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/blkpg.h>
  33 #include <linux/cpumask.h>
  34 #include "drbd_int.h"
  35 #include "drbd_req.h"
  36 #include "drbd_wrappers.h"
  37 #include <asm/unaligned.h>
  38 #include <linux/drbd_limits.h>
  39 #include <linux/kthread.h>
  40
  41 #include <net/genetlink.h>
  42
  43 /* .doit */
  44 // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info);
  45 // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info);
  46
  47 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info);
  48 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info);
  49
  50 int drbd_adm_create_connection(struct sk_buff *skb, struct genl_info *info);
  51 int drbd_adm_delete_connection(struct sk_buff *skb, struct genl_info *info);
  52 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info);
  53
  54 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info);
  55 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info);
  56 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info);
  57 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info);
  58 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info);
  59 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info);
  60 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info);
  61 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info);
  62 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info);
  63 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info);
  64 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info);
  65 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info);
  66 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info);
  67 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info);
  68 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info);
  69 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info);
  70 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info);
  71 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info);
  72 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info);
  73 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info);
  74 /* .dumpit */
  75 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb);
  76
  77 #include <linux/drbd_genl_api.h>
  78 #include <linux/genl_magic_func.h>
  79
  80 /* used blkdev_get_by_path, to claim our meta data device(s) */
  81 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
  82
  83 /* Configuration is strictly serialized, because generic netlink message
  84  * processing is strictly serialized by the genl_lock().
  85  * Which means we can use one static global drbd_config_context struct.
  86  */
  87 static struct drbd_config_context {
  88         /* assigned from drbd_genlmsghdr */
  89         unsigned int minor;
  90         /* assigned from request attributes, if present */
  91         unsigned int volume;
  92 #define VOLUME_UNSPECIFIED              (-1U)
  93         /* pointer into the request skb,
  94          * limited lifetime! */
  95         char *conn_name;
  96
  97         /* reply buffer */
  98         struct sk_buff *reply_skb;
  99         /* pointer into reply buffer */
 100         struct drbd_genlmsghdr *reply_dh;
 101         /* resolved from attributes, if possible */
 102         struct drbd_conf *mdev;
 103         struct drbd_tconn *tconn;
 104 } adm_ctx;
 105
 106 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
 107 {
 108         genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
 109         if (genlmsg_reply(skb, info))
 110                 printk(KERN_ERR "drbd: error sending genl reply\n");
 111 }
 112
 113 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
 114  * reason it could fail was no space in skb, and there are 4k available. */
 115 int drbd_msg_put_info(const char *info)
 116 {
 117         struct sk_buff *skb = adm_ctx.reply_skb;
 118         struct nlattr *nla;
 119         int err = -EMSGSIZE;
 120
 121         if (!info || !info[0])
 122                 return 0;
 123
 124         nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY);
 125         if (!nla)
 126                 return err;
 127
 128         err = nla_put_string(skb, T_info_text, info);
 129         if (err) {
 130                 nla_nest_cancel(skb, nla);
 131                 return err;
 132         } else
 133                 nla_nest_end(skb, nla);
 134         return 0;
 135 }
 136
 137 /* This would be a good candidate for a "pre_doit" hook,
 138  * and per-family private info->pointers.
 139  * But we need to stay compatible with older kernels.
 140  * If it returns successfully, adm_ctx members are valid.
 141  */
 142 #define DRBD_ADM_NEED_MINOR     1
 143 #define DRBD_ADM_NEED_CONN      2
 144 static int drbd_adm_prepare(struct sk_buff *skb, struct genl_info *info,
 145                 unsigned flags)
 146 {
 147         struct drbd_genlmsghdr *d_in = info->userhdr;
 148         const u8 cmd = info->genlhdr->cmd;
 149         int err;
 150
 151         memset(&adm_ctx, 0, sizeof(adm_ctx));
 152
 153         /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
 154         if (cmd != DRBD_ADM_GET_STATUS
 155         && security_netlink_recv(skb, CAP_SYS_ADMIN))
 156                return -EPERM;
 157
 158         adm_ctx.reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
 159         if (!adm_ctx.reply_skb)
 160                 goto fail;
 161
 162         adm_ctx.reply_dh = genlmsg_put_reply(adm_ctx.reply_skb,
 163                                         info, &drbd_genl_family, 0, cmd);
 164         /* put of a few bytes into a fresh skb of >= 4k will always succeed.
 165          * but anyways */
 166         if (!adm_ctx.reply_dh)
 167                 goto fail;
 168
 169         adm_ctx.reply_dh->minor = d_in->minor;
 170         adm_ctx.reply_dh->ret_code = NO_ERROR;
 171
 172         if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
 173                 struct nlattr *nla;
 174                 /* parse and validate only */
 175                 err = drbd_cfg_context_from_attrs(NULL, info);
 176                 if (err)
 177                         goto fail;
 178
 179                 /* It was present, and valid,
 180                  * copy it over to the reply skb. */
 181                 err = nla_put_nohdr(adm_ctx.reply_skb,
 182                                 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
 183                                 info->attrs[DRBD_NLA_CFG_CONTEXT]);
 184                 if (err)
 185                         goto fail;
 186
 187                 /* and assign stuff to the global adm_ctx */
 188                 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
 189                 adm_ctx.volume = nla ? nla_get_u32(nla) : VOLUME_UNSPECIFIED;
 190                 nla = nested_attr_tb[__nla_type(T_ctx_conn_name)];
 191                 if (nla)
 192                         adm_ctx.conn_name = nla_data(nla);
 193         } else
 194                 adm_ctx.volume = VOLUME_UNSPECIFIED;
 195
 196         adm_ctx.minor = d_in->minor;
 197         adm_ctx.mdev = minor_to_mdev(d_in->minor);
 198         adm_ctx.tconn = conn_get_by_name(adm_ctx.conn_name);
 199
 200         if (!adm_ctx.mdev && (flags & DRBD_ADM_NEED_MINOR)) {
 201                 drbd_msg_put_info("unknown minor");
 202                 return ERR_MINOR_INVALID;
 203         }
 204         if (!adm_ctx.tconn && (flags & DRBD_ADM_NEED_CONN)) {
 205                 drbd_msg_put_info("unknown connection");
 206                 return ERR_INVALID_REQUEST;
 207         }
 208
 209         /* some more paranoia, if the request was over-determined */
 210         if (adm_ctx.mdev && adm_ctx.tconn &&
 211             adm_ctx.mdev->tconn != adm_ctx.tconn) {
 212                 pr_warning("request: minor=%u, conn=%s; but that minor belongs to connection %s\n",
 213                                 adm_ctx.minor, adm_ctx.conn_name, adm_ctx.mdev->tconn->name);
 214                 drbd_msg_put_info("minor exists in different connection");
 215                 return ERR_INVALID_REQUEST;
 216         }
 217         if (adm_ctx.mdev &&
 218             adm_ctx.volume != VOLUME_UNSPECIFIED &&
 219             adm_ctx.volume != adm_ctx.mdev->vnr) {
 220                 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
 221                                 adm_ctx.minor, adm_ctx.volume,
 222                                 adm_ctx.mdev->vnr, adm_ctx.mdev->tconn->name);
 223                 drbd_msg_put_info("minor exists as different volume");
 224                 return ERR_INVALID_REQUEST;
 225         }
 226
 227         return NO_ERROR;
 228
 229 fail:
 230         nlmsg_free(adm_ctx.reply_skb);
 231         adm_ctx.reply_skb = NULL;
 232         return -ENOMEM;
 233 }
 234
 235 static int drbd_adm_finish(struct genl_info *info, int retcode)
 236 {
 237         struct nlattr *nla;
 238         const char *conn_name = NULL;
 239
 240         if (adm_ctx.tconn) {
 241                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
 242                 adm_ctx.tconn = NULL;
 243         }
 244
 245         if (!adm_ctx.reply_skb)
 246                 return -ENOMEM;
 247
 248         adm_ctx.reply_dh->ret_code = retcode;
 249
 250         nla = info->attrs[DRBD_NLA_CFG_CONTEXT];
 251         if (nla) {
 252                 nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
 253                 if (nla)
 254                         conn_name = nla_data(nla);
 255         }
 256
 257         drbd_adm_send_reply(adm_ctx.reply_skb, info);
 258         return 0;
 259 }
 260
 261 static void setup_khelper_env(struct drbd_tconn *tconn, char **envp)
 262 {
 263         char *afs;
 264         struct net_conf *nc;
 265
 266         rcu_read_lock();
 267         nc = rcu_dereference(tconn->net_conf);
 268         if (nc) {
 269                 switch (((struct sockaddr *)nc->peer_addr)->sa_family) {
 270                 case AF_INET6:
 271                         afs = "ipv6";
 272                         snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
 273                                  &((struct sockaddr_in6 *)nc->peer_addr)->sin6_addr);
 274                         break;
 275                 case AF_INET:
 276                         afs = "ipv4";
 277                         snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
 278                                  &((struct sockaddr_in *)nc->peer_addr)->sin_addr);
 279                         break;
 280                 default:
 281                         afs = "ssocks";
 282                         snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
 283                                  &((struct sockaddr_in *)nc->peer_addr)->sin_addr);
 284                 }
 285                 snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
 286         }
 287         rcu_read_unlock();
 288 }
 289
 290 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
 291 {
 292         char *envp[] = { "HOME=/",
 293                         "TERM=linux",
 294                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 295                          (char[20]) { }, /* address family */
 296                          (char[60]) { }, /* address */
 297                         NULL };
 298         char mb[12];
 299         char *argv[] = {usermode_helper, cmd, mb, NULL };
 300         struct sib_info sib;
 301         int ret;
 302
 303         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 304         setup_khelper_env(mdev->tconn, envp);
 305
 306         /* The helper may take some time.
 307          * write out any unsynced meta data changes now */
 308         drbd_md_sync(mdev);
 309
 310         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 311         sib.sib_reason = SIB_HELPER_PRE;
 312         sib.helper_name = cmd;
 313         drbd_bcast_event(mdev, &sib);
 314         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 315         if (ret)
 316                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 317                                 usermode_helper, cmd, mb,
 318                                 (ret >> 8) & 0xff, ret);
 319         else
 320                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 321                                 usermode_helper, cmd, mb,
 322                                 (ret >> 8) & 0xff, ret);
 323         sib.sib_reason = SIB_HELPER_POST;
 324         sib.helper_exit_code = ret;
 325         drbd_bcast_event(mdev, &sib);
 326
 327         if (ret < 0) /* Ignore any ERRNOs we got. */
 328                 ret = 0;
 329
 330         return ret;
 331 }
 332
 333 static void conn_md_sync(struct drbd_tconn *tconn)
 334 {
 335         struct drbd_conf *mdev;
 336         int vnr;
 337
 338         down_read(&drbd_cfg_rwsem);
 339         idr_for_each_entry(&tconn->volumes, mdev, vnr)
 340                 drbd_md_sync(mdev);
 341         up_read(&drbd_cfg_rwsem);
 342 }
 343
 344 int conn_khelper(struct drbd_tconn *tconn, char *cmd)
 345 {
 346         char *envp[] = { "HOME=/",
 347                         "TERM=linux",
 348                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 349                          (char[20]) { }, /* address family */
 350                          (char[60]) { }, /* address */
 351                         NULL };
 352         char *argv[] = {usermode_helper, cmd, tconn->name, NULL };
 353         int ret;
 354
 355         setup_khelper_env(tconn, envp);
 356         conn_md_sync(tconn);
 357
 358         conn_info(tconn, "helper command: %s %s %s\n", usermode_helper, cmd, tconn->name);
 359         /* TODO: conn_bcast_event() ?? */
 360
 361         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 362         if (ret)
 363                 conn_warn(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
 364                           usermode_helper, cmd, tconn->name,
 365                           (ret >> 8) & 0xff, ret);
 366         else
 367                 conn_info(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
 368                           usermode_helper, cmd, tconn->name,
 369                           (ret >> 8) & 0xff, ret);
 370         /* TODO: conn_bcast_event() ?? */
 371
 372         if (ret < 0) /* Ignore any ERRNOs we got. */
 373                 ret = 0;
 374
 375         return ret;
 376 }
 377
 378 static enum drbd_fencing_p highest_fencing_policy(struct drbd_tconn *tconn)
 379 {
 380         enum drbd_fencing_p fp = FP_NOT_AVAIL;
 381         struct drbd_conf *mdev;
 382         int vnr;
 383
 384         rcu_read_lock();
 385         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
 386                 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 387                         fp = max_t(enum drbd_fencing_p, fp, mdev->ldev->dc.fencing);
 388                         put_ldev(mdev);
 389                 }
 390         }
 391         rcu_read_unlock();
 392
 393         return fp;
 394 }
 395
 396 bool conn_try_outdate_peer(struct drbd_tconn *tconn)
 397 {
 398         union drbd_state mask = { };
 399         union drbd_state val = { };
 400         enum drbd_fencing_p fp;
 401         char *ex_to_string;
 402         int r;
 403
 404         if (tconn->cstate >= C_WF_REPORT_PARAMS) {
 405                 conn_err(tconn, "Expected cstate < C_WF_REPORT_PARAMS\n");
 406                 return false;
 407         }
 408
 409         fp = highest_fencing_policy(tconn);
 410         switch (fp) {
 411         case FP_NOT_AVAIL:
 412                 conn_warn(tconn, "Not fencing peer, I'm not even Consistent myself.\n");
 413                 goto out;
 414         case FP_DONT_CARE:
 415                 return true;
 416         default: ;
 417         }
 418
 419         r = conn_khelper(tconn, "fence-peer");
 420
 421         switch ((r>>8) & 0xff) {
 422         case 3: /* peer is inconsistent */
 423                 ex_to_string = "peer is inconsistent or worse";
 424                 mask.pdsk = D_MASK;
 425                 val.pdsk = D_INCONSISTENT;
 426                 break;
 427         case 4: /* peer got outdated, or was already outdated */
 428                 ex_to_string = "peer was fenced";
 429                 mask.pdsk = D_MASK;
 430                 val.pdsk = D_OUTDATED;
 431                 break;
 432         case 5: /* peer was down */
 433                 if (conn_highest_disk(tconn) == D_UP_TO_DATE) {
 434                         /* we will(have) create(d) a new UUID anyways... */
 435                         ex_to_string = "peer is unreachable, assumed to be dead";
 436                         mask.pdsk = D_MASK;
 437                         val.pdsk = D_OUTDATED;
 438                 } else {
 439                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 440                 }
 441                 break;
 442         case 6: /* Peer is primary, voluntarily outdate myself.
 443                  * This is useful when an unconnected R_SECONDARY is asked to
 444                  * become R_PRIMARY, but finds the other peer being active. */
 445                 ex_to_string = "peer is active";
 446                 conn_warn(tconn, "Peer is primary, outdating myself.\n");
 447                 mask.disk = D_MASK;
 448                 val.disk = D_OUTDATED;
 449                 break;
 450         case 7:
 451                 if (fp != FP_STONITH)
 452                         conn_err(tconn, "fence-peer() = 7 && fencing != Stonith !!!\n");
 453                 ex_to_string = "peer was stonithed";
 454                 mask.pdsk = D_MASK;
 455                 val.pdsk = D_OUTDATED;
 456                 break;
 457         default:
 458                 /* The script is broken ... */
 459                 conn_err(tconn, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 460                 return false; /* Eventually leave IO frozen */
 461         }
 462
 463         conn_info(tconn, "fence-peer helper returned %d (%s)\n",
 464                   (r>>8) & 0xff, ex_to_string);
 465
 466  out:
 467
 468         /* Not using
 469            conn_request_state(tconn, mask, val, CS_VERBOSE);
 470            here, because we might were able to re-establish the connection in the
 471            meantime. */
 472         spin_lock_irq(&tconn->req_lock);
 473         if (tconn->cstate < C_WF_REPORT_PARAMS)
 474                 _conn_request_state(tconn, mask, val, CS_VERBOSE);
 475         spin_unlock_irq(&tconn->req_lock);
 476
 477         return conn_highest_pdsk(tconn) <= D_OUTDATED;
 478 }
 479
 480 static int _try_outdate_peer_async(void *data)
 481 {
 482         struct drbd_tconn *tconn = (struct drbd_tconn *)data;
 483
 484         conn_try_outdate_peer(tconn);
 485
 486         kref_put(&tconn->kref, &conn_destroy);
 487         return 0;
 488 }
 489
 490 void conn_try_outdate_peer_async(struct drbd_tconn *tconn)
 491 {
 492         struct task_struct *opa;
 493
 494         kref_get(&tconn->kref);
 495         opa = kthread_run(_try_outdate_peer_async, tconn, "drbd_async_h");
 496         if (IS_ERR(opa)) {
 497                 conn_err(tconn, "out of mem, failed to invoke fence-peer helper\n");
 498                 kref_put(&tconn->kref, &conn_destroy);
 499         }
 500 }
 501
 502 enum drbd_state_rv
 503 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 504 {
 505         const int max_tries = 4;
 506         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
 507         struct net_conf *nc;
 508         int try = 0;
 509         int forced = 0;
 510         union drbd_state mask, val;
 511
 512         if (new_role == R_PRIMARY)
 513                 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
 514
 515         mutex_lock(mdev->state_mutex);
 516
 517         mask.i = 0; mask.role = R_MASK;
 518         val.i  = 0; val.role  = new_role;
 519
 520         while (try++ < max_tries) {
 521                 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 522
 523                 /* in case we first succeeded to outdate,
 524                  * but now suddenly could establish a connection */
 525                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 526                         val.pdsk = 0;
 527                         mask.pdsk = 0;
 528                         continue;
 529                 }
 530
 531                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
 532                     (mdev->state.disk < D_UP_TO_DATE &&
 533                      mdev->state.disk >= D_INCONSISTENT)) {
 534                         mask.disk = D_MASK;
 535                         val.disk  = D_UP_TO_DATE;
 536                         forced = 1;
 537                         continue;
 538                 }
 539
 540                 if (rv == SS_NO_UP_TO_DATE_DISK &&
 541                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 542                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 543
 544                         if (conn_try_outdate_peer(mdev->tconn)) {
 545                                 val.disk = D_UP_TO_DATE;
 546                                 mask.disk = D_MASK;
 547                         }
 548                         continue;
 549                 }
 550
 551                 if (rv == SS_NOTHING_TO_DO)
 552                         goto out;
 553                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
 554                         if (!conn_try_outdate_peer(mdev->tconn) && force) {
 555                                 dev_warn(DEV, "Forced into split brain situation!\n");
 556                                 mask.pdsk = D_MASK;
 557                                 val.pdsk  = D_OUTDATED;
 558
 559                         }
 560                         continue;
 561                 }
 562                 if (rv == SS_TWO_PRIMARIES) {
 563                         /* Maybe the peer is detected as dead very soon...
 564                            retry at most once more in this case. */
 565                         int timeo;
 566                         rcu_read_lock();
 567                         nc = rcu_dereference(mdev->tconn->net_conf);
 568                         timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
 569                         rcu_read_unlock();
 570                         schedule_timeout_interruptible(timeo);
 571                         if (try < max_tries)
 572                                 try = max_tries - 1;
 573                         continue;
 574                 }
 575                 if (rv < SS_SUCCESS) {
 576                         rv = _drbd_request_state(mdev, mask, val,
 577                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 578                         if (rv < SS_SUCCESS)
 579                                 goto out;
 580                 }
 581                 break;
 582         }
 583
 584         if (rv < SS_SUCCESS)
 585                 goto out;
 586
 587         if (forced)
 588                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 589
 590         /* Wait until nothing is on the fly :) */
 591         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 592
 593         if (new_role == R_SECONDARY) {
 594                 set_disk_ro(mdev->vdisk, true);
 595                 if (get_ldev(mdev)) {
 596                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 597                         put_ldev(mdev);
 598                 }
 599         } else {
 600                 mutex_lock(&mdev->tconn->net_conf_update);
 601                 nc = mdev->tconn->net_conf;
 602                 if (nc)
 603                         nc->want_lose = 0; /* without copy; single bit op is atomic */
 604                 mutex_unlock(&mdev->tconn->net_conf_update);
 605
 606                 set_disk_ro(mdev->vdisk, false);
 607                 if (get_ldev(mdev)) {
 608                         if (((mdev->state.conn < C_CONNECTED ||
 609                                mdev->state.pdsk <= D_FAILED)
 610                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 611                                 drbd_uuid_new_current(mdev);
 612
 613                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 614                         put_ldev(mdev);
 615                 }
 616         }
 617
 618         /* writeout of activity log covered areas of the bitmap
 619          * to stable storage done in after state change already */
 620
 621         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 622                 /* if this was forced, we should consider sync */
 623                 if (forced)
 624                         drbd_send_uuids(mdev);
 625                 drbd_send_state(mdev);
 626         }
 627
 628         drbd_md_sync(mdev);
 629
 630         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 631 out:
 632         mutex_unlock(mdev->state_mutex);
 633         return rv;
 634 }
 635
 636 static const char *from_attrs_err_to_txt(int err)
 637 {
 638         return  err == -ENOMSG ? "required attribute missing" :
 639                 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
 640                 err == -EEXIST ? "can not change invariant setting" :
 641                 "invalid attribute value";
 642 }
 643
 644 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
 645 {
 646         struct set_role_parms parms;
 647         int err;
 648         enum drbd_ret_code retcode;
 649
 650         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
 651         if (!adm_ctx.reply_skb)
 652                 return retcode;
 653         if (retcode != NO_ERROR)
 654                 goto out;
 655
 656         memset(&parms, 0, sizeof(parms));
 657         if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
 658                 err = set_role_parms_from_attrs(&parms, info);
 659                 if (err) {
 660                         retcode = ERR_MANDATORY_TAG;
 661                         drbd_msg_put_info(from_attrs_err_to_txt(err));
 662                         goto out;
 663                 }
 664         }
 665
 666         if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
 667                 retcode = drbd_set_role(adm_ctx.mdev, R_PRIMARY, parms.assume_uptodate);
 668         else
 669                 retcode = drbd_set_role(adm_ctx.mdev, R_SECONDARY, 0);
 670 out:
 671         drbd_adm_finish(info, retcode);
 672         return 0;
 673 }
 674
 675 /* initializes the md.*_offset members, so we are able to find
 676  * the on disk meta data */
 677 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 678                                        struct drbd_backing_dev *bdev)
 679 {
 680         sector_t md_size_sect = 0;
 681         switch (bdev->dc.meta_dev_idx) {
 682         default:
 683                 /* v07 style fixed size indexed meta data */
 684                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 685                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 686                 bdev->md.al_offset = MD_AL_OFFSET;
 687                 bdev->md.bm_offset = MD_BM_OFFSET;
 688                 break;
 689         case DRBD_MD_INDEX_FLEX_EXT:
 690                 /* just occupy the full device; unit: sectors */
 691                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 692                 bdev->md.md_offset = 0;
 693                 bdev->md.al_offset = MD_AL_OFFSET;
 694                 bdev->md.bm_offset = MD_BM_OFFSET;
 695                 break;
 696         case DRBD_MD_INDEX_INTERNAL:
 697         case DRBD_MD_INDEX_FLEX_INT:
 698                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 699                 /* al size is still fixed */
 700                 bdev->md.al_offset = -MD_AL_SECTORS;
 701                 /* we need (slightly less than) ~ this much bitmap sectors: */
 702                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 703                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 704                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 705                 md_size_sect = ALIGN(md_size_sect, 8);
 706
 707                 /* plus the "drbd meta data super block",
 708                  * and the activity log; */
 709                 md_size_sect += MD_BM_OFFSET;
 710
 711                 bdev->md.md_size_sect = md_size_sect;
 712                 /* bitmap offset is adjusted by 'super' block size */
 713                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 714                 break;
 715         }
 716 }
 717
 718 /* input size is expected to be in KB */
 719 char *ppsize(char *buf, unsigned long long size)
 720 {
 721         /* Needs 9 bytes at max including trailing NUL:
 722          * -1ULL ==> "16384 EB" */
 723         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 724         int base = 0;
 725         while (size >= 10000 && base < sizeof(units)-1) {
 726                 /* shift + round */
 727                 size = (size >> 10) + !!(size & (1<<9));
 728                 base++;
 729         }
 730         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
 731
 732         return buf;
 733 }
 734
 735 /* there is still a theoretical deadlock when called from receiver
 736  * on an D_INCONSISTENT R_PRIMARY:
 737  *  remote READ does inc_ap_bio, receiver would need to receive answer
 738  *  packet from remote to dec_ap_bio again.
 739  *  receiver receive_sizes(), comes here,
 740  *  waits for ap_bio_cnt == 0. -> deadlock.
 741  * but this cannot happen, actually, because:
 742  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 743  *  (not connected, or bad/no disk on peer):
 744  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 745  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 746  *  peer may not initiate a resize.
 747  */
 748 /* Note these are not to be confused with
 749  * drbd_adm_suspend_io/drbd_adm_resume_io,
 750  * which are (sub) state changes triggered by admin (drbdsetup),
 751  * and can be long lived.
 752  * This changes an mdev->flag, is triggered by drbd internals,
 753  * and should be short-lived. */
 754 void drbd_suspend_io(struct drbd_conf *mdev)
 755 {
 756         set_bit(SUSPEND_IO, &mdev->flags);
 757         if (drbd_suspended(mdev))
 758                 return;
 759         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 760 }
 761
 762 void drbd_resume_io(struct drbd_conf *mdev)
 763 {
 764         clear_bit(SUSPEND_IO, &mdev->flags);
 765         wake_up(&mdev->misc_wait);
 766 }
 767
 768 /**
 769  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 770  * @mdev:       DRBD device.
 771  *
 772  * Returns 0 on success, negative return values indicate errors.
 773  * You should call drbd_md_sync() after calling this function.
 774  */
 775 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 776 {
 777         sector_t prev_first_sect, prev_size; /* previous meta location */
 778         sector_t la_size;
 779         sector_t size;
 780         char ppb[10];
 781
 782         int md_moved, la_size_changed;
 783         enum determine_dev_size rv = unchanged;
 784
 785         /* race:
 786          * application request passes inc_ap_bio,
 787          * but then cannot get an AL-reference.
 788          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 789          *
 790          * to avoid that:
 791          * Suspend IO right here.
 792          * still lock the act_log to not trigger ASSERTs there.
 793          */
 794         drbd_suspend_io(mdev);
 795
 796         /* no wait necessary anymore, actually we could assert that */
 797         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 798
 799         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 800         prev_size = mdev->ldev->md.md_size_sect;
 801         la_size = mdev->ldev->md.la_size_sect;
 802
 803         /* TODO: should only be some assert here, not (re)init... */
 804         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 805
 806         size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
 807
 808         if (drbd_get_capacity(mdev->this_bdev) != size ||
 809             drbd_bm_capacity(mdev) != size) {
 810                 int err;
 811                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 812                 if (unlikely(err)) {
 813                         /* currently there is only one error: ENOMEM! */
 814                         size = drbd_bm_capacity(mdev)>>1;
 815                         if (size == 0) {
 816                                 dev_err(DEV, "OUT OF MEMORY! "
 817                                     "Could not allocate bitmap!\n");
 818                         } else {
 819                                 dev_err(DEV, "BM resizing failed. "
 820                                     "Leaving size unchanged at size = %lu KB\n",
 821                                     (unsigned long)size);
 822                         }
 823                         rv = dev_size_error;
 824                 }
 825                 /* racy, see comments above. */
 826                 drbd_set_my_capacity(mdev, size);
 827                 mdev->ldev->md.la_size_sect = size;
 828                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 829                      (unsigned long long)size>>1);
 830         }
 831         if (rv == dev_size_error)
 832                 goto out;
 833
 834         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 835
 836         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 837                 || prev_size       != mdev->ldev->md.md_size_sect;
 838
 839         if (la_size_changed || md_moved) {
 840                 int err;
 841
 842                 drbd_al_shrink(mdev); /* All extents inactive. */
 843                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 844                          la_size_changed && md_moved ? "size changed and md moved" :
 845                          la_size_changed ? "size changed" : "md moved");
 846                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
 847                 err = drbd_bitmap_io(mdev, &drbd_bm_write,
 848                                 "size changed", BM_LOCKED_MASK);
 849                 if (err) {
 850                         rv = dev_size_error;
 851                         goto out;
 852                 }
 853                 drbd_md_mark_dirty(mdev);
 854         }
 855
 856         if (size > la_size)
 857                 rv = grew;
 858         if (size < la_size)
 859                 rv = shrunk;
 860 out:
 861         lc_unlock(mdev->act_log);
 862         wake_up(&mdev->al_wait);
 863         drbd_resume_io(mdev);
 864
 865         return rv;
 866 }
 867
 868 sector_t
 869 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
 870 {
 871         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 872         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 873         sector_t m_size; /* my size */
 874         sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
 875         sector_t size = 0;
 876
 877         m_size = drbd_get_max_capacity(bdev);
 878
 879         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 880                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 881                 p_size = m_size;
 882         }
 883
 884         if (p_size && m_size) {
 885                 size = min_t(sector_t, p_size, m_size);
 886         } else {
 887                 if (la_size) {
 888                         size = la_size;
 889                         if (m_size && m_size < size)
 890                                 size = m_size;
 891                         if (p_size && p_size < size)
 892                                 size = p_size;
 893                 } else {
 894                         if (m_size)
 895                                 size = m_size;
 896                         if (p_size)
 897                                 size = p_size;
 898                 }
 899         }
 900
 901         if (size == 0)
 902                 dev_err(DEV, "Both nodes diskless!\n");
 903
 904         if (u_size) {
 905                 if (u_size > size)
 906                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 907                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 908                 else
 909                         size = u_size;
 910         }
 911
 912         return size;
 913 }
 914
 915 /**
 916  * drbd_check_al_size() - Ensures that the AL is of the right size
 917  * @mdev:       DRBD device.
 918  *
 919  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 920  * failed, and 0 on success. You should call drbd_md_sync() after you called
 921  * this function.
 922  */
 923 static int drbd_check_al_size(struct drbd_conf *mdev, struct disk_conf *dc)
 924 {
 925         struct lru_cache *n, *t;
 926         struct lc_element *e;
 927         unsigned int in_use;
 928         int i;
 929
 930         if (!expect(dc->al_extents >= DRBD_AL_EXTENTS_MIN))
 931                 dc->al_extents = DRBD_AL_EXTENTS_MIN;
 932
 933         if (mdev->act_log &&
 934             mdev->act_log->nr_elements == dc->al_extents)
 935                 return 0;
 936
 937         in_use = 0;
 938         t = mdev->act_log;
 939         n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
 940                 dc->al_extents, sizeof(struct lc_element), 0);
 941
 942         if (n == NULL) {
 943                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 944                 return -ENOMEM;
 945         }
 946         spin_lock_irq(&mdev->al_lock);
 947         if (t) {
 948                 for (i = 0; i < t->nr_elements; i++) {
 949                         e = lc_element_by_index(t, i);
 950                         if (e->refcnt)
 951                                 dev_err(DEV, "refcnt(%d)==%d\n",
 952                                     e->lc_number, e->refcnt);
 953                         in_use += e->refcnt;
 954                 }
 955         }
 956         if (!in_use)
 957                 mdev->act_log = n;
 958         spin_unlock_irq(&mdev->al_lock);
 959         if (in_use) {
 960                 dev_err(DEV, "Activity log still in use!\n");
 961                 lc_destroy(n);
 962                 return -EBUSY;
 963         } else {
 964                 if (t)
 965                         lc_destroy(t);
 966         }
 967         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 968         return 0;
 969 }
 970
 971 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
 972 {
 973         struct request_queue * const q = mdev->rq_queue;
 974         int max_hw_sectors = max_bio_size >> 9;
 975         int max_segments = 0;
 976
 977         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 978                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 979
 980                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
 981                 max_segments = mdev->ldev->dc.max_bio_bvecs;
 982                 put_ldev(mdev);
 983         }
 984
 985         blk_queue_logical_block_size(q, 512);
 986         blk_queue_max_hw_sectors(q, max_hw_sectors);
 987         /* This is the workaround for "bio would need to, but cannot, be split" */
 988         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
 989         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
 990
 991         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 992                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 993
 994                 blk_queue_stack_limits(q, b);
 995
 996                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
 997                         dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
 998                                  q->backing_dev_info.ra_pages,
 999                                  b->backing_dev_info.ra_pages);
1000                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
1001                 }
1002                 put_ldev(mdev);
1003         }
1004 }
1005
1006 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
1007 {
1008         int now, new, local, peer;
1009
1010         now = queue_max_hw_sectors(mdev->rq_queue) << 9;
1011         local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
1012         peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
1013
1014         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1015                 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
1016                 mdev->local_max_bio_size = local;
1017                 put_ldev(mdev);
1018         }
1019
1020         /* We may ignore peer limits if the peer is modern enough.
1021            Because new from 8.3.8 onwards the peer can use multiple
1022            BIOs for a single peer_request */
1023         if (mdev->state.conn >= C_CONNECTED) {
1024                 if (mdev->tconn->agreed_pro_version < 94)
1025                         peer = mdev->peer_max_bio_size;
1026                 else if (mdev->tconn->agreed_pro_version == 94)
1027                         peer = DRBD_MAX_SIZE_H80_PACKET;
1028                 else /* drbd 8.3.8 onwards */
1029                         peer = DRBD_MAX_BIO_SIZE;
1030         }
1031
1032         new = min_t(int, local, peer);
1033
1034         if (mdev->state.role == R_PRIMARY && new < now)
1035                 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
1036
1037         if (new != now)
1038                 dev_info(DEV, "max BIO size = %u\n", new);
1039
1040         drbd_setup_queue_param(mdev, new);
1041 }
1042
1043 /* serialize deconfig (worker exiting, doing cleanup)
1044  * and reconfig (drbdsetup disk, drbdsetup net)
1045  *
1046  * Wait for a potentially exiting worker, then restart it,
1047  * or start a new one.  Flush any pending work, there may still be an
1048  * after_state_change queued.
1049  */
1050 static void conn_reconfig_start(struct drbd_tconn *tconn)
1051 {
1052         wait_event(tconn->ping_wait, !test_and_set_bit(CONFIG_PENDING, &tconn->flags));
1053         wait_event(tconn->ping_wait, !test_bit(OBJECT_DYING, &tconn->flags));
1054         drbd_thread_start(&tconn->worker);
1055         conn_flush_workqueue(tconn);
1056 }
1057
1058 /* if still unconfigured, stops worker again.
1059  * if configured now, clears CONFIG_PENDING.
1060  * wakes potential waiters */
1061 static void conn_reconfig_done(struct drbd_tconn *tconn)
1062 {
1063         spin_lock_irq(&tconn->req_lock);
1064         if (conn_all_vols_unconf(tconn)) {
1065                 set_bit(OBJECT_DYING, &tconn->flags);
1066                 drbd_thread_stop_nowait(&tconn->worker);
1067         } else
1068                 clear_bit(CONFIG_PENDING, &tconn->flags);
1069         spin_unlock_irq(&tconn->req_lock);
1070         wake_up(&tconn->ping_wait);
1071 }
1072
1073 /* Make sure IO is suspended before calling this function(). */
1074 static void drbd_suspend_al(struct drbd_conf *mdev)
1075 {
1076         int s = 0;
1077
1078         if (!lc_try_lock(mdev->act_log)) {
1079                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
1080                 return;
1081         }
1082
1083         drbd_al_shrink(mdev);
1084         spin_lock_irq(&mdev->tconn->req_lock);
1085         if (mdev->state.conn < C_CONNECTED)
1086                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
1087         spin_unlock_irq(&mdev->tconn->req_lock);
1088         lc_unlock(mdev->act_log);
1089
1090         if (s)
1091                 dev_info(DEV, "Suspended AL updates\n");
1092 }
1093
1094 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1095 {
1096         enum drbd_ret_code retcode;
1097         struct drbd_conf *mdev;
1098         struct disk_conf *ndc; /* new disk conf */
1099         int err, fifo_size;
1100         int *rs_plan_s = NULL;
1101
1102         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1103         if (!adm_ctx.reply_skb)
1104                 return retcode;
1105         if (retcode != NO_ERROR)
1106                 goto out;
1107
1108         mdev = adm_ctx.mdev;
1109
1110         /* we also need a disk
1111          * to change the options on */
1112         if (!get_ldev(mdev)) {
1113                 retcode = ERR_NO_DISK;
1114                 goto out;
1115         }
1116
1117 /* FIXME freeze IO, cluster wide.
1118  *
1119  * We should make sure no-one uses
1120  * some half-updated struct when we
1121  * assign it later. */
1122
1123         ndc = kmalloc(sizeof(*ndc), GFP_KERNEL);
1124         if (!ndc) {
1125                 retcode = ERR_NOMEM;
1126                 goto fail;
1127         }
1128
1129         memcpy(ndc, &mdev->ldev->dc, sizeof(*ndc));
1130         err = disk_conf_from_attrs_for_change(ndc, info);
1131         if (err) {
1132                 retcode = ERR_MANDATORY_TAG;
1133                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1134         }
1135
1136         if (!expect(ndc->resync_rate >= 1))
1137                 ndc->resync_rate = 1;
1138
1139         /* clip to allowed range */
1140         if (!expect(ndc->al_extents >= DRBD_AL_EXTENTS_MIN))
1141                 ndc->al_extents = DRBD_AL_EXTENTS_MIN;
1142         if (!expect(ndc->al_extents <= DRBD_AL_EXTENTS_MAX))
1143                 ndc->al_extents = DRBD_AL_EXTENTS_MAX;
1144
1145         /* most sanity checks done, try to assign the new sync-after
1146          * dependency.  need to hold the global lock in there,
1147          * to avoid a race in the dependency loop check. */
1148         retcode = drbd_alter_sa(mdev, ndc->resync_after);
1149         if (retcode != NO_ERROR)
1150                 goto fail;
1151
1152         fifo_size = (ndc->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1153         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1154                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1155                 if (!rs_plan_s) {
1156                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1157                         retcode = ERR_NOMEM;
1158                         goto fail;
1159                 }
1160         }
1161
1162         if (fifo_size != mdev->rs_plan_s.size) {
1163                 kfree(mdev->rs_plan_s.values);
1164                 mdev->rs_plan_s.values = rs_plan_s;
1165                 mdev->rs_plan_s.size   = fifo_size;
1166                 mdev->rs_planed = 0;
1167                 rs_plan_s = NULL;
1168         }
1169
1170         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1171         drbd_al_shrink(mdev);
1172         err = drbd_check_al_size(mdev, ndc);
1173         lc_unlock(mdev->act_log);
1174         wake_up(&mdev->al_wait);
1175
1176         if (err) {
1177                 retcode = ERR_NOMEM;
1178                 goto fail;
1179         }
1180
1181         /* FIXME
1182          * To avoid someone looking at a half-updated struct, we probably
1183          * should have a rw-semaphor on net_conf and disk_conf.
1184          */
1185         mdev->ldev->dc = *ndc;
1186
1187         drbd_md_sync(mdev);
1188
1189
1190         if (mdev->state.conn >= C_CONNECTED)
1191                 drbd_send_sync_param(mdev);
1192
1193  fail:
1194         put_ldev(mdev);
1195         kfree(ndc);
1196         kfree(rs_plan_s);
1197  out:
1198         drbd_adm_finish(info, retcode);
1199         return 0;
1200 }
1201
1202 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1203 {
1204         struct drbd_conf *mdev;
1205         int err;
1206         enum drbd_ret_code retcode;
1207         enum determine_dev_size dd;
1208         sector_t max_possible_sectors;
1209         sector_t min_md_device_sectors;
1210         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1211         struct block_device *bdev;
1212         struct lru_cache *resync_lru = NULL;
1213         union drbd_state ns, os;
1214         enum drbd_state_rv rv;
1215         struct net_conf *nc;
1216         int cp_discovered = 0;
1217
1218         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1219         if (!adm_ctx.reply_skb)
1220                 return retcode;
1221         if (retcode != NO_ERROR)
1222                 goto finish;
1223
1224         mdev = adm_ctx.mdev;
1225         conn_reconfig_start(mdev->tconn);
1226
1227         /* if you want to reconfigure, please tear down first */
1228         if (mdev->state.disk > D_DISKLESS) {
1229                 retcode = ERR_DISK_CONFIGURED;
1230                 goto fail;
1231         }
1232         /* It may just now have detached because of IO error.  Make sure
1233          * drbd_ldev_destroy is done already, we may end up here very fast,
1234          * e.g. if someone calls attach from the on-io-error handler,
1235          * to realize a "hot spare" feature (not that I'd recommend that) */
1236         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1237
1238         /* allocation not in the IO path, drbdsetup context */
1239         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1240         if (!nbc) {
1241                 retcode = ERR_NOMEM;
1242                 goto fail;
1243         }
1244
1245         nbc->dc = (struct disk_conf) {
1246                 {}, 0, /* backing_dev */
1247                 {}, 0, /* meta_dev */
1248                 0, /* meta_dev_idx */
1249                 DRBD_DISK_SIZE_SECT_DEF, /* disk_size */
1250                 DRBD_MAX_BIO_BVECS_DEF, /* max_bio_bvecs */
1251                 DRBD_ON_IO_ERROR_DEF, /* on_io_error */
1252                 DRBD_FENCING_DEF, /* fencing */
1253                 DRBD_RATE_DEF, /* resync_rate */
1254                 DRBD_AFTER_DEF, /* resync_after */
1255                 DRBD_AL_EXTENTS_DEF, /* al_extents */
1256                 DRBD_C_PLAN_AHEAD_DEF, /* c_plan_ahead */
1257                 DRBD_C_DELAY_TARGET_DEF, /* c_delay_target */
1258                 DRBD_C_FILL_TARGET_DEF, /* c_fill_target */
1259                 DRBD_C_MAX_RATE_DEF, /* c_max_rate */
1260                 DRBD_C_MIN_RATE_DEF, /* c_min_rate */
1261                 0, /* no_disk_barrier */
1262                 0, /* no_disk_flush */
1263                 0, /* no_disk_drain */
1264                 0, /* no_md_flush */
1265         };
1266
1267         err = disk_conf_from_attrs(&nbc->dc, info);
1268         if (err) {
1269                 retcode = ERR_MANDATORY_TAG;
1270                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1271                 goto fail;
1272         }
1273
1274         if ((int)nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1275                 retcode = ERR_MD_IDX_INVALID;
1276                 goto fail;
1277         }
1278
1279         rcu_read_lock();
1280         nc = rcu_dereference(mdev->tconn->net_conf);
1281         if (nc) {
1282                 if (nbc->dc.fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1283                         rcu_read_unlock();
1284                         retcode = ERR_STONITH_AND_PROT_A;
1285                         goto fail;
1286                 }
1287         }
1288         rcu_read_unlock();
1289
1290         bdev = blkdev_get_by_path(nbc->dc.backing_dev,
1291                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1292         if (IS_ERR(bdev)) {
1293                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
1294                         PTR_ERR(bdev));
1295                 retcode = ERR_OPEN_DISK;
1296                 goto fail;
1297         }
1298         nbc->backing_bdev = bdev;
1299
1300         /*
1301          * meta_dev_idx >= 0: external fixed size, possibly multiple
1302          * drbd sharing one meta device.  TODO in that case, paranoia
1303          * check that [md_bdev, meta_dev_idx] is not yet used by some
1304          * other drbd minor!  (if you use drbd.conf + drbdadm, that
1305          * should check it for you already; but if you don't, or
1306          * someone fooled it, we need to double check here)
1307          */
1308         bdev = blkdev_get_by_path(nbc->dc.meta_dev,
1309                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1310                                   ((int)nbc->dc.meta_dev_idx < 0) ?
1311                                   (void *)mdev : (void *)drbd_m_holder);
1312         if (IS_ERR(bdev)) {
1313                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
1314                         PTR_ERR(bdev));
1315                 retcode = ERR_OPEN_MD_DISK;
1316                 goto fail;
1317         }
1318         nbc->md_bdev = bdev;
1319
1320         if ((nbc->backing_bdev == nbc->md_bdev) !=
1321             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1322              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1323                 retcode = ERR_MD_IDX_INVALID;
1324                 goto fail;
1325         }
1326
1327         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1328                         1, 61, sizeof(struct bm_extent),
1329                         offsetof(struct bm_extent, lce));
1330         if (!resync_lru) {
1331                 retcode = ERR_NOMEM;
1332                 goto fail;
1333         }
1334
1335         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1336         drbd_md_set_sector_offsets(mdev, nbc);
1337
1338         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
1339                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1340                         (unsigned long long) drbd_get_max_capacity(nbc),
1341                         (unsigned long long) nbc->dc.disk_size);
1342                 retcode = ERR_DISK_TO_SMALL;
1343                 goto fail;
1344         }
1345
1346         if ((int)nbc->dc.meta_dev_idx < 0) {
1347                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1348                 /* at least one MB, otherwise it does not make sense */
1349                 min_md_device_sectors = (2<<10);
1350         } else {
1351                 max_possible_sectors = DRBD_MAX_SECTORS;
1352                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
1353         }
1354
1355         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1356                 retcode = ERR_MD_DISK_TO_SMALL;
1357                 dev_warn(DEV, "refusing attach: md-device too small, "
1358                      "at least %llu sectors needed for this meta-disk type\n",
1359                      (unsigned long long) min_md_device_sectors);
1360                 goto fail;
1361         }
1362
1363         /* Make sure the new disk is big enough
1364          * (we may currently be R_PRIMARY with no local disk...) */
1365         if (drbd_get_max_capacity(nbc) <
1366             drbd_get_capacity(mdev->this_bdev)) {
1367                 retcode = ERR_DISK_TO_SMALL;
1368                 goto fail;
1369         }
1370
1371         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1372
1373         if (nbc->known_size > max_possible_sectors) {
1374                 dev_warn(DEV, "==> truncating very big lower level device "
1375                         "to currently maximum possible %llu sectors <==\n",
1376                         (unsigned long long) max_possible_sectors);
1377                 if ((int)nbc->dc.meta_dev_idx >= 0)
1378                         dev_warn(DEV, "==>> using internal or flexible "
1379                                       "meta data may help <<==\n");
1380         }
1381
1382         drbd_suspend_io(mdev);
1383         /* also wait for the last barrier ack. */
1384         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || drbd_suspended(mdev));
1385         /* and for any other previously queued work */
1386         drbd_flush_workqueue(mdev);
1387
1388         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1389         retcode = rv;  /* FIXME: Type mismatch. */
1390         drbd_resume_io(mdev);
1391         if (rv < SS_SUCCESS)
1392                 goto fail;
1393
1394         if (!get_ldev_if_state(mdev, D_ATTACHING))
1395                 goto force_diskless;
1396
1397         drbd_md_set_sector_offsets(mdev, nbc);
1398
1399         if (!mdev->bitmap) {
1400                 if (drbd_bm_init(mdev)) {
1401                         retcode = ERR_NOMEM;
1402                         goto force_diskless_dec;
1403                 }
1404         }
1405
1406         retcode = drbd_md_read(mdev, nbc);
1407         if (retcode != NO_ERROR)
1408                 goto force_diskless_dec;
1409
1410         if (mdev->state.conn < C_CONNECTED &&
1411             mdev->state.role == R_PRIMARY &&
1412             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1413                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1414                     (unsigned long long)mdev->ed_uuid);
1415                 retcode = ERR_DATA_NOT_CURRENT;
1416                 goto force_diskless_dec;
1417         }
1418
1419         /* Since we are diskless, fix the activity log first... */
1420         if (drbd_check_al_size(mdev, &nbc->dc)) {
1421                 retcode = ERR_NOMEM;
1422                 goto force_diskless_dec;
1423         }
1424
1425         /* Prevent shrinking of consistent devices ! */
1426         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1427             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1428                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1429                 retcode = ERR_DISK_TO_SMALL;
1430                 goto force_diskless_dec;
1431         }
1432
1433         if (!drbd_al_read_log(mdev, nbc)) {
1434                 retcode = ERR_IO_MD_DISK;
1435                 goto force_diskless_dec;
1436         }
1437
1438         /* Reset the "barriers don't work" bits here, then force meta data to
1439          * be written, to ensure we determine if barriers are supported. */
1440         if (nbc->dc.no_md_flush)
1441                 set_bit(MD_NO_FUA, &mdev->flags);
1442         else
1443                 clear_bit(MD_NO_FUA, &mdev->flags);
1444
1445         /* Point of no return reached.
1446          * Devices and memory are no longer released by error cleanup below.
1447          * now mdev takes over responsibility, and the state engine should
1448          * clean it up somewhere.  */
1449         D_ASSERT(mdev->ldev == NULL);
1450         mdev->ldev = nbc;
1451         mdev->resync = resync_lru;
1452         nbc = NULL;
1453         resync_lru = NULL;
1454
1455         mdev->write_ordering = WO_bdev_flush;
1456         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1457
1458         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1459                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1460         else
1461                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1462
1463         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1464             !(mdev->state.role == R_PRIMARY && mdev->tconn->susp_nod)) {
1465                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1466                 cp_discovered = 1;
1467         }
1468
1469         mdev->send_cnt = 0;
1470         mdev->recv_cnt = 0;
1471         mdev->read_cnt = 0;
1472         mdev->writ_cnt = 0;
1473
1474         drbd_reconsider_max_bio_size(mdev);
1475
1476         /* If I am currently not R_PRIMARY,
1477          * but meta data primary indicator is set,
1478          * I just now recover from a hard crash,
1479          * and have been R_PRIMARY before that crash.
1480          *
1481          * Now, if I had no connection before that crash
1482          * (have been degraded R_PRIMARY), chances are that
1483          * I won't find my peer now either.
1484          *
1485          * In that case, and _only_ in that case,
1486          * we use the degr-wfc-timeout instead of the default,
1487          * so we can automatically recover from a crash of a
1488          * degraded but active "cluster" after a certain timeout.
1489          */
1490         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1491         if (mdev->state.role != R_PRIMARY &&
1492              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1493             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1494                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1495
1496         dd = drbd_determine_dev_size(mdev, 0);
1497         if (dd == dev_size_error) {
1498                 retcode = ERR_NOMEM_BITMAP;
1499                 goto force_diskless_dec;
1500         } else if (dd == grew)
1501                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1502
1503         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1504                 dev_info(DEV, "Assuming that all blocks are out of sync "
1505                      "(aka FullSync)\n");
1506                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1507                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1508                         retcode = ERR_IO_MD_DISK;
1509                         goto force_diskless_dec;
1510                 }
1511         } else {
1512                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1513                         "read from attaching", BM_LOCKED_MASK)) {
1514                         retcode = ERR_IO_MD_DISK;
1515                         goto force_diskless_dec;
1516                 }
1517         }
1518
1519         if (cp_discovered) {
1520                 drbd_al_apply_to_bm(mdev);
1521                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1522                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1523                         retcode = ERR_IO_MD_DISK;
1524                         goto force_diskless_dec;
1525                 }
1526         }
1527
1528         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1529                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1530
1531         spin_lock_irq(&mdev->tconn->req_lock);
1532         os = drbd_read_state(mdev);
1533         ns = os;
1534         /* If MDF_CONSISTENT is not set go into inconsistent state,
1535            otherwise investigate MDF_WasUpToDate...
1536            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1537            otherwise into D_CONSISTENT state.
1538         */
1539         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1540                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1541                         ns.disk = D_CONSISTENT;
1542                 else
1543                         ns.disk = D_OUTDATED;
1544         } else {
1545                 ns.disk = D_INCONSISTENT;
1546         }
1547
1548         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1549                 ns.pdsk = D_OUTDATED;
1550
1551         if ( ns.disk == D_CONSISTENT &&
1552             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1553                 ns.disk = D_UP_TO_DATE;
1554
1555         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1556            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1557            this point, because drbd_request_state() modifies these
1558            flags. */
1559
1560         /* In case we are C_CONNECTED postpone any decision on the new disk
1561            state after the negotiation phase. */
1562         if (mdev->state.conn == C_CONNECTED) {
1563                 mdev->new_state_tmp.i = ns.i;
1564                 ns.i = os.i;
1565                 ns.disk = D_NEGOTIATING;
1566
1567                 /* We expect to receive up-to-date UUIDs soon.
1568                    To avoid a race in receive_state, free p_uuid while
1569                    holding req_lock. I.e. atomic with the state change */
1570                 kfree(mdev->p_uuid);
1571                 mdev->p_uuid = NULL;
1572         }
1573
1574         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1575         spin_unlock_irq(&mdev->tconn->req_lock);
1576
1577         if (rv < SS_SUCCESS)
1578                 goto force_diskless_dec;
1579
1580         if (mdev->state.role == R_PRIMARY)
1581                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1582         else
1583                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1584
1585         drbd_md_mark_dirty(mdev);
1586         drbd_md_sync(mdev);
1587
1588         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1589         put_ldev(mdev);
1590         conn_reconfig_done(mdev->tconn);
1591         drbd_adm_finish(info, retcode);
1592         return 0;
1593
1594  force_diskless_dec:
1595         put_ldev(mdev);
1596  force_diskless:
1597         drbd_force_state(mdev, NS(disk, D_FAILED));
1598         drbd_md_sync(mdev);
1599  fail:
1600         conn_reconfig_done(mdev->tconn);
1601         if (nbc) {
1602                 if (nbc->backing_bdev)
1603                         blkdev_put(nbc->backing_bdev,
1604                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1605                 if (nbc->md_bdev)
1606                         blkdev_put(nbc->md_bdev,
1607                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1608                 kfree(nbc);
1609         }
1610         lc_destroy(resync_lru);
1611
1612  finish:
1613         drbd_adm_finish(info, retcode);
1614         return 0;
1615 }
1616
1617 static int adm_detach(struct drbd_conf *mdev)
1618 {
1619         enum drbd_state_rv retcode;
1620         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1621         retcode = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1622         wait_event(mdev->misc_wait,
1623                         mdev->state.disk != D_DISKLESS ||
1624                         !atomic_read(&mdev->local_cnt));
1625         drbd_resume_io(mdev);
1626         return retcode;
1627 }
1628
1629 /* Detaching the disk is a process in multiple stages.  First we need to lock
1630  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1631  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1632  * internal references as well.
1633  * Only then we have finally detached. */
1634 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1635 {
1636         enum drbd_ret_code retcode;
1637
1638         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1639         if (!adm_ctx.reply_skb)
1640                 return retcode;
1641         if (retcode != NO_ERROR)
1642                 goto out;
1643
1644         retcode = adm_detach(adm_ctx.mdev);
1645 out:
1646         drbd_adm_finish(info, retcode);
1647         return 0;
1648 }
1649
1650 static bool conn_resync_running(struct drbd_tconn *tconn)
1651 {
1652         struct drbd_conf *mdev;
1653         bool rv = false;
1654         int vnr;
1655
1656         rcu_read_lock();
1657         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1658                 if (mdev->state.conn == C_SYNC_SOURCE ||
1659                     mdev->state.conn == C_SYNC_TARGET ||
1660                     mdev->state.conn == C_PAUSED_SYNC_S ||
1661                     mdev->state.conn == C_PAUSED_SYNC_T) {
1662                         rv = true;
1663                         break;
1664                 }
1665         }
1666         rcu_read_unlock();
1667
1668         return rv;
1669 }
1670
1671 static bool conn_ov_running(struct drbd_tconn *tconn)
1672 {
1673         struct drbd_conf *mdev;
1674         bool rv = false;
1675         int vnr;
1676
1677         rcu_read_lock();
1678         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1679                 if (mdev->state.conn == C_VERIFY_S ||
1680                     mdev->state.conn == C_VERIFY_T) {
1681                         rv = true;
1682                         break;
1683                 }
1684         }
1685         rcu_read_unlock();
1686
1687         return rv;
1688 }
1689
1690 static enum drbd_ret_code
1691 _check_net_options(struct drbd_tconn *tconn, struct net_conf *old_conf, struct net_conf *new_conf)
1692 {
1693         struct drbd_conf *mdev;
1694         int i;
1695
1696         if (old_conf && tconn->agreed_pro_version < 100 &&
1697             tconn->cstate == C_WF_REPORT_PARAMS &&
1698             new_conf->wire_protocol != old_conf->wire_protocol)
1699                 return ERR_NEED_APV_100;
1700
1701         if (new_conf->two_primaries &&
1702             (new_conf->wire_protocol != DRBD_PROT_C))
1703                 return ERR_NOT_PROTO_C;
1704
1705         idr_for_each_entry(&tconn->volumes, mdev, i) {
1706                 if (get_ldev(mdev)) {
1707                         enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1708                         put_ldev(mdev);
1709                         if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
1710                                 return ERR_STONITH_AND_PROT_A;
1711                 }
1712                 if (mdev->state.role == R_PRIMARY && new_conf->want_lose)
1713                         return ERR_DISCARD;
1714         }
1715
1716         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A)
1717                 return ERR_CONG_NOT_PROTO_A;
1718
1719         return NO_ERROR;
1720 }
1721
1722 static enum drbd_ret_code
1723 check_net_options(struct drbd_tconn *tconn, struct net_conf *new_conf)
1724 {
1725         static enum drbd_ret_code rv;
1726         struct drbd_conf *mdev;
1727         int i;
1728
1729         rcu_read_lock();
1730         rv = _check_net_options(tconn, rcu_dereference(tconn->net_conf), new_conf);
1731         rcu_read_unlock();
1732
1733         /* tconn->volumes protected by genl_lock() here */
1734         idr_for_each_entry(&tconn->volumes, mdev, i) {
1735                 if (!mdev->bitmap) {
1736                         if(drbd_bm_init(mdev))
1737                                 return ERR_NOMEM;
1738                 }
1739         }
1740
1741         return rv;
1742 }
1743
1744 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
1745 {
1746         enum drbd_ret_code retcode;
1747         struct drbd_tconn *tconn;
1748         struct net_conf *old_conf, *new_conf = NULL;
1749         int err;
1750         int ovr; /* online verify running */
1751         int rsr; /* re-sync running */
1752         struct crypto_hash *verify_tfm = NULL;
1753         struct crypto_hash *csums_tfm = NULL;
1754
1755
1756         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1757         if (!adm_ctx.reply_skb)
1758                 return retcode;
1759         if (retcode != NO_ERROR)
1760                 goto out;
1761
1762         tconn = adm_ctx.tconn;
1763
1764         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1765         if (!new_conf) {
1766                 retcode = ERR_NOMEM;
1767                 goto out;
1768         }
1769
1770         conn_reconfig_start(tconn);
1771
1772         mutex_lock(&tconn->net_conf_update);
1773         old_conf = tconn->net_conf;
1774
1775         if (!old_conf) {
1776                 drbd_msg_put_info("net conf missing, try connect");
1777                 retcode = ERR_INVALID_REQUEST;
1778                 goto fail;
1779         }
1780
1781         *new_conf = *old_conf;
1782
1783         err = net_conf_from_attrs_for_change(new_conf, info);
1784         if (err) {
1785                 retcode = ERR_MANDATORY_TAG;
1786                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1787                 goto fail;
1788         }
1789
1790         retcode = check_net_options(tconn, new_conf);
1791         if (retcode != NO_ERROR)
1792                 goto fail;
1793
1794         /* re-sync running */
1795         rsr = conn_resync_running(tconn);
1796         if (rsr && old_conf && strcmp(new_conf->csums_alg, old_conf->csums_alg)) {
1797                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1798                 goto fail;
1799         }
1800
1801         if (!rsr && new_conf->csums_alg[0]) {
1802                 csums_tfm = crypto_alloc_hash(new_conf->csums_alg, 0, CRYPTO_ALG_ASYNC);
1803                 if (IS_ERR(csums_tfm)) {
1804                         csums_tfm = NULL;
1805                         retcode = ERR_CSUMS_ALG;
1806                         goto fail;
1807                 }
1808
1809                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1810                         retcode = ERR_CSUMS_ALG_ND;
1811                         goto fail;
1812                 }
1813         }
1814
1815         /* online verify running */
1816         ovr = conn_ov_running(tconn);
1817         if (ovr) {
1818                 if (strcmp(new_conf->verify_alg, old_conf->verify_alg)) {
1819                         retcode = ERR_VERIFY_RUNNING;
1820                         goto fail;
1821                 }
1822         }
1823
1824         if (!ovr && new_conf->verify_alg[0]) {
1825                 verify_tfm = crypto_alloc_hash(new_conf->verify_alg, 0, CRYPTO_ALG_ASYNC);
1826                 if (IS_ERR(verify_tfm)) {
1827                         verify_tfm = NULL;
1828                         retcode = ERR_VERIFY_ALG;
1829                         goto fail;
1830                 }
1831
1832                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1833                         retcode = ERR_VERIFY_ALG_ND;
1834                         goto fail;
1835                 }
1836         }
1837
1838         rcu_assign_pointer(tconn->net_conf, new_conf);
1839
1840         if (!rsr) {
1841                 crypto_free_hash(tconn->csums_tfm);
1842                 tconn->csums_tfm = csums_tfm;
1843                 csums_tfm = NULL;
1844         }
1845         if (!ovr) {
1846                 crypto_free_hash(tconn->verify_tfm);
1847                 tconn->verify_tfm = verify_tfm;
1848                 verify_tfm = NULL;
1849         }
1850
1851         mutex_unlock(&tconn->net_conf_update);
1852         synchronize_rcu();
1853         kfree(old_conf);
1854
1855         if (tconn->cstate >= C_WF_REPORT_PARAMS)
1856                 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn)));
1857
1858         goto done;
1859
1860  fail:
1861         mutex_unlock(&tconn->net_conf_update);
1862         crypto_free_hash(csums_tfm);
1863         crypto_free_hash(verify_tfm);
1864         kfree(new_conf);
1865  done:
1866         conn_reconfig_done(tconn);
1867  out:
1868         drbd_adm_finish(info, retcode);
1869         return 0;
1870 }
1871
1872 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
1873 {
1874         char hmac_name[CRYPTO_MAX_ALG_NAME];
1875         struct drbd_conf *mdev;
1876         struct net_conf *old_conf, *new_conf = NULL;
1877         struct crypto_hash *tfm = NULL;
1878         struct crypto_hash *integrity_w_tfm = NULL;
1879         struct crypto_hash *integrity_r_tfm = NULL;
1880         void *int_dig_in = NULL;
1881         void *int_dig_vv = NULL;
1882         struct drbd_tconn *oconn;
1883         struct drbd_tconn *tconn;
1884         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1885         enum drbd_ret_code retcode;
1886         int i;
1887         int err;
1888
1889         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1890         if (!adm_ctx.reply_skb)
1891                 return retcode;
1892         if (retcode != NO_ERROR)
1893                 goto out;
1894
1895         tconn = adm_ctx.tconn;
1896         conn_reconfig_start(tconn);
1897
1898         if (tconn->cstate > C_STANDALONE) {
1899                 retcode = ERR_NET_CONFIGURED;
1900                 goto fail;
1901         }
1902
1903         /* allocation not in the IO path, cqueue thread context */
1904         new_conf = kmalloc(sizeof(struct net_conf), GFP_KERNEL);
1905         if (!new_conf) {
1906                 retcode = ERR_NOMEM;
1907                 goto fail;
1908         }
1909
1910         *new_conf = (struct net_conf) {
1911                 {}, 0, /* my_addr */
1912                 {}, 0, /* peer_addr */
1913                 {}, 0, /* shared_secret */
1914                 {}, 0, /* cram_hmac_alg */
1915                 {}, 0, /* integrity_alg */
1916                 {}, 0, /* verify_alg */
1917                 {}, 0, /* csums_alg */
1918                 DRBD_PROTOCOL_DEF, /* wire_protocol */
1919                 DRBD_CONNECT_INT_DEF, /* try_connect_int */
1920                 DRBD_TIMEOUT_DEF, /* timeout */
1921                 DRBD_PING_INT_DEF, /* ping_int */
1922                 DRBD_PING_TIMEO_DEF, /* ping_timeo */
1923                 DRBD_SNDBUF_SIZE_DEF, /* sndbuf_size */
1924                 DRBD_RCVBUF_SIZE_DEF, /* rcvbuf_size */
1925                 DRBD_KO_COUNT_DEF, /* ko_count */
1926                 DRBD_MAX_BUFFERS_DEF, /* max_buffers */
1927                 DRBD_MAX_EPOCH_SIZE_DEF, /* max_epoch_size */
1928                 DRBD_UNPLUG_WATERMARK_DEF, /* unplug_watermark */
1929                 DRBD_AFTER_SB_0P_DEF, /* after_sb_0p */
1930                 DRBD_AFTER_SB_1P_DEF, /* after_sb_1p */
1931                 DRBD_AFTER_SB_2P_DEF, /* after_sb_2p */
1932                 DRBD_RR_CONFLICT_DEF, /* rr_conflict */
1933                 DRBD_ON_CONGESTION_DEF, /* on_congestion */
1934                 DRBD_CONG_FILL_DEF, /* cong_fill */
1935                 DRBD_CONG_EXTENTS_DEF, /* cong_extents */
1936                 0, /* two_primaries */
1937                 0, /* want_lose */
1938                 0, /* no_cork */
1939                 0, /* always_asbp */
1940                 0, /* dry_run */
1941                 0, /* use_rle */
1942         };
1943
1944         err = net_conf_from_attrs(new_conf, info);
1945         if (err) {
1946                 retcode = ERR_MANDATORY_TAG;
1947                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1948                 goto fail;
1949         }
1950
1951         retcode = check_net_options(tconn, new_conf);
1952         if (retcode != NO_ERROR)
1953                 goto fail;
1954
1955         retcode = NO_ERROR;
1956
1957         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1958         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1959
1960         /* No need to take drbd_cfg_rwsem here.  All reconfiguration is
1961          * strictly serialized on genl_lock(). We are protected against
1962          * concurrent reconfiguration/addition/deletion */
1963         list_for_each_entry(oconn, &drbd_tconns, all_tconn) {
1964                 struct net_conf *nc;
1965                 if (oconn == tconn)
1966                         continue;
1967
1968                 rcu_read_lock();
1969                 nc = rcu_dereference(oconn->net_conf);
1970                 if (nc) {
1971                         taken_addr = (struct sockaddr *)&nc->my_addr;
1972                         if (new_conf->my_addr_len == nc->my_addr_len &&
1973                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1974                                 retcode = ERR_LOCAL_ADDR;
1975
1976                         taken_addr = (struct sockaddr *)&nc->peer_addr;
1977                         if (new_conf->peer_addr_len == nc->peer_addr_len &&
1978                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1979                                 retcode = ERR_PEER_ADDR;
1980                 }
1981                 rcu_read_unlock();
1982                 if (retcode != NO_ERROR)
1983                         goto fail;
1984         }
1985
1986         if (new_conf->cram_hmac_alg[0] != 0) {
1987                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1988                         new_conf->cram_hmac_alg);
1989                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1990                 if (IS_ERR(tfm)) {
1991                         tfm = NULL;
1992                         retcode = ERR_AUTH_ALG;
1993                         goto fail;
1994                 }
1995
1996                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1997                         retcode = ERR_AUTH_ALG_ND;
1998                         goto fail;
1999                 }
2000         }
2001
2002         if (new_conf->integrity_alg[0]) {
2003                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
2004                 if (IS_ERR(integrity_w_tfm)) {
2005                         integrity_w_tfm = NULL;
2006                         retcode=ERR_INTEGRITY_ALG;
2007                         goto fail;
2008                 }
2009
2010                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
2011                         retcode=ERR_INTEGRITY_ALG_ND;
2012                         goto fail;
2013                 }
2014
2015                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
2016                 if (IS_ERR(integrity_r_tfm)) {
2017                         integrity_r_tfm = NULL;
2018                         retcode=ERR_INTEGRITY_ALG;
2019                         goto fail;
2020                 }
2021         }
2022
2023         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2024
2025         /* allocation not in the IO path, cqueue thread context */
2026         if (integrity_w_tfm) {
2027                 i = crypto_hash_digestsize(integrity_w_tfm);
2028                 int_dig_in = kmalloc(i, GFP_KERNEL);
2029                 if (!int_dig_in) {
2030                         retcode = ERR_NOMEM;
2031                         goto fail;
2032                 }
2033                 int_dig_vv = kmalloc(i, GFP_KERNEL);
2034                 if (!int_dig_vv) {
2035                         retcode = ERR_NOMEM;
2036                         goto fail;
2037                 }
2038         }
2039
2040         conn_flush_workqueue(tconn);
2041
2042         mutex_lock(&tconn->net_conf_update);
2043         old_conf = tconn->net_conf;
2044         if (old_conf) {
2045                 retcode = ERR_NET_CONFIGURED;
2046                 mutex_unlock(&tconn->net_conf_update);
2047                 goto fail;
2048         }
2049         rcu_assign_pointer(tconn->net_conf, new_conf);
2050
2051         conn_free_crypto(tconn);
2052         tconn->cram_hmac_tfm = tfm;
2053         tconn->integrity_w_tfm = integrity_w_tfm;
2054         tconn->integrity_r_tfm = integrity_r_tfm;
2055         tconn->int_dig_in = int_dig_in;
2056         tconn->int_dig_vv = int_dig_vv;
2057
2058         mutex_unlock(&tconn->net_conf_update);
2059
2060         retcode = conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2061
2062         rcu_read_lock();
2063         idr_for_each_entry(&tconn->volumes, mdev, i) {
2064                 mdev->send_cnt = 0;
2065                 mdev->recv_cnt = 0;
2066                 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
2067         }
2068         rcu_read_unlock();
2069         conn_reconfig_done(tconn);
2070         drbd_adm_finish(info, retcode);
2071         return 0;
2072
2073 fail:
2074         kfree(int_dig_in);
2075         kfree(int_dig_vv);
2076         crypto_free_hash(tfm);
2077         crypto_free_hash(integrity_w_tfm);
2078         crypto_free_hash(integrity_r_tfm);
2079         kfree(new_conf);
2080
2081         conn_reconfig_done(tconn);
2082 out:
2083         drbd_adm_finish(info, retcode);
2084         return 0;
2085 }
2086
2087 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
2088 {
2089         enum drbd_state_rv rv;
2090         if (force) {
2091                 spin_lock_irq(&tconn->req_lock);
2092                 if (tconn->cstate >= C_WF_CONNECTION)
2093                         _conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
2094                 spin_unlock_irq(&tconn->req_lock);
2095                 return SS_SUCCESS;
2096         }
2097
2098         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING), 0);
2099
2100         switch (rv) {
2101         case SS_NOTHING_TO_DO:
2102         case SS_ALREADY_STANDALONE:
2103                 return SS_SUCCESS;
2104         case SS_PRIMARY_NOP:
2105                 /* Our state checking code wants to see the peer outdated. */
2106                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2107                                                         pdsk, D_OUTDATED), CS_VERBOSE);
2108                 break;
2109         case SS_CW_FAILED_BY_PEER:
2110                 /* The peer probably wants to see us outdated. */
2111                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2112                                                         disk, D_OUTDATED), 0);
2113                 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2114                         conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
2115                         rv = SS_SUCCESS;
2116                 }
2117                 break;
2118         default:;
2119                 /* no special handling necessary */
2120         }
2121
2122         return rv;
2123 }
2124
2125 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2126 {
2127         struct disconnect_parms parms;
2128         struct drbd_tconn *tconn;
2129         enum drbd_state_rv rv;
2130         enum drbd_ret_code retcode;
2131         int err;
2132
2133         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2134         if (!adm_ctx.reply_skb)
2135                 return retcode;
2136         if (retcode != NO_ERROR)
2137                 goto fail;
2138
2139         tconn = adm_ctx.tconn;
2140         memset(&parms, 0, sizeof(parms));
2141         if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2142                 err = disconnect_parms_from_attrs(&parms, info);
2143                 if (err) {
2144                         retcode = ERR_MANDATORY_TAG;
2145                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2146                         goto fail;
2147                 }
2148         }
2149
2150         rv = conn_try_disconnect(tconn, parms.force_disconnect);
2151         if (rv < SS_SUCCESS)
2152                 goto fail;
2153
2154         if (wait_event_interruptible(tconn->ping_wait,
2155                                      tconn->cstate != C_DISCONNECTING)) {
2156                 /* Do not test for mdev->state.conn == C_STANDALONE, since
2157                    someone else might connect us in the mean time! */
2158                 retcode = ERR_INTR;
2159                 goto fail;
2160         }
2161
2162         retcode = NO_ERROR;
2163  fail:
2164         drbd_adm_finish(info, retcode);
2165         return 0;
2166 }
2167
2168 void resync_after_online_grow(struct drbd_conf *mdev)
2169 {
2170         int iass; /* I am sync source */
2171
2172         dev_info(DEV, "Resync of new storage after online grow\n");
2173         if (mdev->state.role != mdev->state.peer)
2174                 iass = (mdev->state.role == R_PRIMARY);
2175         else
2176                 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
2177
2178         if (iass)
2179                 drbd_start_resync(mdev, C_SYNC_SOURCE);
2180         else
2181                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2182 }
2183
2184 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2185 {
2186         struct resize_parms rs;
2187         struct drbd_conf *mdev;
2188         enum drbd_ret_code retcode;
2189         enum determine_dev_size dd;
2190         enum dds_flags ddsf;
2191         int err;
2192
2193         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2194         if (!adm_ctx.reply_skb)
2195                 return retcode;
2196         if (retcode != NO_ERROR)
2197                 goto fail;
2198
2199         memset(&rs, 0, sizeof(struct resize_parms));
2200         if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2201                 err = resize_parms_from_attrs(&rs, info);
2202                 if (err) {
2203                         retcode = ERR_MANDATORY_TAG;
2204                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2205                         goto fail;
2206                 }
2207         }
2208
2209         mdev = adm_ctx.mdev;
2210         if (mdev->state.conn > C_CONNECTED) {
2211                 retcode = ERR_RESIZE_RESYNC;
2212                 goto fail;
2213         }
2214
2215         if (mdev->state.role == R_SECONDARY &&
2216             mdev->state.peer == R_SECONDARY) {
2217                 retcode = ERR_NO_PRIMARY;
2218                 goto fail;
2219         }
2220
2221         if (!get_ldev(mdev)) {
2222                 retcode = ERR_NO_DISK;
2223                 goto fail;
2224         }
2225
2226         if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
2227                 retcode = ERR_NEED_APV_93;
2228                 goto fail;
2229         }
2230
2231         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
2232                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2233
2234         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
2235         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2236         dd = drbd_determine_dev_size(mdev, ddsf);
2237         drbd_md_sync(mdev);
2238         put_ldev(mdev);
2239         if (dd == dev_size_error) {
2240                 retcode = ERR_NOMEM_BITMAP;
2241                 goto fail;
2242         }
2243
2244         if (mdev->state.conn == C_CONNECTED) {
2245                 if (dd == grew)
2246                         set_bit(RESIZE_PENDING, &mdev->flags);
2247
2248                 drbd_send_uuids(mdev);
2249                 drbd_send_sizes(mdev, 1, ddsf);
2250         }
2251
2252  fail:
2253         drbd_adm_finish(info, retcode);
2254         return 0;
2255 }
2256
2257 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2258 {
2259         enum drbd_ret_code retcode;
2260         cpumask_var_t new_cpu_mask;
2261         struct drbd_tconn *tconn;
2262         int *rs_plan_s = NULL;
2263         struct res_opts sc;
2264         int err;
2265
2266         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2267         if (!adm_ctx.reply_skb)
2268                 return retcode;
2269         if (retcode != NO_ERROR)
2270                 goto fail;
2271         tconn = adm_ctx.tconn;
2272
2273         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
2274                 retcode = ERR_NOMEM;
2275                 drbd_msg_put_info("unable to allocate cpumask");
2276                 goto fail;
2277         }
2278
2279         if (((struct drbd_genlmsghdr*)info->userhdr)->flags
2280                         & DRBD_GENL_F_SET_DEFAULTS) {
2281                 memset(&sc, 0, sizeof(struct res_opts));
2282                 sc.on_no_data  = DRBD_ON_NO_DATA_DEF;
2283         } else
2284                 sc = tconn->res_opts;
2285
2286         err = res_opts_from_attrs(&sc, info);
2287         if (err) {
2288                 retcode = ERR_MANDATORY_TAG;
2289                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2290                 goto fail;
2291         }
2292
2293         /* silently ignore cpu mask on UP kernel */
2294         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
2295                 err = __bitmap_parse(sc.cpu_mask, 32, 0,
2296                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2297                 if (err) {
2298                         conn_warn(tconn, "__bitmap_parse() failed with %d\n", err);
2299                         retcode = ERR_CPU_MASK_PARSE;
2300                         goto fail;
2301                 }
2302         }
2303
2304
2305         tconn->res_opts = sc;
2306
2307         if (!cpumask_equal(tconn->cpu_mask, new_cpu_mask)) {
2308                 cpumask_copy(tconn->cpu_mask, new_cpu_mask);
2309                 drbd_calc_cpu_mask(tconn);
2310                 tconn->receiver.reset_cpu_mask = 1;
2311                 tconn->asender.reset_cpu_mask = 1;
2312                 tconn->worker.reset_cpu_mask = 1;
2313         }
2314
2315 fail:
2316         kfree(rs_plan_s);
2317         free_cpumask_var(new_cpu_mask);
2318
2319         drbd_adm_finish(info, retcode);
2320         return 0;
2321 }
2322
2323 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2324 {
2325         struct drbd_conf *mdev;
2326         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2327
2328         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2329         if (!adm_ctx.reply_skb)
2330                 return retcode;
2331         if (retcode != NO_ERROR)
2332                 goto out;
2333
2334         mdev = adm_ctx.mdev;
2335
2336         /* If there is still bitmap IO pending, probably because of a previous
2337          * resync just being finished, wait for it before requesting a new resync. */
2338         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2339
2340         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2341
2342         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2343                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2344
2345         while (retcode == SS_NEED_CONNECTION) {
2346                 spin_lock_irq(&mdev->tconn->req_lock);
2347                 if (mdev->state.conn < C_CONNECTED)
2348                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2349                 spin_unlock_irq(&mdev->tconn->req_lock);
2350
2351                 if (retcode != SS_NEED_CONNECTION)
2352                         break;
2353
2354                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2355         }
2356
2357 out:
2358         drbd_adm_finish(info, retcode);
2359         return 0;
2360 }
2361
2362 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2363 {
2364         int rv;
2365
2366         rv = drbd_bmio_set_n_write(mdev);
2367         drbd_suspend_al(mdev);
2368         return rv;
2369 }
2370
2371 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2372                 union drbd_state mask, union drbd_state val)
2373 {
2374         enum drbd_ret_code retcode;
2375
2376         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2377         if (!adm_ctx.reply_skb)
2378                 return retcode;
2379         if (retcode != NO_ERROR)
2380                 goto out;
2381
2382         retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2383 out:
2384         drbd_adm_finish(info, retcode);
2385         return 0;
2386 }
2387
2388 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2389 {
2390         return drbd_adm_simple_request_state(skb, info, NS(conn, C_STARTING_SYNC_S));
2391 }
2392
2393 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2394 {
2395         enum drbd_ret_code retcode;
2396
2397         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2398         if (!adm_ctx.reply_skb)
2399                 return retcode;
2400         if (retcode != NO_ERROR)
2401                 goto out;
2402
2403         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2404                 retcode = ERR_PAUSE_IS_SET;
2405 out:
2406         drbd_adm_finish(info, retcode);
2407         return 0;
2408 }
2409
2410 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2411 {
2412         union drbd_dev_state s;
2413         enum drbd_ret_code retcode;
2414
2415         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2416         if (!adm_ctx.reply_skb)
2417                 return retcode;
2418         if (retcode != NO_ERROR)
2419                 goto out;
2420
2421         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2422                 s = adm_ctx.mdev->state;
2423                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2424                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2425                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2426                 } else {
2427                         retcode = ERR_PAUSE_IS_CLEAR;
2428                 }
2429         }
2430
2431 out:
2432         drbd_adm_finish(info, retcode);
2433         return 0;
2434 }
2435
2436 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2437 {
2438         return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2439 }
2440
2441 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2442 {
2443         struct drbd_conf *mdev;
2444         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2445
2446         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2447         if (!adm_ctx.reply_skb)
2448                 return retcode;
2449         if (retcode != NO_ERROR)
2450                 goto out;
2451
2452         mdev = adm_ctx.mdev;
2453         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2454                 drbd_uuid_new_current(mdev);
2455                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2456         }
2457         drbd_suspend_io(mdev);
2458         retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2459         if (retcode == SS_SUCCESS) {
2460                 if (mdev->state.conn < C_CONNECTED)
2461                         tl_clear(mdev->tconn);
2462                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2463                         tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2464         }
2465         drbd_resume_io(mdev);
2466
2467 out:
2468         drbd_adm_finish(info, retcode);
2469         return 0;
2470 }
2471
2472 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2473 {
2474         return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2475 }
2476
2477 int nla_put_drbd_cfg_context(struct sk_buff *skb, const char *conn_name, unsigned vnr)
2478 {
2479         struct nlattr *nla;
2480         nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2481         if (!nla)
2482                 goto nla_put_failure;
2483         if (vnr != VOLUME_UNSPECIFIED)
2484                 NLA_PUT_U32(skb, T_ctx_volume, vnr);
2485         NLA_PUT_STRING(skb, T_ctx_conn_name, conn_name);
2486         nla_nest_end(skb, nla);
2487         return 0;
2488
2489 nla_put_failure:
2490         if (nla)
2491                 nla_nest_cancel(skb, nla);
2492         return -EMSGSIZE;
2493 }
2494
2495 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2496                 const struct sib_info *sib)
2497 {
2498         struct state_info *si = NULL; /* for sizeof(si->member); */
2499         struct net_conf *nc;
2500         struct nlattr *nla;
2501         int got_ldev;
2502         int err = 0;
2503         int exclude_sensitive;
2504
2505         /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2506          * to.  So we better exclude_sensitive information.
2507          *
2508          * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2509          * in the context of the requesting user process. Exclude sensitive
2510          * information, unless current has superuser.
2511          *
2512          * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2513          * relies on the current implementation of netlink_dump(), which
2514          * executes the dump callback successively from netlink_recvmsg(),
2515          * always in the context of the receiving process */
2516         exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2517
2518         got_ldev = get_ldev(mdev);
2519
2520         /* We need to add connection name and volume number information still.
2521          * Minor number is in drbd_genlmsghdr. */
2522         if (nla_put_drbd_cfg_context(skb, mdev->tconn->name, mdev->vnr))
2523                 goto nla_put_failure;
2524
2525         if (res_opts_to_skb(skb, &mdev->tconn->res_opts, exclude_sensitive))
2526                 goto nla_put_failure;
2527
2528         if (got_ldev)
2529                 if (disk_conf_to_skb(skb, &mdev->ldev->dc, exclude_sensitive))
2530                         goto nla_put_failure;
2531
2532         rcu_read_lock();
2533         nc = rcu_dereference(mdev->tconn->net_conf);
2534         if (nc)
2535                 err = net_conf_to_skb(skb, nc, exclude_sensitive);
2536         rcu_read_unlock();
2537         if (err)
2538                 goto nla_put_failure;
2539
2540         nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2541         if (!nla)
2542                 goto nla_put_failure;
2543         NLA_PUT_U32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY);
2544         NLA_PUT_U32(skb, T_current_state, mdev->state.i);
2545         NLA_PUT_U64(skb, T_ed_uuid, mdev->ed_uuid);
2546         NLA_PUT_U64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev));
2547
2548         if (got_ldev) {
2549                 NLA_PUT_U32(skb, T_disk_flags, mdev->ldev->md.flags);
2550                 NLA_PUT(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2551                 NLA_PUT_U64(skb, T_bits_total, drbd_bm_bits(mdev));
2552                 NLA_PUT_U64(skb, T_bits_oos, drbd_bm_total_weight(mdev));
2553                 if (C_SYNC_SOURCE <= mdev->state.conn &&
2554                     C_PAUSED_SYNC_T >= mdev->state.conn) {
2555                         NLA_PUT_U64(skb, T_bits_rs_total, mdev->rs_total);
2556                         NLA_PUT_U64(skb, T_bits_rs_failed, mdev->rs_failed);
2557                 }
2558         }
2559
2560         if (sib) {
2561                 switch(sib->sib_reason) {
2562                 case SIB_SYNC_PROGRESS:
2563                 case SIB_GET_STATUS_REPLY:
2564                         break;
2565                 case SIB_STATE_CHANGE:
2566                         NLA_PUT_U32(skb, T_prev_state, sib->os.i);
2567                         NLA_PUT_U32(skb, T_new_state, sib->ns.i);
2568                         break;
2569                 case SIB_HELPER_POST:
2570                         NLA_PUT_U32(skb,
2571                                 T_helper_exit_code, sib->helper_exit_code);
2572                         /* fall through */
2573                 case SIB_HELPER_PRE:
2574                         NLA_PUT_STRING(skb, T_helper, sib->helper_name);
2575                         break;
2576                 }
2577         }
2578         nla_nest_end(skb, nla);
2579
2580         if (0)
2581 nla_put_failure:
2582                 err = -EMSGSIZE;
2583         if (got_ldev)
2584                 put_ldev(mdev);
2585         return err;
2586 }
2587
2588 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2589 {
2590         enum drbd_ret_code retcode;
2591         int err;
2592
2593         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2594         if (!adm_ctx.reply_skb)
2595                 return retcode;
2596         if (retcode != NO_ERROR)
2597                 goto out;
2598
2599         err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2600         if (err) {
2601                 nlmsg_free(adm_ctx.reply_skb);
2602                 return err;
2603         }
2604 out:
2605         drbd_adm_finish(info, retcode);
2606         return 0;
2607 }
2608
2609 int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
2610 {
2611         struct drbd_conf *mdev;
2612         struct drbd_genlmsghdr *dh;
2613         struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2614         struct drbd_tconn *tconn = NULL;
2615         struct drbd_tconn *tmp;
2616         unsigned volume = cb->args[1];
2617
2618         /* Open coded, deferred, iteration:
2619          * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2620          *      idr_for_each_entry(&tconn->volumes, mdev, i) {
2621          *        ...
2622          *      }
2623          * }
2624          * where tconn is cb->args[0];
2625          * and i is cb->args[1];
2626          *
2627          * cb->args[2] indicates if we shall loop over all resources,
2628          * or just dump all volumes of a single resource.
2629          *
2630          * This may miss entries inserted after this dump started,
2631          * or entries deleted before they are reached.
2632          *
2633          * We need to make sure the mdev won't disappear while
2634          * we are looking at it, and revalidate our iterators
2635          * on each iteration.
2636          */
2637
2638         /* synchronize with conn_create()/conn_destroy() */
2639         down_read(&drbd_cfg_rwsem);
2640         /* revalidate iterator position */
2641         list_for_each_entry(tmp, &drbd_tconns, all_tconn) {
2642                 if (pos == NULL) {
2643                         /* first iteration */
2644                         pos = tmp;
2645                         tconn = pos;
2646                         break;
2647                 }
2648                 if (tmp == pos) {
2649                         tconn = pos;
2650                         break;
2651                 }
2652         }
2653         if (tconn) {
2654 next_tconn:
2655                 mdev = idr_get_next(&tconn->volumes, &volume);
2656                 if (!mdev) {
2657                         /* No more volumes to dump on this tconn.
2658                          * Advance tconn iterator. */
2659                         pos = list_entry(tconn->all_tconn.next,
2660                                         struct drbd_tconn, all_tconn);
2661                         /* Did we dump any volume on this tconn yet? */
2662                         if (volume != 0) {
2663                                 /* If we reached the end of the list,
2664                                  * or only a single resource dump was requested,
2665                                  * we are done. */
2666                                 if (&pos->all_tconn == &drbd_tconns || cb->args[2])
2667                                         goto out;
2668                                 volume = 0;
2669                                 tconn = pos;
2670                                 goto next_tconn;
2671                         }
2672                 }
2673
2674                 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).pid,
2675                                 cb->nlh->nlmsg_seq, &drbd_genl_family,
2676                                 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2677                 if (!dh)
2678                         goto out;
2679
2680                 if (!mdev) {
2681                         /* this is a tconn without a single volume */
2682                         dh->minor = -1U;
2683                         dh->ret_code = NO_ERROR;
2684                         if (nla_put_drbd_cfg_context(skb, tconn->name, VOLUME_UNSPECIFIED))
2685                                 genlmsg_cancel(skb, dh);
2686                         else
2687                                 genlmsg_end(skb, dh);
2688                         goto out;
2689                 }
2690
2691                 D_ASSERT(mdev->vnr == volume);
2692                 D_ASSERT(mdev->tconn == tconn);
2693
2694                 dh->minor = mdev_to_minor(mdev);
2695                 dh->ret_code = NO_ERROR;
2696
2697                 if (nla_put_status_info(skb, mdev, NULL)) {
2698                         genlmsg_cancel(skb, dh);
2699                         goto out;
2700                 }
2701                 genlmsg_end(skb, dh);
2702         }
2703
2704 out:
2705         up_read(&drbd_cfg_rwsem);
2706         /* where to start the next iteration */
2707         cb->args[0] = (long)pos;
2708         cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2709
2710         /* No more tconns/volumes/minors found results in an empty skb.
2711          * Which will terminate the dump. */
2712         return skb->len;
2713 }
2714
2715 /*
2716  * Request status of all resources, or of all volumes within a single resource.
2717  *
2718  * This is a dump, as the answer may not fit in a single reply skb otherwise.
2719  * Which means we cannot use the family->attrbuf or other such members, because
2720  * dump is NOT protected by the genl_lock().  During dump, we only have access
2721  * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
2722  *
2723  * Once things are setup properly, we call into get_one_status().
2724  */
2725 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2726 {
2727         const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
2728         struct nlattr *nla;
2729         const char *conn_name;
2730         struct drbd_tconn *tconn;
2731
2732         /* Is this a followup call? */
2733         if (cb->args[0]) {
2734                 /* ... of a single resource dump,
2735                  * and the resource iterator has been advanced already? */
2736                 if (cb->args[2] && cb->args[2] != cb->args[0])
2737                         return 0; /* DONE. */
2738                 goto dump;
2739         }
2740
2741         /* First call (from netlink_dump_start).  We need to figure out
2742          * which resource(s) the user wants us to dump. */
2743         nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
2744                         nlmsg_attrlen(cb->nlh, hdrlen),
2745                         DRBD_NLA_CFG_CONTEXT);
2746
2747         /* No explicit context given.  Dump all. */
2748         if (!nla)
2749                 goto dump;
2750         nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
2751         /* context given, but no name present? */
2752         if (!nla)
2753                 return -EINVAL;
2754         conn_name = nla_data(nla);
2755         tconn = conn_get_by_name(conn_name);
2756
2757         if (!tconn)
2758                 return -ENODEV;
2759
2760         kref_put(&tconn->kref, &conn_destroy); /* get_one_status() (re)validates tconn by itself */
2761
2762         /* prime iterators, and set "filter" mode mark:
2763          * only dump this tconn. */
2764         cb->args[0] = (long)tconn;
2765         /* cb->args[1] = 0; passed in this way. */
2766         cb->args[2] = (long)tconn;
2767
2768 dump:
2769         return get_one_status(skb, cb);
2770 }
2771
2772 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2773 {
2774         enum drbd_ret_code retcode;
2775         struct timeout_parms tp;
2776         int err;
2777
2778         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2779         if (!adm_ctx.reply_skb)
2780                 return retcode;
2781         if (retcode != NO_ERROR)
2782                 goto out;
2783
2784         tp.timeout_type =
2785                 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2786                 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2787                 UT_DEFAULT;
2788
2789         err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2790         if (err) {
2791                 nlmsg_free(adm_ctx.reply_skb);
2792                 return err;
2793         }
2794 out:
2795         drbd_adm_finish(info, retcode);
2796         return 0;
2797 }
2798
2799 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2800 {
2801         struct drbd_conf *mdev;
2802         enum drbd_ret_code retcode;
2803
2804         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2805         if (!adm_ctx.reply_skb)
2806                 return retcode;
2807         if (retcode != NO_ERROR)
2808                 goto out;
2809
2810         mdev = adm_ctx.mdev;
2811         if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2812                 /* resume from last known position, if possible */
2813                 struct start_ov_parms parms =
2814                         { .ov_start_sector = mdev->ov_start_sector };
2815                 int err = start_ov_parms_from_attrs(&parms, info);
2816                 if (err) {
2817                         retcode = ERR_MANDATORY_TAG;
2818                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2819                         goto out;
2820                 }
2821                 /* w_make_ov_request expects position to be aligned */
2822                 mdev->ov_start_sector = parms.ov_start_sector & ~BM_SECT_PER_BIT;
2823         }
2824         /* If there is still bitmap IO pending, e.g. previous resync or verify
2825          * just being finished, wait for it before requesting a new resync. */
2826         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2827         retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2828 out:
2829         drbd_adm_finish(info, retcode);
2830         return 0;
2831 }
2832
2833
2834 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
2835 {
2836         struct drbd_conf *mdev;
2837         enum drbd_ret_code retcode;
2838         int skip_initial_sync = 0;
2839         int err;
2840         struct new_c_uuid_parms args;
2841
2842         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2843         if (!adm_ctx.reply_skb)
2844                 return retcode;
2845         if (retcode != NO_ERROR)
2846                 goto out_nolock;
2847
2848         mdev = adm_ctx.mdev;
2849         memset(&args, 0, sizeof(args));
2850         if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
2851                 err = new_c_uuid_parms_from_attrs(&args, info);
2852                 if (err) {
2853                         retcode = ERR_MANDATORY_TAG;
2854                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2855                         goto out_nolock;
2856                 }
2857         }
2858
2859         mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2860
2861         if (!get_ldev(mdev)) {
2862                 retcode = ERR_NO_DISK;
2863                 goto out;
2864         }
2865
2866         /* this is "skip initial sync", assume to be clean */
2867         if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2868             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2869                 dev_info(DEV, "Preparing to skip initial sync\n");
2870                 skip_initial_sync = 1;
2871         } else if (mdev->state.conn != C_STANDALONE) {
2872                 retcode = ERR_CONNECTED;
2873                 goto out_dec;
2874         }
2875
2876         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2877         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2878
2879         if (args.clear_bm) {
2880                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2881                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2882                 if (err) {
2883                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2884                         retcode = ERR_IO_MD_DISK;
2885                 }
2886                 if (skip_initial_sync) {
2887                         drbd_send_uuids_skip_initial_sync(mdev);
2888                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2889                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2890                         spin_lock_irq(&mdev->tconn->req_lock);
2891                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2892                                         CS_VERBOSE, NULL);
2893                         spin_unlock_irq(&mdev->tconn->req_lock);
2894                 }
2895         }
2896
2897         drbd_md_sync(mdev);
2898 out_dec:
2899         put_ldev(mdev);
2900 out:
2901         mutex_unlock(mdev->state_mutex);
2902 out_nolock:
2903         drbd_adm_finish(info, retcode);
2904         return 0;
2905 }
2906
2907 static enum drbd_ret_code
2908 drbd_check_conn_name(const char *name)
2909 {
2910         if (!name || !name[0]) {
2911                 drbd_msg_put_info("connection name missing");
2912                 return ERR_MANDATORY_TAG;
2913         }
2914         /* if we want to use these in sysfs/configfs/debugfs some day,
2915          * we must not allow slashes */
2916         if (strchr(name, '/')) {
2917                 drbd_msg_put_info("invalid connection name");
2918                 return ERR_INVALID_REQUEST;
2919         }
2920         return NO_ERROR;
2921 }
2922
2923 int drbd_adm_create_connection(struct sk_buff *skb, struct genl_info *info)
2924 {
2925         enum drbd_ret_code retcode;
2926
2927         retcode = drbd_adm_prepare(skb, info, 0);
2928         if (!adm_ctx.reply_skb)
2929                 return retcode;
2930         if (retcode != NO_ERROR)
2931                 goto out;
2932
2933         retcode = drbd_check_conn_name(adm_ctx.conn_name);
2934         if (retcode != NO_ERROR)
2935                 goto out;
2936
2937         if (adm_ctx.tconn) {
2938                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
2939                         retcode = ERR_INVALID_REQUEST;
2940                         drbd_msg_put_info("connection exists");
2941                 }
2942                 /* else: still NO_ERROR */
2943                 goto out;
2944         }
2945
2946         if (!conn_create(adm_ctx.conn_name))
2947                 retcode = ERR_NOMEM;
2948 out:
2949         drbd_adm_finish(info, retcode);
2950         return 0;
2951 }
2952
2953 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
2954 {
2955         struct drbd_genlmsghdr *dh = info->userhdr;
2956         enum drbd_ret_code retcode;
2957
2958         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2959         if (!adm_ctx.reply_skb)
2960                 return retcode;
2961         if (retcode != NO_ERROR)
2962                 goto out;
2963
2964         /* FIXME drop minor_count parameter, limit to MINORMASK */
2965         if (dh->minor >= minor_count) {
2966                 drbd_msg_put_info("requested minor out of range");
2967                 retcode = ERR_INVALID_REQUEST;
2968                 goto out;
2969         }
2970         if (adm_ctx.volume > DRBD_VOLUME_MAX) {
2971                 drbd_msg_put_info("requested volume id out of range");
2972                 retcode = ERR_INVALID_REQUEST;
2973                 goto out;
2974         }
2975
2976         /* drbd_adm_prepare made sure already
2977          * that mdev->tconn and mdev->vnr match the request. */
2978         if (adm_ctx.mdev) {
2979                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
2980                         retcode = ERR_MINOR_EXISTS;
2981                 /* else: still NO_ERROR */
2982                 goto out;
2983         }
2984
2985         down_write(&drbd_cfg_rwsem);
2986         retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
2987         up_write(&drbd_cfg_rwsem);
2988 out:
2989         drbd_adm_finish(info, retcode);
2990         return 0;
2991 }
2992
2993 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
2994 {
2995         if (mdev->state.disk == D_DISKLESS &&
2996             /* no need to be mdev->state.conn == C_STANDALONE &&
2997              * we may want to delete a minor from a live replication group.
2998              */
2999             mdev->state.role == R_SECONDARY) {
3000                 drbd_delete_device(mdev);
3001                 return NO_ERROR;
3002         } else
3003                 return ERR_MINOR_CONFIGURED;
3004 }
3005
3006 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
3007 {
3008         enum drbd_ret_code retcode;
3009
3010         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3011         if (!adm_ctx.reply_skb)
3012                 return retcode;
3013         if (retcode != NO_ERROR)
3014                 goto out;
3015
3016         down_write(&drbd_cfg_rwsem);
3017         retcode = adm_delete_minor(adm_ctx.mdev);
3018         up_write(&drbd_cfg_rwsem);
3019 out:
3020         drbd_adm_finish(info, retcode);
3021         return 0;
3022 }
3023
3024 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
3025 {
3026         enum drbd_ret_code retcode;
3027         enum drbd_state_rv rv;
3028         struct drbd_conf *mdev;
3029         unsigned i;
3030
3031         retcode = drbd_adm_prepare(skb, info, 0);
3032         if (!adm_ctx.reply_skb)
3033                 return retcode;
3034         if (retcode != NO_ERROR)
3035                 goto out;
3036
3037         if (!adm_ctx.tconn) {
3038                 retcode = ERR_CONN_NOT_KNOWN;
3039                 goto out;
3040         }
3041
3042         down_read(&drbd_cfg_rwsem);
3043         /* demote */
3044         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3045                 retcode = drbd_set_role(mdev, R_SECONDARY, 0);
3046                 if (retcode < SS_SUCCESS) {
3047                         drbd_msg_put_info("failed to demote");
3048                         goto out_unlock;
3049                 }
3050         }
3051
3052         /* disconnect */
3053         rv = conn_try_disconnect(adm_ctx.tconn, 0);
3054         if (rv < SS_SUCCESS) {
3055                 retcode = rv; /* enum type mismatch! */
3056                 drbd_msg_put_info("failed to disconnect");
3057                 goto out_unlock;
3058         }
3059
3060         /* detach */
3061         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3062                 rv = adm_detach(mdev);
3063                 if (rv < SS_SUCCESS) {
3064                         retcode = rv; /* enum type mismatch! */
3065                         drbd_msg_put_info("failed to detach");
3066                         goto out_unlock;
3067                 }
3068         }
3069         up_read(&drbd_cfg_rwsem);
3070
3071         /* delete volumes */
3072         down_write(&drbd_cfg_rwsem);
3073         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3074                 retcode = adm_delete_minor(mdev);
3075                 if (retcode != NO_ERROR) {
3076                         /* "can not happen" */
3077                         drbd_msg_put_info("failed to delete volume");
3078                         up_write(&drbd_cfg_rwsem);
3079                         goto out;
3080                 }
3081         }
3082
3083         /* stop all threads */
3084         conn_reconfig_done(adm_ctx.tconn);
3085
3086         /* delete connection */
3087         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3088                 list_del(&adm_ctx.tconn->all_tconn);
3089                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3090
3091                 retcode = NO_ERROR;
3092         } else {
3093                 /* "can not happen" */
3094                 retcode = ERR_CONN_IN_USE;
3095                 drbd_msg_put_info("failed to delete connection");
3096         }
3097
3098         up_write(&drbd_cfg_rwsem);
3099         goto out;
3100 out_unlock:
3101         up_read(&drbd_cfg_rwsem);
3102 out:
3103         drbd_adm_finish(info, retcode);
3104         return 0;
3105 }
3106
3107 int drbd_adm_delete_connection(struct sk_buff *skb, struct genl_info *info)
3108 {
3109         enum drbd_ret_code retcode;
3110
3111         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
3112         if (!adm_ctx.reply_skb)
3113                 return retcode;
3114         if (retcode != NO_ERROR)
3115                 goto out;
3116
3117         down_write(&drbd_cfg_rwsem);
3118         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3119                 list_del(&adm_ctx.tconn->all_tconn);
3120                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3121
3122                 retcode = NO_ERROR;
3123         } else {
3124                 retcode = ERR_CONN_IN_USE;
3125         }
3126         up_write(&drbd_cfg_rwsem);
3127
3128 out:
3129         drbd_adm_finish(info, retcode);
3130         return 0;
3131 }
3132
3133 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
3134 {
3135         static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3136         struct sk_buff *msg;
3137         struct drbd_genlmsghdr *d_out;
3138         unsigned seq;
3139         int err = -ENOMEM;
3140
3141         seq = atomic_inc_return(&drbd_genl_seq);
3142         msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3143         if (!msg)
3144                 goto failed;
3145
3146         err = -EMSGSIZE;
3147         d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3148         if (!d_out) /* cannot happen, but anyways. */
3149                 goto nla_put_failure;
3150         d_out->minor = mdev_to_minor(mdev);
3151         d_out->ret_code = 0;
3152
3153         if (nla_put_status_info(msg, mdev, sib))
3154                 goto nla_put_failure;
3155         genlmsg_end(msg, d_out);
3156         err = drbd_genl_multicast_events(msg, 0);
3157         /* msg has been consumed or freed in netlink_broadcast() */
3158         if (err && err != -ESRCH)
3159                 goto failed;
3160
3161         return;
3162
3163 nla_put_failure:
3164         nlmsg_free(msg);
3165 failed:
3166         dev_err(DEV, "Error %d while broadcasting event. "
3167                         "Event seq:%u sib_reason:%u\n",
3168                         err, seq, sib->sib_reason);
3169 }