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 drbd_tconn *tconn = mdev->tconn;
 327         struct sib_info sib;
 328         int ret;
 329
 330         if (current == tconn->worker.task)
 331                 set_bit(CALLBACK_PENDING, &tconn->flags);
 332
 333         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 334         setup_khelper_env(tconn, envp);
 335
 336         /* The helper may take some time.
 337          * write out any unsynced meta data changes now */
 338         drbd_md_sync(mdev);
 339
 340         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 341         sib.sib_reason = SIB_HELPER_PRE;
 342         sib.helper_name = cmd;
 343         drbd_bcast_event(mdev, &sib);
 344         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 345         if (ret)
 346                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 347                                 usermode_helper, cmd, mb,
 348                                 (ret >> 8) & 0xff, ret);
 349         else
 350                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 351                                 usermode_helper, cmd, mb,
 352                                 (ret >> 8) & 0xff, ret);
 353         sib.sib_reason = SIB_HELPER_POST;
 354         sib.helper_exit_code = ret;
 355         drbd_bcast_event(mdev, &sib);
 356
 357         if (current == tconn->worker.task)
 358                 clear_bit(CALLBACK_PENDING, &tconn->flags);
 359
 360         if (ret < 0) /* Ignore any ERRNOs we got. */
 361                 ret = 0;
 362
 363         return ret;
 364 }
 365
 366 int conn_khelper(struct drbd_tconn *tconn, char *cmd)
 367 {
 368         char *envp[] = { "HOME=/",
 369                         "TERM=linux",
 370                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 371                          (char[20]) { }, /* address family */
 372                          (char[60]) { }, /* address */
 373                         NULL };
 374         char *argv[] = {usermode_helper, cmd, tconn->name, NULL };
 375         int ret;
 376
 377         setup_khelper_env(tconn, envp);
 378         conn_md_sync(tconn);
 379
 380         conn_info(tconn, "helper command: %s %s %s\n", usermode_helper, cmd, tconn->name);
 381         /* TODO: conn_bcast_event() ?? */
 382
 383         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 384         if (ret)
 385                 conn_warn(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
 386                           usermode_helper, cmd, tconn->name,
 387                           (ret >> 8) & 0xff, ret);
 388         else
 389                 conn_info(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
 390                           usermode_helper, cmd, tconn->name,
 391                           (ret >> 8) & 0xff, ret);
 392         /* TODO: conn_bcast_event() ?? */
 393
 394         if (ret < 0) /* Ignore any ERRNOs we got. */
 395                 ret = 0;
 396
 397         return ret;
 398 }
 399
 400 static enum drbd_fencing_p highest_fencing_policy(struct drbd_tconn *tconn)
 401 {
 402         enum drbd_fencing_p fp = FP_NOT_AVAIL;
 403         struct drbd_conf *mdev;
 404         int vnr;
 405
 406         rcu_read_lock();
 407         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
 408                 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 409                         fp = max_t(enum drbd_fencing_p, fp,
 410                                    rcu_dereference(mdev->ldev->disk_conf)->fencing);
 411                         put_ldev(mdev);
 412                 }
 413         }
 414         rcu_read_unlock();
 415
 416         return fp;
 417 }
 418
 419 bool conn_try_outdate_peer(struct drbd_tconn *tconn)
 420 {
 421         union drbd_state mask = { };
 422         union drbd_state val = { };
 423         enum drbd_fencing_p fp;
 424         char *ex_to_string;
 425         int r;
 426
 427         if (tconn->cstate >= C_WF_REPORT_PARAMS) {
 428                 conn_err(tconn, "Expected cstate < C_WF_REPORT_PARAMS\n");
 429                 return false;
 430         }
 431
 432         fp = highest_fencing_policy(tconn);
 433         switch (fp) {
 434         case FP_NOT_AVAIL:
 435                 conn_warn(tconn, "Not fencing peer, I'm not even Consistent myself.\n");
 436                 goto out;
 437         case FP_DONT_CARE:
 438                 return true;
 439         default: ;
 440         }
 441
 442         r = conn_khelper(tconn, "fence-peer");
 443
 444         switch ((r>>8) & 0xff) {
 445         case 3: /* peer is inconsistent */
 446                 ex_to_string = "peer is inconsistent or worse";
 447                 mask.pdsk = D_MASK;
 448                 val.pdsk = D_INCONSISTENT;
 449                 break;
 450         case 4: /* peer got outdated, or was already outdated */
 451                 ex_to_string = "peer was fenced";
 452                 mask.pdsk = D_MASK;
 453                 val.pdsk = D_OUTDATED;
 454                 break;
 455         case 5: /* peer was down */
 456                 if (conn_highest_disk(tconn) == D_UP_TO_DATE) {
 457                         /* we will(have) create(d) a new UUID anyways... */
 458                         ex_to_string = "peer is unreachable, assumed to be dead";
 459                         mask.pdsk = D_MASK;
 460                         val.pdsk = D_OUTDATED;
 461                 } else {
 462                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 463                 }
 464                 break;
 465         case 6: /* Peer is primary, voluntarily outdate myself.
 466                  * This is useful when an unconnected R_SECONDARY is asked to
 467                  * become R_PRIMARY, but finds the other peer being active. */
 468                 ex_to_string = "peer is active";
 469                 conn_warn(tconn, "Peer is primary, outdating myself.\n");
 470                 mask.disk = D_MASK;
 471                 val.disk = D_OUTDATED;
 472                 break;
 473         case 7:
 474                 if (fp != FP_STONITH)
 475                         conn_err(tconn, "fence-peer() = 7 && fencing != Stonith !!!\n");
 476                 ex_to_string = "peer was stonithed";
 477                 mask.pdsk = D_MASK;
 478                 val.pdsk = D_OUTDATED;
 479                 break;
 480         default:
 481                 /* The script is broken ... */
 482                 conn_err(tconn, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 483                 return false; /* Eventually leave IO frozen */
 484         }
 485
 486         conn_info(tconn, "fence-peer helper returned %d (%s)\n",
 487                   (r>>8) & 0xff, ex_to_string);
 488
 489  out:
 490
 491         /* Not using
 492            conn_request_state(tconn, mask, val, CS_VERBOSE);
 493            here, because we might were able to re-establish the connection in the
 494            meantime. */
 495         spin_lock_irq(&tconn->req_lock);
 496         if (tconn->cstate < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &tconn->flags))
 497                 _conn_request_state(tconn, mask, val, CS_VERBOSE);
 498         spin_unlock_irq(&tconn->req_lock);
 499
 500         return conn_highest_pdsk(tconn) <= D_OUTDATED;
 501 }
 502
 503 static int _try_outdate_peer_async(void *data)
 504 {
 505         struct drbd_tconn *tconn = (struct drbd_tconn *)data;
 506
 507         conn_try_outdate_peer(tconn);
 508
 509         kref_put(&tconn->kref, &conn_destroy);
 510         return 0;
 511 }
 512
 513 void conn_try_outdate_peer_async(struct drbd_tconn *tconn)
 514 {
 515         struct task_struct *opa;
 516
 517         kref_get(&tconn->kref);
 518         opa = kthread_run(_try_outdate_peer_async, tconn, "drbd_async_h");
 519         if (IS_ERR(opa)) {
 520                 conn_err(tconn, "out of mem, failed to invoke fence-peer helper\n");
 521                 kref_put(&tconn->kref, &conn_destroy);
 522         }
 523 }
 524
 525 enum drbd_state_rv
 526 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 527 {
 528         const int max_tries = 4;
 529         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
 530         struct net_conf *nc;
 531         int try = 0;
 532         int forced = 0;
 533         union drbd_state mask, val;
 534
 535         if (new_role == R_PRIMARY)
 536                 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
 537
 538         mutex_lock(mdev->state_mutex);
 539
 540         mask.i = 0; mask.role = R_MASK;
 541         val.i  = 0; val.role  = new_role;
 542
 543         while (try++ < max_tries) {
 544                 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 545
 546                 /* in case we first succeeded to outdate,
 547                  * but now suddenly could establish a connection */
 548                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 549                         val.pdsk = 0;
 550                         mask.pdsk = 0;
 551                         continue;
 552                 }
 553
 554                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
 555                     (mdev->state.disk < D_UP_TO_DATE &&
 556                      mdev->state.disk >= D_INCONSISTENT)) {
 557                         mask.disk = D_MASK;
 558                         val.disk  = D_UP_TO_DATE;
 559                         forced = 1;
 560                         continue;
 561                 }
 562
 563                 if (rv == SS_NO_UP_TO_DATE_DISK &&
 564                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 565                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 566
 567                         if (conn_try_outdate_peer(mdev->tconn)) {
 568                                 val.disk = D_UP_TO_DATE;
 569                                 mask.disk = D_MASK;
 570                         }
 571                         continue;
 572                 }
 573
 574                 if (rv == SS_NOTHING_TO_DO)
 575                         goto out;
 576                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
 577                         if (!conn_try_outdate_peer(mdev->tconn) && force) {
 578                                 dev_warn(DEV, "Forced into split brain situation!\n");
 579                                 mask.pdsk = D_MASK;
 580                                 val.pdsk  = D_OUTDATED;
 581
 582                         }
 583                         continue;
 584                 }
 585                 if (rv == SS_TWO_PRIMARIES) {
 586                         /* Maybe the peer is detected as dead very soon...
 587                            retry at most once more in this case. */
 588                         int timeo;
 589                         rcu_read_lock();
 590                         nc = rcu_dereference(mdev->tconn->net_conf);
 591                         timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
 592                         rcu_read_unlock();
 593                         schedule_timeout_interruptible(timeo);
 594                         if (try < max_tries)
 595                                 try = max_tries - 1;
 596                         continue;
 597                 }
 598                 if (rv < SS_SUCCESS) {
 599                         rv = _drbd_request_state(mdev, mask, val,
 600                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 601                         if (rv < SS_SUCCESS)
 602                                 goto out;
 603                 }
 604                 break;
 605         }
 606
 607         if (rv < SS_SUCCESS)
 608                 goto out;
 609
 610         if (forced)
 611                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 612
 613         /* Wait until nothing is on the fly :) */
 614         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 615
 616         /* FIXME also wait for all pending P_BARRIER_ACK? */
 617
 618         if (new_role == R_SECONDARY) {
 619                 set_disk_ro(mdev->vdisk, true);
 620                 if (get_ldev(mdev)) {
 621                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 622                         put_ldev(mdev);
 623                 }
 624         } else {
 625                 mutex_lock(&mdev->tconn->conf_update);
 626                 nc = mdev->tconn->net_conf;
 627                 if (nc)
 628                         nc->discard_my_data = 0; /* without copy; single bit op is atomic */
 629                 mutex_unlock(&mdev->tconn->conf_update);
 630
 631                 set_disk_ro(mdev->vdisk, false);
 632                 if (get_ldev(mdev)) {
 633                         if (((mdev->state.conn < C_CONNECTED ||
 634                                mdev->state.pdsk <= D_FAILED)
 635                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 636                                 drbd_uuid_new_current(mdev);
 637
 638                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 639                         put_ldev(mdev);
 640                 }
 641         }
 642
 643         /* writeout of activity log covered areas of the bitmap
 644          * to stable storage done in after state change already */
 645
 646         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 647                 /* if this was forced, we should consider sync */
 648                 if (forced)
 649                         drbd_send_uuids(mdev);
 650                 drbd_send_current_state(mdev);
 651         }
 652
 653         drbd_md_sync(mdev);
 654
 655         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 656 out:
 657         mutex_unlock(mdev->state_mutex);
 658         return rv;
 659 }
 660
 661 static const char *from_attrs_err_to_txt(int err)
 662 {
 663         return  err == -ENOMSG ? "required attribute missing" :
 664                 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
 665                 err == -EEXIST ? "can not change invariant setting" :
 666                 "invalid attribute value";
 667 }
 668
 669 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
 670 {
 671         struct set_role_parms parms;
 672         int err;
 673         enum drbd_ret_code retcode;
 674
 675         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
 676         if (!adm_ctx.reply_skb)
 677                 return retcode;
 678         if (retcode != NO_ERROR)
 679                 goto out;
 680
 681         memset(&parms, 0, sizeof(parms));
 682         if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
 683                 err = set_role_parms_from_attrs(&parms, info);
 684                 if (err) {
 685                         retcode = ERR_MANDATORY_TAG;
 686                         drbd_msg_put_info(from_attrs_err_to_txt(err));
 687                         goto out;
 688                 }
 689         }
 690
 691         if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
 692                 retcode = drbd_set_role(adm_ctx.mdev, R_PRIMARY, parms.assume_uptodate);
 693         else
 694                 retcode = drbd_set_role(adm_ctx.mdev, R_SECONDARY, 0);
 695 out:
 696         drbd_adm_finish(info, retcode);
 697         return 0;
 698 }
 699
 700 /* initializes the md.*_offset members, so we are able to find
 701  * the on disk meta data */
 702 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 703                                        struct drbd_backing_dev *bdev)
 704 {
 705         sector_t md_size_sect = 0;
 706         int meta_dev_idx;
 707
 708         rcu_read_lock();
 709         meta_dev_idx = rcu_dereference(bdev->disk_conf)->meta_dev_idx;
 710
 711         switch (meta_dev_idx) {
 712         default:
 713                 /* v07 style fixed size indexed meta data */
 714                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 715                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 716                 bdev->md.al_offset = MD_AL_OFFSET;
 717                 bdev->md.bm_offset = MD_BM_OFFSET;
 718                 break;
 719         case DRBD_MD_INDEX_FLEX_EXT:
 720                 /* just occupy the full device; unit: sectors */
 721                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 722                 bdev->md.md_offset = 0;
 723                 bdev->md.al_offset = MD_AL_OFFSET;
 724                 bdev->md.bm_offset = MD_BM_OFFSET;
 725                 break;
 726         case DRBD_MD_INDEX_INTERNAL:
 727         case DRBD_MD_INDEX_FLEX_INT:
 728                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 729                 /* al size is still fixed */
 730                 bdev->md.al_offset = -MD_AL_SECTORS;
 731                 /* we need (slightly less than) ~ this much bitmap sectors: */
 732                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 733                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 734                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 735                 md_size_sect = ALIGN(md_size_sect, 8);
 736
 737                 /* plus the "drbd meta data super block",
 738                  * and the activity log; */
 739                 md_size_sect += MD_BM_OFFSET;
 740
 741                 bdev->md.md_size_sect = md_size_sect;
 742                 /* bitmap offset is adjusted by 'super' block size */
 743                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 744                 break;
 745         }
 746         rcu_read_unlock();
 747 }
 748
 749 /* input size is expected to be in KB */
 750 char *ppsize(char *buf, unsigned long long size)
 751 {
 752         /* Needs 9 bytes at max including trailing NUL:
 753          * -1ULL ==> "16384 EB" */
 754         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 755         int base = 0;
 756         while (size >= 10000 && base < sizeof(units)-1) {
 757                 /* shift + round */
 758                 size = (size >> 10) + !!(size & (1<<9));
 759                 base++;
 760         }
 761         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
 762
 763         return buf;
 764 }
 765
 766 /* there is still a theoretical deadlock when called from receiver
 767  * on an D_INCONSISTENT R_PRIMARY:
 768  *  remote READ does inc_ap_bio, receiver would need to receive answer
 769  *  packet from remote to dec_ap_bio again.
 770  *  receiver receive_sizes(), comes here,
 771  *  waits for ap_bio_cnt == 0. -> deadlock.
 772  * but this cannot happen, actually, because:
 773  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 774  *  (not connected, or bad/no disk on peer):
 775  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 776  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 777  *  peer may not initiate a resize.
 778  */
 779 /* Note these are not to be confused with
 780  * drbd_adm_suspend_io/drbd_adm_resume_io,
 781  * which are (sub) state changes triggered by admin (drbdsetup),
 782  * and can be long lived.
 783  * This changes an mdev->flag, is triggered by drbd internals,
 784  * and should be short-lived. */
 785 void drbd_suspend_io(struct drbd_conf *mdev)
 786 {
 787         set_bit(SUSPEND_IO, &mdev->flags);
 788         if (drbd_suspended(mdev))
 789                 return;
 790         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 791 }
 792
 793 void drbd_resume_io(struct drbd_conf *mdev)
 794 {
 795         clear_bit(SUSPEND_IO, &mdev->flags);
 796         wake_up(&mdev->misc_wait);
 797 }
 798
 799 /**
 800  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 801  * @mdev:       DRBD device.
 802  *
 803  * Returns 0 on success, negative return values indicate errors.
 804  * You should call drbd_md_sync() after calling this function.
 805  */
 806 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 807 {
 808         sector_t prev_first_sect, prev_size; /* previous meta location */
 809         sector_t la_size, u_size;
 810         sector_t size;
 811         char ppb[10];
 812
 813         int md_moved, la_size_changed;
 814         enum determine_dev_size rv = unchanged;
 815
 816         /* race:
 817          * application request passes inc_ap_bio,
 818          * but then cannot get an AL-reference.
 819          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 820          *
 821          * to avoid that:
 822          * Suspend IO right here.
 823          * still lock the act_log to not trigger ASSERTs there.
 824          */
 825         drbd_suspend_io(mdev);
 826
 827         /* no wait necessary anymore, actually we could assert that */
 828         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 829
 830         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 831         prev_size = mdev->ldev->md.md_size_sect;
 832         la_size = mdev->ldev->md.la_size_sect;
 833
 834         /* TODO: should only be some assert here, not (re)init... */
 835         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 836
 837         rcu_read_lock();
 838         u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
 839         rcu_read_unlock();
 840         size = drbd_new_dev_size(mdev, mdev->ldev, u_size, flags & DDSF_FORCED);
 841
 842         if (drbd_get_capacity(mdev->this_bdev) != size ||
 843             drbd_bm_capacity(mdev) != size) {
 844                 int err;
 845                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 846                 if (unlikely(err)) {
 847                         /* currently there is only one error: ENOMEM! */
 848                         size = drbd_bm_capacity(mdev)>>1;
 849                         if (size == 0) {
 850                                 dev_err(DEV, "OUT OF MEMORY! "
 851                                     "Could not allocate bitmap!\n");
 852                         } else {
 853                                 dev_err(DEV, "BM resizing failed. "
 854                                     "Leaving size unchanged at size = %lu KB\n",
 855                                     (unsigned long)size);
 856                         }
 857                         rv = dev_size_error;
 858                 }
 859                 /* racy, see comments above. */
 860                 drbd_set_my_capacity(mdev, size);
 861                 mdev->ldev->md.la_size_sect = size;
 862                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 863                      (unsigned long long)size>>1);
 864         }
 865         if (rv == dev_size_error)
 866                 goto out;
 867
 868         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 869
 870         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 871                 || prev_size       != mdev->ldev->md.md_size_sect;
 872
 873         if (la_size_changed || md_moved) {
 874                 int err;
 875
 876                 drbd_al_shrink(mdev); /* All extents inactive. */
 877                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 878                          la_size_changed && md_moved ? "size changed and md moved" :
 879                          la_size_changed ? "size changed" : "md moved");
 880                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
 881                 err = drbd_bitmap_io(mdev, md_moved ? &drbd_bm_write_all : &drbd_bm_write,
 882                                      "size changed", BM_LOCKED_MASK);
 883                 if (err) {
 884                         rv = dev_size_error;
 885                         goto out;
 886                 }
 887                 drbd_md_mark_dirty(mdev);
 888         }
 889
 890         if (size > la_size)
 891                 rv = grew;
 892         if (size < la_size)
 893                 rv = shrunk;
 894 out:
 895         lc_unlock(mdev->act_log);
 896         wake_up(&mdev->al_wait);
 897         drbd_resume_io(mdev);
 898
 899         return rv;
 900 }
 901
 902 sector_t
 903 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
 904                   sector_t u_size, int assume_peer_has_space)
 905 {
 906         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 907         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 908         sector_t m_size; /* my size */
 909         sector_t size = 0;
 910
 911         m_size = drbd_get_max_capacity(bdev);
 912
 913         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 914                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 915                 p_size = m_size;
 916         }
 917
 918         if (p_size && m_size) {
 919                 size = min_t(sector_t, p_size, m_size);
 920         } else {
 921                 if (la_size) {
 922                         size = la_size;
 923                         if (m_size && m_size < size)
 924                                 size = m_size;
 925                         if (p_size && p_size < size)
 926                                 size = p_size;
 927                 } else {
 928                         if (m_size)
 929                                 size = m_size;
 930                         if (p_size)
 931                                 size = p_size;
 932                 }
 933         }
 934
 935         if (size == 0)
 936                 dev_err(DEV, "Both nodes diskless!\n");
 937
 938         if (u_size) {
 939                 if (u_size > size)
 940                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 941                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 942                 else
 943                         size = u_size;
 944         }
 945
 946         return size;
 947 }
 948
 949 /**
 950  * drbd_check_al_size() - Ensures that the AL is of the right size
 951  * @mdev:       DRBD device.
 952  *
 953  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 954  * failed, and 0 on success. You should call drbd_md_sync() after you called
 955  * this function.
 956  */
 957 static int drbd_check_al_size(struct drbd_conf *mdev, struct disk_conf *dc)
 958 {
 959         struct lru_cache *n, *t;
 960         struct lc_element *e;
 961         unsigned int in_use;
 962         int i;
 963
 964         if (mdev->act_log &&
 965             mdev->act_log->nr_elements == dc->al_extents)
 966                 return 0;
 967
 968         in_use = 0;
 969         t = mdev->act_log;
 970         n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
 971                 dc->al_extents, sizeof(struct lc_element), 0);
 972
 973         if (n == NULL) {
 974                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 975                 return -ENOMEM;
 976         }
 977         spin_lock_irq(&mdev->al_lock);
 978         if (t) {
 979                 for (i = 0; i < t->nr_elements; i++) {
 980                         e = lc_element_by_index(t, i);
 981                         if (e->refcnt)
 982                                 dev_err(DEV, "refcnt(%d)==%d\n",
 983                                     e->lc_number, e->refcnt);
 984                         in_use += e->refcnt;
 985                 }
 986         }
 987         if (!in_use)
 988                 mdev->act_log = n;
 989         spin_unlock_irq(&mdev->al_lock);
 990         if (in_use) {
 991                 dev_err(DEV, "Activity log still in use!\n");
 992                 lc_destroy(n);
 993                 return -EBUSY;
 994         } else {
 995                 if (t)
 996                         lc_destroy(t);
 997         }
 998         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 999         return 0;
1000 }
1001
1002 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
1003 {
1004         struct request_queue * const q = mdev->rq_queue;
1005         int max_hw_sectors = max_bio_size >> 9;
1006         int max_segments = 0;
1007
1008         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1009                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
1010
1011                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1012                 rcu_read_lock();
1013                 max_segments = rcu_dereference(mdev->ldev->disk_conf)->max_bio_bvecs;
1014                 rcu_read_unlock();
1015                 put_ldev(mdev);
1016         }
1017
1018         blk_queue_logical_block_size(q, 512);
1019         blk_queue_max_hw_sectors(q, max_hw_sectors);
1020         /* This is the workaround for "bio would need to, but cannot, be split" */
1021         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1022         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
1023
1024         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1025                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
1026
1027                 blk_queue_stack_limits(q, b);
1028
1029                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
1030                         dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1031                                  q->backing_dev_info.ra_pages,
1032                                  b->backing_dev_info.ra_pages);
1033                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
1034                 }
1035                 put_ldev(mdev);
1036         }
1037 }
1038
1039 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
1040 {
1041         int now, new, local, peer;
1042
1043         now = queue_max_hw_sectors(mdev->rq_queue) << 9;
1044         local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
1045         peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
1046
1047         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1048                 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
1049                 mdev->local_max_bio_size = local;
1050                 put_ldev(mdev);
1051         }
1052
1053         /* We may ignore peer limits if the peer is modern enough.
1054            Because new from 8.3.8 onwards the peer can use multiple
1055            BIOs for a single peer_request */
1056         if (mdev->state.conn >= C_CONNECTED) {
1057                 if (mdev->tconn->agreed_pro_version < 94)
1058                         peer = min_t(int, mdev->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1059                         /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1060                 else if (mdev->tconn->agreed_pro_version == 94)
1061                         peer = DRBD_MAX_SIZE_H80_PACKET;
1062                 else if (mdev->tconn->agreed_pro_version < 100)
1063                         peer = DRBD_MAX_BIO_SIZE_P95;  /* drbd 8.3.8 onwards, before 8.4.0 */
1064                 else
1065                         peer = DRBD_MAX_BIO_SIZE;
1066         }
1067
1068         new = min_t(int, local, peer);
1069
1070         if (mdev->state.role == R_PRIMARY && new < now)
1071                 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
1072
1073         if (new != now)
1074                 dev_info(DEV, "max BIO size = %u\n", new);
1075
1076         drbd_setup_queue_param(mdev, new);
1077 }
1078
1079 /* Starts the worker thread */
1080 static void conn_reconfig_start(struct drbd_tconn *tconn)
1081 {
1082         drbd_thread_start(&tconn->worker);
1083         conn_flush_workqueue(tconn);
1084 }
1085
1086 /* if still unconfigured, stops worker again. */
1087 static void conn_reconfig_done(struct drbd_tconn *tconn)
1088 {
1089         bool stop_threads;
1090         spin_lock_irq(&tconn->req_lock);
1091         stop_threads = conn_all_vols_unconf(tconn) &&
1092                 tconn->cstate == C_STANDALONE;
1093         spin_unlock_irq(&tconn->req_lock);
1094         if (stop_threads) {
1095                 /* asender is implicitly stopped by receiver
1096                  * in conn_disconnect() */
1097                 drbd_thread_stop(&tconn->receiver);
1098                 drbd_thread_stop(&tconn->worker);
1099         }
1100 }
1101
1102 /* Make sure IO is suspended before calling this function(). */
1103 static void drbd_suspend_al(struct drbd_conf *mdev)
1104 {
1105         int s = 0;
1106
1107         if (!lc_try_lock(mdev->act_log)) {
1108                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
1109                 return;
1110         }
1111
1112         drbd_al_shrink(mdev);
1113         spin_lock_irq(&mdev->tconn->req_lock);
1114         if (mdev->state.conn < C_CONNECTED)
1115                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
1116         spin_unlock_irq(&mdev->tconn->req_lock);
1117         lc_unlock(mdev->act_log);
1118
1119         if (s)
1120                 dev_info(DEV, "Suspended AL updates\n");
1121 }
1122
1123
1124 static bool should_set_defaults(struct genl_info *info)
1125 {
1126         unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1127         return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1128 }
1129
1130 static void enforce_disk_conf_limits(struct disk_conf *dc)
1131 {
1132         if (dc->al_extents < DRBD_AL_EXTENTS_MIN)
1133                 dc->al_extents = DRBD_AL_EXTENTS_MIN;
1134         if (dc->al_extents > DRBD_AL_EXTENTS_MAX)
1135                 dc->al_extents = DRBD_AL_EXTENTS_MAX;
1136
1137         if (dc->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1138                 dc->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1139 }
1140
1141 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1142 {
1143         enum drbd_ret_code retcode;
1144         struct drbd_conf *mdev;
1145         struct disk_conf *new_disk_conf, *old_disk_conf;
1146         struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1147         int err, fifo_size;
1148
1149         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1150         if (!adm_ctx.reply_skb)
1151                 return retcode;
1152         if (retcode != NO_ERROR)
1153                 goto out;
1154
1155         mdev = adm_ctx.mdev;
1156
1157         /* we also need a disk
1158          * to change the options on */
1159         if (!get_ldev(mdev)) {
1160                 retcode = ERR_NO_DISK;
1161                 goto out;
1162         }
1163
1164         new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1165         if (!new_disk_conf) {
1166                 retcode = ERR_NOMEM;
1167                 goto fail;
1168         }
1169
1170         mutex_lock(&mdev->tconn->conf_update);
1171         old_disk_conf = mdev->ldev->disk_conf;
1172         *new_disk_conf = *old_disk_conf;
1173         if (should_set_defaults(info))
1174                 set_disk_conf_defaults(new_disk_conf);
1175
1176         err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1177         if (err && err != -ENOMSG) {
1178                 retcode = ERR_MANDATORY_TAG;
1179                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1180         }
1181
1182         if (!expect(new_disk_conf->resync_rate >= 1))
1183                 new_disk_conf->resync_rate = 1;
1184
1185         enforce_disk_conf_limits(new_disk_conf);
1186
1187         fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1188         if (fifo_size != mdev->rs_plan_s->size) {
1189                 new_plan = fifo_alloc(fifo_size);
1190                 if (!new_plan) {
1191                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1192                         retcode = ERR_NOMEM;
1193                         goto fail_unlock;
1194                 }
1195         }
1196
1197         drbd_suspend_io(mdev);
1198         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1199         drbd_al_shrink(mdev);
1200         err = drbd_check_al_size(mdev, new_disk_conf);
1201         lc_unlock(mdev->act_log);
1202         wake_up(&mdev->al_wait);
1203         drbd_resume_io(mdev);
1204
1205         if (err) {
1206                 retcode = ERR_NOMEM;
1207                 goto fail_unlock;
1208         }
1209
1210         write_lock_irq(&global_state_lock);
1211         retcode = drbd_resync_after_valid(mdev, new_disk_conf->resync_after);
1212         if (retcode == NO_ERROR) {
1213                 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
1214                 drbd_resync_after_changed(mdev);
1215         }
1216         write_unlock_irq(&global_state_lock);
1217
1218         if (retcode != NO_ERROR)
1219                 goto fail_unlock;
1220
1221         if (new_plan) {
1222                 old_plan = mdev->rs_plan_s;
1223                 rcu_assign_pointer(mdev->rs_plan_s, new_plan);
1224         }
1225
1226         mutex_unlock(&mdev->tconn->conf_update);
1227
1228         if (new_disk_conf->al_updates)
1229                 mdev->ldev->md.flags &= MDF_AL_DISABLED;
1230         else
1231                 mdev->ldev->md.flags |= MDF_AL_DISABLED;
1232
1233         drbd_bump_write_ordering(mdev->tconn, WO_bdev_flush);
1234
1235         drbd_md_sync(mdev);
1236
1237         if (mdev->state.conn >= C_CONNECTED)
1238                 drbd_send_sync_param(mdev);
1239
1240         synchronize_rcu();
1241         kfree(old_disk_conf);
1242         kfree(old_plan);
1243         mod_timer(&mdev->request_timer, jiffies + HZ);
1244         goto success;
1245
1246 fail_unlock:
1247         mutex_unlock(&mdev->tconn->conf_update);
1248  fail:
1249         kfree(new_disk_conf);
1250         kfree(new_plan);
1251 success:
1252         put_ldev(mdev);
1253  out:
1254         drbd_adm_finish(info, retcode);
1255         return 0;
1256 }
1257
1258 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1259 {
1260         struct drbd_conf *mdev;
1261         int err;
1262         enum drbd_ret_code retcode;
1263         enum determine_dev_size dd;
1264         sector_t max_possible_sectors;
1265         sector_t min_md_device_sectors;
1266         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1267         struct disk_conf *new_disk_conf = NULL;
1268         struct block_device *bdev;
1269         struct lru_cache *resync_lru = NULL;
1270         struct fifo_buffer *new_plan = NULL;
1271         union drbd_state ns, os;
1272         enum drbd_state_rv rv;
1273         struct net_conf *nc;
1274
1275         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1276         if (!adm_ctx.reply_skb)
1277                 return retcode;
1278         if (retcode != NO_ERROR)
1279                 goto finish;
1280
1281         mdev = adm_ctx.mdev;
1282         conn_reconfig_start(mdev->tconn);
1283
1284         /* if you want to reconfigure, please tear down first */
1285         if (mdev->state.disk > D_DISKLESS) {
1286                 retcode = ERR_DISK_CONFIGURED;
1287                 goto fail;
1288         }
1289         /* It may just now have detached because of IO error.  Make sure
1290          * drbd_ldev_destroy is done already, we may end up here very fast,
1291          * e.g. if someone calls attach from the on-io-error handler,
1292          * to realize a "hot spare" feature (not that I'd recommend that) */
1293         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1294
1295         /* make sure there is no leftover from previous force-detach attempts */
1296         clear_bit(FORCE_DETACH, &mdev->flags);
1297
1298         /* and no leftover from previously aborted resync or verify, either */
1299         mdev->rs_total = 0;
1300         mdev->rs_failed = 0;
1301         atomic_set(&mdev->rs_pending_cnt, 0);
1302
1303         /* allocation not in the IO path, drbdsetup context */
1304         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1305         if (!nbc) {
1306                 retcode = ERR_NOMEM;
1307                 goto fail;
1308         }
1309         spin_lock_init(&nbc->md.uuid_lock);
1310
1311         new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1312         if (!new_disk_conf) {
1313                 retcode = ERR_NOMEM;
1314                 goto fail;
1315         }
1316         nbc->disk_conf = new_disk_conf;
1317
1318         set_disk_conf_defaults(new_disk_conf);
1319         err = disk_conf_from_attrs(new_disk_conf, info);
1320         if (err) {
1321                 retcode = ERR_MANDATORY_TAG;
1322                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1323                 goto fail;
1324         }
1325
1326         enforce_disk_conf_limits(new_disk_conf);
1327
1328         new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1329         if (!new_plan) {
1330                 retcode = ERR_NOMEM;
1331                 goto fail;
1332         }
1333
1334         if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1335                 retcode = ERR_MD_IDX_INVALID;
1336                 goto fail;
1337         }
1338
1339         rcu_read_lock();
1340         nc = rcu_dereference(mdev->tconn->net_conf);
1341         if (nc) {
1342                 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1343                         rcu_read_unlock();
1344                         retcode = ERR_STONITH_AND_PROT_A;
1345                         goto fail;
1346                 }
1347         }
1348         rcu_read_unlock();
1349
1350         bdev = blkdev_get_by_path(new_disk_conf->backing_dev,
1351                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1352         if (IS_ERR(bdev)) {
1353                 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->backing_dev,
1354                         PTR_ERR(bdev));
1355                 retcode = ERR_OPEN_DISK;
1356                 goto fail;
1357         }
1358         nbc->backing_bdev = bdev;
1359
1360         /*
1361          * meta_dev_idx >= 0: external fixed size, possibly multiple
1362          * drbd sharing one meta device.  TODO in that case, paranoia
1363          * check that [md_bdev, meta_dev_idx] is not yet used by some
1364          * other drbd minor!  (if you use drbd.conf + drbdadm, that
1365          * should check it for you already; but if you don't, or
1366          * someone fooled it, we need to double check here)
1367          */
1368         bdev = blkdev_get_by_path(new_disk_conf->meta_dev,
1369                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1370                                   (new_disk_conf->meta_dev_idx < 0) ?
1371                                   (void *)mdev : (void *)drbd_m_holder);
1372         if (IS_ERR(bdev)) {
1373                 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->meta_dev,
1374                         PTR_ERR(bdev));
1375                 retcode = ERR_OPEN_MD_DISK;
1376                 goto fail;
1377         }
1378         nbc->md_bdev = bdev;
1379
1380         if ((nbc->backing_bdev == nbc->md_bdev) !=
1381             (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1382              new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1383                 retcode = ERR_MD_IDX_INVALID;
1384                 goto fail;
1385         }
1386
1387         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1388                         1, 61, sizeof(struct bm_extent),
1389                         offsetof(struct bm_extent, lce));
1390         if (!resync_lru) {
1391                 retcode = ERR_NOMEM;
1392                 goto fail;
1393         }
1394
1395         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1396         drbd_md_set_sector_offsets(mdev, nbc);
1397
1398         if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1399                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1400                         (unsigned long long) drbd_get_max_capacity(nbc),
1401                         (unsigned long long) new_disk_conf->disk_size);
1402                 retcode = ERR_DISK_TOO_SMALL;
1403                 goto fail;
1404         }
1405
1406         if (new_disk_conf->meta_dev_idx < 0) {
1407                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1408                 /* at least one MB, otherwise it does not make sense */
1409                 min_md_device_sectors = (2<<10);
1410         } else {
1411                 max_possible_sectors = DRBD_MAX_SECTORS;
1412                 min_md_device_sectors = MD_RESERVED_SECT * (new_disk_conf->meta_dev_idx + 1);
1413         }
1414
1415         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1416                 retcode = ERR_MD_DISK_TOO_SMALL;
1417                 dev_warn(DEV, "refusing attach: md-device too small, "
1418                      "at least %llu sectors needed for this meta-disk type\n",
1419                      (unsigned long long) min_md_device_sectors);
1420                 goto fail;
1421         }
1422
1423         /* Make sure the new disk is big enough
1424          * (we may currently be R_PRIMARY with no local disk...) */
1425         if (drbd_get_max_capacity(nbc) <
1426             drbd_get_capacity(mdev->this_bdev)) {
1427                 retcode = ERR_DISK_TOO_SMALL;
1428                 goto fail;
1429         }
1430
1431         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1432
1433         if (nbc->known_size > max_possible_sectors) {
1434                 dev_warn(DEV, "==> truncating very big lower level device "
1435                         "to currently maximum possible %llu sectors <==\n",
1436                         (unsigned long long) max_possible_sectors);
1437                 if (new_disk_conf->meta_dev_idx >= 0)
1438                         dev_warn(DEV, "==>> using internal or flexible "
1439                                       "meta data may help <<==\n");
1440         }
1441
1442         drbd_suspend_io(mdev);
1443         /* also wait for the last barrier ack. */
1444         /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1445          * We need a way to either ignore barrier acks for barriers sent before a device
1446          * was attached, or a way to wait for all pending barrier acks to come in.
1447          * As barriers are counted per resource,
1448          * we'd need to suspend io on all devices of a resource.
1449          */
1450         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || drbd_suspended(mdev));
1451         /* and for any other previously queued work */
1452         drbd_flush_workqueue(mdev);
1453
1454         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1455         retcode = rv;  /* FIXME: Type mismatch. */
1456         drbd_resume_io(mdev);
1457         if (rv < SS_SUCCESS)
1458                 goto fail;
1459
1460         if (!get_ldev_if_state(mdev, D_ATTACHING))
1461                 goto force_diskless;
1462
1463         drbd_md_set_sector_offsets(mdev, nbc);
1464
1465         if (!mdev->bitmap) {
1466                 if (drbd_bm_init(mdev)) {
1467                         retcode = ERR_NOMEM;
1468                         goto force_diskless_dec;
1469                 }
1470         }
1471
1472         retcode = drbd_md_read(mdev, nbc);
1473         if (retcode != NO_ERROR)
1474                 goto force_diskless_dec;
1475
1476         if (mdev->state.conn < C_CONNECTED &&
1477             mdev->state.role == R_PRIMARY &&
1478             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1479                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1480                     (unsigned long long)mdev->ed_uuid);
1481                 retcode = ERR_DATA_NOT_CURRENT;
1482                 goto force_diskless_dec;
1483         }
1484
1485         /* Since we are diskless, fix the activity log first... */
1486         if (drbd_check_al_size(mdev, new_disk_conf)) {
1487                 retcode = ERR_NOMEM;
1488                 goto force_diskless_dec;
1489         }
1490
1491         /* Prevent shrinking of consistent devices ! */
1492         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1493             drbd_new_dev_size(mdev, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1494                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1495                 retcode = ERR_DISK_TOO_SMALL;
1496                 goto force_diskless_dec;
1497         }
1498
1499         /* Reset the "barriers don't work" bits here, then force meta data to
1500          * be written, to ensure we determine if barriers are supported. */
1501         if (new_disk_conf->md_flushes)
1502                 clear_bit(MD_NO_FUA, &mdev->flags);
1503         else
1504                 set_bit(MD_NO_FUA, &mdev->flags);
1505
1506         /* Point of no return reached.
1507          * Devices and memory are no longer released by error cleanup below.
1508          * now mdev takes over responsibility, and the state engine should
1509          * clean it up somewhere.  */
1510         D_ASSERT(mdev->ldev == NULL);
1511         mdev->ldev = nbc;
1512         mdev->resync = resync_lru;
1513         mdev->rs_plan_s = new_plan;
1514         nbc = NULL;
1515         resync_lru = NULL;
1516         new_disk_conf = NULL;
1517         new_plan = NULL;
1518
1519         drbd_bump_write_ordering(mdev->tconn, WO_bdev_flush);
1520
1521         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1522                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1523         else
1524                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1525
1526         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1527             !(mdev->state.role == R_PRIMARY && mdev->tconn->susp_nod))
1528                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1529
1530         mdev->send_cnt = 0;
1531         mdev->recv_cnt = 0;
1532         mdev->read_cnt = 0;
1533         mdev->writ_cnt = 0;
1534
1535         drbd_reconsider_max_bio_size(mdev);
1536
1537         /* If I am currently not R_PRIMARY,
1538          * but meta data primary indicator is set,
1539          * I just now recover from a hard crash,
1540          * and have been R_PRIMARY before that crash.
1541          *
1542          * Now, if I had no connection before that crash
1543          * (have been degraded R_PRIMARY), chances are that
1544          * I won't find my peer now either.
1545          *
1546          * In that case, and _only_ in that case,
1547          * we use the degr-wfc-timeout instead of the default,
1548          * so we can automatically recover from a crash of a
1549          * degraded but active "cluster" after a certain timeout.
1550          */
1551         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1552         if (mdev->state.role != R_PRIMARY &&
1553              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1554             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1555                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1556
1557         dd = drbd_determine_dev_size(mdev, 0);
1558         if (dd == dev_size_error) {
1559                 retcode = ERR_NOMEM_BITMAP;
1560                 goto force_diskless_dec;
1561         } else if (dd == grew)
1562                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1563
1564         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC) ||
1565             (test_bit(CRASHED_PRIMARY, &mdev->flags) &&
1566              drbd_md_test_flag(mdev->ldev, MDF_AL_DISABLED))) {
1567                 dev_info(DEV, "Assuming that all blocks are out of sync "
1568                      "(aka FullSync)\n");
1569                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1570                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1571                         retcode = ERR_IO_MD_DISK;
1572                         goto force_diskless_dec;
1573                 }
1574         } else {
1575                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1576                         "read from attaching", BM_LOCKED_MASK)) {
1577                         retcode = ERR_IO_MD_DISK;
1578                         goto force_diskless_dec;
1579                 }
1580         }
1581
1582         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1583                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1584
1585         spin_lock_irq(&mdev->tconn->req_lock);
1586         os = drbd_read_state(mdev);
1587         ns = os;
1588         /* If MDF_CONSISTENT is not set go into inconsistent state,
1589            otherwise investigate MDF_WasUpToDate...
1590            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1591            otherwise into D_CONSISTENT state.
1592         */
1593         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1594                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1595                         ns.disk = D_CONSISTENT;
1596                 else
1597                         ns.disk = D_OUTDATED;
1598         } else {
1599                 ns.disk = D_INCONSISTENT;
1600         }
1601
1602         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1603                 ns.pdsk = D_OUTDATED;
1604
1605         rcu_read_lock();
1606         if (ns.disk == D_CONSISTENT &&
1607             (ns.pdsk == D_OUTDATED || rcu_dereference(mdev->ldev->disk_conf)->fencing == FP_DONT_CARE))
1608                 ns.disk = D_UP_TO_DATE;
1609
1610         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1611            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1612            this point, because drbd_request_state() modifies these
1613            flags. */
1614
1615         if (rcu_dereference(mdev->ldev->disk_conf)->al_updates)
1616                 mdev->ldev->md.flags &= MDF_AL_DISABLED;
1617         else
1618                 mdev->ldev->md.flags |= MDF_AL_DISABLED;
1619
1620         rcu_read_unlock();
1621
1622         /* In case we are C_CONNECTED postpone any decision on the new disk
1623            state after the negotiation phase. */
1624         if (mdev->state.conn == C_CONNECTED) {
1625                 mdev->new_state_tmp.i = ns.i;
1626                 ns.i = os.i;
1627                 ns.disk = D_NEGOTIATING;
1628
1629                 /* We expect to receive up-to-date UUIDs soon.
1630                    To avoid a race in receive_state, free p_uuid while
1631                    holding req_lock. I.e. atomic with the state change */
1632                 kfree(mdev->p_uuid);
1633                 mdev->p_uuid = NULL;
1634         }
1635
1636         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1637         spin_unlock_irq(&mdev->tconn->req_lock);
1638
1639         if (rv < SS_SUCCESS)
1640                 goto force_diskless_dec;
1641
1642         mod_timer(&mdev->request_timer, jiffies + HZ);
1643
1644         if (mdev->state.role == R_PRIMARY)
1645                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1646         else
1647                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1648
1649         drbd_md_mark_dirty(mdev);
1650         drbd_md_sync(mdev);
1651
1652         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1653         put_ldev(mdev);
1654         conn_reconfig_done(mdev->tconn);
1655         drbd_adm_finish(info, retcode);
1656         return 0;
1657
1658  force_diskless_dec:
1659         put_ldev(mdev);
1660  force_diskless:
1661         drbd_force_state(mdev, NS(disk, D_DISKLESS));
1662         drbd_md_sync(mdev);
1663  fail:
1664         conn_reconfig_done(mdev->tconn);
1665         if (nbc) {
1666                 if (nbc->backing_bdev)
1667                         blkdev_put(nbc->backing_bdev,
1668                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1669                 if (nbc->md_bdev)
1670                         blkdev_put(nbc->md_bdev,
1671                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1672                 kfree(nbc);
1673         }
1674         kfree(new_disk_conf);
1675         lc_destroy(resync_lru);
1676         kfree(new_plan);
1677
1678  finish:
1679         drbd_adm_finish(info, retcode);
1680         return 0;
1681 }
1682
1683 static int adm_detach(struct drbd_conf *mdev, int force)
1684 {
1685         enum drbd_state_rv retcode;
1686         int ret;
1687
1688         if (force) {
1689                 set_bit(FORCE_DETACH, &mdev->flags);
1690                 drbd_force_state(mdev, NS(disk, D_FAILED));
1691                 retcode = SS_SUCCESS;
1692                 goto out;
1693         }
1694
1695         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1696         drbd_md_get_buffer(mdev); /* make sure there is no in-flight meta-data IO */
1697         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1698         drbd_md_put_buffer(mdev);
1699         /* D_FAILED will transition to DISKLESS. */
1700         ret = wait_event_interruptible(mdev->misc_wait,
1701                         mdev->state.disk != D_FAILED);
1702         drbd_resume_io(mdev);
1703         if ((int)retcode == (int)SS_IS_DISKLESS)
1704                 retcode = SS_NOTHING_TO_DO;
1705         if (ret)
1706                 retcode = ERR_INTR;
1707 out:
1708         return retcode;
1709 }
1710
1711 /* Detaching the disk is a process in multiple stages.  First we need to lock
1712  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1713  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1714  * internal references as well.
1715  * Only then we have finally detached. */
1716 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1717 {
1718         enum drbd_ret_code retcode;
1719         struct detach_parms parms = { };
1720         int err;
1721
1722         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1723         if (!adm_ctx.reply_skb)
1724                 return retcode;
1725         if (retcode != NO_ERROR)
1726                 goto out;
1727
1728         if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
1729                 err = detach_parms_from_attrs(&parms, info);
1730                 if (err) {
1731                         retcode = ERR_MANDATORY_TAG;
1732                         drbd_msg_put_info(from_attrs_err_to_txt(err));
1733                         goto out;
1734                 }
1735         }
1736
1737         retcode = adm_detach(adm_ctx.mdev, parms.force_detach);
1738 out:
1739         drbd_adm_finish(info, retcode);
1740         return 0;
1741 }
1742
1743 static bool conn_resync_running(struct drbd_tconn *tconn)
1744 {
1745         struct drbd_conf *mdev;
1746         bool rv = false;
1747         int vnr;
1748
1749         rcu_read_lock();
1750         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1751                 if (mdev->state.conn == C_SYNC_SOURCE ||
1752                     mdev->state.conn == C_SYNC_TARGET ||
1753                     mdev->state.conn == C_PAUSED_SYNC_S ||
1754                     mdev->state.conn == C_PAUSED_SYNC_T) {
1755                         rv = true;
1756                         break;
1757                 }
1758         }
1759         rcu_read_unlock();
1760
1761         return rv;
1762 }
1763
1764 static bool conn_ov_running(struct drbd_tconn *tconn)
1765 {
1766         struct drbd_conf *mdev;
1767         bool rv = false;
1768         int vnr;
1769
1770         rcu_read_lock();
1771         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1772                 if (mdev->state.conn == C_VERIFY_S ||
1773                     mdev->state.conn == C_VERIFY_T) {
1774                         rv = true;
1775                         break;
1776                 }
1777         }
1778         rcu_read_unlock();
1779
1780         return rv;
1781 }
1782
1783 static enum drbd_ret_code
1784 _check_net_options(struct drbd_tconn *tconn, struct net_conf *old_conf, struct net_conf *new_conf)
1785 {
1786         struct drbd_conf *mdev;
1787         int i;
1788
1789         if (old_conf && tconn->cstate == C_WF_REPORT_PARAMS && tconn->agreed_pro_version < 100) {
1790                 if (new_conf->wire_protocol != old_conf->wire_protocol)
1791                         return ERR_NEED_APV_100;
1792
1793                 if (new_conf->two_primaries != old_conf->two_primaries)
1794                         return ERR_NEED_APV_100;
1795
1796                 if (!new_conf->integrity_alg != !old_conf->integrity_alg)
1797                         return ERR_NEED_APV_100;
1798
1799                 if (strcmp(new_conf->integrity_alg, old_conf->integrity_alg))
1800                         return ERR_NEED_APV_100;
1801         }
1802
1803         if (!new_conf->two_primaries &&
1804             conn_highest_role(tconn) == R_PRIMARY &&
1805             conn_highest_peer(tconn) == R_PRIMARY)
1806                 return ERR_NEED_ALLOW_TWO_PRI;
1807
1808         if (new_conf->two_primaries &&
1809             (new_conf->wire_protocol != DRBD_PROT_C))
1810                 return ERR_NOT_PROTO_C;
1811
1812         idr_for_each_entry(&tconn->volumes, mdev, i) {
1813                 if (get_ldev(mdev)) {
1814                         enum drbd_fencing_p fp = rcu_dereference(mdev->ldev->disk_conf)->fencing;
1815                         put_ldev(mdev);
1816                         if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
1817                                 return ERR_STONITH_AND_PROT_A;
1818                 }
1819                 if (mdev->state.role == R_PRIMARY && new_conf->discard_my_data)
1820                         return ERR_DISCARD_IMPOSSIBLE;
1821         }
1822
1823         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A)
1824                 return ERR_CONG_NOT_PROTO_A;
1825
1826         return NO_ERROR;
1827 }
1828
1829 static enum drbd_ret_code
1830 check_net_options(struct drbd_tconn *tconn, struct net_conf *new_conf)
1831 {
1832         static enum drbd_ret_code rv;
1833         struct drbd_conf *mdev;
1834         int i;
1835
1836         rcu_read_lock();
1837         rv = _check_net_options(tconn, rcu_dereference(tconn->net_conf), new_conf);
1838         rcu_read_unlock();
1839
1840         /* tconn->volumes protected by genl_lock() here */
1841         idr_for_each_entry(&tconn->volumes, mdev, i) {
1842                 if (!mdev->bitmap) {
1843                         if(drbd_bm_init(mdev))
1844                                 return ERR_NOMEM;
1845                 }
1846         }
1847
1848         return rv;
1849 }
1850
1851 struct crypto {
1852         struct crypto_hash *verify_tfm;
1853         struct crypto_hash *csums_tfm;
1854         struct crypto_hash *cram_hmac_tfm;
1855         struct crypto_hash *integrity_tfm;
1856 };
1857
1858 static int
1859 alloc_hash(struct crypto_hash **tfm, char *tfm_name, int err_alg)
1860 {
1861         if (!tfm_name[0])
1862                 return NO_ERROR;
1863
1864         *tfm = crypto_alloc_hash(tfm_name, 0, CRYPTO_ALG_ASYNC);
1865         if (IS_ERR(*tfm)) {
1866                 *tfm = NULL;
1867                 return err_alg;
1868         }
1869
1870         return NO_ERROR;
1871 }
1872
1873 static enum drbd_ret_code
1874 alloc_crypto(struct crypto *crypto, struct net_conf *new_conf)
1875 {
1876         char hmac_name[CRYPTO_MAX_ALG_NAME];
1877         enum drbd_ret_code rv;
1878
1879         rv = alloc_hash(&crypto->csums_tfm, new_conf->csums_alg,
1880                        ERR_CSUMS_ALG);
1881         if (rv != NO_ERROR)
1882                 return rv;
1883         rv = alloc_hash(&crypto->verify_tfm, new_conf->verify_alg,
1884                        ERR_VERIFY_ALG);
1885         if (rv != NO_ERROR)
1886                 return rv;
1887         rv = alloc_hash(&crypto->integrity_tfm, new_conf->integrity_alg,
1888                        ERR_INTEGRITY_ALG);
1889         if (rv != NO_ERROR)
1890                 return rv;
1891         if (new_conf->cram_hmac_alg[0] != 0) {
1892                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1893                          new_conf->cram_hmac_alg);
1894
1895                 rv = alloc_hash(&crypto->cram_hmac_tfm, hmac_name,
1896                                ERR_AUTH_ALG);
1897         }
1898
1899         return rv;
1900 }
1901
1902 static void free_crypto(struct crypto *crypto)
1903 {
1904         crypto_free_hash(crypto->cram_hmac_tfm);
1905         crypto_free_hash(crypto->integrity_tfm);
1906         crypto_free_hash(crypto->csums_tfm);
1907         crypto_free_hash(crypto->verify_tfm);
1908 }
1909
1910 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
1911 {
1912         enum drbd_ret_code retcode;
1913         struct drbd_tconn *tconn;
1914         struct net_conf *old_conf, *new_conf = NULL;
1915         int err;
1916         int ovr; /* online verify running */
1917         int rsr; /* re-sync running */
1918         struct crypto crypto = { };
1919
1920         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONNECTION);
1921         if (!adm_ctx.reply_skb)
1922                 return retcode;
1923         if (retcode != NO_ERROR)
1924                 goto out;
1925
1926         tconn = adm_ctx.tconn;
1927
1928         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1929         if (!new_conf) {
1930                 retcode = ERR_NOMEM;
1931                 goto out;
1932         }
1933
1934         conn_reconfig_start(tconn);
1935
1936         mutex_lock(&tconn->data.mutex);
1937         mutex_lock(&tconn->conf_update);
1938         old_conf = tconn->net_conf;
1939
1940         if (!old_conf) {
1941                 drbd_msg_put_info("net conf missing, try connect");
1942                 retcode = ERR_INVALID_REQUEST;
1943                 goto fail;
1944         }
1945
1946         *new_conf = *old_conf;
1947         if (should_set_defaults(info))
1948                 set_net_conf_defaults(new_conf);
1949
1950         err = net_conf_from_attrs_for_change(new_conf, info);
1951         if (err && err != -ENOMSG) {
1952                 retcode = ERR_MANDATORY_TAG;
1953                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1954                 goto fail;
1955         }
1956
1957         retcode = check_net_options(tconn, new_conf);
1958         if (retcode != NO_ERROR)
1959                 goto fail;
1960
1961         /* re-sync running */
1962         rsr = conn_resync_running(tconn);
1963         if (rsr && strcmp(new_conf->csums_alg, old_conf->csums_alg)) {
1964                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1965                 goto fail;
1966         }
1967
1968         /* online verify running */
1969         ovr = conn_ov_running(tconn);
1970         if (ovr && strcmp(new_conf->verify_alg, old_conf->verify_alg)) {
1971                 retcode = ERR_VERIFY_RUNNING;
1972                 goto fail;
1973         }
1974
1975         retcode = alloc_crypto(&crypto, new_conf);
1976         if (retcode != NO_ERROR)
1977                 goto fail;
1978
1979         rcu_assign_pointer(tconn->net_conf, new_conf);
1980
1981         if (!rsr) {
1982                 crypto_free_hash(tconn->csums_tfm);
1983                 tconn->csums_tfm = crypto.csums_tfm;
1984                 crypto.csums_tfm = NULL;
1985         }
1986         if (!ovr) {
1987                 crypto_free_hash(tconn->verify_tfm);
1988                 tconn->verify_tfm = crypto.verify_tfm;
1989                 crypto.verify_tfm = NULL;
1990         }
1991
1992         crypto_free_hash(tconn->integrity_tfm);
1993         tconn->integrity_tfm = crypto.integrity_tfm;
1994         if (tconn->cstate >= C_WF_REPORT_PARAMS && tconn->agreed_pro_version >= 100)
1995                 /* Do this without trying to take tconn->data.mutex again.  */
1996                 __drbd_send_protocol(tconn, P_PROTOCOL_UPDATE);
1997
1998         crypto_free_hash(tconn->cram_hmac_tfm);
1999         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
2000
2001         mutex_unlock(&tconn->conf_update);
2002         mutex_unlock(&tconn->data.mutex);
2003         synchronize_rcu();
2004         kfree(old_conf);
2005
2006         if (tconn->cstate >= C_WF_REPORT_PARAMS)
2007                 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn)));
2008
2009         goto done;
2010
2011  fail:
2012         mutex_unlock(&tconn->conf_update);
2013         mutex_unlock(&tconn->data.mutex);
2014         free_crypto(&crypto);
2015         kfree(new_conf);
2016  done:
2017         conn_reconfig_done(tconn);
2018  out:
2019         drbd_adm_finish(info, retcode);
2020         return 0;
2021 }
2022
2023 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2024 {
2025         struct drbd_conf *mdev;
2026         struct net_conf *old_conf, *new_conf = NULL;
2027         struct crypto crypto = { };
2028         struct drbd_tconn *tconn;
2029         enum drbd_ret_code retcode;
2030         int i;
2031         int err;
2032
2033         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
2034
2035         if (!adm_ctx.reply_skb)
2036                 return retcode;
2037         if (retcode != NO_ERROR)
2038                 goto out;
2039         if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2040                 drbd_msg_put_info("connection endpoint(s) missing");
2041                 retcode = ERR_INVALID_REQUEST;
2042                 goto out;
2043         }
2044
2045         /* No need for _rcu here. All reconfiguration is
2046          * strictly serialized on genl_lock(). We are protected against
2047          * concurrent reconfiguration/addition/deletion */
2048         list_for_each_entry(tconn, &drbd_tconns, all_tconn) {
2049                 if (nla_len(adm_ctx.my_addr) == tconn->my_addr_len &&
2050                     !memcmp(nla_data(adm_ctx.my_addr), &tconn->my_addr, tconn->my_addr_len)) {
2051                         retcode = ERR_LOCAL_ADDR;
2052                         goto out;
2053                 }
2054
2055                 if (nla_len(adm_ctx.peer_addr) == tconn->peer_addr_len &&
2056                     !memcmp(nla_data(adm_ctx.peer_addr), &tconn->peer_addr, tconn->peer_addr_len)) {
2057                         retcode = ERR_PEER_ADDR;
2058                         goto out;
2059                 }
2060         }
2061
2062         tconn = adm_ctx.tconn;
2063         conn_reconfig_start(tconn);
2064
2065         if (tconn->cstate > C_STANDALONE) {
2066                 retcode = ERR_NET_CONFIGURED;
2067                 goto fail;
2068         }
2069
2070         /* allocation not in the IO path, drbdsetup / netlink process context */
2071         new_conf = kzalloc(sizeof(*new_conf), GFP_KERNEL);
2072         if (!new_conf) {
2073                 retcode = ERR_NOMEM;
2074                 goto fail;
2075         }
2076
2077         set_net_conf_defaults(new_conf);
2078
2079         err = net_conf_from_attrs(new_conf, info);
2080         if (err && err != -ENOMSG) {
2081                 retcode = ERR_MANDATORY_TAG;
2082                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2083                 goto fail;
2084         }
2085
2086         retcode = check_net_options(tconn, new_conf);
2087         if (retcode != NO_ERROR)
2088                 goto fail;
2089
2090         retcode = alloc_crypto(&crypto, new_conf);
2091         if (retcode != NO_ERROR)
2092                 goto fail;
2093
2094         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2095
2096         conn_flush_workqueue(tconn);
2097
2098         mutex_lock(&tconn->conf_update);
2099         old_conf = tconn->net_conf;
2100         if (old_conf) {
2101                 retcode = ERR_NET_CONFIGURED;
2102                 mutex_unlock(&tconn->conf_update);
2103                 goto fail;
2104         }
2105         rcu_assign_pointer(tconn->net_conf, new_conf);
2106
2107         conn_free_crypto(tconn);
2108         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
2109         tconn->integrity_tfm = crypto.integrity_tfm;
2110         tconn->csums_tfm = crypto.csums_tfm;
2111         tconn->verify_tfm = crypto.verify_tfm;
2112
2113         tconn->my_addr_len = nla_len(adm_ctx.my_addr);
2114         memcpy(&tconn->my_addr, nla_data(adm_ctx.my_addr), tconn->my_addr_len);
2115         tconn->peer_addr_len = nla_len(adm_ctx.peer_addr);
2116         memcpy(&tconn->peer_addr, nla_data(adm_ctx.peer_addr), tconn->peer_addr_len);
2117
2118         mutex_unlock(&tconn->conf_update);
2119
2120         rcu_read_lock();
2121         idr_for_each_entry(&tconn->volumes, mdev, i) {
2122                 mdev->send_cnt = 0;
2123                 mdev->recv_cnt = 0;
2124         }
2125         rcu_read_unlock();
2126
2127         retcode = conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2128
2129         conn_reconfig_done(tconn);
2130         drbd_adm_finish(info, retcode);
2131         return 0;
2132
2133 fail:
2134         free_crypto(&crypto);
2135         kfree(new_conf);
2136
2137         conn_reconfig_done(tconn);
2138 out:
2139         drbd_adm_finish(info, retcode);
2140         return 0;
2141 }
2142
2143 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
2144 {
2145         enum drbd_state_rv rv;
2146
2147         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2148                         force ? CS_HARD : 0);
2149
2150         switch (rv) {
2151         case SS_NOTHING_TO_DO:
2152                 break;
2153         case SS_ALREADY_STANDALONE:
2154                 return SS_SUCCESS;
2155         case SS_PRIMARY_NOP:
2156                 /* Our state checking code wants to see the peer outdated. */
2157                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2158                                                 pdsk, D_OUTDATED), CS_VERBOSE);
2159                 break;
2160         case SS_CW_FAILED_BY_PEER:
2161                 /* The peer probably wants to see us outdated. */
2162                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2163                                                         disk, D_OUTDATED), 0);
2164                 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2165                         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2166                                         CS_HARD);
2167                 }
2168                 break;
2169         default:;
2170                 /* no special handling necessary */
2171         }
2172
2173         if (rv >= SS_SUCCESS) {
2174                 enum drbd_state_rv rv2;
2175                 /* No one else can reconfigure the network while I am here.
2176                  * The state handling only uses drbd_thread_stop_nowait(),
2177                  * we want to really wait here until the receiver is no more.
2178                  */
2179                 drbd_thread_stop(&adm_ctx.tconn->receiver);
2180
2181                 /* Race breaker.  This additional state change request may be
2182                  * necessary, if this was a forced disconnect during a receiver
2183                  * restart.  We may have "killed" the receiver thread just
2184                  * after drbdd_init() returned.  Typically, we should be
2185                  * C_STANDALONE already, now, and this becomes a no-op.
2186                  */
2187                 rv2 = conn_request_state(tconn, NS(conn, C_STANDALONE),
2188                                 CS_VERBOSE | CS_HARD);
2189                 if (rv2 < SS_SUCCESS)
2190                         conn_err(tconn,
2191                                 "unexpected rv2=%d in conn_try_disconnect()\n",
2192                                 rv2);
2193         }
2194         return rv;
2195 }
2196
2197 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2198 {
2199         struct disconnect_parms parms;
2200         struct drbd_tconn *tconn;
2201         enum drbd_state_rv rv;
2202         enum drbd_ret_code retcode;
2203         int err;
2204
2205         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONNECTION);
2206         if (!adm_ctx.reply_skb)
2207                 return retcode;
2208         if (retcode != NO_ERROR)
2209                 goto fail;
2210
2211         tconn = adm_ctx.tconn;
2212         memset(&parms, 0, sizeof(parms));
2213         if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2214                 err = disconnect_parms_from_attrs(&parms, info);
2215                 if (err) {
2216                         retcode = ERR_MANDATORY_TAG;
2217                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2218                         goto fail;
2219                 }
2220         }
2221
2222         rv = conn_try_disconnect(tconn, parms.force_disconnect);
2223         if (rv < SS_SUCCESS)
2224                 retcode = rv;  /* FIXME: Type mismatch. */
2225         else
2226                 retcode = NO_ERROR;
2227  fail:
2228         drbd_adm_finish(info, retcode);
2229         return 0;
2230 }
2231
2232 void resync_after_online_grow(struct drbd_conf *mdev)
2233 {
2234         int iass; /* I am sync source */
2235
2236         dev_info(DEV, "Resync of new storage after online grow\n");
2237         if (mdev->state.role != mdev->state.peer)
2238                 iass = (mdev->state.role == R_PRIMARY);
2239         else
2240                 iass = test_bit(RESOLVE_CONFLICTS, &mdev->tconn->flags);
2241
2242         if (iass)
2243                 drbd_start_resync(mdev, C_SYNC_SOURCE);
2244         else
2245                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2246 }
2247
2248 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2249 {
2250         struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2251         struct resize_parms rs;
2252         struct drbd_conf *mdev;
2253         enum drbd_ret_code retcode;
2254         enum determine_dev_size dd;
2255         enum dds_flags ddsf;
2256         sector_t u_size;
2257         int err;
2258
2259         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2260         if (!adm_ctx.reply_skb)
2261                 return retcode;
2262         if (retcode != NO_ERROR)
2263                 goto fail;
2264
2265         memset(&rs, 0, sizeof(struct resize_parms));
2266         if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2267                 err = resize_parms_from_attrs(&rs, info);
2268                 if (err) {
2269                         retcode = ERR_MANDATORY_TAG;
2270                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2271                         goto fail;
2272                 }
2273         }
2274
2275         mdev = adm_ctx.mdev;
2276         if (mdev->state.conn > C_CONNECTED) {
2277                 retcode = ERR_RESIZE_RESYNC;
2278                 goto fail;
2279         }
2280
2281         if (mdev->state.role == R_SECONDARY &&
2282             mdev->state.peer == R_SECONDARY) {
2283                 retcode = ERR_NO_PRIMARY;
2284                 goto fail;
2285         }
2286
2287         if (!get_ldev(mdev)) {
2288                 retcode = ERR_NO_DISK;
2289                 goto fail;
2290         }
2291
2292         if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
2293                 retcode = ERR_NEED_APV_93;
2294                 goto fail_ldev;
2295         }
2296
2297         rcu_read_lock();
2298         u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
2299         rcu_read_unlock();
2300         if (u_size != (sector_t)rs.resize_size) {
2301                 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2302                 if (!new_disk_conf) {
2303                         retcode = ERR_NOMEM;
2304                         goto fail_ldev;
2305                 }
2306         }
2307
2308         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
2309                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2310
2311         if (new_disk_conf) {
2312                 mutex_lock(&mdev->tconn->conf_update);
2313                 old_disk_conf = mdev->ldev->disk_conf;
2314                 *new_disk_conf = *old_disk_conf;
2315                 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2316                 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
2317                 mutex_unlock(&mdev->tconn->conf_update);
2318                 synchronize_rcu();
2319                 kfree(old_disk_conf);
2320         }
2321
2322         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2323         dd = drbd_determine_dev_size(mdev, ddsf);
2324         drbd_md_sync(mdev);
2325         put_ldev(mdev);
2326         if (dd == dev_size_error) {
2327                 retcode = ERR_NOMEM_BITMAP;
2328                 goto fail;
2329         }
2330
2331         if (mdev->state.conn == C_CONNECTED) {
2332                 if (dd == grew)
2333                         set_bit(RESIZE_PENDING, &mdev->flags);
2334
2335                 drbd_send_uuids(mdev);
2336                 drbd_send_sizes(mdev, 1, ddsf);
2337         }
2338
2339  fail:
2340         drbd_adm_finish(info, retcode);
2341         return 0;
2342
2343  fail_ldev:
2344         put_ldev(mdev);
2345         goto fail;
2346 }
2347
2348 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2349 {
2350         enum drbd_ret_code retcode;
2351         struct drbd_tconn *tconn;
2352         struct res_opts res_opts;
2353         int err;
2354
2355         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
2356         if (!adm_ctx.reply_skb)
2357                 return retcode;
2358         if (retcode != NO_ERROR)
2359                 goto fail;
2360         tconn = adm_ctx.tconn;
2361
2362         res_opts = tconn->res_opts;
2363         if (should_set_defaults(info))
2364                 set_res_opts_defaults(&res_opts);
2365
2366         err = res_opts_from_attrs(&res_opts, info);
2367         if (err && err != -ENOMSG) {
2368                 retcode = ERR_MANDATORY_TAG;
2369                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2370                 goto fail;
2371         }
2372
2373         err = set_resource_options(tconn, &res_opts);
2374         if (err) {
2375                 retcode = ERR_INVALID_REQUEST;
2376                 if (err == -ENOMEM)
2377                         retcode = ERR_NOMEM;
2378         }
2379
2380 fail:
2381         drbd_adm_finish(info, retcode);
2382         return 0;
2383 }
2384
2385 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2386 {
2387         struct drbd_conf *mdev;
2388         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2389
2390         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2391         if (!adm_ctx.reply_skb)
2392                 return retcode;
2393         if (retcode != NO_ERROR)
2394                 goto out;
2395
2396         mdev = adm_ctx.mdev;
2397
2398         /* If there is still bitmap IO pending, probably because of a previous
2399          * resync just being finished, wait for it before requesting a new resync.
2400          * Also wait for it's after_state_ch(). */
2401         drbd_suspend_io(mdev);
2402         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2403         drbd_flush_workqueue(mdev);
2404
2405         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2406
2407         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2408                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2409
2410         while (retcode == SS_NEED_CONNECTION) {
2411                 spin_lock_irq(&mdev->tconn->req_lock);
2412                 if (mdev->state.conn < C_CONNECTED)
2413                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2414                 spin_unlock_irq(&mdev->tconn->req_lock);
2415
2416                 if (retcode != SS_NEED_CONNECTION)
2417                         break;
2418
2419                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2420         }
2421         drbd_resume_io(mdev);
2422
2423 out:
2424         drbd_adm_finish(info, retcode);
2425         return 0;
2426 }
2427
2428 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2429                 union drbd_state mask, union drbd_state val)
2430 {
2431         enum drbd_ret_code retcode;
2432
2433         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2434         if (!adm_ctx.reply_skb)
2435                 return retcode;
2436         if (retcode != NO_ERROR)
2437                 goto out;
2438
2439         retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2440 out:
2441         drbd_adm_finish(info, retcode);
2442         return 0;
2443 }
2444
2445 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2446 {
2447         int rv;
2448
2449         rv = drbd_bmio_set_n_write(mdev);
2450         drbd_suspend_al(mdev);
2451         return rv;
2452 }
2453
2454 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2455 {
2456         int retcode; /* drbd_ret_code, drbd_state_rv */
2457         struct drbd_conf *mdev;
2458
2459         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2460         if (!adm_ctx.reply_skb)
2461                 return retcode;
2462         if (retcode != NO_ERROR)
2463                 goto out;
2464
2465         mdev = adm_ctx.mdev;
2466
2467         /* If there is still bitmap IO pending, probably because of a previous
2468          * resync just being finished, wait for it before requesting a new resync.
2469          * Also wait for it's after_state_ch(). */
2470         drbd_suspend_io(mdev);
2471         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2472         drbd_flush_workqueue(mdev);
2473
2474         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
2475         if (retcode < SS_SUCCESS) {
2476                 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
2477                         /* The peer will get a resync upon connect anyways.
2478                          * Just make that into a full resync. */
2479                         retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
2480                         if (retcode >= SS_SUCCESS) {
2481                                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
2482                                                    "set_n_write from invalidate_peer",
2483                                                    BM_LOCKED_SET_ALLOWED))
2484                                         retcode = ERR_IO_MD_DISK;
2485                         }
2486                 } else
2487                         retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
2488         }
2489         drbd_resume_io(mdev);
2490
2491 out:
2492         drbd_adm_finish(info, retcode);
2493         return 0;
2494 }
2495
2496 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2497 {
2498         enum drbd_ret_code retcode;
2499
2500         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2501         if (!adm_ctx.reply_skb)
2502                 return retcode;
2503         if (retcode != NO_ERROR)
2504                 goto out;
2505
2506         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2507                 retcode = ERR_PAUSE_IS_SET;
2508 out:
2509         drbd_adm_finish(info, retcode);
2510         return 0;
2511 }
2512
2513 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2514 {
2515         union drbd_dev_state s;
2516         enum drbd_ret_code retcode;
2517
2518         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2519         if (!adm_ctx.reply_skb)
2520                 return retcode;
2521         if (retcode != NO_ERROR)
2522                 goto out;
2523
2524         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2525                 s = adm_ctx.mdev->state;
2526                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2527                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2528                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2529                 } else {
2530                         retcode = ERR_PAUSE_IS_CLEAR;
2531                 }
2532         }
2533
2534 out:
2535         drbd_adm_finish(info, retcode);
2536         return 0;
2537 }
2538
2539 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2540 {
2541         return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2542 }
2543
2544 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2545 {
2546         struct drbd_conf *mdev;
2547         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2548
2549         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2550         if (!adm_ctx.reply_skb)
2551                 return retcode;
2552         if (retcode != NO_ERROR)
2553                 goto out;
2554
2555         mdev = adm_ctx.mdev;
2556         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2557                 drbd_uuid_new_current(mdev);
2558                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2559         }
2560         drbd_suspend_io(mdev);
2561         retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2562         if (retcode == SS_SUCCESS) {
2563                 if (mdev->state.conn < C_CONNECTED)
2564                         tl_clear(mdev->tconn);
2565                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2566                         tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2567         }
2568         drbd_resume_io(mdev);
2569
2570 out:
2571         drbd_adm_finish(info, retcode);
2572         return 0;
2573 }
2574
2575 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2576 {
2577         return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2578 }
2579
2580 int nla_put_drbd_cfg_context(struct sk_buff *skb, struct drbd_tconn *tconn, unsigned vnr)
2581 {
2582         struct nlattr *nla;
2583         nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2584         if (!nla)
2585                 goto nla_put_failure;
2586         if (vnr != VOLUME_UNSPECIFIED &&
2587             nla_put_u32(skb, T_ctx_volume, vnr))
2588                 goto nla_put_failure;
2589         if (nla_put_string(skb, T_ctx_resource_name, tconn->name))
2590                 goto nla_put_failure;
2591         if (tconn->my_addr_len &&
2592             nla_put(skb, T_ctx_my_addr, tconn->my_addr_len, &tconn->my_addr))
2593                 goto nla_put_failure;
2594         if (tconn->peer_addr_len &&
2595             nla_put(skb, T_ctx_peer_addr, tconn->peer_addr_len, &tconn->peer_addr))
2596                 goto nla_put_failure;
2597         nla_nest_end(skb, nla);
2598         return 0;
2599
2600 nla_put_failure:
2601         if (nla)
2602                 nla_nest_cancel(skb, nla);
2603         return -EMSGSIZE;
2604 }
2605
2606 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2607                 const struct sib_info *sib)
2608 {
2609         struct state_info *si = NULL; /* for sizeof(si->member); */
2610         struct net_conf *nc;
2611         struct nlattr *nla;
2612         int got_ldev;
2613         int err = 0;
2614         int exclude_sensitive;
2615
2616         /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2617          * to.  So we better exclude_sensitive information.
2618          *
2619          * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2620          * in the context of the requesting user process. Exclude sensitive
2621          * information, unless current has superuser.
2622          *
2623          * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2624          * relies on the current implementation of netlink_dump(), which
2625          * executes the dump callback successively from netlink_recvmsg(),
2626          * always in the context of the receiving process */
2627         exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2628
2629         got_ldev = get_ldev(mdev);
2630
2631         /* We need to add connection name and volume number information still.
2632          * Minor number is in drbd_genlmsghdr. */
2633         if (nla_put_drbd_cfg_context(skb, mdev->tconn, mdev->vnr))
2634                 goto nla_put_failure;
2635
2636         if (res_opts_to_skb(skb, &mdev->tconn->res_opts, exclude_sensitive))
2637                 goto nla_put_failure;
2638
2639         rcu_read_lock();
2640         if (got_ldev)
2641                 if (disk_conf_to_skb(skb, rcu_dereference(mdev->ldev->disk_conf), exclude_sensitive))
2642                         goto nla_put_failure;
2643
2644         nc = rcu_dereference(mdev->tconn->net_conf);
2645         if (nc)
2646                 err = net_conf_to_skb(skb, nc, exclude_sensitive);
2647         rcu_read_unlock();
2648         if (err)
2649                 goto nla_put_failure;
2650
2651         nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2652         if (!nla)
2653                 goto nla_put_failure;
2654         if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) ||
2655             nla_put_u32(skb, T_current_state, mdev->state.i) ||
2656             nla_put_u64(skb, T_ed_uuid, mdev->ed_uuid) ||
2657             nla_put_u64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev)) ||
2658             nla_put_u64(skb, T_send_cnt, mdev->send_cnt) ||
2659             nla_put_u64(skb, T_recv_cnt, mdev->recv_cnt) ||
2660             nla_put_u64(skb, T_read_cnt, mdev->read_cnt) ||
2661             nla_put_u64(skb, T_writ_cnt, mdev->writ_cnt) ||
2662             nla_put_u64(skb, T_al_writ_cnt, mdev->al_writ_cnt) ||
2663             nla_put_u64(skb, T_bm_writ_cnt, mdev->bm_writ_cnt) ||
2664             nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&mdev->ap_bio_cnt)) ||
2665             nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&mdev->ap_pending_cnt)) ||
2666             nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&mdev->rs_pending_cnt)))
2667                 goto nla_put_failure;
2668
2669         if (got_ldev) {
2670                 int err;
2671
2672                 spin_lock_irq(&mdev->ldev->md.uuid_lock);
2673                 err = nla_put(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2674                 spin_unlock_irq(&mdev->ldev->md.uuid_lock);
2675
2676                 if (err)
2677                         goto nla_put_failure;
2678
2679                 if (nla_put_u32(skb, T_disk_flags, mdev->ldev->md.flags) ||
2680                     nla_put_u64(skb, T_bits_total, drbd_bm_bits(mdev)) ||
2681                     nla_put_u64(skb, T_bits_oos, drbd_bm_total_weight(mdev)))
2682                         goto nla_put_failure;
2683                 if (C_SYNC_SOURCE <= mdev->state.conn &&
2684                     C_PAUSED_SYNC_T >= mdev->state.conn) {
2685                         if (nla_put_u64(skb, T_bits_rs_total, mdev->rs_total) ||
2686                             nla_put_u64(skb, T_bits_rs_failed, mdev->rs_failed))
2687                                 goto nla_put_failure;
2688                 }
2689         }
2690
2691         if (sib) {
2692                 switch(sib->sib_reason) {
2693                 case SIB_SYNC_PROGRESS:
2694                 case SIB_GET_STATUS_REPLY:
2695                         break;
2696                 case SIB_STATE_CHANGE:
2697                         if (nla_put_u32(skb, T_prev_state, sib->os.i) ||
2698                             nla_put_u32(skb, T_new_state, sib->ns.i))
2699                                 goto nla_put_failure;
2700                         break;
2701                 case SIB_HELPER_POST:
2702                         if (nla_put_u32(skb, T_helper_exit_code,
2703                                         sib->helper_exit_code))
2704                                 goto nla_put_failure;
2705                         /* fall through */
2706                 case SIB_HELPER_PRE:
2707                         if (nla_put_string(skb, T_helper, sib->helper_name))
2708                                 goto nla_put_failure;
2709                         break;
2710                 }
2711         }
2712         nla_nest_end(skb, nla);
2713
2714         if (0)
2715 nla_put_failure:
2716                 err = -EMSGSIZE;
2717         if (got_ldev)
2718                 put_ldev(mdev);
2719         return err;
2720 }
2721
2722 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2723 {
2724         enum drbd_ret_code retcode;
2725         int err;
2726
2727         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2728         if (!adm_ctx.reply_skb)
2729                 return retcode;
2730         if (retcode != NO_ERROR)
2731                 goto out;
2732
2733         err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2734         if (err) {
2735                 nlmsg_free(adm_ctx.reply_skb);
2736                 return err;
2737         }
2738 out:
2739         drbd_adm_finish(info, retcode);
2740         return 0;
2741 }
2742
2743 int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
2744 {
2745         struct drbd_conf *mdev;
2746         struct drbd_genlmsghdr *dh;
2747         struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2748         struct drbd_tconn *tconn = NULL;
2749         struct drbd_tconn *tmp;
2750         unsigned volume = cb->args[1];
2751
2752         /* Open coded, deferred, iteration:
2753          * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2754          *      idr_for_each_entry(&tconn->volumes, mdev, i) {
2755          *        ...
2756          *      }
2757          * }
2758          * where tconn is cb->args[0];
2759          * and i is cb->args[1];
2760          *
2761          * cb->args[2] indicates if we shall loop over all resources,
2762          * or just dump all volumes of a single resource.
2763          *
2764          * This may miss entries inserted after this dump started,
2765          * or entries deleted before they are reached.
2766          *
2767          * We need to make sure the mdev won't disappear while
2768          * we are looking at it, and revalidate our iterators
2769          * on each iteration.
2770          */
2771
2772         /* synchronize with conn_create()/conn_destroy() */
2773         rcu_read_lock();
2774         /* revalidate iterator position */
2775         list_for_each_entry_rcu(tmp, &drbd_tconns, all_tconn) {
2776                 if (pos == NULL) {
2777                         /* first iteration */
2778                         pos = tmp;
2779                         tconn = pos;
2780                         break;
2781                 }
2782                 if (tmp == pos) {
2783                         tconn = pos;
2784                         break;
2785                 }
2786         }
2787         if (tconn) {
2788 next_tconn:
2789                 mdev = idr_get_next(&tconn->volumes, &volume);
2790                 if (!mdev) {
2791                         /* No more volumes to dump on this tconn.
2792                          * Advance tconn iterator. */
2793                         pos = list_entry_rcu(tconn->all_tconn.next,
2794                                              struct drbd_tconn, all_tconn);
2795                         /* Did we dump any volume on this tconn yet? */
2796                         if (volume != 0) {
2797                                 /* If we reached the end of the list,
2798                                  * or only a single resource dump was requested,
2799                                  * we are done. */
2800                                 if (&pos->all_tconn == &drbd_tconns || cb->args[2])
2801                                         goto out;
2802                                 volume = 0;
2803                                 tconn = pos;
2804                                 goto next_tconn;
2805                         }
2806                 }
2807
2808                 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).pid,
2809                                 cb->nlh->nlmsg_seq, &drbd_genl_family,
2810                                 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2811                 if (!dh)
2812                         goto out;
2813
2814                 if (!mdev) {
2815                         /* This is a tconn without a single volume.
2816                          * Suprisingly enough, it may have a network
2817                          * configuration. */
2818                         struct net_conf *nc;
2819                         dh->minor = -1U;
2820                         dh->ret_code = NO_ERROR;
2821                         if (nla_put_drbd_cfg_context(skb, tconn, VOLUME_UNSPECIFIED))
2822                                 goto cancel;
2823                         nc = rcu_dereference(tconn->net_conf);
2824                         if (nc && net_conf_to_skb(skb, nc, 1) != 0)
2825                                 goto cancel;
2826                         goto done;
2827                 }
2828
2829                 D_ASSERT(mdev->vnr == volume);
2830                 D_ASSERT(mdev->tconn == tconn);
2831
2832                 dh->minor = mdev_to_minor(mdev);
2833                 dh->ret_code = NO_ERROR;
2834
2835                 if (nla_put_status_info(skb, mdev, NULL)) {
2836 cancel:
2837                         genlmsg_cancel(skb, dh);
2838                         goto out;
2839                 }
2840 done:
2841                 genlmsg_end(skb, dh);
2842         }
2843
2844 out:
2845         rcu_read_unlock();
2846         /* where to start the next iteration */
2847         cb->args[0] = (long)pos;
2848         cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2849
2850         /* No more tconns/volumes/minors found results in an empty skb.
2851          * Which will terminate the dump. */
2852         return skb->len;
2853 }
2854
2855 /*
2856  * Request status of all resources, or of all volumes within a single resource.
2857  *
2858  * This is a dump, as the answer may not fit in a single reply skb otherwise.
2859  * Which means we cannot use the family->attrbuf or other such members, because
2860  * dump is NOT protected by the genl_lock().  During dump, we only have access
2861  * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
2862  *
2863  * Once things are setup properly, we call into get_one_status().
2864  */
2865 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2866 {
2867         const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
2868         struct nlattr *nla;
2869         const char *resource_name;
2870         struct drbd_tconn *tconn;
2871         int maxtype;
2872
2873         /* Is this a followup call? */
2874         if (cb->args[0]) {
2875                 /* ... of a single resource dump,
2876                  * and the resource iterator has been advanced already? */
2877                 if (cb->args[2] && cb->args[2] != cb->args[0])
2878                         return 0; /* DONE. */
2879                 goto dump;
2880         }
2881
2882         /* First call (from netlink_dump_start).  We need to figure out
2883          * which resource(s) the user wants us to dump. */
2884         nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
2885                         nlmsg_attrlen(cb->nlh, hdrlen),
2886                         DRBD_NLA_CFG_CONTEXT);
2887
2888         /* No explicit context given.  Dump all. */
2889         if (!nla)
2890                 goto dump;
2891         maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
2892         nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
2893         if (IS_ERR(nla))
2894                 return PTR_ERR(nla);
2895         /* context given, but no name present? */
2896         if (!nla)
2897                 return -EINVAL;
2898         resource_name = nla_data(nla);
2899         tconn = conn_get_by_name(resource_name);
2900
2901         if (!tconn)
2902                 return -ENODEV;
2903
2904         kref_put(&tconn->kref, &conn_destroy); /* get_one_status() (re)validates tconn by itself */
2905
2906         /* prime iterators, and set "filter" mode mark:
2907          * only dump this tconn. */
2908         cb->args[0] = (long)tconn;
2909         /* cb->args[1] = 0; passed in this way. */
2910         cb->args[2] = (long)tconn;
2911
2912 dump:
2913         return get_one_status(skb, cb);
2914 }
2915
2916 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2917 {
2918         enum drbd_ret_code retcode;
2919         struct timeout_parms tp;
2920         int err;
2921
2922         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2923         if (!adm_ctx.reply_skb)
2924                 return retcode;
2925         if (retcode != NO_ERROR)
2926                 goto out;
2927
2928         tp.timeout_type =
2929                 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2930                 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2931                 UT_DEFAULT;
2932
2933         err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2934         if (err) {
2935                 nlmsg_free(adm_ctx.reply_skb);
2936                 return err;
2937         }
2938 out:
2939         drbd_adm_finish(info, retcode);
2940         return 0;
2941 }
2942
2943 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2944 {
2945         struct drbd_conf *mdev;
2946         enum drbd_ret_code retcode;
2947         struct start_ov_parms parms;
2948
2949         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2950         if (!adm_ctx.reply_skb)
2951                 return retcode;
2952         if (retcode != NO_ERROR)
2953                 goto out;
2954
2955         mdev = adm_ctx.mdev;
2956
2957         /* resume from last known position, if possible */
2958         parms.ov_start_sector = mdev->ov_start_sector;
2959         parms.ov_stop_sector = ULLONG_MAX;
2960         if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2961                 int err = start_ov_parms_from_attrs(&parms, info);
2962                 if (err) {
2963                         retcode = ERR_MANDATORY_TAG;
2964                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2965                         goto out;
2966                 }
2967         }
2968         /* w_make_ov_request expects position to be aligned */
2969         mdev->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1);
2970         mdev->ov_stop_sector = parms.ov_stop_sector;
2971
2972         /* If there is still bitmap IO pending, e.g. previous resync or verify
2973          * just being finished, wait for it before requesting a new resync. */
2974         drbd_suspend_io(mdev);
2975         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2976         retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2977         drbd_resume_io(mdev);
2978 out:
2979         drbd_adm_finish(info, retcode);
2980         return 0;
2981 }
2982
2983
2984 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
2985 {
2986         struct drbd_conf *mdev;
2987         enum drbd_ret_code retcode;
2988         int skip_initial_sync = 0;
2989         int err;
2990         struct new_c_uuid_parms args;
2991
2992         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2993         if (!adm_ctx.reply_skb)
2994                 return retcode;
2995         if (retcode != NO_ERROR)
2996                 goto out_nolock;
2997
2998         mdev = adm_ctx.mdev;
2999         memset(&args, 0, sizeof(args));
3000         if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
3001                 err = new_c_uuid_parms_from_attrs(&args, info);
3002                 if (err) {
3003                         retcode = ERR_MANDATORY_TAG;
3004                         drbd_msg_put_info(from_attrs_err_to_txt(err));
3005                         goto out_nolock;
3006                 }
3007         }
3008
3009         mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
3010
3011         if (!get_ldev(mdev)) {
3012                 retcode = ERR_NO_DISK;
3013                 goto out;
3014         }
3015
3016         /* this is "skip initial sync", assume to be clean */
3017         if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
3018             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
3019                 dev_info(DEV, "Preparing to skip initial sync\n");
3020                 skip_initial_sync = 1;
3021         } else if (mdev->state.conn != C_STANDALONE) {
3022                 retcode = ERR_CONNECTED;
3023                 goto out_dec;
3024         }
3025
3026         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
3027         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
3028
3029         if (args.clear_bm) {
3030                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
3031                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
3032                 if (err) {
3033                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
3034                         retcode = ERR_IO_MD_DISK;
3035                 }
3036                 if (skip_initial_sync) {
3037                         drbd_send_uuids_skip_initial_sync(mdev);
3038                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
3039                         drbd_print_uuids(mdev, "cleared bitmap UUID");
3040                         spin_lock_irq(&mdev->tconn->req_lock);
3041                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
3042                                         CS_VERBOSE, NULL);
3043                         spin_unlock_irq(&mdev->tconn->req_lock);
3044                 }
3045         }
3046
3047         drbd_md_sync(mdev);
3048 out_dec:
3049         put_ldev(mdev);
3050 out:
3051         mutex_unlock(mdev->state_mutex);
3052 out_nolock:
3053         drbd_adm_finish(info, retcode);
3054         return 0;
3055 }
3056
3057 static enum drbd_ret_code
3058 drbd_check_resource_name(const char *name)
3059 {
3060         if (!name || !name[0]) {
3061                 drbd_msg_put_info("resource name missing");
3062                 return ERR_MANDATORY_TAG;
3063         }
3064         /* if we want to use these in sysfs/configfs/debugfs some day,
3065          * we must not allow slashes */
3066         if (strchr(name, '/')) {
3067                 drbd_msg_put_info("invalid resource name");
3068                 return ERR_INVALID_REQUEST;
3069         }
3070         return NO_ERROR;
3071 }
3072
3073 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
3074 {
3075         enum drbd_ret_code retcode;
3076         struct res_opts res_opts;
3077         int err;
3078
3079         retcode = drbd_adm_prepare(skb, info, 0);
3080         if (!adm_ctx.reply_skb)
3081                 return retcode;
3082         if (retcode != NO_ERROR)
3083                 goto out;
3084
3085         set_res_opts_defaults(&res_opts);
3086         err = res_opts_from_attrs(&res_opts, info);
3087         if (err && err != -ENOMSG) {
3088                 retcode = ERR_MANDATORY_TAG;
3089                 drbd_msg_put_info(from_attrs_err_to_txt(err));
3090                 goto out;
3091         }
3092
3093         retcode = drbd_check_resource_name(adm_ctx.resource_name);
3094         if (retcode != NO_ERROR)
3095                 goto out;
3096
3097         if (adm_ctx.tconn) {
3098                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
3099                         retcode = ERR_INVALID_REQUEST;
3100                         drbd_msg_put_info("resource exists");
3101                 }
3102                 /* else: still NO_ERROR */
3103                 goto out;
3104         }
3105
3106         if (!conn_create(adm_ctx.resource_name, &res_opts))
3107                 retcode = ERR_NOMEM;
3108 out:
3109         drbd_adm_finish(info, retcode);
3110         return 0;
3111 }
3112
3113 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
3114 {
3115         struct drbd_genlmsghdr *dh = info->userhdr;
3116         enum drbd_ret_code retcode;
3117
3118         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
3119         if (!adm_ctx.reply_skb)
3120                 return retcode;
3121         if (retcode != NO_ERROR)
3122                 goto out;
3123
3124         if (dh->minor > MINORMASK) {
3125                 drbd_msg_put_info("requested minor out of range");
3126                 retcode = ERR_INVALID_REQUEST;
3127                 goto out;
3128         }
3129         if (adm_ctx.volume > DRBD_VOLUME_MAX) {
3130                 drbd_msg_put_info("requested volume id out of range");
3131                 retcode = ERR_INVALID_REQUEST;
3132                 goto out;
3133         }
3134
3135         /* drbd_adm_prepare made sure already
3136          * that mdev->tconn and mdev->vnr match the request. */
3137         if (adm_ctx.mdev) {
3138                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
3139                         retcode = ERR_MINOR_EXISTS;
3140                 /* else: still NO_ERROR */
3141                 goto out;
3142         }
3143
3144         retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
3145 out:
3146         drbd_adm_finish(info, retcode);
3147         return 0;
3148 }
3149
3150 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
3151 {
3152         if (mdev->state.disk == D_DISKLESS &&
3153             /* no need to be mdev->state.conn == C_STANDALONE &&
3154              * we may want to delete a minor from a live replication group.
3155              */
3156             mdev->state.role == R_SECONDARY) {
3157                 _drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS),
3158                                     CS_VERBOSE + CS_WAIT_COMPLETE);
3159                 idr_remove(&mdev->tconn->volumes, mdev->vnr);
3160                 idr_remove(&minors, mdev_to_minor(mdev));
3161                 del_gendisk(mdev->vdisk);
3162                 synchronize_rcu();
3163                 kref_put(&mdev->kref, &drbd_minor_destroy);
3164                 return NO_ERROR;
3165         } else
3166                 return ERR_MINOR_CONFIGURED;
3167 }
3168
3169 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
3170 {
3171         enum drbd_ret_code retcode;
3172
3173         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3174         if (!adm_ctx.reply_skb)
3175                 return retcode;
3176         if (retcode != NO_ERROR)
3177                 goto out;
3178
3179         retcode = adm_delete_minor(adm_ctx.mdev);
3180 out:
3181         drbd_adm_finish(info, retcode);
3182         return 0;
3183 }
3184
3185 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
3186 {
3187         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3188         struct drbd_conf *mdev;
3189         unsigned i;
3190
3191         retcode = drbd_adm_prepare(skb, info, 0);
3192         if (!adm_ctx.reply_skb)
3193                 return retcode;
3194         if (retcode != NO_ERROR)
3195                 goto out;
3196
3197         if (!adm_ctx.tconn) {
3198                 retcode = ERR_RES_NOT_KNOWN;
3199                 goto out;
3200         }
3201
3202         /* demote */
3203         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3204                 retcode = drbd_set_role(mdev, R_SECONDARY, 0);
3205                 if (retcode < SS_SUCCESS) {
3206                         drbd_msg_put_info("failed to demote");
3207                         goto out;
3208                 }
3209         }
3210
3211         retcode = conn_try_disconnect(adm_ctx.tconn, 0);
3212         if (retcode < SS_SUCCESS) {
3213                 drbd_msg_put_info("failed to disconnect");
3214                 goto out;
3215         }
3216
3217         /* detach */
3218         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3219                 retcode = adm_detach(mdev, 0);
3220                 if (retcode < SS_SUCCESS || retcode > NO_ERROR) {
3221                         drbd_msg_put_info("failed to detach");
3222                         goto out;
3223                 }
3224         }
3225
3226         /* If we reach this, all volumes (of this tconn) are Secondary,
3227          * Disconnected, Diskless, aka Unconfigured. Make sure all threads have
3228          * actually stopped, state handling only does drbd_thread_stop_nowait(). */
3229         drbd_thread_stop(&adm_ctx.tconn->worker);
3230
3231         /* Now, nothing can fail anymore */
3232
3233         /* delete volumes */
3234         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3235                 retcode = adm_delete_minor(mdev);
3236                 if (retcode != NO_ERROR) {
3237                         /* "can not happen" */
3238                         drbd_msg_put_info("failed to delete volume");
3239                         goto out;
3240                 }
3241         }
3242
3243         /* delete connection */
3244         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3245                 list_del_rcu(&adm_ctx.tconn->all_tconn);
3246                 synchronize_rcu();
3247                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3248
3249                 retcode = NO_ERROR;
3250         } else {
3251                 /* "can not happen" */
3252                 retcode = ERR_RES_IN_USE;
3253                 drbd_msg_put_info("failed to delete connection");
3254         }
3255         goto out;
3256 out:
3257         drbd_adm_finish(info, retcode);
3258         return 0;
3259 }
3260
3261 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
3262 {
3263         enum drbd_ret_code retcode;
3264
3265         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
3266         if (!adm_ctx.reply_skb)
3267                 return retcode;
3268         if (retcode != NO_ERROR)
3269                 goto out;
3270
3271         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3272                 list_del_rcu(&adm_ctx.tconn->all_tconn);
3273                 synchronize_rcu();
3274                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3275
3276                 retcode = NO_ERROR;
3277         } else {
3278                 retcode = ERR_RES_IN_USE;
3279         }
3280
3281         if (retcode == NO_ERROR)
3282                 drbd_thread_stop(&adm_ctx.tconn->worker);
3283 out:
3284         drbd_adm_finish(info, retcode);
3285         return 0;
3286 }
3287
3288 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
3289 {
3290         static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3291         struct sk_buff *msg;
3292         struct drbd_genlmsghdr *d_out;
3293         unsigned seq;
3294         int err = -ENOMEM;
3295
3296         seq = atomic_inc_return(&drbd_genl_seq);
3297         msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3298         if (!msg)
3299                 goto failed;
3300
3301         err = -EMSGSIZE;
3302         d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3303         if (!d_out) /* cannot happen, but anyways. */
3304                 goto nla_put_failure;
3305         d_out->minor = mdev_to_minor(mdev);
3306         d_out->ret_code = NO_ERROR;
3307
3308         if (nla_put_status_info(msg, mdev, sib))
3309                 goto nla_put_failure;
3310         genlmsg_end(msg, d_out);
3311         err = drbd_genl_multicast_events(msg, 0);
3312         /* msg has been consumed or freed in netlink_broadcast() */
3313         if (err && err != -ESRCH)
3314                 goto failed;
3315
3316         return;
3317
3318 nla_put_failure:
3319         nlmsg_free(msg);
3320 failed:
3321         dev_err(DEV, "Error %d while broadcasting event. "
3322                         "Event seq:%u sib_reason:%u\n",
3323                         err, seq, sib->sib_reason);
3324 }