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