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_new_resource(struct sk_buff *skb, struct genl_info *info);
  51 int drbd_adm_del_resource(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 "drbd_nla.h"
  79 #include <linux/genl_magic_func.h>
  80
  81 /* used blkdev_get_by_path, to claim our meta data device(s) */
  82 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
  83
  84 /* Configuration is strictly serialized, because generic netlink message
  85  * processing is strictly serialized by the genl_lock().
  86  * Which means we can use one static global drbd_config_context struct.
  87  */
  88 static struct drbd_config_context {
  89         /* assigned from drbd_genlmsghdr */
  90         unsigned int minor;
  91         /* assigned from request attributes, if present */
  92         unsigned int volume;
  93 #define VOLUME_UNSPECIFIED              (-1U)
  94         /* pointer into the request skb,
  95          * limited lifetime! */
  96         char *resource_name;
  97         struct nlattr *my_addr;
  98         struct nlattr *peer_addr;
  99
 100         /* reply buffer */
 101         struct sk_buff *reply_skb;
 102         /* pointer into reply buffer */
 103         struct drbd_genlmsghdr *reply_dh;
 104         /* resolved from attributes, if possible */
 105         struct drbd_conf *mdev;
 106         struct drbd_tconn *tconn;
 107 } adm_ctx;
 108
 109 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
 110 {
 111         genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
 112         if (genlmsg_reply(skb, info))
 113                 printk(KERN_ERR "drbd: error sending genl reply\n");
 114 }
 115
 116 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
 117  * reason it could fail was no space in skb, and there are 4k available. */
 118 int drbd_msg_put_info(const char *info)
 119 {
 120         struct sk_buff *skb = adm_ctx.reply_skb;
 121         struct nlattr *nla;
 122         int err = -EMSGSIZE;
 123
 124         if (!info || !info[0])
 125                 return 0;
 126
 127         nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY);
 128         if (!nla)
 129                 return err;
 130
 131         err = nla_put_string(skb, T_info_text, info);
 132         if (err) {
 133                 nla_nest_cancel(skb, nla);
 134                 return err;
 135         } else
 136                 nla_nest_end(skb, nla);
 137         return 0;
 138 }
 139
 140 /* This would be a good candidate for a "pre_doit" hook,
 141  * and per-family private info->pointers.
 142  * But we need to stay compatible with older kernels.
 143  * If it returns successfully, adm_ctx members are valid.
 144  */
 145 #define DRBD_ADM_NEED_MINOR     1
 146 #define DRBD_ADM_NEED_RESOURCE  2
 147 #define DRBD_ADM_NEED_CONNECTION 4
 148 static int drbd_adm_prepare(struct sk_buff *skb, struct genl_info *info,
 149                 unsigned flags)
 150 {
 151         struct drbd_genlmsghdr *d_in = info->userhdr;
 152         const u8 cmd = info->genlhdr->cmd;
 153         int err;
 154
 155         memset(&adm_ctx, 0, sizeof(adm_ctx));
 156
 157         /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
 158         if (cmd != DRBD_ADM_GET_STATUS
 159         && security_netlink_recv(skb, CAP_SYS_ADMIN))
 160                return -EPERM;
 161
 162         adm_ctx.reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
 163         if (!adm_ctx.reply_skb) {
 164                 err = -ENOMEM;
 165                 goto fail;
 166         }
 167
 168         adm_ctx.reply_dh = genlmsg_put_reply(adm_ctx.reply_skb,
 169                                         info, &drbd_genl_family, 0, cmd);
 170         /* put of a few bytes into a fresh skb of >= 4k will always succeed.
 171          * but anyways */
 172         if (!adm_ctx.reply_dh) {
 173                 err = -ENOMEM;
 174                 goto fail;
 175         }
 176
 177         adm_ctx.reply_dh->minor = d_in->minor;
 178         adm_ctx.reply_dh->ret_code = NO_ERROR;
 179
 180         adm_ctx.volume = VOLUME_UNSPECIFIED;
 181         if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
 182                 struct nlattr *nla;
 183                 /* parse and validate only */
 184                 err = drbd_cfg_context_from_attrs(NULL, info);
 185                 if (err)
 186                         goto fail;
 187
 188                 /* It was present, and valid,
 189                  * copy it over to the reply skb. */
 190                 err = nla_put_nohdr(adm_ctx.reply_skb,
 191                                 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
 192                                 info->attrs[DRBD_NLA_CFG_CONTEXT]);
 193                 if (err)
 194                         goto fail;
 195
 196                 /* and assign stuff to the global adm_ctx */
 197                 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
 198                 if (nla)
 199                         adm_ctx.volume = nla_get_u32(nla);
 200                 nla = nested_attr_tb[__nla_type(T_ctx_resource_name)];
 201                 if (nla)
 202                         adm_ctx.resource_name = nla_data(nla);
 203                 adm_ctx.my_addr = nested_attr_tb[__nla_type(T_ctx_my_addr)];
 204                 adm_ctx.peer_addr = nested_attr_tb[__nla_type(T_ctx_peer_addr)];
 205                 if ((adm_ctx.my_addr &&
 206                      nla_len(adm_ctx.my_addr) > sizeof(adm_ctx.tconn->my_addr)) ||
 207                     (adm_ctx.peer_addr &&
 208                      nla_len(adm_ctx.peer_addr) > sizeof(adm_ctx.tconn->peer_addr))) {
 209                         err = -EINVAL;
 210                         goto fail;
 211                 }
 212         }
 213
 214         adm_ctx.minor = d_in->minor;
 215         adm_ctx.mdev = minor_to_mdev(d_in->minor);
 216         adm_ctx.tconn = conn_get_by_name(adm_ctx.resource_name);
 217
 218         if (!adm_ctx.mdev && (flags & DRBD_ADM_NEED_MINOR)) {
 219                 drbd_msg_put_info("unknown minor");
 220                 return ERR_MINOR_INVALID;
 221         }
 222         if (!adm_ctx.tconn && (flags & DRBD_ADM_NEED_RESOURCE)) {
 223                 drbd_msg_put_info("unknown resource");
 224                 return ERR_INVALID_REQUEST;
 225         }
 226
 227         if (flags & DRBD_ADM_NEED_CONNECTION) {
 228                 if (adm_ctx.tconn && !(flags & DRBD_ADM_NEED_RESOURCE)) {
 229                         drbd_msg_put_info("no resource name expected");
 230                         return ERR_INVALID_REQUEST;
 231                 }
 232                 if (adm_ctx.mdev) {
 233                         drbd_msg_put_info("no minor number expected");
 234                         return ERR_INVALID_REQUEST;
 235                 }
 236                 if (adm_ctx.my_addr && adm_ctx.peer_addr)
 237                         adm_ctx.tconn = conn_get_by_addrs(nla_data(adm_ctx.my_addr),
 238                                                           nla_len(adm_ctx.my_addr),
 239                                                           nla_data(adm_ctx.peer_addr),
 240                                                           nla_len(adm_ctx.peer_addr));
 241                 if (!adm_ctx.tconn) {
 242                         drbd_msg_put_info("unknown connection");
 243                         return ERR_INVALID_REQUEST;
 244                 }
 245         }
 246
 247         /* some more paranoia, if the request was over-determined */
 248         if (adm_ctx.mdev && adm_ctx.tconn &&
 249             adm_ctx.mdev->tconn != adm_ctx.tconn) {
 250                 pr_warning("request: minor=%u, resource=%s; but that minor belongs to connection %s\n",
 251                                 adm_ctx.minor, adm_ctx.resource_name,
 252                                 adm_ctx.mdev->tconn->name);
 253                 drbd_msg_put_info("minor exists in different resource");
 254                 return ERR_INVALID_REQUEST;
 255         }
 256         if (adm_ctx.mdev &&
 257             adm_ctx.volume != VOLUME_UNSPECIFIED &&
 258             adm_ctx.volume != adm_ctx.mdev->vnr) {
 259                 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
 260                                 adm_ctx.minor, adm_ctx.volume,
 261                                 adm_ctx.mdev->vnr, adm_ctx.mdev->tconn->name);
 262                 drbd_msg_put_info("minor exists as different volume");
 263                 return ERR_INVALID_REQUEST;
 264         }
 265
 266         return NO_ERROR;
 267
 268 fail:
 269         nlmsg_free(adm_ctx.reply_skb);
 270         adm_ctx.reply_skb = NULL;
 271         return err;
 272 }
 273
 274 static int drbd_adm_finish(struct genl_info *info, int retcode)
 275 {
 276         if (adm_ctx.tconn) {
 277                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
 278                 adm_ctx.tconn = NULL;
 279         }
 280
 281         if (!adm_ctx.reply_skb)
 282                 return -ENOMEM;
 283
 284         adm_ctx.reply_dh->ret_code = retcode;
 285         drbd_adm_send_reply(adm_ctx.reply_skb, info);
 286         return 0;
 287 }
 288
 289 static void setup_khelper_env(struct drbd_tconn *tconn, char **envp)
 290 {
 291         char *afs;
 292
 293         /* FIXME: A future version will not allow this case. */
 294         if (tconn->my_addr_len == 0 || tconn->peer_addr_len == 0)
 295                 return;
 296
 297         switch (((struct sockaddr *)&tconn->peer_addr)->sa_family) {
 298         case AF_INET6:
 299                 afs = "ipv6";
 300                 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
 301                          &((struct sockaddr_in6 *)&tconn->peer_addr)->sin6_addr);
 302                 break;
 303         case AF_INET:
 304                 afs = "ipv4";
 305                 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
 306                          &((struct sockaddr_in *)&tconn->peer_addr)->sin_addr);
 307                 break;
 308         default:
 309                 afs = "ssocks";
 310                 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
 311                          &((struct sockaddr_in *)&tconn->peer_addr)->sin_addr);
 312         }
 313         snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
 314 }
 315
 316 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
 317 {
 318         char *envp[] = { "HOME=/",
 319                         "TERM=linux",
 320                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 321                          (char[20]) { }, /* address family */
 322                          (char[60]) { }, /* address */
 323                         NULL };
 324         char mb[12];
 325         char *argv[] = {usermode_helper, cmd, mb, NULL };
 326         struct sib_info sib;
 327         int ret;
 328
 329         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 330         setup_khelper_env(mdev->tconn, envp);
 331
 332         /* The helper may take some time.
 333          * write out any unsynced meta data changes now */
 334         drbd_md_sync(mdev);
 335
 336         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 337         sib.sib_reason = SIB_HELPER_PRE;
 338         sib.helper_name = cmd;
 339         drbd_bcast_event(mdev, &sib);
 340         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 341         if (ret)
 342                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 343                                 usermode_helper, cmd, mb,
 344                                 (ret >> 8) & 0xff, ret);
 345         else
 346                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 347                                 usermode_helper, cmd, mb,
 348                                 (ret >> 8) & 0xff, ret);
 349         sib.sib_reason = SIB_HELPER_POST;
 350         sib.helper_exit_code = ret;
 351         drbd_bcast_event(mdev, &sib);
 352
 353         if (ret < 0) /* Ignore any ERRNOs we got. */
 354                 ret = 0;
 355
 356         return ret;
 357 }
 358
 359 static void conn_md_sync(struct drbd_tconn *tconn)
 360 {
 361         struct drbd_conf *mdev;
 362         int vnr;
 363
 364         rcu_read_lock();
 365         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
 366                 kref_get(&mdev->kref);
 367                 rcu_read_unlock();
 368                 drbd_md_sync(mdev);
 369                 kref_put(&mdev->kref, &drbd_minor_destroy);
 370                 rcu_read_lock();
 371         }
 372         rcu_read_unlock();
 373 }
 374
 375 int conn_khelper(struct drbd_tconn *tconn, char *cmd)
 376 {
 377         char *envp[] = { "HOME=/",
 378                         "TERM=linux",
 379                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 380                          (char[20]) { }, /* address family */
 381                          (char[60]) { }, /* address */
 382                         NULL };
 383         char *argv[] = {usermode_helper, cmd, tconn->name, NULL };
 384         int ret;
 385
 386         setup_khelper_env(tconn, envp);
 387         conn_md_sync(tconn);
 388
 389         conn_info(tconn, "helper command: %s %s %s\n", usermode_helper, cmd, tconn->name);
 390         /* TODO: conn_bcast_event() ?? */
 391
 392         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 393         if (ret)
 394                 conn_warn(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
 395                           usermode_helper, cmd, tconn->name,
 396                           (ret >> 8) & 0xff, ret);
 397         else
 398                 conn_info(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
 399                           usermode_helper, cmd, tconn->name,
 400                           (ret >> 8) & 0xff, ret);
 401         /* TODO: conn_bcast_event() ?? */
 402
 403         if (ret < 0) /* Ignore any ERRNOs we got. */
 404                 ret = 0;
 405
 406         return ret;
 407 }
 408
 409 static enum drbd_fencing_p highest_fencing_policy(struct drbd_tconn *tconn)
 410 {
 411         enum drbd_fencing_p fp = FP_NOT_AVAIL;
 412         struct drbd_conf *mdev;
 413         int vnr;
 414
 415         rcu_read_lock();
 416         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
 417                 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 418                         fp = max_t(enum drbd_fencing_p, fp,
 419                                    rcu_dereference(mdev->ldev->disk_conf)->fencing);
 420                         put_ldev(mdev);
 421                 }
 422         }
 423         rcu_read_unlock();
 424
 425         return fp;
 426 }
 427
 428 bool conn_try_outdate_peer(struct drbd_tconn *tconn)
 429 {
 430         union drbd_state mask = { };
 431         union drbd_state val = { };
 432         enum drbd_fencing_p fp;
 433         char *ex_to_string;
 434         int r;
 435
 436         if (tconn->cstate >= C_WF_REPORT_PARAMS) {
 437                 conn_err(tconn, "Expected cstate < C_WF_REPORT_PARAMS\n");
 438                 return false;
 439         }
 440
 441         fp = highest_fencing_policy(tconn);
 442         switch (fp) {
 443         case FP_NOT_AVAIL:
 444                 conn_warn(tconn, "Not fencing peer, I'm not even Consistent myself.\n");
 445                 goto out;
 446         case FP_DONT_CARE:
 447                 return true;
 448         default: ;
 449         }
 450
 451         r = conn_khelper(tconn, "fence-peer");
 452
 453         switch ((r>>8) & 0xff) {
 454         case 3: /* peer is inconsistent */
 455                 ex_to_string = "peer is inconsistent or worse";
 456                 mask.pdsk = D_MASK;
 457                 val.pdsk = D_INCONSISTENT;
 458                 break;
 459         case 4: /* peer got outdated, or was already outdated */
 460                 ex_to_string = "peer was fenced";
 461                 mask.pdsk = D_MASK;
 462                 val.pdsk = D_OUTDATED;
 463                 break;
 464         case 5: /* peer was down */
 465                 if (conn_highest_disk(tconn) == D_UP_TO_DATE) {
 466                         /* we will(have) create(d) a new UUID anyways... */
 467                         ex_to_string = "peer is unreachable, assumed to be dead";
 468                         mask.pdsk = D_MASK;
 469                         val.pdsk = D_OUTDATED;
 470                 } else {
 471                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 472                 }
 473                 break;
 474         case 6: /* Peer is primary, voluntarily outdate myself.
 475                  * This is useful when an unconnected R_SECONDARY is asked to
 476                  * become R_PRIMARY, but finds the other peer being active. */
 477                 ex_to_string = "peer is active";
 478                 conn_warn(tconn, "Peer is primary, outdating myself.\n");
 479                 mask.disk = D_MASK;
 480                 val.disk = D_OUTDATED;
 481                 break;
 482         case 7:
 483                 if (fp != FP_STONITH)
 484                         conn_err(tconn, "fence-peer() = 7 && fencing != Stonith !!!\n");
 485                 ex_to_string = "peer was stonithed";
 486                 mask.pdsk = D_MASK;
 487                 val.pdsk = D_OUTDATED;
 488                 break;
 489         default:
 490                 /* The script is broken ... */
 491                 conn_err(tconn, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 492                 return false; /* Eventually leave IO frozen */
 493         }
 494
 495         conn_info(tconn, "fence-peer helper returned %d (%s)\n",
 496                   (r>>8) & 0xff, ex_to_string);
 497
 498  out:
 499
 500         /* Not using
 501            conn_request_state(tconn, mask, val, CS_VERBOSE);
 502            here, because we might were able to re-establish the connection in the
 503            meantime. */
 504         spin_lock_irq(&tconn->req_lock);
 505         if (tconn->cstate < C_WF_REPORT_PARAMS)
 506                 _conn_request_state(tconn, mask, val, CS_VERBOSE);
 507         spin_unlock_irq(&tconn->req_lock);
 508
 509         return conn_highest_pdsk(tconn) <= D_OUTDATED;
 510 }
 511
 512 static int _try_outdate_peer_async(void *data)
 513 {
 514         struct drbd_tconn *tconn = (struct drbd_tconn *)data;
 515
 516         conn_try_outdate_peer(tconn);
 517
 518         kref_put(&tconn->kref, &conn_destroy);
 519         return 0;
 520 }
 521
 522 void conn_try_outdate_peer_async(struct drbd_tconn *tconn)
 523 {
 524         struct task_struct *opa;
 525
 526         kref_get(&tconn->kref);
 527         opa = kthread_run(_try_outdate_peer_async, tconn, "drbd_async_h");
 528         if (IS_ERR(opa)) {
 529                 conn_err(tconn, "out of mem, failed to invoke fence-peer helper\n");
 530                 kref_put(&tconn->kref, &conn_destroy);
 531         }
 532 }
 533
 534 enum drbd_state_rv
 535 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 536 {
 537         const int max_tries = 4;
 538         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
 539         struct net_conf *nc;
 540         int try = 0;
 541         int forced = 0;
 542         union drbd_state mask, val;
 543
 544         if (new_role == R_PRIMARY)
 545                 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
 546
 547         mutex_lock(mdev->state_mutex);
 548
 549         mask.i = 0; mask.role = R_MASK;
 550         val.i  = 0; val.role  = new_role;
 551
 552         while (try++ < max_tries) {
 553                 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 554
 555                 /* in case we first succeeded to outdate,
 556                  * but now suddenly could establish a connection */
 557                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 558                         val.pdsk = 0;
 559                         mask.pdsk = 0;
 560                         continue;
 561                 }
 562
 563                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
 564                     (mdev->state.disk < D_UP_TO_DATE &&
 565                      mdev->state.disk >= D_INCONSISTENT)) {
 566                         mask.disk = D_MASK;
 567                         val.disk  = D_UP_TO_DATE;
 568                         forced = 1;
 569                         continue;
 570                 }
 571
 572                 if (rv == SS_NO_UP_TO_DATE_DISK &&
 573                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 574                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 575
 576                         if (conn_try_outdate_peer(mdev->tconn)) {
 577                                 val.disk = D_UP_TO_DATE;
 578                                 mask.disk = D_MASK;
 579                         }
 580                         continue;
 581                 }
 582
 583                 if (rv == SS_NOTHING_TO_DO)
 584                         goto out;
 585                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
 586                         if (!conn_try_outdate_peer(mdev->tconn) && force) {
 587                                 dev_warn(DEV, "Forced into split brain situation!\n");
 588                                 mask.pdsk = D_MASK;
 589                                 val.pdsk  = D_OUTDATED;
 590
 591                         }
 592                         continue;
 593                 }
 594                 if (rv == SS_TWO_PRIMARIES) {
 595                         /* Maybe the peer is detected as dead very soon...
 596                            retry at most once more in this case. */
 597                         int timeo;
 598                         rcu_read_lock();
 599                         nc = rcu_dereference(mdev->tconn->net_conf);
 600                         timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
 601                         rcu_read_unlock();
 602                         schedule_timeout_interruptible(timeo);
 603                         if (try < max_tries)
 604                                 try = max_tries - 1;
 605                         continue;
 606                 }
 607                 if (rv < SS_SUCCESS) {
 608                         rv = _drbd_request_state(mdev, mask, val,
 609                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 610                         if (rv < SS_SUCCESS)
 611                                 goto out;
 612                 }
 613                 break;
 614         }
 615
 616         if (rv < SS_SUCCESS)
 617                 goto out;
 618
 619         if (forced)
 620                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 621
 622         /* Wait until nothing is on the fly :) */
 623         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 624
 625         if (new_role == R_SECONDARY) {
 626                 set_disk_ro(mdev->vdisk, true);
 627                 if (get_ldev(mdev)) {
 628                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 629                         put_ldev(mdev);
 630                 }
 631         } else {
 632                 mutex_lock(&mdev->tconn->conf_update);
 633                 nc = mdev->tconn->net_conf;
 634                 if (nc)
 635                         nc->discard_my_data = 0; /* without copy; single bit op is atomic */
 636                 mutex_unlock(&mdev->tconn->conf_update);
 637
 638                 set_disk_ro(mdev->vdisk, false);
 639                 if (get_ldev(mdev)) {
 640                         if (((mdev->state.conn < C_CONNECTED ||
 641                                mdev->state.pdsk <= D_FAILED)
 642                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 643                                 drbd_uuid_new_current(mdev);
 644
 645                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 646                         put_ldev(mdev);
 647                 }
 648         }
 649
 650         /* writeout of activity log covered areas of the bitmap
 651          * to stable storage done in after state change already */
 652
 653         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 654                 /* if this was forced, we should consider sync */
 655                 if (forced)
 656                         drbd_send_uuids(mdev);
 657                 drbd_send_state(mdev);
 658         }
 659
 660         drbd_md_sync(mdev);
 661
 662         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 663 out:
 664         mutex_unlock(mdev->state_mutex);
 665         return rv;
 666 }
 667
 668 static const char *from_attrs_err_to_txt(int err)
 669 {
 670         return  err == -ENOMSG ? "required attribute missing" :
 671                 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
 672                 err == -EEXIST ? "can not change invariant setting" :
 673                 "invalid attribute value";
 674 }
 675
 676 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
 677 {
 678         struct set_role_parms parms;
 679         int err;
 680         enum drbd_ret_code retcode;
 681
 682         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
 683         if (!adm_ctx.reply_skb)
 684                 return retcode;
 685         if (retcode != NO_ERROR)
 686                 goto out;
 687
 688         memset(&parms, 0, sizeof(parms));
 689         if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
 690                 err = set_role_parms_from_attrs(&parms, info);
 691                 if (err) {
 692                         retcode = ERR_MANDATORY_TAG;
 693                         drbd_msg_put_info(from_attrs_err_to_txt(err));
 694                         goto out;
 695                 }
 696         }
 697
 698         if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
 699                 retcode = drbd_set_role(adm_ctx.mdev, R_PRIMARY, parms.assume_uptodate);
 700         else
 701                 retcode = drbd_set_role(adm_ctx.mdev, R_SECONDARY, 0);
 702 out:
 703         drbd_adm_finish(info, retcode);
 704         return 0;
 705 }
 706
 707 /* initializes the md.*_offset members, so we are able to find
 708  * the on disk meta data */
 709 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 710                                        struct drbd_backing_dev *bdev)
 711 {
 712         sector_t md_size_sect = 0;
 713         int meta_dev_idx;
 714
 715         rcu_read_lock();
 716         meta_dev_idx = rcu_dereference(bdev->disk_conf)->meta_dev_idx;
 717
 718         switch (meta_dev_idx) {
 719         default:
 720                 /* v07 style fixed size indexed meta data */
 721                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 722                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 723                 bdev->md.al_offset = MD_AL_OFFSET;
 724                 bdev->md.bm_offset = MD_BM_OFFSET;
 725                 break;
 726         case DRBD_MD_INDEX_FLEX_EXT:
 727                 /* just occupy the full device; unit: sectors */
 728                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 729                 bdev->md.md_offset = 0;
 730                 bdev->md.al_offset = MD_AL_OFFSET;
 731                 bdev->md.bm_offset = MD_BM_OFFSET;
 732                 break;
 733         case DRBD_MD_INDEX_INTERNAL:
 734         case DRBD_MD_INDEX_FLEX_INT:
 735                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 736                 /* al size is still fixed */
 737                 bdev->md.al_offset = -MD_AL_SECTORS;
 738                 /* we need (slightly less than) ~ this much bitmap sectors: */
 739                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 740                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 741                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 742                 md_size_sect = ALIGN(md_size_sect, 8);
 743
 744                 /* plus the "drbd meta data super block",
 745                  * and the activity log; */
 746                 md_size_sect += MD_BM_OFFSET;
 747
 748                 bdev->md.md_size_sect = md_size_sect;
 749                 /* bitmap offset is adjusted by 'super' block size */
 750                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 751                 break;
 752         }
 753         rcu_read_unlock();
 754 }
 755
 756 /* input size is expected to be in KB */
 757 char *ppsize(char *buf, unsigned long long size)
 758 {
 759         /* Needs 9 bytes at max including trailing NUL:
 760          * -1ULL ==> "16384 EB" */
 761         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 762         int base = 0;
 763         while (size >= 10000 && base < sizeof(units)-1) {
 764                 /* shift + round */
 765                 size = (size >> 10) + !!(size & (1<<9));
 766                 base++;
 767         }
 768         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
 769
 770         return buf;
 771 }
 772
 773 /* there is still a theoretical deadlock when called from receiver
 774  * on an D_INCONSISTENT R_PRIMARY:
 775  *  remote READ does inc_ap_bio, receiver would need to receive answer
 776  *  packet from remote to dec_ap_bio again.
 777  *  receiver receive_sizes(), comes here,
 778  *  waits for ap_bio_cnt == 0. -> deadlock.
 779  * but this cannot happen, actually, because:
 780  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 781  *  (not connected, or bad/no disk on peer):
 782  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 783  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 784  *  peer may not initiate a resize.
 785  */
 786 /* Note these are not to be confused with
 787  * drbd_adm_suspend_io/drbd_adm_resume_io,
 788  * which are (sub) state changes triggered by admin (drbdsetup),
 789  * and can be long lived.
 790  * This changes an mdev->flag, is triggered by drbd internals,
 791  * and should be short-lived. */
 792 void drbd_suspend_io(struct drbd_conf *mdev)
 793 {
 794         set_bit(SUSPEND_IO, &mdev->flags);
 795         if (drbd_suspended(mdev))
 796                 return;
 797         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 798 }
 799
 800 void drbd_resume_io(struct drbd_conf *mdev)
 801 {
 802         clear_bit(SUSPEND_IO, &mdev->flags);
 803         wake_up(&mdev->misc_wait);
 804 }
 805
 806 /**
 807  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 808  * @mdev:       DRBD device.
 809  *
 810  * Returns 0 on success, negative return values indicate errors.
 811  * You should call drbd_md_sync() after calling this function.
 812  */
 813 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 814 {
 815         sector_t prev_first_sect, prev_size; /* previous meta location */
 816         sector_t la_size, u_size;
 817         sector_t size;
 818         char ppb[10];
 819
 820         int md_moved, la_size_changed;
 821         enum determine_dev_size rv = unchanged;
 822
 823         /* race:
 824          * application request passes inc_ap_bio,
 825          * but then cannot get an AL-reference.
 826          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 827          *
 828          * to avoid that:
 829          * Suspend IO right here.
 830          * still lock the act_log to not trigger ASSERTs there.
 831          */
 832         drbd_suspend_io(mdev);
 833
 834         /* no wait necessary anymore, actually we could assert that */
 835         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 836
 837         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 838         prev_size = mdev->ldev->md.md_size_sect;
 839         la_size = mdev->ldev->md.la_size_sect;
 840
 841         /* TODO: should only be some assert here, not (re)init... */
 842         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 843
 844         rcu_read_lock();
 845         u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
 846         rcu_read_unlock();
 847         size = drbd_new_dev_size(mdev, mdev->ldev, u_size, flags & DDSF_FORCED);
 848
 849         if (drbd_get_capacity(mdev->this_bdev) != size ||
 850             drbd_bm_capacity(mdev) != size) {
 851                 int err;
 852                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 853                 if (unlikely(err)) {
 854                         /* currently there is only one error: ENOMEM! */
 855                         size = drbd_bm_capacity(mdev)>>1;
 856                         if (size == 0) {
 857                                 dev_err(DEV, "OUT OF MEMORY! "
 858                                     "Could not allocate bitmap!\n");
 859                         } else {
 860                                 dev_err(DEV, "BM resizing failed. "
 861                                     "Leaving size unchanged at size = %lu KB\n",
 862                                     (unsigned long)size);
 863                         }
 864                         rv = dev_size_error;
 865                 }
 866                 /* racy, see comments above. */
 867                 drbd_set_my_capacity(mdev, size);
 868                 mdev->ldev->md.la_size_sect = size;
 869                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 870                      (unsigned long long)size>>1);
 871         }
 872         if (rv == dev_size_error)
 873                 goto out;
 874
 875         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 876
 877         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 878                 || prev_size       != mdev->ldev->md.md_size_sect;
 879
 880         if (la_size_changed || md_moved) {
 881                 int err;
 882
 883                 drbd_al_shrink(mdev); /* All extents inactive. */
 884                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 885                          la_size_changed && md_moved ? "size changed and md moved" :
 886                          la_size_changed ? "size changed" : "md moved");
 887                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
 888                 err = drbd_bitmap_io(mdev, &drbd_bm_write,
 889                                 "size changed", BM_LOCKED_MASK);
 890                 if (err) {
 891                         rv = dev_size_error;
 892                         goto out;
 893                 }
 894                 drbd_md_mark_dirty(mdev);
 895         }
 896
 897         if (size > la_size)
 898                 rv = grew;
 899         if (size < la_size)
 900                 rv = shrunk;
 901 out:
 902         lc_unlock(mdev->act_log);
 903         wake_up(&mdev->al_wait);
 904         drbd_resume_io(mdev);
 905
 906         return rv;
 907 }
 908
 909 sector_t
 910 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
 911                   sector_t u_size, int assume_peer_has_space)
 912 {
 913         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 914         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 915         sector_t m_size; /* my size */
 916         sector_t size = 0;
 917
 918         m_size = drbd_get_max_capacity(bdev);
 919
 920         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 921                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 922                 p_size = m_size;
 923         }
 924
 925         if (p_size && m_size) {
 926                 size = min_t(sector_t, p_size, m_size);
 927         } else {
 928                 if (la_size) {
 929                         size = la_size;
 930                         if (m_size && m_size < size)
 931                                 size = m_size;
 932                         if (p_size && p_size < size)
 933                                 size = p_size;
 934                 } else {
 935                         if (m_size)
 936                                 size = m_size;
 937                         if (p_size)
 938                                 size = p_size;
 939                 }
 940         }
 941
 942         if (size == 0)
 943                 dev_err(DEV, "Both nodes diskless!\n");
 944
 945         if (u_size) {
 946                 if (u_size > size)
 947                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 948                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 949                 else
 950                         size = u_size;
 951         }
 952
 953         return size;
 954 }
 955
 956 /**
 957  * drbd_check_al_size() - Ensures that the AL is of the right size
 958  * @mdev:       DRBD device.
 959  *
 960  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 961  * failed, and 0 on success. You should call drbd_md_sync() after you called
 962  * this function.
 963  */
 964 static int drbd_check_al_size(struct drbd_conf *mdev, struct disk_conf *dc)
 965 {
 966         struct lru_cache *n, *t;
 967         struct lc_element *e;
 968         unsigned int in_use;
 969         int i;
 970
 971         if (mdev->act_log &&
 972             mdev->act_log->nr_elements == dc->al_extents)
 973                 return 0;
 974
 975         in_use = 0;
 976         t = mdev->act_log;
 977         n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
 978                 dc->al_extents, sizeof(struct lc_element), 0);
 979
 980         if (n == NULL) {
 981                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 982                 return -ENOMEM;
 983         }
 984         spin_lock_irq(&mdev->al_lock);
 985         if (t) {
 986                 for (i = 0; i < t->nr_elements; i++) {
 987                         e = lc_element_by_index(t, i);
 988                         if (e->refcnt)
 989                                 dev_err(DEV, "refcnt(%d)==%d\n",
 990                                     e->lc_number, e->refcnt);
 991                         in_use += e->refcnt;
 992                 }
 993         }
 994         if (!in_use)
 995                 mdev->act_log = n;
 996         spin_unlock_irq(&mdev->al_lock);
 997         if (in_use) {
 998                 dev_err(DEV, "Activity log still in use!\n");
 999                 lc_destroy(n);
1000                 return -EBUSY;
1001         } else {
1002                 if (t)
1003                         lc_destroy(t);
1004         }
1005         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
1006         return 0;
1007 }
1008
1009 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
1010 {
1011         struct request_queue * const q = mdev->rq_queue;
1012         int max_hw_sectors = max_bio_size >> 9;
1013         int max_segments = 0;
1014
1015         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1016                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
1017
1018                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1019                 rcu_read_lock();
1020                 max_segments = rcu_dereference(mdev->ldev->disk_conf)->max_bio_bvecs;
1021                 rcu_read_unlock();
1022                 put_ldev(mdev);
1023         }
1024
1025         blk_queue_logical_block_size(q, 512);
1026         blk_queue_max_hw_sectors(q, max_hw_sectors);
1027         /* This is the workaround for "bio would need to, but cannot, be split" */
1028         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1029         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
1030
1031         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1032                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
1033
1034                 blk_queue_stack_limits(q, b);
1035
1036                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
1037                         dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1038                                  q->backing_dev_info.ra_pages,
1039                                  b->backing_dev_info.ra_pages);
1040                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
1041                 }
1042                 put_ldev(mdev);
1043         }
1044 }
1045
1046 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
1047 {
1048         int now, new, local, peer;
1049
1050         now = queue_max_hw_sectors(mdev->rq_queue) << 9;
1051         local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
1052         peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
1053
1054         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1055                 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
1056                 mdev->local_max_bio_size = local;
1057                 put_ldev(mdev);
1058         }
1059
1060         /* We may ignore peer limits if the peer is modern enough.
1061            Because new from 8.3.8 onwards the peer can use multiple
1062            BIOs for a single peer_request */
1063         if (mdev->state.conn >= C_CONNECTED) {
1064                 if (mdev->tconn->agreed_pro_version < 94)
1065                         peer = min_t(int, mdev->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1066                         /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1067                 else if (mdev->tconn->agreed_pro_version == 94)
1068                         peer = DRBD_MAX_SIZE_H80_PACKET;
1069                 else if (mdev->tconn->agreed_pro_version < 100)
1070                         peer = DRBD_MAX_BIO_SIZE_P95;  /* drbd 8.3.8 onwards, before 8.4.0 */
1071                 else
1072                         peer = DRBD_MAX_BIO_SIZE;
1073         }
1074
1075         new = min_t(int, local, peer);
1076
1077         if (mdev->state.role == R_PRIMARY && new < now)
1078                 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
1079
1080         if (new != now)
1081                 dev_info(DEV, "max BIO size = %u\n", new);
1082
1083         drbd_setup_queue_param(mdev, new);
1084 }
1085
1086 /* Starts the worker thread */
1087 static void conn_reconfig_start(struct drbd_tconn *tconn)
1088 {
1089         drbd_thread_start(&tconn->worker);
1090         conn_flush_workqueue(tconn);
1091 }
1092
1093 /* if still unconfigured, stops worker again. */
1094 static void conn_reconfig_done(struct drbd_tconn *tconn)
1095 {
1096         bool stop_threads;
1097         spin_lock_irq(&tconn->req_lock);
1098         stop_threads = conn_all_vols_unconf(tconn);
1099         spin_unlock_irq(&tconn->req_lock);
1100         if (stop_threads) {
1101                 /* asender is implicitly stopped by receiver
1102                  * in conn_disconnect() */
1103                 drbd_thread_stop(&tconn->receiver);
1104                 drbd_thread_stop(&tconn->worker);
1105         }
1106 }
1107
1108 /* Make sure IO is suspended before calling this function(). */
1109 static void drbd_suspend_al(struct drbd_conf *mdev)
1110 {
1111         int s = 0;
1112
1113         if (!lc_try_lock(mdev->act_log)) {
1114                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
1115                 return;
1116         }
1117
1118         drbd_al_shrink(mdev);
1119         spin_lock_irq(&mdev->tconn->req_lock);
1120         if (mdev->state.conn < C_CONNECTED)
1121                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
1122         spin_unlock_irq(&mdev->tconn->req_lock);
1123         lc_unlock(mdev->act_log);
1124
1125         if (s)
1126                 dev_info(DEV, "Suspended AL updates\n");
1127 }
1128
1129
1130 static bool should_set_defaults(struct genl_info *info)
1131 {
1132         unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1133         return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1134 }
1135
1136 static void enforce_disk_conf_limits(struct disk_conf *dc)
1137 {
1138         if (dc->al_extents < DRBD_AL_EXTENTS_MIN)
1139                 dc->al_extents = DRBD_AL_EXTENTS_MIN;
1140         if (dc->al_extents > DRBD_AL_EXTENTS_MAX)
1141                 dc->al_extents = DRBD_AL_EXTENTS_MAX;
1142
1143         if (dc->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1144                 dc->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1145 }
1146
1147 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1148 {
1149         enum drbd_ret_code retcode;
1150         struct drbd_conf *mdev;
1151         struct disk_conf *new_disk_conf, *old_disk_conf;
1152         struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1153         int err, fifo_size;
1154
1155         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1156         if (!adm_ctx.reply_skb)
1157                 return retcode;
1158         if (retcode != NO_ERROR)
1159                 goto out;
1160
1161         mdev = adm_ctx.mdev;
1162
1163         /* we also need a disk
1164          * to change the options on */
1165         if (!get_ldev(mdev)) {
1166                 retcode = ERR_NO_DISK;
1167                 goto out;
1168         }
1169
1170         new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1171         if (!new_disk_conf) {
1172                 retcode = ERR_NOMEM;
1173                 goto fail;
1174         }
1175
1176         mutex_lock(&mdev->tconn->conf_update);
1177         old_disk_conf = mdev->ldev->disk_conf;
1178         *new_disk_conf = *old_disk_conf;
1179         if (should_set_defaults(info))
1180                 set_disk_conf_defaults(new_disk_conf);
1181
1182         err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1183         if (err && err != -ENOMSG) {
1184                 retcode = ERR_MANDATORY_TAG;
1185                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1186         }
1187
1188         if (!expect(new_disk_conf->resync_rate >= 1))
1189                 new_disk_conf->resync_rate = 1;
1190
1191         enforce_disk_conf_limits(new_disk_conf);
1192
1193         fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1194         if (fifo_size != mdev->rs_plan_s->size) {
1195                 new_plan = fifo_alloc(fifo_size);
1196                 if (!new_plan) {
1197                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1198                         retcode = ERR_NOMEM;
1199                         goto fail_unlock;
1200                 }
1201         }
1202
1203         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1204         drbd_al_shrink(mdev);
1205         err = drbd_check_al_size(mdev, new_disk_conf);
1206         lc_unlock(mdev->act_log);
1207         wake_up(&mdev->al_wait);
1208
1209         if (err) {
1210                 retcode = ERR_NOMEM;
1211                 goto fail_unlock;
1212         }
1213
1214         write_lock_irq(&global_state_lock);
1215         retcode = drbd_resync_after_valid(mdev, new_disk_conf->resync_after);
1216         if (retcode == NO_ERROR) {
1217                 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
1218                 drbd_resync_after_changed(mdev);
1219         }
1220         write_unlock_irq(&global_state_lock);
1221
1222         if (retcode != NO_ERROR)
1223                 goto fail_unlock;
1224
1225         if (new_plan) {
1226                 old_plan = mdev->rs_plan_s;
1227                 rcu_assign_pointer(mdev->rs_plan_s, new_plan);
1228         }
1229
1230         mutex_unlock(&mdev->tconn->conf_update);
1231         drbd_md_sync(mdev);
1232
1233         if (mdev->state.conn >= C_CONNECTED)
1234                 drbd_send_sync_param(mdev);
1235
1236         synchronize_rcu();
1237         kfree(old_disk_conf);
1238         kfree(old_plan);
1239         goto success;
1240
1241 fail_unlock:
1242         mutex_unlock(&mdev->tconn->conf_update);
1243  fail:
1244         kfree(new_disk_conf);
1245         kfree(new_plan);
1246 success:
1247         put_ldev(mdev);
1248  out:
1249         drbd_adm_finish(info, retcode);
1250         return 0;
1251 }
1252
1253 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1254 {
1255         struct drbd_conf *mdev;
1256         int err;
1257         enum drbd_ret_code retcode;
1258         enum determine_dev_size dd;
1259         sector_t max_possible_sectors;
1260         sector_t min_md_device_sectors;
1261         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1262         struct disk_conf *new_disk_conf = NULL;
1263         struct block_device *bdev;
1264         struct lru_cache *resync_lru = NULL;
1265         struct fifo_buffer *new_plan = NULL;
1266         union drbd_state ns, os;
1267         enum drbd_state_rv rv;
1268         struct net_conf *nc;
1269         int cp_discovered = 0;
1270
1271         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1272         if (!adm_ctx.reply_skb)
1273                 return retcode;
1274         if (retcode != NO_ERROR)
1275                 goto finish;
1276
1277         mdev = adm_ctx.mdev;
1278         conn_reconfig_start(mdev->tconn);
1279
1280         /* if you want to reconfigure, please tear down first */
1281         if (mdev->state.disk > D_DISKLESS) {
1282                 retcode = ERR_DISK_CONFIGURED;
1283                 goto fail;
1284         }
1285         /* It may just now have detached because of IO error.  Make sure
1286          * drbd_ldev_destroy is done already, we may end up here very fast,
1287          * e.g. if someone calls attach from the on-io-error handler,
1288          * to realize a "hot spare" feature (not that I'd recommend that) */
1289         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1290
1291         /* allocation not in the IO path, drbdsetup context */
1292         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1293         if (!nbc) {
1294                 retcode = ERR_NOMEM;
1295                 goto fail;
1296         }
1297         new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1298         if (!new_disk_conf) {
1299                 retcode = ERR_NOMEM;
1300                 goto fail;
1301         }
1302         nbc->disk_conf = new_disk_conf;
1303
1304         set_disk_conf_defaults(new_disk_conf);
1305         err = disk_conf_from_attrs(new_disk_conf, info);
1306         if (err) {
1307                 retcode = ERR_MANDATORY_TAG;
1308                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1309                 goto fail;
1310         }
1311
1312         enforce_disk_conf_limits(new_disk_conf);
1313
1314         new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1315         if (!new_plan) {
1316                 retcode = ERR_NOMEM;
1317                 goto fail;
1318         }
1319
1320         if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1321                 retcode = ERR_MD_IDX_INVALID;
1322                 goto fail;
1323         }
1324
1325         rcu_read_lock();
1326         nc = rcu_dereference(mdev->tconn->net_conf);
1327         if (nc) {
1328                 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1329                         rcu_read_unlock();
1330                         retcode = ERR_STONITH_AND_PROT_A;
1331                         goto fail;
1332                 }
1333         }
1334         rcu_read_unlock();
1335
1336         bdev = blkdev_get_by_path(new_disk_conf->backing_dev,
1337                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1338         if (IS_ERR(bdev)) {
1339                 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->backing_dev,
1340                         PTR_ERR(bdev));
1341                 retcode = ERR_OPEN_DISK;
1342                 goto fail;
1343         }
1344         nbc->backing_bdev = bdev;
1345
1346         /*
1347          * meta_dev_idx >= 0: external fixed size, possibly multiple
1348          * drbd sharing one meta device.  TODO in that case, paranoia
1349          * check that [md_bdev, meta_dev_idx] is not yet used by some
1350          * other drbd minor!  (if you use drbd.conf + drbdadm, that
1351          * should check it for you already; but if you don't, or
1352          * someone fooled it, we need to double check here)
1353          */
1354         bdev = blkdev_get_by_path(new_disk_conf->meta_dev,
1355                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1356                                   (new_disk_conf->meta_dev_idx < 0) ?
1357                                   (void *)mdev : (void *)drbd_m_holder);
1358         if (IS_ERR(bdev)) {
1359                 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->meta_dev,
1360                         PTR_ERR(bdev));
1361                 retcode = ERR_OPEN_MD_DISK;
1362                 goto fail;
1363         }
1364         nbc->md_bdev = bdev;
1365
1366         if ((nbc->backing_bdev == nbc->md_bdev) !=
1367             (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1368              new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1369                 retcode = ERR_MD_IDX_INVALID;
1370                 goto fail;
1371         }
1372
1373         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1374                         1, 61, sizeof(struct bm_extent),
1375                         offsetof(struct bm_extent, lce));
1376         if (!resync_lru) {
1377                 retcode = ERR_NOMEM;
1378                 goto fail;
1379         }
1380
1381         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1382         drbd_md_set_sector_offsets(mdev, nbc);
1383
1384         if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1385                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1386                         (unsigned long long) drbd_get_max_capacity(nbc),
1387                         (unsigned long long) new_disk_conf->disk_size);
1388                 retcode = ERR_DISK_TOO_SMALL;
1389                 goto fail;
1390         }
1391
1392         if (new_disk_conf->meta_dev_idx < 0) {
1393                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1394                 /* at least one MB, otherwise it does not make sense */
1395                 min_md_device_sectors = (2<<10);
1396         } else {
1397                 max_possible_sectors = DRBD_MAX_SECTORS;
1398                 min_md_device_sectors = MD_RESERVED_SECT * (new_disk_conf->meta_dev_idx + 1);
1399         }
1400
1401         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1402                 retcode = ERR_MD_DISK_TOO_SMALL;
1403                 dev_warn(DEV, "refusing attach: md-device too small, "
1404                      "at least %llu sectors needed for this meta-disk type\n",
1405                      (unsigned long long) min_md_device_sectors);
1406                 goto fail;
1407         }
1408
1409         /* Make sure the new disk is big enough
1410          * (we may currently be R_PRIMARY with no local disk...) */
1411         if (drbd_get_max_capacity(nbc) <
1412             drbd_get_capacity(mdev->this_bdev)) {
1413                 retcode = ERR_DISK_TOO_SMALL;
1414                 goto fail;
1415         }
1416
1417         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1418
1419         if (nbc->known_size > max_possible_sectors) {
1420                 dev_warn(DEV, "==> truncating very big lower level device "
1421                         "to currently maximum possible %llu sectors <==\n",
1422                         (unsigned long long) max_possible_sectors);
1423                 if (new_disk_conf->meta_dev_idx >= 0)
1424                         dev_warn(DEV, "==>> using internal or flexible "
1425                                       "meta data may help <<==\n");
1426         }
1427
1428         drbd_suspend_io(mdev);
1429         /* also wait for the last barrier ack. */
1430         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || drbd_suspended(mdev));
1431         /* and for any other previously queued work */
1432         drbd_flush_workqueue(mdev);
1433
1434         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1435         retcode = rv;  /* FIXME: Type mismatch. */
1436         drbd_resume_io(mdev);
1437         if (rv < SS_SUCCESS)
1438                 goto fail;
1439
1440         if (!get_ldev_if_state(mdev, D_ATTACHING))
1441                 goto force_diskless;
1442
1443         drbd_md_set_sector_offsets(mdev, nbc);
1444
1445         if (!mdev->bitmap) {
1446                 if (drbd_bm_init(mdev)) {
1447                         retcode = ERR_NOMEM;
1448                         goto force_diskless_dec;
1449                 }
1450         }
1451
1452         retcode = drbd_md_read(mdev, nbc);
1453         if (retcode != NO_ERROR)
1454                 goto force_diskless_dec;
1455
1456         if (mdev->state.conn < C_CONNECTED &&
1457             mdev->state.role == R_PRIMARY &&
1458             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1459                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1460                     (unsigned long long)mdev->ed_uuid);
1461                 retcode = ERR_DATA_NOT_CURRENT;
1462                 goto force_diskless_dec;
1463         }
1464
1465         /* Since we are diskless, fix the activity log first... */
1466         if (drbd_check_al_size(mdev, new_disk_conf)) {
1467                 retcode = ERR_NOMEM;
1468                 goto force_diskless_dec;
1469         }
1470
1471         /* Prevent shrinking of consistent devices ! */
1472         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1473             drbd_new_dev_size(mdev, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1474                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1475                 retcode = ERR_DISK_TOO_SMALL;
1476                 goto force_diskless_dec;
1477         }
1478
1479         if (!drbd_al_read_log(mdev, nbc)) {
1480                 retcode = ERR_IO_MD_DISK;
1481                 goto force_diskless_dec;
1482         }
1483
1484         /* Reset the "barriers don't work" bits here, then force meta data to
1485          * be written, to ensure we determine if barriers are supported. */
1486         if (new_disk_conf->md_flushes)
1487                 clear_bit(MD_NO_FUA, &mdev->flags);
1488         else
1489                 set_bit(MD_NO_FUA, &mdev->flags);
1490
1491         /* Point of no return reached.
1492          * Devices and memory are no longer released by error cleanup below.
1493          * now mdev takes over responsibility, and the state engine should
1494          * clean it up somewhere.  */
1495         D_ASSERT(mdev->ldev == NULL);
1496         mdev->ldev = nbc;
1497         mdev->resync = resync_lru;
1498         mdev->rs_plan_s = new_plan;
1499         nbc = NULL;
1500         resync_lru = NULL;
1501         new_disk_conf = NULL;
1502         new_plan = NULL;
1503
1504         mdev->write_ordering = WO_bdev_flush;
1505         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1506
1507         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1508                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1509         else
1510                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1511
1512         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1513             !(mdev->state.role == R_PRIMARY && mdev->tconn->susp_nod)) {
1514                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1515                 cp_discovered = 1;
1516         }
1517
1518         mdev->send_cnt = 0;
1519         mdev->recv_cnt = 0;
1520         mdev->read_cnt = 0;
1521         mdev->writ_cnt = 0;
1522
1523         drbd_reconsider_max_bio_size(mdev);
1524
1525         /* If I am currently not R_PRIMARY,
1526          * but meta data primary indicator is set,
1527          * I just now recover from a hard crash,
1528          * and have been R_PRIMARY before that crash.
1529          *
1530          * Now, if I had no connection before that crash
1531          * (have been degraded R_PRIMARY), chances are that
1532          * I won't find my peer now either.
1533          *
1534          * In that case, and _only_ in that case,
1535          * we use the degr-wfc-timeout instead of the default,
1536          * so we can automatically recover from a crash of a
1537          * degraded but active "cluster" after a certain timeout.
1538          */
1539         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1540         if (mdev->state.role != R_PRIMARY &&
1541              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1542             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1543                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1544
1545         dd = drbd_determine_dev_size(mdev, 0);
1546         if (dd == dev_size_error) {
1547                 retcode = ERR_NOMEM_BITMAP;
1548                 goto force_diskless_dec;
1549         } else if (dd == grew)
1550                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1551
1552         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1553                 dev_info(DEV, "Assuming that all blocks are out of sync "
1554                      "(aka FullSync)\n");
1555                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1556                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1557                         retcode = ERR_IO_MD_DISK;
1558                         goto force_diskless_dec;
1559                 }
1560         } else {
1561                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1562                         "read from attaching", BM_LOCKED_MASK)) {
1563                         retcode = ERR_IO_MD_DISK;
1564                         goto force_diskless_dec;
1565                 }
1566         }
1567
1568         if (cp_discovered) {
1569                 drbd_al_apply_to_bm(mdev);
1570                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1571                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1572                         retcode = ERR_IO_MD_DISK;
1573                         goto force_diskless_dec;
1574                 }
1575         }
1576
1577         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1578                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1579
1580         spin_lock_irq(&mdev->tconn->req_lock);
1581         os = drbd_read_state(mdev);
1582         ns = os;
1583         /* If MDF_CONSISTENT is not set go into inconsistent state,
1584            otherwise investigate MDF_WasUpToDate...
1585            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1586            otherwise into D_CONSISTENT state.
1587         */
1588         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1589                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1590                         ns.disk = D_CONSISTENT;
1591                 else
1592                         ns.disk = D_OUTDATED;
1593         } else {
1594                 ns.disk = D_INCONSISTENT;
1595         }
1596
1597         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1598                 ns.pdsk = D_OUTDATED;
1599
1600         rcu_read_lock();
1601         if (ns.disk == D_CONSISTENT &&
1602             (ns.pdsk == D_OUTDATED || rcu_dereference(mdev->ldev->disk_conf)->fencing == FP_DONT_CARE))
1603                 ns.disk = D_UP_TO_DATE;
1604         rcu_read_unlock();
1605
1606         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1607            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1608            this point, because drbd_request_state() modifies these
1609            flags. */
1610
1611         /* In case we are C_CONNECTED postpone any decision on the new disk
1612            state after the negotiation phase. */
1613         if (mdev->state.conn == C_CONNECTED) {
1614                 mdev->new_state_tmp.i = ns.i;
1615                 ns.i = os.i;
1616                 ns.disk = D_NEGOTIATING;
1617
1618                 /* We expect to receive up-to-date UUIDs soon.
1619                    To avoid a race in receive_state, free p_uuid while
1620                    holding req_lock. I.e. atomic with the state change */
1621                 kfree(mdev->p_uuid);
1622                 mdev->p_uuid = NULL;
1623         }
1624
1625         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1626         spin_unlock_irq(&mdev->tconn->req_lock);
1627
1628         if (rv < SS_SUCCESS)
1629                 goto force_diskless_dec;
1630
1631         if (mdev->state.role == R_PRIMARY)
1632                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1633         else
1634                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1635
1636         drbd_md_mark_dirty(mdev);
1637         drbd_md_sync(mdev);
1638
1639         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1640         put_ldev(mdev);
1641         conn_reconfig_done(mdev->tconn);
1642         drbd_adm_finish(info, retcode);
1643         return 0;
1644
1645  force_diskless_dec:
1646         put_ldev(mdev);
1647  force_diskless:
1648         drbd_force_state(mdev, NS(disk, D_FAILED));
1649         drbd_md_sync(mdev);
1650  fail:
1651         conn_reconfig_done(mdev->tconn);
1652         if (nbc) {
1653                 if (nbc->backing_bdev)
1654                         blkdev_put(nbc->backing_bdev,
1655                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1656                 if (nbc->md_bdev)
1657                         blkdev_put(nbc->md_bdev,
1658                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1659                 kfree(nbc);
1660         }
1661         kfree(new_disk_conf);
1662         lc_destroy(resync_lru);
1663         kfree(new_plan);
1664
1665  finish:
1666         drbd_adm_finish(info, retcode);
1667         return 0;
1668 }
1669
1670 static int adm_detach(struct drbd_conf *mdev)
1671 {
1672         enum drbd_state_rv retcode;
1673         int ret;
1674         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1675         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1676         /* D_FAILED will transition to DISKLESS. */
1677         ret = wait_event_interruptible(mdev->misc_wait,
1678                         mdev->state.disk != D_FAILED);
1679         drbd_resume_io(mdev);
1680         if ((int)retcode == (int)SS_IS_DISKLESS)
1681                 retcode = SS_NOTHING_TO_DO;
1682         if (ret)
1683                 retcode = ERR_INTR;
1684         return retcode;
1685 }
1686
1687 /* Detaching the disk is a process in multiple stages.  First we need to lock
1688  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1689  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1690  * internal references as well.
1691  * Only then we have finally detached. */
1692 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1693 {
1694         enum drbd_ret_code retcode;
1695
1696         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1697         if (!adm_ctx.reply_skb)
1698                 return retcode;
1699         if (retcode != NO_ERROR)
1700                 goto out;
1701
1702         retcode = adm_detach(adm_ctx.mdev);
1703 out:
1704         drbd_adm_finish(info, retcode);
1705         return 0;
1706 }
1707
1708 static bool conn_resync_running(struct drbd_tconn *tconn)
1709 {
1710         struct drbd_conf *mdev;
1711         bool rv = false;
1712         int vnr;
1713
1714         rcu_read_lock();
1715         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1716                 if (mdev->state.conn == C_SYNC_SOURCE ||
1717                     mdev->state.conn == C_SYNC_TARGET ||
1718                     mdev->state.conn == C_PAUSED_SYNC_S ||
1719                     mdev->state.conn == C_PAUSED_SYNC_T) {
1720                         rv = true;
1721                         break;
1722                 }
1723         }
1724         rcu_read_unlock();
1725
1726         return rv;
1727 }
1728
1729 static bool conn_ov_running(struct drbd_tconn *tconn)
1730 {
1731         struct drbd_conf *mdev;
1732         bool rv = false;
1733         int vnr;
1734
1735         rcu_read_lock();
1736         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1737                 if (mdev->state.conn == C_VERIFY_S ||
1738                     mdev->state.conn == C_VERIFY_T) {
1739                         rv = true;
1740                         break;
1741                 }
1742         }
1743         rcu_read_unlock();
1744
1745         return rv;
1746 }
1747
1748 static enum drbd_ret_code
1749 _check_net_options(struct drbd_tconn *tconn, struct net_conf *old_conf, struct net_conf *new_conf)
1750 {
1751         struct drbd_conf *mdev;
1752         int i;
1753
1754         if (old_conf && tconn->cstate == C_WF_REPORT_PARAMS && tconn->agreed_pro_version < 100) {
1755                 if (new_conf->wire_protocol != old_conf->wire_protocol)
1756                         return ERR_NEED_APV_100;
1757
1758                 if (new_conf->two_primaries != old_conf->two_primaries)
1759                         return ERR_NEED_APV_100;
1760
1761                 if (!new_conf->integrity_alg != !old_conf->integrity_alg)
1762                         return ERR_NEED_APV_100;
1763
1764                 if (strcmp(new_conf->integrity_alg, old_conf->integrity_alg))
1765                         return ERR_NEED_APV_100;
1766         }
1767
1768         if (!new_conf->two_primaries &&
1769             conn_highest_role(tconn) == R_PRIMARY &&
1770             conn_highest_peer(tconn) == R_PRIMARY)
1771                 return ERR_NEED_ALLOW_TWO_PRI;
1772
1773         if (new_conf->two_primaries &&
1774             (new_conf->wire_protocol != DRBD_PROT_C))
1775                 return ERR_NOT_PROTO_C;
1776
1777         idr_for_each_entry(&tconn->volumes, mdev, i) {
1778                 if (get_ldev(mdev)) {
1779                         enum drbd_fencing_p fp = rcu_dereference(mdev->ldev->disk_conf)->fencing;
1780                         put_ldev(mdev);
1781                         if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
1782                                 return ERR_STONITH_AND_PROT_A;
1783                 }
1784                 if (mdev->state.role == R_PRIMARY && new_conf->discard_my_data)
1785                         return ERR_DISCARD;
1786         }
1787
1788         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A)
1789                 return ERR_CONG_NOT_PROTO_A;
1790
1791         return NO_ERROR;
1792 }
1793
1794 static enum drbd_ret_code
1795 check_net_options(struct drbd_tconn *tconn, struct net_conf *new_conf)
1796 {
1797         static enum drbd_ret_code rv;
1798         struct drbd_conf *mdev;
1799         int i;
1800
1801         rcu_read_lock();
1802         rv = _check_net_options(tconn, rcu_dereference(tconn->net_conf), new_conf);
1803         rcu_read_unlock();
1804
1805         /* tconn->volumes protected by genl_lock() here */
1806         idr_for_each_entry(&tconn->volumes, mdev, i) {
1807                 if (!mdev->bitmap) {
1808                         if(drbd_bm_init(mdev))
1809                                 return ERR_NOMEM;
1810                 }
1811         }
1812
1813         return rv;
1814 }
1815
1816 struct crypto {
1817         struct crypto_hash *verify_tfm;
1818         struct crypto_hash *csums_tfm;
1819         struct crypto_hash *cram_hmac_tfm;
1820         struct crypto_hash *integrity_tfm;
1821         void *int_dig_in;
1822         void *int_dig_vv;
1823 };
1824
1825 static int
1826 alloc_hash(struct crypto_hash **tfm, char *tfm_name, int err_alg)
1827 {
1828         if (!tfm_name[0])
1829                 return NO_ERROR;
1830
1831         *tfm = crypto_alloc_hash(tfm_name, 0, CRYPTO_ALG_ASYNC);
1832         if (IS_ERR(*tfm)) {
1833                 *tfm = NULL;
1834                 return err_alg;
1835         }
1836
1837         return NO_ERROR;
1838 }
1839
1840 static enum drbd_ret_code
1841 alloc_crypto(struct crypto *crypto, struct net_conf *new_conf)
1842 {
1843         char hmac_name[CRYPTO_MAX_ALG_NAME];
1844         enum drbd_ret_code rv;
1845         int hash_size;
1846
1847         rv = alloc_hash(&crypto->csums_tfm, new_conf->csums_alg,
1848                        ERR_CSUMS_ALG);
1849         if (rv != NO_ERROR)
1850                 return rv;
1851         rv = alloc_hash(&crypto->verify_tfm, new_conf->verify_alg,
1852                        ERR_VERIFY_ALG);
1853         if (rv != NO_ERROR)
1854                 return rv;
1855         rv = alloc_hash(&crypto->integrity_tfm, new_conf->integrity_alg,
1856                        ERR_INTEGRITY_ALG);
1857         if (rv != NO_ERROR)
1858                 return rv;
1859         if (new_conf->cram_hmac_alg[0] != 0) {
1860                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1861                          new_conf->cram_hmac_alg);
1862
1863                 rv = alloc_hash(&crypto->cram_hmac_tfm, hmac_name,
1864                                ERR_AUTH_ALG);
1865         }
1866         if (crypto->integrity_tfm) {
1867                 hash_size = crypto_hash_digestsize(crypto->integrity_tfm);
1868                 crypto->int_dig_in = kmalloc(hash_size, GFP_KERNEL);
1869                 if (!crypto->int_dig_in)
1870                         return ERR_NOMEM;
1871                 crypto->int_dig_vv = kmalloc(hash_size, GFP_KERNEL);
1872                 if (!crypto->int_dig_vv)
1873                         return ERR_NOMEM;
1874         }
1875
1876         return rv;
1877 }
1878
1879 static void free_crypto(struct crypto *crypto)
1880 {
1881         kfree(crypto->int_dig_in);
1882         kfree(crypto->int_dig_vv);
1883         crypto_free_hash(crypto->cram_hmac_tfm);
1884         crypto_free_hash(crypto->integrity_tfm);
1885         crypto_free_hash(crypto->csums_tfm);
1886         crypto_free_hash(crypto->verify_tfm);
1887 }
1888
1889 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
1890 {
1891         enum drbd_ret_code retcode;
1892         struct drbd_tconn *tconn;
1893         struct net_conf *old_conf, *new_conf = NULL;
1894         int err;
1895         int ovr; /* online verify running */
1896         int rsr; /* re-sync running */
1897         struct crypto crypto = { };
1898
1899         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONNECTION);
1900         if (!adm_ctx.reply_skb)
1901                 return retcode;
1902         if (retcode != NO_ERROR)
1903                 goto out;
1904
1905         tconn = adm_ctx.tconn;
1906
1907         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1908         if (!new_conf) {
1909                 retcode = ERR_NOMEM;
1910                 goto out;
1911         }
1912
1913         conn_reconfig_start(tconn);
1914
1915         mutex_lock(&tconn->data.mutex);
1916         mutex_lock(&tconn->conf_update);
1917         old_conf = tconn->net_conf;
1918
1919         if (!old_conf) {
1920                 drbd_msg_put_info("net conf missing, try connect");
1921                 retcode = ERR_INVALID_REQUEST;
1922                 goto fail;
1923         }
1924
1925         *new_conf = *old_conf;
1926         if (should_set_defaults(info))
1927                 set_net_conf_defaults(new_conf);
1928
1929         err = net_conf_from_attrs_for_change(new_conf, info);
1930         if (err && err != -ENOMSG) {
1931                 retcode = ERR_MANDATORY_TAG;
1932                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1933                 goto fail;
1934         }
1935
1936         retcode = check_net_options(tconn, new_conf);
1937         if (retcode != NO_ERROR)
1938                 goto fail;
1939
1940         /* re-sync running */
1941         rsr = conn_resync_running(tconn);
1942         if (rsr && strcmp(new_conf->csums_alg, old_conf->csums_alg)) {
1943                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1944                 goto fail;
1945         }
1946
1947         /* online verify running */
1948         ovr = conn_ov_running(tconn);
1949         if (ovr && strcmp(new_conf->verify_alg, old_conf->verify_alg)) {
1950                 retcode = ERR_VERIFY_RUNNING;
1951                 goto fail;
1952         }
1953
1954         retcode = alloc_crypto(&crypto, new_conf);
1955         if (retcode != NO_ERROR)
1956                 goto fail;
1957
1958         rcu_assign_pointer(tconn->net_conf, new_conf);
1959
1960         if (!rsr) {
1961                 crypto_free_hash(tconn->csums_tfm);
1962                 tconn->csums_tfm = crypto.csums_tfm;
1963                 crypto.csums_tfm = NULL;
1964         }
1965         if (!ovr) {
1966                 crypto_free_hash(tconn->verify_tfm);
1967                 tconn->verify_tfm = crypto.verify_tfm;
1968                 crypto.verify_tfm = NULL;
1969         }
1970
1971         kfree(tconn->int_dig_in);
1972         tconn->int_dig_in = crypto.int_dig_in;
1973         kfree(tconn->int_dig_vv);
1974         tconn->int_dig_vv = crypto.int_dig_vv;
1975         crypto_free_hash(tconn->integrity_tfm);
1976         tconn->integrity_tfm = crypto.integrity_tfm;
1977         if (tconn->cstate >= C_WF_REPORT_PARAMS && tconn->agreed_pro_version >= 100)
1978                 /* Do this without trying to take tconn->data.mutex again.  */
1979                 __drbd_send_protocol(tconn, P_PROTOCOL_UPDATE);
1980
1981         crypto_free_hash(tconn->cram_hmac_tfm);
1982         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
1983
1984         mutex_unlock(&tconn->conf_update);
1985         mutex_unlock(&tconn->data.mutex);
1986         synchronize_rcu();
1987         kfree(old_conf);
1988
1989         if (tconn->cstate >= C_WF_REPORT_PARAMS)
1990                 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn)));
1991
1992         goto done;
1993
1994  fail:
1995         mutex_unlock(&tconn->conf_update);
1996         mutex_unlock(&tconn->data.mutex);
1997         free_crypto(&crypto);
1998         kfree(new_conf);
1999  done:
2000         conn_reconfig_done(tconn);
2001  out:
2002         drbd_adm_finish(info, retcode);
2003         return 0;
2004 }
2005
2006 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2007 {
2008         struct drbd_conf *mdev;
2009         struct net_conf *old_conf, *new_conf = NULL;
2010         struct crypto crypto = { };
2011         struct drbd_tconn *tconn;
2012         enum drbd_ret_code retcode;
2013         int i;
2014         int err;
2015
2016         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
2017
2018         if (!adm_ctx.reply_skb)
2019                 return retcode;
2020         if (retcode != NO_ERROR)
2021                 goto out;
2022         if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2023                 drbd_msg_put_info("connection endpoint(s) missing");
2024                 retcode = ERR_INVALID_REQUEST;
2025                 goto out;
2026         }
2027
2028         /* No need for _rcu here. All reconfiguration is
2029          * strictly serialized on genl_lock(). We are protected against
2030          * concurrent reconfiguration/addition/deletion */
2031         list_for_each_entry(tconn, &drbd_tconns, all_tconn) {
2032                 if (nla_len(adm_ctx.my_addr) == tconn->my_addr_len &&
2033                     !memcmp(nla_data(adm_ctx.my_addr), &tconn->my_addr, tconn->my_addr_len)) {
2034                         retcode = ERR_LOCAL_ADDR;
2035                         goto out;
2036                 }
2037
2038                 if (nla_len(adm_ctx.peer_addr) == tconn->peer_addr_len &&
2039                     !memcmp(nla_data(adm_ctx.peer_addr), &tconn->peer_addr, tconn->peer_addr_len)) {
2040                         retcode = ERR_PEER_ADDR;
2041                         goto out;
2042                 }
2043         }
2044
2045         tconn = adm_ctx.tconn;
2046         conn_reconfig_start(tconn);
2047
2048         if (tconn->cstate > C_STANDALONE) {
2049                 retcode = ERR_NET_CONFIGURED;
2050                 goto fail;
2051         }
2052
2053         /* allocation not in the IO path, cqueue thread context */
2054         new_conf = kzalloc(sizeof(*new_conf), GFP_KERNEL);
2055         if (!new_conf) {
2056                 retcode = ERR_NOMEM;
2057                 goto fail;
2058         }
2059
2060         set_net_conf_defaults(new_conf);
2061
2062         err = net_conf_from_attrs(new_conf, info);
2063         if (err) {
2064                 retcode = ERR_MANDATORY_TAG;
2065                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2066                 goto fail;
2067         }
2068
2069         retcode = check_net_options(tconn, new_conf);
2070         if (retcode != NO_ERROR)
2071                 goto fail;
2072
2073         retcode = alloc_crypto(&crypto, new_conf);
2074         if (retcode != NO_ERROR)
2075                 goto fail;
2076
2077         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2078
2079         conn_flush_workqueue(tconn);
2080
2081         mutex_lock(&tconn->conf_update);
2082         old_conf = tconn->net_conf;
2083         if (old_conf) {
2084                 retcode = ERR_NET_CONFIGURED;
2085                 mutex_unlock(&tconn->conf_update);
2086                 goto fail;
2087         }
2088         rcu_assign_pointer(tconn->net_conf, new_conf);
2089
2090         conn_free_crypto(tconn);
2091         tconn->int_dig_in = crypto.int_dig_in;
2092         tconn->int_dig_vv = crypto.int_dig_vv;
2093         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
2094         tconn->integrity_tfm = crypto.integrity_tfm;
2095         tconn->csums_tfm = crypto.csums_tfm;
2096         tconn->verify_tfm = crypto.verify_tfm;
2097
2098         tconn->my_addr_len = nla_len(adm_ctx.my_addr);
2099         memcpy(&tconn->my_addr, nla_data(adm_ctx.my_addr), tconn->my_addr_len);
2100         tconn->peer_addr_len = nla_len(adm_ctx.peer_addr);
2101         memcpy(&tconn->peer_addr, nla_data(adm_ctx.peer_addr), tconn->peer_addr_len);
2102
2103         mutex_unlock(&tconn->conf_update);
2104
2105         rcu_read_lock();
2106         idr_for_each_entry(&tconn->volumes, mdev, i) {
2107                 mdev->send_cnt = 0;
2108                 mdev->recv_cnt = 0;
2109         }
2110         rcu_read_unlock();
2111
2112         retcode = conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2113
2114         conn_reconfig_done(tconn);
2115         drbd_adm_finish(info, retcode);
2116         return 0;
2117
2118 fail:
2119         free_crypto(&crypto);
2120         kfree(new_conf);
2121
2122         conn_reconfig_done(tconn);
2123 out:
2124         drbd_adm_finish(info, retcode);
2125         return 0;
2126 }
2127
2128 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
2129 {
2130         enum drbd_state_rv rv;
2131
2132         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2133                         force ? CS_HARD : 0);
2134
2135         switch (rv) {
2136         case SS_NOTHING_TO_DO:
2137                 break;
2138         case SS_ALREADY_STANDALONE:
2139                 return SS_SUCCESS;
2140         case SS_PRIMARY_NOP:
2141                 /* Our state checking code wants to see the peer outdated. */
2142                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2143                                                 pdsk, D_OUTDATED), CS_VERBOSE);
2144                 break;
2145         case SS_CW_FAILED_BY_PEER:
2146                 /* The peer probably wants to see us outdated. */
2147                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2148                                                         disk, D_OUTDATED), 0);
2149                 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2150                         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2151                                         CS_HARD);
2152                 }
2153                 break;
2154         default:;
2155                 /* no special handling necessary */
2156         }
2157
2158         if (rv >= SS_SUCCESS) {
2159                 enum drbd_state_rv rv2;
2160                 /* No one else can reconfigure the network while I am here.
2161                  * The state handling only uses drbd_thread_stop_nowait(),
2162                  * we want to really wait here until the receiver is no more.
2163                  */
2164                 drbd_thread_stop(&adm_ctx.tconn->receiver);
2165
2166                 /* Race breaker.  This additional state change request may be
2167                  * necessary, if this was a forced disconnect during a receiver
2168                  * restart.  We may have "killed" the receiver thread just
2169                  * after drbdd_init() returned.  Typically, we should be
2170                  * C_STANDALONE already, now, and this becomes a no-op.
2171                  */
2172                 rv2 = conn_request_state(tconn, NS(conn, C_STANDALONE),
2173                                 CS_VERBOSE | CS_HARD);
2174                 if (rv2 < SS_SUCCESS)
2175                         conn_err(tconn,
2176                                 "unexpected rv2=%d in conn_try_disconnect()\n",
2177                                 rv2);
2178         }
2179         return rv;
2180 }
2181
2182 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2183 {
2184         struct disconnect_parms parms;
2185         struct drbd_tconn *tconn;
2186         enum drbd_state_rv rv;
2187         enum drbd_ret_code retcode;
2188         int err;
2189
2190         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONNECTION);
2191         if (!adm_ctx.reply_skb)
2192                 return retcode;
2193         if (retcode != NO_ERROR)
2194                 goto fail;
2195
2196         tconn = adm_ctx.tconn;
2197         memset(&parms, 0, sizeof(parms));
2198         if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2199                 err = disconnect_parms_from_attrs(&parms, info);
2200                 if (err) {
2201                         retcode = ERR_MANDATORY_TAG;
2202                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2203                         goto fail;
2204                 }
2205         }
2206
2207         rv = conn_try_disconnect(tconn, parms.force_disconnect);
2208         if (rv < SS_SUCCESS)
2209                 retcode = rv;  /* FIXME: Type mismatch. */
2210         else
2211                 retcode = NO_ERROR;
2212  fail:
2213         drbd_adm_finish(info, retcode);
2214         return 0;
2215 }
2216
2217 void resync_after_online_grow(struct drbd_conf *mdev)
2218 {
2219         int iass; /* I am sync source */
2220
2221         dev_info(DEV, "Resync of new storage after online grow\n");
2222         if (mdev->state.role != mdev->state.peer)
2223                 iass = (mdev->state.role == R_PRIMARY);
2224         else
2225                 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
2226
2227         if (iass)
2228                 drbd_start_resync(mdev, C_SYNC_SOURCE);
2229         else
2230                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2231 }
2232
2233 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2234 {
2235         struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2236         struct resize_parms rs;
2237         struct drbd_conf *mdev;
2238         enum drbd_ret_code retcode;
2239         enum determine_dev_size dd;
2240         enum dds_flags ddsf;
2241         sector_t u_size;
2242         int err;
2243
2244         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2245         if (!adm_ctx.reply_skb)
2246                 return retcode;
2247         if (retcode != NO_ERROR)
2248                 goto fail;
2249
2250         memset(&rs, 0, sizeof(struct resize_parms));
2251         if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2252                 err = resize_parms_from_attrs(&rs, info);
2253                 if (err) {
2254                         retcode = ERR_MANDATORY_TAG;
2255                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2256                         goto fail;
2257                 }
2258         }
2259
2260         mdev = adm_ctx.mdev;
2261         if (mdev->state.conn > C_CONNECTED) {
2262                 retcode = ERR_RESIZE_RESYNC;
2263                 goto fail;
2264         }
2265
2266         if (mdev->state.role == R_SECONDARY &&
2267             mdev->state.peer == R_SECONDARY) {
2268                 retcode = ERR_NO_PRIMARY;
2269                 goto fail;
2270         }
2271
2272         if (!get_ldev(mdev)) {
2273                 retcode = ERR_NO_DISK;
2274                 goto fail;
2275         }
2276
2277         if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
2278                 retcode = ERR_NEED_APV_93;
2279                 goto fail;
2280         }
2281
2282         rcu_read_lock();
2283         u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
2284         rcu_read_unlock();
2285         if (u_size != (sector_t)rs.resize_size) {
2286                 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2287                 if (!new_disk_conf) {
2288                         retcode = ERR_NOMEM;
2289                         goto fail;
2290                 }
2291         }
2292
2293         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
2294                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2295
2296         if (new_disk_conf) {
2297                 mutex_lock(&mdev->tconn->conf_update);
2298                 old_disk_conf = mdev->ldev->disk_conf;
2299                 *new_disk_conf = *old_disk_conf;
2300                 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2301                 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
2302                 mutex_unlock(&mdev->tconn->conf_update);
2303                 synchronize_rcu();
2304                 kfree(old_disk_conf);
2305         }
2306
2307         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2308         dd = drbd_determine_dev_size(mdev, ddsf);
2309         drbd_md_sync(mdev);
2310         put_ldev(mdev);
2311         if (dd == dev_size_error) {
2312                 retcode = ERR_NOMEM_BITMAP;
2313                 goto fail;
2314         }
2315
2316         if (mdev->state.conn == C_CONNECTED) {
2317                 if (dd == grew)
2318                         set_bit(RESIZE_PENDING, &mdev->flags);
2319
2320                 drbd_send_uuids(mdev);
2321                 drbd_send_sizes(mdev, 1, ddsf);
2322         }
2323
2324  fail:
2325         drbd_adm_finish(info, retcode);
2326         return 0;
2327 }
2328
2329 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2330 {
2331         enum drbd_ret_code retcode;
2332         struct drbd_tconn *tconn;
2333         struct res_opts res_opts;
2334         int err;
2335
2336         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
2337         if (!adm_ctx.reply_skb)
2338                 return retcode;
2339         if (retcode != NO_ERROR)
2340                 goto fail;
2341         tconn = adm_ctx.tconn;
2342
2343         res_opts = tconn->res_opts;
2344         if (should_set_defaults(info))
2345                 set_res_opts_defaults(&res_opts);
2346
2347         err = res_opts_from_attrs(&res_opts, info);
2348         if (err && err != -ENOMSG) {
2349                 retcode = ERR_MANDATORY_TAG;
2350                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2351                 goto fail;
2352         }
2353
2354         err = set_resource_options(tconn, &res_opts);
2355         if (err) {
2356                 retcode = ERR_INVALID_REQUEST;
2357                 if (err == -ENOMEM)
2358                         retcode = ERR_NOMEM;
2359         }
2360
2361 fail:
2362         drbd_adm_finish(info, retcode);
2363         return 0;
2364 }
2365
2366 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2367 {
2368         struct drbd_conf *mdev;
2369         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2370
2371         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2372         if (!adm_ctx.reply_skb)
2373                 return retcode;
2374         if (retcode != NO_ERROR)
2375                 goto out;
2376
2377         mdev = adm_ctx.mdev;
2378
2379         /* If there is still bitmap IO pending, probably because of a previous
2380          * resync just being finished, wait for it before requesting a new resync. */
2381         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2382
2383         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2384
2385         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2386                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2387
2388         while (retcode == SS_NEED_CONNECTION) {
2389                 spin_lock_irq(&mdev->tconn->req_lock);
2390                 if (mdev->state.conn < C_CONNECTED)
2391                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2392                 spin_unlock_irq(&mdev->tconn->req_lock);
2393
2394                 if (retcode != SS_NEED_CONNECTION)
2395                         break;
2396
2397                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2398         }
2399
2400 out:
2401         drbd_adm_finish(info, retcode);
2402         return 0;
2403 }
2404
2405 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2406 {
2407         int rv;
2408
2409         rv = drbd_bmio_set_n_write(mdev);
2410         drbd_suspend_al(mdev);
2411         return rv;
2412 }
2413
2414 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2415                 union drbd_state mask, union drbd_state val)
2416 {
2417         enum drbd_ret_code retcode;
2418
2419         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2420         if (!adm_ctx.reply_skb)
2421                 return retcode;
2422         if (retcode != NO_ERROR)
2423                 goto out;
2424
2425         retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2426 out:
2427         drbd_adm_finish(info, retcode);
2428         return 0;
2429 }
2430
2431 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2432 {
2433         return drbd_adm_simple_request_state(skb, info, NS(conn, C_STARTING_SYNC_S));
2434 }
2435
2436 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2437 {
2438         enum drbd_ret_code retcode;
2439
2440         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2441         if (!adm_ctx.reply_skb)
2442                 return retcode;
2443         if (retcode != NO_ERROR)
2444                 goto out;
2445
2446         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2447                 retcode = ERR_PAUSE_IS_SET;
2448 out:
2449         drbd_adm_finish(info, retcode);
2450         return 0;
2451 }
2452
2453 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2454 {
2455         union drbd_dev_state s;
2456         enum drbd_ret_code retcode;
2457
2458         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2459         if (!adm_ctx.reply_skb)
2460                 return retcode;
2461         if (retcode != NO_ERROR)
2462                 goto out;
2463
2464         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2465                 s = adm_ctx.mdev->state;
2466                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2467                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2468                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2469                 } else {
2470                         retcode = ERR_PAUSE_IS_CLEAR;
2471                 }
2472         }
2473
2474 out:
2475         drbd_adm_finish(info, retcode);
2476         return 0;
2477 }
2478
2479 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2480 {
2481         return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2482 }
2483
2484 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2485 {
2486         struct drbd_conf *mdev;
2487         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2488
2489         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2490         if (!adm_ctx.reply_skb)
2491                 return retcode;
2492         if (retcode != NO_ERROR)
2493                 goto out;
2494
2495         mdev = adm_ctx.mdev;
2496         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2497                 drbd_uuid_new_current(mdev);
2498                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2499         }
2500         drbd_suspend_io(mdev);
2501         retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2502         if (retcode == SS_SUCCESS) {
2503                 if (mdev->state.conn < C_CONNECTED)
2504                         tl_clear(mdev->tconn);
2505                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2506                         tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2507         }
2508         drbd_resume_io(mdev);
2509
2510 out:
2511         drbd_adm_finish(info, retcode);
2512         return 0;
2513 }
2514
2515 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2516 {
2517         return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2518 }
2519
2520 int nla_put_drbd_cfg_context(struct sk_buff *skb, struct drbd_tconn *tconn, unsigned vnr)
2521 {
2522         struct nlattr *nla;
2523         nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2524         if (!nla)
2525                 goto nla_put_failure;
2526         if (vnr != VOLUME_UNSPECIFIED)
2527                 NLA_PUT_U32(skb, T_ctx_volume, vnr);
2528         NLA_PUT_STRING(skb, T_ctx_resource_name, tconn->name);
2529         if (tconn->my_addr_len)
2530                 NLA_PUT(skb, T_ctx_my_addr, tconn->my_addr_len, &tconn->my_addr);
2531         if (tconn->peer_addr_len)
2532                 NLA_PUT(skb, T_ctx_peer_addr, tconn->peer_addr_len, &tconn->peer_addr);
2533         nla_nest_end(skb, nla);
2534         return 0;
2535
2536 nla_put_failure:
2537         if (nla)
2538                 nla_nest_cancel(skb, nla);
2539         return -EMSGSIZE;
2540 }
2541
2542 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2543                 const struct sib_info *sib)
2544 {
2545         struct state_info *si = NULL; /* for sizeof(si->member); */
2546         struct net_conf *nc;
2547         struct nlattr *nla;
2548         int got_ldev;
2549         int err = 0;
2550         int exclude_sensitive;
2551
2552         /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2553          * to.  So we better exclude_sensitive information.
2554          *
2555          * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2556          * in the context of the requesting user process. Exclude sensitive
2557          * information, unless current has superuser.
2558          *
2559          * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2560          * relies on the current implementation of netlink_dump(), which
2561          * executes the dump callback successively from netlink_recvmsg(),
2562          * always in the context of the receiving process */
2563         exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2564
2565         got_ldev = get_ldev(mdev);
2566
2567         /* We need to add connection name and volume number information still.
2568          * Minor number is in drbd_genlmsghdr. */
2569         if (nla_put_drbd_cfg_context(skb, mdev->tconn, mdev->vnr))
2570                 goto nla_put_failure;
2571
2572         if (res_opts_to_skb(skb, &mdev->tconn->res_opts, exclude_sensitive))
2573                 goto nla_put_failure;
2574
2575         rcu_read_lock();
2576         if (got_ldev)
2577                 if (disk_conf_to_skb(skb, rcu_dereference(mdev->ldev->disk_conf), exclude_sensitive))
2578                         goto nla_put_failure;
2579
2580         nc = rcu_dereference(mdev->tconn->net_conf);
2581         if (nc)
2582                 err = net_conf_to_skb(skb, nc, exclude_sensitive);
2583         rcu_read_unlock();
2584         if (err)
2585                 goto nla_put_failure;
2586
2587         nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2588         if (!nla)
2589                 goto nla_put_failure;
2590         NLA_PUT_U32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY);
2591         NLA_PUT_U32(skb, T_current_state, mdev->state.i);
2592         NLA_PUT_U64(skb, T_ed_uuid, mdev->ed_uuid);
2593         NLA_PUT_U64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev));
2594
2595         if (got_ldev) {
2596                 NLA_PUT_U32(skb, T_disk_flags, mdev->ldev->md.flags);
2597                 NLA_PUT(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2598                 NLA_PUT_U64(skb, T_bits_total, drbd_bm_bits(mdev));
2599                 NLA_PUT_U64(skb, T_bits_oos, drbd_bm_total_weight(mdev));
2600                 if (C_SYNC_SOURCE <= mdev->state.conn &&
2601                     C_PAUSED_SYNC_T >= mdev->state.conn) {
2602                         NLA_PUT_U64(skb, T_bits_rs_total, mdev->rs_total);
2603                         NLA_PUT_U64(skb, T_bits_rs_failed, mdev->rs_failed);
2604                 }
2605         }
2606
2607         if (sib) {
2608                 switch(sib->sib_reason) {
2609                 case SIB_SYNC_PROGRESS:
2610                 case SIB_GET_STATUS_REPLY:
2611                         break;
2612                 case SIB_STATE_CHANGE:
2613                         NLA_PUT_U32(skb, T_prev_state, sib->os.i);
2614                         NLA_PUT_U32(skb, T_new_state, sib->ns.i);
2615                         break;
2616                 case SIB_HELPER_POST:
2617                         NLA_PUT_U32(skb,
2618                                 T_helper_exit_code, sib->helper_exit_code);
2619                         /* fall through */
2620                 case SIB_HELPER_PRE:
2621                         NLA_PUT_STRING(skb, T_helper, sib->helper_name);
2622                         break;
2623                 }
2624         }
2625         nla_nest_end(skb, nla);
2626
2627         if (0)
2628 nla_put_failure:
2629                 err = -EMSGSIZE;
2630         if (got_ldev)
2631                 put_ldev(mdev);
2632         return err;
2633 }
2634
2635 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2636 {
2637         enum drbd_ret_code retcode;
2638         int err;
2639
2640         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2641         if (!adm_ctx.reply_skb)
2642                 return retcode;
2643         if (retcode != NO_ERROR)
2644                 goto out;
2645
2646         err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2647         if (err) {
2648                 nlmsg_free(adm_ctx.reply_skb);
2649                 return err;
2650         }
2651 out:
2652         drbd_adm_finish(info, retcode);
2653         return 0;
2654 }
2655
2656 int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
2657 {
2658         struct drbd_conf *mdev;
2659         struct drbd_genlmsghdr *dh;
2660         struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2661         struct drbd_tconn *tconn = NULL;
2662         struct drbd_tconn *tmp;
2663         unsigned volume = cb->args[1];
2664
2665         /* Open coded, deferred, iteration:
2666          * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2667          *      idr_for_each_entry(&tconn->volumes, mdev, i) {
2668          *        ...
2669          *      }
2670          * }
2671          * where tconn is cb->args[0];
2672          * and i is cb->args[1];
2673          *
2674          * cb->args[2] indicates if we shall loop over all resources,
2675          * or just dump all volumes of a single resource.
2676          *
2677          * This may miss entries inserted after this dump started,
2678          * or entries deleted before they are reached.
2679          *
2680          * We need to make sure the mdev won't disappear while
2681          * we are looking at it, and revalidate our iterators
2682          * on each iteration.
2683          */
2684
2685         /* synchronize with conn_create()/conn_destroy() */
2686         rcu_read_lock();
2687         /* revalidate iterator position */
2688         list_for_each_entry_rcu(tmp, &drbd_tconns, all_tconn) {
2689                 if (pos == NULL) {
2690                         /* first iteration */
2691                         pos = tmp;
2692                         tconn = pos;
2693                         break;
2694                 }
2695                 if (tmp == pos) {
2696                         tconn = pos;
2697                         break;
2698                 }
2699         }
2700         if (tconn) {
2701 next_tconn:
2702                 mdev = idr_get_next(&tconn->volumes, &volume);
2703                 if (!mdev) {
2704                         /* No more volumes to dump on this tconn.
2705                          * Advance tconn iterator. */
2706                         pos = list_entry_rcu(tconn->all_tconn.next,
2707                                              struct drbd_tconn, all_tconn);
2708                         /* Did we dump any volume on this tconn yet? */
2709                         if (volume != 0) {
2710                                 /* If we reached the end of the list,
2711                                  * or only a single resource dump was requested,
2712                                  * we are done. */
2713                                 if (&pos->all_tconn == &drbd_tconns || cb->args[2])
2714                                         goto out;
2715                                 volume = 0;
2716                                 tconn = pos;
2717                                 goto next_tconn;
2718                         }
2719                 }
2720
2721                 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).pid,
2722                                 cb->nlh->nlmsg_seq, &drbd_genl_family,
2723                                 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2724                 if (!dh)
2725                         goto out;
2726
2727                 if (!mdev) {
2728                         /* this is a tconn without a single volume */
2729                         dh->minor = -1U;
2730                         dh->ret_code = NO_ERROR;
2731                         if (nla_put_drbd_cfg_context(skb, tconn, VOLUME_UNSPECIFIED))
2732                                 genlmsg_cancel(skb, dh);
2733                         else
2734                                 genlmsg_end(skb, dh);
2735                         goto out;
2736                 }
2737
2738                 D_ASSERT(mdev->vnr == volume);
2739                 D_ASSERT(mdev->tconn == tconn);
2740
2741                 dh->minor = mdev_to_minor(mdev);
2742                 dh->ret_code = NO_ERROR;
2743
2744                 if (nla_put_status_info(skb, mdev, NULL)) {
2745                         genlmsg_cancel(skb, dh);
2746                         goto out;
2747                 }
2748                 genlmsg_end(skb, dh);
2749         }
2750
2751 out:
2752         rcu_read_unlock();
2753         /* where to start the next iteration */
2754         cb->args[0] = (long)pos;
2755         cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2756
2757         /* No more tconns/volumes/minors found results in an empty skb.
2758          * Which will terminate the dump. */
2759         return skb->len;
2760 }
2761
2762 /*
2763  * Request status of all resources, or of all volumes within a single resource.
2764  *
2765  * This is a dump, as the answer may not fit in a single reply skb otherwise.
2766  * Which means we cannot use the family->attrbuf or other such members, because
2767  * dump is NOT protected by the genl_lock().  During dump, we only have access
2768  * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
2769  *
2770  * Once things are setup properly, we call into get_one_status().
2771  */
2772 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2773 {
2774         const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
2775         struct nlattr *nla;
2776         const char *resource_name;
2777         struct drbd_tconn *tconn;
2778         int maxtype;
2779
2780         /* Is this a followup call? */
2781         if (cb->args[0]) {
2782                 /* ... of a single resource dump,
2783                  * and the resource iterator has been advanced already? */
2784                 if (cb->args[2] && cb->args[2] != cb->args[0])
2785                         return 0; /* DONE. */
2786                 goto dump;
2787         }
2788
2789         /* First call (from netlink_dump_start).  We need to figure out
2790          * which resource(s) the user wants us to dump. */
2791         nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
2792                         nlmsg_attrlen(cb->nlh, hdrlen),
2793                         DRBD_NLA_CFG_CONTEXT);
2794
2795         /* No explicit context given.  Dump all. */
2796         if (!nla)
2797                 goto dump;
2798         maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
2799         nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
2800         if (IS_ERR(nla))
2801                 return PTR_ERR(nla);
2802         /* context given, but no name present? */
2803         if (!nla)
2804                 return -EINVAL;
2805         resource_name = nla_data(nla);
2806         tconn = conn_get_by_name(resource_name);
2807
2808         if (!tconn)
2809                 return -ENODEV;
2810
2811         kref_put(&tconn->kref, &conn_destroy); /* get_one_status() (re)validates tconn by itself */
2812
2813         /* prime iterators, and set "filter" mode mark:
2814          * only dump this tconn. */
2815         cb->args[0] = (long)tconn;
2816         /* cb->args[1] = 0; passed in this way. */
2817         cb->args[2] = (long)tconn;
2818
2819 dump:
2820         return get_one_status(skb, cb);
2821 }
2822
2823 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2824 {
2825         enum drbd_ret_code retcode;
2826         struct timeout_parms tp;
2827         int err;
2828
2829         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2830         if (!adm_ctx.reply_skb)
2831                 return retcode;
2832         if (retcode != NO_ERROR)
2833                 goto out;
2834
2835         tp.timeout_type =
2836                 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2837                 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2838                 UT_DEFAULT;
2839
2840         err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2841         if (err) {
2842                 nlmsg_free(adm_ctx.reply_skb);
2843                 return err;
2844         }
2845 out:
2846         drbd_adm_finish(info, retcode);
2847         return 0;
2848 }
2849
2850 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2851 {
2852         struct drbd_conf *mdev;
2853         enum drbd_ret_code retcode;
2854
2855         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2856         if (!adm_ctx.reply_skb)
2857                 return retcode;
2858         if (retcode != NO_ERROR)
2859                 goto out;
2860
2861         mdev = adm_ctx.mdev;
2862         if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2863                 /* resume from last known position, if possible */
2864                 struct start_ov_parms parms =
2865                         { .ov_start_sector = mdev->ov_start_sector };
2866                 int err = start_ov_parms_from_attrs(&parms, info);
2867                 if (err) {
2868                         retcode = ERR_MANDATORY_TAG;
2869                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2870                         goto out;
2871                 }
2872                 /* w_make_ov_request expects position to be aligned */
2873                 mdev->ov_start_sector = parms.ov_start_sector & ~BM_SECT_PER_BIT;
2874         }
2875         /* If there is still bitmap IO pending, e.g. previous resync or verify
2876          * just being finished, wait for it before requesting a new resync. */
2877         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2878         retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2879 out:
2880         drbd_adm_finish(info, retcode);
2881         return 0;
2882 }
2883
2884
2885 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
2886 {
2887         struct drbd_conf *mdev;
2888         enum drbd_ret_code retcode;
2889         int skip_initial_sync = 0;
2890         int err;
2891         struct new_c_uuid_parms args;
2892
2893         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2894         if (!adm_ctx.reply_skb)
2895                 return retcode;
2896         if (retcode != NO_ERROR)
2897                 goto out_nolock;
2898
2899         mdev = adm_ctx.mdev;
2900         memset(&args, 0, sizeof(args));
2901         if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
2902                 err = new_c_uuid_parms_from_attrs(&args, info);
2903                 if (err) {
2904                         retcode = ERR_MANDATORY_TAG;
2905                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2906                         goto out_nolock;
2907                 }
2908         }
2909
2910         mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2911
2912         if (!get_ldev(mdev)) {
2913                 retcode = ERR_NO_DISK;
2914                 goto out;
2915         }
2916
2917         /* this is "skip initial sync", assume to be clean */
2918         if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2919             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2920                 dev_info(DEV, "Preparing to skip initial sync\n");
2921                 skip_initial_sync = 1;
2922         } else if (mdev->state.conn != C_STANDALONE) {
2923                 retcode = ERR_CONNECTED;
2924                 goto out_dec;
2925         }
2926
2927         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2928         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2929
2930         if (args.clear_bm) {
2931                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2932                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2933                 if (err) {
2934                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2935                         retcode = ERR_IO_MD_DISK;
2936                 }
2937                 if (skip_initial_sync) {
2938                         drbd_send_uuids_skip_initial_sync(mdev);
2939                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2940                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2941                         spin_lock_irq(&mdev->tconn->req_lock);
2942                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2943                                         CS_VERBOSE, NULL);
2944                         spin_unlock_irq(&mdev->tconn->req_lock);
2945                 }
2946         }
2947
2948         drbd_md_sync(mdev);
2949 out_dec:
2950         put_ldev(mdev);
2951 out:
2952         mutex_unlock(mdev->state_mutex);
2953 out_nolock:
2954         drbd_adm_finish(info, retcode);
2955         return 0;
2956 }
2957
2958 static enum drbd_ret_code
2959 drbd_check_resource_name(const char *name)
2960 {
2961         if (!name || !name[0]) {
2962                 drbd_msg_put_info("resource name missing");
2963                 return ERR_MANDATORY_TAG;
2964         }
2965         /* if we want to use these in sysfs/configfs/debugfs some day,
2966          * we must not allow slashes */
2967         if (strchr(name, '/')) {
2968                 drbd_msg_put_info("invalid resource name");
2969                 return ERR_INVALID_REQUEST;
2970         }
2971         return NO_ERROR;
2972 }
2973
2974 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
2975 {
2976         enum drbd_ret_code retcode;
2977         struct res_opts res_opts;
2978         int err;
2979
2980         retcode = drbd_adm_prepare(skb, info, 0);
2981         if (!adm_ctx.reply_skb)
2982                 return retcode;
2983         if (retcode != NO_ERROR)
2984                 goto out;
2985
2986         set_res_opts_defaults(&res_opts);
2987         err = res_opts_from_attrs(&res_opts, info);
2988         if (err && err != -ENOMSG) {
2989                 retcode = ERR_MANDATORY_TAG;
2990                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2991                 goto out;
2992         }
2993
2994         retcode = drbd_check_resource_name(adm_ctx.resource_name);
2995         if (retcode != NO_ERROR)
2996                 goto out;
2997
2998         if (adm_ctx.tconn) {
2999                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
3000                         retcode = ERR_INVALID_REQUEST;
3001                         drbd_msg_put_info("resource exists");
3002                 }
3003                 /* else: still NO_ERROR */
3004                 goto out;
3005         }
3006
3007         if (!conn_create(adm_ctx.resource_name, &res_opts))
3008                 retcode = ERR_NOMEM;
3009 out:
3010         drbd_adm_finish(info, retcode);
3011         return 0;
3012 }
3013
3014 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
3015 {
3016         struct drbd_genlmsghdr *dh = info->userhdr;
3017         enum drbd_ret_code retcode;
3018
3019         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
3020         if (!adm_ctx.reply_skb)
3021                 return retcode;
3022         if (retcode != NO_ERROR)
3023                 goto out;
3024
3025         /* FIXME drop minor_count parameter, limit to MINORMASK */
3026         if (dh->minor >= minor_count) {
3027                 drbd_msg_put_info("requested minor out of range");
3028                 retcode = ERR_INVALID_REQUEST;
3029                 goto out;
3030         }
3031         if (adm_ctx.volume > DRBD_VOLUME_MAX) {
3032                 drbd_msg_put_info("requested volume id out of range");
3033                 retcode = ERR_INVALID_REQUEST;
3034                 goto out;
3035         }
3036
3037         /* drbd_adm_prepare made sure already
3038          * that mdev->tconn and mdev->vnr match the request. */
3039         if (adm_ctx.mdev) {
3040                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
3041                         retcode = ERR_MINOR_EXISTS;
3042                 /* else: still NO_ERROR */
3043                 goto out;
3044         }
3045
3046         retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
3047 out:
3048         drbd_adm_finish(info, retcode);
3049         return 0;
3050 }
3051
3052 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
3053 {
3054         if (mdev->state.disk == D_DISKLESS &&
3055             /* no need to be mdev->state.conn == C_STANDALONE &&
3056              * we may want to delete a minor from a live replication group.
3057              */
3058             mdev->state.role == R_SECONDARY) {
3059                 idr_remove(&mdev->tconn->volumes, mdev->vnr);
3060                 idr_remove(&minors, mdev_to_minor(mdev));
3061                 del_gendisk(mdev->vdisk);
3062                 synchronize_rcu();
3063                 kref_put(&mdev->kref, &drbd_minor_destroy);
3064                 return NO_ERROR;
3065         } else
3066                 return ERR_MINOR_CONFIGURED;
3067 }
3068
3069 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
3070 {
3071         enum drbd_ret_code retcode;
3072
3073         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3074         if (!adm_ctx.reply_skb)
3075                 return retcode;
3076         if (retcode != NO_ERROR)
3077                 goto out;
3078
3079         retcode = adm_delete_minor(adm_ctx.mdev);
3080 out:
3081         drbd_adm_finish(info, retcode);
3082         return 0;
3083 }
3084
3085 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
3086 {
3087         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3088         struct drbd_conf *mdev;
3089         unsigned i;
3090
3091         retcode = drbd_adm_prepare(skb, info, 0);
3092         if (!adm_ctx.reply_skb)
3093                 return retcode;
3094         if (retcode != NO_ERROR)
3095                 goto out;
3096
3097         if (!adm_ctx.tconn) {
3098                 retcode = ERR_RES_NOT_KNOWN;
3099                 goto out;
3100         }
3101
3102         /* demote */
3103         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3104                 retcode = drbd_set_role(mdev, R_SECONDARY, 0);
3105                 if (retcode < SS_SUCCESS) {
3106                         drbd_msg_put_info("failed to demote");
3107                         goto out;
3108                 }
3109         }
3110
3111         retcode = conn_try_disconnect(adm_ctx.tconn, 0);
3112         if (retcode < SS_SUCCESS) {
3113                 drbd_msg_put_info("failed to disconnect");
3114                 goto out;
3115         }
3116
3117         /* detach */
3118         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3119                 retcode = adm_detach(mdev);
3120                 if (retcode < SS_SUCCESS) {
3121                         drbd_msg_put_info("failed to detach");
3122                         goto out;
3123                 }
3124         }
3125
3126         /* If we reach this, all volumes (of this tconn) are Secondary,
3127          * Disconnected, Diskless, aka Unconfigured. Make sure all threads have
3128          * actually stopped, state handling only does drbd_thread_stop_nowait(). */
3129         drbd_thread_stop(&adm_ctx.tconn->worker);
3130
3131         /* Now, nothing can fail anymore */
3132
3133         /* delete volumes */
3134         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3135                 retcode = adm_delete_minor(mdev);
3136                 if (retcode != NO_ERROR) {
3137                         /* "can not happen" */
3138                         drbd_msg_put_info("failed to delete volume");
3139                         goto out;
3140                 }
3141         }
3142
3143         /* delete connection */
3144         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3145                 list_del_rcu(&adm_ctx.tconn->all_tconn);
3146                 synchronize_rcu();
3147                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3148
3149                 retcode = NO_ERROR;
3150         } else {
3151                 /* "can not happen" */
3152                 retcode = ERR_RES_IN_USE;
3153                 drbd_msg_put_info("failed to delete connection");
3154         }
3155         goto out;
3156 out:
3157         drbd_adm_finish(info, retcode);
3158         return 0;
3159 }
3160
3161 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
3162 {
3163         enum drbd_ret_code retcode;
3164
3165         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
3166         if (!adm_ctx.reply_skb)
3167                 return retcode;
3168         if (retcode != NO_ERROR)
3169                 goto out;
3170
3171         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3172                 list_del_rcu(&adm_ctx.tconn->all_tconn);
3173                 synchronize_rcu();
3174                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3175
3176                 retcode = NO_ERROR;
3177         } else {
3178                 retcode = ERR_RES_IN_USE;
3179         }
3180
3181         if (retcode == NO_ERROR)
3182                 drbd_thread_stop(&adm_ctx.tconn->worker);
3183 out:
3184         drbd_adm_finish(info, retcode);
3185         return 0;
3186 }
3187
3188 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
3189 {
3190         static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3191         struct sk_buff *msg;
3192         struct drbd_genlmsghdr *d_out;
3193         unsigned seq;
3194         int err = -ENOMEM;
3195
3196         seq = atomic_inc_return(&drbd_genl_seq);
3197         msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3198         if (!msg)
3199                 goto failed;
3200
3201         err = -EMSGSIZE;
3202         d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3203         if (!d_out) /* cannot happen, but anyways. */
3204                 goto nla_put_failure;
3205         d_out->minor = mdev_to_minor(mdev);
3206         d_out->ret_code = NO_ERROR;
3207
3208         if (nla_put_status_info(msg, mdev, sib))
3209                 goto nla_put_failure;
3210         genlmsg_end(msg, d_out);
3211         err = drbd_genl_multicast_events(msg, 0);
3212         /* msg has been consumed or freed in netlink_broadcast() */
3213         if (err && err != -ESRCH)
3214                 goto failed;
3215
3216         return;
3217
3218 nla_put_failure:
3219         nlmsg_free(msg);
3220 failed:
3221         dev_err(DEV, "Error %d while broadcasting event. "
3222                         "Event seq:%u sib_reason:%u\n",
3223                         err, seq, sib->sib_reason);
3224 }