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         mod_timer(&mdev->request_timer, jiffies + HZ);
1240         goto success;
1241
1242 fail_unlock:
1243         mutex_unlock(&mdev->tconn->conf_update);
1244  fail:
1245         kfree(new_disk_conf);
1246         kfree(new_plan);
1247 success:
1248         put_ldev(mdev);
1249  out:
1250         drbd_adm_finish(info, retcode);
1251         return 0;
1252 }
1253
1254 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1255 {
1256         struct drbd_conf *mdev;
1257         int err;
1258         enum drbd_ret_code retcode;
1259         enum determine_dev_size dd;
1260         sector_t max_possible_sectors;
1261         sector_t min_md_device_sectors;
1262         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1263         struct disk_conf *new_disk_conf = NULL;
1264         struct block_device *bdev;
1265         struct lru_cache *resync_lru = NULL;
1266         struct fifo_buffer *new_plan = NULL;
1267         union drbd_state ns, os;
1268         enum drbd_state_rv rv;
1269         struct net_conf *nc;
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         /* Reset the "barriers don't work" bits here, then force meta data to
1480          * be written, to ensure we determine if barriers are supported. */
1481         if (new_disk_conf->md_flushes)
1482                 clear_bit(MD_NO_FUA, &mdev->flags);
1483         else
1484                 set_bit(MD_NO_FUA, &mdev->flags);
1485
1486         /* Point of no return reached.
1487          * Devices and memory are no longer released by error cleanup below.
1488          * now mdev takes over responsibility, and the state engine should
1489          * clean it up somewhere.  */
1490         D_ASSERT(mdev->ldev == NULL);
1491         mdev->ldev = nbc;
1492         mdev->resync = resync_lru;
1493         mdev->rs_plan_s = new_plan;
1494         nbc = NULL;
1495         resync_lru = NULL;
1496         new_disk_conf = NULL;
1497         new_plan = NULL;
1498
1499         mdev->write_ordering = WO_bdev_flush;
1500         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1501
1502         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1503                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1504         else
1505                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1506
1507         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1508             !(mdev->state.role == R_PRIMARY && mdev->tconn->susp_nod))
1509                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1510
1511         mdev->send_cnt = 0;
1512         mdev->recv_cnt = 0;
1513         mdev->read_cnt = 0;
1514         mdev->writ_cnt = 0;
1515
1516         drbd_reconsider_max_bio_size(mdev);
1517
1518         /* If I am currently not R_PRIMARY,
1519          * but meta data primary indicator is set,
1520          * I just now recover from a hard crash,
1521          * and have been R_PRIMARY before that crash.
1522          *
1523          * Now, if I had no connection before that crash
1524          * (have been degraded R_PRIMARY), chances are that
1525          * I won't find my peer now either.
1526          *
1527          * In that case, and _only_ in that case,
1528          * we use the degr-wfc-timeout instead of the default,
1529          * so we can automatically recover from a crash of a
1530          * degraded but active "cluster" after a certain timeout.
1531          */
1532         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1533         if (mdev->state.role != R_PRIMARY &&
1534              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1535             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1536                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1537
1538         dd = drbd_determine_dev_size(mdev, 0);
1539         if (dd == dev_size_error) {
1540                 retcode = ERR_NOMEM_BITMAP;
1541                 goto force_diskless_dec;
1542         } else if (dd == grew)
1543                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1544
1545         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1546                 dev_info(DEV, "Assuming that all blocks are out of sync "
1547                      "(aka FullSync)\n");
1548                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1549                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1550                         retcode = ERR_IO_MD_DISK;
1551                         goto force_diskless_dec;
1552                 }
1553         } else {
1554                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1555                         "read from attaching", BM_LOCKED_MASK)) {
1556                         retcode = ERR_IO_MD_DISK;
1557                         goto force_diskless_dec;
1558                 }
1559         }
1560
1561         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1562                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1563
1564         spin_lock_irq(&mdev->tconn->req_lock);
1565         os = drbd_read_state(mdev);
1566         ns = os;
1567         /* If MDF_CONSISTENT is not set go into inconsistent state,
1568            otherwise investigate MDF_WasUpToDate...
1569            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1570            otherwise into D_CONSISTENT state.
1571         */
1572         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1573                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1574                         ns.disk = D_CONSISTENT;
1575                 else
1576                         ns.disk = D_OUTDATED;
1577         } else {
1578                 ns.disk = D_INCONSISTENT;
1579         }
1580
1581         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1582                 ns.pdsk = D_OUTDATED;
1583
1584         rcu_read_lock();
1585         if (ns.disk == D_CONSISTENT &&
1586             (ns.pdsk == D_OUTDATED || rcu_dereference(mdev->ldev->disk_conf)->fencing == FP_DONT_CARE))
1587                 ns.disk = D_UP_TO_DATE;
1588         rcu_read_unlock();
1589
1590         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1591            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1592            this point, because drbd_request_state() modifies these
1593            flags. */
1594
1595         /* In case we are C_CONNECTED postpone any decision on the new disk
1596            state after the negotiation phase. */
1597         if (mdev->state.conn == C_CONNECTED) {
1598                 mdev->new_state_tmp.i = ns.i;
1599                 ns.i = os.i;
1600                 ns.disk = D_NEGOTIATING;
1601
1602                 /* We expect to receive up-to-date UUIDs soon.
1603                    To avoid a race in receive_state, free p_uuid while
1604                    holding req_lock. I.e. atomic with the state change */
1605                 kfree(mdev->p_uuid);
1606                 mdev->p_uuid = NULL;
1607         }
1608
1609         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1610         spin_unlock_irq(&mdev->tconn->req_lock);
1611
1612         if (rv < SS_SUCCESS)
1613                 goto force_diskless_dec;
1614
1615         mod_timer(&mdev->request_timer, jiffies + HZ);
1616
1617         if (mdev->state.role == R_PRIMARY)
1618                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1619         else
1620                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1621
1622         drbd_md_mark_dirty(mdev);
1623         drbd_md_sync(mdev);
1624
1625         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1626         put_ldev(mdev);
1627         conn_reconfig_done(mdev->tconn);
1628         drbd_adm_finish(info, retcode);
1629         return 0;
1630
1631  force_diskless_dec:
1632         put_ldev(mdev);
1633  force_diskless:
1634         drbd_force_state(mdev, NS(disk, D_DISKLESS));
1635         drbd_md_sync(mdev);
1636  fail:
1637         conn_reconfig_done(mdev->tconn);
1638         if (nbc) {
1639                 if (nbc->backing_bdev)
1640                         blkdev_put(nbc->backing_bdev,
1641                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1642                 if (nbc->md_bdev)
1643                         blkdev_put(nbc->md_bdev,
1644                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1645                 kfree(nbc);
1646         }
1647         kfree(new_disk_conf);
1648         lc_destroy(resync_lru);
1649         kfree(new_plan);
1650
1651  finish:
1652         drbd_adm_finish(info, retcode);
1653         return 0;
1654 }
1655
1656 static int adm_detach(struct drbd_conf *mdev, int force)
1657 {
1658         enum drbd_state_rv retcode;
1659         int ret;
1660
1661         if (force) {
1662                 drbd_force_state(mdev, NS(disk, D_FAILED));
1663                 retcode = SS_SUCCESS;
1664                 goto out;
1665         }
1666
1667         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1668         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1669         /* D_FAILED will transition to DISKLESS. */
1670         ret = wait_event_interruptible(mdev->misc_wait,
1671                         mdev->state.disk != D_FAILED);
1672         drbd_resume_io(mdev);
1673         if ((int)retcode == (int)SS_IS_DISKLESS)
1674                 retcode = SS_NOTHING_TO_DO;
1675         if (ret)
1676                 retcode = ERR_INTR;
1677 out:
1678         return retcode;
1679 }
1680
1681 /* Detaching the disk is a process in multiple stages.  First we need to lock
1682  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1683  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1684  * internal references as well.
1685  * Only then we have finally detached. */
1686 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1687 {
1688         enum drbd_ret_code retcode;
1689         struct detach_parms parms = { };
1690         int err;
1691
1692         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1693         if (!adm_ctx.reply_skb)
1694                 return retcode;
1695         if (retcode != NO_ERROR)
1696                 goto out;
1697
1698         if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
1699                 err = detach_parms_from_attrs(&parms, info);
1700                 if (err) {
1701                         retcode = ERR_MANDATORY_TAG;
1702                         drbd_msg_put_info(from_attrs_err_to_txt(err));
1703                         goto out;
1704                 }
1705         }
1706
1707         retcode = adm_detach(adm_ctx.mdev, parms.force_detach);
1708 out:
1709         drbd_adm_finish(info, retcode);
1710         return 0;
1711 }
1712
1713 static bool conn_resync_running(struct drbd_tconn *tconn)
1714 {
1715         struct drbd_conf *mdev;
1716         bool rv = false;
1717         int vnr;
1718
1719         rcu_read_lock();
1720         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1721                 if (mdev->state.conn == C_SYNC_SOURCE ||
1722                     mdev->state.conn == C_SYNC_TARGET ||
1723                     mdev->state.conn == C_PAUSED_SYNC_S ||
1724                     mdev->state.conn == C_PAUSED_SYNC_T) {
1725                         rv = true;
1726                         break;
1727                 }
1728         }
1729         rcu_read_unlock();
1730
1731         return rv;
1732 }
1733
1734 static bool conn_ov_running(struct drbd_tconn *tconn)
1735 {
1736         struct drbd_conf *mdev;
1737         bool rv = false;
1738         int vnr;
1739
1740         rcu_read_lock();
1741         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1742                 if (mdev->state.conn == C_VERIFY_S ||
1743                     mdev->state.conn == C_VERIFY_T) {
1744                         rv = true;
1745                         break;
1746                 }
1747         }
1748         rcu_read_unlock();
1749
1750         return rv;
1751 }
1752
1753 static enum drbd_ret_code
1754 _check_net_options(struct drbd_tconn *tconn, struct net_conf *old_conf, struct net_conf *new_conf)
1755 {
1756         struct drbd_conf *mdev;
1757         int i;
1758
1759         if (old_conf && tconn->cstate == C_WF_REPORT_PARAMS && tconn->agreed_pro_version < 100) {
1760                 if (new_conf->wire_protocol != old_conf->wire_protocol)
1761                         return ERR_NEED_APV_100;
1762
1763                 if (new_conf->two_primaries != old_conf->two_primaries)
1764                         return ERR_NEED_APV_100;
1765
1766                 if (!new_conf->integrity_alg != !old_conf->integrity_alg)
1767                         return ERR_NEED_APV_100;
1768
1769                 if (strcmp(new_conf->integrity_alg, old_conf->integrity_alg))
1770                         return ERR_NEED_APV_100;
1771         }
1772
1773         if (!new_conf->two_primaries &&
1774             conn_highest_role(tconn) == R_PRIMARY &&
1775             conn_highest_peer(tconn) == R_PRIMARY)
1776                 return ERR_NEED_ALLOW_TWO_PRI;
1777
1778         if (new_conf->two_primaries &&
1779             (new_conf->wire_protocol != DRBD_PROT_C))
1780                 return ERR_NOT_PROTO_C;
1781
1782         idr_for_each_entry(&tconn->volumes, mdev, i) {
1783                 if (get_ldev(mdev)) {
1784                         enum drbd_fencing_p fp = rcu_dereference(mdev->ldev->disk_conf)->fencing;
1785                         put_ldev(mdev);
1786                         if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
1787                                 return ERR_STONITH_AND_PROT_A;
1788                 }
1789                 if (mdev->state.role == R_PRIMARY && new_conf->discard_my_data)
1790                         return ERR_DISCARD;
1791         }
1792
1793         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A)
1794                 return ERR_CONG_NOT_PROTO_A;
1795
1796         return NO_ERROR;
1797 }
1798
1799 static enum drbd_ret_code
1800 check_net_options(struct drbd_tconn *tconn, struct net_conf *new_conf)
1801 {
1802         static enum drbd_ret_code rv;
1803         struct drbd_conf *mdev;
1804         int i;
1805
1806         rcu_read_lock();
1807         rv = _check_net_options(tconn, rcu_dereference(tconn->net_conf), new_conf);
1808         rcu_read_unlock();
1809
1810         /* tconn->volumes protected by genl_lock() here */
1811         idr_for_each_entry(&tconn->volumes, mdev, i) {
1812                 if (!mdev->bitmap) {
1813                         if(drbd_bm_init(mdev))
1814                                 return ERR_NOMEM;
1815                 }
1816         }
1817
1818         return rv;
1819 }
1820
1821 struct crypto {
1822         struct crypto_hash *verify_tfm;
1823         struct crypto_hash *csums_tfm;
1824         struct crypto_hash *cram_hmac_tfm;
1825         struct crypto_hash *integrity_tfm;
1826         void *int_dig_in;
1827         void *int_dig_vv;
1828 };
1829
1830 static int
1831 alloc_hash(struct crypto_hash **tfm, char *tfm_name, int err_alg)
1832 {
1833         if (!tfm_name[0])
1834                 return NO_ERROR;
1835
1836         *tfm = crypto_alloc_hash(tfm_name, 0, CRYPTO_ALG_ASYNC);
1837         if (IS_ERR(*tfm)) {
1838                 *tfm = NULL;
1839                 return err_alg;
1840         }
1841
1842         return NO_ERROR;
1843 }
1844
1845 static enum drbd_ret_code
1846 alloc_crypto(struct crypto *crypto, struct net_conf *new_conf)
1847 {
1848         char hmac_name[CRYPTO_MAX_ALG_NAME];
1849         enum drbd_ret_code rv;
1850         int hash_size;
1851
1852         rv = alloc_hash(&crypto->csums_tfm, new_conf->csums_alg,
1853                        ERR_CSUMS_ALG);
1854         if (rv != NO_ERROR)
1855                 return rv;
1856         rv = alloc_hash(&crypto->verify_tfm, new_conf->verify_alg,
1857                        ERR_VERIFY_ALG);
1858         if (rv != NO_ERROR)
1859                 return rv;
1860         rv = alloc_hash(&crypto->integrity_tfm, new_conf->integrity_alg,
1861                        ERR_INTEGRITY_ALG);
1862         if (rv != NO_ERROR)
1863                 return rv;
1864         if (new_conf->cram_hmac_alg[0] != 0) {
1865                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1866                          new_conf->cram_hmac_alg);
1867
1868                 rv = alloc_hash(&crypto->cram_hmac_tfm, hmac_name,
1869                                ERR_AUTH_ALG);
1870         }
1871         if (crypto->integrity_tfm) {
1872                 hash_size = crypto_hash_digestsize(crypto->integrity_tfm);
1873                 crypto->int_dig_in = kmalloc(hash_size, GFP_KERNEL);
1874                 if (!crypto->int_dig_in)
1875                         return ERR_NOMEM;
1876                 crypto->int_dig_vv = kmalloc(hash_size, GFP_KERNEL);
1877                 if (!crypto->int_dig_vv)
1878                         return ERR_NOMEM;
1879         }
1880
1881         return rv;
1882 }
1883
1884 static void free_crypto(struct crypto *crypto)
1885 {
1886         kfree(crypto->int_dig_in);
1887         kfree(crypto->int_dig_vv);
1888         crypto_free_hash(crypto->cram_hmac_tfm);
1889         crypto_free_hash(crypto->integrity_tfm);
1890         crypto_free_hash(crypto->csums_tfm);
1891         crypto_free_hash(crypto->verify_tfm);
1892 }
1893
1894 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
1895 {
1896         enum drbd_ret_code retcode;
1897         struct drbd_tconn *tconn;
1898         struct net_conf *old_conf, *new_conf = NULL;
1899         int err;
1900         int ovr; /* online verify running */
1901         int rsr; /* re-sync running */
1902         struct crypto crypto = { };
1903
1904         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONNECTION);
1905         if (!adm_ctx.reply_skb)
1906                 return retcode;
1907         if (retcode != NO_ERROR)
1908                 goto out;
1909
1910         tconn = adm_ctx.tconn;
1911
1912         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1913         if (!new_conf) {
1914                 retcode = ERR_NOMEM;
1915                 goto out;
1916         }
1917
1918         conn_reconfig_start(tconn);
1919
1920         mutex_lock(&tconn->data.mutex);
1921         mutex_lock(&tconn->conf_update);
1922         old_conf = tconn->net_conf;
1923
1924         if (!old_conf) {
1925                 drbd_msg_put_info("net conf missing, try connect");
1926                 retcode = ERR_INVALID_REQUEST;
1927                 goto fail;
1928         }
1929
1930         *new_conf = *old_conf;
1931         if (should_set_defaults(info))
1932                 set_net_conf_defaults(new_conf);
1933
1934         err = net_conf_from_attrs_for_change(new_conf, info);
1935         if (err && err != -ENOMSG) {
1936                 retcode = ERR_MANDATORY_TAG;
1937                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1938                 goto fail;
1939         }
1940
1941         retcode = check_net_options(tconn, new_conf);
1942         if (retcode != NO_ERROR)
1943                 goto fail;
1944
1945         /* re-sync running */
1946         rsr = conn_resync_running(tconn);
1947         if (rsr && strcmp(new_conf->csums_alg, old_conf->csums_alg)) {
1948                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1949                 goto fail;
1950         }
1951
1952         /* online verify running */
1953         ovr = conn_ov_running(tconn);
1954         if (ovr && strcmp(new_conf->verify_alg, old_conf->verify_alg)) {
1955                 retcode = ERR_VERIFY_RUNNING;
1956                 goto fail;
1957         }
1958
1959         retcode = alloc_crypto(&crypto, new_conf);
1960         if (retcode != NO_ERROR)
1961                 goto fail;
1962
1963         rcu_assign_pointer(tconn->net_conf, new_conf);
1964
1965         if (!rsr) {
1966                 crypto_free_hash(tconn->csums_tfm);
1967                 tconn->csums_tfm = crypto.csums_tfm;
1968                 crypto.csums_tfm = NULL;
1969         }
1970         if (!ovr) {
1971                 crypto_free_hash(tconn->verify_tfm);
1972                 tconn->verify_tfm = crypto.verify_tfm;
1973                 crypto.verify_tfm = NULL;
1974         }
1975
1976         kfree(tconn->int_dig_in);
1977         tconn->int_dig_in = crypto.int_dig_in;
1978         kfree(tconn->int_dig_vv);
1979         tconn->int_dig_vv = crypto.int_dig_vv;
1980         crypto_free_hash(tconn->integrity_tfm);
1981         tconn->integrity_tfm = crypto.integrity_tfm;
1982         if (tconn->cstate >= C_WF_REPORT_PARAMS && tconn->agreed_pro_version >= 100)
1983                 /* Do this without trying to take tconn->data.mutex again.  */
1984                 __drbd_send_protocol(tconn, P_PROTOCOL_UPDATE);
1985
1986         crypto_free_hash(tconn->cram_hmac_tfm);
1987         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
1988
1989         mutex_unlock(&tconn->conf_update);
1990         mutex_unlock(&tconn->data.mutex);
1991         synchronize_rcu();
1992         kfree(old_conf);
1993
1994         if (tconn->cstate >= C_WF_REPORT_PARAMS)
1995                 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn)));
1996
1997         goto done;
1998
1999  fail:
2000         mutex_unlock(&tconn->conf_update);
2001         mutex_unlock(&tconn->data.mutex);
2002         free_crypto(&crypto);
2003         kfree(new_conf);
2004  done:
2005         conn_reconfig_done(tconn);
2006  out:
2007         drbd_adm_finish(info, retcode);
2008         return 0;
2009 }
2010
2011 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2012 {
2013         struct drbd_conf *mdev;
2014         struct net_conf *old_conf, *new_conf = NULL;
2015         struct crypto crypto = { };
2016         struct drbd_tconn *tconn;
2017         enum drbd_ret_code retcode;
2018         int i;
2019         int err;
2020
2021         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
2022
2023         if (!adm_ctx.reply_skb)
2024                 return retcode;
2025         if (retcode != NO_ERROR)
2026                 goto out;
2027         if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2028                 drbd_msg_put_info("connection endpoint(s) missing");
2029                 retcode = ERR_INVALID_REQUEST;
2030                 goto out;
2031         }
2032
2033         /* No need for _rcu here. All reconfiguration is
2034          * strictly serialized on genl_lock(). We are protected against
2035          * concurrent reconfiguration/addition/deletion */
2036         list_for_each_entry(tconn, &drbd_tconns, all_tconn) {
2037                 if (nla_len(adm_ctx.my_addr) == tconn->my_addr_len &&
2038                     !memcmp(nla_data(adm_ctx.my_addr), &tconn->my_addr, tconn->my_addr_len)) {
2039                         retcode = ERR_LOCAL_ADDR;
2040                         goto out;
2041                 }
2042
2043                 if (nla_len(adm_ctx.peer_addr) == tconn->peer_addr_len &&
2044                     !memcmp(nla_data(adm_ctx.peer_addr), &tconn->peer_addr, tconn->peer_addr_len)) {
2045                         retcode = ERR_PEER_ADDR;
2046                         goto out;
2047                 }
2048         }
2049
2050         tconn = adm_ctx.tconn;
2051         conn_reconfig_start(tconn);
2052
2053         if (tconn->cstate > C_STANDALONE) {
2054                 retcode = ERR_NET_CONFIGURED;
2055                 goto fail;
2056         }
2057
2058         /* allocation not in the IO path, cqueue thread context */
2059         new_conf = kzalloc(sizeof(*new_conf), GFP_KERNEL);
2060         if (!new_conf) {
2061                 retcode = ERR_NOMEM;
2062                 goto fail;
2063         }
2064
2065         set_net_conf_defaults(new_conf);
2066
2067         err = net_conf_from_attrs(new_conf, info);
2068         if (err) {
2069                 retcode = ERR_MANDATORY_TAG;
2070                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2071                 goto fail;
2072         }
2073
2074         retcode = check_net_options(tconn, new_conf);
2075         if (retcode != NO_ERROR)
2076                 goto fail;
2077
2078         retcode = alloc_crypto(&crypto, new_conf);
2079         if (retcode != NO_ERROR)
2080                 goto fail;
2081
2082         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2083
2084         conn_flush_workqueue(tconn);
2085
2086         mutex_lock(&tconn->conf_update);
2087         old_conf = tconn->net_conf;
2088         if (old_conf) {
2089                 retcode = ERR_NET_CONFIGURED;
2090                 mutex_unlock(&tconn->conf_update);
2091                 goto fail;
2092         }
2093         rcu_assign_pointer(tconn->net_conf, new_conf);
2094
2095         conn_free_crypto(tconn);
2096         tconn->int_dig_in = crypto.int_dig_in;
2097         tconn->int_dig_vv = crypto.int_dig_vv;
2098         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
2099         tconn->integrity_tfm = crypto.integrity_tfm;
2100         tconn->csums_tfm = crypto.csums_tfm;
2101         tconn->verify_tfm = crypto.verify_tfm;
2102
2103         tconn->my_addr_len = nla_len(adm_ctx.my_addr);
2104         memcpy(&tconn->my_addr, nla_data(adm_ctx.my_addr), tconn->my_addr_len);
2105         tconn->peer_addr_len = nla_len(adm_ctx.peer_addr);
2106         memcpy(&tconn->peer_addr, nla_data(adm_ctx.peer_addr), tconn->peer_addr_len);
2107
2108         mutex_unlock(&tconn->conf_update);
2109
2110         rcu_read_lock();
2111         idr_for_each_entry(&tconn->volumes, mdev, i) {
2112                 mdev->send_cnt = 0;
2113                 mdev->recv_cnt = 0;
2114         }
2115         rcu_read_unlock();
2116
2117         retcode = conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2118
2119         conn_reconfig_done(tconn);
2120         drbd_adm_finish(info, retcode);
2121         return 0;
2122
2123 fail:
2124         free_crypto(&crypto);
2125         kfree(new_conf);
2126
2127         conn_reconfig_done(tconn);
2128 out:
2129         drbd_adm_finish(info, retcode);
2130         return 0;
2131 }
2132
2133 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
2134 {
2135         enum drbd_state_rv rv;
2136
2137         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2138                         force ? CS_HARD : 0);
2139
2140         switch (rv) {
2141         case SS_NOTHING_TO_DO:
2142                 break;
2143         case SS_ALREADY_STANDALONE:
2144                 return SS_SUCCESS;
2145         case SS_PRIMARY_NOP:
2146                 /* Our state checking code wants to see the peer outdated. */
2147                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2148                                                 pdsk, D_OUTDATED), CS_VERBOSE);
2149                 break;
2150         case SS_CW_FAILED_BY_PEER:
2151                 /* The peer probably wants to see us outdated. */
2152                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2153                                                         disk, D_OUTDATED), 0);
2154                 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2155                         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2156                                         CS_HARD);
2157                 }
2158                 break;
2159         default:;
2160                 /* no special handling necessary */
2161         }
2162
2163         if (rv >= SS_SUCCESS) {
2164                 enum drbd_state_rv rv2;
2165                 /* No one else can reconfigure the network while I am here.
2166                  * The state handling only uses drbd_thread_stop_nowait(),
2167                  * we want to really wait here until the receiver is no more.
2168                  */
2169                 drbd_thread_stop(&adm_ctx.tconn->receiver);
2170
2171                 /* Race breaker.  This additional state change request may be
2172                  * necessary, if this was a forced disconnect during a receiver
2173                  * restart.  We may have "killed" the receiver thread just
2174                  * after drbdd_init() returned.  Typically, we should be
2175                  * C_STANDALONE already, now, and this becomes a no-op.
2176                  */
2177                 rv2 = conn_request_state(tconn, NS(conn, C_STANDALONE),
2178                                 CS_VERBOSE | CS_HARD);
2179                 if (rv2 < SS_SUCCESS)
2180                         conn_err(tconn,
2181                                 "unexpected rv2=%d in conn_try_disconnect()\n",
2182                                 rv2);
2183         }
2184         return rv;
2185 }
2186
2187 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2188 {
2189         struct disconnect_parms parms;
2190         struct drbd_tconn *tconn;
2191         enum drbd_state_rv rv;
2192         enum drbd_ret_code retcode;
2193         int err;
2194
2195         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONNECTION);
2196         if (!adm_ctx.reply_skb)
2197                 return retcode;
2198         if (retcode != NO_ERROR)
2199                 goto fail;
2200
2201         tconn = adm_ctx.tconn;
2202         memset(&parms, 0, sizeof(parms));
2203         if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2204                 err = disconnect_parms_from_attrs(&parms, info);
2205                 if (err) {
2206                         retcode = ERR_MANDATORY_TAG;
2207                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2208                         goto fail;
2209                 }
2210         }
2211
2212         rv = conn_try_disconnect(tconn, parms.force_disconnect);
2213         if (rv < SS_SUCCESS)
2214                 retcode = rv;  /* FIXME: Type mismatch. */
2215         else
2216                 retcode = NO_ERROR;
2217  fail:
2218         drbd_adm_finish(info, retcode);
2219         return 0;
2220 }
2221
2222 void resync_after_online_grow(struct drbd_conf *mdev)
2223 {
2224         int iass; /* I am sync source */
2225
2226         dev_info(DEV, "Resync of new storage after online grow\n");
2227         if (mdev->state.role != mdev->state.peer)
2228                 iass = (mdev->state.role == R_PRIMARY);
2229         else
2230                 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
2231
2232         if (iass)
2233                 drbd_start_resync(mdev, C_SYNC_SOURCE);
2234         else
2235                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2236 }
2237
2238 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2239 {
2240         struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2241         struct resize_parms rs;
2242         struct drbd_conf *mdev;
2243         enum drbd_ret_code retcode;
2244         enum determine_dev_size dd;
2245         enum dds_flags ddsf;
2246         sector_t u_size;
2247         int err;
2248
2249         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2250         if (!adm_ctx.reply_skb)
2251                 return retcode;
2252         if (retcode != NO_ERROR)
2253                 goto fail;
2254
2255         memset(&rs, 0, sizeof(struct resize_parms));
2256         if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2257                 err = resize_parms_from_attrs(&rs, info);
2258                 if (err) {
2259                         retcode = ERR_MANDATORY_TAG;
2260                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2261                         goto fail;
2262                 }
2263         }
2264
2265         mdev = adm_ctx.mdev;
2266         if (mdev->state.conn > C_CONNECTED) {
2267                 retcode = ERR_RESIZE_RESYNC;
2268                 goto fail;
2269         }
2270
2271         if (mdev->state.role == R_SECONDARY &&
2272             mdev->state.peer == R_SECONDARY) {
2273                 retcode = ERR_NO_PRIMARY;
2274                 goto fail;
2275         }
2276
2277         if (!get_ldev(mdev)) {
2278                 retcode = ERR_NO_DISK;
2279                 goto fail;
2280         }
2281
2282         if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
2283                 retcode = ERR_NEED_APV_93;
2284                 goto fail;
2285         }
2286
2287         rcu_read_lock();
2288         u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
2289         rcu_read_unlock();
2290         if (u_size != (sector_t)rs.resize_size) {
2291                 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2292                 if (!new_disk_conf) {
2293                         retcode = ERR_NOMEM;
2294                         goto fail;
2295                 }
2296         }
2297
2298         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
2299                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2300
2301         if (new_disk_conf) {
2302                 mutex_lock(&mdev->tconn->conf_update);
2303                 old_disk_conf = mdev->ldev->disk_conf;
2304                 *new_disk_conf = *old_disk_conf;
2305                 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2306                 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
2307                 mutex_unlock(&mdev->tconn->conf_update);
2308                 synchronize_rcu();
2309                 kfree(old_disk_conf);
2310         }
2311
2312         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2313         dd = drbd_determine_dev_size(mdev, ddsf);
2314         drbd_md_sync(mdev);
2315         put_ldev(mdev);
2316         if (dd == dev_size_error) {
2317                 retcode = ERR_NOMEM_BITMAP;
2318                 goto fail;
2319         }
2320
2321         if (mdev->state.conn == C_CONNECTED) {
2322                 if (dd == grew)
2323                         set_bit(RESIZE_PENDING, &mdev->flags);
2324
2325                 drbd_send_uuids(mdev);
2326                 drbd_send_sizes(mdev, 1, ddsf);
2327         }
2328
2329  fail:
2330         drbd_adm_finish(info, retcode);
2331         return 0;
2332 }
2333
2334 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2335 {
2336         enum drbd_ret_code retcode;
2337         struct drbd_tconn *tconn;
2338         struct res_opts res_opts;
2339         int err;
2340
2341         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
2342         if (!adm_ctx.reply_skb)
2343                 return retcode;
2344         if (retcode != NO_ERROR)
2345                 goto fail;
2346         tconn = adm_ctx.tconn;
2347
2348         res_opts = tconn->res_opts;
2349         if (should_set_defaults(info))
2350                 set_res_opts_defaults(&res_opts);
2351
2352         err = res_opts_from_attrs(&res_opts, info);
2353         if (err && err != -ENOMSG) {
2354                 retcode = ERR_MANDATORY_TAG;
2355                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2356                 goto fail;
2357         }
2358
2359         err = set_resource_options(tconn, &res_opts);
2360         if (err) {
2361                 retcode = ERR_INVALID_REQUEST;
2362                 if (err == -ENOMEM)
2363                         retcode = ERR_NOMEM;
2364         }
2365
2366 fail:
2367         drbd_adm_finish(info, retcode);
2368         return 0;
2369 }
2370
2371 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2372 {
2373         struct drbd_conf *mdev;
2374         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2375
2376         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2377         if (!adm_ctx.reply_skb)
2378                 return retcode;
2379         if (retcode != NO_ERROR)
2380                 goto out;
2381
2382         mdev = adm_ctx.mdev;
2383
2384         /* If there is still bitmap IO pending, probably because of a previous
2385          * resync just being finished, wait for it before requesting a new resync. */
2386         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2387
2388         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2389
2390         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2391                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2392
2393         while (retcode == SS_NEED_CONNECTION) {
2394                 spin_lock_irq(&mdev->tconn->req_lock);
2395                 if (mdev->state.conn < C_CONNECTED)
2396                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2397                 spin_unlock_irq(&mdev->tconn->req_lock);
2398
2399                 if (retcode != SS_NEED_CONNECTION)
2400                         break;
2401
2402                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2403         }
2404
2405 out:
2406         drbd_adm_finish(info, retcode);
2407         return 0;
2408 }
2409
2410 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2411 {
2412         int rv;
2413
2414         rv = drbd_bmio_set_n_write(mdev);
2415         drbd_suspend_al(mdev);
2416         return rv;
2417 }
2418
2419 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2420                 union drbd_state mask, union drbd_state val)
2421 {
2422         enum drbd_ret_code retcode;
2423
2424         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2425         if (!adm_ctx.reply_skb)
2426                 return retcode;
2427         if (retcode != NO_ERROR)
2428                 goto out;
2429
2430         retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2431 out:
2432         drbd_adm_finish(info, retcode);
2433         return 0;
2434 }
2435
2436 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2437 {
2438         return drbd_adm_simple_request_state(skb, info, NS(conn, C_STARTING_SYNC_S));
2439 }
2440
2441 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2442 {
2443         enum drbd_ret_code retcode;
2444
2445         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2446         if (!adm_ctx.reply_skb)
2447                 return retcode;
2448         if (retcode != NO_ERROR)
2449                 goto out;
2450
2451         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2452                 retcode = ERR_PAUSE_IS_SET;
2453 out:
2454         drbd_adm_finish(info, retcode);
2455         return 0;
2456 }
2457
2458 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2459 {
2460         union drbd_dev_state s;
2461         enum drbd_ret_code retcode;
2462
2463         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2464         if (!adm_ctx.reply_skb)
2465                 return retcode;
2466         if (retcode != NO_ERROR)
2467                 goto out;
2468
2469         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2470                 s = adm_ctx.mdev->state;
2471                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2472                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2473                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2474                 } else {
2475                         retcode = ERR_PAUSE_IS_CLEAR;
2476                 }
2477         }
2478
2479 out:
2480         drbd_adm_finish(info, retcode);
2481         return 0;
2482 }
2483
2484 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2485 {
2486         return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2487 }
2488
2489 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2490 {
2491         struct drbd_conf *mdev;
2492         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2493
2494         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2495         if (!adm_ctx.reply_skb)
2496                 return retcode;
2497         if (retcode != NO_ERROR)
2498                 goto out;
2499
2500         mdev = adm_ctx.mdev;
2501         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2502                 drbd_uuid_new_current(mdev);
2503                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2504         }
2505         drbd_suspend_io(mdev);
2506         retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2507         if (retcode == SS_SUCCESS) {
2508                 if (mdev->state.conn < C_CONNECTED)
2509                         tl_clear(mdev->tconn);
2510                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2511                         tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2512         }
2513         drbd_resume_io(mdev);
2514
2515 out:
2516         drbd_adm_finish(info, retcode);
2517         return 0;
2518 }
2519
2520 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2521 {
2522         return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2523 }
2524
2525 int nla_put_drbd_cfg_context(struct sk_buff *skb, struct drbd_tconn *tconn, unsigned vnr)
2526 {
2527         struct nlattr *nla;
2528         nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2529         if (!nla)
2530                 goto nla_put_failure;
2531         if (vnr != VOLUME_UNSPECIFIED)
2532                 NLA_PUT_U32(skb, T_ctx_volume, vnr);
2533         NLA_PUT_STRING(skb, T_ctx_resource_name, tconn->name);
2534         if (tconn->my_addr_len)
2535                 NLA_PUT(skb, T_ctx_my_addr, tconn->my_addr_len, &tconn->my_addr);
2536         if (tconn->peer_addr_len)
2537                 NLA_PUT(skb, T_ctx_peer_addr, tconn->peer_addr_len, &tconn->peer_addr);
2538         nla_nest_end(skb, nla);
2539         return 0;
2540
2541 nla_put_failure:
2542         if (nla)
2543                 nla_nest_cancel(skb, nla);
2544         return -EMSGSIZE;
2545 }
2546
2547 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2548                 const struct sib_info *sib)
2549 {
2550         struct state_info *si = NULL; /* for sizeof(si->member); */
2551         struct net_conf *nc;
2552         struct nlattr *nla;
2553         int got_ldev;
2554         int err = 0;
2555         int exclude_sensitive;
2556
2557         /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2558          * to.  So we better exclude_sensitive information.
2559          *
2560          * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2561          * in the context of the requesting user process. Exclude sensitive
2562          * information, unless current has superuser.
2563          *
2564          * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2565          * relies on the current implementation of netlink_dump(), which
2566          * executes the dump callback successively from netlink_recvmsg(),
2567          * always in the context of the receiving process */
2568         exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2569
2570         got_ldev = get_ldev(mdev);
2571
2572         /* We need to add connection name and volume number information still.
2573          * Minor number is in drbd_genlmsghdr. */
2574         if (nla_put_drbd_cfg_context(skb, mdev->tconn, mdev->vnr))
2575                 goto nla_put_failure;
2576
2577         if (res_opts_to_skb(skb, &mdev->tconn->res_opts, exclude_sensitive))
2578                 goto nla_put_failure;
2579
2580         rcu_read_lock();
2581         if (got_ldev)
2582                 if (disk_conf_to_skb(skb, rcu_dereference(mdev->ldev->disk_conf), exclude_sensitive))
2583                         goto nla_put_failure;
2584
2585         nc = rcu_dereference(mdev->tconn->net_conf);
2586         if (nc)
2587                 err = net_conf_to_skb(skb, nc, exclude_sensitive);
2588         rcu_read_unlock();
2589         if (err)
2590                 goto nla_put_failure;
2591
2592         nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2593         if (!nla)
2594                 goto nla_put_failure;
2595         NLA_PUT_U32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY);
2596         NLA_PUT_U32(skb, T_current_state, mdev->state.i);
2597         NLA_PUT_U64(skb, T_ed_uuid, mdev->ed_uuid);
2598         NLA_PUT_U64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev));
2599
2600         if (got_ldev) {
2601                 NLA_PUT_U32(skb, T_disk_flags, mdev->ldev->md.flags);
2602                 NLA_PUT(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2603                 NLA_PUT_U64(skb, T_bits_total, drbd_bm_bits(mdev));
2604                 NLA_PUT_U64(skb, T_bits_oos, drbd_bm_total_weight(mdev));
2605                 if (C_SYNC_SOURCE <= mdev->state.conn &&
2606                     C_PAUSED_SYNC_T >= mdev->state.conn) {
2607                         NLA_PUT_U64(skb, T_bits_rs_total, mdev->rs_total);
2608                         NLA_PUT_U64(skb, T_bits_rs_failed, mdev->rs_failed);
2609                 }
2610         }
2611
2612         if (sib) {
2613                 switch(sib->sib_reason) {
2614                 case SIB_SYNC_PROGRESS:
2615                 case SIB_GET_STATUS_REPLY:
2616                         break;
2617                 case SIB_STATE_CHANGE:
2618                         NLA_PUT_U32(skb, T_prev_state, sib->os.i);
2619                         NLA_PUT_U32(skb, T_new_state, sib->ns.i);
2620                         break;
2621                 case SIB_HELPER_POST:
2622                         NLA_PUT_U32(skb,
2623                                 T_helper_exit_code, sib->helper_exit_code);
2624                         /* fall through */
2625                 case SIB_HELPER_PRE:
2626                         NLA_PUT_STRING(skb, T_helper, sib->helper_name);
2627                         break;
2628                 }
2629         }
2630         nla_nest_end(skb, nla);
2631
2632         if (0)
2633 nla_put_failure:
2634                 err = -EMSGSIZE;
2635         if (got_ldev)
2636                 put_ldev(mdev);
2637         return err;
2638 }
2639
2640 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2641 {
2642         enum drbd_ret_code retcode;
2643         int err;
2644
2645         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2646         if (!adm_ctx.reply_skb)
2647                 return retcode;
2648         if (retcode != NO_ERROR)
2649                 goto out;
2650
2651         err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2652         if (err) {
2653                 nlmsg_free(adm_ctx.reply_skb);
2654                 return err;
2655         }
2656 out:
2657         drbd_adm_finish(info, retcode);
2658         return 0;
2659 }
2660
2661 int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
2662 {
2663         struct drbd_conf *mdev;
2664         struct drbd_genlmsghdr *dh;
2665         struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2666         struct drbd_tconn *tconn = NULL;
2667         struct drbd_tconn *tmp;
2668         unsigned volume = cb->args[1];
2669
2670         /* Open coded, deferred, iteration:
2671          * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2672          *      idr_for_each_entry(&tconn->volumes, mdev, i) {
2673          *        ...
2674          *      }
2675          * }
2676          * where tconn is cb->args[0];
2677          * and i is cb->args[1];
2678          *
2679          * cb->args[2] indicates if we shall loop over all resources,
2680          * or just dump all volumes of a single resource.
2681          *
2682          * This may miss entries inserted after this dump started,
2683          * or entries deleted before they are reached.
2684          *
2685          * We need to make sure the mdev won't disappear while
2686          * we are looking at it, and revalidate our iterators
2687          * on each iteration.
2688          */
2689
2690         /* synchronize with conn_create()/conn_destroy() */
2691         rcu_read_lock();
2692         /* revalidate iterator position */
2693         list_for_each_entry_rcu(tmp, &drbd_tconns, all_tconn) {
2694                 if (pos == NULL) {
2695                         /* first iteration */
2696                         pos = tmp;
2697                         tconn = pos;
2698                         break;
2699                 }
2700                 if (tmp == pos) {
2701                         tconn = pos;
2702                         break;
2703                 }
2704         }
2705         if (tconn) {
2706 next_tconn:
2707                 mdev = idr_get_next(&tconn->volumes, &volume);
2708                 if (!mdev) {
2709                         /* No more volumes to dump on this tconn.
2710                          * Advance tconn iterator. */
2711                         pos = list_entry_rcu(tconn->all_tconn.next,
2712                                              struct drbd_tconn, all_tconn);
2713                         /* Did we dump any volume on this tconn yet? */
2714                         if (volume != 0) {
2715                                 /* If we reached the end of the list,
2716                                  * or only a single resource dump was requested,
2717                                  * we are done. */
2718                                 if (&pos->all_tconn == &drbd_tconns || cb->args[2])
2719                                         goto out;
2720                                 volume = 0;
2721                                 tconn = pos;
2722                                 goto next_tconn;
2723                         }
2724                 }
2725
2726                 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).pid,
2727                                 cb->nlh->nlmsg_seq, &drbd_genl_family,
2728                                 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2729                 if (!dh)
2730                         goto out;
2731
2732                 if (!mdev) {
2733                         /* this is a tconn without a single volume */
2734                         dh->minor = -1U;
2735                         dh->ret_code = NO_ERROR;
2736                         if (nla_put_drbd_cfg_context(skb, tconn, VOLUME_UNSPECIFIED))
2737                                 genlmsg_cancel(skb, dh);
2738                         else
2739                                 genlmsg_end(skb, dh);
2740                         goto out;
2741                 }
2742
2743                 D_ASSERT(mdev->vnr == volume);
2744                 D_ASSERT(mdev->tconn == tconn);
2745
2746                 dh->minor = mdev_to_minor(mdev);
2747                 dh->ret_code = NO_ERROR;
2748
2749                 if (nla_put_status_info(skb, mdev, NULL)) {
2750                         genlmsg_cancel(skb, dh);
2751                         goto out;
2752                 }
2753                 genlmsg_end(skb, dh);
2754         }
2755
2756 out:
2757         rcu_read_unlock();
2758         /* where to start the next iteration */
2759         cb->args[0] = (long)pos;
2760         cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2761
2762         /* No more tconns/volumes/minors found results in an empty skb.
2763          * Which will terminate the dump. */
2764         return skb->len;
2765 }
2766
2767 /*
2768  * Request status of all resources, or of all volumes within a single resource.
2769  *
2770  * This is a dump, as the answer may not fit in a single reply skb otherwise.
2771  * Which means we cannot use the family->attrbuf or other such members, because
2772  * dump is NOT protected by the genl_lock().  During dump, we only have access
2773  * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
2774  *
2775  * Once things are setup properly, we call into get_one_status().
2776  */
2777 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2778 {
2779         const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
2780         struct nlattr *nla;
2781         const char *resource_name;
2782         struct drbd_tconn *tconn;
2783         int maxtype;
2784
2785         /* Is this a followup call? */
2786         if (cb->args[0]) {
2787                 /* ... of a single resource dump,
2788                  * and the resource iterator has been advanced already? */
2789                 if (cb->args[2] && cb->args[2] != cb->args[0])
2790                         return 0; /* DONE. */
2791                 goto dump;
2792         }
2793
2794         /* First call (from netlink_dump_start).  We need to figure out
2795          * which resource(s) the user wants us to dump. */
2796         nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
2797                         nlmsg_attrlen(cb->nlh, hdrlen),
2798                         DRBD_NLA_CFG_CONTEXT);
2799
2800         /* No explicit context given.  Dump all. */
2801         if (!nla)
2802                 goto dump;
2803         maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
2804         nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
2805         if (IS_ERR(nla))
2806                 return PTR_ERR(nla);
2807         /* context given, but no name present? */
2808         if (!nla)
2809                 return -EINVAL;
2810         resource_name = nla_data(nla);
2811         tconn = conn_get_by_name(resource_name);
2812
2813         if (!tconn)
2814                 return -ENODEV;
2815
2816         kref_put(&tconn->kref, &conn_destroy); /* get_one_status() (re)validates tconn by itself */
2817
2818         /* prime iterators, and set "filter" mode mark:
2819          * only dump this tconn. */
2820         cb->args[0] = (long)tconn;
2821         /* cb->args[1] = 0; passed in this way. */
2822         cb->args[2] = (long)tconn;
2823
2824 dump:
2825         return get_one_status(skb, cb);
2826 }
2827
2828 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2829 {
2830         enum drbd_ret_code retcode;
2831         struct timeout_parms tp;
2832         int err;
2833
2834         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2835         if (!adm_ctx.reply_skb)
2836                 return retcode;
2837         if (retcode != NO_ERROR)
2838                 goto out;
2839
2840         tp.timeout_type =
2841                 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2842                 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2843                 UT_DEFAULT;
2844
2845         err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2846         if (err) {
2847                 nlmsg_free(adm_ctx.reply_skb);
2848                 return err;
2849         }
2850 out:
2851         drbd_adm_finish(info, retcode);
2852         return 0;
2853 }
2854
2855 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2856 {
2857         struct drbd_conf *mdev;
2858         enum drbd_ret_code retcode;
2859
2860         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2861         if (!adm_ctx.reply_skb)
2862                 return retcode;
2863         if (retcode != NO_ERROR)
2864                 goto out;
2865
2866         mdev = adm_ctx.mdev;
2867         if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2868                 /* resume from last known position, if possible */
2869                 struct start_ov_parms parms =
2870                         { .ov_start_sector = mdev->ov_start_sector };
2871                 int err = start_ov_parms_from_attrs(&parms, info);
2872                 if (err) {
2873                         retcode = ERR_MANDATORY_TAG;
2874                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2875                         goto out;
2876                 }
2877                 /* w_make_ov_request expects position to be aligned */
2878                 mdev->ov_start_sector = parms.ov_start_sector & ~BM_SECT_PER_BIT;
2879         }
2880         /* If there is still bitmap IO pending, e.g. previous resync or verify
2881          * just being finished, wait for it before requesting a new resync. */
2882         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2883         retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2884 out:
2885         drbd_adm_finish(info, retcode);
2886         return 0;
2887 }
2888
2889
2890 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
2891 {
2892         struct drbd_conf *mdev;
2893         enum drbd_ret_code retcode;
2894         int skip_initial_sync = 0;
2895         int err;
2896         struct new_c_uuid_parms args;
2897
2898         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2899         if (!adm_ctx.reply_skb)
2900                 return retcode;
2901         if (retcode != NO_ERROR)
2902                 goto out_nolock;
2903
2904         mdev = adm_ctx.mdev;
2905         memset(&args, 0, sizeof(args));
2906         if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
2907                 err = new_c_uuid_parms_from_attrs(&args, info);
2908                 if (err) {
2909                         retcode = ERR_MANDATORY_TAG;
2910                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2911                         goto out_nolock;
2912                 }
2913         }
2914
2915         mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2916
2917         if (!get_ldev(mdev)) {
2918                 retcode = ERR_NO_DISK;
2919                 goto out;
2920         }
2921
2922         /* this is "skip initial sync", assume to be clean */
2923         if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2924             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2925                 dev_info(DEV, "Preparing to skip initial sync\n");
2926                 skip_initial_sync = 1;
2927         } else if (mdev->state.conn != C_STANDALONE) {
2928                 retcode = ERR_CONNECTED;
2929                 goto out_dec;
2930         }
2931
2932         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2933         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2934
2935         if (args.clear_bm) {
2936                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2937                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2938                 if (err) {
2939                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2940                         retcode = ERR_IO_MD_DISK;
2941                 }
2942                 if (skip_initial_sync) {
2943                         drbd_send_uuids_skip_initial_sync(mdev);
2944                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2945                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2946                         spin_lock_irq(&mdev->tconn->req_lock);
2947                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2948                                         CS_VERBOSE, NULL);
2949                         spin_unlock_irq(&mdev->tconn->req_lock);
2950                 }
2951         }
2952
2953         drbd_md_sync(mdev);
2954 out_dec:
2955         put_ldev(mdev);
2956 out:
2957         mutex_unlock(mdev->state_mutex);
2958 out_nolock:
2959         drbd_adm_finish(info, retcode);
2960         return 0;
2961 }
2962
2963 static enum drbd_ret_code
2964 drbd_check_resource_name(const char *name)
2965 {
2966         if (!name || !name[0]) {
2967                 drbd_msg_put_info("resource name missing");
2968                 return ERR_MANDATORY_TAG;
2969         }
2970         /* if we want to use these in sysfs/configfs/debugfs some day,
2971          * we must not allow slashes */
2972         if (strchr(name, '/')) {
2973                 drbd_msg_put_info("invalid resource name");
2974                 return ERR_INVALID_REQUEST;
2975         }
2976         return NO_ERROR;
2977 }
2978
2979 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
2980 {
2981         enum drbd_ret_code retcode;
2982         struct res_opts res_opts;
2983         int err;
2984
2985         retcode = drbd_adm_prepare(skb, info, 0);
2986         if (!adm_ctx.reply_skb)
2987                 return retcode;
2988         if (retcode != NO_ERROR)
2989                 goto out;
2990
2991         set_res_opts_defaults(&res_opts);
2992         err = res_opts_from_attrs(&res_opts, info);
2993         if (err && err != -ENOMSG) {
2994                 retcode = ERR_MANDATORY_TAG;
2995                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2996                 goto out;
2997         }
2998
2999         retcode = drbd_check_resource_name(adm_ctx.resource_name);
3000         if (retcode != NO_ERROR)
3001                 goto out;
3002
3003         if (adm_ctx.tconn) {
3004                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
3005                         retcode = ERR_INVALID_REQUEST;
3006                         drbd_msg_put_info("resource exists");
3007                 }
3008                 /* else: still NO_ERROR */
3009                 goto out;
3010         }
3011
3012         if (!conn_create(adm_ctx.resource_name, &res_opts))
3013                 retcode = ERR_NOMEM;
3014 out:
3015         drbd_adm_finish(info, retcode);
3016         return 0;
3017 }
3018
3019 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
3020 {
3021         struct drbd_genlmsghdr *dh = info->userhdr;
3022         enum drbd_ret_code retcode;
3023
3024         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
3025         if (!adm_ctx.reply_skb)
3026                 return retcode;
3027         if (retcode != NO_ERROR)
3028                 goto out;
3029
3030         /* FIXME drop minor_count parameter, limit to MINORMASK */
3031         if (dh->minor >= minor_count) {
3032                 drbd_msg_put_info("requested minor out of range");
3033                 retcode = ERR_INVALID_REQUEST;
3034                 goto out;
3035         }
3036         if (adm_ctx.volume > DRBD_VOLUME_MAX) {
3037                 drbd_msg_put_info("requested volume id out of range");
3038                 retcode = ERR_INVALID_REQUEST;
3039                 goto out;
3040         }
3041
3042         /* drbd_adm_prepare made sure already
3043          * that mdev->tconn and mdev->vnr match the request. */
3044         if (adm_ctx.mdev) {
3045                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
3046                         retcode = ERR_MINOR_EXISTS;
3047                 /* else: still NO_ERROR */
3048                 goto out;
3049         }
3050
3051         retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
3052 out:
3053         drbd_adm_finish(info, retcode);
3054         return 0;
3055 }
3056
3057 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
3058 {
3059         if (mdev->state.disk == D_DISKLESS &&
3060             /* no need to be mdev->state.conn == C_STANDALONE &&
3061              * we may want to delete a minor from a live replication group.
3062              */
3063             mdev->state.role == R_SECONDARY) {
3064                 _drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS),
3065                                     CS_VERBOSE + CS_WAIT_COMPLETE);
3066                 idr_remove(&mdev->tconn->volumes, mdev->vnr);
3067                 idr_remove(&minors, mdev_to_minor(mdev));
3068                 del_gendisk(mdev->vdisk);
3069                 synchronize_rcu();
3070                 kref_put(&mdev->kref, &drbd_minor_destroy);
3071                 return NO_ERROR;
3072         } else
3073                 return ERR_MINOR_CONFIGURED;
3074 }
3075
3076 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
3077 {
3078         enum drbd_ret_code retcode;
3079
3080         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3081         if (!adm_ctx.reply_skb)
3082                 return retcode;
3083         if (retcode != NO_ERROR)
3084                 goto out;
3085
3086         retcode = adm_delete_minor(adm_ctx.mdev);
3087 out:
3088         drbd_adm_finish(info, retcode);
3089         return 0;
3090 }
3091
3092 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
3093 {
3094         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3095         struct drbd_conf *mdev;
3096         unsigned i;
3097
3098         retcode = drbd_adm_prepare(skb, info, 0);
3099         if (!adm_ctx.reply_skb)
3100                 return retcode;
3101         if (retcode != NO_ERROR)
3102                 goto out;
3103
3104         if (!adm_ctx.tconn) {
3105                 retcode = ERR_RES_NOT_KNOWN;
3106                 goto out;
3107         }
3108
3109         /* demote */
3110         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3111                 retcode = drbd_set_role(mdev, R_SECONDARY, 0);
3112                 if (retcode < SS_SUCCESS) {
3113                         drbd_msg_put_info("failed to demote");
3114                         goto out;
3115                 }
3116         }
3117
3118         retcode = conn_try_disconnect(adm_ctx.tconn, 0);
3119         if (retcode < SS_SUCCESS) {
3120                 drbd_msg_put_info("failed to disconnect");
3121                 goto out;
3122         }
3123
3124         /* detach */
3125         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3126                 retcode = adm_detach(mdev, 0);
3127                 if (retcode < SS_SUCCESS) {
3128                         drbd_msg_put_info("failed to detach");
3129                         goto out;
3130                 }
3131         }
3132
3133         /* If we reach this, all volumes (of this tconn) are Secondary,
3134          * Disconnected, Diskless, aka Unconfigured. Make sure all threads have
3135          * actually stopped, state handling only does drbd_thread_stop_nowait(). */
3136         drbd_thread_stop(&adm_ctx.tconn->worker);
3137
3138         /* Now, nothing can fail anymore */
3139
3140         /* delete volumes */
3141         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3142                 retcode = adm_delete_minor(mdev);
3143                 if (retcode != NO_ERROR) {
3144                         /* "can not happen" */
3145                         drbd_msg_put_info("failed to delete volume");
3146                         goto out;
3147                 }
3148         }
3149
3150         /* delete connection */
3151         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3152                 list_del_rcu(&adm_ctx.tconn->all_tconn);
3153                 synchronize_rcu();
3154                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3155
3156                 retcode = NO_ERROR;
3157         } else {
3158                 /* "can not happen" */
3159                 retcode = ERR_RES_IN_USE;
3160                 drbd_msg_put_info("failed to delete connection");
3161         }
3162         goto out;
3163 out:
3164         drbd_adm_finish(info, retcode);
3165         return 0;
3166 }
3167
3168 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
3169 {
3170         enum drbd_ret_code retcode;
3171
3172         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
3173         if (!adm_ctx.reply_skb)
3174                 return retcode;
3175         if (retcode != NO_ERROR)
3176                 goto out;
3177
3178         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3179                 list_del_rcu(&adm_ctx.tconn->all_tconn);
3180                 synchronize_rcu();
3181                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3182
3183                 retcode = NO_ERROR;
3184         } else {
3185                 retcode = ERR_RES_IN_USE;
3186         }
3187
3188         if (retcode == NO_ERROR)
3189                 drbd_thread_stop(&adm_ctx.tconn->worker);
3190 out:
3191         drbd_adm_finish(info, retcode);
3192         return 0;
3193 }
3194
3195 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
3196 {
3197         static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3198         struct sk_buff *msg;
3199         struct drbd_genlmsghdr *d_out;
3200         unsigned seq;
3201         int err = -ENOMEM;
3202
3203         seq = atomic_inc_return(&drbd_genl_seq);
3204         msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3205         if (!msg)
3206                 goto failed;
3207
3208         err = -EMSGSIZE;
3209         d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3210         if (!d_out) /* cannot happen, but anyways. */
3211                 goto nla_put_failure;
3212         d_out->minor = mdev_to_minor(mdev);
3213         d_out->ret_code = NO_ERROR;
3214
3215         if (nla_put_status_info(msg, mdev, sib))
3216                 goto nla_put_failure;
3217         genlmsg_end(msg, d_out);
3218         err = drbd_genl_multicast_events(msg, 0);
3219         /* msg has been consumed or freed in netlink_broadcast() */
3220         if (err && err != -ESRCH)
3221                 goto failed;
3222
3223         return;
3224
3225 nla_put_failure:
3226         nlmsg_free(msg);
3227 failed:
3228         dev_err(DEV, "Error %d while broadcasting event. "
3229                         "Event seq:%u sib_reason:%u\n",
3230                         err, seq, sib->sib_reason);
3231 }