drivers/block/drbd/drbd_nl.c

   1 /*
   2    drbd_nl.c
   3
   4    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
   5
   6    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
   7    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
   8    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
   9
  10    drbd is free software; you can redistribute it and/or modify
  11    it under the terms of the GNU General Public License as published by
  12    the Free Software Foundation; either version 2, or (at your option)
  13    any later version.
  14
  15    drbd is distributed in the hope that it will be useful,
  16    but WITHOUT ANY WARRANTY; without even the implied warranty of
  17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18    GNU General Public License for more details.
  19
  20    You should have received a copy of the GNU General Public License
  21    along with drbd; see the file COPYING.  If not, write to
  22    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  23
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/drbd.h>
  28 #include <linux/in.h>
  29 #include <linux/fs.h>
  30 #include <linux/file.h>
  31 #include <linux/slab.h>
  32 #include <linux/blkpg.h>
  33 #include <linux/cpumask.h>
  34 #include "drbd_int.h"
  35 #include "drbd_req.h"
  36 #include "drbd_wrappers.h"
  37 #include <asm/unaligned.h>
  38 #include <linux/drbd_limits.h>
  39 #include <linux/kthread.h>
  40
  41 #include <net/genetlink.h>
  42
  43 /* .doit */
  44 // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info);
  45 // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info);
  46
  47 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info);
  48 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info);
  49
  50 int drbd_adm_create_connection(struct sk_buff *skb, struct genl_info *info);
  51 int drbd_adm_delete_connection(struct sk_buff *skb, struct genl_info *info);
  52 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info);
  53
  54 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info);
  55 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info);
  56 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info);
  57 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info);
  58 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info);
  59 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info);
  60 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info);
  61 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info);
  62 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info);
  63 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info);
  64 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info);
  65 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info);
  66 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info);
  67 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info);
  68 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info);
  69 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info);
  70 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info);
  71 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info);
  72 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info);
  73 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info);
  74 /* .dumpit */
  75 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb);
  76
  77 #include <linux/drbd_genl_api.h>
  78 #include <linux/genl_magic_func.h>
  79
  80 /* used blkdev_get_by_path, to claim our meta data device(s) */
  81 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
  82
  83 /* Configuration is strictly serialized, because generic netlink message
  84  * processing is strictly serialized by the genl_lock().
  85  * Which means we can use one static global drbd_config_context struct.
  86  */
  87 static struct drbd_config_context {
  88         /* assigned from drbd_genlmsghdr */
  89         unsigned int minor;
  90         /* assigned from request attributes, if present */
  91         unsigned int volume;
  92 #define VOLUME_UNSPECIFIED              (-1U)
  93         /* pointer into the request skb,
  94          * limited lifetime! */
  95         char *conn_name;
  96
  97         /* reply buffer */
  98         struct sk_buff *reply_skb;
  99         /* pointer into reply buffer */
 100         struct drbd_genlmsghdr *reply_dh;
 101         /* resolved from attributes, if possible */
 102         struct drbd_conf *mdev;
 103         struct drbd_tconn *tconn;
 104 } adm_ctx;
 105
 106 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
 107 {
 108         genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
 109         if (genlmsg_reply(skb, info))
 110                 printk(KERN_ERR "drbd: error sending genl reply\n");
 111 }
 112
 113 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
 114  * reason it could fail was no space in skb, and there are 4k available. */
 115 int drbd_msg_put_info(const char *info)
 116 {
 117         struct sk_buff *skb = adm_ctx.reply_skb;
 118         struct nlattr *nla;
 119         int err = -EMSGSIZE;
 120
 121         if (!info || !info[0])
 122                 return 0;
 123
 124         nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY);
 125         if (!nla)
 126                 return err;
 127
 128         err = nla_put_string(skb, T_info_text, info);
 129         if (err) {
 130                 nla_nest_cancel(skb, nla);
 131                 return err;
 132         } else
 133                 nla_nest_end(skb, nla);
 134         return 0;
 135 }
 136
 137 /* This would be a good candidate for a "pre_doit" hook,
 138  * and per-family private info->pointers.
 139  * But we need to stay compatible with older kernels.
 140  * If it returns successfully, adm_ctx members are valid.
 141  */
 142 #define DRBD_ADM_NEED_MINOR     1
 143 #define DRBD_ADM_NEED_CONN      2
 144 static int drbd_adm_prepare(struct sk_buff *skb, struct genl_info *info,
 145                 unsigned flags)
 146 {
 147         struct drbd_genlmsghdr *d_in = info->userhdr;
 148         const u8 cmd = info->genlhdr->cmd;
 149         int err;
 150
 151         memset(&adm_ctx, 0, sizeof(adm_ctx));
 152
 153         /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
 154         if (cmd != DRBD_ADM_GET_STATUS
 155         && security_netlink_recv(skb, CAP_SYS_ADMIN))
 156                return -EPERM;
 157
 158         adm_ctx.reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
 159         if (!adm_ctx.reply_skb)
 160                 goto fail;
 161
 162         adm_ctx.reply_dh = genlmsg_put_reply(adm_ctx.reply_skb,
 163                                         info, &drbd_genl_family, 0, cmd);
 164         /* put of a few bytes into a fresh skb of >= 4k will always succeed.
 165          * but anyways */
 166         if (!adm_ctx.reply_dh)
 167                 goto fail;
 168
 169         adm_ctx.reply_dh->minor = d_in->minor;
 170         adm_ctx.reply_dh->ret_code = NO_ERROR;
 171
 172         if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
 173                 struct nlattr *nla;
 174                 /* parse and validate only */
 175                 err = drbd_cfg_context_from_attrs(NULL, info);
 176                 if (err)
 177                         goto fail;
 178
 179                 /* It was present, and valid,
 180                  * copy it over to the reply skb. */
 181                 err = nla_put_nohdr(adm_ctx.reply_skb,
 182                                 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
 183                                 info->attrs[DRBD_NLA_CFG_CONTEXT]);
 184                 if (err)
 185                         goto fail;
 186
 187                 /* and assign stuff to the global adm_ctx */
 188                 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
 189                 adm_ctx.volume = nla ? nla_get_u32(nla) : VOLUME_UNSPECIFIED;
 190                 nla = nested_attr_tb[__nla_type(T_ctx_conn_name)];
 191                 if (nla)
 192                         adm_ctx.conn_name = nla_data(nla);
 193         } else
 194                 adm_ctx.volume = VOLUME_UNSPECIFIED;
 195
 196         adm_ctx.minor = d_in->minor;
 197         adm_ctx.mdev = minor_to_mdev(d_in->minor);
 198         adm_ctx.tconn = conn_get_by_name(adm_ctx.conn_name);
 199
 200         if (!adm_ctx.mdev && (flags & DRBD_ADM_NEED_MINOR)) {
 201                 drbd_msg_put_info("unknown minor");
 202                 return ERR_MINOR_INVALID;
 203         }
 204         if (!adm_ctx.tconn && (flags & DRBD_ADM_NEED_CONN)) {
 205                 drbd_msg_put_info("unknown connection");
 206                 return ERR_INVALID_REQUEST;
 207         }
 208
 209         /* some more paranoia, if the request was over-determined */
 210         if (adm_ctx.mdev && adm_ctx.tconn &&
 211             adm_ctx.mdev->tconn != adm_ctx.tconn) {
 212                 pr_warning("request: minor=%u, conn=%s; but that minor belongs to connection %s\n",
 213                                 adm_ctx.minor, adm_ctx.conn_name, adm_ctx.mdev->tconn->name);
 214                 drbd_msg_put_info("minor exists in different connection");
 215                 return ERR_INVALID_REQUEST;
 216         }
 217         if (adm_ctx.mdev &&
 218             adm_ctx.volume != VOLUME_UNSPECIFIED &&
 219             adm_ctx.volume != adm_ctx.mdev->vnr) {
 220                 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
 221                                 adm_ctx.minor, adm_ctx.volume,
 222                                 adm_ctx.mdev->vnr, adm_ctx.mdev->tconn->name);
 223                 drbd_msg_put_info("minor exists as different volume");
 224                 return ERR_INVALID_REQUEST;
 225         }
 226
 227         return NO_ERROR;
 228
 229 fail:
 230         nlmsg_free(adm_ctx.reply_skb);
 231         adm_ctx.reply_skb = NULL;
 232         return -ENOMEM;
 233 }
 234
 235 static int drbd_adm_finish(struct genl_info *info, int retcode)
 236 {
 237         struct nlattr *nla;
 238         const char *conn_name = NULL;
 239
 240         if (adm_ctx.tconn) {
 241                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
 242                 adm_ctx.tconn = NULL;
 243         }
 244
 245         if (!adm_ctx.reply_skb)
 246                 return -ENOMEM;
 247
 248         adm_ctx.reply_dh->ret_code = retcode;
 249
 250         nla = info->attrs[DRBD_NLA_CFG_CONTEXT];
 251         if (nla) {
 252                 nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
 253                 if (nla)
 254                         conn_name = nla_data(nla);
 255         }
 256
 257         drbd_adm_send_reply(adm_ctx.reply_skb, info);
 258         return 0;
 259 }
 260
 261 static void setup_khelper_env(struct drbd_tconn *tconn, char **envp)
 262 {
 263         char *afs;
 264         struct net_conf *nc;
 265
 266         rcu_read_lock();
 267         nc = rcu_dereference(tconn->net_conf);
 268         if (nc) {
 269                 switch (((struct sockaddr *)nc->peer_addr)->sa_family) {
 270                 case AF_INET6:
 271                         afs = "ipv6";
 272                         snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
 273                                  &((struct sockaddr_in6 *)nc->peer_addr)->sin6_addr);
 274                         break;
 275                 case AF_INET:
 276                         afs = "ipv4";
 277                         snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
 278                                  &((struct sockaddr_in *)nc->peer_addr)->sin_addr);
 279                         break;
 280                 default:
 281                         afs = "ssocks";
 282                         snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
 283                                  &((struct sockaddr_in *)nc->peer_addr)->sin_addr);
 284                 }
 285                 snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
 286         }
 287         rcu_read_unlock();
 288 }
 289
 290 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
 291 {
 292         char *envp[] = { "HOME=/",
 293                         "TERM=linux",
 294                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 295                          (char[20]) { }, /* address family */
 296                          (char[60]) { }, /* address */
 297                         NULL };
 298         char mb[12];
 299         char *argv[] = {usermode_helper, cmd, mb, NULL };
 300         struct sib_info sib;
 301         int ret;
 302
 303         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 304         setup_khelper_env(mdev->tconn, envp);
 305
 306         /* The helper may take some time.
 307          * write out any unsynced meta data changes now */
 308         drbd_md_sync(mdev);
 309
 310         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 311         sib.sib_reason = SIB_HELPER_PRE;
 312         sib.helper_name = cmd;
 313         drbd_bcast_event(mdev, &sib);
 314         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 315         if (ret)
 316                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 317                                 usermode_helper, cmd, mb,
 318                                 (ret >> 8) & 0xff, ret);
 319         else
 320                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 321                                 usermode_helper, cmd, mb,
 322                                 (ret >> 8) & 0xff, ret);
 323         sib.sib_reason = SIB_HELPER_POST;
 324         sib.helper_exit_code = ret;
 325         drbd_bcast_event(mdev, &sib);
 326
 327         if (ret < 0) /* Ignore any ERRNOs we got. */
 328                 ret = 0;
 329
 330         return ret;
 331 }
 332
 333 static void conn_md_sync(struct drbd_tconn *tconn)
 334 {
 335         struct drbd_conf *mdev;
 336         int vnr;
 337
 338         down_read(&drbd_cfg_rwsem);
 339         idr_for_each_entry(&tconn->volumes, mdev, vnr)
 340                 drbd_md_sync(mdev);
 341         up_read(&drbd_cfg_rwsem);
 342 }
 343
 344 int conn_khelper(struct drbd_tconn *tconn, char *cmd)
 345 {
 346         char *envp[] = { "HOME=/",
 347                         "TERM=linux",
 348                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 349                          (char[20]) { }, /* address family */
 350                          (char[60]) { }, /* address */
 351                         NULL };
 352         char *argv[] = {usermode_helper, cmd, tconn->name, NULL };
 353         int ret;
 354
 355         setup_khelper_env(tconn, envp);
 356         conn_md_sync(tconn);
 357
 358         conn_info(tconn, "helper command: %s %s %s\n", usermode_helper, cmd, tconn->name);
 359         /* TODO: conn_bcast_event() ?? */
 360
 361         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 362         if (ret)
 363                 conn_warn(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
 364                           usermode_helper, cmd, tconn->name,
 365                           (ret >> 8) & 0xff, ret);
 366         else
 367                 conn_info(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
 368                           usermode_helper, cmd, tconn->name,
 369                           (ret >> 8) & 0xff, ret);
 370         /* TODO: conn_bcast_event() ?? */
 371
 372         if (ret < 0) /* Ignore any ERRNOs we got. */
 373                 ret = 0;
 374
 375         return ret;
 376 }
 377
 378 static enum drbd_fencing_p highest_fencing_policy(struct drbd_tconn *tconn)
 379 {
 380         enum drbd_fencing_p fp = FP_NOT_AVAIL;
 381         struct drbd_conf *mdev;
 382         int vnr;
 383
 384         rcu_read_lock();
 385         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
 386                 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 387                         fp = max_t(enum drbd_fencing_p, fp,
 388                                    rcu_dereference(mdev->ldev->disk_conf)->fencing);
 389                         put_ldev(mdev);
 390                 }
 391         }
 392         rcu_read_unlock();
 393
 394         return fp;
 395 }
 396
 397 bool conn_try_outdate_peer(struct drbd_tconn *tconn)
 398 {
 399         union drbd_state mask = { };
 400         union drbd_state val = { };
 401         enum drbd_fencing_p fp;
 402         char *ex_to_string;
 403         int r;
 404
 405         if (tconn->cstate >= C_WF_REPORT_PARAMS) {
 406                 conn_err(tconn, "Expected cstate < C_WF_REPORT_PARAMS\n");
 407                 return false;
 408         }
 409
 410         fp = highest_fencing_policy(tconn);
 411         switch (fp) {
 412         case FP_NOT_AVAIL:
 413                 conn_warn(tconn, "Not fencing peer, I'm not even Consistent myself.\n");
 414                 goto out;
 415         case FP_DONT_CARE:
 416                 return true;
 417         default: ;
 418         }
 419
 420         r = conn_khelper(tconn, "fence-peer");
 421
 422         switch ((r>>8) & 0xff) {
 423         case 3: /* peer is inconsistent */
 424                 ex_to_string = "peer is inconsistent or worse";
 425                 mask.pdsk = D_MASK;
 426                 val.pdsk = D_INCONSISTENT;
 427                 break;
 428         case 4: /* peer got outdated, or was already outdated */
 429                 ex_to_string = "peer was fenced";
 430                 mask.pdsk = D_MASK;
 431                 val.pdsk = D_OUTDATED;
 432                 break;
 433         case 5: /* peer was down */
 434                 if (conn_highest_disk(tconn) == D_UP_TO_DATE) {
 435                         /* we will(have) create(d) a new UUID anyways... */
 436                         ex_to_string = "peer is unreachable, assumed to be dead";
 437                         mask.pdsk = D_MASK;
 438                         val.pdsk = D_OUTDATED;
 439                 } else {
 440                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 441                 }
 442                 break;
 443         case 6: /* Peer is primary, voluntarily outdate myself.
 444                  * This is useful when an unconnected R_SECONDARY is asked to
 445                  * become R_PRIMARY, but finds the other peer being active. */
 446                 ex_to_string = "peer is active";
 447                 conn_warn(tconn, "Peer is primary, outdating myself.\n");
 448                 mask.disk = D_MASK;
 449                 val.disk = D_OUTDATED;
 450                 break;
 451         case 7:
 452                 if (fp != FP_STONITH)
 453                         conn_err(tconn, "fence-peer() = 7 && fencing != Stonith !!!\n");
 454                 ex_to_string = "peer was stonithed";
 455                 mask.pdsk = D_MASK;
 456                 val.pdsk = D_OUTDATED;
 457                 break;
 458         default:
 459                 /* The script is broken ... */
 460                 conn_err(tconn, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 461                 return false; /* Eventually leave IO frozen */
 462         }
 463
 464         conn_info(tconn, "fence-peer helper returned %d (%s)\n",
 465                   (r>>8) & 0xff, ex_to_string);
 466
 467  out:
 468
 469         /* Not using
 470            conn_request_state(tconn, mask, val, CS_VERBOSE);
 471            here, because we might were able to re-establish the connection in the
 472            meantime. */
 473         spin_lock_irq(&tconn->req_lock);
 474         if (tconn->cstate < C_WF_REPORT_PARAMS)
 475                 _conn_request_state(tconn, mask, val, CS_VERBOSE);
 476         spin_unlock_irq(&tconn->req_lock);
 477
 478         return conn_highest_pdsk(tconn) <= D_OUTDATED;
 479 }
 480
 481 static int _try_outdate_peer_async(void *data)
 482 {
 483         struct drbd_tconn *tconn = (struct drbd_tconn *)data;
 484
 485         conn_try_outdate_peer(tconn);
 486
 487         kref_put(&tconn->kref, &conn_destroy);
 488         return 0;
 489 }
 490
 491 void conn_try_outdate_peer_async(struct drbd_tconn *tconn)
 492 {
 493         struct task_struct *opa;
 494
 495         kref_get(&tconn->kref);
 496         opa = kthread_run(_try_outdate_peer_async, tconn, "drbd_async_h");
 497         if (IS_ERR(opa)) {
 498                 conn_err(tconn, "out of mem, failed to invoke fence-peer helper\n");
 499                 kref_put(&tconn->kref, &conn_destroy);
 500         }
 501 }
 502
 503 enum drbd_state_rv
 504 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 505 {
 506         const int max_tries = 4;
 507         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
 508         struct net_conf *nc;
 509         int try = 0;
 510         int forced = 0;
 511         union drbd_state mask, val;
 512
 513         if (new_role == R_PRIMARY)
 514                 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
 515
 516         mutex_lock(mdev->state_mutex);
 517
 518         mask.i = 0; mask.role = R_MASK;
 519         val.i  = 0; val.role  = new_role;
 520
 521         while (try++ < max_tries) {
 522                 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 523
 524                 /* in case we first succeeded to outdate,
 525                  * but now suddenly could establish a connection */
 526                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 527                         val.pdsk = 0;
 528                         mask.pdsk = 0;
 529                         continue;
 530                 }
 531
 532                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
 533                     (mdev->state.disk < D_UP_TO_DATE &&
 534                      mdev->state.disk >= D_INCONSISTENT)) {
 535                         mask.disk = D_MASK;
 536                         val.disk  = D_UP_TO_DATE;
 537                         forced = 1;
 538                         continue;
 539                 }
 540
 541                 if (rv == SS_NO_UP_TO_DATE_DISK &&
 542                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 543                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 544
 545                         if (conn_try_outdate_peer(mdev->tconn)) {
 546                                 val.disk = D_UP_TO_DATE;
 547                                 mask.disk = D_MASK;
 548                         }
 549                         continue;
 550                 }
 551
 552                 if (rv == SS_NOTHING_TO_DO)
 553                         goto out;
 554                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
 555                         if (!conn_try_outdate_peer(mdev->tconn) && force) {
 556                                 dev_warn(DEV, "Forced into split brain situation!\n");
 557                                 mask.pdsk = D_MASK;
 558                                 val.pdsk  = D_OUTDATED;
 559
 560                         }
 561                         continue;
 562                 }
 563                 if (rv == SS_TWO_PRIMARIES) {
 564                         /* Maybe the peer is detected as dead very soon...
 565                            retry at most once more in this case. */
 566                         int timeo;
 567                         rcu_read_lock();
 568                         nc = rcu_dereference(mdev->tconn->net_conf);
 569                         timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
 570                         rcu_read_unlock();
 571                         schedule_timeout_interruptible(timeo);
 572                         if (try < max_tries)
 573                                 try = max_tries - 1;
 574                         continue;
 575                 }
 576                 if (rv < SS_SUCCESS) {
 577                         rv = _drbd_request_state(mdev, mask, val,
 578                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 579                         if (rv < SS_SUCCESS)
 580                                 goto out;
 581                 }
 582                 break;
 583         }
 584
 585         if (rv < SS_SUCCESS)
 586                 goto out;
 587
 588         if (forced)
 589                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 590
 591         /* Wait until nothing is on the fly :) */
 592         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 593
 594         if (new_role == R_SECONDARY) {
 595                 set_disk_ro(mdev->vdisk, true);
 596                 if (get_ldev(mdev)) {
 597                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 598                         put_ldev(mdev);
 599                 }
 600         } else {
 601                 mutex_lock(&mdev->tconn->conf_update);
 602                 nc = mdev->tconn->net_conf;
 603                 if (nc)
 604                         nc->want_lose = 0; /* without copy; single bit op is atomic */
 605                 mutex_unlock(&mdev->tconn->conf_update);
 606
 607                 set_disk_ro(mdev->vdisk, false);
 608                 if (get_ldev(mdev)) {
 609                         if (((mdev->state.conn < C_CONNECTED ||
 610                                mdev->state.pdsk <= D_FAILED)
 611                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 612                                 drbd_uuid_new_current(mdev);
 613
 614                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 615                         put_ldev(mdev);
 616                 }
 617         }
 618
 619         /* writeout of activity log covered areas of the bitmap
 620          * to stable storage done in after state change already */
 621
 622         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 623                 /* if this was forced, we should consider sync */
 624                 if (forced)
 625                         drbd_send_uuids(mdev);
 626                 drbd_send_state(mdev);
 627         }
 628
 629         drbd_md_sync(mdev);
 630
 631         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 632 out:
 633         mutex_unlock(mdev->state_mutex);
 634         return rv;
 635 }
 636
 637 static const char *from_attrs_err_to_txt(int err)
 638 {
 639         return  err == -ENOMSG ? "required attribute missing" :
 640                 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
 641                 err == -EEXIST ? "can not change invariant setting" :
 642                 "invalid attribute value";
 643 }
 644
 645 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
 646 {
 647         struct set_role_parms parms;
 648         int err;
 649         enum drbd_ret_code retcode;
 650
 651         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
 652         if (!adm_ctx.reply_skb)
 653                 return retcode;
 654         if (retcode != NO_ERROR)
 655                 goto out;
 656
 657         memset(&parms, 0, sizeof(parms));
 658         if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
 659                 err = set_role_parms_from_attrs(&parms, info);
 660                 if (err) {
 661                         retcode = ERR_MANDATORY_TAG;
 662                         drbd_msg_put_info(from_attrs_err_to_txt(err));
 663                         goto out;
 664                 }
 665         }
 666
 667         if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
 668                 retcode = drbd_set_role(adm_ctx.mdev, R_PRIMARY, parms.assume_uptodate);
 669         else
 670                 retcode = drbd_set_role(adm_ctx.mdev, R_SECONDARY, 0);
 671 out:
 672         drbd_adm_finish(info, retcode);
 673         return 0;
 674 }
 675
 676 /* initializes the md.*_offset members, so we are able to find
 677  * the on disk meta data */
 678 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 679                                        struct drbd_backing_dev *bdev)
 680 {
 681         sector_t md_size_sect = 0;
 682         int meta_dev_idx;
 683
 684         rcu_read_lock();
 685         meta_dev_idx = rcu_dereference(bdev->disk_conf)->meta_dev_idx;
 686
 687         switch (meta_dev_idx) {
 688         default:
 689                 /* v07 style fixed size indexed meta data */
 690                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 691                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 692                 bdev->md.al_offset = MD_AL_OFFSET;
 693                 bdev->md.bm_offset = MD_BM_OFFSET;
 694                 break;
 695         case DRBD_MD_INDEX_FLEX_EXT:
 696                 /* just occupy the full device; unit: sectors */
 697                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 698                 bdev->md.md_offset = 0;
 699                 bdev->md.al_offset = MD_AL_OFFSET;
 700                 bdev->md.bm_offset = MD_BM_OFFSET;
 701                 break;
 702         case DRBD_MD_INDEX_INTERNAL:
 703         case DRBD_MD_INDEX_FLEX_INT:
 704                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 705                 /* al size is still fixed */
 706                 bdev->md.al_offset = -MD_AL_SECTORS;
 707                 /* we need (slightly less than) ~ this much bitmap sectors: */
 708                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 709                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 710                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 711                 md_size_sect = ALIGN(md_size_sect, 8);
 712
 713                 /* plus the "drbd meta data super block",
 714                  * and the activity log; */
 715                 md_size_sect += MD_BM_OFFSET;
 716
 717                 bdev->md.md_size_sect = md_size_sect;
 718                 /* bitmap offset is adjusted by 'super' block size */
 719                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 720                 break;
 721         }
 722         rcu_read_unlock();
 723 }
 724
 725 /* input size is expected to be in KB */
 726 char *ppsize(char *buf, unsigned long long size)
 727 {
 728         /* Needs 9 bytes at max including trailing NUL:
 729          * -1ULL ==> "16384 EB" */
 730         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 731         int base = 0;
 732         while (size >= 10000 && base < sizeof(units)-1) {
 733                 /* shift + round */
 734                 size = (size >> 10) + !!(size & (1<<9));
 735                 base++;
 736         }
 737         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
 738
 739         return buf;
 740 }
 741
 742 /* there is still a theoretical deadlock when called from receiver
 743  * on an D_INCONSISTENT R_PRIMARY:
 744  *  remote READ does inc_ap_bio, receiver would need to receive answer
 745  *  packet from remote to dec_ap_bio again.
 746  *  receiver receive_sizes(), comes here,
 747  *  waits for ap_bio_cnt == 0. -> deadlock.
 748  * but this cannot happen, actually, because:
 749  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 750  *  (not connected, or bad/no disk on peer):
 751  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 752  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 753  *  peer may not initiate a resize.
 754  */
 755 /* Note these are not to be confused with
 756  * drbd_adm_suspend_io/drbd_adm_resume_io,
 757  * which are (sub) state changes triggered by admin (drbdsetup),
 758  * and can be long lived.
 759  * This changes an mdev->flag, is triggered by drbd internals,
 760  * and should be short-lived. */
 761 void drbd_suspend_io(struct drbd_conf *mdev)
 762 {
 763         set_bit(SUSPEND_IO, &mdev->flags);
 764         if (drbd_suspended(mdev))
 765                 return;
 766         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 767 }
 768
 769 void drbd_resume_io(struct drbd_conf *mdev)
 770 {
 771         clear_bit(SUSPEND_IO, &mdev->flags);
 772         wake_up(&mdev->misc_wait);
 773 }
 774
 775 /**
 776  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 777  * @mdev:       DRBD device.
 778  *
 779  * Returns 0 on success, negative return values indicate errors.
 780  * You should call drbd_md_sync() after calling this function.
 781  */
 782 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 783 {
 784         sector_t prev_first_sect, prev_size; /* previous meta location */
 785         sector_t la_size, u_size;
 786         sector_t size;
 787         char ppb[10];
 788
 789         int md_moved, la_size_changed;
 790         enum determine_dev_size rv = unchanged;
 791
 792         /* race:
 793          * application request passes inc_ap_bio,
 794          * but then cannot get an AL-reference.
 795          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 796          *
 797          * to avoid that:
 798          * Suspend IO right here.
 799          * still lock the act_log to not trigger ASSERTs there.
 800          */
 801         drbd_suspend_io(mdev);
 802
 803         /* no wait necessary anymore, actually we could assert that */
 804         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 805
 806         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 807         prev_size = mdev->ldev->md.md_size_sect;
 808         la_size = mdev->ldev->md.la_size_sect;
 809
 810         /* TODO: should only be some assert here, not (re)init... */
 811         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 812
 813         rcu_read_lock();
 814         u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
 815         rcu_read_unlock();
 816         size = drbd_new_dev_size(mdev, mdev->ldev, u_size, flags & DDSF_FORCED);
 817
 818         if (drbd_get_capacity(mdev->this_bdev) != size ||
 819             drbd_bm_capacity(mdev) != size) {
 820                 int err;
 821                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 822                 if (unlikely(err)) {
 823                         /* currently there is only one error: ENOMEM! */
 824                         size = drbd_bm_capacity(mdev)>>1;
 825                         if (size == 0) {
 826                                 dev_err(DEV, "OUT OF MEMORY! "
 827                                     "Could not allocate bitmap!\n");
 828                         } else {
 829                                 dev_err(DEV, "BM resizing failed. "
 830                                     "Leaving size unchanged at size = %lu KB\n",
 831                                     (unsigned long)size);
 832                         }
 833                         rv = dev_size_error;
 834                 }
 835                 /* racy, see comments above. */
 836                 drbd_set_my_capacity(mdev, size);
 837                 mdev->ldev->md.la_size_sect = size;
 838                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 839                      (unsigned long long)size>>1);
 840         }
 841         if (rv == dev_size_error)
 842                 goto out;
 843
 844         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 845
 846         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 847                 || prev_size       != mdev->ldev->md.md_size_sect;
 848
 849         if (la_size_changed || md_moved) {
 850                 int err;
 851
 852                 drbd_al_shrink(mdev); /* All extents inactive. */
 853                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 854                          la_size_changed && md_moved ? "size changed and md moved" :
 855                          la_size_changed ? "size changed" : "md moved");
 856                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
 857                 err = drbd_bitmap_io(mdev, &drbd_bm_write,
 858                                 "size changed", BM_LOCKED_MASK);
 859                 if (err) {
 860                         rv = dev_size_error;
 861                         goto out;
 862                 }
 863                 drbd_md_mark_dirty(mdev);
 864         }
 865
 866         if (size > la_size)
 867                 rv = grew;
 868         if (size < la_size)
 869                 rv = shrunk;
 870 out:
 871         lc_unlock(mdev->act_log);
 872         wake_up(&mdev->al_wait);
 873         drbd_resume_io(mdev);
 874
 875         return rv;
 876 }
 877
 878 sector_t
 879 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
 880                   sector_t u_size, int assume_peer_has_space)
 881 {
 882         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 883         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 884         sector_t m_size; /* my size */
 885         sector_t size = 0;
 886
 887         m_size = drbd_get_max_capacity(bdev);
 888
 889         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 890                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 891                 p_size = m_size;
 892         }
 893
 894         if (p_size && m_size) {
 895                 size = min_t(sector_t, p_size, m_size);
 896         } else {
 897                 if (la_size) {
 898                         size = la_size;
 899                         if (m_size && m_size < size)
 900                                 size = m_size;
 901                         if (p_size && p_size < size)
 902                                 size = p_size;
 903                 } else {
 904                         if (m_size)
 905                                 size = m_size;
 906                         if (p_size)
 907                                 size = p_size;
 908                 }
 909         }
 910
 911         if (size == 0)
 912                 dev_err(DEV, "Both nodes diskless!\n");
 913
 914         if (u_size) {
 915                 if (u_size > size)
 916                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 917                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 918                 else
 919                         size = u_size;
 920         }
 921
 922         return size;
 923 }
 924
 925 /**
 926  * drbd_check_al_size() - Ensures that the AL is of the right size
 927  * @mdev:       DRBD device.
 928  *
 929  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 930  * failed, and 0 on success. You should call drbd_md_sync() after you called
 931  * this function.
 932  */
 933 static int drbd_check_al_size(struct drbd_conf *mdev, struct disk_conf *dc)
 934 {
 935         struct lru_cache *n, *t;
 936         struct lc_element *e;
 937         unsigned int in_use;
 938         int i;
 939
 940         if (mdev->act_log &&
 941             mdev->act_log->nr_elements == dc->al_extents)
 942                 return 0;
 943
 944         in_use = 0;
 945         t = mdev->act_log;
 946         n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
 947                 dc->al_extents, sizeof(struct lc_element), 0);
 948
 949         if (n == NULL) {
 950                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 951                 return -ENOMEM;
 952         }
 953         spin_lock_irq(&mdev->al_lock);
 954         if (t) {
 955                 for (i = 0; i < t->nr_elements; i++) {
 956                         e = lc_element_by_index(t, i);
 957                         if (e->refcnt)
 958                                 dev_err(DEV, "refcnt(%d)==%d\n",
 959                                     e->lc_number, e->refcnt);
 960                         in_use += e->refcnt;
 961                 }
 962         }
 963         if (!in_use)
 964                 mdev->act_log = n;
 965         spin_unlock_irq(&mdev->al_lock);
 966         if (in_use) {
 967                 dev_err(DEV, "Activity log still in use!\n");
 968                 lc_destroy(n);
 969                 return -EBUSY;
 970         } else {
 971                 if (t)
 972                         lc_destroy(t);
 973         }
 974         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 975         return 0;
 976 }
 977
 978 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
 979 {
 980         struct request_queue * const q = mdev->rq_queue;
 981         int max_hw_sectors = max_bio_size >> 9;
 982         int max_segments = 0;
 983
 984         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 985                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 986
 987                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
 988                 rcu_read_lock();
 989                 max_segments = rcu_dereference(mdev->ldev->disk_conf)->max_bio_bvecs;
 990                 rcu_read_unlock();
 991                 put_ldev(mdev);
 992         }
 993
 994         blk_queue_logical_block_size(q, 512);
 995         blk_queue_max_hw_sectors(q, max_hw_sectors);
 996         /* This is the workaround for "bio would need to, but cannot, be split" */
 997         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
 998         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
 999
1000         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1001                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
1002
1003                 blk_queue_stack_limits(q, b);
1004
1005                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
1006                         dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1007                                  q->backing_dev_info.ra_pages,
1008                                  b->backing_dev_info.ra_pages);
1009                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
1010                 }
1011                 put_ldev(mdev);
1012         }
1013 }
1014
1015 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
1016 {
1017         int now, new, local, peer;
1018
1019         now = queue_max_hw_sectors(mdev->rq_queue) << 9;
1020         local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
1021         peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
1022
1023         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1024                 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
1025                 mdev->local_max_bio_size = local;
1026                 put_ldev(mdev);
1027         }
1028
1029         /* We may ignore peer limits if the peer is modern enough.
1030            Because new from 8.3.8 onwards the peer can use multiple
1031            BIOs for a single peer_request */
1032         if (mdev->state.conn >= C_CONNECTED) {
1033                 if (mdev->tconn->agreed_pro_version < 94)
1034                         peer = mdev->peer_max_bio_size;
1035                 else if (mdev->tconn->agreed_pro_version == 94)
1036                         peer = DRBD_MAX_SIZE_H80_PACKET;
1037                 else /* drbd 8.3.8 onwards */
1038                         peer = DRBD_MAX_BIO_SIZE;
1039         }
1040
1041         new = min_t(int, local, peer);
1042
1043         if (mdev->state.role == R_PRIMARY && new < now)
1044                 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
1045
1046         if (new != now)
1047                 dev_info(DEV, "max BIO size = %u\n", new);
1048
1049         drbd_setup_queue_param(mdev, new);
1050 }
1051
1052 /* Starts the worker thread */
1053 static void conn_reconfig_start(struct drbd_tconn *tconn)
1054 {
1055         drbd_thread_start(&tconn->worker);
1056         conn_flush_workqueue(tconn);
1057 }
1058
1059 /* if still unconfigured, stops worker again. */
1060 static void conn_reconfig_done(struct drbd_tconn *tconn)
1061 {
1062         bool stop_threads;
1063         spin_lock_irq(&tconn->req_lock);
1064         stop_threads = conn_all_vols_unconf(tconn);
1065         spin_unlock_irq(&tconn->req_lock);
1066         if (stop_threads) {
1067                 /* asender is implicitly stopped by receiver
1068                  * in conn_disconnect() */
1069                 drbd_thread_stop(&tconn->receiver);
1070                 drbd_thread_stop(&tconn->worker);
1071         }
1072 }
1073
1074 /* Make sure IO is suspended before calling this function(). */
1075 static void drbd_suspend_al(struct drbd_conf *mdev)
1076 {
1077         int s = 0;
1078
1079         if (!lc_try_lock(mdev->act_log)) {
1080                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
1081                 return;
1082         }
1083
1084         drbd_al_shrink(mdev);
1085         spin_lock_irq(&mdev->tconn->req_lock);
1086         if (mdev->state.conn < C_CONNECTED)
1087                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
1088         spin_unlock_irq(&mdev->tconn->req_lock);
1089         lc_unlock(mdev->act_log);
1090
1091         if (s)
1092                 dev_info(DEV, "Suspended AL updates\n");
1093 }
1094
1095
1096 static bool should_set_defaults(struct genl_info *info)
1097 {
1098         unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1099         return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1100 }
1101
1102 static void enforce_disk_conf_limits(struct disk_conf *dc)
1103 {
1104         if (dc->al_extents < DRBD_AL_EXTENTS_MIN)
1105                 dc->al_extents = DRBD_AL_EXTENTS_MIN;
1106         if (dc->al_extents > DRBD_AL_EXTENTS_MAX)
1107                 dc->al_extents = DRBD_AL_EXTENTS_MAX;
1108
1109         if (dc->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1110                 dc->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1111 }
1112
1113 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1114 {
1115         enum drbd_ret_code retcode;
1116         struct drbd_conf *mdev;
1117         struct disk_conf *new_disk_conf, *old_disk_conf;
1118         struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1119         int err, fifo_size;
1120
1121         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1122         if (!adm_ctx.reply_skb)
1123                 return retcode;
1124         if (retcode != NO_ERROR)
1125                 goto out;
1126
1127         mdev = adm_ctx.mdev;
1128
1129         /* we also need a disk
1130          * to change the options on */
1131         if (!get_ldev(mdev)) {
1132                 retcode = ERR_NO_DISK;
1133                 goto out;
1134         }
1135
1136         new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1137         if (!new_disk_conf) {
1138                 retcode = ERR_NOMEM;
1139                 goto fail;
1140         }
1141
1142         mutex_lock(&mdev->tconn->conf_update);
1143         old_disk_conf = mdev->ldev->disk_conf;
1144         *new_disk_conf = *old_disk_conf;
1145         if (should_set_defaults(info))
1146                 set_disk_conf_defaults(new_disk_conf);
1147
1148         err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1149         if (err) {
1150                 retcode = ERR_MANDATORY_TAG;
1151                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1152         }
1153
1154         if (!expect(new_disk_conf->resync_rate >= 1))
1155                 new_disk_conf->resync_rate = 1;
1156
1157         enforce_disk_conf_limits(new_disk_conf);
1158
1159         fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1160         if (fifo_size != mdev->rs_plan_s->size) {
1161                 new_plan = fifo_alloc(fifo_size);
1162                 if (!new_plan) {
1163                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1164                         retcode = ERR_NOMEM;
1165                         goto fail_unlock;
1166                 }
1167         }
1168
1169         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1170         drbd_al_shrink(mdev);
1171         err = drbd_check_al_size(mdev, new_disk_conf);
1172         lc_unlock(mdev->act_log);
1173         wake_up(&mdev->al_wait);
1174
1175         if (err) {
1176                 retcode = ERR_NOMEM;
1177                 goto fail_unlock;
1178         }
1179
1180         write_lock_irq(&global_state_lock);
1181         retcode = drbd_sync_after_valid(mdev, new_disk_conf->resync_after);
1182         if (retcode == NO_ERROR) {
1183                 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
1184                 drbd_sync_after_changed(mdev);
1185         }
1186         write_unlock_irq(&global_state_lock);
1187
1188         if (retcode != NO_ERROR)
1189                 goto fail_unlock;
1190
1191         if (new_plan) {
1192                 old_plan = mdev->rs_plan_s;
1193                 rcu_assign_pointer(mdev->rs_plan_s, new_plan);
1194         }
1195
1196         drbd_md_sync(mdev);
1197
1198         if (mdev->state.conn >= C_CONNECTED)
1199                 drbd_send_sync_param(mdev);
1200
1201         mutex_unlock(&mdev->tconn->conf_update);
1202         synchronize_rcu();
1203         kfree(old_disk_conf);
1204         kfree(old_plan);
1205         goto success;
1206
1207 fail_unlock:
1208         mutex_unlock(&mdev->tconn->conf_update);
1209  fail:
1210         kfree(new_disk_conf);
1211         kfree(new_plan);
1212 success:
1213         put_ldev(mdev);
1214  out:
1215         drbd_adm_finish(info, retcode);
1216         return 0;
1217 }
1218
1219 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1220 {
1221         struct drbd_conf *mdev;
1222         int err;
1223         enum drbd_ret_code retcode;
1224         enum determine_dev_size dd;
1225         sector_t max_possible_sectors;
1226         sector_t min_md_device_sectors;
1227         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1228         struct disk_conf *new_disk_conf = NULL;
1229         struct block_device *bdev;
1230         struct lru_cache *resync_lru = NULL;
1231         struct fifo_buffer *new_plan = NULL;
1232         union drbd_state ns, os;
1233         enum drbd_state_rv rv;
1234         struct net_conf *nc;
1235         int cp_discovered = 0;
1236
1237         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1238         if (!adm_ctx.reply_skb)
1239                 return retcode;
1240         if (retcode != NO_ERROR)
1241                 goto finish;
1242
1243         mdev = adm_ctx.mdev;
1244         conn_reconfig_start(mdev->tconn);
1245
1246         /* if you want to reconfigure, please tear down first */
1247         if (mdev->state.disk > D_DISKLESS) {
1248                 retcode = ERR_DISK_CONFIGURED;
1249                 goto fail;
1250         }
1251         /* It may just now have detached because of IO error.  Make sure
1252          * drbd_ldev_destroy is done already, we may end up here very fast,
1253          * e.g. if someone calls attach from the on-io-error handler,
1254          * to realize a "hot spare" feature (not that I'd recommend that) */
1255         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1256
1257         /* allocation not in the IO path, drbdsetup context */
1258         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1259         if (!nbc) {
1260                 retcode = ERR_NOMEM;
1261                 goto fail;
1262         }
1263         new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1264         if (!new_disk_conf) {
1265                 retcode = ERR_NOMEM;
1266                 goto fail;
1267         }
1268         nbc->disk_conf = new_disk_conf;
1269
1270         set_disk_conf_defaults(new_disk_conf);
1271         err = disk_conf_from_attrs(new_disk_conf, info);
1272         if (err) {
1273                 retcode = ERR_MANDATORY_TAG;
1274                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1275                 goto fail;
1276         }
1277
1278         enforce_disk_conf_limits(new_disk_conf);
1279
1280         new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1281         if (!new_plan) {
1282                 retcode = ERR_NOMEM;
1283                 goto fail;
1284         }
1285
1286         if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1287                 retcode = ERR_MD_IDX_INVALID;
1288                 goto fail;
1289         }
1290
1291         rcu_read_lock();
1292         nc = rcu_dereference(mdev->tconn->net_conf);
1293         if (nc) {
1294                 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1295                         rcu_read_unlock();
1296                         retcode = ERR_STONITH_AND_PROT_A;
1297                         goto fail;
1298                 }
1299         }
1300         rcu_read_unlock();
1301
1302         bdev = blkdev_get_by_path(new_disk_conf->backing_dev,
1303                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1304         if (IS_ERR(bdev)) {
1305                 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->backing_dev,
1306                         PTR_ERR(bdev));
1307                 retcode = ERR_OPEN_DISK;
1308                 goto fail;
1309         }
1310         nbc->backing_bdev = bdev;
1311
1312         /*
1313          * meta_dev_idx >= 0: external fixed size, possibly multiple
1314          * drbd sharing one meta device.  TODO in that case, paranoia
1315          * check that [md_bdev, meta_dev_idx] is not yet used by some
1316          * other drbd minor!  (if you use drbd.conf + drbdadm, that
1317          * should check it for you already; but if you don't, or
1318          * someone fooled it, we need to double check here)
1319          */
1320         bdev = blkdev_get_by_path(new_disk_conf->meta_dev,
1321                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1322                                   (new_disk_conf->meta_dev_idx < 0) ?
1323                                   (void *)mdev : (void *)drbd_m_holder);
1324         if (IS_ERR(bdev)) {
1325                 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->meta_dev,
1326                         PTR_ERR(bdev));
1327                 retcode = ERR_OPEN_MD_DISK;
1328                 goto fail;
1329         }
1330         nbc->md_bdev = bdev;
1331
1332         if ((nbc->backing_bdev == nbc->md_bdev) !=
1333             (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1334              new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1335                 retcode = ERR_MD_IDX_INVALID;
1336                 goto fail;
1337         }
1338
1339         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1340                         1, 61, sizeof(struct bm_extent),
1341                         offsetof(struct bm_extent, lce));
1342         if (!resync_lru) {
1343                 retcode = ERR_NOMEM;
1344                 goto fail;
1345         }
1346
1347         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1348         drbd_md_set_sector_offsets(mdev, nbc);
1349
1350         if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1351                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1352                         (unsigned long long) drbd_get_max_capacity(nbc),
1353                         (unsigned long long) new_disk_conf->disk_size);
1354                 retcode = ERR_DISK_TO_SMALL;
1355                 goto fail;
1356         }
1357
1358         if (new_disk_conf->meta_dev_idx < 0) {
1359                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1360                 /* at least one MB, otherwise it does not make sense */
1361                 min_md_device_sectors = (2<<10);
1362         } else {
1363                 max_possible_sectors = DRBD_MAX_SECTORS;
1364                 min_md_device_sectors = MD_RESERVED_SECT * (new_disk_conf->meta_dev_idx + 1);
1365         }
1366
1367         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1368                 retcode = ERR_MD_DISK_TO_SMALL;
1369                 dev_warn(DEV, "refusing attach: md-device too small, "
1370                      "at least %llu sectors needed for this meta-disk type\n",
1371                      (unsigned long long) min_md_device_sectors);
1372                 goto fail;
1373         }
1374
1375         /* Make sure the new disk is big enough
1376          * (we may currently be R_PRIMARY with no local disk...) */
1377         if (drbd_get_max_capacity(nbc) <
1378             drbd_get_capacity(mdev->this_bdev)) {
1379                 retcode = ERR_DISK_TO_SMALL;
1380                 goto fail;
1381         }
1382
1383         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1384
1385         if (nbc->known_size > max_possible_sectors) {
1386                 dev_warn(DEV, "==> truncating very big lower level device "
1387                         "to currently maximum possible %llu sectors <==\n",
1388                         (unsigned long long) max_possible_sectors);
1389                 if (new_disk_conf->meta_dev_idx >= 0)
1390                         dev_warn(DEV, "==>> using internal or flexible "
1391                                       "meta data may help <<==\n");
1392         }
1393
1394         drbd_suspend_io(mdev);
1395         /* also wait for the last barrier ack. */
1396         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || drbd_suspended(mdev));
1397         /* and for any other previously queued work */
1398         drbd_flush_workqueue(mdev);
1399
1400         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1401         retcode = rv;  /* FIXME: Type mismatch. */
1402         drbd_resume_io(mdev);
1403         if (rv < SS_SUCCESS)
1404                 goto fail;
1405
1406         if (!get_ldev_if_state(mdev, D_ATTACHING))
1407                 goto force_diskless;
1408
1409         drbd_md_set_sector_offsets(mdev, nbc);
1410
1411         if (!mdev->bitmap) {
1412                 if (drbd_bm_init(mdev)) {
1413                         retcode = ERR_NOMEM;
1414                         goto force_diskless_dec;
1415                 }
1416         }
1417
1418         retcode = drbd_md_read(mdev, nbc);
1419         if (retcode != NO_ERROR)
1420                 goto force_diskless_dec;
1421
1422         if (mdev->state.conn < C_CONNECTED &&
1423             mdev->state.role == R_PRIMARY &&
1424             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1425                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1426                     (unsigned long long)mdev->ed_uuid);
1427                 retcode = ERR_DATA_NOT_CURRENT;
1428                 goto force_diskless_dec;
1429         }
1430
1431         /* Since we are diskless, fix the activity log first... */
1432         if (drbd_check_al_size(mdev, new_disk_conf)) {
1433                 retcode = ERR_NOMEM;
1434                 goto force_diskless_dec;
1435         }
1436
1437         /* Prevent shrinking of consistent devices ! */
1438         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1439             drbd_new_dev_size(mdev, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1440                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1441                 retcode = ERR_DISK_TO_SMALL;
1442                 goto force_diskless_dec;
1443         }
1444
1445         if (!drbd_al_read_log(mdev, nbc)) {
1446                 retcode = ERR_IO_MD_DISK;
1447                 goto force_diskless_dec;
1448         }
1449
1450         /* Reset the "barriers don't work" bits here, then force meta data to
1451          * be written, to ensure we determine if barriers are supported. */
1452         if (new_disk_conf->md_flushes)
1453                 clear_bit(MD_NO_FUA, &mdev->flags);
1454         else
1455                 set_bit(MD_NO_FUA, &mdev->flags);
1456
1457         /* Point of no return reached.
1458          * Devices and memory are no longer released by error cleanup below.
1459          * now mdev takes over responsibility, and the state engine should
1460          * clean it up somewhere.  */
1461         D_ASSERT(mdev->ldev == NULL);
1462         mdev->ldev = nbc;
1463         mdev->resync = resync_lru;
1464         mdev->rs_plan_s = new_plan;
1465         nbc = NULL;
1466         resync_lru = NULL;
1467         new_disk_conf = NULL;
1468         new_plan = NULL;
1469
1470         mdev->write_ordering = WO_bdev_flush;
1471         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1472
1473         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1474                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1475         else
1476                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1477
1478         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1479             !(mdev->state.role == R_PRIMARY && mdev->tconn->susp_nod)) {
1480                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1481                 cp_discovered = 1;
1482         }
1483
1484         mdev->send_cnt = 0;
1485         mdev->recv_cnt = 0;
1486         mdev->read_cnt = 0;
1487         mdev->writ_cnt = 0;
1488
1489         drbd_reconsider_max_bio_size(mdev);
1490
1491         /* If I am currently not R_PRIMARY,
1492          * but meta data primary indicator is set,
1493          * I just now recover from a hard crash,
1494          * and have been R_PRIMARY before that crash.
1495          *
1496          * Now, if I had no connection before that crash
1497          * (have been degraded R_PRIMARY), chances are that
1498          * I won't find my peer now either.
1499          *
1500          * In that case, and _only_ in that case,
1501          * we use the degr-wfc-timeout instead of the default,
1502          * so we can automatically recover from a crash of a
1503          * degraded but active "cluster" after a certain timeout.
1504          */
1505         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1506         if (mdev->state.role != R_PRIMARY &&
1507              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1508             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1509                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1510
1511         dd = drbd_determine_dev_size(mdev, 0);
1512         if (dd == dev_size_error) {
1513                 retcode = ERR_NOMEM_BITMAP;
1514                 goto force_diskless_dec;
1515         } else if (dd == grew)
1516                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1517
1518         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1519                 dev_info(DEV, "Assuming that all blocks are out of sync "
1520                      "(aka FullSync)\n");
1521                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1522                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1523                         retcode = ERR_IO_MD_DISK;
1524                         goto force_diskless_dec;
1525                 }
1526         } else {
1527                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1528                         "read from attaching", BM_LOCKED_MASK)) {
1529                         retcode = ERR_IO_MD_DISK;
1530                         goto force_diskless_dec;
1531                 }
1532         }
1533
1534         if (cp_discovered) {
1535                 drbd_al_apply_to_bm(mdev);
1536                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1537                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1538                         retcode = ERR_IO_MD_DISK;
1539                         goto force_diskless_dec;
1540                 }
1541         }
1542
1543         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1544                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1545
1546         spin_lock_irq(&mdev->tconn->req_lock);
1547         os = drbd_read_state(mdev);
1548         ns = os;
1549         /* If MDF_CONSISTENT is not set go into inconsistent state,
1550            otherwise investigate MDF_WasUpToDate...
1551            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1552            otherwise into D_CONSISTENT state.
1553         */
1554         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1555                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1556                         ns.disk = D_CONSISTENT;
1557                 else
1558                         ns.disk = D_OUTDATED;
1559         } else {
1560                 ns.disk = D_INCONSISTENT;
1561         }
1562
1563         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1564                 ns.pdsk = D_OUTDATED;
1565
1566         rcu_read_lock();
1567         if (ns.disk == D_CONSISTENT &&
1568             (ns.pdsk == D_OUTDATED || rcu_dereference(mdev->ldev->disk_conf)->fencing == FP_DONT_CARE))
1569                 ns.disk = D_UP_TO_DATE;
1570         rcu_read_unlock();
1571
1572         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1573            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1574            this point, because drbd_request_state() modifies these
1575            flags. */
1576
1577         /* In case we are C_CONNECTED postpone any decision on the new disk
1578            state after the negotiation phase. */
1579         if (mdev->state.conn == C_CONNECTED) {
1580                 mdev->new_state_tmp.i = ns.i;
1581                 ns.i = os.i;
1582                 ns.disk = D_NEGOTIATING;
1583
1584                 /* We expect to receive up-to-date UUIDs soon.
1585                    To avoid a race in receive_state, free p_uuid while
1586                    holding req_lock. I.e. atomic with the state change */
1587                 kfree(mdev->p_uuid);
1588                 mdev->p_uuid = NULL;
1589         }
1590
1591         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1592         spin_unlock_irq(&mdev->tconn->req_lock);
1593
1594         if (rv < SS_SUCCESS)
1595                 goto force_diskless_dec;
1596
1597         if (mdev->state.role == R_PRIMARY)
1598                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1599         else
1600                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1601
1602         drbd_md_mark_dirty(mdev);
1603         drbd_md_sync(mdev);
1604
1605         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1606         put_ldev(mdev);
1607         conn_reconfig_done(mdev->tconn);
1608         drbd_adm_finish(info, retcode);
1609         return 0;
1610
1611  force_diskless_dec:
1612         put_ldev(mdev);
1613  force_diskless:
1614         drbd_force_state(mdev, NS(disk, D_FAILED));
1615         drbd_md_sync(mdev);
1616  fail:
1617         conn_reconfig_done(mdev->tconn);
1618         if (nbc) {
1619                 if (nbc->backing_bdev)
1620                         blkdev_put(nbc->backing_bdev,
1621                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1622                 if (nbc->md_bdev)
1623                         blkdev_put(nbc->md_bdev,
1624                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1625                 kfree(nbc);
1626         }
1627         kfree(new_disk_conf);
1628         lc_destroy(resync_lru);
1629         kfree(new_plan);
1630
1631  finish:
1632         drbd_adm_finish(info, retcode);
1633         return 0;
1634 }
1635
1636 static int adm_detach(struct drbd_conf *mdev)
1637 {
1638         enum drbd_state_rv retcode;
1639         int ret;
1640         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1641         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1642         /* D_FAILED will transition to DISKLESS. */
1643         ret = wait_event_interruptible(mdev->misc_wait,
1644                         mdev->state.disk != D_FAILED);
1645         drbd_resume_io(mdev);
1646         if ((int)retcode == (int)SS_IS_DISKLESS)
1647                 retcode = SS_NOTHING_TO_DO;
1648         if (ret)
1649                 retcode = ERR_INTR;
1650         return retcode;
1651 }
1652
1653 /* Detaching the disk is a process in multiple stages.  First we need to lock
1654  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1655  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1656  * internal references as well.
1657  * Only then we have finally detached. */
1658 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1659 {
1660         enum drbd_ret_code retcode;
1661
1662         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1663         if (!adm_ctx.reply_skb)
1664                 return retcode;
1665         if (retcode != NO_ERROR)
1666                 goto out;
1667
1668         retcode = adm_detach(adm_ctx.mdev);
1669 out:
1670         drbd_adm_finish(info, retcode);
1671         return 0;
1672 }
1673
1674 static bool conn_resync_running(struct drbd_tconn *tconn)
1675 {
1676         struct drbd_conf *mdev;
1677         bool rv = false;
1678         int vnr;
1679
1680         rcu_read_lock();
1681         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1682                 if (mdev->state.conn == C_SYNC_SOURCE ||
1683                     mdev->state.conn == C_SYNC_TARGET ||
1684                     mdev->state.conn == C_PAUSED_SYNC_S ||
1685                     mdev->state.conn == C_PAUSED_SYNC_T) {
1686                         rv = true;
1687                         break;
1688                 }
1689         }
1690         rcu_read_unlock();
1691
1692         return rv;
1693 }
1694
1695 static bool conn_ov_running(struct drbd_tconn *tconn)
1696 {
1697         struct drbd_conf *mdev;
1698         bool rv = false;
1699         int vnr;
1700
1701         rcu_read_lock();
1702         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1703                 if (mdev->state.conn == C_VERIFY_S ||
1704                     mdev->state.conn == C_VERIFY_T) {
1705                         rv = true;
1706                         break;
1707                 }
1708         }
1709         rcu_read_unlock();
1710
1711         return rv;
1712 }
1713
1714 static enum drbd_ret_code
1715 _check_net_options(struct drbd_tconn *tconn, struct net_conf *old_conf, struct net_conf *new_conf)
1716 {
1717         struct drbd_conf *mdev;
1718         int i;
1719
1720         if (old_conf && tconn->agreed_pro_version < 100 &&
1721             tconn->cstate == C_WF_REPORT_PARAMS &&
1722             new_conf->wire_protocol != old_conf->wire_protocol)
1723                 return ERR_NEED_APV_100;
1724
1725         if (new_conf->two_primaries &&
1726             (new_conf->wire_protocol != DRBD_PROT_C))
1727                 return ERR_NOT_PROTO_C;
1728
1729         idr_for_each_entry(&tconn->volumes, mdev, i) {
1730                 if (get_ldev(mdev)) {
1731                         enum drbd_fencing_p fp = rcu_dereference(mdev->ldev->disk_conf)->fencing;
1732                         put_ldev(mdev);
1733                         if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
1734                                 return ERR_STONITH_AND_PROT_A;
1735                 }
1736                 if (mdev->state.role == R_PRIMARY && new_conf->want_lose)
1737                         return ERR_DISCARD;
1738         }
1739
1740         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A)
1741                 return ERR_CONG_NOT_PROTO_A;
1742
1743         return NO_ERROR;
1744 }
1745
1746 static enum drbd_ret_code
1747 check_net_options(struct drbd_tconn *tconn, struct net_conf *new_conf)
1748 {
1749         static enum drbd_ret_code rv;
1750         struct drbd_conf *mdev;
1751         int i;
1752
1753         rcu_read_lock();
1754         rv = _check_net_options(tconn, rcu_dereference(tconn->net_conf), new_conf);
1755         rcu_read_unlock();
1756
1757         /* tconn->volumes protected by genl_lock() here */
1758         idr_for_each_entry(&tconn->volumes, mdev, i) {
1759                 if (!mdev->bitmap) {
1760                         if(drbd_bm_init(mdev))
1761                                 return ERR_NOMEM;
1762                 }
1763         }
1764
1765         return rv;
1766 }
1767
1768 struct crypto {
1769         struct crypto_hash *verify_tfm;
1770         struct crypto_hash *csums_tfm;
1771         struct crypto_hash *cram_hmac_tfm;
1772         struct crypto_hash *integrity_tfm;
1773         void *int_dig_in;
1774         void *int_dig_vv;
1775 };
1776
1777 static int
1778 alloc_hash(struct crypto_hash **tfm, char *tfm_name, int err_alg)
1779 {
1780         if (!tfm_name[0])
1781                 return NO_ERROR;
1782
1783         *tfm = crypto_alloc_hash(tfm_name, 0, CRYPTO_ALG_ASYNC);
1784         if (IS_ERR(*tfm)) {
1785                 *tfm = NULL;
1786                 return err_alg;
1787         }
1788
1789         return NO_ERROR;
1790 }
1791
1792 static enum drbd_ret_code
1793 alloc_crypto(struct crypto *crypto, struct net_conf *new_conf)
1794 {
1795         char hmac_name[CRYPTO_MAX_ALG_NAME];
1796         enum drbd_ret_code rv;
1797         int hash_size;
1798
1799         rv = alloc_hash(&crypto->csums_tfm, new_conf->csums_alg,
1800                        ERR_CSUMS_ALG);
1801         if (rv != NO_ERROR)
1802                 return rv;
1803         rv = alloc_hash(&crypto->verify_tfm, new_conf->verify_alg,
1804                        ERR_VERIFY_ALG);
1805         if (rv != NO_ERROR)
1806                 return rv;
1807         rv = alloc_hash(&crypto->integrity_tfm, new_conf->integrity_alg,
1808                        ERR_INTEGRITY_ALG);
1809         if (rv != NO_ERROR)
1810                 return rv;
1811         if (new_conf->cram_hmac_alg[0] != 0) {
1812                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1813                          new_conf->cram_hmac_alg);
1814
1815                 rv = alloc_hash(&crypto->cram_hmac_tfm, hmac_name,
1816                                ERR_AUTH_ALG);
1817         }
1818         if (crypto->integrity_tfm) {
1819                 hash_size = crypto_hash_digestsize(crypto->integrity_tfm);
1820                 crypto->int_dig_in = kmalloc(hash_size, GFP_KERNEL);
1821                 if (!crypto->int_dig_in)
1822                         return ERR_NOMEM;
1823                 crypto->int_dig_vv = kmalloc(hash_size, GFP_KERNEL);
1824                 if (!crypto->int_dig_vv)
1825                         return ERR_NOMEM;
1826         }
1827
1828         return rv;
1829 }
1830
1831 static void free_crypto(struct crypto *crypto)
1832 {
1833         kfree(crypto->int_dig_in);
1834         kfree(crypto->int_dig_vv);
1835         crypto_free_hash(crypto->cram_hmac_tfm);
1836         crypto_free_hash(crypto->integrity_tfm);
1837         crypto_free_hash(crypto->csums_tfm);
1838         crypto_free_hash(crypto->verify_tfm);
1839 }
1840
1841 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
1842 {
1843         enum drbd_ret_code retcode;
1844         struct drbd_tconn *tconn;
1845         struct net_conf *old_conf, *new_conf = NULL;
1846         int err;
1847         int ovr; /* online verify running */
1848         int rsr; /* re-sync running */
1849         struct crypto crypto = { };
1850         bool change_integrity_alg;
1851
1852         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1853         if (!adm_ctx.reply_skb)
1854                 return retcode;
1855         if (retcode != NO_ERROR)
1856                 goto out;
1857
1858         tconn = adm_ctx.tconn;
1859
1860         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1861         if (!new_conf) {
1862                 retcode = ERR_NOMEM;
1863                 goto out;
1864         }
1865
1866         conn_reconfig_start(tconn);
1867
1868         mutex_lock(&tconn->data.mutex);
1869         mutex_lock(&tconn->conf_update);
1870         old_conf = tconn->net_conf;
1871
1872         if (!old_conf) {
1873                 drbd_msg_put_info("net conf missing, try connect");
1874                 retcode = ERR_INVALID_REQUEST;
1875                 goto fail;
1876         }
1877
1878         *new_conf = *old_conf;
1879         if (should_set_defaults(info))
1880                 set_net_conf_defaults(new_conf);
1881
1882         err = net_conf_from_attrs_for_change(new_conf, info);
1883         if (err) {
1884                 retcode = ERR_MANDATORY_TAG;
1885                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1886                 goto fail;
1887         }
1888
1889         retcode = check_net_options(tconn, new_conf);
1890         if (retcode != NO_ERROR)
1891                 goto fail;
1892
1893         /* re-sync running */
1894         rsr = conn_resync_running(tconn);
1895         if (rsr && strcmp(new_conf->csums_alg, old_conf->csums_alg)) {
1896                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1897                 goto fail;
1898         }
1899
1900         /* online verify running */
1901         ovr = conn_ov_running(tconn);
1902         if (ovr && strcmp(new_conf->verify_alg, old_conf->verify_alg)) {
1903                 retcode = ERR_VERIFY_RUNNING;
1904                 goto fail;
1905         }
1906
1907         change_integrity_alg = strcmp(old_conf->integrity_alg,
1908                                       new_conf->integrity_alg);
1909
1910         retcode = alloc_crypto(&crypto, new_conf);
1911         if (retcode != NO_ERROR)
1912                 goto fail;
1913
1914         rcu_assign_pointer(tconn->net_conf, new_conf);
1915
1916         if (!rsr) {
1917                 crypto_free_hash(tconn->csums_tfm);
1918                 tconn->csums_tfm = crypto.csums_tfm;
1919                 crypto.csums_tfm = NULL;
1920         }
1921         if (!ovr) {
1922                 crypto_free_hash(tconn->verify_tfm);
1923                 tconn->verify_tfm = crypto.verify_tfm;
1924                 crypto.verify_tfm = NULL;
1925         }
1926
1927         kfree(tconn->int_dig_in);
1928         tconn->int_dig_in = crypto.int_dig_in;
1929         kfree(tconn->int_dig_vv);
1930         tconn->int_dig_vv = crypto.int_dig_vv;
1931         crypto_free_hash(tconn->integrity_tfm);
1932         tconn->integrity_tfm = crypto.integrity_tfm;
1933         if (change_integrity_alg) {
1934                 /* Do this without trying to take tconn->data.mutex again.  */
1935                 if (__drbd_send_protocol(tconn))
1936                         goto fail;
1937         }
1938
1939         /* FIXME Changing cram_hmac while the connection is established is useless */
1940         crypto_free_hash(tconn->cram_hmac_tfm);
1941         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
1942
1943         mutex_unlock(&tconn->conf_update);
1944         mutex_unlock(&tconn->data.mutex);
1945         synchronize_rcu();
1946         kfree(old_conf);
1947
1948         if (tconn->cstate >= C_WF_REPORT_PARAMS)
1949                 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn)));
1950
1951         goto done;
1952
1953  fail:
1954         mutex_unlock(&tconn->conf_update);
1955         mutex_unlock(&tconn->data.mutex);
1956         free_crypto(&crypto);
1957         kfree(new_conf);
1958  done:
1959         conn_reconfig_done(tconn);
1960  out:
1961         drbd_adm_finish(info, retcode);
1962         return 0;
1963 }
1964
1965 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
1966 {
1967         struct drbd_conf *mdev;
1968         struct net_conf *old_conf, *new_conf = NULL;
1969         struct crypto crypto = { };
1970         struct drbd_tconn *oconn;
1971         struct drbd_tconn *tconn;
1972         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1973         enum drbd_ret_code retcode;
1974         int i;
1975         int err;
1976
1977         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1978         if (!adm_ctx.reply_skb)
1979                 return retcode;
1980         if (retcode != NO_ERROR)
1981                 goto out;
1982
1983         tconn = adm_ctx.tconn;
1984         conn_reconfig_start(tconn);
1985
1986         if (tconn->cstate > C_STANDALONE) {
1987                 retcode = ERR_NET_CONFIGURED;
1988                 goto fail;
1989         }
1990
1991         /* allocation not in the IO path, cqueue thread context */
1992         new_conf = kzalloc(sizeof(*new_conf), GFP_KERNEL);
1993         if (!new_conf) {
1994                 retcode = ERR_NOMEM;
1995                 goto fail;
1996         }
1997
1998         set_net_conf_defaults(new_conf);
1999
2000         err = net_conf_from_attrs(new_conf, info);
2001         if (err) {
2002                 retcode = ERR_MANDATORY_TAG;
2003                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2004                 goto fail;
2005         }
2006
2007         retcode = check_net_options(tconn, new_conf);
2008         if (retcode != NO_ERROR)
2009                 goto fail;
2010
2011         retcode = NO_ERROR;
2012
2013         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
2014         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
2015
2016         /* No need to take drbd_cfg_rwsem here.  All reconfiguration is
2017          * strictly serialized on genl_lock(). We are protected against
2018          * concurrent reconfiguration/addition/deletion */
2019         list_for_each_entry(oconn, &drbd_tconns, all_tconn) {
2020                 struct net_conf *nc;
2021                 if (oconn == tconn)
2022                         continue;
2023
2024                 rcu_read_lock();
2025                 nc = rcu_dereference(oconn->net_conf);
2026                 if (nc) {
2027                         taken_addr = (struct sockaddr *)&nc->my_addr;
2028                         if (new_conf->my_addr_len == nc->my_addr_len &&
2029                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
2030                                 retcode = ERR_LOCAL_ADDR;
2031
2032                         taken_addr = (struct sockaddr *)&nc->peer_addr;
2033                         if (new_conf->peer_addr_len == nc->peer_addr_len &&
2034                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
2035                                 retcode = ERR_PEER_ADDR;
2036                 }
2037                 rcu_read_unlock();
2038                 if (retcode != NO_ERROR)
2039                         goto fail;
2040         }
2041
2042         retcode = alloc_crypto(&crypto, new_conf);
2043         if (retcode != NO_ERROR)
2044                 goto fail;
2045
2046         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2047
2048         conn_flush_workqueue(tconn);
2049
2050         mutex_lock(&tconn->conf_update);
2051         old_conf = tconn->net_conf;
2052         if (old_conf) {
2053                 retcode = ERR_NET_CONFIGURED;
2054                 mutex_unlock(&tconn->conf_update);
2055                 goto fail;
2056         }
2057         rcu_assign_pointer(tconn->net_conf, new_conf);
2058
2059         conn_free_crypto(tconn);
2060         tconn->int_dig_in = crypto.int_dig_in;
2061         tconn->int_dig_vv = crypto.int_dig_vv;
2062         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
2063         tconn->integrity_tfm = crypto.integrity_tfm;
2064         tconn->csums_tfm = crypto.csums_tfm;
2065         tconn->verify_tfm = crypto.verify_tfm;
2066
2067         mutex_unlock(&tconn->conf_update);
2068
2069         rcu_read_lock();
2070         idr_for_each_entry(&tconn->volumes, mdev, i) {
2071                 mdev->send_cnt = 0;
2072                 mdev->recv_cnt = 0;
2073         }
2074         rcu_read_unlock();
2075
2076         retcode = conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2077
2078         conn_reconfig_done(tconn);
2079         drbd_adm_finish(info, retcode);
2080         return 0;
2081
2082 fail:
2083         free_crypto(&crypto);
2084         kfree(new_conf);
2085
2086         conn_reconfig_done(tconn);
2087 out:
2088         drbd_adm_finish(info, retcode);
2089         return 0;
2090 }
2091
2092 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
2093 {
2094         enum drbd_state_rv rv;
2095
2096         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2097                         force ? CS_HARD : 0);
2098
2099         switch (rv) {
2100         case SS_NOTHING_TO_DO:
2101                 break;
2102         case SS_ALREADY_STANDALONE:
2103                 return SS_SUCCESS;
2104         case SS_PRIMARY_NOP:
2105                 /* Our state checking code wants to see the peer outdated. */
2106                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2107                                                 pdsk, D_OUTDATED), CS_VERBOSE);
2108                 break;
2109         case SS_CW_FAILED_BY_PEER:
2110                 /* The peer probably wants to see us outdated. */
2111                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2112                                                         disk, D_OUTDATED), 0);
2113                 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2114                         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2115                                         CS_HARD);
2116                 }
2117                 break;
2118         default:;
2119                 /* no special handling necessary */
2120         }
2121
2122         if (rv >= SS_SUCCESS) {
2123                 enum drbd_state_rv rv2;
2124                 /* No one else can reconfigure the network while I am here.
2125                  * The state handling only uses drbd_thread_stop_nowait(),
2126                  * we want to really wait here until the receiver is no more.
2127                  */
2128                 drbd_thread_stop(&adm_ctx.tconn->receiver);
2129
2130                 /* Race breaker.  This additional state change request may be
2131                  * necessary, if this was a forced disconnect during a receiver
2132                  * restart.  We may have "killed" the receiver thread just
2133                  * after drbdd_init() returned.  Typically, we should be
2134                  * C_STANDALONE already, now, and this becomes a no-op.
2135                  */
2136                 rv2 = conn_request_state(tconn, NS(conn, C_STANDALONE),
2137                                 CS_VERBOSE | CS_HARD);
2138                 if (rv2 < SS_SUCCESS)
2139                         conn_err(tconn,
2140                                 "unexpected rv2=%d in conn_try_disconnect()\n",
2141                                 rv2);
2142         }
2143         return rv;
2144 }
2145
2146 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2147 {
2148         struct disconnect_parms parms;
2149         struct drbd_tconn *tconn;
2150         enum drbd_state_rv rv;
2151         enum drbd_ret_code retcode;
2152         int err;
2153
2154         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2155         if (!adm_ctx.reply_skb)
2156                 return retcode;
2157         if (retcode != NO_ERROR)
2158                 goto fail;
2159
2160         tconn = adm_ctx.tconn;
2161         memset(&parms, 0, sizeof(parms));
2162         if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2163                 err = disconnect_parms_from_attrs(&parms, info);
2164                 if (err) {
2165                         retcode = ERR_MANDATORY_TAG;
2166                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2167                         goto fail;
2168                 }
2169         }
2170
2171         rv = conn_try_disconnect(tconn, parms.force_disconnect);
2172         if (rv < SS_SUCCESS)
2173                 retcode = rv;  /* FIXME: Type mismatch. */
2174         else
2175                 retcode = NO_ERROR;
2176  fail:
2177         drbd_adm_finish(info, retcode);
2178         return 0;
2179 }
2180
2181 void resync_after_online_grow(struct drbd_conf *mdev)
2182 {
2183         int iass; /* I am sync source */
2184
2185         dev_info(DEV, "Resync of new storage after online grow\n");
2186         if (mdev->state.role != mdev->state.peer)
2187                 iass = (mdev->state.role == R_PRIMARY);
2188         else
2189                 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
2190
2191         if (iass)
2192                 drbd_start_resync(mdev, C_SYNC_SOURCE);
2193         else
2194                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2195 }
2196
2197 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2198 {
2199         struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2200         struct resize_parms rs;
2201         struct drbd_conf *mdev;
2202         enum drbd_ret_code retcode;
2203         enum determine_dev_size dd;
2204         enum dds_flags ddsf;
2205         sector_t u_size;
2206         int err;
2207
2208         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2209         if (!adm_ctx.reply_skb)
2210                 return retcode;
2211         if (retcode != NO_ERROR)
2212                 goto fail;
2213
2214         memset(&rs, 0, sizeof(struct resize_parms));
2215         if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2216                 err = resize_parms_from_attrs(&rs, info);
2217                 if (err) {
2218                         retcode = ERR_MANDATORY_TAG;
2219                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2220                         goto fail;
2221                 }
2222         }
2223
2224         mdev = adm_ctx.mdev;
2225         if (mdev->state.conn > C_CONNECTED) {
2226                 retcode = ERR_RESIZE_RESYNC;
2227                 goto fail;
2228         }
2229
2230         if (mdev->state.role == R_SECONDARY &&
2231             mdev->state.peer == R_SECONDARY) {
2232                 retcode = ERR_NO_PRIMARY;
2233                 goto fail;
2234         }
2235
2236         if (!get_ldev(mdev)) {
2237                 retcode = ERR_NO_DISK;
2238                 goto fail;
2239         }
2240
2241         if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
2242                 retcode = ERR_NEED_APV_93;
2243                 goto fail;
2244         }
2245
2246         rcu_read_lock();
2247         u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
2248         rcu_read_unlock();
2249         if (u_size != (sector_t)rs.resize_size) {
2250                 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2251                 if (!new_disk_conf) {
2252                         retcode = ERR_NOMEM;
2253                         goto fail;
2254                 }
2255         }
2256
2257         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
2258                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2259
2260         if (new_disk_conf) {
2261                 mutex_lock(&mdev->tconn->conf_update);
2262                 old_disk_conf = mdev->ldev->disk_conf;
2263                 *new_disk_conf = *old_disk_conf;
2264                 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2265                 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
2266                 mutex_unlock(&mdev->tconn->conf_update);
2267                 synchronize_rcu();
2268                 kfree(old_disk_conf);
2269         }
2270
2271         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2272         dd = drbd_determine_dev_size(mdev, ddsf);
2273         drbd_md_sync(mdev);
2274         put_ldev(mdev);
2275         if (dd == dev_size_error) {
2276                 retcode = ERR_NOMEM_BITMAP;
2277                 goto fail;
2278         }
2279
2280         if (mdev->state.conn == C_CONNECTED) {
2281                 if (dd == grew)
2282                         set_bit(RESIZE_PENDING, &mdev->flags);
2283
2284                 drbd_send_uuids(mdev);
2285                 drbd_send_sizes(mdev, 1, ddsf);
2286         }
2287
2288  fail:
2289         drbd_adm_finish(info, retcode);
2290         return 0;
2291 }
2292
2293 void drbd_set_res_opts_defaults(struct res_opts *r)
2294 {
2295         return set_res_opts_defaults(r);
2296 }
2297
2298 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2299 {
2300         enum drbd_ret_code retcode;
2301         cpumask_var_t new_cpu_mask;
2302         struct drbd_tconn *tconn;
2303         struct res_opts res_opts;
2304         int err;
2305
2306         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2307         if (!adm_ctx.reply_skb)
2308                 return retcode;
2309         if (retcode != NO_ERROR)
2310                 goto fail;
2311         tconn = adm_ctx.tconn;
2312
2313         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
2314                 retcode = ERR_NOMEM;
2315                 drbd_msg_put_info("unable to allocate cpumask");
2316                 goto fail;
2317         }
2318
2319         res_opts = tconn->res_opts;
2320         if (should_set_defaults(info))
2321                 set_res_opts_defaults(&res_opts);
2322
2323         err = res_opts_from_attrs(&res_opts, info);
2324         if (err) {
2325                 retcode = ERR_MANDATORY_TAG;
2326                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2327                 goto fail;
2328         }
2329
2330         /* silently ignore cpu mask on UP kernel */
2331         if (nr_cpu_ids > 1 && res_opts.cpu_mask[0] != 0) {
2332                 err = __bitmap_parse(res_opts.cpu_mask, 32, 0,
2333                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2334                 if (err) {
2335                         conn_warn(tconn, "__bitmap_parse() failed with %d\n", err);
2336                         retcode = ERR_CPU_MASK_PARSE;
2337                         goto fail;
2338                 }
2339         }
2340
2341
2342         tconn->res_opts = res_opts;
2343
2344         if (!cpumask_equal(tconn->cpu_mask, new_cpu_mask)) {
2345                 cpumask_copy(tconn->cpu_mask, new_cpu_mask);
2346                 drbd_calc_cpu_mask(tconn);
2347                 tconn->receiver.reset_cpu_mask = 1;
2348                 tconn->asender.reset_cpu_mask = 1;
2349                 tconn->worker.reset_cpu_mask = 1;
2350         }
2351
2352 fail:
2353         free_cpumask_var(new_cpu_mask);
2354
2355         drbd_adm_finish(info, retcode);
2356         return 0;
2357 }
2358
2359 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2360 {
2361         struct drbd_conf *mdev;
2362         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2363
2364         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2365         if (!adm_ctx.reply_skb)
2366                 return retcode;
2367         if (retcode != NO_ERROR)
2368                 goto out;
2369
2370         mdev = adm_ctx.mdev;
2371
2372         /* If there is still bitmap IO pending, probably because of a previous
2373          * resync just being finished, wait for it before requesting a new resync. */
2374         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2375
2376         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2377
2378         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2379                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2380
2381         while (retcode == SS_NEED_CONNECTION) {
2382                 spin_lock_irq(&mdev->tconn->req_lock);
2383                 if (mdev->state.conn < C_CONNECTED)
2384                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2385                 spin_unlock_irq(&mdev->tconn->req_lock);
2386
2387                 if (retcode != SS_NEED_CONNECTION)
2388                         break;
2389
2390                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2391         }
2392
2393 out:
2394         drbd_adm_finish(info, retcode);
2395         return 0;
2396 }
2397
2398 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2399 {
2400         int rv;
2401
2402         rv = drbd_bmio_set_n_write(mdev);
2403         drbd_suspend_al(mdev);
2404         return rv;
2405 }
2406
2407 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2408                 union drbd_state mask, union drbd_state val)
2409 {
2410         enum drbd_ret_code retcode;
2411
2412         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2413         if (!adm_ctx.reply_skb)
2414                 return retcode;
2415         if (retcode != NO_ERROR)
2416                 goto out;
2417
2418         retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2419 out:
2420         drbd_adm_finish(info, retcode);
2421         return 0;
2422 }
2423
2424 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2425 {
2426         return drbd_adm_simple_request_state(skb, info, NS(conn, C_STARTING_SYNC_S));
2427 }
2428
2429 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2430 {
2431         enum drbd_ret_code retcode;
2432
2433         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2434         if (!adm_ctx.reply_skb)
2435                 return retcode;
2436         if (retcode != NO_ERROR)
2437                 goto out;
2438
2439         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2440                 retcode = ERR_PAUSE_IS_SET;
2441 out:
2442         drbd_adm_finish(info, retcode);
2443         return 0;
2444 }
2445
2446 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2447 {
2448         union drbd_dev_state s;
2449         enum drbd_ret_code retcode;
2450
2451         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2452         if (!adm_ctx.reply_skb)
2453                 return retcode;
2454         if (retcode != NO_ERROR)
2455                 goto out;
2456
2457         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2458                 s = adm_ctx.mdev->state;
2459                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2460                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2461                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2462                 } else {
2463                         retcode = ERR_PAUSE_IS_CLEAR;
2464                 }
2465         }
2466
2467 out:
2468         drbd_adm_finish(info, retcode);
2469         return 0;
2470 }
2471
2472 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2473 {
2474         return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2475 }
2476
2477 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2478 {
2479         struct drbd_conf *mdev;
2480         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2481
2482         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2483         if (!adm_ctx.reply_skb)
2484                 return retcode;
2485         if (retcode != NO_ERROR)
2486                 goto out;
2487
2488         mdev = adm_ctx.mdev;
2489         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2490                 drbd_uuid_new_current(mdev);
2491                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2492         }
2493         drbd_suspend_io(mdev);
2494         retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2495         if (retcode == SS_SUCCESS) {
2496                 if (mdev->state.conn < C_CONNECTED)
2497                         tl_clear(mdev->tconn);
2498                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2499                         tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2500         }
2501         drbd_resume_io(mdev);
2502
2503 out:
2504         drbd_adm_finish(info, retcode);
2505         return 0;
2506 }
2507
2508 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2509 {
2510         return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2511 }
2512
2513 int nla_put_drbd_cfg_context(struct sk_buff *skb, const char *conn_name, unsigned vnr)
2514 {
2515         struct nlattr *nla;
2516         nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2517         if (!nla)
2518                 goto nla_put_failure;
2519         if (vnr != VOLUME_UNSPECIFIED)
2520                 NLA_PUT_U32(skb, T_ctx_volume, vnr);
2521         NLA_PUT_STRING(skb, T_ctx_conn_name, conn_name);
2522         nla_nest_end(skb, nla);
2523         return 0;
2524
2525 nla_put_failure:
2526         if (nla)
2527                 nla_nest_cancel(skb, nla);
2528         return -EMSGSIZE;
2529 }
2530
2531 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2532                 const struct sib_info *sib)
2533 {
2534         struct state_info *si = NULL; /* for sizeof(si->member); */
2535         struct net_conf *nc;
2536         struct nlattr *nla;
2537         int got_ldev;
2538         int err = 0;
2539         int exclude_sensitive;
2540
2541         /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2542          * to.  So we better exclude_sensitive information.
2543          *
2544          * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2545          * in the context of the requesting user process. Exclude sensitive
2546          * information, unless current has superuser.
2547          *
2548          * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2549          * relies on the current implementation of netlink_dump(), which
2550          * executes the dump callback successively from netlink_recvmsg(),
2551          * always in the context of the receiving process */
2552         exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2553
2554         got_ldev = get_ldev(mdev);
2555
2556         /* We need to add connection name and volume number information still.
2557          * Minor number is in drbd_genlmsghdr. */
2558         if (nla_put_drbd_cfg_context(skb, mdev->tconn->name, mdev->vnr))
2559                 goto nla_put_failure;
2560
2561         if (res_opts_to_skb(skb, &mdev->tconn->res_opts, exclude_sensitive))
2562                 goto nla_put_failure;
2563
2564         rcu_read_lock();
2565         if (got_ldev)
2566                 if (disk_conf_to_skb(skb, rcu_dereference(mdev->ldev->disk_conf), exclude_sensitive))
2567                         goto nla_put_failure;
2568
2569         nc = rcu_dereference(mdev->tconn->net_conf);
2570         if (nc)
2571                 err = net_conf_to_skb(skb, nc, exclude_sensitive);
2572         rcu_read_unlock();
2573         if (err)
2574                 goto nla_put_failure;
2575
2576         nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2577         if (!nla)
2578                 goto nla_put_failure;
2579         NLA_PUT_U32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY);
2580         NLA_PUT_U32(skb, T_current_state, mdev->state.i);
2581         NLA_PUT_U64(skb, T_ed_uuid, mdev->ed_uuid);
2582         NLA_PUT_U64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev));
2583
2584         if (got_ldev) {
2585                 NLA_PUT_U32(skb, T_disk_flags, mdev->ldev->md.flags);
2586                 NLA_PUT(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2587                 NLA_PUT_U64(skb, T_bits_total, drbd_bm_bits(mdev));
2588                 NLA_PUT_U64(skb, T_bits_oos, drbd_bm_total_weight(mdev));
2589                 if (C_SYNC_SOURCE <= mdev->state.conn &&
2590                     C_PAUSED_SYNC_T >= mdev->state.conn) {
2591                         NLA_PUT_U64(skb, T_bits_rs_total, mdev->rs_total);
2592                         NLA_PUT_U64(skb, T_bits_rs_failed, mdev->rs_failed);
2593                 }
2594         }
2595
2596         if (sib) {
2597                 switch(sib->sib_reason) {
2598                 case SIB_SYNC_PROGRESS:
2599                 case SIB_GET_STATUS_REPLY:
2600                         break;
2601                 case SIB_STATE_CHANGE:
2602                         NLA_PUT_U32(skb, T_prev_state, sib->os.i);
2603                         NLA_PUT_U32(skb, T_new_state, sib->ns.i);
2604                         break;
2605                 case SIB_HELPER_POST:
2606                         NLA_PUT_U32(skb,
2607                                 T_helper_exit_code, sib->helper_exit_code);
2608                         /* fall through */
2609                 case SIB_HELPER_PRE:
2610                         NLA_PUT_STRING(skb, T_helper, sib->helper_name);
2611                         break;
2612                 }
2613         }
2614         nla_nest_end(skb, nla);
2615
2616         if (0)
2617 nla_put_failure:
2618                 err = -EMSGSIZE;
2619         if (got_ldev)
2620                 put_ldev(mdev);
2621         return err;
2622 }
2623
2624 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2625 {
2626         enum drbd_ret_code retcode;
2627         int err;
2628
2629         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2630         if (!adm_ctx.reply_skb)
2631                 return retcode;
2632         if (retcode != NO_ERROR)
2633                 goto out;
2634
2635         err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2636         if (err) {
2637                 nlmsg_free(adm_ctx.reply_skb);
2638                 return err;
2639         }
2640 out:
2641         drbd_adm_finish(info, retcode);
2642         return 0;
2643 }
2644
2645 int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
2646 {
2647         struct drbd_conf *mdev;
2648         struct drbd_genlmsghdr *dh;
2649         struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2650         struct drbd_tconn *tconn = NULL;
2651         struct drbd_tconn *tmp;
2652         unsigned volume = cb->args[1];
2653
2654         /* Open coded, deferred, iteration:
2655          * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2656          *      idr_for_each_entry(&tconn->volumes, mdev, i) {
2657          *        ...
2658          *      }
2659          * }
2660          * where tconn is cb->args[0];
2661          * and i is cb->args[1];
2662          *
2663          * cb->args[2] indicates if we shall loop over all resources,
2664          * or just dump all volumes of a single resource.
2665          *
2666          * This may miss entries inserted after this dump started,
2667          * or entries deleted before they are reached.
2668          *
2669          * We need to make sure the mdev won't disappear while
2670          * we are looking at it, and revalidate our iterators
2671          * on each iteration.
2672          */
2673
2674         /* synchronize with conn_create()/conn_destroy() */
2675         down_read(&drbd_cfg_rwsem);
2676         /* revalidate iterator position */
2677         list_for_each_entry_rcu(tmp, &drbd_tconns, all_tconn) {
2678                 if (pos == NULL) {
2679                         /* first iteration */
2680                         pos = tmp;
2681                         tconn = pos;
2682                         break;
2683                 }
2684                 if (tmp == pos) {
2685                         tconn = pos;
2686                         break;
2687                 }
2688         }
2689         if (tconn) {
2690 next_tconn:
2691                 mdev = idr_get_next(&tconn->volumes, &volume);
2692                 if (!mdev) {
2693                         /* No more volumes to dump on this tconn.
2694                          * Advance tconn iterator. */
2695                         pos = list_entry_rcu(tconn->all_tconn.next,
2696                                              struct drbd_tconn, all_tconn);
2697                         /* Did we dump any volume on this tconn yet? */
2698                         if (volume != 0) {
2699                                 /* If we reached the end of the list,
2700                                  * or only a single resource dump was requested,
2701                                  * we are done. */
2702                                 if (&pos->all_tconn == &drbd_tconns || cb->args[2])
2703                                         goto out;
2704                                 volume = 0;
2705                                 tconn = pos;
2706                                 goto next_tconn;
2707                         }
2708                 }
2709
2710                 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).pid,
2711                                 cb->nlh->nlmsg_seq, &drbd_genl_family,
2712                                 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2713                 if (!dh)
2714                         goto out;
2715
2716                 if (!mdev) {
2717                         /* this is a tconn without a single volume */
2718                         dh->minor = -1U;
2719                         dh->ret_code = NO_ERROR;
2720                         if (nla_put_drbd_cfg_context(skb, tconn->name, VOLUME_UNSPECIFIED))
2721                                 genlmsg_cancel(skb, dh);
2722                         else
2723                                 genlmsg_end(skb, dh);
2724                         goto out;
2725                 }
2726
2727                 D_ASSERT(mdev->vnr == volume);
2728                 D_ASSERT(mdev->tconn == tconn);
2729
2730                 dh->minor = mdev_to_minor(mdev);
2731                 dh->ret_code = NO_ERROR;
2732
2733                 if (nla_put_status_info(skb, mdev, NULL)) {
2734                         genlmsg_cancel(skb, dh);
2735                         goto out;
2736                 }
2737                 genlmsg_end(skb, dh);
2738         }
2739
2740 out:
2741         up_read(&drbd_cfg_rwsem);
2742         /* where to start the next iteration */
2743         cb->args[0] = (long)pos;
2744         cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2745
2746         /* No more tconns/volumes/minors found results in an empty skb.
2747          * Which will terminate the dump. */
2748         return skb->len;
2749 }
2750
2751 /*
2752  * Request status of all resources, or of all volumes within a single resource.
2753  *
2754  * This is a dump, as the answer may not fit in a single reply skb otherwise.
2755  * Which means we cannot use the family->attrbuf or other such members, because
2756  * dump is NOT protected by the genl_lock().  During dump, we only have access
2757  * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
2758  *
2759  * Once things are setup properly, we call into get_one_status().
2760  */
2761 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2762 {
2763         const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
2764         struct nlattr *nla;
2765         const char *conn_name;
2766         struct drbd_tconn *tconn;
2767
2768         /* Is this a followup call? */
2769         if (cb->args[0]) {
2770                 /* ... of a single resource dump,
2771                  * and the resource iterator has been advanced already? */
2772                 if (cb->args[2] && cb->args[2] != cb->args[0])
2773                         return 0; /* DONE. */
2774                 goto dump;
2775         }
2776
2777         /* First call (from netlink_dump_start).  We need to figure out
2778          * which resource(s) the user wants us to dump. */
2779         nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
2780                         nlmsg_attrlen(cb->nlh, hdrlen),
2781                         DRBD_NLA_CFG_CONTEXT);
2782
2783         /* No explicit context given.  Dump all. */
2784         if (!nla)
2785                 goto dump;
2786         nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
2787         /* context given, but no name present? */
2788         if (!nla)
2789                 return -EINVAL;
2790         conn_name = nla_data(nla);
2791         tconn = conn_get_by_name(conn_name);
2792
2793         if (!tconn)
2794                 return -ENODEV;
2795
2796         kref_put(&tconn->kref, &conn_destroy); /* get_one_status() (re)validates tconn by itself */
2797
2798         /* prime iterators, and set "filter" mode mark:
2799          * only dump this tconn. */
2800         cb->args[0] = (long)tconn;
2801         /* cb->args[1] = 0; passed in this way. */
2802         cb->args[2] = (long)tconn;
2803
2804 dump:
2805         return get_one_status(skb, cb);
2806 }
2807
2808 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2809 {
2810         enum drbd_ret_code retcode;
2811         struct timeout_parms tp;
2812         int err;
2813
2814         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2815         if (!adm_ctx.reply_skb)
2816                 return retcode;
2817         if (retcode != NO_ERROR)
2818                 goto out;
2819
2820         tp.timeout_type =
2821                 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2822                 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2823                 UT_DEFAULT;
2824
2825         err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2826         if (err) {
2827                 nlmsg_free(adm_ctx.reply_skb);
2828                 return err;
2829         }
2830 out:
2831         drbd_adm_finish(info, retcode);
2832         return 0;
2833 }
2834
2835 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2836 {
2837         struct drbd_conf *mdev;
2838         enum drbd_ret_code retcode;
2839
2840         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2841         if (!adm_ctx.reply_skb)
2842                 return retcode;
2843         if (retcode != NO_ERROR)
2844                 goto out;
2845
2846         mdev = adm_ctx.mdev;
2847         if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2848                 /* resume from last known position, if possible */
2849                 struct start_ov_parms parms =
2850                         { .ov_start_sector = mdev->ov_start_sector };
2851                 int err = start_ov_parms_from_attrs(&parms, info);
2852                 if (err) {
2853                         retcode = ERR_MANDATORY_TAG;
2854                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2855                         goto out;
2856                 }
2857                 /* w_make_ov_request expects position to be aligned */
2858                 mdev->ov_start_sector = parms.ov_start_sector & ~BM_SECT_PER_BIT;
2859         }
2860         /* If there is still bitmap IO pending, e.g. previous resync or verify
2861          * just being finished, wait for it before requesting a new resync. */
2862         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2863         retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2864 out:
2865         drbd_adm_finish(info, retcode);
2866         return 0;
2867 }
2868
2869
2870 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
2871 {
2872         struct drbd_conf *mdev;
2873         enum drbd_ret_code retcode;
2874         int skip_initial_sync = 0;
2875         int err;
2876         struct new_c_uuid_parms args;
2877
2878         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2879         if (!adm_ctx.reply_skb)
2880                 return retcode;
2881         if (retcode != NO_ERROR)
2882                 goto out_nolock;
2883
2884         mdev = adm_ctx.mdev;
2885         memset(&args, 0, sizeof(args));
2886         if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
2887                 err = new_c_uuid_parms_from_attrs(&args, info);
2888                 if (err) {
2889                         retcode = ERR_MANDATORY_TAG;
2890                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2891                         goto out_nolock;
2892                 }
2893         }
2894
2895         mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2896
2897         if (!get_ldev(mdev)) {
2898                 retcode = ERR_NO_DISK;
2899                 goto out;
2900         }
2901
2902         /* this is "skip initial sync", assume to be clean */
2903         if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2904             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2905                 dev_info(DEV, "Preparing to skip initial sync\n");
2906                 skip_initial_sync = 1;
2907         } else if (mdev->state.conn != C_STANDALONE) {
2908                 retcode = ERR_CONNECTED;
2909                 goto out_dec;
2910         }
2911
2912         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2913         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2914
2915         if (args.clear_bm) {
2916                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2917                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2918                 if (err) {
2919                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2920                         retcode = ERR_IO_MD_DISK;
2921                 }
2922                 if (skip_initial_sync) {
2923                         drbd_send_uuids_skip_initial_sync(mdev);
2924                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2925                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2926                         spin_lock_irq(&mdev->tconn->req_lock);
2927                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2928                                         CS_VERBOSE, NULL);
2929                         spin_unlock_irq(&mdev->tconn->req_lock);
2930                 }
2931         }
2932
2933         drbd_md_sync(mdev);
2934 out_dec:
2935         put_ldev(mdev);
2936 out:
2937         mutex_unlock(mdev->state_mutex);
2938 out_nolock:
2939         drbd_adm_finish(info, retcode);
2940         return 0;
2941 }
2942
2943 static enum drbd_ret_code
2944 drbd_check_conn_name(const char *name)
2945 {
2946         if (!name || !name[0]) {
2947                 drbd_msg_put_info("connection name missing");
2948                 return ERR_MANDATORY_TAG;
2949         }
2950         /* if we want to use these in sysfs/configfs/debugfs some day,
2951          * we must not allow slashes */
2952         if (strchr(name, '/')) {
2953                 drbd_msg_put_info("invalid connection name");
2954                 return ERR_INVALID_REQUEST;
2955         }
2956         return NO_ERROR;
2957 }
2958
2959 int drbd_adm_create_connection(struct sk_buff *skb, struct genl_info *info)
2960 {
2961         enum drbd_ret_code retcode;
2962
2963         retcode = drbd_adm_prepare(skb, info, 0);
2964         if (!adm_ctx.reply_skb)
2965                 return retcode;
2966         if (retcode != NO_ERROR)
2967                 goto out;
2968
2969         retcode = drbd_check_conn_name(adm_ctx.conn_name);
2970         if (retcode != NO_ERROR)
2971                 goto out;
2972
2973         if (adm_ctx.tconn) {
2974                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
2975                         retcode = ERR_INVALID_REQUEST;
2976                         drbd_msg_put_info("connection exists");
2977                 }
2978                 /* else: still NO_ERROR */
2979                 goto out;
2980         }
2981
2982         if (!conn_create(adm_ctx.conn_name))
2983                 retcode = ERR_NOMEM;
2984 out:
2985         drbd_adm_finish(info, retcode);
2986         return 0;
2987 }
2988
2989 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
2990 {
2991         struct drbd_genlmsghdr *dh = info->userhdr;
2992         enum drbd_ret_code retcode;
2993
2994         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2995         if (!adm_ctx.reply_skb)
2996                 return retcode;
2997         if (retcode != NO_ERROR)
2998                 goto out;
2999
3000         /* FIXME drop minor_count parameter, limit to MINORMASK */
3001         if (dh->minor >= minor_count) {
3002                 drbd_msg_put_info("requested minor out of range");
3003                 retcode = ERR_INVALID_REQUEST;
3004                 goto out;
3005         }
3006         if (adm_ctx.volume > DRBD_VOLUME_MAX) {
3007                 drbd_msg_put_info("requested volume id out of range");
3008                 retcode = ERR_INVALID_REQUEST;
3009                 goto out;
3010         }
3011
3012         /* drbd_adm_prepare made sure already
3013          * that mdev->tconn and mdev->vnr match the request. */
3014         if (adm_ctx.mdev) {
3015                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
3016                         retcode = ERR_MINOR_EXISTS;
3017                 /* else: still NO_ERROR */
3018                 goto out;
3019         }
3020
3021         down_write(&drbd_cfg_rwsem);
3022         retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
3023         up_write(&drbd_cfg_rwsem);
3024 out:
3025         drbd_adm_finish(info, retcode);
3026         return 0;
3027 }
3028
3029 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
3030 {
3031         if (mdev->state.disk == D_DISKLESS &&
3032             /* no need to be mdev->state.conn == C_STANDALONE &&
3033              * we may want to delete a minor from a live replication group.
3034              */
3035             mdev->state.role == R_SECONDARY) {
3036                 idr_remove(&mdev->tconn->volumes, mdev->vnr);
3037                 idr_remove(&minors, mdev_to_minor(mdev));
3038                 del_gendisk(mdev->vdisk);
3039                 synchronize_rcu();
3040                 kref_put(&mdev->kref, &drbd_minor_destroy);
3041                 return NO_ERROR;
3042         } else
3043                 return ERR_MINOR_CONFIGURED;
3044 }
3045
3046 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
3047 {
3048         enum drbd_ret_code retcode;
3049
3050         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3051         if (!adm_ctx.reply_skb)
3052                 return retcode;
3053         if (retcode != NO_ERROR)
3054                 goto out;
3055
3056         down_write(&drbd_cfg_rwsem);
3057         retcode = adm_delete_minor(adm_ctx.mdev);
3058         up_write(&drbd_cfg_rwsem);
3059 out:
3060         drbd_adm_finish(info, retcode);
3061         return 0;
3062 }
3063
3064 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
3065 {
3066         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3067         struct drbd_conf *mdev;
3068         unsigned i;
3069
3070         retcode = drbd_adm_prepare(skb, info, 0);
3071         if (!adm_ctx.reply_skb)
3072                 return retcode;
3073         if (retcode != NO_ERROR)
3074                 goto out;
3075
3076         if (!adm_ctx.tconn) {
3077                 retcode = ERR_CONN_NOT_KNOWN;
3078                 goto out;
3079         }
3080
3081         down_read(&drbd_cfg_rwsem);
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_unlock;
3088                 }
3089         }
3090         up_read(&drbd_cfg_rwsem);
3091
3092         /* disconnect; may stop the receiver;
3093          * must not hold the drbd_cfg_rwsem */
3094         retcode = conn_try_disconnect(adm_ctx.tconn, 0);
3095         if (retcode < SS_SUCCESS) {
3096                 drbd_msg_put_info("failed to disconnect");
3097                 goto out;
3098         }
3099
3100         down_read(&drbd_cfg_rwsem);
3101         /* detach */
3102         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3103                 retcode = adm_detach(mdev);
3104                 if (retcode < SS_SUCCESS) {
3105                         drbd_msg_put_info("failed to detach");
3106                         goto out_unlock;
3107                 }
3108         }
3109         up_read(&drbd_cfg_rwsem);
3110
3111         /* If we reach this, all volumes (of this tconn) are Secondary,
3112          * Disconnected, Diskless, aka Unconfigured. Make sure all threads have
3113          * actually stopped, state handling only does drbd_thread_stop_nowait().
3114          * This needs to be done without holding drbd_cfg_rwsem. */
3115         drbd_thread_stop(&adm_ctx.tconn->worker);
3116
3117         /* Now, nothing can fail anymore */
3118
3119         /* delete volumes */
3120         down_write(&drbd_cfg_rwsem);
3121         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3122                 retcode = adm_delete_minor(mdev);
3123                 if (retcode != NO_ERROR) {
3124                         /* "can not happen" */
3125                         drbd_msg_put_info("failed to delete volume");
3126                         up_write(&drbd_cfg_rwsem);
3127                         goto out;
3128                 }
3129         }
3130
3131         /* delete connection */
3132         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3133                 list_del_rcu(&adm_ctx.tconn->all_tconn);
3134                 synchronize_rcu();
3135                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3136
3137                 retcode = NO_ERROR;
3138         } else {
3139                 /* "can not happen" */
3140                 retcode = ERR_CONN_IN_USE;
3141                 drbd_msg_put_info("failed to delete connection");
3142         }
3143         up_write(&drbd_cfg_rwsem);
3144         goto out;
3145 out_unlock:
3146         up_read(&drbd_cfg_rwsem);
3147 out:
3148         drbd_adm_finish(info, retcode);
3149         return 0;
3150 }
3151
3152 int drbd_adm_delete_connection(struct sk_buff *skb, struct genl_info *info)
3153 {
3154         enum drbd_ret_code retcode;
3155
3156         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
3157         if (!adm_ctx.reply_skb)
3158                 return retcode;
3159         if (retcode != NO_ERROR)
3160                 goto out;
3161
3162         down_write(&drbd_cfg_rwsem);
3163         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3164                 list_del_rcu(&adm_ctx.tconn->all_tconn);
3165                 synchronize_rcu();
3166                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3167
3168                 retcode = NO_ERROR;
3169         } else {
3170                 retcode = ERR_CONN_IN_USE;
3171         }
3172         up_write(&drbd_cfg_rwsem);
3173
3174         if (retcode == NO_ERROR)
3175                 drbd_thread_stop(&adm_ctx.tconn->worker);
3176 out:
3177         drbd_adm_finish(info, retcode);
3178         return 0;
3179 }
3180
3181 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
3182 {
3183         static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3184         struct sk_buff *msg;
3185         struct drbd_genlmsghdr *d_out;
3186         unsigned seq;
3187         int err = -ENOMEM;
3188
3189         seq = atomic_inc_return(&drbd_genl_seq);
3190         msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3191         if (!msg)
3192                 goto failed;
3193
3194         err = -EMSGSIZE;
3195         d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3196         if (!d_out) /* cannot happen, but anyways. */
3197                 goto nla_put_failure;
3198         d_out->minor = mdev_to_minor(mdev);
3199         d_out->ret_code = 0;
3200
3201         if (nla_put_status_info(msg, mdev, sib))
3202                 goto nla_put_failure;
3203         genlmsg_end(msg, d_out);
3204         err = drbd_genl_multicast_events(msg, 0);
3205         /* msg has been consumed or freed in netlink_broadcast() */
3206         if (err && err != -ESRCH)
3207                 goto failed;
3208
3209         return;
3210
3211 nla_put_failure:
3212         nlmsg_free(msg);
3213 failed:
3214         dev_err(DEV, "Error %d while broadcasting event. "
3215                         "Event seq:%u sib_reason:%u\n",
3216                         err, seq, sib->sib_reason);
3217 }