(mdev, 1, ddsf);
1635         }
1636
1637  fail:
1638         reply->ret_code = retcode;
1639         return 0;
1640 }
1641
1642 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1643                                struct drbd_nl_cfg_reply *reply)
1644 {
1645         int retcode = NO_ERROR;
1646         int err;
1647         int ovr; /* online verify running */
1648         int rsr; /* re-sync running */
1649         struct crypto_hash *verify_tfm = NULL;
1650         struct crypto_hash *csums_tfm = NULL;
1651         struct syncer_conf sc;
1652         cpumask_var_t new_cpu_mask;
1653         int *rs_plan_s = NULL;
1654         int fifo_size;
1655
1656         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1657                 retcode = ERR_NOMEM;
1658                 goto fail;
1659         }
1660
1661         if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1662                 memset(&sc, 0, sizeof(struct syncer_conf));
1663                 sc.rate       = DRBD_RATE_DEF;
1664                 sc.after      = DRBD_AFTER_DEF;
1665                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1666                 sc.on_no_data  = DRBD_ON_NO_DATA_DEF;
1667                 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1668                 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1669                 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1670                 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1671                 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1672         } else
1673                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1674
1675         if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1676                 retcode = ERR_MANDATORY_TAG;
1677                 goto fail;
1678         }
1679
1680         /* re-sync running */
1681         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1682                 mdev->state.conn == C_SYNC_TARGET ||
1683                 mdev->state.conn == C_PAUSED_SYNC_S ||
1684                 mdev->state.conn == C_PAUSED_SYNC_T );
1685
1686         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1687                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1688                 goto fail;
1689         }
1690
1691         if (!rsr && sc.csums_alg[0]) {
1692                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1693                 if (IS_ERR(csums_tfm)) {
1694                         csums_tfm = NULL;
1695                         retcode = ERR_CSUMS_ALG;
1696                         goto fail;
1697                 }
1698
1699                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1700                         retcode = ERR_CSUMS_ALG_ND;
1701                         goto fail;
1702                 }
1703         }
1704
1705         /* online verify running */
1706         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1707
1708         if (ovr) {
1709                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1710                         retcode = ERR_VERIFY_RUNNING;
1711                         goto fail;
1712                 }
1713         }
1714
1715         if (!ovr && sc.verify_alg[0]) {
1716                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1717                 if (IS_ERR(verify_tfm)) {
1718                         verify_tfm = NULL;
1719                         retcode = ERR_VERIFY_ALG;
1720                         goto fail;
1721                 }
1722
1723                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1724                         retcode = ERR_VERIFY_ALG_ND;
1725                         goto fail;
1726                 }
1727         }
1728
1729         /* silently ignore cpu mask on UP kernel */
1730         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1731                 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1732                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1733                 if (err) {
1734                         dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1735                         retcode = ERR_CPU_MASK_PARSE;
1736                         goto fail;
1737                 }
1738         }
1739
1740         ERR_IF (sc.rate < 1) sc.rate = 1;
1741         ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1742 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1743         if (sc.al_extents > AL_MAX) {
1744                 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1745                 sc.al_extents = AL_MAX;
1746         }
1747 #undef AL_MAX
1748
1749         /* to avoid spurious errors when configuring minors before configuring
1750          * the minors they depend on: if necessary, first create the minor we
1751          * depend on */
1752         if (sc.after >= 0)
1753                 ensure_mdev(sc.after, 1);
1754
1755         /* most sanity checks done, try to assign the new sync-after
1756          * dependency.  need to hold the global lock in there,
1757          * to avoid a race in the dependency loop check. */
1758         retcode = drbd_alter_sa(mdev, sc.after);
1759         if (retcode != NO_ERROR)
1760                 goto fail;
1761
1762         fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1763         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1764                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1765                 if (!rs_plan_s) {
1766                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1767                         retcode = ERR_NOMEM;
1768                         goto fail;
1769                 }
1770         }
1771
1772         /* ok, assign the rest of it as well.
1773          * lock against receive_SyncParam() */
1774         spin_lock(&mdev->peer_seq_lock);
1775         mdev->sync_conf = sc;
1776
1777         if (!rsr) {
1778                 crypto_free_hash(mdev->csums_tfm);
1779                 mdev->csums_tfm = csums_tfm;
1780                 csums_tfm = NULL;
1781         }
1782
1783         if (!ovr) {
1784                 crypto_free_hash(mdev->verify_tfm);
1785                 mdev->verify_tfm = verify_tfm;
1786                 verify_tfm = NULL;
1787         }
1788
1789         if (fifo_size != mdev->rs_plan_s.size) {
1790                 kfree(mdev->rs_plan_s.values);
1791                 mdev->rs_plan_s.values = rs_plan_s;
1792                 mdev->rs_plan_s.size   = fifo_size;
1793                 mdev->rs_planed = 0;
1794                 rs_plan_s = NULL;
1795         }
1796
1797         spin_unlock(&mdev->peer_seq_lock);
1798
1799         if (get_ldev(mdev)) {
1800                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1801                 drbd_al_shrink(mdev);
1802                 err = drbd_check_al_size(mdev);
1803                 lc_unlock(mdev->act_log);
1804                 wake_up(&mdev->al_wait);
1805
1806                 put_ldev(mdev);
1807                 drbd_md_sync(mdev);
1808
1809                 if (err) {
1810                         retcode = ERR_NOMEM;
1811                         goto fail;
1812                 }
1813         }
1814
1815         if (mdev->state.conn >= C_CONNECTED)
1816                 drbd_send_sync_param(mdev, &sc);
1817
1818         if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1819                 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1820                 drbd_calc_cpu_mask(mdev);
1821                 mdev->receiver.reset_cpu_mask = 1;
1822                 mdev->asender.reset_cpu_mask = 1;
1823                 mdev->worker.reset_cpu_mask = 1;
1824         }
1825
1826         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1827 fail:
1828         kfree(rs_plan_s);
1829         free_cpumask_var(new_cpu_mask);
1830         crypto_free_hash(csums_tfm);
1831         crypto_free_hash(verify_tfm);
1832         reply->ret_code = retcode;
1833         return 0;
1834 }
1835
1836 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1837                               struct drbd_nl_cfg_reply *reply)
1838 {
1839         int retcode;
1840
1841         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1842
1843         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1844                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1845
1846         while (retcode == SS_NEED_CONNECTION) {
1847                 spin_lock_irq(&mdev->req_lock);
1848                 if (mdev->state.conn < C_CONNECTED)
1849                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1850                 spin_unlock_irq(&mdev->req_lock);
1851
1852                 if (retcode != SS_NEED_CONNECTION)
1853                         break;
1854
1855                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1856         }
1857
1858         reply->ret_code = retcode;
1859         return 0;
1860 }
1861
1862 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
1863 {
1864         int rv;
1865
1866         rv = drbd_bmio_set_n_write(mdev);
1867         drbd_suspend_al(mdev);
1868         return rv;
1869 }
1870
1871 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1872                                    struct drbd_nl_cfg_reply *reply)
1873 {
1874         int retcode;
1875
1876         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
1877
1878         if (retcode < SS_SUCCESS) {
1879                 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
1880                         /* The peer will get a resync upon connect anyways. Just make that
1881                            into a full resync. */
1882                         retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
1883                         if (retcode >= SS_SUCCESS) {
1884                                 /* open coded drbd_bitmap_io() */
1885                                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
1886                                                    "set_n_write from invalidate_peer"))
1887                                         retcode = ERR_IO_MD_DISK;
1888                         }
1889                 } else
1890                         retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
1891         }
1892
1893         reply->ret_code = retcode;
1894         return 0;
1895 }
1896
1897 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1898                               struct drbd_nl_cfg_reply *reply)
1899 {
1900         int retcode = NO_ERROR;
1901
1902         if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
1903                 retcode = ERR_PAUSE_IS_SET;
1904
1905         reply->ret_code = retcode;
1906         return 0;
1907 }
1908
1909 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1910                                struct drbd_nl_cfg_reply *reply)
1911 {
1912         int retcode = NO_ERROR;
1913
1914         if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO)
1915                 retcode = ERR_PAUSE_IS_CLEAR;
1916
1917         reply->ret_code = retcode;
1918         return 0;
1919 }
1920
1921 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1922                               struct drbd_nl_cfg_reply *reply)
1923 {
1924         reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
1925
1926         return 0;
1927 }
1928
1929 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1930                              struct drbd_nl_cfg_reply *reply)
1931 {
1932         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
1933                 drbd_uuid_new_current(mdev);
1934                 clear_bit(NEW_CUR_UUID, &mdev->flags);
1935                 drbd_md_sync(mdev);
1936         }
1937         drbd_suspend_io(mdev);
1938         reply->ret_code = drbd_request_state(mdev, NS(susp, 0));
1939         if (reply->ret_code == SS_SUCCESS) {
1940                 if (mdev->state.conn < C_CONNECTED)
1941                         tl_clear(mdev);
1942                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
1943                         tl_restart(mdev, fail_frozen_disk_io);
1944         }
1945         drbd_resume_io(mdev);
1946
1947         return 0;
1948 }
1949
1950 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1951                            struct drbd_nl_cfg_reply *reply)
1952 {
1953         reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
1954         return 0;
1955 }
1956
1957 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1958                            struct drbd_nl_cfg_reply *reply)
1959 {
1960         unsigned short *tl;
1961
1962         tl = reply->tag_list;
1963
1964         if (get_ldev(mdev)) {
1965                 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
1966                 put_ldev(mdev);
1967         }
1968
1969         if (get_net_conf(mdev)) {
1970                 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
1971                 put_net_conf(mdev);
1972         }
1973         tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
1974
1975         put_unaligned(TT_END, tl++); /* Close the tag list */
1976
1977         return (int)((char *)tl - (char *)reply->tag_list);
1978 }
1979
1980 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1981                              struct drbd_nl_cfg_reply *reply)
1982 {
1983         unsigned short *tl = reply->tag_list;
1984         union drbd_state s = mdev->state;
1985         unsigned long rs_left;
1986         unsigned int res;
1987
1988         tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
1989
1990         /* no local ref, no bitmap, no syncer progress. */
1991         if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
1992                 if (get_ldev(mdev)) {
1993                         drbd_get_syncer_progress(mdev, &rs_left, &res);
1994                         tl = tl_add_int(tl, T_sync_progress, &res);
1995                         put_ldev(mdev);
1996                 }
1997         }
1998         put_unaligned(TT_END, tl++); /* Close the tag list */
1999
2000         return (int)((char *)tl - (char *)reply->tag_list);
2001 }
2002
2003 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2004                              struct drbd_nl_cfg_reply *reply)
2005 {
2006         unsigned short *tl;
2007
2008         tl = reply->tag_list;
2009
2010         if (get_ldev(mdev)) {
2011                 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
2012                 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
2013                 put_ldev(mdev);
2014         }
2015         put_unaligned(TT_END, tl++); /* Close the tag list */
2016
2017         return (int)((char *)tl - (char *)reply->tag_list);
2018 }
2019
2020 /**
2021  * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
2022  * @mdev:       DRBD device.
2023  * @nlp:        Netlink/connector packet from drbdsetup
2024  * @reply:      Reply packet for drbdsetup
2025  */
2026 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2027                                     struct drbd_nl_cfg_reply *reply)
2028 {
2029         unsigned short *tl;
2030         char rv;
2031
2032         tl = reply->tag_list;
2033
2034         rv = mdev->state.pdsk == D_OUTDATED        ? UT_PEER_OUTDATED :
2035           test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
2036
2037         tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
2038         put_unaligned(TT_END, tl++); /* Close the tag list */
2039
2040         return (int)((char *)tl - (char *)reply->tag_list);
2041 }
2042
2043 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2044                                     struct drbd_nl_cfg_reply *reply)
2045 {
2046         /* default to resume from last known position, if possible */
2047         struct start_ov args =
2048                 { .start_sector = mdev->ov_start_sector };
2049
2050         if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2051                 reply->ret_code = ERR_MANDATORY_TAG;
2052                 return 0;
2053         }
2054         /* w_make_ov_request expects position to be aligned */
2055         mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
2056         reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2057         return 0;
2058 }
2059
2060
2061 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2062                               struct drbd_nl_cfg_reply *reply)
2063 {
2064         int retcode = NO_ERROR;
2065         int skip_initial_sync = 0;
2066         int err;
2067
2068         struct new_c_uuid args;
2069
2070         memset(&args, 0, sizeof(struct new_c_uuid));
2071         if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2072                 reply->ret_code = ERR_MANDATORY_TAG;
2073                 return 0;
2074         }
2075
2076         mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
2077
2078         if (!get_ldev(mdev)) {
2079                 retcode = ERR_NO_DISK;
2080                 goto out;
2081         }
2082
2083         /* this is "skip initial sync", assume to be clean */
2084         if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
2085             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2086                 dev_info(DEV, "Preparing to skip initial sync\n");
2087                 skip_initial_sync = 1;
2088         } else if (mdev->state.conn != C_STANDALONE) {
2089                 retcode = ERR_CONNECTED;
2090                 goto out_dec;
2091         }
2092
2093         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2094         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2095
2096         if (args.clear_bm) {
2097                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, "clear_n_write from new_c_uuid");
2098                 if (err) {
2099                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2100                         retcode = ERR_IO_MD_DISK;
2101                 }
2102                 if (skip_initial_sync) {
2103                         drbd_send_uuids_skip_initial_sync(mdev);
2104                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2105                         spin_lock_irq(&mdev->req_lock);
2106                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2107                                         CS_VERBOSE, NULL);
2108                         spin_unlock_irq(&mdev->req_lock);
2109                 }
2110         }
2111
2112         drbd_md_sync(mdev);
2113 out_dec:
2114         put_ldev(mdev);
2115 out:
2116         mutex_unlock(&mdev->state_mutex);
2117
2118         reply->ret_code = retcode;
2119         return 0;
2120 }
2121
2122 struct cn_handler_struct {
2123         int (*function)(struct drbd_conf *,
2124                          struct drbd_nl_cfg_req *,
2125                          struct drbd_nl_cfg_reply *);
2126         int reply_body_size;
2127 };
2128
2129 static struct cn_handler_struct cnd_table[] = {
2130         [ P_primary ]           = { &drbd_nl_primary,           0 },
2131         [ P_secondary ]         = { &drbd_nl_secondary,         0 },
2132         [ P_disk_conf ]         = { &drbd_nl_disk_conf,         0 },
2133         [ P_detach ]            = { &drbd_nl_detach,            0 },
2134         [ P_net_conf ]          = { &drbd_nl_net_conf,          0 },
2135         [ P_disconnect ]        = { &drbd_nl_disconnect,        0 },
2136         [ P_resize ]            = { &drbd_nl_resize,            0 },
2137         [ P_syncer_conf ]       = { &drbd_nl_syncer_conf,       0 },
2138         [ P_invalidate ]        = { &drbd_nl_invalidate,        0 },
2139         [ P_invalidate_peer ]   = { &drbd_nl_invalidate_peer,   0 },
2140         [ P_pause_sync ]        = { &drbd_nl_pause_sync,        0 },
2141         [ P_resume_sync ]       = { &drbd_nl_resume_sync,       0 },
2142         [ P_suspend_io ]        = { &drbd_nl_suspend_io,        0 },
2143         [ P_resume_io ]         = { &drbd_nl_resume_io,         0 },
2144         [ P_outdate ]           = { &drbd_nl_outdate,           0 },
2145         [ P_get_config ]        = { &drbd_nl_get_config,
2146                                     sizeof(struct syncer_conf_tag_len_struct) +
2147                                     sizeof(struct disk_conf_tag_len_struct) +
2148                                     sizeof(struct net_conf_tag_len_struct) },
2149         [ P_get_state ]         = { &drbd_nl_get_state,
2150                                     sizeof(struct get_state_tag_len_struct) +
2151                                     sizeof(struct sync_progress_tag_len_struct) },
2152         [ P_get_uuids ]         = { &drbd_nl_get_uuids,
2153                                     sizeof(struct get_uuids_tag_len_struct) },
2154         [ P_get_timeout_flag ]  = { &drbd_nl_get_timeout_flag,
2155                                     sizeof(struct get_timeout_flag_tag_len_struct)},
2156         [ P_start_ov ]          = { &drbd_nl_start_ov,          0 },
2157         [ P_new_c_uuid ]        = { &drbd_nl_new_c_uuid,        0 },
2158 };
2159
2160 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2161 {
2162         struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2163         struct cn_handler_struct *cm;
2164         struct cn_msg *cn_reply;
2165         struct drbd_nl_cfg_reply *reply;
2166         struct drbd_conf *mdev;
2167         int retcode, rr;
2168         int reply_size = sizeof(struct cn_msg)
2169                 + sizeof(struct drbd_nl_cfg_reply)
2170                 + sizeof(short int);
2171
2172         if (!try_module_get(THIS_MODULE)) {
2173                 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2174                 return;
2175         }
2176
2177         if (!cap_raised(nsp->eff_cap, CAP_SYS_ADMIN)) {
2178                 retcode = ERR_PERM;
2179                 goto fail;
2180         }
2181
2182         mdev = ensure_mdev(nlp->drbd_minor,
2183                         (nlp->flags & DRBD_NL_CREATE_DEVICE));
2184         if (!mdev) {
2185                 retcode = ERR_MINOR_INVALID;
2186                 goto fail;
2187         }
2188
2189         if (nlp->packet_type >= P_nl_after_last_packet) {
2190                 retcode = ERR_PACKET_NR;
2191                 goto fail;
2192         }
2193
2194         cm = cnd_table + nlp->packet_type;
2195
2196         /* This may happen if packet number is 0: */
2197         if (cm->function == NULL) {
2198                 retcode = ERR_PACKET_NR;
2199                 goto fail;
2200         }
2201
2202         reply_size += cm->reply_body_size;
2203
2204         /* allocation not in the IO path, cqueue thread context */
2205         cn_reply = kmalloc(reply_size, GFP_KERNEL);
2206         if (!cn_reply) {
2207                 retcode = ERR_NOMEM;
2208                 goto fail;
2209         }
2210         reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2211
2212         reply->packet_type =
2213                 cm->reply_body_size ? nlp->packet_type : P_nl_after_last_packet;
2214         reply->minor = nlp->drbd_minor;
2215         reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2216         /* reply->tag_list; might be modified by cm->function. */
2217
2218         rr = cm->function(mdev, nlp, reply);
2219
2220         cn_reply->id = req->id;
2221         cn_reply->seq = req->seq;
2222         cn_reply->ack = req->ack  + 1;
2223         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2224         cn_reply->flags = 0;
2225
2226         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2227         if (rr && rr != -ESRCH)
2228                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2229
2230         kfree(cn_reply);
2231         module_put(THIS_MODULE);
2232         return;
2233  fail:
2234         drbd_nl_send_reply(req, retcode);
2235         module_put(THIS_MODULE);
2236 }
2237
2238 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2239
2240 static unsigned short *
2241 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2242         unsigned short len, int nul_terminated)
2243 {
2244         unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2245         len = (len < l) ? len :  l;
2246         put_unaligned(tag, tl++);
2247         put_unaligned(len, tl++);
2248         memcpy(tl, data, len);
2249         tl = (unsigned short*)((char*)tl + len);
2250         if (nul_terminated)
2251                 *((char*)tl - 1) = 0;
2252         return tl;
2253 }
2254
2255 static unsigned short *
2256 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2257 {
2258         return __tl_add_blob(tl, tag, data, len, 0);
2259 }
2260
2261 static unsigned short *
2262 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2263 {
2264         return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2265 }
2266
2267 static unsigned short *
2268 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2269 {
2270         put_unaligned(tag, tl++);
2271         switch(tag_type(tag)) {
2272         case TT_INTEGER:
2273                 put_unaligned(sizeof(int), tl++);
2274                 put_unaligned(*(int *)val, (int *)tl);
2275                 tl = (unsigned short*)((char*)tl+sizeof(int));
2276                 break;
2277         case TT_INT64:
2278                 put_unaligned(sizeof(u64), tl++);
2279                 put_unaligned(*(u64 *)val, (u64 *)tl);
2280                 tl = (unsigned short*)((char*)tl+sizeof(u64));
2281                 break;
2282         default:
2283                 /* someone did something stupid. */
2284                 ;
2285         }
2286         return tl;
2287 }
2288
2289 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2290 {
2291         char buffer[sizeof(struct cn_msg)+
2292                     sizeof(struct drbd_nl_cfg_reply)+
2293                     sizeof(struct get_state_tag_len_struct)+
2294                     sizeof(short int)];
2295         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2296         struct drbd_nl_cfg_reply *reply =
2297                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2298         unsigned short *tl = reply->tag_list;
2299
2300         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2301
2302         tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2303
2304         put_unaligned(TT_END, tl++); /* Close the tag list */
2305
2306         cn_reply->id.idx = CN_IDX_DRBD;
2307         cn_reply->id.val = CN_VAL_DRBD;
2308
2309         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2310         cn_reply->ack = 0; /* not used here. */
2311         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2312                 (int)((char *)tl - (char *)reply->tag_list);
2313         cn_reply->flags = 0;
2314
2315         reply->packet_type = P_get_state;
2316         reply->minor = mdev_to_minor(mdev);
2317         reply->ret_code = NO_ERROR;
2318
2319         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2320 }
2321
2322 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2323 {
2324         char buffer[sizeof(struct cn_msg)+
2325                     sizeof(struct drbd_nl_cfg_reply)+
2326                     sizeof(struct call_helper_tag_len_struct)+
2327                     sizeof(short int)];
2328         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2329         struct drbd_nl_cfg_reply *reply =
2330                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2331         unsigned short *tl = reply->tag_list;
2332
2333         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2334
2335         tl = tl_add_str(tl, T_helper, helper_name);
2336         put_unaligned(TT_END, tl++); /* Close the tag list */
2337
2338         cn_reply->id.idx = CN_IDX_DRBD;
2339         cn_reply->id.val = CN_VAL_DRBD;
2340
2341         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2342         cn_reply->ack = 0; /* not used here. */
2343         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2344                 (int)((char *)tl - (char *)reply->tag_list);
2345         cn_reply->flags = 0;
2346
2347         reply->packet_type = P_call_helper;
2348         reply->minor = mdev_to_minor(mdev);
2349         reply->ret_code = NO_ERROR;
2350
2351         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2352 }
2353
2354 void drbd_bcast_ee(struct drbd_conf *mdev,
2355                 const char *reason, const int dgs,
2356                 const char* seen_hash, const char* calc_hash,
2357                 const struct drbd_epoch_entry* e)
2358 {
2359         struct cn_msg *cn_reply;
2360         struct drbd_nl_cfg_reply *reply;
2361         unsigned short *tl;
2362         struct page *page;
2363         unsigned len;
2364
2365         if (!e)
2366                 return;
2367         if (!reason || !reason[0])
2368                 return;
2369
2370         /* apparently we have to memcpy twice, first to prepare the data for the
2371          * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2372          * netlink skb. */
2373         /* receiver thread context, which is not in the writeout path (of this node),
2374          * but may be in the writeout path of the _other_ node.
2375          * GFP_NOIO to avoid potential "distributed deadlock". */
2376         cn_reply = kmalloc(
2377                 sizeof(struct cn_msg)+
2378                 sizeof(struct drbd_nl_cfg_reply)+
2379                 sizeof(struct dump_ee_tag_len_struct)+
2380                 sizeof(short int),
2381                 GFP_NOIO);
2382
2383         if (!cn_reply) {
2384                 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2385                                 (unsigned long long)e->sector, e->size);
2386                 return;
2387         }
2388
2389         reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2390         tl = reply->tag_list;
2391
2392         tl = tl_add_str(tl, T_dump_ee_reason, reason);
2393         tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2394         tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2395         tl = tl_add_int(tl, T_ee_sector, &e->sector);
2396         tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2397
2398         put_unaligned(T_ee_data, tl++);
2399         put_unaligned(e->size, tl++);
2400
2401         len = e->size;
2402         page = e->pages;
2403         page_chain_for_each(page) {
2404                 void *d = kmap_atomic(page, KM_USER0);
2405                 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2406                 memcpy(tl, d, l);
2407                 kunmap_atomic(d, KM_USER0);
2408                 tl = (unsigned short*)((char*)tl + l);
2409                 len -= l;
2410         }
2411         put_unaligned(TT_END, tl++); /* Close the tag list */
2412
2413         cn_reply->id.idx = CN_IDX_DRBD;
2414         cn_reply->id.val = CN_VAL_DRBD;
2415
2416         cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2417         cn_reply->ack = 0; // not used here.
2418         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2419                 (int)((char*)tl - (char*)reply->tag_list);
2420         cn_reply->flags = 0;
2421
2422         reply->packet_type = P_dump_ee;
2423         reply->minor = mdev_to_minor(mdev);
2424         reply->ret_code = NO_ERROR;
2425
2426         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2427         kfree(cn_reply);
2428 }
2429
2430 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2431 {
2432         char buffer[sizeof(struct cn_msg)+
2433                     sizeof(struct drbd_nl_cfg_reply)+
2434                     sizeof(struct sync_progress_tag_len_struct)+
2435                     sizeof(short int)];
2436         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2437         struct drbd_nl_cfg_reply *reply =
2438                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2439         unsigned short *tl = reply->tag_list;
2440         unsigned long rs_left;
2441         unsigned int res;
2442
2443         /* no local ref, no bitmap, no syncer progress, no broadcast. */
2444         if (!get_ldev(mdev))
2445                 return;
2446         drbd_get_syncer_progress(mdev, &rs_left, &res);
2447         put_ldev(mdev);
2448
2449         tl = tl_add_int(tl, T_sync_progress, &res);
2450         put_unaligned(TT_END, tl++); /* Close the tag list */
2451
2452         cn_reply->id.idx = CN_IDX_DRBD;
2453         cn_reply->id.val = CN_VAL_DRBD;
2454
2455         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2456         cn_reply->ack = 0; /* not used here. */
2457         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2458                 (int)((char *)tl - (char *)reply->tag_list);
2459         cn_reply->flags = 0;
2460
2461         reply->packet_type = P_sync_progress;
2462         reply->minor = mdev_to_minor(mdev);
2463         reply->ret_code = NO_ERROR;
2464
2465         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2466 }
2467
2468 int __init drbd_nl_init(void)
2469 {
2470         static struct cb_id cn_id_drbd;
2471         int err, try=10;
2472
2473         cn_id_drbd.val = CN_VAL_DRBD;
2474         do {
2475                 cn_id_drbd.idx = cn_idx;
2476                 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2477                 if (!err)
2478                         break;
2479                 cn_idx = (cn_idx + CN_IDX_STEP);
2480         } while (try--);
2481
2482         if (err) {
2483                 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2484                 return err;
2485         }
2486
2487         return 0;
2488 }
2489
2490 void drbd_nl_cleanup(void)
2491 {
2492         static struct cb_id cn_id_drbd;
2493
2494         cn_id_drbd.idx = cn_idx;
2495         cn_id_drbd.val = CN_VAL_DRBD;
2496
2497         cn_del_callback(&cn_id_drbd);
2498 }
2499
2500 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2501 {
2502         char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2503         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2504         struct drbd_nl_cfg_reply *reply =
2505                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2506         int rr;
2507
2508         cn_reply->id = req->id;
2509
2510         cn_reply->seq = req->seq;
2511         cn_reply->ack = req->ack  + 1;
2512         cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2513         cn_reply->flags = 0;
2514
2515         reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2516         reply->ret_code = ret_code;
2517
2518         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2519         if (rr && rr != -ESRCH)
2520                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2521 }
2522