drivers/block/drbd/drbd_nl.c

   1 /*
   2    drbd_nl.c
   3
   4    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
   5
   6    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
   7    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
   8    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
   9
  10    drbd is free software; you can redistribute it and/or modify
  11    it under the terms of the GNU General Public License as published by
  12    the Free Software Foundation; either version 2, or (at your option)
  13    any later version.
  14
  15    drbd is distributed in the hope that it will be useful,
  16    but WITHOUT ANY WARRANTY; without even the implied warranty of
  17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18    GNU General Public License for more details.
  19
  20    You should have received a copy of the GNU General Public License
  21    along with drbd; see the file COPYING.  If not, write to
  22    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  23
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/drbd.h>
  28 #include <linux/in.h>
  29 #include <linux/fs.h>
  30 #include <linux/file.h>
  31 #include <linux/slab.h>
  32 #include <linux/blkpg.h>
  33 #include <linux/cpumask.h>
  34 #include "drbd_int.h"
  35 #include "drbd_req.h"
  36 #include "drbd_wrappers.h"
  37 #include <asm/unaligned.h>
  38 #include <linux/drbd_limits.h>
  39 #include <linux/kthread.h>
  40
  41 #include <net/genetlink.h>
  42
  43 /* .doit */
  44 // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info);
  45 // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info);
  46
  47 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info);
  48 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info);
  49
  50 int drbd_adm_create_connection(struct sk_buff *skb, struct genl_info *info);
  51 int drbd_adm_delete_connection(struct sk_buff *skb, struct genl_info *info);
  52 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info);
  53
  54 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info);
  55 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info);
  56 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info);
  57 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info);
  58 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info);
  59 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info);
  60 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info);
  61 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info);
  62 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info);
  63 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info);
  64 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info);
  65 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info);
  66 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info);
  67 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info);
  68 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info);
  69 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info);
  70 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info);
  71 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info);
  72 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info);
  73 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info);
  74 /* .dumpit */
  75 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb);
  76
  77 #include <linux/drbd_genl_api.h>
  78 #include <linux/genl_magic_func.h>
  79
  80 /* used blkdev_get_by_path, to claim our meta data device(s) */
  81 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
  82
  83 /* Configuration is strictly serialized, because generic netlink message
  84  * processing is strictly serialized by the genl_lock().
  85  * Which means we can use one static global drbd_config_context struct.
  86  */
  87 static struct drbd_config_context {
  88         /* assigned from drbd_genlmsghdr */
  89         unsigned int minor;
  90         /* assigned from request attributes, if present */
  91         unsigned int volume;
  92 #define VOLUME_UNSPECIFIED              (-1U)
  93         /* pointer into the request skb,
  94          * limited lifetime! */
  95         char *conn_name;
  96
  97         /* reply buffer */
  98         struct sk_buff *reply_skb;
  99         /* pointer into reply buffer */
 100         struct drbd_genlmsghdr *reply_dh;
 101         /* resolved from attributes, if possible */
 102         struct drbd_conf *mdev;
 103         struct drbd_tconn *tconn;
 104 } adm_ctx;
 105
 106 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
 107 {
 108         genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
 109         if (genlmsg_reply(skb, info))
 110                 printk(KERN_ERR "drbd: error sending genl reply\n");
 111 }
 112
 113 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
 114  * reason it could fail was no space in skb, and there are 4k available. */
 115 int drbd_msg_put_info(const char *info)
 116 {
 117         struct sk_buff *skb = adm_ctx.reply_skb;
 118         struct nlattr *nla;
 119         int err = -EMSGSIZE;
 120
 121         if (!info || !info[0])
 122                 return 0;
 123
 124         nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY);
 125         if (!nla)
 126                 return err;
 127
 128         err = nla_put_string(skb, T_info_text, info);
 129         if (err) {
 130                 nla_nest_cancel(skb, nla);
 131                 return err;
 132         } else
 133                 nla_nest_end(skb, nla);
 134         return 0;
 135 }
 136
 137 /* This would be a good candidate for a "pre_doit" hook,
 138  * and per-family private info->pointers.
 139  * But we need to stay compatible with older kernels.
 140  * If it returns successfully, adm_ctx members are valid.
 141  */
 142 #define DRBD_ADM_NEED_MINOR     1
 143 #define DRBD_ADM_NEED_CONN      2
 144 static int drbd_adm_prepare(struct sk_buff *skb, struct genl_info *info,
 145                 unsigned flags)
 146 {
 147         struct drbd_genlmsghdr *d_in = info->userhdr;
 148         const u8 cmd = info->genlhdr->cmd;
 149         int err;
 150
 151         memset(&adm_ctx, 0, sizeof(adm_ctx));
 152
 153         /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
 154         if (cmd != DRBD_ADM_GET_STATUS
 155         && security_netlink_recv(skb, CAP_SYS_ADMIN))
 156                return -EPERM;
 157
 158         adm_ctx.reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
 159         if (!adm_ctx.reply_skb)
 160                 goto fail;
 161
 162         adm_ctx.reply_dh = genlmsg_put_reply(adm_ctx.reply_skb,
 163                                         info, &drbd_genl_family, 0, cmd);
 164         /* put of a few bytes into a fresh skb of >= 4k will always succeed.
 165          * but anyways */
 166         if (!adm_ctx.reply_dh)
 167                 goto fail;
 168
 169         adm_ctx.reply_dh->minor = d_in->minor;
 170         adm_ctx.reply_dh->ret_code = NO_ERROR;
 171
 172         if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
 173                 struct nlattr *nla;
 174                 /* parse and validate only */
 175                 err = drbd_cfg_context_from_attrs(NULL, info);
 176                 if (err)
 177                         goto fail;
 178
 179                 /* It was present, and valid,
 180                  * copy it over to the reply skb. */
 181                 err = nla_put_nohdr(adm_ctx.reply_skb,
 182                                 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
 183                                 info->attrs[DRBD_NLA_CFG_CONTEXT]);
 184                 if (err)
 185                         goto fail;
 186
 187                 /* and assign stuff to the global adm_ctx */
 188                 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
 189                 adm_ctx.volume = nla ? nla_get_u32(nla) : VOLUME_UNSPECIFIED;
 190                 nla = nested_attr_tb[__nla_type(T_ctx_conn_name)];
 191                 if (nla)
 192                         adm_ctx.conn_name = nla_data(nla);
 193         } else
 194                 adm_ctx.volume = VOLUME_UNSPECIFIED;
 195
 196         adm_ctx.minor = d_in->minor;
 197         adm_ctx.mdev = minor_to_mdev(d_in->minor);
 198         adm_ctx.tconn = conn_get_by_name(adm_ctx.conn_name);
 199
 200         if (!adm_ctx.mdev && (flags & DRBD_ADM_NEED_MINOR)) {
 201                 drbd_msg_put_info("unknown minor");
 202                 return ERR_MINOR_INVALID;
 203         }
 204         if (!adm_ctx.tconn && (flags & DRBD_ADM_NEED_CONN)) {
 205                 drbd_msg_put_info("unknown connection");
 206                 return ERR_INVALID_REQUEST;
 207         }
 208
 209         /* some more paranoia, if the request was over-determined */
 210         if (adm_ctx.mdev && adm_ctx.tconn &&
 211             adm_ctx.mdev->tconn != adm_ctx.tconn) {
 212                 pr_warning("request: minor=%u, conn=%s; but that minor belongs to connection %s\n",
 213                                 adm_ctx.minor, adm_ctx.conn_name, adm_ctx.mdev->tconn->name);
 214                 drbd_msg_put_info("minor exists in different connection");
 215                 return ERR_INVALID_REQUEST;
 216         }
 217         if (adm_ctx.mdev &&
 218             adm_ctx.volume != VOLUME_UNSPECIFIED &&
 219             adm_ctx.volume != adm_ctx.mdev->vnr) {
 220                 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
 221                                 adm_ctx.minor, adm_ctx.volume,
 222                                 adm_ctx.mdev->vnr, adm_ctx.mdev->tconn->name);
 223                 drbd_msg_put_info("minor exists as different volume");
 224                 return ERR_INVALID_REQUEST;
 225         }
 226
 227         return NO_ERROR;
 228
 229 fail:
 230         nlmsg_free(adm_ctx.reply_skb);
 231         adm_ctx.reply_skb = NULL;
 232         return -ENOMEM;
 233 }
 234
 235 static int drbd_adm_finish(struct genl_info *info, int retcode)
 236 {
 237         struct nlattr *nla;
 238         const char *conn_name = NULL;
 239
 240         if (adm_ctx.tconn) {
 241                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
 242                 adm_ctx.tconn = NULL;
 243         }
 244
 245         if (!adm_ctx.reply_skb)
 246                 return -ENOMEM;
 247
 248         adm_ctx.reply_dh->ret_code = retcode;
 249
 250         nla = info->attrs[DRBD_NLA_CFG_CONTEXT];
 251         if (nla) {
 252                 nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
 253                 if (nla)
 254                         conn_name = nla_data(nla);
 255         }
 256
 257         drbd_adm_send_reply(adm_ctx.reply_skb, info);
 258         return 0;
 259 }
 260
 261 static void setup_khelper_env(struct drbd_tconn *tconn, char **envp)
 262 {
 263         char *afs;
 264         struct net_conf *nc;
 265
 266         rcu_read_lock();
 267         nc = rcu_dereference(tconn->net_conf);
 268         if (nc) {
 269                 switch (((struct sockaddr *)nc->peer_addr)->sa_family) {
 270                 case AF_INET6:
 271                         afs = "ipv6";
 272                         snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
 273                                  &((struct sockaddr_in6 *)nc->peer_addr)->sin6_addr);
 274                         break;
 275                 case AF_INET:
 276                         afs = "ipv4";
 277                         snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
 278                                  &((struct sockaddr_in *)nc->peer_addr)->sin_addr);
 279                         break;
 280                 default:
 281                         afs = "ssocks";
 282                         snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
 283                                  &((struct sockaddr_in *)nc->peer_addr)->sin_addr);
 284                 }
 285                 snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
 286         }
 287         rcu_read_unlock();
 288 }
 289
 290 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
 291 {
 292         char *envp[] = { "HOME=/",
 293                         "TERM=linux",
 294                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 295                          (char[20]) { }, /* address family */
 296                          (char[60]) { }, /* address */
 297                         NULL };
 298         char mb[12];
 299         char *argv[] = {usermode_helper, cmd, mb, NULL };
 300         struct sib_info sib;
 301         int ret;
 302
 303         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 304         setup_khelper_env(mdev->tconn, envp);
 305
 306         /* The helper may take some time.
 307          * write out any unsynced meta data changes now */
 308         drbd_md_sync(mdev);
 309
 310         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 311         sib.sib_reason = SIB_HELPER_PRE;
 312         sib.helper_name = cmd;
 313         drbd_bcast_event(mdev, &sib);
 314         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 315         if (ret)
 316                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 317                                 usermode_helper, cmd, mb,
 318                                 (ret >> 8) & 0xff, ret);
 319         else
 320                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 321                                 usermode_helper, cmd, mb,
 322                                 (ret >> 8) & 0xff, ret);
 323         sib.sib_reason = SIB_HELPER_POST;
 324         sib.helper_exit_code = ret;
 325         drbd_bcast_event(mdev, &sib);
 326
 327         if (ret < 0) /* Ignore any ERRNOs we got. */
 328                 ret = 0;
 329
 330         return ret;
 331 }
 332
 333 static void conn_md_sync(struct drbd_tconn *tconn)
 334 {
 335         struct drbd_conf *mdev;
 336         int vnr;
 337
 338         rcu_read_lock();
 339         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
 340                 kref_get(&mdev->kref);
 341                 rcu_read_unlock();
 342                 drbd_md_sync(mdev);
 343                 kref_put(&mdev->kref, &drbd_minor_destroy);
 344                 rcu_read_lock();
 345         }
 346         rcu_read_unlock();
 347 }
 348
 349 int conn_khelper(struct drbd_tconn *tconn, char *cmd)
 350 {
 351         char *envp[] = { "HOME=/",
 352                         "TERM=linux",
 353                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 354                          (char[20]) { }, /* address family */
 355                          (char[60]) { }, /* address */
 356                         NULL };
 357         char *argv[] = {usermode_helper, cmd, tconn->name, NULL };
 358         int ret;
 359
 360         setup_khelper_env(tconn, envp);
 361         conn_md_sync(tconn);
 362
 363         conn_info(tconn, "helper command: %s %s %s\n", usermode_helper, cmd, tconn->name);
 364         /* TODO: conn_bcast_event() ?? */
 365
 366         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 367         if (ret)
 368                 conn_warn(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
 369                           usermode_helper, cmd, tconn->name,
 370                           (ret >> 8) & 0xff, ret);
 371         else
 372                 conn_info(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
 373                           usermode_helper, cmd, tconn->name,
 374                           (ret >> 8) & 0xff, ret);
 375         /* TODO: conn_bcast_event() ?? */
 376
 377         if (ret < 0) /* Ignore any ERRNOs we got. */
 378                 ret = 0;
 379
 380         return ret;
 381 }
 382
 383 static enum drbd_fencing_p highest_fencing_policy(struct drbd_tconn *tconn)
 384 {
 385         enum drbd_fencing_p fp = FP_NOT_AVAIL;
 386         struct drbd_conf *mdev;
 387         int vnr;
 388
 389         rcu_read_lock();
 390         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
 391                 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 392                         fp = max_t(enum drbd_fencing_p, fp,
 393                                    rcu_dereference(mdev->ldev->disk_conf)->fencing);
 394                         put_ldev(mdev);
 395                 }
 396         }
 397         rcu_read_unlock();
 398
 399         return fp;
 400 }
 401
 402 bool conn_try_outdate_peer(struct drbd_tconn *tconn)
 403 {
 404         union drbd_state mask = { };
 405         union drbd_state val = { };
 406         enum drbd_fencing_p fp;
 407         char *ex_to_string;
 408         int r;
 409
 410         if (tconn->cstate >= C_WF_REPORT_PARAMS) {
 411                 conn_err(tconn, "Expected cstate < C_WF_REPORT_PARAMS\n");
 412                 return false;
 413         }
 414
 415         fp = highest_fencing_policy(tconn);
 416         switch (fp) {
 417         case FP_NOT_AVAIL:
 418                 conn_warn(tconn, "Not fencing peer, I'm not even Consistent myself.\n");
 419                 goto out;
 420         case FP_DONT_CARE:
 421                 return true;
 422         default: ;
 423         }
 424
 425         r = conn_khelper(tconn, "fence-peer");
 426
 427         switch ((r>>8) & 0xff) {
 428         case 3: /* peer is inconsistent */
 429                 ex_to_string = "peer is inconsistent or worse";
 430                 mask.pdsk = D_MASK;
 431                 val.pdsk = D_INCONSISTENT;
 432                 break;
 433         case 4: /* peer got outdated, or was already outdated */
 434                 ex_to_string = "peer was fenced";
 435                 mask.pdsk = D_MASK;
 436                 val.pdsk = D_OUTDATED;
 437                 break;
 438         case 5: /* peer was down */
 439                 if (conn_highest_disk(tconn) == D_UP_TO_DATE) {
 440                         /* we will(have) create(d) a new UUID anyways... */
 441                         ex_to_string = "peer is unreachable, assumed to be dead";
 442                         mask.pdsk = D_MASK;
 443                         val.pdsk = D_OUTDATED;
 444                 } else {
 445                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 446                 }
 447                 break;
 448         case 6: /* Peer is primary, voluntarily outdate myself.
 449                  * This is useful when an unconnected R_SECONDARY is asked to
 450                  * become R_PRIMARY, but finds the other peer being active. */
 451                 ex_to_string = "peer is active";
 452                 conn_warn(tconn, "Peer is primary, outdating myself.\n");
 453                 mask.disk = D_MASK;
 454                 val.disk = D_OUTDATED;
 455                 break;
 456         case 7:
 457                 if (fp != FP_STONITH)
 458                         conn_err(tconn, "fence-peer() = 7 && fencing != Stonith !!!\n");
 459                 ex_to_string = "peer was stonithed";
 460                 mask.pdsk = D_MASK;
 461                 val.pdsk = D_OUTDATED;
 462                 break;
 463         default:
 464                 /* The script is broken ... */
 465                 conn_err(tconn, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 466                 return false; /* Eventually leave IO frozen */
 467         }
 468
 469         conn_info(tconn, "fence-peer helper returned %d (%s)\n",
 470                   (r>>8) & 0xff, ex_to_string);
 471
 472  out:
 473
 474         /* Not using
 475            conn_request_state(tconn, mask, val, CS_VERBOSE);
 476            here, because we might were able to re-establish the connection in the
 477            meantime. */
 478         spin_lock_irq(&tconn->req_lock);
 479         if (tconn->cstate < C_WF_REPORT_PARAMS)
 480                 _conn_request_state(tconn, mask, val, CS_VERBOSE);
 481         spin_unlock_irq(&tconn->req_lock);
 482
 483         return conn_highest_pdsk(tconn) <= D_OUTDATED;
 484 }
 485
 486 static int _try_outdate_peer_async(void *data)
 487 {
 488         struct drbd_tconn *tconn = (struct drbd_tconn *)data;
 489
 490         conn_try_outdate_peer(tconn);
 491
 492         kref_put(&tconn->kref, &conn_destroy);
 493         return 0;
 494 }
 495
 496 void conn_try_outdate_peer_async(struct drbd_tconn *tconn)
 497 {
 498         struct task_struct *opa;
 499
 500         kref_get(&tconn->kref);
 501         opa = kthread_run(_try_outdate_peer_async, tconn, "drbd_async_h");
 502         if (IS_ERR(opa)) {
 503                 conn_err(tconn, "out of mem, failed to invoke fence-peer helper\n");
 504                 kref_put(&tconn->kref, &conn_destroy);
 505         }
 506 }
 507
 508 enum drbd_state_rv
 509 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 510 {
 511         const int max_tries = 4;
 512         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
 513         struct net_conf *nc;
 514         int try = 0;
 515         int forced = 0;
 516         union drbd_state mask, val;
 517
 518         if (new_role == R_PRIMARY)
 519                 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
 520
 521         mutex_lock(mdev->state_mutex);
 522
 523         mask.i = 0; mask.role = R_MASK;
 524         val.i  = 0; val.role  = new_role;
 525
 526         while (try++ < max_tries) {
 527                 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 528
 529                 /* in case we first succeeded to outdate,
 530                  * but now suddenly could establish a connection */
 531                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 532                         val.pdsk = 0;
 533                         mask.pdsk = 0;
 534                         continue;
 535                 }
 536
 537                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
 538                     (mdev->state.disk < D_UP_TO_DATE &&
 539                      mdev->state.disk >= D_INCONSISTENT)) {
 540                         mask.disk = D_MASK;
 541                         val.disk  = D_UP_TO_DATE;
 542                         forced = 1;
 543                         continue;
 544                 }
 545
 546                 if (rv == SS_NO_UP_TO_DATE_DISK &&
 547                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 548                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 549
 550                         if (conn_try_outdate_peer(mdev->tconn)) {
 551                                 val.disk = D_UP_TO_DATE;
 552                                 mask.disk = D_MASK;
 553                         }
 554                         continue;
 555                 }
 556
 557                 if (rv == SS_NOTHING_TO_DO)
 558                         goto out;
 559                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
 560                         if (!conn_try_outdate_peer(mdev->tconn) && force) {
 561                                 dev_warn(DEV, "Forced into split brain situation!\n");
 562                                 mask.pdsk = D_MASK;
 563                                 val.pdsk  = D_OUTDATED;
 564
 565                         }
 566                         continue;
 567                 }
 568                 if (rv == SS_TWO_PRIMARIES) {
 569                         /* Maybe the peer is detected as dead very soon...
 570                            retry at most once more in this case. */
 571                         int timeo;
 572                         rcu_read_lock();
 573                         nc = rcu_dereference(mdev->tconn->net_conf);
 574                         timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
 575                         rcu_read_unlock();
 576                         schedule_timeout_interruptible(timeo);
 577                         if (try < max_tries)
 578                                 try = max_tries - 1;
 579                         continue;
 580                 }
 581                 if (rv < SS_SUCCESS) {
 582                         rv = _drbd_request_state(mdev, mask, val,
 583                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 584                         if (rv < SS_SUCCESS)
 585                                 goto out;
 586                 }
 587                 break;
 588         }
 589
 590         if (rv < SS_SUCCESS)
 591                 goto out;
 592
 593         if (forced)
 594                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 595
 596         /* Wait until nothing is on the fly :) */
 597         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 598
 599         if (new_role == R_SECONDARY) {
 600                 set_disk_ro(mdev->vdisk, true);
 601                 if (get_ldev(mdev)) {
 602                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 603                         put_ldev(mdev);
 604                 }
 605         } else {
 606                 mutex_lock(&mdev->tconn->conf_update);
 607                 nc = mdev->tconn->net_conf;
 608                 if (nc)
 609                         nc->want_lose = 0; /* without copy; single bit op is atomic */
 610                 mutex_unlock(&mdev->tconn->conf_update);
 611
 612                 set_disk_ro(mdev->vdisk, false);
 613                 if (get_ldev(mdev)) {
 614                         if (((mdev->state.conn < C_CONNECTED ||
 615                                mdev->state.pdsk <= D_FAILED)
 616                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 617                                 drbd_uuid_new_current(mdev);
 618
 619                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 620                         put_ldev(mdev);
 621                 }
 622         }
 623
 624         /* writeout of activity log covered areas of the bitmap
 625          * to stable storage done in after state change already */
 626
 627         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 628                 /* if this was forced, we should consider sync */
 629                 if (forced)
 630                         drbd_send_uuids(mdev);
 631                 drbd_send_state(mdev);
 632         }
 633
 634         drbd_md_sync(mdev);
 635
 636         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 637 out:
 638         mutex_unlock(mdev->state_mutex);
 639         return rv;
 640 }
 641
 642 static const char *from_attrs_err_to_txt(int err)
 643 {
 644         return  err == -ENOMSG ? "required attribute missing" :
 645                 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
 646                 err == -EEXIST ? "can not change invariant setting" :
 647                 "invalid attribute value";
 648 }
 649
 650 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
 651 {
 652         struct set_role_parms parms;
 653         int err;
 654         enum drbd_ret_code retcode;
 655
 656         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
 657         if (!adm_ctx.reply_skb)
 658                 return retcode;
 659         if (retcode != NO_ERROR)
 660                 goto out;
 661
 662         memset(&parms, 0, sizeof(parms));
 663         if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
 664                 err = set_role_parms_from_attrs(&parms, info);
 665                 if (err) {
 666                         retcode = ERR_MANDATORY_TAG;
 667                         drbd_msg_put_info(from_attrs_err_to_txt(err));
 668                         goto out;
 669                 }
 670         }
 671
 672         if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
 673                 retcode = drbd_set_role(adm_ctx.mdev, R_PRIMARY, parms.assume_uptodate);
 674         else
 675                 retcode = drbd_set_role(adm_ctx.mdev, R_SECONDARY, 0);
 676 out:
 677         drbd_adm_finish(info, retcode);
 678         return 0;
 679 }
 680
 681 /* initializes the md.*_offset members, so we are able to find
 682  * the on disk meta data */
 683 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 684                                        struct drbd_backing_dev *bdev)
 685 {
 686         sector_t md_size_sect = 0;
 687         int meta_dev_idx;
 688
 689         rcu_read_lock();
 690         meta_dev_idx = rcu_dereference(bdev->disk_conf)->meta_dev_idx;
 691
 692         switch (meta_dev_idx) {
 693         default:
 694                 /* v07 style fixed size indexed meta data */
 695                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 696                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 697                 bdev->md.al_offset = MD_AL_OFFSET;
 698                 bdev->md.bm_offset = MD_BM_OFFSET;
 699                 break;
 700         case DRBD_MD_INDEX_FLEX_EXT:
 701                 /* just occupy the full device; unit: sectors */
 702                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 703                 bdev->md.md_offset = 0;
 704                 bdev->md.al_offset = MD_AL_OFFSET;
 705                 bdev->md.bm_offset = MD_BM_OFFSET;
 706                 break;
 707         case DRBD_MD_INDEX_INTERNAL:
 708         case DRBD_MD_INDEX_FLEX_INT:
 709                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 710                 /* al size is still fixed */
 711                 bdev->md.al_offset = -MD_AL_SECTORS;
 712                 /* we need (slightly less than) ~ this much bitmap sectors: */
 713                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 714                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 715                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 716                 md_size_sect = ALIGN(md_size_sect, 8);
 717
 718                 /* plus the "drbd meta data super block",
 719                  * and the activity log; */
 720                 md_size_sect += MD_BM_OFFSET;
 721
 722                 bdev->md.md_size_sect = md_size_sect;
 723                 /* bitmap offset is adjusted by 'super' block size */
 724                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 725                 break;
 726         }
 727         rcu_read_unlock();
 728 }
 729
 730 /* input size is expected to be in KB */
 731 char *ppsize(char *buf, unsigned long long size)
 732 {
 733         /* Needs 9 bytes at max including trailing NUL:
 734          * -1ULL ==> "16384 EB" */
 735         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 736         int base = 0;
 737         while (size >= 10000 && base < sizeof(units)-1) {
 738                 /* shift + round */
 739                 size = (size >> 10) + !!(size & (1<<9));
 740                 base++;
 741         }
 742         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
 743
 744         return buf;
 745 }
 746
 747 /* there is still a theoretical deadlock when called from receiver
 748  * on an D_INCONSISTENT R_PRIMARY:
 749  *  remote READ does inc_ap_bio, receiver would need to receive answer
 750  *  packet from remote to dec_ap_bio again.
 751  *  receiver receive_sizes(), comes here,
 752  *  waits for ap_bio_cnt == 0. -> deadlock.
 753  * but this cannot happen, actually, because:
 754  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 755  *  (not connected, or bad/no disk on peer):
 756  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 757  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 758  *  peer may not initiate a resize.
 759  */
 760 /* Note these are not to be confused with
 761  * drbd_adm_suspend_io/drbd_adm_resume_io,
 762  * which are (sub) state changes triggered by admin (drbdsetup),
 763  * and can be long lived.
 764  * This changes an mdev->flag, is triggered by drbd internals,
 765  * and should be short-lived. */
 766 void drbd_suspend_io(struct drbd_conf *mdev)
 767 {
 768         set_bit(SUSPEND_IO, &mdev->flags);
 769         if (drbd_suspended(mdev))
 770                 return;
 771         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 772 }
 773
 774 void drbd_resume_io(struct drbd_conf *mdev)
 775 {
 776         clear_bit(SUSPEND_IO, &mdev->flags);
 777         wake_up(&mdev->misc_wait);
 778 }
 779
 780 /**
 781  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 782  * @mdev:       DRBD device.
 783  *
 784  * Returns 0 on success, negative return values indicate errors.
 785  * You should call drbd_md_sync() after calling this function.
 786  */
 787 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 788 {
 789         sector_t prev_first_sect, prev_size; /* previous meta location */
 790         sector_t la_size, u_size;
 791         sector_t size;
 792         char ppb[10];
 793
 794         int md_moved, la_size_changed;
 795         enum determine_dev_size rv = unchanged;
 796
 797         /* race:
 798          * application request passes inc_ap_bio,
 799          * but then cannot get an AL-reference.
 800          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 801          *
 802          * to avoid that:
 803          * Suspend IO right here.
 804          * still lock the act_log to not trigger ASSERTs there.
 805          */
 806         drbd_suspend_io(mdev);
 807
 808         /* no wait necessary anymore, actually we could assert that */
 809         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 810
 811         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 812         prev_size = mdev->ldev->md.md_size_sect;
 813         la_size = mdev->ldev->md.la_size_sect;
 814
 815         /* TODO: should only be some assert here, not (re)init... */
 816         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 817
 818         rcu_read_lock();
 819         u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
 820         rcu_read_unlock();
 821         size = drbd_new_dev_size(mdev, mdev->ldev, u_size, flags & DDSF_FORCED);
 822
 823         if (drbd_get_capacity(mdev->this_bdev) != size ||
 824             drbd_bm_capacity(mdev) != size) {
 825                 int err;
 826                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 827                 if (unlikely(err)) {
 828                         /* currently there is only one error: ENOMEM! */
 829                         size = drbd_bm_capacity(mdev)>>1;
 830                         if (size == 0) {
 831                                 dev_err(DEV, "OUT OF MEMORY! "
 832                                     "Could not allocate bitmap!\n");
 833                         } else {
 834                                 dev_err(DEV, "BM resizing failed. "
 835                                     "Leaving size unchanged at size = %lu KB\n",
 836                                     (unsigned long)size);
 837                         }
 838                         rv = dev_size_error;
 839                 }
 840                 /* racy, see comments above. */
 841                 drbd_set_my_capacity(mdev, size);
 842                 mdev->ldev->md.la_size_sect = size;
 843                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 844                      (unsigned long long)size>>1);
 845         }
 846         if (rv == dev_size_error)
 847                 goto out;
 848
 849         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 850
 851         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 852                 || prev_size       != mdev->ldev->md.md_size_sect;
 853
 854         if (la_size_changed || md_moved) {
 855                 int err;
 856
 857                 drbd_al_shrink(mdev); /* All extents inactive. */
 858                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 859                          la_size_changed && md_moved ? "size changed and md moved" :
 860                          la_size_changed ? "size changed" : "md moved");
 861                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
 862                 err = drbd_bitmap_io(mdev, &drbd_bm_write,
 863                                 "size changed", BM_LOCKED_MASK);
 864                 if (err) {
 865                         rv = dev_size_error;
 866                         goto out;
 867                 }
 868                 drbd_md_mark_dirty(mdev);
 869         }
 870
 871         if (size > la_size)
 872                 rv = grew;
 873         if (size < la_size)
 874                 rv = shrunk;
 875 out:
 876         lc_unlock(mdev->act_log);
 877         wake_up(&mdev->al_wait);
 878         drbd_resume_io(mdev);
 879
 880         return rv;
 881 }
 882
 883 sector_t
 884 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
 885                   sector_t u_size, int assume_peer_has_space)
 886 {
 887         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 888         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 889         sector_t m_size; /* my size */
 890         sector_t size = 0;
 891
 892         m_size = drbd_get_max_capacity(bdev);
 893
 894         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 895                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 896                 p_size = m_size;
 897         }
 898
 899         if (p_size && m_size) {
 900                 size = min_t(sector_t, p_size, m_size);
 901         } else {
 902                 if (la_size) {
 903                         size = la_size;
 904                         if (m_size && m_size < size)
 905                                 size = m_size;
 906                         if (p_size && p_size < size)
 907                                 size = p_size;
 908                 } else {
 909                         if (m_size)
 910                                 size = m_size;
 911                         if (p_size)
 912                                 size = p_size;
 913                 }
 914         }
 915
 916         if (size == 0)
 917                 dev_err(DEV, "Both nodes diskless!\n");
 918
 919         if (u_size) {
 920                 if (u_size > size)
 921                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 922                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 923                 else
 924                         size = u_size;
 925         }
 926
 927         return size;
 928 }
 929
 930 /**
 931  * drbd_check_al_size() - Ensures that the AL is of the right size
 932  * @mdev:       DRBD device.
 933  *
 934  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 935  * failed, and 0 on success. You should call drbd_md_sync() after you called
 936  * this function.
 937  */
 938 static int drbd_check_al_size(struct drbd_conf *mdev, struct disk_conf *dc)
 939 {
 940         struct lru_cache *n, *t;
 941         struct lc_element *e;
 942         unsigned int in_use;
 943         int i;
 944
 945         if (mdev->act_log &&
 946             mdev->act_log->nr_elements == dc->al_extents)
 947                 return 0;
 948
 949         in_use = 0;
 950         t = mdev->act_log;
 951         n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
 952                 dc->al_extents, sizeof(struct lc_element), 0);
 953
 954         if (n == NULL) {
 955                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 956                 return -ENOMEM;
 957         }
 958         spin_lock_irq(&mdev->al_lock);
 959         if (t) {
 960                 for (i = 0; i < t->nr_elements; i++) {
 961                         e = lc_element_by_index(t, i);
 962                         if (e->refcnt)
 963                                 dev_err(DEV, "refcnt(%d)==%d\n",
 964                                     e->lc_number, e->refcnt);
 965                         in_use += e->refcnt;
 966                 }
 967         }
 968         if (!in_use)
 969                 mdev->act_log = n;
 970         spin_unlock_irq(&mdev->al_lock);
 971         if (in_use) {
 972                 dev_err(DEV, "Activity log still in use!\n");
 973                 lc_destroy(n);
 974                 return -EBUSY;
 975         } else {
 976                 if (t)
 977                         lc_destroy(t);
 978         }
 979         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 980         return 0;
 981 }
 982
 983 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
 984 {
 985         struct request_queue * const q = mdev->rq_queue;
 986         int max_hw_sectors = max_bio_size >> 9;
 987         int max_segments = 0;
 988
 989         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 990                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 991
 992                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
 993                 rcu_read_lock();
 994                 max_segments = rcu_dereference(mdev->ldev->disk_conf)->max_bio_bvecs;
 995                 rcu_read_unlock();
 996                 put_ldev(mdev);
 997         }
 998
 999         blk_queue_logical_block_size(q, 512);
1000         blk_queue_max_hw_sectors(q, max_hw_sectors);
1001         /* This is the workaround for "bio would need to, but cannot, be split" */
1002         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1003         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
1004
1005         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1006                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
1007
1008                 blk_queue_stack_limits(q, b);
1009
1010                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
1011                         dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1012                                  q->backing_dev_info.ra_pages,
1013                                  b->backing_dev_info.ra_pages);
1014                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
1015                 }
1016                 put_ldev(mdev);
1017         }
1018 }
1019
1020 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
1021 {
1022         int now, new, local, peer;
1023
1024         now = queue_max_hw_sectors(mdev->rq_queue) << 9;
1025         local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
1026         peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
1027
1028         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1029                 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
1030                 mdev->local_max_bio_size = local;
1031                 put_ldev(mdev);
1032         }
1033
1034         /* We may ignore peer limits if the peer is modern enough.
1035            Because new from 8.3.8 onwards the peer can use multiple
1036            BIOs for a single peer_request */
1037         if (mdev->state.conn >= C_CONNECTED) {
1038                 if (mdev->tconn->agreed_pro_version < 94)
1039                         peer = mdev->peer_max_bio_size;
1040                 else if (mdev->tconn->agreed_pro_version == 94)
1041                         peer = DRBD_MAX_SIZE_H80_PACKET;
1042                 else /* drbd 8.3.8 onwards */
1043                         peer = DRBD_MAX_BIO_SIZE;
1044         }
1045
1046         new = min_t(int, local, peer);
1047
1048         if (mdev->state.role == R_PRIMARY && new < now)
1049                 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
1050
1051         if (new != now)
1052                 dev_info(DEV, "max BIO size = %u\n", new);
1053
1054         drbd_setup_queue_param(mdev, new);
1055 }
1056
1057 /* Starts the worker thread */
1058 static void conn_reconfig_start(struct drbd_tconn *tconn)
1059 {
1060         drbd_thread_start(&tconn->worker);
1061         conn_flush_workqueue(tconn);
1062 }
1063
1064 /* if still unconfigured, stops worker again. */
1065 static void conn_reconfig_done(struct drbd_tconn *tconn)
1066 {
1067         bool stop_threads;
1068         spin_lock_irq(&tconn->req_lock);
1069         stop_threads = conn_all_vols_unconf(tconn);
1070         spin_unlock_irq(&tconn->req_lock);
1071         if (stop_threads) {
1072                 /* asender is implicitly stopped by receiver
1073                  * in conn_disconnect() */
1074                 drbd_thread_stop(&tconn->receiver);
1075                 drbd_thread_stop(&tconn->worker);
1076         }
1077 }
1078
1079 /* Make sure IO is suspended before calling this function(). */
1080 static void drbd_suspend_al(struct drbd_conf *mdev)
1081 {
1082         int s = 0;
1083
1084         if (!lc_try_lock(mdev->act_log)) {
1085                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
1086                 return;
1087         }
1088
1089         drbd_al_shrink(mdev);
1090         spin_lock_irq(&mdev->tconn->req_lock);
1091         if (mdev->state.conn < C_CONNECTED)
1092                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
1093         spin_unlock_irq(&mdev->tconn->req_lock);
1094         lc_unlock(mdev->act_log);
1095
1096         if (s)
1097                 dev_info(DEV, "Suspended AL updates\n");
1098 }
1099
1100
1101 static bool should_set_defaults(struct genl_info *info)
1102 {
1103         unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1104         return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1105 }
1106
1107 static void enforce_disk_conf_limits(struct disk_conf *dc)
1108 {
1109         if (dc->al_extents < DRBD_AL_EXTENTS_MIN)
1110                 dc->al_extents = DRBD_AL_EXTENTS_MIN;
1111         if (dc->al_extents > DRBD_AL_EXTENTS_MAX)
1112                 dc->al_extents = DRBD_AL_EXTENTS_MAX;
1113
1114         if (dc->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1115                 dc->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1116 }
1117
1118 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1119 {
1120         enum drbd_ret_code retcode;
1121         struct drbd_conf *mdev;
1122         struct disk_conf *new_disk_conf, *old_disk_conf;
1123         struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1124         int err, fifo_size;
1125
1126         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1127         if (!adm_ctx.reply_skb)
1128                 return retcode;
1129         if (retcode != NO_ERROR)
1130                 goto out;
1131
1132         mdev = adm_ctx.mdev;
1133
1134         /* we also need a disk
1135          * to change the options on */
1136         if (!get_ldev(mdev)) {
1137                 retcode = ERR_NO_DISK;
1138                 goto out;
1139         }
1140
1141         new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1142         if (!new_disk_conf) {
1143                 retcode = ERR_NOMEM;
1144                 goto fail;
1145         }
1146
1147         mutex_lock(&mdev->tconn->conf_update);
1148         old_disk_conf = mdev->ldev->disk_conf;
1149         *new_disk_conf = *old_disk_conf;
1150         if (should_set_defaults(info))
1151                 set_disk_conf_defaults(new_disk_conf);
1152
1153         err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1154         if (err) {
1155                 retcode = ERR_MANDATORY_TAG;
1156                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1157         }
1158
1159         if (!expect(new_disk_conf->resync_rate >= 1))
1160                 new_disk_conf->resync_rate = 1;
1161
1162         enforce_disk_conf_limits(new_disk_conf);
1163
1164         fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1165         if (fifo_size != mdev->rs_plan_s->size) {
1166                 new_plan = fifo_alloc(fifo_size);
1167                 if (!new_plan) {
1168                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1169                         retcode = ERR_NOMEM;
1170                         goto fail_unlock;
1171                 }
1172         }
1173
1174         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1175         drbd_al_shrink(mdev);
1176         err = drbd_check_al_size(mdev, new_disk_conf);
1177         lc_unlock(mdev->act_log);
1178         wake_up(&mdev->al_wait);
1179
1180         if (err) {
1181                 retcode = ERR_NOMEM;
1182                 goto fail_unlock;
1183         }
1184
1185         write_lock_irq(&global_state_lock);
1186         retcode = drbd_sync_after_valid(mdev, new_disk_conf->resync_after);
1187         if (retcode == NO_ERROR) {
1188                 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
1189                 drbd_sync_after_changed(mdev);
1190         }
1191         write_unlock_irq(&global_state_lock);
1192
1193         if (retcode != NO_ERROR)
1194                 goto fail_unlock;
1195
1196         if (new_plan) {
1197                 old_plan = mdev->rs_plan_s;
1198                 rcu_assign_pointer(mdev->rs_plan_s, new_plan);
1199         }
1200
1201         mutex_unlock(&mdev->tconn->conf_update);
1202         drbd_md_sync(mdev);
1203
1204         if (mdev->state.conn >= C_CONNECTED)
1205                 drbd_send_sync_param(mdev);
1206
1207         synchronize_rcu();
1208         kfree(old_disk_conf);
1209         kfree(old_plan);
1210         goto success;
1211
1212 fail_unlock:
1213         mutex_unlock(&mdev->tconn->conf_update);
1214  fail:
1215         kfree(new_disk_conf);
1216         kfree(new_plan);
1217 success:
1218         put_ldev(mdev);
1219  out:
1220         drbd_adm_finish(info, retcode);
1221         return 0;
1222 }
1223
1224 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1225 {
1226         struct drbd_conf *mdev;
1227         int err;
1228         enum drbd_ret_code retcode;
1229         enum determine_dev_size dd;
1230         sector_t max_possible_sectors;
1231         sector_t min_md_device_sectors;
1232         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1233         struct disk_conf *new_disk_conf = NULL;
1234         struct block_device *bdev;
1235         struct lru_cache *resync_lru = NULL;
1236         struct fifo_buffer *new_plan = NULL;
1237         union drbd_state ns, os;
1238         enum drbd_state_rv rv;
1239         struct net_conf *nc;
1240         int cp_discovered = 0;
1241
1242         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1243         if (!adm_ctx.reply_skb)
1244                 return retcode;
1245         if (retcode != NO_ERROR)
1246                 goto finish;
1247
1248         mdev = adm_ctx.mdev;
1249         conn_reconfig_start(mdev->tconn);
1250
1251         /* if you want to reconfigure, please tear down first */
1252         if (mdev->state.disk > D_DISKLESS) {
1253                 retcode = ERR_DISK_CONFIGURED;
1254                 goto fail;
1255         }
1256         /* It may just now have detached because of IO error.  Make sure
1257          * drbd_ldev_destroy is done already, we may end up here very fast,
1258          * e.g. if someone calls attach from the on-io-error handler,
1259          * to realize a "hot spare" feature (not that I'd recommend that) */
1260         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1261
1262         /* allocation not in the IO path, drbdsetup context */
1263         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1264         if (!nbc) {
1265                 retcode = ERR_NOMEM;
1266                 goto fail;
1267         }
1268         new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1269         if (!new_disk_conf) {
1270                 retcode = ERR_NOMEM;
1271                 goto fail;
1272         }
1273         nbc->disk_conf = new_disk_conf;
1274
1275         set_disk_conf_defaults(new_disk_conf);
1276         err = disk_conf_from_attrs(new_disk_conf, info);
1277         if (err) {
1278                 retcode = ERR_MANDATORY_TAG;
1279                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1280                 goto fail;
1281         }
1282
1283         enforce_disk_conf_limits(new_disk_conf);
1284
1285         new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1286         if (!new_plan) {
1287                 retcode = ERR_NOMEM;
1288                 goto fail;
1289         }
1290
1291         if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1292                 retcode = ERR_MD_IDX_INVALID;
1293                 goto fail;
1294         }
1295
1296         rcu_read_lock();
1297         nc = rcu_dereference(mdev->tconn->net_conf);
1298         if (nc) {
1299                 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1300                         rcu_read_unlock();
1301                         retcode = ERR_STONITH_AND_PROT_A;
1302                         goto fail;
1303                 }
1304         }
1305         rcu_read_unlock();
1306
1307         bdev = blkdev_get_by_path(new_disk_conf->backing_dev,
1308                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1309         if (IS_ERR(bdev)) {
1310                 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->backing_dev,
1311                         PTR_ERR(bdev));
1312                 retcode = ERR_OPEN_DISK;
1313                 goto fail;
1314         }
1315         nbc->backing_bdev = bdev;
1316
1317         /*
1318          * meta_dev_idx >= 0: external fixed size, possibly multiple
1319          * drbd sharing one meta device.  TODO in that case, paranoia
1320          * check that [md_bdev, meta_dev_idx] is not yet used by some
1321          * other drbd minor!  (if you use drbd.conf + drbdadm, that
1322          * should check it for you already; but if you don't, or
1323          * someone fooled it, we need to double check here)
1324          */
1325         bdev = blkdev_get_by_path(new_disk_conf->meta_dev,
1326                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1327                                   (new_disk_conf->meta_dev_idx < 0) ?
1328                                   (void *)mdev : (void *)drbd_m_holder);
1329         if (IS_ERR(bdev)) {
1330                 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->meta_dev,
1331                         PTR_ERR(bdev));
1332                 retcode = ERR_OPEN_MD_DISK;
1333                 goto fail;
1334         }
1335         nbc->md_bdev = bdev;
1336
1337         if ((nbc->backing_bdev == nbc->md_bdev) !=
1338             (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1339              new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1340                 retcode = ERR_MD_IDX_INVALID;
1341                 goto fail;
1342         }
1343
1344         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1345                         1, 61, sizeof(struct bm_extent),
1346                         offsetof(struct bm_extent, lce));
1347         if (!resync_lru) {
1348                 retcode = ERR_NOMEM;
1349                 goto fail;
1350         }
1351
1352         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1353         drbd_md_set_sector_offsets(mdev, nbc);
1354
1355         if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1356                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1357                         (unsigned long long) drbd_get_max_capacity(nbc),
1358                         (unsigned long long) new_disk_conf->disk_size);
1359                 retcode = ERR_DISK_TO_SMALL;
1360                 goto fail;
1361         }
1362
1363         if (new_disk_conf->meta_dev_idx < 0) {
1364                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1365                 /* at least one MB, otherwise it does not make sense */
1366                 min_md_device_sectors = (2<<10);
1367         } else {
1368                 max_possible_sectors = DRBD_MAX_SECTORS;
1369                 min_md_device_sectors = MD_RESERVED_SECT * (new_disk_conf->meta_dev_idx + 1);
1370         }
1371
1372         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1373                 retcode = ERR_MD_DISK_TO_SMALL;
1374                 dev_warn(DEV, "refusing attach: md-device too small, "
1375                      "at least %llu sectors needed for this meta-disk type\n",
1376                      (unsigned long long) min_md_device_sectors);
1377                 goto fail;
1378         }
1379
1380         /* Make sure the new disk is big enough
1381          * (we may currently be R_PRIMARY with no local disk...) */
1382         if (drbd_get_max_capacity(nbc) <
1383             drbd_get_capacity(mdev->this_bdev)) {
1384                 retcode = ERR_DISK_TO_SMALL;
1385                 goto fail;
1386         }
1387
1388         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1389
1390         if (nbc->known_size > max_possible_sectors) {
1391                 dev_warn(DEV, "==> truncating very big lower level device "
1392                         "to currently maximum possible %llu sectors <==\n",
1393                         (unsigned long long) max_possible_sectors);
1394                 if (new_disk_conf->meta_dev_idx >= 0)
1395                         dev_warn(DEV, "==>> using internal or flexible "
1396                                       "meta data may help <<==\n");
1397         }
1398
1399         drbd_suspend_io(mdev);
1400         /* also wait for the last barrier ack. */
1401         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || drbd_suspended(mdev));
1402         /* and for any other previously queued work */
1403         drbd_flush_workqueue(mdev);
1404
1405         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1406         retcode = rv;  /* FIXME: Type mismatch. */
1407         drbd_resume_io(mdev);
1408         if (rv < SS_SUCCESS)
1409                 goto fail;
1410
1411         if (!get_ldev_if_state(mdev, D_ATTACHING))
1412                 goto force_diskless;
1413
1414         drbd_md_set_sector_offsets(mdev, nbc);
1415
1416         if (!mdev->bitmap) {
1417                 if (drbd_bm_init(mdev)) {
1418                         retcode = ERR_NOMEM;
1419                         goto force_diskless_dec;
1420                 }
1421         }
1422
1423         retcode = drbd_md_read(mdev, nbc);
1424         if (retcode != NO_ERROR)
1425                 goto force_diskless_dec;
1426
1427         if (mdev->state.conn < C_CONNECTED &&
1428             mdev->state.role == R_PRIMARY &&
1429             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1430                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1431                     (unsigned long long)mdev->ed_uuid);
1432                 retcode = ERR_DATA_NOT_CURRENT;
1433                 goto force_diskless_dec;
1434         }
1435
1436         /* Since we are diskless, fix the activity log first... */
1437         if (drbd_check_al_size(mdev, new_disk_conf)) {
1438                 retcode = ERR_NOMEM;
1439                 goto force_diskless_dec;
1440         }
1441
1442         /* Prevent shrinking of consistent devices ! */
1443         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1444             drbd_new_dev_size(mdev, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1445                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1446                 retcode = ERR_DISK_TO_SMALL;
1447                 goto force_diskless_dec;
1448         }
1449
1450         if (!drbd_al_read_log(mdev, nbc)) {
1451                 retcode = ERR_IO_MD_DISK;
1452                 goto force_diskless_dec;
1453         }
1454
1455         /* Reset the "barriers don't work" bits here, then force meta data to
1456          * be written, to ensure we determine if barriers are supported. */
1457         if (new_disk_conf->md_flushes)
1458                 clear_bit(MD_NO_FUA, &mdev->flags);
1459         else
1460                 set_bit(MD_NO_FUA, &mdev->flags);
1461
1462         /* Point of no return reached.
1463          * Devices and memory are no longer released by error cleanup below.
1464          * now mdev takes over responsibility, and the state engine should
1465          * clean it up somewhere.  */
1466         D_ASSERT(mdev->ldev == NULL);
1467         mdev->ldev = nbc;
1468         mdev->resync = resync_lru;
1469         mdev->rs_plan_s = new_plan;
1470         nbc = NULL;
1471         resync_lru = NULL;
1472         new_disk_conf = NULL;
1473         new_plan = NULL;
1474
1475         mdev->write_ordering = WO_bdev_flush;
1476         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1477
1478         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1479                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1480         else
1481                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1482
1483         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1484             !(mdev->state.role == R_PRIMARY && mdev->tconn->susp_nod)) {
1485                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1486                 cp_discovered = 1;
1487         }
1488
1489         mdev->send_cnt = 0;
1490         mdev->recv_cnt = 0;
1491         mdev->read_cnt = 0;
1492         mdev->writ_cnt = 0;
1493
1494         drbd_reconsider_max_bio_size(mdev);
1495
1496         /* If I am currently not R_PRIMARY,
1497          * but meta data primary indicator is set,
1498          * I just now recover from a hard crash,
1499          * and have been R_PRIMARY before that crash.
1500          *
1501          * Now, if I had no connection before that crash
1502          * (have been degraded R_PRIMARY), chances are that
1503          * I won't find my peer now either.
1504          *
1505          * In that case, and _only_ in that case,
1506          * we use the degr-wfc-timeout instead of the default,
1507          * so we can automatically recover from a crash of a
1508          * degraded but active "cluster" after a certain timeout.
1509          */
1510         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1511         if (mdev->state.role != R_PRIMARY &&
1512              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1513             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1514                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1515
1516         dd = drbd_determine_dev_size(mdev, 0);
1517         if (dd == dev_size_error) {
1518                 retcode = ERR_NOMEM_BITMAP;
1519                 goto force_diskless_dec;
1520         } else if (dd == grew)
1521                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1522
1523         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1524                 dev_info(DEV, "Assuming that all blocks are out of sync "
1525                      "(aka FullSync)\n");
1526                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1527                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1528                         retcode = ERR_IO_MD_DISK;
1529                         goto force_diskless_dec;
1530                 }
1531         } else {
1532                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1533                         "read from attaching", BM_LOCKED_MASK)) {
1534                         retcode = ERR_IO_MD_DISK;
1535                         goto force_diskless_dec;
1536                 }
1537         }
1538
1539         if (cp_discovered) {
1540                 drbd_al_apply_to_bm(mdev);
1541                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1542                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1543                         retcode = ERR_IO_MD_DISK;
1544                         goto force_diskless_dec;
1545                 }
1546         }
1547
1548         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1549                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1550
1551         spin_lock_irq(&mdev->tconn->req_lock);
1552         os = drbd_read_state(mdev);
1553         ns = os;
1554         /* If MDF_CONSISTENT is not set go into inconsistent state,
1555            otherwise investigate MDF_WasUpToDate...
1556            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1557            otherwise into D_CONSISTENT state.
1558         */
1559         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1560                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1561                         ns.disk = D_CONSISTENT;
1562                 else
1563                         ns.disk = D_OUTDATED;
1564         } else {
1565                 ns.disk = D_INCONSISTENT;
1566         }
1567
1568         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1569                 ns.pdsk = D_OUTDATED;
1570
1571         rcu_read_lock();
1572         if (ns.disk == D_CONSISTENT &&
1573             (ns.pdsk == D_OUTDATED || rcu_dereference(mdev->ldev->disk_conf)->fencing == FP_DONT_CARE))
1574                 ns.disk = D_UP_TO_DATE;
1575         rcu_read_unlock();
1576
1577         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1578            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1579            this point, because drbd_request_state() modifies these
1580            flags. */
1581
1582         /* In case we are C_CONNECTED postpone any decision on the new disk
1583            state after the negotiation phase. */
1584         if (mdev->state.conn == C_CONNECTED) {
1585                 mdev->new_state_tmp.i = ns.i;
1586                 ns.i = os.i;
1587                 ns.disk = D_NEGOTIATING;
1588
1589                 /* We expect to receive up-to-date UUIDs soon.
1590                    To avoid a race in receive_state, free p_uuid while
1591                    holding req_lock. I.e. atomic with the state change */
1592                 kfree(mdev->p_uuid);
1593                 mdev->p_uuid = NULL;
1594         }
1595
1596         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1597         spin_unlock_irq(&mdev->tconn->req_lock);
1598
1599         if (rv < SS_SUCCESS)
1600                 goto force_diskless_dec;
1601
1602         if (mdev->state.role == R_PRIMARY)
1603                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1604         else
1605                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1606
1607         drbd_md_mark_dirty(mdev);
1608         drbd_md_sync(mdev);
1609
1610         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1611         put_ldev(mdev);
1612         conn_reconfig_done(mdev->tconn);
1613         drbd_adm_finish(info, retcode);
1614         return 0;
1615
1616  force_diskless_dec:
1617         put_ldev(mdev);
1618  force_diskless:
1619         drbd_force_state(mdev, NS(disk, D_FAILED));
1620         drbd_md_sync(mdev);
1621  fail:
1622         conn_reconfig_done(mdev->tconn);
1623         if (nbc) {
1624                 if (nbc->backing_bdev)
1625                         blkdev_put(nbc->backing_bdev,
1626                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1627                 if (nbc->md_bdev)
1628                         blkdev_put(nbc->md_bdev,
1629                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1630                 kfree(nbc);
1631         }
1632         kfree(new_disk_conf);
1633         lc_destroy(resync_lru);
1634         kfree(new_plan);
1635
1636  finish:
1637         drbd_adm_finish(info, retcode);
1638         return 0;
1639 }
1640
1641 static int adm_detach(struct drbd_conf *mdev)
1642 {
1643         enum drbd_state_rv retcode;
1644         int ret;
1645         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1646         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1647         /* D_FAILED will transition to DISKLESS. */
1648         ret = wait_event_interruptible(mdev->misc_wait,
1649                         mdev->state.disk != D_FAILED);
1650         drbd_resume_io(mdev);
1651         if ((int)retcode == (int)SS_IS_DISKLESS)
1652                 retcode = SS_NOTHING_TO_DO;
1653         if (ret)
1654                 retcode = ERR_INTR;
1655         return retcode;
1656 }
1657
1658 /* Detaching the disk is a process in multiple stages.  First we need to lock
1659  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1660  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1661  * internal references as well.
1662  * Only then we have finally detached. */
1663 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1664 {
1665         enum drbd_ret_code retcode;
1666
1667         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1668         if (!adm_ctx.reply_skb)
1669                 return retcode;
1670         if (retcode != NO_ERROR)
1671                 goto out;
1672
1673         retcode = adm_detach(adm_ctx.mdev);
1674 out:
1675         drbd_adm_finish(info, retcode);
1676         return 0;
1677 }
1678
1679 static bool conn_resync_running(struct drbd_tconn *tconn)
1680 {
1681         struct drbd_conf *mdev;
1682         bool rv = false;
1683         int vnr;
1684
1685         rcu_read_lock();
1686         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1687                 if (mdev->state.conn == C_SYNC_SOURCE ||
1688                     mdev->state.conn == C_SYNC_TARGET ||
1689                     mdev->state.conn == C_PAUSED_SYNC_S ||
1690                     mdev->state.conn == C_PAUSED_SYNC_T) {
1691                         rv = true;
1692                         break;
1693                 }
1694         }
1695         rcu_read_unlock();
1696
1697         return rv;
1698 }
1699
1700 static bool conn_ov_running(struct drbd_tconn *tconn)
1701 {
1702         struct drbd_conf *mdev;
1703         bool rv = false;
1704         int vnr;
1705
1706         rcu_read_lock();
1707         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1708                 if (mdev->state.conn == C_VERIFY_S ||
1709                     mdev->state.conn == C_VERIFY_T) {
1710                         rv = true;
1711                         break;
1712                 }
1713         }
1714         rcu_read_unlock();
1715
1716         return rv;
1717 }
1718
1719 static enum drbd_ret_code
1720 _check_net_options(struct drbd_tconn *tconn, struct net_conf *old_conf, struct net_conf *new_conf)
1721 {
1722         struct drbd_conf *mdev;
1723         int i;
1724
1725         if (old_conf && tconn->agreed_pro_version < 100 &&
1726             tconn->cstate == C_WF_REPORT_PARAMS &&
1727             new_conf->wire_protocol != old_conf->wire_protocol)
1728                 return ERR_NEED_APV_100;
1729
1730         if (new_conf->two_primaries &&
1731             (new_conf->wire_protocol != DRBD_PROT_C))
1732                 return ERR_NOT_PROTO_C;
1733
1734         idr_for_each_entry(&tconn->volumes, mdev, i) {
1735                 if (get_ldev(mdev)) {
1736                         enum drbd_fencing_p fp = rcu_dereference(mdev->ldev->disk_conf)->fencing;
1737                         put_ldev(mdev);
1738                         if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
1739                                 return ERR_STONITH_AND_PROT_A;
1740                 }
1741                 if (mdev->state.role == R_PRIMARY && new_conf->want_lose)
1742                         return ERR_DISCARD;
1743         }
1744
1745         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A)
1746                 return ERR_CONG_NOT_PROTO_A;
1747
1748         return NO_ERROR;
1749 }
1750
1751 static enum drbd_ret_code
1752 check_net_options(struct drbd_tconn *tconn, struct net_conf *new_conf)
1753 {
1754         static enum drbd_ret_code rv;
1755         struct drbd_conf *mdev;
1756         int i;
1757
1758         rcu_read_lock();
1759         rv = _check_net_options(tconn, rcu_dereference(tconn->net_conf), new_conf);
1760         rcu_read_unlock();
1761
1762         /* tconn->volumes protected by genl_lock() here */
1763         idr_for_each_entry(&tconn->volumes, mdev, i) {
1764                 if (!mdev->bitmap) {
1765                         if(drbd_bm_init(mdev))
1766                                 return ERR_NOMEM;
1767                 }
1768         }
1769
1770         return rv;
1771 }
1772
1773 struct crypto {
1774         struct crypto_hash *verify_tfm;
1775         struct crypto_hash *csums_tfm;
1776         struct crypto_hash *cram_hmac_tfm;
1777         struct crypto_hash *integrity_tfm;
1778         void *int_dig_in;
1779         void *int_dig_vv;
1780 };
1781
1782 static int
1783 alloc_hash(struct crypto_hash **tfm, char *tfm_name, int err_alg)
1784 {
1785         if (!tfm_name[0])
1786                 return NO_ERROR;
1787
1788         *tfm = crypto_alloc_hash(tfm_name, 0, CRYPTO_ALG_ASYNC);
1789         if (IS_ERR(*tfm)) {
1790                 *tfm = NULL;
1791                 return err_alg;
1792         }
1793
1794         return NO_ERROR;
1795 }
1796
1797 static enum drbd_ret_code
1798 alloc_crypto(struct crypto *crypto, struct net_conf *new_conf)
1799 {
1800         char hmac_name[CRYPTO_MAX_ALG_NAME];
1801         enum drbd_ret_code rv;
1802         int hash_size;
1803
1804         rv = alloc_hash(&crypto->csums_tfm, new_conf->csums_alg,
1805                        ERR_CSUMS_ALG);
1806         if (rv != NO_ERROR)
1807                 return rv;
1808         rv = alloc_hash(&crypto->verify_tfm, new_conf->verify_alg,
1809                        ERR_VERIFY_ALG);
1810         if (rv != NO_ERROR)
1811                 return rv;
1812         rv = alloc_hash(&crypto->integrity_tfm, new_conf->integrity_alg,
1813                        ERR_INTEGRITY_ALG);
1814         if (rv != NO_ERROR)
1815                 return rv;
1816         if (new_conf->cram_hmac_alg[0] != 0) {
1817                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1818                          new_conf->cram_hmac_alg);
1819
1820                 rv = alloc_hash(&crypto->cram_hmac_tfm, hmac_name,
1821                                ERR_AUTH_ALG);
1822         }
1823         if (crypto->integrity_tfm) {
1824                 hash_size = crypto_hash_digestsize(crypto->integrity_tfm);
1825                 crypto->int_dig_in = kmalloc(hash_size, GFP_KERNEL);
1826                 if (!crypto->int_dig_in)
1827                         return ERR_NOMEM;
1828                 crypto->int_dig_vv = kmalloc(hash_size, GFP_KERNEL);
1829                 if (!crypto->int_dig_vv)
1830                         return ERR_NOMEM;
1831         }
1832
1833         return rv;
1834 }
1835
1836 static void free_crypto(struct crypto *crypto)
1837 {
1838         kfree(crypto->int_dig_in);
1839         kfree(crypto->int_dig_vv);
1840         crypto_free_hash(crypto->cram_hmac_tfm);
1841         crypto_free_hash(crypto->integrity_tfm);
1842         crypto_free_hash(crypto->csums_tfm);
1843         crypto_free_hash(crypto->verify_tfm);
1844 }
1845
1846 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
1847 {
1848         enum drbd_ret_code retcode;
1849         struct drbd_tconn *tconn;
1850         struct net_conf *old_conf, *new_conf = NULL;
1851         int err;
1852         int ovr; /* online verify running */
1853         int rsr; /* re-sync running */
1854         struct crypto crypto = { };
1855         bool change_integrity_alg;
1856
1857         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1858         if (!adm_ctx.reply_skb)
1859                 return retcode;
1860         if (retcode != NO_ERROR)
1861                 goto out;
1862
1863         tconn = adm_ctx.tconn;
1864
1865         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1866         if (!new_conf) {
1867                 retcode = ERR_NOMEM;
1868                 goto out;
1869         }
1870
1871         conn_reconfig_start(tconn);
1872
1873         mutex_lock(&tconn->data.mutex);
1874         mutex_lock(&tconn->conf_update);
1875         old_conf = tconn->net_conf;
1876
1877         if (!old_conf) {
1878                 drbd_msg_put_info("net conf missing, try connect");
1879                 retcode = ERR_INVALID_REQUEST;
1880                 goto fail;
1881         }
1882
1883         *new_conf = *old_conf;
1884         if (should_set_defaults(info))
1885                 set_net_conf_defaults(new_conf);
1886
1887         err = net_conf_from_attrs_for_change(new_conf, info);
1888         if (err) {
1889                 retcode = ERR_MANDATORY_TAG;
1890                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1891                 goto fail;
1892         }
1893
1894         retcode = check_net_options(tconn, new_conf);
1895         if (retcode != NO_ERROR)
1896                 goto fail;
1897
1898         /* re-sync running */
1899         rsr = conn_resync_running(tconn);
1900         if (rsr && strcmp(new_conf->csums_alg, old_conf->csums_alg)) {
1901                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1902                 goto fail;
1903         }
1904
1905         /* online verify running */
1906         ovr = conn_ov_running(tconn);
1907         if (ovr && strcmp(new_conf->verify_alg, old_conf->verify_alg)) {
1908                 retcode = ERR_VERIFY_RUNNING;
1909                 goto fail;
1910         }
1911
1912         change_integrity_alg = strcmp(old_conf->integrity_alg,
1913                                       new_conf->integrity_alg);
1914
1915         retcode = alloc_crypto(&crypto, new_conf);
1916         if (retcode != NO_ERROR)
1917                 goto fail;
1918
1919         rcu_assign_pointer(tconn->net_conf, new_conf);
1920
1921         if (!rsr) {
1922                 crypto_free_hash(tconn->csums_tfm);
1923                 tconn->csums_tfm = crypto.csums_tfm;
1924                 crypto.csums_tfm = NULL;
1925         }
1926         if (!ovr) {
1927                 crypto_free_hash(tconn->verify_tfm);
1928                 tconn->verify_tfm = crypto.verify_tfm;
1929                 crypto.verify_tfm = NULL;
1930         }
1931
1932         kfree(tconn->int_dig_in);
1933         tconn->int_dig_in = crypto.int_dig_in;
1934         kfree(tconn->int_dig_vv);
1935         tconn->int_dig_vv = crypto.int_dig_vv;
1936         crypto_free_hash(tconn->integrity_tfm);
1937         tconn->integrity_tfm = crypto.integrity_tfm;
1938         if (change_integrity_alg) {
1939                 /* Do this without trying to take tconn->data.mutex again.  */
1940                 if (__drbd_send_protocol(tconn))
1941                         goto fail;
1942         }
1943
1944         /* FIXME Changing cram_hmac while the connection is established is useless */
1945         crypto_free_hash(tconn->cram_hmac_tfm);
1946         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
1947
1948         mutex_unlock(&tconn->conf_update);
1949         mutex_unlock(&tconn->data.mutex);
1950         synchronize_rcu();
1951         kfree(old_conf);
1952
1953         if (tconn->cstate >= C_WF_REPORT_PARAMS)
1954                 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn)));
1955
1956         goto done;
1957
1958  fail:
1959         mutex_unlock(&tconn->conf_update);
1960         mutex_unlock(&tconn->data.mutex);
1961         free_crypto(&crypto);
1962         kfree(new_conf);
1963  done:
1964         conn_reconfig_done(tconn);
1965  out:
1966         drbd_adm_finish(info, retcode);
1967         return 0;
1968 }
1969
1970 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
1971 {
1972         struct drbd_conf *mdev;
1973         struct net_conf *old_conf, *new_conf = NULL;
1974         struct crypto crypto = { };
1975         struct drbd_tconn *oconn;
1976         struct drbd_tconn *tconn;
1977         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1978         enum drbd_ret_code retcode;
1979         int i;
1980         int err;
1981
1982         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1983         if (!adm_ctx.reply_skb)
1984                 return retcode;
1985         if (retcode != NO_ERROR)
1986                 goto out;
1987
1988         tconn = adm_ctx.tconn;
1989         conn_reconfig_start(tconn);
1990
1991         if (tconn->cstate > C_STANDALONE) {
1992                 retcode = ERR_NET_CONFIGURED;
1993                 goto fail;
1994         }
1995
1996         /* allocation not in the IO path, cqueue thread context */
1997         new_conf = kzalloc(sizeof(*new_conf), GFP_KERNEL);
1998         if (!new_conf) {
1999                 retcode = ERR_NOMEM;
2000                 goto fail;
2001         }
2002
2003         set_net_conf_defaults(new_conf);
2004
2005         err = net_conf_from_attrs(new_conf, info);
2006         if (err) {
2007                 retcode = ERR_MANDATORY_TAG;
2008                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2009                 goto fail;
2010         }
2011
2012         retcode = check_net_options(tconn, new_conf);
2013         if (retcode != NO_ERROR)
2014                 goto fail;
2015
2016         retcode = NO_ERROR;
2017
2018         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
2019         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
2020
2021         /* No need for _rcu here. All reconfiguration is
2022          * strictly serialized on genl_lock(). We are protected against
2023          * concurrent reconfiguration/addition/deletion */
2024         list_for_each_entry(oconn, &drbd_tconns, all_tconn) {
2025                 struct net_conf *nc;
2026                 if (oconn == tconn)
2027                         continue;
2028
2029                 rcu_read_lock();
2030                 nc = rcu_dereference(oconn->net_conf);
2031                 if (nc) {
2032                         taken_addr = (struct sockaddr *)&nc->my_addr;
2033                         if (new_conf->my_addr_len == nc->my_addr_len &&
2034                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
2035                                 retcode = ERR_LOCAL_ADDR;
2036
2037                         taken_addr = (struct sockaddr *)&nc->peer_addr;
2038                         if (new_conf->peer_addr_len == nc->peer_addr_len &&
2039                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
2040                                 retcode = ERR_PEER_ADDR;
2041                 }
2042                 rcu_read_unlock();
2043                 if (retcode != NO_ERROR)
2044                         goto fail;
2045         }
2046
2047         retcode = alloc_crypto(&crypto, new_conf);
2048         if (retcode != NO_ERROR)
2049                 goto fail;
2050
2051         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2052
2053         conn_flush_workqueue(tconn);
2054
2055         mutex_lock(&tconn->conf_update);
2056         old_conf = tconn->net_conf;
2057         if (old_conf) {
2058                 retcode = ERR_NET_CONFIGURED;
2059                 mutex_unlock(&tconn->conf_update);
2060                 goto fail;
2061         }
2062         rcu_assign_pointer(tconn->net_conf, new_conf);
2063
2064         conn_free_crypto(tconn);
2065         tconn->int_dig_in = crypto.int_dig_in;
2066         tconn->int_dig_vv = crypto.int_dig_vv;
2067         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
2068         tconn->integrity_tfm = crypto.integrity_tfm;
2069         tconn->csums_tfm = crypto.csums_tfm;
2070         tconn->verify_tfm = crypto.verify_tfm;
2071
2072         mutex_unlock(&tconn->conf_update);
2073
2074         rcu_read_lock();
2075         idr_for_each_entry(&tconn->volumes, mdev, i) {
2076                 mdev->send_cnt = 0;
2077                 mdev->recv_cnt = 0;
2078         }
2079         rcu_read_unlock();
2080
2081         retcode = conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2082
2083         conn_reconfig_done(tconn);
2084         drbd_adm_finish(info, retcode);
2085         return 0;
2086
2087 fail:
2088         free_crypto(&crypto);
2089         kfree(new_conf);
2090
2091         conn_reconfig_done(tconn);
2092 out:
2093         drbd_adm_finish(info, retcode);
2094         return 0;
2095 }
2096
2097 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
2098 {
2099         enum drbd_state_rv rv;
2100
2101         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2102                         force ? CS_HARD : 0);
2103
2104         switch (rv) {
2105         case SS_NOTHING_TO_DO:
2106                 break;
2107         case SS_ALREADY_STANDALONE:
2108                 return SS_SUCCESS;
2109         case SS_PRIMARY_NOP:
2110                 /* Our state checking code wants to see the peer outdated. */
2111                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2112                                                 pdsk, D_OUTDATED), CS_VERBOSE);
2113                 break;
2114         case SS_CW_FAILED_BY_PEER:
2115                 /* The peer probably wants to see us outdated. */
2116                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2117                                                         disk, D_OUTDATED), 0);
2118                 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2119                         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2120                                         CS_HARD);
2121                 }
2122                 break;
2123         default:;
2124                 /* no special handling necessary */
2125         }
2126
2127         if (rv >= SS_SUCCESS) {
2128                 enum drbd_state_rv rv2;
2129                 /* No one else can reconfigure the network while I am here.
2130                  * The state handling only uses drbd_thread_stop_nowait(),
2131                  * we want to really wait here until the receiver is no more.
2132                  */
2133                 drbd_thread_stop(&adm_ctx.tconn->receiver);
2134
2135                 /* Race breaker.  This additional state change request may be
2136                  * necessary, if this was a forced disconnect during a receiver
2137                  * restart.  We may have "killed" the receiver thread just
2138                  * after drbdd_init() returned.  Typically, we should be
2139                  * C_STANDALONE already, now, and this becomes a no-op.
2140                  */
2141                 rv2 = conn_request_state(tconn, NS(conn, C_STANDALONE),
2142                                 CS_VERBOSE | CS_HARD);
2143                 if (rv2 < SS_SUCCESS)
2144                         conn_err(tconn,
2145                                 "unexpected rv2=%d in conn_try_disconnect()\n",
2146                                 rv2);
2147         }
2148         return rv;
2149 }
2150
2151 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2152 {
2153         struct disconnect_parms parms;
2154         struct drbd_tconn *tconn;
2155         enum drbd_state_rv rv;
2156         enum drbd_ret_code retcode;
2157         int err;
2158
2159         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2160         if (!adm_ctx.reply_skb)
2161                 return retcode;
2162         if (retcode != NO_ERROR)
2163                 goto fail;
2164
2165         tconn = adm_ctx.tconn;
2166         memset(&parms, 0, sizeof(parms));
2167         if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2168                 err = disconnect_parms_from_attrs(&parms, info);
2169                 if (err) {
2170                         retcode = ERR_MANDATORY_TAG;
2171                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2172                         goto fail;
2173                 }
2174         }
2175
2176         rv = conn_try_disconnect(tconn, parms.force_disconnect);
2177         if (rv < SS_SUCCESS)
2178                 retcode = rv;  /* FIXME: Type mismatch. */
2179         else
2180                 retcode = NO_ERROR;
2181  fail:
2182         drbd_adm_finish(info, retcode);
2183         return 0;
2184 }
2185
2186 void resync_after_online_grow(struct drbd_conf *mdev)
2187 {
2188         int iass; /* I am sync source */
2189
2190         dev_info(DEV, "Resync of new storage after online grow\n");
2191         if (mdev->state.role != mdev->state.peer)
2192                 iass = (mdev->state.role == R_PRIMARY);
2193         else
2194                 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
2195
2196         if (iass)
2197                 drbd_start_resync(mdev, C_SYNC_SOURCE);
2198         else
2199                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2200 }
2201
2202 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2203 {
2204         struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2205         struct resize_parms rs;
2206         struct drbd_conf *mdev;
2207         enum drbd_ret_code retcode;
2208         enum determine_dev_size dd;
2209         enum dds_flags ddsf;
2210         sector_t u_size;
2211         int err;
2212
2213         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2214         if (!adm_ctx.reply_skb)
2215                 return retcode;
2216         if (retcode != NO_ERROR)
2217                 goto fail;
2218
2219         memset(&rs, 0, sizeof(struct resize_parms));
2220         if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2221                 err = resize_parms_from_attrs(&rs, info);
2222                 if (err) {
2223                         retcode = ERR_MANDATORY_TAG;
2224                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2225                         goto fail;
2226                 }
2227         }
2228
2229         mdev = adm_ctx.mdev;
2230         if (mdev->state.conn > C_CONNECTED) {
2231                 retcode = ERR_RESIZE_RESYNC;
2232                 goto fail;
2233         }
2234
2235         if (mdev->state.role == R_SECONDARY &&
2236             mdev->state.peer == R_SECONDARY) {
2237                 retcode = ERR_NO_PRIMARY;
2238                 goto fail;
2239         }
2240
2241         if (!get_ldev(mdev)) {
2242                 retcode = ERR_NO_DISK;
2243                 goto fail;
2244         }
2245
2246         if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
2247                 retcode = ERR_NEED_APV_93;
2248                 goto fail;
2249         }
2250
2251         rcu_read_lock();
2252         u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
2253         rcu_read_unlock();
2254         if (u_size != (sector_t)rs.resize_size) {
2255                 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2256                 if (!new_disk_conf) {
2257                         retcode = ERR_NOMEM;
2258                         goto fail;
2259                 }
2260         }
2261
2262         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
2263                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2264
2265         if (new_disk_conf) {
2266                 mutex_lock(&mdev->tconn->conf_update);
2267                 old_disk_conf = mdev->ldev->disk_conf;
2268                 *new_disk_conf = *old_disk_conf;
2269                 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2270                 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
2271                 mutex_unlock(&mdev->tconn->conf_update);
2272                 synchronize_rcu();
2273                 kfree(old_disk_conf);
2274         }
2275
2276         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2277         dd = drbd_determine_dev_size(mdev, ddsf);
2278         drbd_md_sync(mdev);
2279         put_ldev(mdev);
2280         if (dd == dev_size_error) {
2281                 retcode = ERR_NOMEM_BITMAP;
2282                 goto fail;
2283         }
2284
2285         if (mdev->state.conn == C_CONNECTED) {
2286                 if (dd == grew)
2287                         set_bit(RESIZE_PENDING, &mdev->flags);
2288
2289                 drbd_send_uuids(mdev);
2290                 drbd_send_sizes(mdev, 1, ddsf);
2291         }
2292
2293  fail:
2294         drbd_adm_finish(info, retcode);
2295         return 0;
2296 }
2297
2298 void drbd_set_res_opts_defaults(struct res_opts *r)
2299 {
2300         return set_res_opts_defaults(r);
2301 }
2302
2303 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2304 {
2305         enum drbd_ret_code retcode;
2306         cpumask_var_t new_cpu_mask;
2307         struct drbd_tconn *tconn;
2308         struct res_opts res_opts;
2309         int err;
2310
2311         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2312         if (!adm_ctx.reply_skb)
2313                 return retcode;
2314         if (retcode != NO_ERROR)
2315                 goto fail;
2316         tconn = adm_ctx.tconn;
2317
2318         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
2319                 retcode = ERR_NOMEM;
2320                 drbd_msg_put_info("unable to allocate cpumask");
2321                 goto fail;
2322         }
2323
2324         res_opts = tconn->res_opts;
2325         if (should_set_defaults(info))
2326                 set_res_opts_defaults(&res_opts);
2327
2328         err = res_opts_from_attrs(&res_opts, info);
2329         if (err) {
2330                 retcode = ERR_MANDATORY_TAG;
2331                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2332                 goto fail;
2333         }
2334
2335         /* silently ignore cpu mask on UP kernel */
2336         if (nr_cpu_ids > 1 && res_opts.cpu_mask[0] != 0) {
2337                 err = __bitmap_parse(res_opts.cpu_mask, 32, 0,
2338                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2339                 if (err) {
2340                         conn_warn(tconn, "__bitmap_parse() failed with %d\n", err);
2341                         retcode = ERR_CPU_MASK_PARSE;
2342                         goto fail;
2343                 }
2344         }
2345
2346
2347         tconn->res_opts = res_opts;
2348
2349         if (!cpumask_equal(tconn->cpu_mask, new_cpu_mask)) {
2350                 cpumask_copy(tconn->cpu_mask, new_cpu_mask);
2351                 drbd_calc_cpu_mask(tconn);
2352                 tconn->receiver.reset_cpu_mask = 1;
2353                 tconn->asender.reset_cpu_mask = 1;
2354                 tconn->worker.reset_cpu_mask = 1;
2355         }
2356
2357 fail:
2358         free_cpumask_var(new_cpu_mask);
2359
2360         drbd_adm_finish(info, retcode);
2361         return 0;
2362 }
2363
2364 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2365 {
2366         struct drbd_conf *mdev;
2367         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2368
2369         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2370         if (!adm_ctx.reply_skb)
2371                 return retcode;
2372         if (retcode != NO_ERROR)
2373                 goto out;
2374
2375         mdev = adm_ctx.mdev;
2376
2377         /* If there is still bitmap IO pending, probably because of a previous
2378          * resync just being finished, wait for it before requesting a new resync. */
2379         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2380
2381         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2382
2383         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2384                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2385
2386         while (retcode == SS_NEED_CONNECTION) {
2387                 spin_lock_irq(&mdev->tconn->req_lock);
2388                 if (mdev->state.conn < C_CONNECTED)
2389                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2390                 spin_unlock_irq(&mdev->tconn->req_lock);
2391
2392                 if (retcode != SS_NEED_CONNECTION)
2393                         break;
2394
2395                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2396         }
2397
2398 out:
2399         drbd_adm_finish(info, retcode);
2400         return 0;
2401 }
2402
2403 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2404 {
2405         int rv;
2406
2407         rv = drbd_bmio_set_n_write(mdev);
2408         drbd_suspend_al(mdev);
2409         return rv;
2410 }
2411
2412 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2413                 union drbd_state mask, union drbd_state val)
2414 {
2415         enum drbd_ret_code retcode;
2416
2417         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2418         if (!adm_ctx.reply_skb)
2419                 return retcode;
2420         if (retcode != NO_ERROR)
2421                 goto out;
2422
2423         retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2424 out:
2425         drbd_adm_finish(info, retcode);
2426         return 0;
2427 }
2428
2429 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2430 {
2431         return drbd_adm_simple_request_state(skb, info, NS(conn, C_STARTING_SYNC_S));
2432 }
2433
2434 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2435 {
2436         enum drbd_ret_code retcode;
2437
2438         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2439         if (!adm_ctx.reply_skb)
2440                 return retcode;
2441         if (retcode != NO_ERROR)
2442                 goto out;
2443
2444         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2445                 retcode = ERR_PAUSE_IS_SET;
2446 out:
2447         drbd_adm_finish(info, retcode);
2448         return 0;
2449 }
2450
2451 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2452 {
2453         union drbd_dev_state s;
2454         enum drbd_ret_code retcode;
2455
2456         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2457         if (!adm_ctx.reply_skb)
2458                 return retcode;
2459         if (retcode != NO_ERROR)
2460                 goto out;
2461
2462         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2463                 s = adm_ctx.mdev->state;
2464                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2465                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2466                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2467                 } else {
2468                         retcode = ERR_PAUSE_IS_CLEAR;
2469                 }
2470         }
2471
2472 out:
2473         drbd_adm_finish(info, retcode);
2474         return 0;
2475 }
2476
2477 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2478 {
2479         return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2480 }
2481
2482 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2483 {
2484         struct drbd_conf *mdev;
2485         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2486
2487         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2488         if (!adm_ctx.reply_skb)
2489                 return retcode;
2490         if (retcode != NO_ERROR)
2491                 goto out;
2492
2493         mdev = adm_ctx.mdev;
2494         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2495                 drbd_uuid_new_current(mdev);
2496                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2497         }
2498         drbd_suspend_io(mdev);
2499         retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2500         if (retcode == SS_SUCCESS) {
2501                 if (mdev->state.conn < C_CONNECTED)
2502                         tl_clear(mdev->tconn);
2503                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2504                         tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2505         }
2506         drbd_resume_io(mdev);
2507
2508 out:
2509         drbd_adm_finish(info, retcode);
2510         return 0;
2511 }
2512
2513 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2514 {
2515         return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2516 }
2517
2518 int nla_put_drbd_cfg_context(struct sk_buff *skb, const char *conn_name, unsigned vnr)
2519 {
2520         struct nlattr *nla;
2521         nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2522         if (!nla)
2523                 goto nla_put_failure;
2524         if (vnr != VOLUME_UNSPECIFIED)
2525                 NLA_PUT_U32(skb, T_ctx_volume, vnr);
2526         NLA_PUT_STRING(skb, T_ctx_conn_name, conn_name);
2527         nla_nest_end(skb, nla);
2528         return 0;
2529
2530 nla_put_failure:
2531         if (nla)
2532                 nla_nest_cancel(skb, nla);
2533         return -EMSGSIZE;
2534 }
2535
2536 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2537                 const struct sib_info *sib)
2538 {
2539         struct state_info *si = NULL; /* for sizeof(si->member); */
2540         struct net_conf *nc;
2541         struct nlattr *nla;
2542         int got_ldev;
2543         int err = 0;
2544         int exclude_sensitive;
2545
2546         /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2547          * to.  So we better exclude_sensitive information.
2548          *
2549          * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2550          * in the context of the requesting user process. Exclude sensitive
2551          * information, unless current has superuser.
2552          *
2553          * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2554          * relies on the current implementation of netlink_dump(), which
2555          * executes the dump callback successively from netlink_recvmsg(),
2556          * always in the context of the receiving process */
2557         exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2558
2559         got_ldev = get_ldev(mdev);
2560
2561         /* We need to add connection name and volume number information still.
2562          * Minor number is in drbd_genlmsghdr. */
2563         if (nla_put_drbd_cfg_context(skb, mdev->tconn->name, mdev->vnr))
2564                 goto nla_put_failure;
2565
2566         if (res_opts_to_skb(skb, &mdev->tconn->res_opts, exclude_sensitive))
2567                 goto nla_put_failure;
2568
2569         rcu_read_lock();
2570         if (got_ldev)
2571                 if (disk_conf_to_skb(skb, rcu_dereference(mdev->ldev->disk_conf), exclude_sensitive))
2572                         goto nla_put_failure;
2573
2574         nc = rcu_dereference(mdev->tconn->net_conf);
2575         if (nc)
2576                 err = net_conf_to_skb(skb, nc, exclude_sensitive);
2577         rcu_read_unlock();
2578         if (err)
2579                 goto nla_put_failure;
2580
2581         nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2582         if (!nla)
2583                 goto nla_put_failure;
2584         NLA_PUT_U32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY);
2585         NLA_PUT_U32(skb, T_current_state, mdev->state.i);
2586         NLA_PUT_U64(skb, T_ed_uuid, mdev->ed_uuid);
2587         NLA_PUT_U64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev));
2588
2589         if (got_ldev) {
2590                 NLA_PUT_U32(skb, T_disk_flags, mdev->ldev->md.flags);
2591                 NLA_PUT(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2592                 NLA_PUT_U64(skb, T_bits_total, drbd_bm_bits(mdev));
2593                 NLA_PUT_U64(skb, T_bits_oos, drbd_bm_total_weight(mdev));
2594                 if (C_SYNC_SOURCE <= mdev->state.conn &&
2595                     C_PAUSED_SYNC_T >= mdev->state.conn) {
2596                         NLA_PUT_U64(skb, T_bits_rs_total, mdev->rs_total);
2597                         NLA_PUT_U64(skb, T_bits_rs_failed, mdev->rs_failed);
2598                 }
2599         }
2600
2601         if (sib) {
2602                 switch(sib->sib_reason) {
2603                 case SIB_SYNC_PROGRESS:
2604                 case SIB_GET_STATUS_REPLY:
2605                         break;
2606                 case SIB_STATE_CHANGE:
2607                         NLA_PUT_U32(skb, T_prev_state, sib->os.i);
2608                         NLA_PUT_U32(skb, T_new_state, sib->ns.i);
2609                         break;
2610                 case SIB_HELPER_POST:
2611                         NLA_PUT_U32(skb,
2612                                 T_helper_exit_code, sib->helper_exit_code);
2613                         /* fall through */
2614                 case SIB_HELPER_PRE:
2615                         NLA_PUT_STRING(skb, T_helper, sib->helper_name);
2616                         break;
2617                 }
2618         }
2619         nla_nest_end(skb, nla);
2620
2621         if (0)
2622 nla_put_failure:
2623                 err = -EMSGSIZE;
2624         if (got_ldev)
2625                 put_ldev(mdev);
2626         return err;
2627 }
2628
2629 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2630 {
2631         enum drbd_ret_code retcode;
2632         int err;
2633
2634         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2635         if (!adm_ctx.reply_skb)
2636                 return retcode;
2637         if (retcode != NO_ERROR)
2638                 goto out;
2639
2640         err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2641         if (err) {
2642                 nlmsg_free(adm_ctx.reply_skb);
2643                 return err;
2644         }
2645 out:
2646         drbd_adm_finish(info, retcode);
2647         return 0;
2648 }
2649
2650 int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
2651 {
2652         struct drbd_conf *mdev;
2653         struct drbd_genlmsghdr *dh;
2654         struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2655         struct drbd_tconn *tconn = NULL;
2656         struct drbd_tconn *tmp;
2657         unsigned volume = cb->args[1];
2658
2659         /* Open coded, deferred, iteration:
2660          * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2661          *      idr_for_each_entry(&tconn->volumes, mdev, i) {
2662          *        ...
2663          *      }
2664          * }
2665          * where tconn is cb->args[0];
2666          * and i is cb->args[1];
2667          *
2668          * cb->args[2] indicates if we shall loop over all resources,
2669          * or just dump all volumes of a single resource.
2670          *
2671          * This may miss entries inserted after this dump started,
2672          * or entries deleted before they are reached.
2673          *
2674          * We need to make sure the mdev won't disappear while
2675          * we are looking at it, and revalidate our iterators
2676          * on each iteration.
2677          */
2678
2679         /* synchronize with conn_create()/conn_destroy() */
2680         rcu_read_lock();
2681         /* revalidate iterator position */
2682         list_for_each_entry_rcu(tmp, &drbd_tconns, all_tconn) {
2683                 if (pos == NULL) {
2684                         /* first iteration */
2685                         pos = tmp;
2686                         tconn = pos;
2687                         break;
2688                 }
2689                 if (tmp == pos) {
2690                         tconn = pos;
2691                         break;
2692                 }
2693         }
2694         if (tconn) {
2695 next_tconn:
2696                 mdev = idr_get_next(&tconn->volumes, &volume);
2697                 if (!mdev) {
2698                         /* No more volumes to dump on this tconn.
2699                          * Advance tconn iterator. */
2700                         pos = list_entry_rcu(tconn->all_tconn.next,
2701                                              struct drbd_tconn, all_tconn);
2702                         /* Did we dump any volume on this tconn yet? */
2703                         if (volume != 0) {
2704                                 /* If we reached the end of the list,
2705                                  * or only a single resource dump was requested,
2706                                  * we are done. */
2707                                 if (&pos->all_tconn == &drbd_tconns || cb->args[2])
2708                                         goto out;
2709                                 volume = 0;
2710                                 tconn = pos;
2711                                 goto next_tconn;
2712                         }
2713                 }
2714
2715                 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).pid,
2716                                 cb->nlh->nlmsg_seq, &drbd_genl_family,
2717                                 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2718                 if (!dh)
2719                         goto out;
2720
2721                 if (!mdev) {
2722                         /* this is a tconn without a single volume */
2723                         dh->minor = -1U;
2724                         dh->ret_code = NO_ERROR;
2725                         if (nla_put_drbd_cfg_context(skb, tconn->name, VOLUME_UNSPECIFIED))
2726                                 genlmsg_cancel(skb, dh);
2727                         else
2728                                 genlmsg_end(skb, dh);
2729                         goto out;
2730                 }
2731
2732                 D_ASSERT(mdev->vnr == volume);
2733                 D_ASSERT(mdev->tconn == tconn);
2734
2735                 dh->minor = mdev_to_minor(mdev);
2736                 dh->ret_code = NO_ERROR;
2737
2738                 if (nla_put_status_info(skb, mdev, NULL)) {
2739                         genlmsg_cancel(skb, dh);
2740                         goto out;
2741                 }
2742                 genlmsg_end(skb, dh);
2743         }
2744
2745 out:
2746         rcu_read_unlock();
2747         /* where to start the next iteration */
2748         cb->args[0] = (long)pos;
2749         cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2750
2751         /* No more tconns/volumes/minors found results in an empty skb.
2752          * Which will terminate the dump. */
2753         return skb->len;
2754 }
2755
2756 /*
2757  * Request status of all resources, or of all volumes within a single resource.
2758  *
2759  * This is a dump, as the answer may not fit in a single reply skb otherwise.
2760  * Which means we cannot use the family->attrbuf or other such members, because
2761  * dump is NOT protected by the genl_lock().  During dump, we only have access
2762  * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
2763  *
2764  * Once things are setup properly, we call into get_one_status().
2765  */
2766 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2767 {
2768         const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
2769         struct nlattr *nla;
2770         const char *conn_name;
2771         struct drbd_tconn *tconn;
2772
2773         /* Is this a followup call? */
2774         if (cb->args[0]) {
2775                 /* ... of a single resource dump,
2776                  * and the resource iterator has been advanced already? */
2777                 if (cb->args[2] && cb->args[2] != cb->args[0])
2778                         return 0; /* DONE. */
2779                 goto dump;
2780         }
2781
2782         /* First call (from netlink_dump_start).  We need to figure out
2783          * which resource(s) the user wants us to dump. */
2784         nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
2785                         nlmsg_attrlen(cb->nlh, hdrlen),
2786                         DRBD_NLA_CFG_CONTEXT);
2787
2788         /* No explicit context given.  Dump all. */
2789         if (!nla)
2790                 goto dump;
2791         nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
2792         /* context given, but no name present? */
2793         if (!nla)
2794                 return -EINVAL;
2795         conn_name = nla_data(nla);
2796         tconn = conn_get_by_name(conn_name);
2797
2798         if (!tconn)
2799                 return -ENODEV;
2800
2801         kref_put(&tconn->kref, &conn_destroy); /* get_one_status() (re)validates tconn by itself */
2802
2803         /* prime iterators, and set "filter" mode mark:
2804          * only dump this tconn. */
2805         cb->args[0] = (long)tconn;
2806         /* cb->args[1] = 0; passed in this way. */
2807         cb->args[2] = (long)tconn;
2808
2809 dump:
2810         return get_one_status(skb, cb);
2811 }
2812
2813 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2814 {
2815         enum drbd_ret_code retcode;
2816         struct timeout_parms tp;
2817         int err;
2818
2819         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2820         if (!adm_ctx.reply_skb)
2821                 return retcode;
2822         if (retcode != NO_ERROR)
2823                 goto out;
2824
2825         tp.timeout_type =
2826                 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2827                 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2828                 UT_DEFAULT;
2829
2830         err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2831         if (err) {
2832                 nlmsg_free(adm_ctx.reply_skb);
2833                 return err;
2834         }
2835 out:
2836         drbd_adm_finish(info, retcode);
2837         return 0;
2838 }
2839
2840 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2841 {
2842         struct drbd_conf *mdev;
2843         enum drbd_ret_code retcode;
2844
2845         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2846         if (!adm_ctx.reply_skb)
2847                 return retcode;
2848         if (retcode != NO_ERROR)
2849                 goto out;
2850
2851         mdev = adm_ctx.mdev;
2852         if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2853                 /* resume from last known position, if possible */
2854                 struct start_ov_parms parms =
2855                         { .ov_start_sector = mdev->ov_start_sector };
2856                 int err = start_ov_parms_from_attrs(&parms, info);
2857                 if (err) {
2858                         retcode = ERR_MANDATORY_TAG;
2859                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2860                         goto out;
2861                 }
2862                 /* w_make_ov_request expects position to be aligned */
2863                 mdev->ov_start_sector = parms.ov_start_sector & ~BM_SECT_PER_BIT;
2864         }
2865         /* If there is still bitmap IO pending, e.g. previous resync or verify
2866          * just being finished, wait for it before requesting a new resync. */
2867         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2868         retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2869 out:
2870         drbd_adm_finish(info, retcode);
2871         return 0;
2872 }
2873
2874
2875 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
2876 {
2877         struct drbd_conf *mdev;
2878         enum drbd_ret_code retcode;
2879         int skip_initial_sync = 0;
2880         int err;
2881         struct new_c_uuid_parms args;
2882
2883         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2884         if (!adm_ctx.reply_skb)
2885                 return retcode;
2886         if (retcode != NO_ERROR)
2887                 goto out_nolock;
2888
2889         mdev = adm_ctx.mdev;
2890         memset(&args, 0, sizeof(args));
2891         if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
2892                 err = new_c_uuid_parms_from_attrs(&args, info);
2893                 if (err) {
2894                         retcode = ERR_MANDATORY_TAG;
2895                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2896                         goto out_nolock;
2897                 }
2898         }
2899
2900         mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2901
2902         if (!get_ldev(mdev)) {
2903                 retcode = ERR_NO_DISK;
2904                 goto out;
2905         }
2906
2907         /* this is "skip initial sync", assume to be clean */
2908         if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2909             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2910                 dev_info(DEV, "Preparing to skip initial sync\n");
2911                 skip_initial_sync = 1;
2912         } else if (mdev->state.conn != C_STANDALONE) {
2913                 retcode = ERR_CONNECTED;
2914                 goto out_dec;
2915         }
2916
2917         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2918         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2919
2920         if (args.clear_bm) {
2921                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2922                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2923                 if (err) {
2924                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2925                         retcode = ERR_IO_MD_DISK;
2926                 }
2927                 if (skip_initial_sync) {
2928                         drbd_send_uuids_skip_initial_sync(mdev);
2929                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2930                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2931                         spin_lock_irq(&mdev->tconn->req_lock);
2932                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2933                                         CS_VERBOSE, NULL);
2934                         spin_unlock_irq(&mdev->tconn->req_lock);
2935                 }
2936         }
2937
2938         drbd_md_sync(mdev);
2939 out_dec:
2940         put_ldev(mdev);
2941 out:
2942         mutex_unlock(mdev->state_mutex);
2943 out_nolock:
2944         drbd_adm_finish(info, retcode);
2945         return 0;
2946 }
2947
2948 static enum drbd_ret_code
2949 drbd_check_conn_name(const char *name)
2950 {
2951         if (!name || !name[0]) {
2952                 drbd_msg_put_info("connection name missing");
2953                 return ERR_MANDATORY_TAG;
2954         }
2955         /* if we want to use these in sysfs/configfs/debugfs some day,
2956          * we must not allow slashes */
2957         if (strchr(name, '/')) {
2958                 drbd_msg_put_info("invalid connection name");
2959                 return ERR_INVALID_REQUEST;
2960         }
2961         return NO_ERROR;
2962 }
2963
2964 int drbd_adm_create_connection(struct sk_buff *skb, struct genl_info *info)
2965 {
2966         enum drbd_ret_code retcode;
2967
2968         retcode = drbd_adm_prepare(skb, info, 0);
2969         if (!adm_ctx.reply_skb)
2970                 return retcode;
2971         if (retcode != NO_ERROR)
2972                 goto out;
2973
2974         retcode = drbd_check_conn_name(adm_ctx.conn_name);
2975         if (retcode != NO_ERROR)
2976                 goto out;
2977
2978         if (adm_ctx.tconn) {
2979                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
2980                         retcode = ERR_INVALID_REQUEST;
2981                         drbd_msg_put_info("connection exists");
2982                 }
2983                 /* else: still NO_ERROR */
2984                 goto out;
2985         }
2986
2987         if (!conn_create(adm_ctx.conn_name))
2988                 retcode = ERR_NOMEM;
2989 out:
2990         drbd_adm_finish(info, retcode);
2991         return 0;
2992 }
2993
2994 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
2995 {
2996         struct drbd_genlmsghdr *dh = info->userhdr;
2997         enum drbd_ret_code retcode;
2998
2999         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
3000         if (!adm_ctx.reply_skb)
3001                 return retcode;
3002         if (retcode != NO_ERROR)
3003                 goto out;
3004
3005         /* FIXME drop minor_count parameter, limit to MINORMASK */
3006         if (dh->minor >= minor_count) {
3007                 drbd_msg_put_info("requested minor out of range");
3008                 retcode = ERR_INVALID_REQUEST;
3009                 goto out;
3010         }
3011         if (adm_ctx.volume > DRBD_VOLUME_MAX) {
3012                 drbd_msg_put_info("requested volume id out of range");
3013                 retcode = ERR_INVALID_REQUEST;
3014                 goto out;
3015         }
3016
3017         /* drbd_adm_prepare made sure already
3018          * that mdev->tconn and mdev->vnr match the request. */
3019         if (adm_ctx.mdev) {
3020                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
3021                         retcode = ERR_MINOR_EXISTS;
3022                 /* else: still NO_ERROR */
3023                 goto out;
3024         }
3025
3026         retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
3027 out:
3028         drbd_adm_finish(info, retcode);
3029         return 0;
3030 }
3031
3032 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
3033 {
3034         if (mdev->state.disk == D_DISKLESS &&
3035             /* no need to be mdev->state.conn == C_STANDALONE &&
3036              * we may want to delete a minor from a live replication group.
3037              */
3038             mdev->state.role == R_SECONDARY) {
3039                 idr_remove(&mdev->tconn->volumes, mdev->vnr);
3040                 idr_remove(&minors, mdev_to_minor(mdev));
3041                 del_gendisk(mdev->vdisk);
3042                 synchronize_rcu();
3043                 kref_put(&mdev->kref, &drbd_minor_destroy);
3044                 return NO_ERROR;
3045         } else
3046                 return ERR_MINOR_CONFIGURED;
3047 }
3048
3049 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
3050 {
3051         enum drbd_ret_code retcode;
3052
3053         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3054         if (!adm_ctx.reply_skb)
3055                 return retcode;
3056         if (retcode != NO_ERROR)
3057                 goto out;
3058
3059         retcode = adm_delete_minor(adm_ctx.mdev);
3060 out:
3061         drbd_adm_finish(info, retcode);
3062         return 0;
3063 }
3064
3065 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
3066 {
3067         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3068         struct drbd_conf *mdev;
3069         unsigned i;
3070
3071         retcode = drbd_adm_prepare(skb, info, 0);
3072         if (!adm_ctx.reply_skb)
3073                 return retcode;
3074         if (retcode != NO_ERROR)
3075                 goto out;
3076
3077         if (!adm_ctx.tconn) {
3078                 retcode = ERR_CONN_NOT_KNOWN;
3079                 goto out;
3080         }
3081
3082         /* demote */
3083         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3084                 retcode = drbd_set_role(mdev, R_SECONDARY, 0);
3085                 if (retcode < SS_SUCCESS) {
3086                         drbd_msg_put_info("failed to demote");
3087                         goto out;
3088                 }
3089         }
3090
3091         retcode = conn_try_disconnect(adm_ctx.tconn, 0);
3092         if (retcode < SS_SUCCESS) {
3093                 drbd_msg_put_info("failed to disconnect");
3094                 goto out;
3095         }
3096
3097         /* detach */
3098         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3099                 retcode = adm_detach(mdev);
3100                 if (retcode < SS_SUCCESS) {
3101                         drbd_msg_put_info("failed to detach");
3102                         goto out;
3103                 }
3104         }
3105
3106         /* If we reach this, all volumes (of this tconn) are Secondary,
3107          * Disconnected, Diskless, aka Unconfigured. Make sure all threads have
3108          * actually stopped, state handling only does drbd_thread_stop_nowait(). */
3109         drbd_thread_stop(&adm_ctx.tconn->worker);
3110
3111         /* Now, nothing can fail anymore */
3112
3113         /* delete volumes */
3114         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3115                 retcode = adm_delete_minor(mdev);
3116                 if (retcode != NO_ERROR) {
3117                         /* "can not happen" */
3118                         drbd_msg_put_info("failed to delete volume");
3119                         goto out;
3120                 }
3121         }
3122
3123         /* delete connection */
3124         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3125                 list_del_rcu(&adm_ctx.tconn->all_tconn);
3126                 synchronize_rcu();
3127                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3128
3129                 retcode = NO_ERROR;
3130         } else {
3131                 /* "can not happen" */
3132                 retcode = ERR_CONN_IN_USE;
3133                 drbd_msg_put_info("failed to delete connection");
3134         }
3135         goto out;
3136 out:
3137         drbd_adm_finish(info, retcode);
3138         return 0;
3139 }
3140
3141 int drbd_adm_delete_connection(struct sk_buff *skb, struct genl_info *info)
3142 {
3143         enum drbd_ret_code retcode;
3144
3145         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
3146         if (!adm_ctx.reply_skb)
3147                 return retcode;
3148         if (retcode != NO_ERROR)
3149                 goto out;
3150
3151         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3152                 list_del_rcu(&adm_ctx.tconn->all_tconn);
3153                 synchronize_rcu();
3154                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3155
3156                 retcode = NO_ERROR;
3157         } else {
3158                 retcode = ERR_CONN_IN_USE;
3159         }
3160
3161         if (retcode == NO_ERROR)
3162                 drbd_thread_stop(&adm_ctx.tconn->worker);
3163 out:
3164         drbd_adm_finish(info, retcode);
3165         return 0;
3166 }
3167
3168 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
3169 {
3170         static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3171         struct sk_buff *msg;
3172         struct drbd_genlmsghdr *d_out;
3173         unsigned seq;
3174         int err = -ENOMEM;
3175
3176         seq = atomic_inc_return(&drbd_genl_seq);
3177         msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3178         if (!msg)
3179                 goto failed;
3180
3181         err = -EMSGSIZE;
3182         d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3183         if (!d_out) /* cannot happen, but anyways. */
3184                 goto nla_put_failure;
3185         d_out->minor = mdev_to_minor(mdev);
3186         d_out->ret_code = 0;
3187
3188         if (nla_put_status_info(msg, mdev, sib))
3189                 goto nla_put_failure;
3190         genlmsg_end(msg, d_out);
3191         err = drbd_genl_multicast_events(msg, 0);
3192         /* msg has been consumed or freed in netlink_broadcast() */
3193         if (err && err != -ESRCH)
3194                 goto failed;
3195
3196         return;
3197
3198 nla_put_failure:
3199         nlmsg_free(msg);
3200 failed:
3201         dev_err(DEV, "Error %d while broadcasting event. "
3202                         "Event seq:%u sib_reason:%u\n",
3203                         err, seq, sib->sib_reason);
3204 }