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 drbd_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 *rs_plan_s = 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                 rs_plan_s = fifo_alloc(fifo_size);
1162                 if (!rs_plan_s) {
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         spin_lock(&mdev->peer_seq_lock);
1192         if (rs_plan_s) {
1193                 kfree(mdev->rs_plan_s);
1194                 mdev->rs_plan_s = rs_plan_s;
1195                 rs_plan_s = NULL;
1196         }
1197         spin_unlock(&mdev->peer_seq_lock);
1198
1199         drbd_md_sync(mdev);
1200
1201         if (mdev->state.conn >= C_CONNECTED)
1202                 drbd_send_sync_param(mdev);
1203
1204         mutex_unlock(&mdev->tconn->conf_update);
1205         synchronize_rcu();
1206         kfree(old_disk_conf);
1207         goto success;
1208
1209 fail_unlock:
1210         mutex_unlock(&mdev->tconn->conf_update);
1211  fail:
1212         kfree(new_disk_conf);
1213         kfree(rs_plan_s);
1214 success:
1215         put_ldev(mdev);
1216  out:
1217         drbd_adm_finish(info, retcode);
1218         return 0;
1219 }
1220
1221 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1222 {
1223         struct drbd_conf *mdev;
1224         int err;
1225         enum drbd_ret_code retcode;
1226         enum determine_dev_size dd;
1227         sector_t max_possible_sectors;
1228         sector_t min_md_device_sectors;
1229         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1230         struct disk_conf *new_disk_conf = NULL;
1231         struct block_device *bdev;
1232         struct lru_cache *resync_lru = NULL;
1233         struct fifo_buffer *new_plan = NULL;
1234         union drbd_state ns, os;
1235         enum drbd_state_rv rv;
1236         struct net_conf *nc;
1237         int cp_discovered = 0;
1238
1239         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1240         if (!adm_ctx.reply_skb)
1241                 return retcode;
1242         if (retcode != NO_ERROR)
1243                 goto finish;
1244
1245         mdev = adm_ctx.mdev;
1246         conn_reconfig_start(mdev->tconn);
1247
1248         /* if you want to reconfigure, please tear down first */
1249         if (mdev->state.disk > D_DISKLESS) {
1250                 retcode = ERR_DISK_CONFIGURED;
1251                 goto fail;
1252         }
1253         /* It may just now have detached because of IO error.  Make sure
1254          * drbd_ldev_destroy is done already, we may end up here very fast,
1255          * e.g. if someone calls attach from the on-io-error handler,
1256          * to realize a "hot spare" feature (not that I'd recommend that) */
1257         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1258
1259         /* allocation not in the IO path, drbdsetup context */
1260         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1261         if (!nbc) {
1262                 retcode = ERR_NOMEM;
1263                 goto fail;
1264         }
1265         new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1266         if (!new_disk_conf) {
1267                 retcode = ERR_NOMEM;
1268                 goto fail;
1269         }
1270         nbc->disk_conf = new_disk_conf;
1271
1272         set_disk_conf_defaults(new_disk_conf);
1273         err = disk_conf_from_attrs(new_disk_conf, info);
1274         if (err) {
1275                 retcode = ERR_MANDATORY_TAG;
1276                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1277                 goto fail;
1278         }
1279
1280         enforce_disk_conf_limits(new_disk_conf);
1281
1282         new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1283         if (!new_plan) {
1284                 retcode = ERR_NOMEM;
1285                 goto fail;
1286         }
1287
1288         if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1289                 retcode = ERR_MD_IDX_INVALID;
1290                 goto fail;
1291         }
1292
1293         rcu_read_lock();
1294         nc = rcu_dereference(mdev->tconn->net_conf);
1295         if (nc) {
1296                 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1297                         rcu_read_unlock();
1298                         retcode = ERR_STONITH_AND_PROT_A;
1299                         goto fail;
1300                 }
1301         }
1302         rcu_read_unlock();
1303
1304         bdev = blkdev_get_by_path(new_disk_conf->backing_dev,
1305                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1306         if (IS_ERR(bdev)) {
1307                 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->backing_dev,
1308                         PTR_ERR(bdev));
1309                 retcode = ERR_OPEN_DISK;
1310                 goto fail;
1311         }
1312         nbc->backing_bdev = bdev;
1313
1314         /*
1315          * meta_dev_idx >= 0: external fixed size, possibly multiple
1316          * drbd sharing one meta device.  TODO in that case, paranoia
1317          * check that [md_bdev, meta_dev_idx] is not yet used by some
1318          * other drbd minor!  (if you use drbd.conf + drbdadm, that
1319          * should check it for you already; but if you don't, or
1320          * someone fooled it, we need to double check here)
1321          */
1322         bdev = blkdev_get_by_path(new_disk_conf->meta_dev,
1323                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1324                                   (new_disk_conf->meta_dev_idx < 0) ?
1325                                   (void *)mdev : (void *)drbd_m_holder);
1326         if (IS_ERR(bdev)) {
1327                 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->meta_dev,
1328                         PTR_ERR(bdev));
1329                 retcode = ERR_OPEN_MD_DISK;
1330                 goto fail;
1331         }
1332         nbc->md_bdev = bdev;
1333
1334         if ((nbc->backing_bdev == nbc->md_bdev) !=
1335             (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1336              new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1337                 retcode = ERR_MD_IDX_INVALID;
1338                 goto fail;
1339         }
1340
1341         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1342                         1, 61, sizeof(struct bm_extent),
1343                         offsetof(struct bm_extent, lce));
1344         if (!resync_lru) {
1345                 retcode = ERR_NOMEM;
1346                 goto fail;
1347         }
1348
1349         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1350         drbd_md_set_sector_offsets(mdev, nbc);
1351
1352         if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1353                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1354                         (unsigned long long) drbd_get_max_capacity(nbc),
1355                         (unsigned long long) new_disk_conf->disk_size);
1356                 retcode = ERR_DISK_TO_SMALL;
1357                 goto fail;
1358         }
1359
1360         if (new_disk_conf->meta_dev_idx < 0) {
1361                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1362                 /* at least one MB, otherwise it does not make sense */
1363                 min_md_device_sectors = (2<<10);
1364         } else {
1365                 max_possible_sectors = DRBD_MAX_SECTORS;
1366                 min_md_device_sectors = MD_RESERVED_SECT * (new_disk_conf->meta_dev_idx + 1);
1367         }
1368
1369         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1370                 retcode = ERR_MD_DISK_TO_SMALL;
1371                 dev_warn(DEV, "refusing attach: md-device too small, "
1372                      "at least %llu sectors needed for this meta-disk type\n",
1373                      (unsigned long long) min_md_device_sectors);
1374                 goto fail;
1375         }
1376
1377         /* Make sure the new disk is big enough
1378          * (we may currently be R_PRIMARY with no local disk...) */
1379         if (drbd_get_max_capacity(nbc) <
1380             drbd_get_capacity(mdev->this_bdev)) {
1381                 retcode = ERR_DISK_TO_SMALL;
1382                 goto fail;
1383         }
1384
1385         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1386
1387         if (nbc->known_size > max_possible_sectors) {
1388                 dev_warn(DEV, "==> truncating very big lower level device "
1389                         "to currently maximum possible %llu sectors <==\n",
1390                         (unsigned long long) max_possible_sectors);
1391                 if (new_disk_conf->meta_dev_idx >= 0)
1392                         dev_warn(DEV, "==>> using internal or flexible "
1393                                       "meta data may help <<==\n");
1394         }
1395
1396         drbd_suspend_io(mdev);
1397         /* also wait for the last barrier ack. */
1398         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || drbd_suspended(mdev));
1399         /* and for any other previously queued work */
1400         drbd_flush_workqueue(mdev);
1401
1402         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1403         retcode = rv;  /* FIXME: Type mismatch. */
1404         drbd_resume_io(mdev);
1405         if (rv < SS_SUCCESS)
1406                 goto fail;
1407
1408         if (!get_ldev_if_state(mdev, D_ATTACHING))
1409                 goto force_diskless;
1410
1411         drbd_md_set_sector_offsets(mdev, nbc);
1412
1413         if (!mdev->bitmap) {
1414                 if (drbd_bm_init(mdev)) {
1415                         retcode = ERR_NOMEM;
1416                         goto force_diskless_dec;
1417                 }
1418         }
1419
1420         retcode = drbd_md_read(mdev, nbc);
1421         if (retcode != NO_ERROR)
1422                 goto force_diskless_dec;
1423
1424         if (mdev->state.conn < C_CONNECTED &&
1425             mdev->state.role == R_PRIMARY &&
1426             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1427                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1428                     (unsigned long long)mdev->ed_uuid);
1429                 retcode = ERR_DATA_NOT_CURRENT;
1430                 goto force_diskless_dec;
1431         }
1432
1433         /* Since we are diskless, fix the activity log first... */
1434         if (drbd_check_al_size(mdev, new_disk_conf)) {
1435                 retcode = ERR_NOMEM;
1436                 goto force_diskless_dec;
1437         }
1438
1439         /* Prevent shrinking of consistent devices ! */
1440         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1441             drbd_new_dev_size(mdev, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1442                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1443                 retcode = ERR_DISK_TO_SMALL;
1444                 goto force_diskless_dec;
1445         }
1446
1447         if (!drbd_al_read_log(mdev, nbc)) {
1448                 retcode = ERR_IO_MD_DISK;
1449                 goto force_diskless_dec;
1450         }
1451
1452         /* Reset the "barriers don't work" bits here, then force meta data to
1453          * be written, to ensure we determine if barriers are supported. */
1454         if (new_disk_conf->no_md_flush)
1455                 set_bit(MD_NO_FUA, &mdev->flags);
1456         else
1457                 clear_bit(MD_NO_FUA, &mdev->flags);
1458
1459         /* Point of no return reached.
1460          * Devices and memory are no longer released by error cleanup below.
1461          * now mdev takes over responsibility, and the state engine should
1462          * clean it up somewhere.  */
1463         D_ASSERT(mdev->ldev == NULL);
1464         mdev->ldev = nbc;
1465         mdev->resync = resync_lru;
1466         mdev->rs_plan_s = new_plan;
1467         nbc = NULL;
1468         resync_lru = NULL;
1469         new_disk_conf = NULL;
1470         new_plan = NULL;
1471
1472         mdev->write_ordering = WO_bdev_flush;
1473         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1474
1475         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1476                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1477         else
1478                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1479
1480         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1481             !(mdev->state.role == R_PRIMARY && mdev->tconn->susp_nod)) {
1482                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1483                 cp_discovered = 1;
1484         }
1485
1486         mdev->send_cnt = 0;
1487         mdev->recv_cnt = 0;
1488         mdev->read_cnt = 0;
1489         mdev->writ_cnt = 0;
1490
1491         drbd_reconsider_max_bio_size(mdev);
1492
1493         /* If I am currently not R_PRIMARY,
1494          * but meta data primary indicator is set,
1495          * I just now recover from a hard crash,
1496          * and have been R_PRIMARY before that crash.
1497          *
1498          * Now, if I had no connection before that crash
1499          * (have been degraded R_PRIMARY), chances are that
1500          * I won't find my peer now either.
1501          *
1502          * In that case, and _only_ in that case,
1503          * we use the degr-wfc-timeout instead of the default,
1504          * so we can automatically recover from a crash of a
1505          * degraded but active "cluster" after a certain timeout.
1506          */
1507         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1508         if (mdev->state.role != R_PRIMARY &&
1509              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1510             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1511                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1512
1513         dd = drbd_determine_dev_size(mdev, 0);
1514         if (dd == dev_size_error) {
1515                 retcode = ERR_NOMEM_BITMAP;
1516                 goto force_diskless_dec;
1517         } else if (dd == grew)
1518                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1519
1520         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1521                 dev_info(DEV, "Assuming that all blocks are out of sync "
1522                      "(aka FullSync)\n");
1523                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1524                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1525                         retcode = ERR_IO_MD_DISK;
1526                         goto force_diskless_dec;
1527                 }
1528         } else {
1529                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1530                         "read from attaching", BM_LOCKED_MASK)) {
1531                         retcode = ERR_IO_MD_DISK;
1532                         goto force_diskless_dec;
1533                 }
1534         }
1535
1536         if (cp_discovered) {
1537                 drbd_al_apply_to_bm(mdev);
1538                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1539                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1540                         retcode = ERR_IO_MD_DISK;
1541                         goto force_diskless_dec;
1542                 }
1543         }
1544
1545         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1546                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1547
1548         spin_lock_irq(&mdev->tconn->req_lock);
1549         os = drbd_read_state(mdev);
1550         ns = os;
1551         /* If MDF_CONSISTENT is not set go into inconsistent state,
1552            otherwise investigate MDF_WasUpToDate...
1553            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1554            otherwise into D_CONSISTENT state.
1555         */
1556         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1557                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1558                         ns.disk = D_CONSISTENT;
1559                 else
1560                         ns.disk = D_OUTDATED;
1561         } else {
1562                 ns.disk = D_INCONSISTENT;
1563         }
1564
1565         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1566                 ns.pdsk = D_OUTDATED;
1567
1568         rcu_read_lock();
1569         if (ns.disk == D_CONSISTENT &&
1570             (ns.pdsk == D_OUTDATED || rcu_dereference(mdev->ldev->disk_conf)->fencing == FP_DONT_CARE))
1571                 ns.disk = D_UP_TO_DATE;
1572         rcu_read_unlock();
1573
1574         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1575            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1576            this point, because drbd_request_state() modifies these
1577            flags. */
1578
1579         /* In case we are C_CONNECTED postpone any decision on the new disk
1580            state after the negotiation phase. */
1581         if (mdev->state.conn == C_CONNECTED) {
1582                 mdev->new_state_tmp.i = ns.i;
1583                 ns.i = os.i;
1584                 ns.disk = D_NEGOTIATING;
1585
1586                 /* We expect to receive up-to-date UUIDs soon.
1587                    To avoid a race in receive_state, free p_uuid while
1588                    holding req_lock. I.e. atomic with the state change */
1589                 kfree(mdev->p_uuid);
1590                 mdev->p_uuid = NULL;
1591         }
1592
1593         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1594         spin_unlock_irq(&mdev->tconn->req_lock);
1595
1596         if (rv < SS_SUCCESS)
1597                 goto force_diskless_dec;
1598
1599         if (mdev->state.role == R_PRIMARY)
1600                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1601         else
1602                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1603
1604         drbd_md_mark_dirty(mdev);
1605         drbd_md_sync(mdev);
1606
1607         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1608         put_ldev(mdev);
1609         conn_reconfig_done(mdev->tconn);
1610         drbd_adm_finish(info, retcode);
1611         return 0;
1612
1613  force_diskless_dec:
1614         put_ldev(mdev);
1615  force_diskless:
1616         drbd_force_state(mdev, NS(disk, D_FAILED));
1617         drbd_md_sync(mdev);
1618  fail:
1619         conn_reconfig_done(mdev->tconn);
1620         if (nbc) {
1621                 if (nbc->backing_bdev)
1622                         blkdev_put(nbc->backing_bdev,
1623                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1624                 if (nbc->md_bdev)
1625                         blkdev_put(nbc->md_bdev,
1626                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1627                 kfree(nbc);
1628         }
1629         kfree(new_disk_conf);
1630         lc_destroy(resync_lru);
1631         kfree(new_plan);
1632
1633  finish:
1634         drbd_adm_finish(info, retcode);
1635         return 0;
1636 }
1637
1638 static int adm_detach(struct drbd_conf *mdev)
1639 {
1640         enum drbd_state_rv retcode;
1641         int ret;
1642         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1643         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1644         /* D_FAILED will transition to DISKLESS. */
1645         ret = wait_event_interruptible(mdev->misc_wait,
1646                         mdev->state.disk != D_FAILED);
1647         drbd_resume_io(mdev);
1648         if ((int)retcode == (int)SS_IS_DISKLESS)
1649                 retcode = SS_NOTHING_TO_DO;
1650         if (ret)
1651                 retcode = ERR_INTR;
1652         return retcode;
1653 }
1654
1655 /* Detaching the disk is a process in multiple stages.  First we need to lock
1656  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1657  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1658  * internal references as well.
1659  * Only then we have finally detached. */
1660 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1661 {
1662         enum drbd_ret_code retcode;
1663
1664         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1665         if (!adm_ctx.reply_skb)
1666                 return retcode;
1667         if (retcode != NO_ERROR)
1668                 goto out;
1669
1670         retcode = adm_detach(adm_ctx.mdev);
1671 out:
1672         drbd_adm_finish(info, retcode);
1673         return 0;
1674 }
1675
1676 static bool conn_resync_running(struct drbd_tconn *tconn)
1677 {
1678         struct drbd_conf *mdev;
1679         bool rv = false;
1680         int vnr;
1681
1682         rcu_read_lock();
1683         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1684                 if (mdev->state.conn == C_SYNC_SOURCE ||
1685                     mdev->state.conn == C_SYNC_TARGET ||
1686                     mdev->state.conn == C_PAUSED_SYNC_S ||
1687                     mdev->state.conn == C_PAUSED_SYNC_T) {
1688                         rv = true;
1689                         break;
1690                 }
1691         }
1692         rcu_read_unlock();
1693
1694         return rv;
1695 }
1696
1697 static bool conn_ov_running(struct drbd_tconn *tconn)
1698 {
1699         struct drbd_conf *mdev;
1700         bool rv = false;
1701         int vnr;
1702
1703         rcu_read_lock();
1704         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1705                 if (mdev->state.conn == C_VERIFY_S ||
1706                     mdev->state.conn == C_VERIFY_T) {
1707                         rv = true;
1708                         break;
1709                 }
1710         }
1711         rcu_read_unlock();
1712
1713         return rv;
1714 }
1715
1716 static enum drbd_ret_code
1717 _check_net_options(struct drbd_tconn *tconn, struct net_conf *old_conf, struct net_conf *new_conf)
1718 {
1719         struct drbd_conf *mdev;
1720         int i;
1721
1722         if (old_conf && tconn->agreed_pro_version < 100 &&
1723             tconn->cstate == C_WF_REPORT_PARAMS &&
1724             new_conf->wire_protocol != old_conf->wire_protocol)
1725                 return ERR_NEED_APV_100;
1726
1727         if (new_conf->two_primaries &&
1728             (new_conf->wire_protocol != DRBD_PROT_C))
1729                 return ERR_NOT_PROTO_C;
1730
1731         idr_for_each_entry(&tconn->volumes, mdev, i) {
1732                 if (get_ldev(mdev)) {
1733                         enum drbd_fencing_p fp = rcu_dereference(mdev->ldev->disk_conf)->fencing;
1734                         put_ldev(mdev);
1735                         if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
1736                                 return ERR_STONITH_AND_PROT_A;
1737                 }
1738                 if (mdev->state.role == R_PRIMARY && new_conf->want_lose)
1739                         return ERR_DISCARD;
1740         }
1741
1742         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A)
1743                 return ERR_CONG_NOT_PROTO_A;
1744
1745         return NO_ERROR;
1746 }
1747
1748 static enum drbd_ret_code
1749 check_net_options(struct drbd_tconn *tconn, struct net_conf *new_conf)
1750 {
1751         static enum drbd_ret_code rv;
1752         struct drbd_conf *mdev;
1753         int i;
1754
1755         rcu_read_lock();
1756         rv = _check_net_options(tconn, rcu_dereference(tconn->net_conf), new_conf);
1757         rcu_read_unlock();
1758
1759         /* tconn->volumes protected by genl_lock() here */
1760         idr_for_each_entry(&tconn->volumes, mdev, i) {
1761                 if (!mdev->bitmap) {
1762                         if(drbd_bm_init(mdev))
1763                                 return ERR_NOMEM;
1764                 }
1765         }
1766
1767         return rv;
1768 }
1769
1770 struct crypto {
1771         struct crypto_hash *verify_tfm;
1772         struct crypto_hash *csums_tfm;
1773         struct crypto_hash *cram_hmac_tfm;
1774         struct crypto_hash *integrity_tfm;
1775         void *int_dig_in;
1776         void *int_dig_vv;
1777 };
1778
1779 static int
1780 alloc_hash(struct crypto_hash **tfm, char *tfm_name, int err_alg)
1781 {
1782         if (!tfm_name[0])
1783                 return NO_ERROR;
1784
1785         *tfm = crypto_alloc_hash(tfm_name, 0, CRYPTO_ALG_ASYNC);
1786         if (IS_ERR(*tfm)) {
1787                 *tfm = NULL;
1788                 return err_alg;
1789         }
1790
1791         return NO_ERROR;
1792 }
1793
1794 static enum drbd_ret_code
1795 alloc_crypto(struct crypto *crypto, struct net_conf *new_conf)
1796 {
1797         char hmac_name[CRYPTO_MAX_ALG_NAME];
1798         enum drbd_ret_code rv;
1799         int hash_size;
1800
1801         rv = alloc_hash(&crypto->csums_tfm, new_conf->csums_alg,
1802                        ERR_CSUMS_ALG);
1803         if (rv != NO_ERROR)
1804                 return rv;
1805         rv = alloc_hash(&crypto->verify_tfm, new_conf->verify_alg,
1806                        ERR_VERIFY_ALG);
1807         if (rv != NO_ERROR)
1808                 return rv;
1809         rv = alloc_hash(&crypto->integrity_tfm, new_conf->integrity_alg,
1810                        ERR_INTEGRITY_ALG);
1811         if (rv != NO_ERROR)
1812                 return rv;
1813         if (new_conf->cram_hmac_alg[0] != 0) {
1814                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1815                          new_conf->cram_hmac_alg);
1816
1817                 rv = alloc_hash(&crypto->cram_hmac_tfm, hmac_name,
1818                                ERR_AUTH_ALG);
1819         }
1820         if (crypto->integrity_tfm) {
1821                 hash_size = crypto_hash_digestsize(crypto->integrity_tfm);
1822                 crypto->int_dig_in = kmalloc(hash_size, GFP_KERNEL);
1823                 if (!crypto->int_dig_in)
1824                         return ERR_NOMEM;
1825                 crypto->int_dig_vv = kmalloc(hash_size, GFP_KERNEL);
1826                 if (!crypto->int_dig_vv)
1827                         return ERR_NOMEM;
1828         }
1829
1830         return rv;
1831 }
1832
1833 static void free_crypto(struct crypto *crypto)
1834 {
1835         kfree(crypto->int_dig_in);
1836         kfree(crypto->int_dig_vv);
1837         crypto_free_hash(crypto->cram_hmac_tfm);
1838         crypto_free_hash(crypto->integrity_tfm);
1839         crypto_free_hash(crypto->csums_tfm);
1840         crypto_free_hash(crypto->verify_tfm);
1841 }
1842
1843 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
1844 {
1845         enum drbd_ret_code retcode;
1846         struct drbd_tconn *tconn;
1847         struct net_conf *old_conf, *new_conf = NULL;
1848         int err;
1849         int ovr; /* online verify running */
1850         int rsr; /* re-sync running */
1851         struct crypto crypto = { };
1852         bool change_integrity_alg;
1853
1854         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1855         if (!adm_ctx.reply_skb)
1856                 return retcode;
1857         if (retcode != NO_ERROR)
1858                 goto out;
1859
1860         tconn = adm_ctx.tconn;
1861
1862         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1863         if (!new_conf) {
1864                 retcode = ERR_NOMEM;
1865                 goto out;
1866         }
1867
1868         conn_reconfig_start(tconn);
1869
1870         mutex_lock(&tconn->data.mutex);
1871         mutex_lock(&tconn->conf_update);
1872         old_conf = tconn->net_conf;
1873
1874         if (!old_conf) {
1875                 drbd_msg_put_info("net conf missing, try connect");
1876                 retcode = ERR_INVALID_REQUEST;
1877                 goto fail;
1878         }
1879
1880         *new_conf = *old_conf;
1881         if (should_set_defaults(info))
1882                 set_net_conf_defaults(new_conf);
1883
1884         err = net_conf_from_attrs_for_change(new_conf, info);
1885         if (err) {
1886                 retcode = ERR_MANDATORY_TAG;
1887                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1888                 goto fail;
1889         }
1890
1891         retcode = check_net_options(tconn, new_conf);
1892         if (retcode != NO_ERROR)
1893                 goto fail;
1894
1895         /* re-sync running */
1896         rsr = conn_resync_running(tconn);
1897         if (rsr && strcmp(new_conf->csums_alg, old_conf->csums_alg)) {
1898                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1899                 goto fail;
1900         }
1901
1902         /* online verify running */
1903         ovr = conn_ov_running(tconn);
1904         if (ovr && strcmp(new_conf->verify_alg, old_conf->verify_alg)) {
1905                 retcode = ERR_VERIFY_RUNNING;
1906                 goto fail;
1907         }
1908
1909         change_integrity_alg = strcmp(old_conf->integrity_alg,
1910                                       new_conf->integrity_alg);
1911
1912         retcode = alloc_crypto(&crypto, new_conf);
1913         if (retcode != NO_ERROR)
1914                 goto fail;
1915
1916         rcu_assign_pointer(tconn->net_conf, new_conf);
1917
1918         if (!rsr) {
1919                 crypto_free_hash(tconn->csums_tfm);
1920                 tconn->csums_tfm = crypto.csums_tfm;
1921                 crypto.csums_tfm = NULL;
1922         }
1923         if (!ovr) {
1924                 crypto_free_hash(tconn->verify_tfm);
1925                 tconn->verify_tfm = crypto.verify_tfm;
1926                 crypto.verify_tfm = NULL;
1927         }
1928
1929         kfree(tconn->int_dig_in);
1930         tconn->int_dig_in = crypto.int_dig_in;
1931         kfree(tconn->int_dig_vv);
1932         tconn->int_dig_vv = crypto.int_dig_vv;
1933         crypto_free_hash(tconn->integrity_tfm);
1934         tconn->integrity_tfm = crypto.integrity_tfm;
1935         if (change_integrity_alg) {
1936                 /* Do this without trying to take tconn->data.mutex again.  */
1937                 if (__drbd_send_protocol(tconn))
1938                         goto fail;
1939         }
1940
1941         /* FIXME Changing cram_hmac while the connection is established is useless */
1942         crypto_free_hash(tconn->cram_hmac_tfm);
1943         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
1944
1945         mutex_unlock(&tconn->conf_update);
1946         mutex_unlock(&tconn->data.mutex);
1947         synchronize_rcu();
1948         kfree(old_conf);
1949
1950         if (tconn->cstate >= C_WF_REPORT_PARAMS)
1951                 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn)));
1952
1953         goto done;
1954
1955  fail:
1956         mutex_unlock(&tconn->conf_update);
1957         mutex_unlock(&tconn->data.mutex);
1958         free_crypto(&crypto);
1959         kfree(new_conf);
1960  done:
1961         conn_reconfig_done(tconn);
1962  out:
1963         drbd_adm_finish(info, retcode);
1964         return 0;
1965 }
1966
1967 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
1968 {
1969         struct drbd_conf *mdev;
1970         struct net_conf *old_conf, *new_conf = NULL;
1971         struct crypto crypto = { };
1972         struct drbd_tconn *oconn;
1973         struct drbd_tconn *tconn;
1974         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1975         enum drbd_ret_code retcode;
1976         int i;
1977         int err;
1978
1979         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1980         if (!adm_ctx.reply_skb)
1981                 return retcode;
1982         if (retcode != NO_ERROR)
1983                 goto out;
1984
1985         tconn = adm_ctx.tconn;
1986         conn_reconfig_start(tconn);
1987
1988         if (tconn->cstate > C_STANDALONE) {
1989                 retcode = ERR_NET_CONFIGURED;
1990                 goto fail;
1991         }
1992
1993         /* allocation not in the IO path, cqueue thread context */
1994         new_conf = kzalloc(sizeof(*new_conf), GFP_KERNEL);
1995         if (!new_conf) {
1996                 retcode = ERR_NOMEM;
1997                 goto fail;
1998         }
1999
2000         set_net_conf_defaults(new_conf);
2001
2002         err = net_conf_from_attrs(new_conf, info);
2003         if (err) {
2004                 retcode = ERR_MANDATORY_TAG;
2005                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2006                 goto fail;
2007         }
2008
2009         retcode = check_net_options(tconn, new_conf);
2010         if (retcode != NO_ERROR)
2011                 goto fail;
2012
2013         retcode = NO_ERROR;
2014
2015         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
2016         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
2017
2018         /* No need to take drbd_cfg_rwsem here.  All reconfiguration is
2019          * strictly serialized on genl_lock(). We are protected against
2020          * concurrent reconfiguration/addition/deletion */
2021         list_for_each_entry(oconn, &drbd_tconns, all_tconn) {
2022                 struct net_conf *nc;
2023                 if (oconn == tconn)
2024                         continue;
2025
2026                 rcu_read_lock();
2027                 nc = rcu_dereference(oconn->net_conf);
2028                 if (nc) {
2029                         taken_addr = (struct sockaddr *)&nc->my_addr;
2030                         if (new_conf->my_addr_len == nc->my_addr_len &&
2031                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
2032                                 retcode = ERR_LOCAL_ADDR;
2033
2034                         taken_addr = (struct sockaddr *)&nc->peer_addr;
2035                         if (new_conf->peer_addr_len == nc->peer_addr_len &&
2036                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
2037                                 retcode = ERR_PEER_ADDR;
2038                 }
2039                 rcu_read_unlock();
2040                 if (retcode != NO_ERROR)
2041                         goto fail;
2042         }
2043
2044         retcode = alloc_crypto(&crypto, new_conf);
2045         if (retcode != NO_ERROR)
2046                 goto fail;
2047
2048         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2049
2050         conn_flush_workqueue(tconn);
2051
2052         mutex_lock(&tconn->conf_update);
2053         old_conf = tconn->net_conf;
2054         if (old_conf) {
2055                 retcode = ERR_NET_CONFIGURED;
2056                 mutex_unlock(&tconn->conf_update);
2057                 goto fail;
2058         }
2059         rcu_assign_pointer(tconn->net_conf, new_conf);
2060
2061         conn_free_crypto(tconn);
2062         tconn->int_dig_in = crypto.int_dig_in;
2063         tconn->int_dig_vv = crypto.int_dig_vv;
2064         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
2065         tconn->integrity_tfm = crypto.integrity_tfm;
2066         tconn->csums_tfm = crypto.csums_tfm;
2067         tconn->verify_tfm = crypto.verify_tfm;
2068
2069         mutex_unlock(&tconn->conf_update);
2070
2071         rcu_read_lock();
2072         idr_for_each_entry(&tconn->volumes, mdev, i) {
2073                 mdev->send_cnt = 0;
2074                 mdev->recv_cnt = 0;
2075         }
2076         rcu_read_unlock();
2077
2078         retcode = conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2079
2080         conn_reconfig_done(tconn);
2081         drbd_adm_finish(info, retcode);
2082         return 0;
2083
2084 fail:
2085         free_crypto(&crypto);
2086         kfree(new_conf);
2087
2088         conn_reconfig_done(tconn);
2089 out:
2090         drbd_adm_finish(info, retcode);
2091         return 0;
2092 }
2093
2094 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
2095 {
2096         enum drbd_state_rv rv;
2097
2098         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2099                         force ? CS_HARD : 0);
2100
2101         switch (rv) {
2102         case SS_NOTHING_TO_DO:
2103                 break;
2104         case SS_ALREADY_STANDALONE:
2105                 return SS_SUCCESS;
2106         case SS_PRIMARY_NOP:
2107                 /* Our state checking code wants to see the peer outdated. */
2108                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2109                                                 pdsk, D_OUTDATED), CS_VERBOSE);
2110                 break;
2111         case SS_CW_FAILED_BY_PEER:
2112                 /* The peer probably wants to see us outdated. */
2113                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2114                                                         disk, D_OUTDATED), 0);
2115                 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2116                         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2117                                         CS_HARD);
2118                 }
2119                 break;
2120         default:;
2121                 /* no special handling necessary */
2122         }
2123
2124         if (rv >= SS_SUCCESS) {
2125                 enum drbd_state_rv rv2;
2126                 /* No one else can reconfigure the network while I am here.
2127                  * The state handling only uses drbd_thread_stop_nowait(),
2128                  * we want to really wait here until the receiver is no more.
2129                  */
2130                 drbd_thread_stop(&adm_ctx.tconn->receiver);
2131
2132                 /* Race breaker.  This additional state change request may be
2133                  * necessary, if this was a forced disconnect during a receiver
2134                  * restart.  We may have "killed" the receiver thread just
2135                  * after drbdd_init() returned.  Typically, we should be
2136                  * C_STANDALONE already, now, and this becomes a no-op.
2137                  */
2138                 rv2 = conn_request_state(tconn, NS(conn, C_STANDALONE),
2139                                 CS_VERBOSE | CS_HARD);
2140                 if (rv2 < SS_SUCCESS)
2141                         conn_err(tconn,
2142                                 "unexpected rv2=%d in conn_try_disconnect()\n",
2143                                 rv2);
2144         }
2145         return rv;
2146 }
2147
2148 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2149 {
2150         struct disconnect_parms parms;
2151         struct drbd_tconn *tconn;
2152         enum drbd_state_rv rv;
2153         enum drbd_ret_code retcode;
2154         int err;
2155
2156         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2157         if (!adm_ctx.reply_skb)
2158                 return retcode;
2159         if (retcode != NO_ERROR)
2160                 goto fail;
2161
2162         tconn = adm_ctx.tconn;
2163         memset(&parms, 0, sizeof(parms));
2164         if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2165                 err = disconnect_parms_from_attrs(&parms, info);
2166                 if (err) {
2167                         retcode = ERR_MANDATORY_TAG;
2168                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2169                         goto fail;
2170                 }
2171         }
2172
2173         rv = conn_try_disconnect(tconn, parms.force_disconnect);
2174         if (rv < SS_SUCCESS)
2175                 retcode = rv;  /* FIXME: Type mismatch. */
2176         else
2177                 retcode = NO_ERROR;
2178  fail:
2179         drbd_adm_finish(info, retcode);
2180         return 0;
2181 }
2182
2183 void resync_after_online_grow(struct drbd_conf *mdev)
2184 {
2185         int iass; /* I am sync source */
2186
2187         dev_info(DEV, "Resync of new storage after online grow\n");
2188         if (mdev->state.role != mdev->state.peer)
2189                 iass = (mdev->state.role == R_PRIMARY);
2190         else
2191                 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
2192
2193         if (iass)
2194                 drbd_start_resync(mdev, C_SYNC_SOURCE);
2195         else
2196                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2197 }
2198
2199 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2200 {
2201         struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2202         struct resize_parms rs;
2203         struct drbd_conf *mdev;
2204         enum drbd_ret_code retcode;
2205         enum determine_dev_size dd;
2206         enum dds_flags ddsf;
2207         sector_t u_size;
2208         int err;
2209
2210         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2211         if (!adm_ctx.reply_skb)
2212                 return retcode;
2213         if (retcode != NO_ERROR)
2214                 goto fail;
2215
2216         memset(&rs, 0, sizeof(struct resize_parms));
2217         if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2218                 err = resize_parms_from_attrs(&rs, info);
2219                 if (err) {
2220                         retcode = ERR_MANDATORY_TAG;
2221                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2222                         goto fail;
2223                 }
2224         }
2225
2226         mdev = adm_ctx.mdev;
2227         if (mdev->state.conn > C_CONNECTED) {
2228                 retcode = ERR_RESIZE_RESYNC;
2229                 goto fail;
2230         }
2231
2232         if (mdev->state.role == R_SECONDARY &&
2233             mdev->state.peer == R_SECONDARY) {
2234                 retcode = ERR_NO_PRIMARY;
2235                 goto fail;
2236         }
2237
2238         if (!get_ldev(mdev)) {
2239                 retcode = ERR_NO_DISK;
2240                 goto fail;
2241         }
2242
2243         if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
2244                 retcode = ERR_NEED_APV_93;
2245                 goto fail;
2246         }
2247
2248         rcu_read_lock();
2249         u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
2250         rcu_read_unlock();
2251         if (u_size != (sector_t)rs.resize_size) {
2252                 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2253                 if (!new_disk_conf) {
2254                         retcode = ERR_NOMEM;
2255                         goto fail;
2256                 }
2257         }
2258
2259         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
2260                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2261
2262         if (new_disk_conf) {
2263                 mutex_lock(&mdev->tconn->conf_update);
2264                 old_disk_conf = mdev->ldev->disk_conf;
2265                 *new_disk_conf = *old_disk_conf;
2266                 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2267                 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
2268                 mutex_unlock(&mdev->tconn->conf_update);
2269                 synchronize_rcu();
2270                 kfree(old_disk_conf);
2271         }
2272
2273         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2274         dd = drbd_determine_dev_size(mdev, ddsf);
2275         drbd_md_sync(mdev);
2276         put_ldev(mdev);
2277         if (dd == dev_size_error) {
2278                 retcode = ERR_NOMEM_BITMAP;
2279                 goto fail;
2280         }
2281
2282         if (mdev->state.conn == C_CONNECTED) {
2283                 if (dd == grew)
2284                         set_bit(RESIZE_PENDING, &mdev->flags);
2285
2286                 drbd_send_uuids(mdev);
2287                 drbd_send_sizes(mdev, 1, ddsf);
2288         }
2289
2290  fail:
2291         drbd_adm_finish(info, retcode);
2292         return 0;
2293 }
2294
2295 void drbd_set_res_opts_defaults(struct res_opts *r)
2296 {
2297         return set_res_opts_defaults(r);
2298 }
2299
2300 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2301 {
2302         enum drbd_ret_code retcode;
2303         cpumask_var_t new_cpu_mask;
2304         struct drbd_tconn *tconn;
2305         struct res_opts res_opts;
2306         int err;
2307
2308         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2309         if (!adm_ctx.reply_skb)
2310                 return retcode;
2311         if (retcode != NO_ERROR)
2312                 goto fail;
2313         tconn = adm_ctx.tconn;
2314
2315         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
2316                 retcode = ERR_NOMEM;
2317                 drbd_msg_put_info("unable to allocate cpumask");
2318                 goto fail;
2319         }
2320
2321         res_opts = tconn->res_opts;
2322         if (should_set_defaults(info))
2323                 set_res_opts_defaults(&res_opts);
2324
2325         err = res_opts_from_attrs(&res_opts, info);
2326         if (err) {
2327                 retcode = ERR_MANDATORY_TAG;
2328                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2329                 goto fail;
2330         }
2331
2332         /* silently ignore cpu mask on UP kernel */
2333         if (nr_cpu_ids > 1 && res_opts.cpu_mask[0] != 0) {
2334                 err = __bitmap_parse(res_opts.cpu_mask, 32, 0,
2335                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2336                 if (err) {
2337                         conn_warn(tconn, "__bitmap_parse() failed with %d\n", err);
2338                         retcode = ERR_CPU_MASK_PARSE;
2339                         goto fail;
2340                 }
2341         }
2342
2343
2344         tconn->res_opts = res_opts;
2345
2346         if (!cpumask_equal(tconn->cpu_mask, new_cpu_mask)) {
2347                 cpumask_copy(tconn->cpu_mask, new_cpu_mask);
2348                 drbd_calc_cpu_mask(tconn);
2349                 tconn->receiver.reset_cpu_mask = 1;
2350                 tconn->asender.reset_cpu_mask = 1;
2351                 tconn->worker.reset_cpu_mask = 1;
2352         }
2353
2354 fail:
2355         free_cpumask_var(new_cpu_mask);
2356
2357         drbd_adm_finish(info, retcode);
2358         return 0;
2359 }
2360
2361 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2362 {
2363         struct drbd_conf *mdev;
2364         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2365
2366         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2367         if (!adm_ctx.reply_skb)
2368                 return retcode;
2369         if (retcode != NO_ERROR)
2370                 goto out;
2371
2372         mdev = adm_ctx.mdev;
2373
2374         /* If there is still bitmap IO pending, probably because of a previous
2375          * resync just being finished, wait for it before requesting a new resync. */
2376         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2377
2378         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2379
2380         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2381                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2382
2383         while (retcode == SS_NEED_CONNECTION) {
2384                 spin_lock_irq(&mdev->tconn->req_lock);
2385                 if (mdev->state.conn < C_CONNECTED)
2386                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2387                 spin_unlock_irq(&mdev->tconn->req_lock);
2388
2389                 if (retcode != SS_NEED_CONNECTION)
2390                         break;
2391
2392                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2393         }
2394
2395 out:
2396         drbd_adm_finish(info, retcode);
2397         return 0;
2398 }
2399
2400 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2401 {
2402         int rv;
2403
2404         rv = drbd_bmio_set_n_write(mdev);
2405         drbd_suspend_al(mdev);
2406         return rv;
2407 }
2408
2409 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2410                 union drbd_state mask, union drbd_state val)
2411 {
2412         enum drbd_ret_code retcode;
2413
2414         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2415         if (!adm_ctx.reply_skb)
2416                 return retcode;
2417         if (retcode != NO_ERROR)
2418                 goto out;
2419
2420         retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2421 out:
2422         drbd_adm_finish(info, retcode);
2423         return 0;
2424 }
2425
2426 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2427 {
2428         return drbd_adm_simple_request_state(skb, info, NS(conn, C_STARTING_SYNC_S));
2429 }
2430
2431 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2432 {
2433         enum drbd_ret_code retcode;
2434
2435         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2436         if (!adm_ctx.reply_skb)
2437                 return retcode;
2438         if (retcode != NO_ERROR)
2439                 goto out;
2440
2441         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2442                 retcode = ERR_PAUSE_IS_SET;
2443 out:
2444         drbd_adm_finish(info, retcode);
2445         return 0;
2446 }
2447
2448 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2449 {
2450         union drbd_dev_state s;
2451         enum drbd_ret_code retcode;
2452
2453         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2454         if (!adm_ctx.reply_skb)
2455                 return retcode;
2456         if (retcode != NO_ERROR)
2457                 goto out;
2458
2459         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2460                 s = adm_ctx.mdev->state;
2461                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2462                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2463                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2464                 } else {
2465                         retcode = ERR_PAUSE_IS_CLEAR;
2466                 }
2467         }
2468
2469 out:
2470         drbd_adm_finish(info, retcode);
2471         return 0;
2472 }
2473
2474 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2475 {
2476         return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2477 }
2478
2479 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2480 {
2481         struct drbd_conf *mdev;
2482         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2483
2484         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2485         if (!adm_ctx.reply_skb)
2486                 return retcode;
2487         if (retcode != NO_ERROR)
2488                 goto out;
2489
2490         mdev = adm_ctx.mdev;
2491         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2492                 drbd_uuid_new_current(mdev);
2493                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2494         }
2495         drbd_suspend_io(mdev);
2496         retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2497         if (retcode == SS_SUCCESS) {
2498                 if (mdev->state.conn < C_CONNECTED)
2499                         tl_clear(mdev->tconn);
2500                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2501                         tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2502         }
2503         drbd_resume_io(mdev);
2504
2505 out:
2506         drbd_adm_finish(info, retcode);
2507         return 0;
2508 }
2509
2510 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2511 {
2512         return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2513 }
2514
2515 int nla_put_drbd_cfg_context(struct sk_buff *skb, const char *conn_name, unsigned vnr)
2516 {
2517         struct nlattr *nla;
2518         nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2519         if (!nla)
2520                 goto nla_put_failure;
2521         if (vnr != VOLUME_UNSPECIFIED)
2522                 NLA_PUT_U32(skb, T_ctx_volume, vnr);
2523         NLA_PUT_STRING(skb, T_ctx_conn_name, conn_name);
2524         nla_nest_end(skb, nla);
2525         return 0;
2526
2527 nla_put_failure:
2528         if (nla)
2529                 nla_nest_cancel(skb, nla);
2530         return -EMSGSIZE;
2531 }
2532
2533 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2534                 const struct sib_info *sib)
2535 {
2536         struct state_info *si = NULL; /* for sizeof(si->member); */
2537         struct net_conf *nc;
2538         struct nlattr *nla;
2539         int got_ldev;
2540         int err = 0;
2541         int exclude_sensitive;
2542
2543         /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2544          * to.  So we better exclude_sensitive information.
2545          *
2546          * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2547          * in the context of the requesting user process. Exclude sensitive
2548          * information, unless current has superuser.
2549          *
2550          * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2551          * relies on the current implementation of netlink_dump(), which
2552          * executes the dump callback successively from netlink_recvmsg(),
2553          * always in the context of the receiving process */
2554         exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2555
2556         got_ldev = get_ldev(mdev);
2557
2558         /* We need to add connection name and volume number information still.
2559          * Minor number is in drbd_genlmsghdr. */
2560         if (nla_put_drbd_cfg_context(skb, mdev->tconn->name, mdev->vnr))
2561                 goto nla_put_failure;
2562
2563         if (res_opts_to_skb(skb, &mdev->tconn->res_opts, exclude_sensitive))
2564                 goto nla_put_failure;
2565
2566         rcu_read_lock();
2567         if (got_ldev)
2568                 if (disk_conf_to_skb(skb, rcu_dereference(mdev->ldev->disk_conf), exclude_sensitive))
2569                         goto nla_put_failure;
2570
2571         nc = rcu_dereference(mdev->tconn->net_conf);
2572         if (nc)
2573                 err = net_conf_to_skb(skb, nc, exclude_sensitive);
2574         rcu_read_unlock();
2575         if (err)
2576                 goto nla_put_failure;
2577
2578         nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2579         if (!nla)
2580                 goto nla_put_failure;
2581         NLA_PUT_U32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY);
2582         NLA_PUT_U32(skb, T_current_state, mdev->state.i);
2583         NLA_PUT_U64(skb, T_ed_uuid, mdev->ed_uuid);
2584         NLA_PUT_U64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev));
2585
2586         if (got_ldev) {
2587                 NLA_PUT_U32(skb, T_disk_flags, mdev->ldev->md.flags);
2588                 NLA_PUT(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2589                 NLA_PUT_U64(skb, T_bits_total, drbd_bm_bits(mdev));
2590                 NLA_PUT_U64(skb, T_bits_oos, drbd_bm_total_weight(mdev));
2591                 if (C_SYNC_SOURCE <= mdev->state.conn &&
2592                     C_PAUSED_SYNC_T >= mdev->state.conn) {
2593                         NLA_PUT_U64(skb, T_bits_rs_total, mdev->rs_total);
2594                         NLA_PUT_U64(skb, T_bits_rs_failed, mdev->rs_failed);
2595                 }
2596         }
2597
2598         if (sib) {
2599                 switch(sib->sib_reason) {
2600                 case SIB_SYNC_PROGRESS:
2601                 case SIB_GET_STATUS_REPLY:
2602                         break;
2603                 case SIB_STATE_CHANGE:
2604                         NLA_PUT_U32(skb, T_prev_state, sib->os.i);
2605                         NLA_PUT_U32(skb, T_new_state, sib->ns.i);
2606                         break;
2607                 case SIB_HELPER_POST:
2608                         NLA_PUT_U32(skb,
2609                                 T_helper_exit_code, sib->helper_exit_code);
2610                         /* fall through */
2611                 case SIB_HELPER_PRE:
2612                         NLA_PUT_STRING(skb, T_helper, sib->helper_name);
2613                         break;
2614                 }
2615         }
2616         nla_nest_end(skb, nla);
2617
2618         if (0)
2619 nla_put_failure:
2620                 err = -EMSGSIZE;
2621         if (got_ldev)
2622                 put_ldev(mdev);
2623         return err;
2624 }
2625
2626 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2627 {
2628         enum drbd_ret_code retcode;
2629         int err;
2630
2631         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2632         if (!adm_ctx.reply_skb)
2633                 return retcode;
2634         if (retcode != NO_ERROR)
2635                 goto out;
2636
2637         err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2638         if (err) {
2639                 nlmsg_free(adm_ctx.reply_skb);
2640                 return err;
2641         }
2642 out:
2643         drbd_adm_finish(info, retcode);
2644         return 0;
2645 }
2646
2647 int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
2648 {
2649         struct drbd_conf *mdev;
2650         struct drbd_genlmsghdr *dh;
2651         struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2652         struct drbd_tconn *tconn = NULL;
2653         struct drbd_tconn *tmp;
2654         unsigned volume = cb->args[1];
2655
2656         /* Open coded, deferred, iteration:
2657          * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2658          *      idr_for_each_entry(&tconn->volumes, mdev, i) {
2659          *        ...
2660          *      }
2661          * }
2662          * where tconn is cb->args[0];
2663          * and i is cb->args[1];
2664          *
2665          * cb->args[2] indicates if we shall loop over all resources,
2666          * or just dump all volumes of a single resource.
2667          *
2668          * This may miss entries inserted after this dump started,
2669          * or entries deleted before they are reached.
2670          *
2671          * We need to make sure the mdev won't disappear while
2672          * we are looking at it, and revalidate our iterators
2673          * on each iteration.
2674          */
2675
2676         /* synchronize with conn_create()/conn_destroy() */
2677         down_read(&drbd_cfg_rwsem);
2678         /* revalidate iterator position */
2679         list_for_each_entry(tmp, &drbd_tconns, all_tconn) {
2680                 if (pos == NULL) {
2681                         /* first iteration */
2682                         pos = tmp;
2683                         tconn = pos;
2684                         break;
2685                 }
2686                 if (tmp == pos) {
2687                         tconn = pos;
2688                         break;
2689                 }
2690         }
2691         if (tconn) {
2692 next_tconn:
2693                 mdev = idr_get_next(&tconn->volumes, &volume);
2694                 if (!mdev) {
2695                         /* No more volumes to dump on this tconn.
2696                          * Advance tconn iterator. */
2697                         pos = list_entry(tconn->all_tconn.next,
2698                                         struct drbd_tconn, all_tconn);
2699                         /* Did we dump any volume on this tconn yet? */
2700                         if (volume != 0) {
2701                                 /* If we reached the end of the list,
2702                                  * or only a single resource dump was requested,
2703                                  * we are done. */
2704                                 if (&pos->all_tconn == &drbd_tconns || cb->args[2])
2705                                         goto out;
2706                                 volume = 0;
2707                                 tconn = pos;
2708                                 goto next_tconn;
2709                         }
2710                 }
2711
2712                 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).pid,
2713                                 cb->nlh->nlmsg_seq, &drbd_genl_family,
2714                                 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2715                 if (!dh)
2716                         goto out;
2717
2718                 if (!mdev) {
2719                         /* this is a tconn without a single volume */
2720                         dh->minor = -1U;
2721                         dh->ret_code = NO_ERROR;
2722                         if (nla_put_drbd_cfg_context(skb, tconn->name, VOLUME_UNSPECIFIED))
2723                                 genlmsg_cancel(skb, dh);
2724                         else
2725                                 genlmsg_end(skb, dh);
2726                         goto out;
2727                 }
2728
2729                 D_ASSERT(mdev->vnr == volume);
2730                 D_ASSERT(mdev->tconn == tconn);
2731
2732                 dh->minor = mdev_to_minor(mdev);
2733                 dh->ret_code = NO_ERROR;
2734
2735                 if (nla_put_status_info(skb, mdev, NULL)) {
2736                         genlmsg_cancel(skb, dh);
2737                         goto out;
2738                 }
2739                 genlmsg_end(skb, dh);
2740         }
2741
2742 out:
2743         up_read(&drbd_cfg_rwsem);
2744         /* where to start the next iteration */
2745         cb->args[0] = (long)pos;
2746         cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2747
2748         /* No more tconns/volumes/minors found results in an empty skb.
2749          * Which will terminate the dump. */
2750         return skb->len;
2751 }
2752
2753 /*
2754  * Request status of all resources, or of all volumes within a single resource.
2755  *
2756  * This is a dump, as the answer may not fit in a single reply skb otherwise.
2757  * Which means we cannot use the family->attrbuf or other such members, because
2758  * dump is NOT protected by the genl_lock().  During dump, we only have access
2759  * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
2760  *
2761  * Once things are setup properly, we call into get_one_status().
2762  */
2763 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2764 {
2765         const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
2766         struct nlattr *nla;
2767         const char *conn_name;
2768         struct drbd_tconn *tconn;
2769
2770         /* Is this a followup call? */
2771         if (cb->args[0]) {
2772                 /* ... of a single resource dump,
2773                  * and the resource iterator has been advanced already? */
2774                 if (cb->args[2] && cb->args[2] != cb->args[0])
2775                         return 0; /* DONE. */
2776                 goto dump;
2777         }
2778
2779         /* First call (from netlink_dump_start).  We need to figure out
2780          * which resource(s) the user wants us to dump. */
2781         nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
2782                         nlmsg_attrlen(cb->nlh, hdrlen),
2783                         DRBD_NLA_CFG_CONTEXT);
2784
2785         /* No explicit context given.  Dump all. */
2786         if (!nla)
2787                 goto dump;
2788         nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
2789         /* context given, but no name present? */
2790         if (!nla)
2791                 return -EINVAL;
2792         conn_name = nla_data(nla);
2793         tconn = conn_get_by_name(conn_name);
2794
2795         if (!tconn)
2796                 return -ENODEV;
2797
2798         kref_put(&tconn->kref, &conn_destroy); /* get_one_status() (re)validates tconn by itself */
2799
2800         /* prime iterators, and set "filter" mode mark:
2801          * only dump this tconn. */
2802         cb->args[0] = (long)tconn;
2803         /* cb->args[1] = 0; passed in this way. */
2804         cb->args[2] = (long)tconn;
2805
2806 dump:
2807         return get_one_status(skb, cb);
2808 }
2809
2810 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2811 {
2812         enum drbd_ret_code retcode;
2813         struct timeout_parms tp;
2814         int err;
2815
2816         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2817         if (!adm_ctx.reply_skb)
2818                 return retcode;
2819         if (retcode != NO_ERROR)
2820                 goto out;
2821
2822         tp.timeout_type =
2823                 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2824                 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2825                 UT_DEFAULT;
2826
2827         err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2828         if (err) {
2829                 nlmsg_free(adm_ctx.reply_skb);
2830                 return err;
2831         }
2832 out:
2833         drbd_adm_finish(info, retcode);
2834         return 0;
2835 }
2836
2837 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2838 {
2839         struct drbd_conf *mdev;
2840         enum drbd_ret_code retcode;
2841
2842         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2843         if (!adm_ctx.reply_skb)
2844                 return retcode;
2845         if (retcode != NO_ERROR)
2846                 goto out;
2847
2848         mdev = adm_ctx.mdev;
2849         if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2850                 /* resume from last known position, if possible */
2851                 struct start_ov_parms parms =
2852                         { .ov_start_sector = mdev->ov_start_sector };
2853                 int err = start_ov_parms_from_attrs(&parms, info);
2854                 if (err) {
2855                         retcode = ERR_MANDATORY_TAG;
2856                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2857                         goto out;
2858                 }
2859                 /* w_make_ov_request expects position to be aligned */
2860                 mdev->ov_start_sector = parms.ov_start_sector & ~BM_SECT_PER_BIT;
2861         }
2862         /* If there is still bitmap IO pending, e.g. previous resync or verify
2863          * just being finished, wait for it before requesting a new resync. */
2864         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2865         retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2866 out:
2867         drbd_adm_finish(info, retcode);
2868         return 0;
2869 }
2870
2871
2872 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
2873 {
2874         struct drbd_conf *mdev;
2875         enum drbd_ret_code retcode;
2876         int skip_initial_sync = 0;
2877         int err;
2878         struct new_c_uuid_parms args;
2879
2880         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2881         if (!adm_ctx.reply_skb)
2882                 return retcode;
2883         if (retcode != NO_ERROR)
2884                 goto out_nolock;
2885
2886         mdev = adm_ctx.mdev;
2887         memset(&args, 0, sizeof(args));
2888         if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
2889                 err = new_c_uuid_parms_from_attrs(&args, info);
2890                 if (err) {
2891                         retcode = ERR_MANDATORY_TAG;
2892                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2893                         goto out_nolock;
2894                 }
2895         }
2896
2897         mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2898
2899         if (!get_ldev(mdev)) {
2900                 retcode = ERR_NO_DISK;
2901                 goto out;
2902         }
2903
2904         /* this is "skip initial sync", assume to be clean */
2905         if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2906             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2907                 dev_info(DEV, "Preparing to skip initial sync\n");
2908                 skip_initial_sync = 1;
2909         } else if (mdev->state.conn != C_STANDALONE) {
2910                 retcode = ERR_CONNECTED;
2911                 goto out_dec;
2912         }
2913
2914         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2915         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2916
2917         if (args.clear_bm) {
2918                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2919                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2920                 if (err) {
2921                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2922                         retcode = ERR_IO_MD_DISK;
2923                 }
2924                 if (skip_initial_sync) {
2925                         drbd_send_uuids_skip_initial_sync(mdev);
2926                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2927                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2928                         spin_lock_irq(&mdev->tconn->req_lock);
2929                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2930                                         CS_VERBOSE, NULL);
2931                         spin_unlock_irq(&mdev->tconn->req_lock);
2932                 }
2933         }
2934
2935         drbd_md_sync(mdev);
2936 out_dec:
2937         put_ldev(mdev);
2938 out:
2939         mutex_unlock(mdev->state_mutex);
2940 out_nolock:
2941         drbd_adm_finish(info, retcode);
2942         return 0;
2943 }
2944
2945 static enum drbd_ret_code
2946 drbd_check_conn_name(const char *name)
2947 {
2948         if (!name || !name[0]) {
2949                 drbd_msg_put_info("connection name missing");
2950                 return ERR_MANDATORY_TAG;
2951         }
2952         /* if we want to use these in sysfs/configfs/debugfs some day,
2953          * we must not allow slashes */
2954         if (strchr(name, '/')) {
2955                 drbd_msg_put_info("invalid connection name");
2956                 return ERR_INVALID_REQUEST;
2957         }
2958         return NO_ERROR;
2959 }
2960
2961 int drbd_adm_create_connection(struct sk_buff *skb, struct genl_info *info)
2962 {
2963         enum drbd_ret_code retcode;
2964
2965         retcode = drbd_adm_prepare(skb, info, 0);
2966         if (!adm_ctx.reply_skb)
2967                 return retcode;
2968         if (retcode != NO_ERROR)
2969                 goto out;
2970
2971         retcode = drbd_check_conn_name(adm_ctx.conn_name);
2972         if (retcode != NO_ERROR)
2973                 goto out;
2974
2975         if (adm_ctx.tconn) {
2976                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
2977                         retcode = ERR_INVALID_REQUEST;
2978                         drbd_msg_put_info("connection exists");
2979                 }
2980                 /* else: still NO_ERROR */
2981                 goto out;
2982         }
2983
2984         if (!conn_create(adm_ctx.conn_name))
2985                 retcode = ERR_NOMEM;
2986 out:
2987         drbd_adm_finish(info, retcode);
2988         return 0;
2989 }
2990
2991 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
2992 {
2993         struct drbd_genlmsghdr *dh = info->userhdr;
2994         enum drbd_ret_code retcode;
2995
2996         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2997         if (!adm_ctx.reply_skb)
2998                 return retcode;
2999         if (retcode != NO_ERROR)
3000                 goto out;
3001
3002         /* FIXME drop minor_count parameter, limit to MINORMASK */
3003         if (dh->minor >= minor_count) {
3004                 drbd_msg_put_info("requested minor out of range");
3005                 retcode = ERR_INVALID_REQUEST;
3006                 goto out;
3007         }
3008         if (adm_ctx.volume > DRBD_VOLUME_MAX) {
3009                 drbd_msg_put_info("requested volume id out of range");
3010                 retcode = ERR_INVALID_REQUEST;
3011                 goto out;
3012         }
3013
3014         /* drbd_adm_prepare made sure already
3015          * that mdev->tconn and mdev->vnr match the request. */
3016         if (adm_ctx.mdev) {
3017                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
3018                         retcode = ERR_MINOR_EXISTS;
3019                 /* else: still NO_ERROR */
3020                 goto out;
3021         }
3022
3023         down_write(&drbd_cfg_rwsem);
3024         retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
3025         up_write(&drbd_cfg_rwsem);
3026 out:
3027         drbd_adm_finish(info, retcode);
3028         return 0;
3029 }
3030
3031 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
3032 {
3033         if (mdev->state.disk == D_DISKLESS &&
3034             /* no need to be mdev->state.conn == C_STANDALONE &&
3035              * we may want to delete a minor from a live replication group.
3036              */
3037             mdev->state.role == R_SECONDARY) {
3038                 drbd_delete_device(mdev);
3039                 return NO_ERROR;
3040         } else
3041                 return ERR_MINOR_CONFIGURED;
3042 }
3043
3044 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
3045 {
3046         enum drbd_ret_code retcode;
3047
3048         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3049         if (!adm_ctx.reply_skb)
3050                 return retcode;
3051         if (retcode != NO_ERROR)
3052                 goto out;
3053
3054         down_write(&drbd_cfg_rwsem);
3055         retcode = adm_delete_minor(adm_ctx.mdev);
3056         up_write(&drbd_cfg_rwsem);
3057 out:
3058         drbd_adm_finish(info, retcode);
3059         return 0;
3060 }
3061
3062 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
3063 {
3064         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3065         struct drbd_conf *mdev;
3066         unsigned i;
3067
3068         retcode = drbd_adm_prepare(skb, info, 0);
3069         if (!adm_ctx.reply_skb)
3070                 return retcode;
3071         if (retcode != NO_ERROR)
3072                 goto out;
3073
3074         if (!adm_ctx.tconn) {
3075                 retcode = ERR_CONN_NOT_KNOWN;
3076                 goto out;
3077         }
3078
3079         down_read(&drbd_cfg_rwsem);
3080         /* demote */
3081         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3082                 retcode = drbd_set_role(mdev, R_SECONDARY, 0);
3083                 if (retcode < SS_SUCCESS) {
3084                         drbd_msg_put_info("failed to demote");
3085                         goto out_unlock;
3086                 }
3087         }
3088         up_read(&drbd_cfg_rwsem);
3089
3090         /* disconnect; may stop the receiver;
3091          * must not hold the drbd_cfg_rwsem */
3092         retcode = conn_try_disconnect(adm_ctx.tconn, 0);
3093         if (retcode < SS_SUCCESS) {
3094                 drbd_msg_put_info("failed to disconnect");
3095                 goto out;
3096         }
3097
3098         down_read(&drbd_cfg_rwsem);
3099         /* detach */
3100         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3101                 retcode = adm_detach(mdev);
3102                 if (retcode < SS_SUCCESS) {
3103                         drbd_msg_put_info("failed to detach");
3104                         goto out_unlock;
3105                 }
3106         }
3107         up_read(&drbd_cfg_rwsem);
3108
3109         /* If we reach this, all volumes (of this tconn) are Secondary,
3110          * Disconnected, Diskless, aka Unconfigured. Make sure all threads have
3111          * actually stopped, state handling only does drbd_thread_stop_nowait().
3112          * This needs to be done without holding drbd_cfg_rwsem. */
3113         drbd_thread_stop(&adm_ctx.tconn->worker);
3114
3115         /* Now, nothing can fail anymore */
3116
3117         /* delete volumes */
3118         down_write(&drbd_cfg_rwsem);
3119         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3120                 retcode = adm_delete_minor(mdev);
3121                 if (retcode != NO_ERROR) {
3122                         /* "can not happen" */
3123                         drbd_msg_put_info("failed to delete volume");
3124                         up_write(&drbd_cfg_rwsem);
3125                         goto out;
3126                 }
3127         }
3128
3129         /* delete connection */
3130         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3131                 list_del(&adm_ctx.tconn->all_tconn);
3132                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3133
3134                 retcode = NO_ERROR;
3135         } else {
3136                 /* "can not happen" */
3137                 retcode = ERR_CONN_IN_USE;
3138                 drbd_msg_put_info("failed to delete connection");
3139         }
3140         up_write(&drbd_cfg_rwsem);
3141         goto out;
3142 out_unlock:
3143         up_read(&drbd_cfg_rwsem);
3144 out:
3145         drbd_adm_finish(info, retcode);
3146         return 0;
3147 }
3148
3149 int drbd_adm_delete_connection(struct sk_buff *skb, struct genl_info *info)
3150 {
3151         enum drbd_ret_code retcode;
3152
3153         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
3154         if (!adm_ctx.reply_skb)
3155                 return retcode;
3156         if (retcode != NO_ERROR)
3157                 goto out;
3158
3159         down_write(&drbd_cfg_rwsem);
3160         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3161                 list_del(&adm_ctx.tconn->all_tconn);
3162                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3163
3164                 retcode = NO_ERROR;
3165         } else {
3166                 retcode = ERR_CONN_IN_USE;
3167         }
3168         up_write(&drbd_cfg_rwsem);
3169
3170         if (retcode == NO_ERROR)
3171                 drbd_thread_stop(&adm_ctx.tconn->worker);
3172 out:
3173         drbd_adm_finish(info, retcode);
3174         return 0;
3175 }
3176
3177 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
3178 {
3179         static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3180         struct sk_buff *msg;
3181         struct drbd_genlmsghdr *d_out;
3182         unsigned seq;
3183         int err = -ENOMEM;
3184
3185         seq = atomic_inc_return(&drbd_genl_seq);
3186         msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3187         if (!msg)
3188                 goto failed;
3189
3190         err = -EMSGSIZE;
3191         d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3192         if (!d_out) /* cannot happen, but anyways. */
3193                 goto nla_put_failure;
3194         d_out->minor = mdev_to_minor(mdev);
3195         d_out->ret_code = 0;
3196
3197         if (nla_put_status_info(msg, mdev, sib))
3198                 goto nla_put_failure;
3199         genlmsg_end(msg, d_out);
3200         err = drbd_genl_multicast_events(msg, 0);
3201         /* msg has been consumed or freed in netlink_broadcast() */
3202         if (err && err != -ESRCH)
3203                 goto failed;
3204
3205         return;
3206
3207 nla_put_failure:
3208         nlmsg_free(msg);
3209 failed:
3210         dev_err(DEV, "Error %d while broadcasting event. "
3211                         "Event seq:%u sib_reason:%u\n",
3212                         err, seq, sib->sib_reason);
3213 }