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 && !capable(CAP_NET_ADMIN))
 159                return -EPERM;
 160
 161         adm_ctx.reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
 162         if (!adm_ctx.reply_skb) {
 163                 err = -ENOMEM;
 164                 goto fail;
 165         }
 166
 167         adm_ctx.reply_dh = genlmsg_put_reply(adm_ctx.reply_skb,
 168                                         info, &drbd_genl_family, 0, cmd);
 169         /* put of a few bytes into a fresh skb of >= 4k will always succeed.
 170          * but anyways */
 171         if (!adm_ctx.reply_dh) {
 172                 err = -ENOMEM;
 173                 goto fail;
 174         }
 175
 176         adm_ctx.reply_dh->minor = d_in->minor;
 177         adm_ctx.reply_dh->ret_code = NO_ERROR;
 178
 179         adm_ctx.volume = VOLUME_UNSPECIFIED;
 180         if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
 181                 struct nlattr *nla;
 182                 /* parse and validate only */
 183                 err = drbd_cfg_context_from_attrs(NULL, info);
 184                 if (err)
 185                         goto fail;
 186
 187                 /* It was present, and valid,
 188                  * copy it over to the reply skb. */
 189                 err = nla_put_nohdr(adm_ctx.reply_skb,
 190                                 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
 191                                 info->attrs[DRBD_NLA_CFG_CONTEXT]);
 192                 if (err)
 193                         goto fail;
 194
 195                 /* and assign stuff to the global adm_ctx */
 196                 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
 197                 if (nla)
 198                         adm_ctx.volume = nla_get_u32(nla);
 199                 nla = nested_attr_tb[__nla_type(T_ctx_resource_name)];
 200                 if (nla)
 201                         adm_ctx.resource_name = nla_data(nla);
 202                 adm_ctx.my_addr = nested_attr_tb[__nla_type(T_ctx_my_addr)];
 203                 adm_ctx.peer_addr = nested_attr_tb[__nla_type(T_ctx_peer_addr)];
 204                 if ((adm_ctx.my_addr &&
 205                      nla_len(adm_ctx.my_addr) > sizeof(adm_ctx.tconn->my_addr)) ||
 206                     (adm_ctx.peer_addr &&
 207                      nla_len(adm_ctx.peer_addr) > sizeof(adm_ctx.tconn->peer_addr))) {
 208                         err = -EINVAL;
 209                         goto fail;
 210                 }
 211         }
 212
 213         adm_ctx.minor = d_in->minor;
 214         adm_ctx.mdev = minor_to_mdev(d_in->minor);
 215         adm_ctx.tconn = conn_get_by_name(adm_ctx.resource_name);
 216
 217         if (!adm_ctx.mdev && (flags & DRBD_ADM_NEED_MINOR)) {
 218                 drbd_msg_put_info("unknown minor");
 219                 return ERR_MINOR_INVALID;
 220         }
 221         if (!adm_ctx.tconn && (flags & DRBD_ADM_NEED_RESOURCE)) {
 222                 drbd_msg_put_info("unknown resource");
 223                 return ERR_INVALID_REQUEST;
 224         }
 225
 226         if (flags & DRBD_ADM_NEED_CONNECTION) {
 227                 if (adm_ctx.tconn && !(flags & DRBD_ADM_NEED_RESOURCE)) {
 228                         drbd_msg_put_info("no resource name expected");
 229                         return ERR_INVALID_REQUEST;
 230                 }
 231                 if (adm_ctx.mdev) {
 232                         drbd_msg_put_info("no minor number expected");
 233                         return ERR_INVALID_REQUEST;
 234                 }
 235                 if (adm_ctx.my_addr && adm_ctx.peer_addr)
 236                         adm_ctx.tconn = conn_get_by_addrs(nla_data(adm_ctx.my_addr),
 237                                                           nla_len(adm_ctx.my_addr),
 238                                                           nla_data(adm_ctx.peer_addr),
 239                                                           nla_len(adm_ctx.peer_addr));
 240                 if (!adm_ctx.tconn) {
 241                         drbd_msg_put_info("unknown connection");
 242                         return ERR_INVALID_REQUEST;
 243                 }
 244         }
 245
 246         /* some more paranoia, if the request was over-determined */
 247         if (adm_ctx.mdev && adm_ctx.tconn &&
 248             adm_ctx.mdev->tconn != adm_ctx.tconn) {
 249                 pr_warning("request: minor=%u, resource=%s; but that minor belongs to connection %s\n",
 250                                 adm_ctx.minor, adm_ctx.resource_name,
 251                                 adm_ctx.mdev->tconn->name);
 252                 drbd_msg_put_info("minor exists in different resource");
 253                 return ERR_INVALID_REQUEST;
 254         }
 255         if (adm_ctx.mdev &&
 256             adm_ctx.volume != VOLUME_UNSPECIFIED &&
 257             adm_ctx.volume != adm_ctx.mdev->vnr) {
 258                 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
 259                                 adm_ctx.minor, adm_ctx.volume,
 260                                 adm_ctx.mdev->vnr, adm_ctx.mdev->tconn->name);
 261                 drbd_msg_put_info("minor exists as different volume");
 262                 return ERR_INVALID_REQUEST;
 263         }
 264
 265         return NO_ERROR;
 266
 267 fail:
 268         nlmsg_free(adm_ctx.reply_skb);
 269         adm_ctx.reply_skb = NULL;
 270         return err;
 271 }
 272
 273 static int drbd_adm_finish(struct genl_info *info, int retcode)
 274 {
 275         if (adm_ctx.tconn) {
 276                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
 277                 adm_ctx.tconn = NULL;
 278         }
 279
 280         if (!adm_ctx.reply_skb)
 281                 return -ENOMEM;
 282
 283         adm_ctx.reply_dh->ret_code = retcode;
 284         drbd_adm_send_reply(adm_ctx.reply_skb, info);
 285         return 0;
 286 }
 287
 288 static void setup_khelper_env(struct drbd_tconn *tconn, char **envp)
 289 {
 290         char *afs;
 291
 292         /* FIXME: A future version will not allow this case. */
 293         if (tconn->my_addr_len == 0 || tconn->peer_addr_len == 0)
 294                 return;
 295
 296         switch (((struct sockaddr *)&tconn->peer_addr)->sa_family) {
 297         case AF_INET6:
 298                 afs = "ipv6";
 299                 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
 300                          &((struct sockaddr_in6 *)&tconn->peer_addr)->sin6_addr);
 301                 break;
 302         case AF_INET:
 303                 afs = "ipv4";
 304                 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
 305                          &((struct sockaddr_in *)&tconn->peer_addr)->sin_addr);
 306                 break;
 307         default:
 308                 afs = "ssocks";
 309                 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
 310                          &((struct sockaddr_in *)&tconn->peer_addr)->sin_addr);
 311         }
 312         snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
 313 }
 314
 315 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
 316 {
 317         char *envp[] = { "HOME=/",
 318                         "TERM=linux",
 319                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 320                          (char[20]) { }, /* address family */
 321                          (char[60]) { }, /* address */
 322                         NULL };
 323         char mb[12];
 324         char *argv[] = {usermode_helper, cmd, mb, NULL };
 325         struct drbd_tconn *tconn = mdev->tconn;
 326         struct sib_info sib;
 327         int ret;
 328
 329         if (current == tconn->worker.task)
 330                 set_bit(CALLBACK_PENDING, &tconn->flags);
 331
 332         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 333         setup_khelper_env(tconn, envp);
 334
 335         /* The helper may take some time.
 336          * write out any unsynced meta data changes now */
 337         drbd_md_sync(mdev);
 338
 339         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 340         sib.sib_reason = SIB_HELPER_PRE;
 341         sib.helper_name = cmd;
 342         drbd_bcast_event(mdev, &sib);
 343         ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC);
 344         if (ret)
 345                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 346                                 usermode_helper, cmd, mb,
 347                                 (ret >> 8) & 0xff, ret);
 348         else
 349                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 350                                 usermode_helper, cmd, mb,
 351                                 (ret >> 8) & 0xff, ret);
 352         sib.sib_reason = SIB_HELPER_POST;
 353         sib.helper_exit_code = ret;
 354         drbd_bcast_event(mdev, &sib);
 355
 356         if (current == tconn->worker.task)
 357                 clear_bit(CALLBACK_PENDING, &tconn->flags);
 358
 359         if (ret < 0) /* Ignore any ERRNOs we got. */
 360                 ret = 0;
 361
 362         return ret;
 363 }
 364
 365 int conn_khelper(struct drbd_tconn *tconn, char *cmd)
 366 {
 367         char *envp[] = { "HOME=/",
 368                         "TERM=linux",
 369                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 370                          (char[20]) { }, /* address family */
 371                          (char[60]) { }, /* address */
 372                         NULL };
 373         char *argv[] = {usermode_helper, cmd, tconn->name, NULL };
 374         int ret;
 375
 376         setup_khelper_env(tconn, envp);
 377         conn_md_sync(tconn);
 378
 379         conn_info(tconn, "helper command: %s %s %s\n", usermode_helper, cmd, tconn->name);
 380         /* TODO: conn_bcast_event() ?? */
 381
 382         ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC);
 383         if (ret)
 384                 conn_warn(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
 385                           usermode_helper, cmd, tconn->name,
 386                           (ret >> 8) & 0xff, ret);
 387         else
 388                 conn_info(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
 389                           usermode_helper, cmd, tconn->name,
 390                           (ret >> 8) & 0xff, ret);
 391         /* TODO: conn_bcast_event() ?? */
 392
 393         if (ret < 0) /* Ignore any ERRNOs we got. */
 394                 ret = 0;
 395
 396         return ret;
 397 }
 398
 399 static enum drbd_fencing_p highest_fencing_policy(struct drbd_tconn *tconn)
 400 {
 401         enum drbd_fencing_p fp = FP_NOT_AVAIL;
 402         struct drbd_conf *mdev;
 403         int vnr;
 404
 405         rcu_read_lock();
 406         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
 407                 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 408                         fp = max_t(enum drbd_fencing_p, fp,
 409                                    rcu_dereference(mdev->ldev->disk_conf)->fencing);
 410                         put_ldev(mdev);
 411                 }
 412         }
 413         rcu_read_unlock();
 414
 415         return fp;
 416 }
 417
 418 bool conn_try_outdate_peer(struct drbd_tconn *tconn)
 419 {
 420         unsigned int connect_cnt;
 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         spin_lock_irq(&tconn->req_lock);
 433         connect_cnt = tconn->connect_cnt;
 434         spin_unlock_irq(&tconn->req_lock);
 435
 436         fp = highest_fencing_policy(tconn);
 437         switch (fp) {
 438         case FP_NOT_AVAIL:
 439                 conn_warn(tconn, "Not fencing peer, I'm not even Consistent myself.\n");
 440                 goto out;
 441         case FP_DONT_CARE:
 442                 return true;
 443         default: ;
 444         }
 445
 446         r = conn_khelper(tconn, "fence-peer");
 447
 448         switch ((r>>8) & 0xff) {
 449         case 3: /* peer is inconsistent */
 450                 ex_to_string = "peer is inconsistent or worse";
 451                 mask.pdsk = D_MASK;
 452                 val.pdsk = D_INCONSISTENT;
 453                 break;
 454         case 4: /* peer got outdated, or was already outdated */
 455                 ex_to_string = "peer was fenced";
 456                 mask.pdsk = D_MASK;
 457                 val.pdsk = D_OUTDATED;
 458                 break;
 459         case 5: /* peer was down */
 460                 if (conn_highest_disk(tconn) == D_UP_TO_DATE) {
 461                         /* we will(have) create(d) a new UUID anyways... */
 462                         ex_to_string = "peer is unreachable, assumed to be dead";
 463                         mask.pdsk = D_MASK;
 464                         val.pdsk = D_OUTDATED;
 465                 } else {
 466                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 467                 }
 468                 break;
 469         case 6: /* Peer is primary, voluntarily outdate myself.
 470                  * This is useful when an unconnected R_SECONDARY is asked to
 471                  * become R_PRIMARY, but finds the other peer being active. */
 472                 ex_to_string = "peer is active";
 473                 conn_warn(tconn, "Peer is primary, outdating myself.\n");
 474                 mask.disk = D_MASK;
 475                 val.disk = D_OUTDATED;
 476                 break;
 477         case 7:
 478                 if (fp != FP_STONITH)
 479                         conn_err(tconn, "fence-peer() = 7 && fencing != Stonith !!!\n");
 480                 ex_to_string = "peer was stonithed";
 481                 mask.pdsk = D_MASK;
 482                 val.pdsk = D_OUTDATED;
 483                 break;
 484         default:
 485                 /* The script is broken ... */
 486                 conn_err(tconn, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 487                 return false; /* Eventually leave IO frozen */
 488         }
 489
 490         conn_info(tconn, "fence-peer helper returned %d (%s)\n",
 491                   (r>>8) & 0xff, ex_to_string);
 492
 493  out:
 494
 495         /* Not using
 496            conn_request_state(tconn, mask, val, CS_VERBOSE);
 497            here, because we might were able to re-establish the connection in the
 498            meantime. */
 499         spin_lock_irq(&tconn->req_lock);
 500         if (tconn->cstate < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &tconn->flags)) {
 501                 if (tconn->connect_cnt != connect_cnt)
 502                         /* In case the connection was established and droped
 503                            while the fence-peer handler was running, ignore it */
 504                         conn_info(tconn, "Ignoring fence-peer exit code\n");
 505                 else
 506                         _conn_request_state(tconn, mask, val, CS_VERBOSE);
 507         }
 508         spin_unlock_irq(&tconn->req_lock);
 509
 510         return conn_highest_pdsk(tconn) <= D_OUTDATED;
 511 }
 512
 513 static int _try_outdate_peer_async(void *data)
 514 {
 515         struct drbd_tconn *tconn = (struct drbd_tconn *)data;
 516
 517         conn_try_outdate_peer(tconn);
 518
 519         kref_put(&tconn->kref, &conn_destroy);
 520         return 0;
 521 }
 522
 523 void conn_try_outdate_peer_async(struct drbd_tconn *tconn)
 524 {
 525         struct task_struct *opa;
 526
 527         kref_get(&tconn->kref);
 528         opa = kthread_run(_try_outdate_peer_async, tconn, "drbd_async_h");
 529         if (IS_ERR(opa)) {
 530                 conn_err(tconn, "out of mem, failed to invoke fence-peer helper\n");
 531                 kref_put(&tconn->kref, &conn_destroy);
 532         }
 533 }
 534
 535 enum drbd_state_rv
 536 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 537 {
 538         const int max_tries = 4;
 539         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
 540         struct net_conf *nc;
 541         int try = 0;
 542         int forced = 0;
 543         union drbd_state mask, val;
 544
 545         if (new_role == R_PRIMARY)
 546                 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
 547
 548         mutex_lock(mdev->state_mutex);
 549
 550         mask.i = 0; mask.role = R_MASK;
 551         val.i  = 0; val.role  = new_role;
 552
 553         while (try++ < max_tries) {
 554                 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 555
 556                 /* in case we first succeeded to outdate,
 557                  * but now suddenly could establish a connection */
 558                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 559                         val.pdsk = 0;
 560                         mask.pdsk = 0;
 561                         continue;
 562                 }
 563
 564                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
 565                     (mdev->state.disk < D_UP_TO_DATE &&
 566                      mdev->state.disk >= D_INCONSISTENT)) {
 567                         mask.disk = D_MASK;
 568                         val.disk  = D_UP_TO_DATE;
 569                         forced = 1;
 570                         continue;
 571                 }
 572
 573                 if (rv == SS_NO_UP_TO_DATE_DISK &&
 574                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 575                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 576
 577                         if (conn_try_outdate_peer(mdev->tconn)) {
 578                                 val.disk = D_UP_TO_DATE;
 579                                 mask.disk = D_MASK;
 580                         }
 581                         continue;
 582                 }
 583
 584                 if (rv == SS_NOTHING_TO_DO)
 585                         goto out;
 586                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
 587                         if (!conn_try_outdate_peer(mdev->tconn) && force) {
 588                                 dev_warn(DEV, "Forced into split brain situation!\n");
 589                                 mask.pdsk = D_MASK;
 590                                 val.pdsk  = D_OUTDATED;
 591
 592                         }
 593                         continue;
 594                 }
 595                 if (rv == SS_TWO_PRIMARIES) {
 596                         /* Maybe the peer is detected as dead very soon...
 597                            retry at most once more in this case. */
 598                         int timeo;
 599                         rcu_read_lock();
 600                         nc = rcu_dereference(mdev->tconn->net_conf);
 601                         timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
 602                         rcu_read_unlock();
 603                         schedule_timeout_interruptible(timeo);
 604                         if (try < max_tries)
 605                                 try = max_tries - 1;
 606                         continue;
 607                 }
 608                 if (rv < SS_SUCCESS) {
 609                         rv = _drbd_request_state(mdev, mask, val,
 610                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 611                         if (rv < SS_SUCCESS)
 612                                 goto out;
 613                 }
 614                 break;
 615         }
 616
 617         if (rv < SS_SUCCESS)
 618                 goto out;
 619
 620         if (forced)
 621                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 622
 623         /* Wait until nothing is on the fly :) */
 624         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 625
 626         /* FIXME also wait for all pending P_BARRIER_ACK? */
 627
 628         if (new_role == R_SECONDARY) {
 629                 set_disk_ro(mdev->vdisk, true);
 630                 if (get_ldev(mdev)) {
 631                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 632                         put_ldev(mdev);
 633                 }
 634         } else {
 635                 mutex_lock(&mdev->tconn->conf_update);
 636                 nc = mdev->tconn->net_conf;
 637                 if (nc)
 638                         nc->discard_my_data = 0; /* without copy; single bit op is atomic */
 639                 mutex_unlock(&mdev->tconn->conf_update);
 640
 641                 set_disk_ro(mdev->vdisk, false);
 642                 if (get_ldev(mdev)) {
 643                         if (((mdev->state.conn < C_CONNECTED ||
 644                                mdev->state.pdsk <= D_FAILED)
 645                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 646                                 drbd_uuid_new_current(mdev);
 647
 648                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 649                         put_ldev(mdev);
 650                 }
 651         }
 652
 653         /* writeout of activity log covered areas of the bitmap
 654          * to stable storage done in after state change already */
 655
 656         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 657                 /* if this was forced, we should consider sync */
 658                 if (forced)
 659                         drbd_send_uuids(mdev);
 660                 drbd_send_current_state(mdev);
 661         }
 662
 663         drbd_md_sync(mdev);
 664
 665         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 666 out:
 667         mutex_unlock(mdev->state_mutex);
 668         return rv;
 669 }
 670
 671 static const char *from_attrs_err_to_txt(int err)
 672 {
 673         return  err == -ENOMSG ? "required attribute missing" :
 674                 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
 675                 err == -EEXIST ? "can not change invariant setting" :
 676                 "invalid attribute value";
 677 }
 678
 679 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
 680 {
 681         struct set_role_parms parms;
 682         int err;
 683         enum drbd_ret_code retcode;
 684
 685         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
 686         if (!adm_ctx.reply_skb)
 687                 return retcode;
 688         if (retcode != NO_ERROR)
 689                 goto out;
 690
 691         memset(&parms, 0, sizeof(parms));
 692         if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
 693                 err = set_role_parms_from_attrs(&parms, info);
 694                 if (err) {
 695                         retcode = ERR_MANDATORY_TAG;
 696                         drbd_msg_put_info(from_attrs_err_to_txt(err));
 697                         goto out;
 698                 }
 699         }
 700
 701         if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
 702                 retcode = drbd_set_role(adm_ctx.mdev, R_PRIMARY, parms.assume_uptodate);
 703         else
 704                 retcode = drbd_set_role(adm_ctx.mdev, R_SECONDARY, 0);
 705 out:
 706         drbd_adm_finish(info, retcode);
 707         return 0;
 708 }
 709
 710 /* Initializes the md.*_offset members, so we are able to find
 711  * the on disk meta data.
 712  *
 713  * We currently have two possible layouts:
 714  * external:
 715  *   |----------- md_size_sect ------------------|
 716  *   [ 4k superblock ][ activity log ][  Bitmap  ]
 717  *   | al_offset == 8 |
 718  *   | bm_offset = al_offset + X      |
 719  *  ==> bitmap sectors = md_size_sect - bm_offset
 720  *
 721  * internal:
 722  *            |----------- md_size_sect ------------------|
 723  * [data.....][  Bitmap  ][ activity log ][ 4k superblock ]
 724  *                        | al_offset < 0 |
 725  *            | bm_offset = al_offset - Y |
 726  *  ==> bitmap sectors = Y = al_offset - bm_offset
 727  *
 728  *  Activity log size used to be fixed 32kB,
 729  *  but is about to become configurable.
 730  */
 731 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 732                                        struct drbd_backing_dev *bdev)
 733 {
 734         sector_t md_size_sect = 0;
 735         unsigned int al_size_sect = bdev->md.al_size_4k * 8;
 736
 737         bdev->md.md_offset = drbd_md_ss(bdev);
 738
 739         switch (bdev->md.meta_dev_idx) {
 740         default:
 741                 /* v07 style fixed size indexed meta data */
 742                 bdev->md.md_size_sect = MD_128MB_SECT;
 743                 bdev->md.al_offset = MD_4kB_SECT;
 744                 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
 745                 break;
 746         case DRBD_MD_INDEX_FLEX_EXT:
 747                 /* just occupy the full device; unit: sectors */
 748                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 749                 bdev->md.al_offset = MD_4kB_SECT;
 750                 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
 751                 break;
 752         case DRBD_MD_INDEX_INTERNAL:
 753         case DRBD_MD_INDEX_FLEX_INT:
 754                 /* al size is still fixed */
 755                 bdev->md.al_offset = -al_size_sect;
 756                 /* we need (slightly less than) ~ this much bitmap sectors: */
 757                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 758                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 759                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 760                 md_size_sect = ALIGN(md_size_sect, 8);
 761
 762                 /* plus the "drbd meta data super block",
 763                  * and the activity log; */
 764                 md_size_sect += MD_4kB_SECT + al_size_sect;
 765
 766                 bdev->md.md_size_sect = md_size_sect;
 767                 /* bitmap offset is adjusted by 'super' block size */
 768                 bdev->md.bm_offset   = -md_size_sect + MD_4kB_SECT;
 769                 break;
 770         }
 771 }
 772
 773 /* input size is expected to be in KB */
 774 char *ppsize(char *buf, unsigned long long size)
 775 {
 776         /* Needs 9 bytes at max including trailing NUL:
 777          * -1ULL ==> "16384 EB" */
 778         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 779         int base = 0;
 780         while (size >= 10000 && base < sizeof(units)-1) {
 781                 /* shift + round */
 782                 size = (size >> 10) + !!(size & (1<<9));
 783                 base++;
 784         }
 785         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
 786
 787         return buf;
 788 }
 789
 790 /* there is still a theoretical deadlock when called from receiver
 791  * on an D_INCONSISTENT R_PRIMARY:
 792  *  remote READ does inc_ap_bio, receiver would need to receive answer
 793  *  packet from remote to dec_ap_bio again.
 794  *  receiver receive_sizes(), comes here,
 795  *  waits for ap_bio_cnt == 0. -> deadlock.
 796  * but this cannot happen, actually, because:
 797  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 798  *  (not connected, or bad/no disk on peer):
 799  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 800  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 801  *  peer may not initiate a resize.
 802  */
 803 /* Note these are not to be confused with
 804  * drbd_adm_suspend_io/drbd_adm_resume_io,
 805  * which are (sub) state changes triggered by admin (drbdsetup),
 806  * and can be long lived.
 807  * This changes an mdev->flag, is triggered by drbd internals,
 808  * and should be short-lived. */
 809 void drbd_suspend_io(struct drbd_conf *mdev)
 810 {
 811         set_bit(SUSPEND_IO, &mdev->flags);
 812         if (drbd_suspended(mdev))
 813                 return;
 814         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 815 }
 816
 817 void drbd_resume_io(struct drbd_conf *mdev)
 818 {
 819         clear_bit(SUSPEND_IO, &mdev->flags);
 820         wake_up(&mdev->misc_wait);
 821 }
 822
 823 /**
 824  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 825  * @mdev:       DRBD device.
 826  *
 827  * Returns 0 on success, negative return values indicate errors.
 828  * You should call drbd_md_sync() after calling this function.
 829  */
 830 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 831 {
 832         sector_t prev_first_sect, prev_size; /* previous meta location */
 833         sector_t la_size_sect, u_size;
 834         sector_t size;
 835         char ppb[10];
 836
 837         int md_moved, la_size_changed;
 838         enum determine_dev_size rv = unchanged;
 839
 840         /* race:
 841          * application request passes inc_ap_bio,
 842          * but then cannot get an AL-reference.
 843          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 844          *
 845          * to avoid that:
 846          * Suspend IO right here.
 847          * still lock the act_log to not trigger ASSERTs there.
 848          */
 849         drbd_suspend_io(mdev);
 850
 851         /* no wait necessary anymore, actually we could assert that */
 852         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 853
 854         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 855         prev_size = mdev->ldev->md.md_size_sect;
 856         la_size_sect = mdev->ldev->md.la_size_sect;
 857
 858         /* TODO: should only be some assert here, not (re)init... */
 859         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 860
 861         rcu_read_lock();
 862         u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
 863         rcu_read_unlock();
 864         size = drbd_new_dev_size(mdev, mdev->ldev, u_size, flags & DDSF_FORCED);
 865
 866         if (drbd_get_capacity(mdev->this_bdev) != size ||
 867             drbd_bm_capacity(mdev) != size) {
 868                 int err;
 869                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 870                 if (unlikely(err)) {
 871                         /* currently there is only one error: ENOMEM! */
 872                         size = drbd_bm_capacity(mdev)>>1;
 873                         if (size == 0) {
 874                                 dev_err(DEV, "OUT OF MEMORY! "
 875                                     "Could not allocate bitmap!\n");
 876                         } else {
 877                                 dev_err(DEV, "BM resizing failed. "
 878                                     "Leaving size unchanged at size = %lu KB\n",
 879                                     (unsigned long)size);
 880                         }
 881                         rv = dev_size_error;
 882                 }
 883                 /* racy, see comments above. */
 884                 drbd_set_my_capacity(mdev, size);
 885                 mdev->ldev->md.la_size_sect = size;
 886                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 887                      (unsigned long long)size>>1);
 888         }
 889         if (rv == dev_size_error)
 890                 goto out;
 891
 892         la_size_changed = (la_size_sect != mdev->ldev->md.la_size_sect);
 893
 894         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 895                 || prev_size       != mdev->ldev->md.md_size_sect;
 896
 897         if (la_size_changed || md_moved) {
 898                 int err;
 899
 900                 drbd_al_shrink(mdev); /* All extents inactive. */
 901                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 902                          la_size_changed && md_moved ? "size changed and md moved" :
 903                          la_size_changed ? "size changed" : "md moved");
 904                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
 905                 err = drbd_bitmap_io(mdev, md_moved ? &drbd_bm_write_all : &drbd_bm_write,
 906                                      "size changed", BM_LOCKED_MASK);
 907                 if (err) {
 908                         rv = dev_size_error;
 909                         goto out;
 910                 }
 911                 drbd_md_mark_dirty(mdev);
 912         }
 913
 914         if (size > la_size_sect)
 915                 rv = grew;
 916         if (size < la_size_sect)
 917                 rv = shrunk;
 918 out:
 919         lc_unlock(mdev->act_log);
 920         wake_up(&mdev->al_wait);
 921         drbd_resume_io(mdev);
 922
 923         return rv;
 924 }
 925
 926 sector_t
 927 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
 928                   sector_t u_size, int assume_peer_has_space)
 929 {
 930         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 931         sector_t la_size_sect = bdev->md.la_size_sect; /* last agreed size. */
 932         sector_t m_size; /* my size */
 933         sector_t size = 0;
 934
 935         m_size = drbd_get_max_capacity(bdev);
 936
 937         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 938                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 939                 p_size = m_size;
 940         }
 941
 942         if (p_size && m_size) {
 943                 size = min_t(sector_t, p_size, m_size);
 944         } else {
 945                 if (la_size_sect) {
 946                         size = la_size_sect;
 947                         if (m_size && m_size < size)
 948                                 size = m_size;
 949                         if (p_size && p_size < size)
 950                                 size = p_size;
 951                 } else {
 952                         if (m_size)
 953                                 size = m_size;
 954                         if (p_size)
 955                                 size = p_size;
 956                 }
 957         }
 958
 959         if (size == 0)
 960                 dev_err(DEV, "Both nodes diskless!\n");
 961
 962         if (u_size) {
 963                 if (u_size > size)
 964                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 965                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 966                 else
 967                         size = u_size;
 968         }
 969
 970         return size;
 971 }
 972
 973 /**
 974  * drbd_check_al_size() - Ensures that the AL is of the right size
 975  * @mdev:       DRBD device.
 976  *
 977  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 978  * failed, and 0 on success. You should call drbd_md_sync() after you called
 979  * this function.
 980  */
 981 static int drbd_check_al_size(struct drbd_conf *mdev, struct disk_conf *dc)
 982 {
 983         struct lru_cache *n, *t;
 984         struct lc_element *e;
 985         unsigned int in_use;
 986         int i;
 987
 988         if (mdev->act_log &&
 989             mdev->act_log->nr_elements == dc->al_extents)
 990                 return 0;
 991
 992         in_use = 0;
 993         t = mdev->act_log;
 994         n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
 995                 dc->al_extents, sizeof(struct lc_element), 0);
 996
 997         if (n == NULL) {
 998                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 999                 return -ENOMEM;
1000         }
1001         spin_lock_irq(&mdev->al_lock);
1002         if (t) {
1003                 for (i = 0; i < t->nr_elements; i++) {
1004                         e = lc_element_by_index(t, i);
1005                         if (e->refcnt)
1006                                 dev_err(DEV, "refcnt(%d)==%d\n",
1007                                     e->lc_number, e->refcnt);
1008                         in_use += e->refcnt;
1009                 }
1010         }
1011         if (!in_use)
1012                 mdev->act_log = n;
1013         spin_unlock_irq(&mdev->al_lock);
1014         if (in_use) {
1015                 dev_err(DEV, "Activity log still in use!\n");
1016                 lc_destroy(n);
1017                 return -EBUSY;
1018         } else {
1019                 if (t)
1020                         lc_destroy(t);
1021         }
1022         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
1023         return 0;
1024 }
1025
1026 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
1027 {
1028         struct request_queue * const q = mdev->rq_queue;
1029         unsigned int max_hw_sectors = max_bio_size >> 9;
1030         unsigned int max_segments = 0;
1031
1032         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1033                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
1034
1035                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1036                 rcu_read_lock();
1037                 max_segments = rcu_dereference(mdev->ldev->disk_conf)->max_bio_bvecs;
1038                 rcu_read_unlock();
1039                 put_ldev(mdev);
1040         }
1041
1042         blk_queue_logical_block_size(q, 512);
1043         blk_queue_max_hw_sectors(q, max_hw_sectors);
1044         /* This is the workaround for "bio would need to, but cannot, be split" */
1045         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1046         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
1047
1048         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1049                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
1050
1051                 blk_queue_stack_limits(q, b);
1052
1053                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
1054                         dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1055                                  q->backing_dev_info.ra_pages,
1056                                  b->backing_dev_info.ra_pages);
1057                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
1058                 }
1059                 put_ldev(mdev);
1060         }
1061 }
1062
1063 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
1064 {
1065         unsigned int now, new, local, peer;
1066
1067         now = queue_max_hw_sectors(mdev->rq_queue) << 9;
1068         local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
1069         peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
1070
1071         if (get_ldev_if_state(mdev, D_ATTACHING)) {
1072                 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
1073                 mdev->local_max_bio_size = local;
1074                 put_ldev(mdev);
1075         }
1076         local = min(local, DRBD_MAX_BIO_SIZE);
1077
1078         /* We may ignore peer limits if the peer is modern enough.
1079            Because new from 8.3.8 onwards the peer can use multiple
1080            BIOs for a single peer_request */
1081         if (mdev->state.conn >= C_CONNECTED) {
1082                 if (mdev->tconn->agreed_pro_version < 94)
1083                         peer = min( mdev->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1084                         /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1085                 else if (mdev->tconn->agreed_pro_version == 94)
1086                         peer = DRBD_MAX_SIZE_H80_PACKET;
1087                 else if (mdev->tconn->agreed_pro_version < 100)
1088                         peer = DRBD_MAX_BIO_SIZE_P95;  /* drbd 8.3.8 onwards, before 8.4.0 */
1089                 else
1090                         peer = DRBD_MAX_BIO_SIZE;
1091         }
1092
1093         new = min(local, peer);
1094
1095         if (mdev->state.role == R_PRIMARY && new < now)
1096                 dev_err(DEV, "ASSERT FAILED new < now; (%u < %u)\n", new, now);
1097
1098         if (new != now)
1099                 dev_info(DEV, "max BIO size = %u\n", new);
1100
1101         drbd_setup_queue_param(mdev, new);
1102 }
1103
1104 /* Starts the worker thread */
1105 static void conn_reconfig_start(struct drbd_tconn *tconn)
1106 {
1107         drbd_thread_start(&tconn->worker);
1108         conn_flush_workqueue(tconn);
1109 }
1110
1111 /* if still unconfigured, stops worker again. */
1112 static void conn_reconfig_done(struct drbd_tconn *tconn)
1113 {
1114         bool stop_threads;
1115         spin_lock_irq(&tconn->req_lock);
1116         stop_threads = conn_all_vols_unconf(tconn) &&
1117                 tconn->cstate == C_STANDALONE;
1118         spin_unlock_irq(&tconn->req_lock);
1119         if (stop_threads) {
1120                 /* asender is implicitly stopped by receiver
1121                  * in conn_disconnect() */
1122                 drbd_thread_stop(&tconn->receiver);
1123                 drbd_thread_stop(&tconn->worker);
1124         }
1125 }
1126
1127 /* Make sure IO is suspended before calling this function(). */
1128 static void drbd_suspend_al(struct drbd_conf *mdev)
1129 {
1130         int s = 0;
1131
1132         if (!lc_try_lock(mdev->act_log)) {
1133                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
1134                 return;
1135         }
1136
1137         drbd_al_shrink(mdev);
1138         spin_lock_irq(&mdev->tconn->req_lock);
1139         if (mdev->state.conn < C_CONNECTED)
1140                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
1141         spin_unlock_irq(&mdev->tconn->req_lock);
1142         lc_unlock(mdev->act_log);
1143
1144         if (s)
1145                 dev_info(DEV, "Suspended AL updates\n");
1146 }
1147
1148
1149 static bool should_set_defaults(struct genl_info *info)
1150 {
1151         unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1152         return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1153 }
1154
1155 static unsigned int drbd_al_extents_max(struct drbd_backing_dev *bdev)
1156 {
1157         /* This is limited by 16 bit "slot" numbers,
1158          * and by available on-disk context storage.
1159          *
1160          * Also (u16)~0 is special (denotes a "free" extent).
1161          *
1162          * One transaction occupies one 4kB on-disk block,
1163          * we have n such blocks in the on disk ring buffer,
1164          * the "current" transaction may fail (n-1),
1165          * and there is 919 slot numbers context information per transaction.
1166          *
1167          * 72 transaction blocks amounts to more than 2**16 context slots,
1168          * so cap there first.
1169          */
1170         const unsigned int max_al_nr = DRBD_AL_EXTENTS_MAX;
1171         const unsigned int sufficient_on_disk =
1172                 (max_al_nr + AL_CONTEXT_PER_TRANSACTION -1)
1173                 /AL_CONTEXT_PER_TRANSACTION;
1174
1175         unsigned int al_size_4k = bdev->md.al_size_4k;
1176
1177         if (al_size_4k > sufficient_on_disk)
1178                 return max_al_nr;
1179
1180         return (al_size_4k - 1) * AL_CONTEXT_PER_TRANSACTION;
1181 }
1182
1183 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1184 {
1185         enum drbd_ret_code retcode;
1186         struct drbd_conf *mdev;
1187         struct disk_conf *new_disk_conf, *old_disk_conf;
1188         struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1189         int err, fifo_size;
1190
1191         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1192         if (!adm_ctx.reply_skb)
1193                 return retcode;
1194         if (retcode != NO_ERROR)
1195                 goto out;
1196
1197         mdev = adm_ctx.mdev;
1198
1199         /* we also need a disk
1200          * to change the options on */
1201         if (!get_ldev(mdev)) {
1202                 retcode = ERR_NO_DISK;
1203                 goto out;
1204         }
1205
1206         new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1207         if (!new_disk_conf) {
1208                 retcode = ERR_NOMEM;
1209                 goto fail;
1210         }
1211
1212         mutex_lock(&mdev->tconn->conf_update);
1213         old_disk_conf = mdev->ldev->disk_conf;
1214         *new_disk_conf = *old_disk_conf;
1215         if (should_set_defaults(info))
1216                 set_disk_conf_defaults(new_disk_conf);
1217
1218         err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1219         if (err && err != -ENOMSG) {
1220                 retcode = ERR_MANDATORY_TAG;
1221                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1222         }
1223
1224         if (!expect(new_disk_conf->resync_rate >= 1))
1225                 new_disk_conf->resync_rate = 1;
1226
1227         if (new_disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1228                 new_disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1229         if (new_disk_conf->al_extents > drbd_al_extents_max(mdev->ldev))
1230                 new_disk_conf->al_extents = drbd_al_extents_max(mdev->ldev);
1231
1232         if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1233                 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1234
1235         fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1236         if (fifo_size != mdev->rs_plan_s->size) {
1237                 new_plan = fifo_alloc(fifo_size);
1238                 if (!new_plan) {
1239                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1240                         retcode = ERR_NOMEM;
1241                         goto fail_unlock;
1242                 }
1243         }
1244
1245         drbd_suspend_io(mdev);
1246         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1247         drbd_al_shrink(mdev);
1248         err = drbd_check_al_size(mdev, new_disk_conf);
1249         lc_unlock(mdev->act_log);
1250         wake_up(&mdev->al_wait);
1251         drbd_resume_io(mdev);
1252
1253         if (err) {
1254                 retcode = ERR_NOMEM;
1255                 goto fail_unlock;
1256         }
1257
1258         write_lock_irq(&global_state_lock);
1259         retcode = drbd_resync_after_valid(mdev, new_disk_conf->resync_after);
1260         if (retcode == NO_ERROR) {
1261                 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
1262                 drbd_resync_after_changed(mdev);
1263         }
1264         write_unlock_irq(&global_state_lock);
1265
1266         if (retcode != NO_ERROR)
1267                 goto fail_unlock;
1268
1269         if (new_plan) {
1270                 old_plan = mdev->rs_plan_s;
1271                 rcu_assign_pointer(mdev->rs_plan_s, new_plan);
1272         }
1273
1274         mutex_unlock(&mdev->tconn->conf_update);
1275
1276         if (new_disk_conf->al_updates)
1277                 mdev->ldev->md.flags &= ~MDF_AL_DISABLED;
1278         else
1279                 mdev->ldev->md.flags |= MDF_AL_DISABLED;
1280
1281         if (new_disk_conf->md_flushes)
1282                 clear_bit(MD_NO_FUA, &mdev->flags);
1283         else
1284                 set_bit(MD_NO_FUA, &mdev->flags);
1285
1286         drbd_bump_write_ordering(mdev->tconn, WO_bdev_flush);
1287
1288         drbd_md_sync(mdev);
1289
1290         if (mdev->state.conn >= C_CONNECTED)
1291                 drbd_send_sync_param(mdev);
1292
1293         synchronize_rcu();
1294         kfree(old_disk_conf);
1295         kfree(old_plan);
1296         mod_timer(&mdev->request_timer, jiffies + HZ);
1297         goto success;
1298
1299 fail_unlock:
1300         mutex_unlock(&mdev->tconn->conf_update);
1301  fail:
1302         kfree(new_disk_conf);
1303         kfree(new_plan);
1304 success:
1305         put_ldev(mdev);
1306  out:
1307         drbd_adm_finish(info, retcode);
1308         return 0;
1309 }
1310
1311 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1312 {
1313         struct drbd_conf *mdev;
1314         int err;
1315         enum drbd_ret_code retcode;
1316         enum determine_dev_size dd;
1317         sector_t max_possible_sectors;
1318         sector_t min_md_device_sectors;
1319         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1320         struct disk_conf *new_disk_conf = NULL;
1321         struct block_device *bdev;
1322         struct lru_cache *resync_lru = NULL;
1323         struct fifo_buffer *new_plan = NULL;
1324         union drbd_state ns, os;
1325         enum drbd_state_rv rv;
1326         struct net_conf *nc;
1327
1328         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1329         if (!adm_ctx.reply_skb)
1330                 return retcode;
1331         if (retcode != NO_ERROR)
1332                 goto finish;
1333
1334         mdev = adm_ctx.mdev;
1335         conn_reconfig_start(mdev->tconn);
1336
1337         /* if you want to reconfigure, please tear down first */
1338         if (mdev->state.disk > D_DISKLESS) {
1339                 retcode = ERR_DISK_CONFIGURED;
1340                 goto fail;
1341         }
1342         /* It may just now have detached because of IO error.  Make sure
1343          * drbd_ldev_destroy is done already, we may end up here very fast,
1344          * e.g. if someone calls attach from the on-io-error handler,
1345          * to realize a "hot spare" feature (not that I'd recommend that) */
1346         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1347
1348         /* make sure there is no leftover from previous force-detach attempts */
1349         clear_bit(FORCE_DETACH, &mdev->flags);
1350         clear_bit(WAS_IO_ERROR, &mdev->flags);
1351         clear_bit(WAS_READ_ERROR, &mdev->flags);
1352
1353         /* and no leftover from previously aborted resync or verify, either */
1354         mdev->rs_total = 0;
1355         mdev->rs_failed = 0;
1356         atomic_set(&mdev->rs_pending_cnt, 0);
1357
1358         /* allocation not in the IO path, drbdsetup context */
1359         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1360         if (!nbc) {
1361                 retcode = ERR_NOMEM;
1362                 goto fail;
1363         }
1364         spin_lock_init(&nbc->md.uuid_lock);
1365
1366         new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1367         if (!new_disk_conf) {
1368                 retcode = ERR_NOMEM;
1369                 goto fail;
1370         }
1371         nbc->disk_conf = new_disk_conf;
1372
1373         set_disk_conf_defaults(new_disk_conf);
1374         err = disk_conf_from_attrs(new_disk_conf, info);
1375         if (err) {
1376                 retcode = ERR_MANDATORY_TAG;
1377                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1378                 goto fail;
1379         }
1380
1381         if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1382                 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1383
1384         new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1385         if (!new_plan) {
1386                 retcode = ERR_NOMEM;
1387                 goto fail;
1388         }
1389
1390         if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1391                 retcode = ERR_MD_IDX_INVALID;
1392                 goto fail;
1393         }
1394
1395         write_lock_irq(&global_state_lock);
1396         retcode = drbd_resync_after_valid(mdev, new_disk_conf->resync_after);
1397         write_unlock_irq(&global_state_lock);
1398         if (retcode != NO_ERROR)
1399                 goto fail;
1400
1401         rcu_read_lock();
1402         nc = rcu_dereference(mdev->tconn->net_conf);
1403         if (nc) {
1404                 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1405                         rcu_read_unlock();
1406                         retcode = ERR_STONITH_AND_PROT_A;
1407                         goto fail;
1408                 }
1409         }
1410         rcu_read_unlock();
1411
1412         bdev = blkdev_get_by_path(new_disk_conf->backing_dev,
1413                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1414         if (IS_ERR(bdev)) {
1415                 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->backing_dev,
1416                         PTR_ERR(bdev));
1417                 retcode = ERR_OPEN_DISK;
1418                 goto fail;
1419         }
1420         nbc->backing_bdev = bdev;
1421
1422         /*
1423          * meta_dev_idx >= 0: external fixed size, possibly multiple
1424          * drbd sharing one meta device.  TODO in that case, paranoia
1425          * check that [md_bdev, meta_dev_idx] is not yet used by some
1426          * other drbd minor!  (if you use drbd.conf + drbdadm, that
1427          * should check it for you already; but if you don't, or
1428          * someone fooled it, we need to double check here)
1429          */
1430         bdev = blkdev_get_by_path(new_disk_conf->meta_dev,
1431                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1432                                   (new_disk_conf->meta_dev_idx < 0) ?
1433                                   (void *)mdev : (void *)drbd_m_holder);
1434         if (IS_ERR(bdev)) {
1435                 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->meta_dev,
1436                         PTR_ERR(bdev));
1437                 retcode = ERR_OPEN_MD_DISK;
1438                 goto fail;
1439         }
1440         nbc->md_bdev = bdev;
1441
1442         if ((nbc->backing_bdev == nbc->md_bdev) !=
1443             (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1444              new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1445                 retcode = ERR_MD_IDX_INVALID;
1446                 goto fail;
1447         }
1448
1449         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1450                         1, 61, sizeof(struct bm_extent),
1451                         offsetof(struct bm_extent, lce));
1452         if (!resync_lru) {
1453                 retcode = ERR_NOMEM;
1454                 goto fail;
1455         }
1456
1457         /* Read our meta data super block early.
1458          * This also sets other on-disk offsets. */
1459         retcode = drbd_md_read(mdev, nbc);
1460         if (retcode != NO_ERROR)
1461                 goto fail;
1462
1463         if (new_disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1464                 new_disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1465         if (new_disk_conf->al_extents > drbd_al_extents_max(nbc))
1466                 new_disk_conf->al_extents = drbd_al_extents_max(nbc);
1467
1468         if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1469                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1470                         (unsigned long long) drbd_get_max_capacity(nbc),
1471                         (unsigned long long) new_disk_conf->disk_size);
1472                 retcode = ERR_DISK_TOO_SMALL;
1473                 goto fail;
1474         }
1475
1476         if (new_disk_conf->meta_dev_idx < 0) {
1477                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1478                 /* at least one MB, otherwise it does not make sense */
1479                 min_md_device_sectors = (2<<10);
1480         } else {
1481                 max_possible_sectors = DRBD_MAX_SECTORS;
1482                 min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1);
1483         }
1484
1485         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1486                 retcode = ERR_MD_DISK_TOO_SMALL;
1487                 dev_warn(DEV, "refusing attach: md-device too small, "
1488                      "at least %llu sectors needed for this meta-disk type\n",
1489                      (unsigned long long) min_md_device_sectors);
1490                 goto fail;
1491         }
1492
1493         /* Make sure the new disk is big enough
1494          * (we may currently be R_PRIMARY with no local disk...) */
1495         if (drbd_get_max_capacity(nbc) <
1496             drbd_get_capacity(mdev->this_bdev)) {
1497                 retcode = ERR_DISK_TOO_SMALL;
1498                 goto fail;
1499         }
1500
1501         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1502
1503         if (nbc->known_size > max_possible_sectors) {
1504                 dev_warn(DEV, "==> truncating very big lower level device "
1505                         "to currently maximum possible %llu sectors <==\n",
1506                         (unsigned long long) max_possible_sectors);
1507                 if (new_disk_conf->meta_dev_idx >= 0)
1508                         dev_warn(DEV, "==>> using internal or flexible "
1509                                       "meta data may help <<==\n");
1510         }
1511
1512         drbd_suspend_io(mdev);
1513         /* also wait for the last barrier ack. */
1514         /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1515          * We need a way to either ignore barrier acks for barriers sent before a device
1516          * was attached, or a way to wait for all pending barrier acks to come in.
1517          * As barriers are counted per resource,
1518          * we'd need to suspend io on all devices of a resource.
1519          */
1520         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || drbd_suspended(mdev));
1521         /* and for any other previously queued work */
1522         drbd_flush_workqueue(mdev);
1523
1524         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1525         retcode = rv;  /* FIXME: Type mismatch. */
1526         drbd_resume_io(mdev);
1527         if (rv < SS_SUCCESS)
1528                 goto fail;
1529
1530         if (!get_ldev_if_state(mdev, D_ATTACHING))
1531                 goto force_diskless;
1532
1533         if (!mdev->bitmap) {
1534                 if (drbd_bm_init(mdev)) {
1535                         retcode = ERR_NOMEM;
1536                         goto force_diskless_dec;
1537                 }
1538         }
1539
1540         if (mdev->state.conn < C_CONNECTED &&
1541             mdev->state.role == R_PRIMARY &&
1542             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1543                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1544                     (unsigned long long)mdev->ed_uuid);
1545                 retcode = ERR_DATA_NOT_CURRENT;
1546                 goto force_diskless_dec;
1547         }
1548
1549         /* Since we are diskless, fix the activity log first... */
1550         if (drbd_check_al_size(mdev, new_disk_conf)) {
1551                 retcode = ERR_NOMEM;
1552                 goto force_diskless_dec;
1553         }
1554
1555         /* Prevent shrinking of consistent devices ! */
1556         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1557             drbd_new_dev_size(mdev, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1558                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1559                 retcode = ERR_DISK_TOO_SMALL;
1560                 goto force_diskless_dec;
1561         }
1562
1563         /* Reset the "barriers don't work" bits here, then force meta data to
1564          * be written, to ensure we determine if barriers are supported. */
1565         if (new_disk_conf->md_flushes)
1566                 clear_bit(MD_NO_FUA, &mdev->flags);
1567         else
1568                 set_bit(MD_NO_FUA, &mdev->flags);
1569
1570         /* Point of no return reached.
1571          * Devices and memory are no longer released by error cleanup below.
1572          * now mdev takes over responsibility, and the state engine should
1573          * clean it up somewhere.  */
1574         D_ASSERT(mdev->ldev == NULL);
1575         mdev->ldev = nbc;
1576         mdev->resync = resync_lru;
1577         mdev->rs_plan_s = new_plan;
1578         nbc = NULL;
1579         resync_lru = NULL;
1580         new_disk_conf = NULL;
1581         new_plan = NULL;
1582
1583         drbd_bump_write_ordering(mdev->tconn, WO_bdev_flush);
1584
1585         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1586                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1587         else
1588                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1589
1590         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1591             !(mdev->state.role == R_PRIMARY && mdev->tconn->susp_nod))
1592                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1593
1594         mdev->send_cnt = 0;
1595         mdev->recv_cnt = 0;
1596         mdev->read_cnt = 0;
1597         mdev->writ_cnt = 0;
1598
1599         drbd_reconsider_max_bio_size(mdev);
1600
1601         /* If I am currently not R_PRIMARY,
1602          * but meta data primary indicator is set,
1603          * I just now recover from a hard crash,
1604          * and have been R_PRIMARY before that crash.
1605          *
1606          * Now, if I had no connection before that crash
1607          * (have been degraded R_PRIMARY), chances are that
1608          * I won't find my peer now either.
1609          *
1610          * In that case, and _only_ in that case,
1611          * we use the degr-wfc-timeout instead of the default,
1612          * so we can automatically recover from a crash of a
1613          * degraded but active "cluster" after a certain timeout.
1614          */
1615         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1616         if (mdev->state.role != R_PRIMARY &&
1617              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1618             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1619                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1620
1621         dd = drbd_determine_dev_size(mdev, 0);
1622         if (dd == dev_size_error) {
1623                 retcode = ERR_NOMEM_BITMAP;
1624                 goto force_diskless_dec;
1625         } else if (dd == grew)
1626                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1627
1628         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC) ||
1629             (test_bit(CRASHED_PRIMARY, &mdev->flags) &&
1630              drbd_md_test_flag(mdev->ldev, MDF_AL_DISABLED))) {
1631                 dev_info(DEV, "Assuming that all blocks are out of sync "
1632                      "(aka FullSync)\n");
1633                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1634                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1635                         retcode = ERR_IO_MD_DISK;
1636                         goto force_diskless_dec;
1637                 }
1638         } else {
1639                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1640                         "read from attaching", BM_LOCKED_MASK)) {
1641                         retcode = ERR_IO_MD_DISK;
1642                         goto force_diskless_dec;
1643                 }
1644         }
1645
1646         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1647                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1648
1649         spin_lock_irq(&mdev->tconn->req_lock);
1650         os = drbd_read_state(mdev);
1651         ns = os;
1652         /* If MDF_CONSISTENT is not set go into inconsistent state,
1653            otherwise investigate MDF_WasUpToDate...
1654            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1655            otherwise into D_CONSISTENT state.
1656         */
1657         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1658                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1659                         ns.disk = D_CONSISTENT;
1660                 else
1661                         ns.disk = D_OUTDATED;
1662         } else {
1663                 ns.disk = D_INCONSISTENT;
1664         }
1665
1666         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1667                 ns.pdsk = D_OUTDATED;
1668
1669         rcu_read_lock();
1670         if (ns.disk == D_CONSISTENT &&
1671             (ns.pdsk == D_OUTDATED || rcu_dereference(mdev->ldev->disk_conf)->fencing == FP_DONT_CARE))
1672                 ns.disk = D_UP_TO_DATE;
1673
1674         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1675            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1676            this point, because drbd_request_state() modifies these
1677            flags. */
1678
1679         if (rcu_dereference(mdev->ldev->disk_conf)->al_updates)
1680                 mdev->ldev->md.flags &= ~MDF_AL_DISABLED;
1681         else
1682                 mdev->ldev->md.flags |= MDF_AL_DISABLED;
1683
1684         rcu_read_unlock();
1685
1686         /* In case we are C_CONNECTED postpone any decision on the new disk
1687            state after the negotiation phase. */
1688         if (mdev->state.conn == C_CONNECTED) {
1689                 mdev->new_state_tmp.i = ns.i;
1690                 ns.i = os.i;
1691                 ns.disk = D_NEGOTIATING;
1692
1693                 /* We expect to receive up-to-date UUIDs soon.
1694                    To avoid a race in receive_state, free p_uuid while
1695                    holding req_lock. I.e. atomic with the state change */
1696                 kfree(mdev->p_uuid);
1697                 mdev->p_uuid = NULL;
1698         }
1699
1700         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1701         spin_unlock_irq(&mdev->tconn->req_lock);
1702
1703         if (rv < SS_SUCCESS)
1704                 goto force_diskless_dec;
1705
1706         mod_timer(&mdev->request_timer, jiffies + HZ);
1707
1708         if (mdev->state.role == R_PRIMARY)
1709                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1710         else
1711                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1712
1713         drbd_md_mark_dirty(mdev);
1714         drbd_md_sync(mdev);
1715
1716         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1717         put_ldev(mdev);
1718         conn_reconfig_done(mdev->tconn);
1719         drbd_adm_finish(info, retcode);
1720         return 0;
1721
1722  force_diskless_dec:
1723         put_ldev(mdev);
1724  force_diskless:
1725         drbd_force_state(mdev, NS(disk, D_DISKLESS));
1726         drbd_md_sync(mdev);
1727  fail:
1728         conn_reconfig_done(mdev->tconn);
1729         if (nbc) {
1730                 if (nbc->backing_bdev)
1731                         blkdev_put(nbc->backing_bdev,
1732                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1733                 if (nbc->md_bdev)
1734                         blkdev_put(nbc->md_bdev,
1735                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1736                 kfree(nbc);
1737         }
1738         kfree(new_disk_conf);
1739         lc_destroy(resync_lru);
1740         kfree(new_plan);
1741
1742  finish:
1743         drbd_adm_finish(info, retcode);
1744         return 0;
1745 }
1746
1747 static int adm_detach(struct drbd_conf *mdev, int force)
1748 {
1749         enum drbd_state_rv retcode;
1750         int ret;
1751
1752         if (force) {
1753                 set_bit(FORCE_DETACH, &mdev->flags);
1754                 drbd_force_state(mdev, NS(disk, D_FAILED));
1755                 retcode = SS_SUCCESS;
1756                 goto out;
1757         }
1758
1759         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1760         drbd_md_get_buffer(mdev); /* make sure there is no in-flight meta-data IO */
1761         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1762         drbd_md_put_buffer(mdev);
1763         /* D_FAILED will transition to DISKLESS. */
1764         ret = wait_event_interruptible(mdev->misc_wait,
1765                         mdev->state.disk != D_FAILED);
1766         drbd_resume_io(mdev);
1767         if ((int)retcode == (int)SS_IS_DISKLESS)
1768                 retcode = SS_NOTHING_TO_DO;
1769         if (ret)
1770                 retcode = ERR_INTR;
1771 out:
1772         return retcode;
1773 }
1774
1775 /* Detaching the disk is a process in multiple stages.  First we need to lock
1776  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1777  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1778  * internal references as well.
1779  * Only then we have finally detached. */
1780 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1781 {
1782         enum drbd_ret_code retcode;
1783         struct detach_parms parms = { };
1784         int err;
1785
1786         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1787         if (!adm_ctx.reply_skb)
1788                 return retcode;
1789         if (retcode != NO_ERROR)
1790                 goto out;
1791
1792         if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
1793                 err = detach_parms_from_attrs(&parms, info);
1794                 if (err) {
1795                         retcode = ERR_MANDATORY_TAG;
1796                         drbd_msg_put_info(from_attrs_err_to_txt(err));
1797                         goto out;
1798                 }
1799         }
1800
1801         retcode = adm_detach(adm_ctx.mdev, parms.force_detach);
1802 out:
1803         drbd_adm_finish(info, retcode);
1804         return 0;
1805 }
1806
1807 static bool conn_resync_running(struct drbd_tconn *tconn)
1808 {
1809         struct drbd_conf *mdev;
1810         bool rv = false;
1811         int vnr;
1812
1813         rcu_read_lock();
1814         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1815                 if (mdev->state.conn == C_SYNC_SOURCE ||
1816                     mdev->state.conn == C_SYNC_TARGET ||
1817                     mdev->state.conn == C_PAUSED_SYNC_S ||
1818                     mdev->state.conn == C_PAUSED_SYNC_T) {
1819                         rv = true;
1820                         break;
1821                 }
1822         }
1823         rcu_read_unlock();
1824
1825         return rv;
1826 }
1827
1828 static bool conn_ov_running(struct drbd_tconn *tconn)
1829 {
1830         struct drbd_conf *mdev;
1831         bool rv = false;
1832         int vnr;
1833
1834         rcu_read_lock();
1835         idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1836                 if (mdev->state.conn == C_VERIFY_S ||
1837                     mdev->state.conn == C_VERIFY_T) {
1838                         rv = true;
1839                         break;
1840                 }
1841         }
1842         rcu_read_unlock();
1843
1844         return rv;
1845 }
1846
1847 static enum drbd_ret_code
1848 _check_net_options(struct drbd_tconn *tconn, struct net_conf *old_conf, struct net_conf *new_conf)
1849 {
1850         struct drbd_conf *mdev;
1851         int i;
1852
1853         if (old_conf && tconn->cstate == C_WF_REPORT_PARAMS && tconn->agreed_pro_version < 100) {
1854                 if (new_conf->wire_protocol != old_conf->wire_protocol)
1855                         return ERR_NEED_APV_100;
1856
1857                 if (new_conf->two_primaries != old_conf->two_primaries)
1858                         return ERR_NEED_APV_100;
1859
1860                 if (strcmp(new_conf->integrity_alg, old_conf->integrity_alg))
1861                         return ERR_NEED_APV_100;
1862         }
1863
1864         if (!new_conf->two_primaries &&
1865             conn_highest_role(tconn) == R_PRIMARY &&
1866             conn_highest_peer(tconn) == R_PRIMARY)
1867                 return ERR_NEED_ALLOW_TWO_PRI;
1868
1869         if (new_conf->two_primaries &&
1870             (new_conf->wire_protocol != DRBD_PROT_C))
1871                 return ERR_NOT_PROTO_C;
1872
1873         idr_for_each_entry(&tconn->volumes, mdev, i) {
1874                 if (get_ldev(mdev)) {
1875                         enum drbd_fencing_p fp = rcu_dereference(mdev->ldev->disk_conf)->fencing;
1876                         put_ldev(mdev);
1877                         if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
1878                                 return ERR_STONITH_AND_PROT_A;
1879                 }
1880                 if (mdev->state.role == R_PRIMARY && new_conf->discard_my_data)
1881                         return ERR_DISCARD_IMPOSSIBLE;
1882         }
1883
1884         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A)
1885                 return ERR_CONG_NOT_PROTO_A;
1886
1887         return NO_ERROR;
1888 }
1889
1890 static enum drbd_ret_code
1891 check_net_options(struct drbd_tconn *tconn, struct net_conf *new_conf)
1892 {
1893         static enum drbd_ret_code rv;
1894         struct drbd_conf *mdev;
1895         int i;
1896
1897         rcu_read_lock();
1898         rv = _check_net_options(tconn, rcu_dereference(tconn->net_conf), new_conf);
1899         rcu_read_unlock();
1900
1901         /* tconn->volumes protected by genl_lock() here */
1902         idr_for_each_entry(&tconn->volumes, mdev, i) {
1903                 if (!mdev->bitmap) {
1904                         if(drbd_bm_init(mdev))
1905                                 return ERR_NOMEM;
1906                 }
1907         }
1908
1909         return rv;
1910 }
1911
1912 struct crypto {
1913         struct crypto_hash *verify_tfm;
1914         struct crypto_hash *csums_tfm;
1915         struct crypto_hash *cram_hmac_tfm;
1916         struct crypto_hash *integrity_tfm;
1917 };
1918
1919 static int
1920 alloc_hash(struct crypto_hash **tfm, char *tfm_name, int err_alg)
1921 {
1922         if (!tfm_name[0])
1923                 return NO_ERROR;
1924
1925         *tfm = crypto_alloc_hash(tfm_name, 0, CRYPTO_ALG_ASYNC);
1926         if (IS_ERR(*tfm)) {
1927                 *tfm = NULL;
1928                 return err_alg;
1929         }
1930
1931         return NO_ERROR;
1932 }
1933
1934 static enum drbd_ret_code
1935 alloc_crypto(struct crypto *crypto, struct net_conf *new_conf)
1936 {
1937         char hmac_name[CRYPTO_MAX_ALG_NAME];
1938         enum drbd_ret_code rv;
1939
1940         rv = alloc_hash(&crypto->csums_tfm, new_conf->csums_alg,
1941                        ERR_CSUMS_ALG);
1942         if (rv != NO_ERROR)
1943                 return rv;
1944         rv = alloc_hash(&crypto->verify_tfm, new_conf->verify_alg,
1945                        ERR_VERIFY_ALG);
1946         if (rv != NO_ERROR)
1947                 return rv;
1948         rv = alloc_hash(&crypto->integrity_tfm, new_conf->integrity_alg,
1949                        ERR_INTEGRITY_ALG);
1950         if (rv != NO_ERROR)
1951                 return rv;
1952         if (new_conf->cram_hmac_alg[0] != 0) {
1953                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1954                          new_conf->cram_hmac_alg);
1955
1956                 rv = alloc_hash(&crypto->cram_hmac_tfm, hmac_name,
1957                                ERR_AUTH_ALG);
1958         }
1959
1960         return rv;
1961 }
1962
1963 static void free_crypto(struct crypto *crypto)
1964 {
1965         crypto_free_hash(crypto->cram_hmac_tfm);
1966         crypto_free_hash(crypto->integrity_tfm);
1967         crypto_free_hash(crypto->csums_tfm);
1968         crypto_free_hash(crypto->verify_tfm);
1969 }
1970
1971 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
1972 {
1973         enum drbd_ret_code retcode;
1974         struct drbd_tconn *tconn;
1975         struct net_conf *old_conf, *new_conf = NULL;
1976         int err;
1977         int ovr; /* online verify running */
1978         int rsr; /* re-sync running */
1979         struct crypto crypto = { };
1980
1981         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONNECTION);
1982         if (!adm_ctx.reply_skb)
1983                 return retcode;
1984         if (retcode != NO_ERROR)
1985                 goto out;
1986
1987         tconn = adm_ctx.tconn;
1988
1989         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1990         if (!new_conf) {
1991                 retcode = ERR_NOMEM;
1992                 goto out;
1993         }
1994
1995         conn_reconfig_start(tconn);
1996
1997         mutex_lock(&tconn->data.mutex);
1998         mutex_lock(&tconn->conf_update);
1999         old_conf = tconn->net_conf;
2000
2001         if (!old_conf) {
2002                 drbd_msg_put_info("net conf missing, try connect");
2003                 retcode = ERR_INVALID_REQUEST;
2004                 goto fail;
2005         }
2006
2007         *new_conf = *old_conf;
2008         if (should_set_defaults(info))
2009                 set_net_conf_defaults(new_conf);
2010
2011         err = net_conf_from_attrs_for_change(new_conf, info);
2012         if (err && err != -ENOMSG) {
2013                 retcode = ERR_MANDATORY_TAG;
2014                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2015                 goto fail;
2016         }
2017
2018         retcode = check_net_options(tconn, new_conf);
2019         if (retcode != NO_ERROR)
2020                 goto fail;
2021
2022         /* re-sync running */
2023         rsr = conn_resync_running(tconn);
2024         if (rsr && strcmp(new_conf->csums_alg, old_conf->csums_alg)) {
2025                 retcode = ERR_CSUMS_RESYNC_RUNNING;
2026                 goto fail;
2027         }
2028
2029         /* online verify running */
2030         ovr = conn_ov_running(tconn);
2031         if (ovr && strcmp(new_conf->verify_alg, old_conf->verify_alg)) {
2032                 retcode = ERR_VERIFY_RUNNING;
2033                 goto fail;
2034         }
2035
2036         retcode = alloc_crypto(&crypto, new_conf);
2037         if (retcode != NO_ERROR)
2038                 goto fail;
2039
2040         rcu_assign_pointer(tconn->net_conf, new_conf);
2041
2042         if (!rsr) {
2043                 crypto_free_hash(tconn->csums_tfm);
2044                 tconn->csums_tfm = crypto.csums_tfm;
2045                 crypto.csums_tfm = NULL;
2046         }
2047         if (!ovr) {
2048                 crypto_free_hash(tconn->verify_tfm);
2049                 tconn->verify_tfm = crypto.verify_tfm;
2050                 crypto.verify_tfm = NULL;
2051         }
2052
2053         crypto_free_hash(tconn->integrity_tfm);
2054         tconn->integrity_tfm = crypto.integrity_tfm;
2055         if (tconn->cstate >= C_WF_REPORT_PARAMS && tconn->agreed_pro_version >= 100)
2056                 /* Do this without trying to take tconn->data.mutex again.  */
2057                 __drbd_send_protocol(tconn, P_PROTOCOL_UPDATE);
2058
2059         crypto_free_hash(tconn->cram_hmac_tfm);
2060         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
2061
2062         mutex_unlock(&tconn->conf_update);
2063         mutex_unlock(&tconn->data.mutex);
2064         synchronize_rcu();
2065         kfree(old_conf);
2066
2067         if (tconn->cstate >= C_WF_REPORT_PARAMS)
2068                 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn)));
2069
2070         goto done;
2071
2072  fail:
2073         mutex_unlock(&tconn->conf_update);
2074         mutex_unlock(&tconn->data.mutex);
2075         free_crypto(&crypto);
2076         kfree(new_conf);
2077  done:
2078         conn_reconfig_done(tconn);
2079  out:
2080         drbd_adm_finish(info, retcode);
2081         return 0;
2082 }
2083
2084 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2085 {
2086         struct drbd_conf *mdev;
2087         struct net_conf *old_conf, *new_conf = NULL;
2088         struct crypto crypto = { };
2089         struct drbd_tconn *tconn;
2090         enum drbd_ret_code retcode;
2091         int i;
2092         int err;
2093
2094         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
2095
2096         if (!adm_ctx.reply_skb)
2097                 return retcode;
2098         if (retcode != NO_ERROR)
2099                 goto out;
2100         if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2101                 drbd_msg_put_info("connection endpoint(s) missing");
2102                 retcode = ERR_INVALID_REQUEST;
2103                 goto out;
2104         }
2105
2106         /* No need for _rcu here. All reconfiguration is
2107          * strictly serialized on genl_lock(). We are protected against
2108          * concurrent reconfiguration/addition/deletion */
2109         list_for_each_entry(tconn, &drbd_tconns, all_tconn) {
2110                 if (nla_len(adm_ctx.my_addr) == tconn->my_addr_len &&
2111                     !memcmp(nla_data(adm_ctx.my_addr), &tconn->my_addr, tconn->my_addr_len)) {
2112                         retcode = ERR_LOCAL_ADDR;
2113                         goto out;
2114                 }
2115
2116                 if (nla_len(adm_ctx.peer_addr) == tconn->peer_addr_len &&
2117                     !memcmp(nla_data(adm_ctx.peer_addr), &tconn->peer_addr, tconn->peer_addr_len)) {
2118                         retcode = ERR_PEER_ADDR;
2119                         goto out;
2120                 }
2121         }
2122
2123         tconn = adm_ctx.tconn;
2124         conn_reconfig_start(tconn);
2125
2126         if (tconn->cstate > C_STANDALONE) {
2127                 retcode = ERR_NET_CONFIGURED;
2128                 goto fail;
2129         }
2130
2131         /* allocation not in the IO path, drbdsetup / netlink process context */
2132         new_conf = kzalloc(sizeof(*new_conf), GFP_KERNEL);
2133         if (!new_conf) {
2134                 retcode = ERR_NOMEM;
2135                 goto fail;
2136         }
2137
2138         set_net_conf_defaults(new_conf);
2139
2140         err = net_conf_from_attrs(new_conf, info);
2141         if (err && err != -ENOMSG) {
2142                 retcode = ERR_MANDATORY_TAG;
2143                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2144                 goto fail;
2145         }
2146
2147         retcode = check_net_options(tconn, new_conf);
2148         if (retcode != NO_ERROR)
2149                 goto fail;
2150
2151         retcode = alloc_crypto(&crypto, new_conf);
2152         if (retcode != NO_ERROR)
2153                 goto fail;
2154
2155         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2156
2157         conn_flush_workqueue(tconn);
2158
2159         mutex_lock(&tconn->conf_update);
2160         old_conf = tconn->net_conf;
2161         if (old_conf) {
2162                 retcode = ERR_NET_CONFIGURED;
2163                 mutex_unlock(&tconn->conf_update);
2164                 goto fail;
2165         }
2166         rcu_assign_pointer(tconn->net_conf, new_conf);
2167
2168         conn_free_crypto(tconn);
2169         tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
2170         tconn->integrity_tfm = crypto.integrity_tfm;
2171         tconn->csums_tfm = crypto.csums_tfm;
2172         tconn->verify_tfm = crypto.verify_tfm;
2173
2174         tconn->my_addr_len = nla_len(adm_ctx.my_addr);
2175         memcpy(&tconn->my_addr, nla_data(adm_ctx.my_addr), tconn->my_addr_len);
2176         tconn->peer_addr_len = nla_len(adm_ctx.peer_addr);
2177         memcpy(&tconn->peer_addr, nla_data(adm_ctx.peer_addr), tconn->peer_addr_len);
2178
2179         mutex_unlock(&tconn->conf_update);
2180
2181         rcu_read_lock();
2182         idr_for_each_entry(&tconn->volumes, mdev, i) {
2183                 mdev->send_cnt = 0;
2184                 mdev->recv_cnt = 0;
2185         }
2186         rcu_read_unlock();
2187
2188         retcode = conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2189
2190         conn_reconfig_done(tconn);
2191         drbd_adm_finish(info, retcode);
2192         return 0;
2193
2194 fail:
2195         free_crypto(&crypto);
2196         kfree(new_conf);
2197
2198         conn_reconfig_done(tconn);
2199 out:
2200         drbd_adm_finish(info, retcode);
2201         return 0;
2202 }
2203
2204 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
2205 {
2206         enum drbd_state_rv rv;
2207
2208         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2209                         force ? CS_HARD : 0);
2210
2211         switch (rv) {
2212         case SS_NOTHING_TO_DO:
2213                 break;
2214         case SS_ALREADY_STANDALONE:
2215                 return SS_SUCCESS;
2216         case SS_PRIMARY_NOP:
2217                 /* Our state checking code wants to see the peer outdated. */
2218                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING, pdsk, D_OUTDATED), 0);
2219
2220                 if (rv == SS_OUTDATE_WO_CONN) /* lost connection before graceful disconnect succeeded */
2221                         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_VERBOSE);
2222
2223                 break;
2224         case SS_CW_FAILED_BY_PEER:
2225                 /* The peer probably wants to see us outdated. */
2226                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2227                                                         disk, D_OUTDATED), 0);
2228                 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2229                         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2230                                         CS_HARD);
2231                 }
2232                 break;
2233         default:;
2234                 /* no special handling necessary */
2235         }
2236
2237         if (rv >= SS_SUCCESS) {
2238                 enum drbd_state_rv rv2;
2239                 /* No one else can reconfigure the network while I am here.
2240                  * The state handling only uses drbd_thread_stop_nowait(),
2241                  * we want to really wait here until the receiver is no more.
2242                  */
2243                 drbd_thread_stop(&adm_ctx.tconn->receiver);
2244
2245                 /* Race breaker.  This additional state change request may be
2246                  * necessary, if this was a forced disconnect during a receiver
2247                  * restart.  We may have "killed" the receiver thread just
2248                  * after drbdd_init() returned.  Typically, we should be
2249                  * C_STANDALONE already, now, and this becomes a no-op.
2250                  */
2251                 rv2 = conn_request_state(tconn, NS(conn, C_STANDALONE),
2252                                 CS_VERBOSE | CS_HARD);
2253                 if (rv2 < SS_SUCCESS)
2254                         conn_err(tconn,
2255                                 "unexpected rv2=%d in conn_try_disconnect()\n",
2256                                 rv2);
2257         }
2258         return rv;
2259 }
2260
2261 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2262 {
2263         struct disconnect_parms parms;
2264         struct drbd_tconn *tconn;
2265         enum drbd_state_rv rv;
2266         enum drbd_ret_code retcode;
2267         int err;
2268
2269         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONNECTION);
2270         if (!adm_ctx.reply_skb)
2271                 return retcode;
2272         if (retcode != NO_ERROR)
2273                 goto fail;
2274
2275         tconn = adm_ctx.tconn;
2276         memset(&parms, 0, sizeof(parms));
2277         if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2278                 err = disconnect_parms_from_attrs(&parms, info);
2279                 if (err) {
2280                         retcode = ERR_MANDATORY_TAG;
2281                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2282                         goto fail;
2283                 }
2284         }
2285
2286         rv = conn_try_disconnect(tconn, parms.force_disconnect);
2287         if (rv < SS_SUCCESS)
2288                 retcode = rv;  /* FIXME: Type mismatch. */
2289         else
2290                 retcode = NO_ERROR;
2291  fail:
2292         drbd_adm_finish(info, retcode);
2293         return 0;
2294 }
2295
2296 void resync_after_online_grow(struct drbd_conf *mdev)
2297 {
2298         int iass; /* I am sync source */
2299
2300         dev_info(DEV, "Resync of new storage after online grow\n");
2301         if (mdev->state.role != mdev->state.peer)
2302                 iass = (mdev->state.role == R_PRIMARY);
2303         else
2304                 iass = test_bit(RESOLVE_CONFLICTS, &mdev->tconn->flags);
2305
2306         if (iass)
2307                 drbd_start_resync(mdev, C_SYNC_SOURCE);
2308         else
2309                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2310 }
2311
2312 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2313 {
2314         struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2315         struct resize_parms rs;
2316         struct drbd_conf *mdev;
2317         enum drbd_ret_code retcode;
2318         enum determine_dev_size dd;
2319         enum dds_flags ddsf;
2320         sector_t u_size;
2321         int err;
2322
2323         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2324         if (!adm_ctx.reply_skb)
2325                 return retcode;
2326         if (retcode != NO_ERROR)
2327                 goto fail;
2328
2329         memset(&rs, 0, sizeof(struct resize_parms));
2330         if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2331                 err = resize_parms_from_attrs(&rs, info);
2332                 if (err) {
2333                         retcode = ERR_MANDATORY_TAG;
2334                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2335                         goto fail;
2336                 }
2337         }
2338
2339         mdev = adm_ctx.mdev;
2340         if (mdev->state.conn > C_CONNECTED) {
2341                 retcode = ERR_RESIZE_RESYNC;
2342                 goto fail;
2343         }
2344
2345         if (mdev->state.role == R_SECONDARY &&
2346             mdev->state.peer == R_SECONDARY) {
2347                 retcode = ERR_NO_PRIMARY;
2348                 goto fail;
2349         }
2350
2351         if (!get_ldev(mdev)) {
2352                 retcode = ERR_NO_DISK;
2353                 goto fail;
2354         }
2355
2356         if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
2357                 retcode = ERR_NEED_APV_93;
2358                 goto fail_ldev;
2359         }
2360
2361         rcu_read_lock();
2362         u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
2363         rcu_read_unlock();
2364         if (u_size != (sector_t)rs.resize_size) {
2365                 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2366                 if (!new_disk_conf) {
2367                         retcode = ERR_NOMEM;
2368                         goto fail_ldev;
2369                 }
2370         }
2371
2372         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
2373                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2374
2375         if (new_disk_conf) {
2376                 mutex_lock(&mdev->tconn->conf_update);
2377                 old_disk_conf = mdev->ldev->disk_conf;
2378                 *new_disk_conf = *old_disk_conf;
2379                 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2380                 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
2381                 mutex_unlock(&mdev->tconn->conf_update);
2382                 synchronize_rcu();
2383                 kfree(old_disk_conf);
2384         }
2385
2386         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2387         dd = drbd_determine_dev_size(mdev, ddsf);
2388         drbd_md_sync(mdev);
2389         put_ldev(mdev);
2390         if (dd == dev_size_error) {
2391                 retcode = ERR_NOMEM_BITMAP;
2392                 goto fail;
2393         }
2394
2395         if (mdev->state.conn == C_CONNECTED) {
2396                 if (dd == grew)
2397                         set_bit(RESIZE_PENDING, &mdev->flags);
2398
2399                 drbd_send_uuids(mdev);
2400                 drbd_send_sizes(mdev, 1, ddsf);
2401         }
2402
2403  fail:
2404         drbd_adm_finish(info, retcode);
2405         return 0;
2406
2407  fail_ldev:
2408         put_ldev(mdev);
2409         goto fail;
2410 }
2411
2412 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2413 {
2414         enum drbd_ret_code retcode;
2415         struct drbd_tconn *tconn;
2416         struct res_opts res_opts;
2417         int err;
2418
2419         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
2420         if (!adm_ctx.reply_skb)
2421                 return retcode;
2422         if (retcode != NO_ERROR)
2423                 goto fail;
2424         tconn = adm_ctx.tconn;
2425
2426         res_opts = tconn->res_opts;
2427         if (should_set_defaults(info))
2428                 set_res_opts_defaults(&res_opts);
2429
2430         err = res_opts_from_attrs(&res_opts, info);
2431         if (err && err != -ENOMSG) {
2432                 retcode = ERR_MANDATORY_TAG;
2433                 drbd_msg_put_info(from_attrs_err_to_txt(err));
2434                 goto fail;
2435         }
2436
2437         err = set_resource_options(tconn, &res_opts);
2438         if (err) {
2439                 retcode = ERR_INVALID_REQUEST;
2440                 if (err == -ENOMEM)
2441                         retcode = ERR_NOMEM;
2442         }
2443
2444 fail:
2445         drbd_adm_finish(info, retcode);
2446         return 0;
2447 }
2448
2449 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2450 {
2451         struct drbd_conf *mdev;
2452         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2453
2454         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2455         if (!adm_ctx.reply_skb)
2456                 return retcode;
2457         if (retcode != NO_ERROR)
2458                 goto out;
2459
2460         mdev = adm_ctx.mdev;
2461
2462         /* If there is still bitmap IO pending, probably because of a previous
2463          * resync just being finished, wait for it before requesting a new resync.
2464          * Also wait for it's after_state_ch(). */
2465         drbd_suspend_io(mdev);
2466         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2467         drbd_flush_workqueue(mdev);
2468
2469         /* If we happen to be C_STANDALONE R_SECONDARY, just change to
2470          * D_INCONSISTENT, and set all bits in the bitmap.  Otherwise,
2471          * try to start a resync handshake as sync target for full sync.
2472          */
2473         if (mdev->state.conn == C_STANDALONE && mdev->state.role == R_SECONDARY) {
2474                 retcode = drbd_request_state(mdev, NS(disk, D_INCONSISTENT));
2475                 if (retcode >= SS_SUCCESS) {
2476                         if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
2477                                 "set_n_write from invalidate", BM_LOCKED_MASK))
2478                                 retcode = ERR_IO_MD_DISK;
2479                 }
2480         } else
2481                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2482         drbd_resume_io(mdev);
2483
2484 out:
2485         drbd_adm_finish(info, retcode);
2486         return 0;
2487 }
2488
2489 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2490                 union drbd_state mask, union drbd_state val)
2491 {
2492         enum drbd_ret_code retcode;
2493
2494         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2495         if (!adm_ctx.reply_skb)
2496                 return retcode;
2497         if (retcode != NO_ERROR)
2498                 goto out;
2499
2500         retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2501 out:
2502         drbd_adm_finish(info, retcode);
2503         return 0;
2504 }
2505
2506 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2507 {
2508         int rv;
2509
2510         rv = drbd_bmio_set_n_write(mdev);
2511         drbd_suspend_al(mdev);
2512         return rv;
2513 }
2514
2515 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2516 {
2517         int retcode; /* drbd_ret_code, drbd_state_rv */
2518         struct drbd_conf *mdev;
2519
2520         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2521         if (!adm_ctx.reply_skb)
2522                 return retcode;
2523         if (retcode != NO_ERROR)
2524                 goto out;
2525
2526         mdev = adm_ctx.mdev;
2527
2528         /* If there is still bitmap IO pending, probably because of a previous
2529          * resync just being finished, wait for it before requesting a new resync.
2530          * Also wait for it's after_state_ch(). */
2531         drbd_suspend_io(mdev);
2532         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2533         drbd_flush_workqueue(mdev);
2534
2535         /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
2536          * in the bitmap.  Otherwise, try to start a resync handshake
2537          * as sync source for full sync.
2538          */
2539         if (mdev->state.conn == C_STANDALONE && mdev->state.role == R_PRIMARY) {
2540                 /* The peer will get a resync upon connect anyways. Just make that
2541                    into a full resync. */
2542                 retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
2543                 if (retcode >= SS_SUCCESS) {
2544                         if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
2545                                 "set_n_write from invalidate_peer",
2546                                 BM_LOCKED_SET_ALLOWED))
2547                                 retcode = ERR_IO_MD_DISK;
2548                 }
2549         } else
2550                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
2551         drbd_resume_io(mdev);
2552
2553 out:
2554         drbd_adm_finish(info, retcode);
2555         return 0;
2556 }
2557
2558 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2559 {
2560         enum drbd_ret_code retcode;
2561
2562         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2563         if (!adm_ctx.reply_skb)
2564                 return retcode;
2565         if (retcode != NO_ERROR)
2566                 goto out;
2567
2568         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2569                 retcode = ERR_PAUSE_IS_SET;
2570 out:
2571         drbd_adm_finish(info, retcode);
2572         return 0;
2573 }
2574
2575 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2576 {
2577         union drbd_dev_state s;
2578         enum drbd_ret_code retcode;
2579
2580         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2581         if (!adm_ctx.reply_skb)
2582                 return retcode;
2583         if (retcode != NO_ERROR)
2584                 goto out;
2585
2586         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2587                 s = adm_ctx.mdev->state;
2588                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2589                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2590                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2591                 } else {
2592                         retcode = ERR_PAUSE_IS_CLEAR;
2593                 }
2594         }
2595
2596 out:
2597         drbd_adm_finish(info, retcode);
2598         return 0;
2599 }
2600
2601 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2602 {
2603         return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2604 }
2605
2606 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2607 {
2608         struct drbd_conf *mdev;
2609         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2610
2611         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2612         if (!adm_ctx.reply_skb)
2613                 return retcode;
2614         if (retcode != NO_ERROR)
2615                 goto out;
2616
2617         mdev = adm_ctx.mdev;
2618         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2619                 drbd_uuid_new_current(mdev);
2620                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2621         }
2622         drbd_suspend_io(mdev);
2623         retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2624         if (retcode == SS_SUCCESS) {
2625                 if (mdev->state.conn < C_CONNECTED)
2626                         tl_clear(mdev->tconn);
2627                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2628                         tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2629         }
2630         drbd_resume_io(mdev);
2631
2632 out:
2633         drbd_adm_finish(info, retcode);
2634         return 0;
2635 }
2636
2637 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2638 {
2639         return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2640 }
2641
2642 int nla_put_drbd_cfg_context(struct sk_buff *skb, struct drbd_tconn *tconn, unsigned vnr)
2643 {
2644         struct nlattr *nla;
2645         nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2646         if (!nla)
2647                 goto nla_put_failure;
2648         if (vnr != VOLUME_UNSPECIFIED &&
2649             nla_put_u32(skb, T_ctx_volume, vnr))
2650                 goto nla_put_failure;
2651         if (nla_put_string(skb, T_ctx_resource_name, tconn->name))
2652                 goto nla_put_failure;
2653         if (tconn->my_addr_len &&
2654             nla_put(skb, T_ctx_my_addr, tconn->my_addr_len, &tconn->my_addr))
2655                 goto nla_put_failure;
2656         if (tconn->peer_addr_len &&
2657             nla_put(skb, T_ctx_peer_addr, tconn->peer_addr_len, &tconn->peer_addr))
2658                 goto nla_put_failure;
2659         nla_nest_end(skb, nla);
2660         return 0;
2661
2662 nla_put_failure:
2663         if (nla)
2664                 nla_nest_cancel(skb, nla);
2665         return -EMSGSIZE;
2666 }
2667
2668 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2669                 const struct sib_info *sib)
2670 {
2671         struct state_info *si = NULL; /* for sizeof(si->member); */
2672         struct nlattr *nla;
2673         int got_ldev;
2674         int err = 0;
2675         int exclude_sensitive;
2676
2677         /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2678          * to.  So we better exclude_sensitive information.
2679          *
2680          * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2681          * in the context of the requesting user process. Exclude sensitive
2682          * information, unless current has superuser.
2683          *
2684          * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2685          * relies on the current implementation of netlink_dump(), which
2686          * executes the dump callback successively from netlink_recvmsg(),
2687          * always in the context of the receiving process */
2688         exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2689
2690         got_ldev = get_ldev(mdev);
2691
2692         /* We need to add connection name and volume number information still.
2693          * Minor number is in drbd_genlmsghdr. */
2694         if (nla_put_drbd_cfg_context(skb, mdev->tconn, mdev->vnr))
2695                 goto nla_put_failure;
2696
2697         if (res_opts_to_skb(skb, &mdev->tconn->res_opts, exclude_sensitive))
2698                 goto nla_put_failure;
2699
2700         rcu_read_lock();
2701         if (got_ldev) {
2702                 struct disk_conf *disk_conf;
2703
2704                 disk_conf = rcu_dereference(mdev->ldev->disk_conf);
2705                 err = disk_conf_to_skb(skb, disk_conf, exclude_sensitive);
2706         }
2707         if (!err) {
2708                 struct net_conf *nc;
2709
2710                 nc = rcu_dereference(mdev->tconn->net_conf);
2711                 if (nc)
2712                         err = net_conf_to_skb(skb, nc, exclude_sensitive);
2713         }
2714         rcu_read_unlock();
2715         if (err)
2716                 goto nla_put_failure;
2717
2718         nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2719         if (!nla)
2720                 goto nla_put_failure;
2721         if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) ||
2722             nla_put_u32(skb, T_current_state, mdev->state.i) ||
2723             nla_put_u64(skb, T_ed_uuid, mdev->ed_uuid) ||
2724             nla_put_u64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev)) ||
2725             nla_put_u64(skb, T_send_cnt, mdev->send_cnt) ||
2726             nla_put_u64(skb, T_recv_cnt, mdev->recv_cnt) ||
2727             nla_put_u64(skb, T_read_cnt, mdev->read_cnt) ||
2728             nla_put_u64(skb, T_writ_cnt, mdev->writ_cnt) ||
2729             nla_put_u64(skb, T_al_writ_cnt, mdev->al_writ_cnt) ||
2730             nla_put_u64(skb, T_bm_writ_cnt, mdev->bm_writ_cnt) ||
2731             nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&mdev->ap_bio_cnt)) ||
2732             nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&mdev->ap_pending_cnt)) ||
2733             nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&mdev->rs_pending_cnt)))
2734                 goto nla_put_failure;
2735
2736         if (got_ldev) {
2737                 int err;
2738
2739                 spin_lock_irq(&mdev->ldev->md.uuid_lock);
2740                 err = nla_put(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2741                 spin_unlock_irq(&mdev->ldev->md.uuid_lock);
2742
2743                 if (err)
2744                         goto nla_put_failure;
2745
2746                 if (nla_put_u32(skb, T_disk_flags, mdev->ldev->md.flags) ||
2747                     nla_put_u64(skb, T_bits_total, drbd_bm_bits(mdev)) ||
2748                     nla_put_u64(skb, T_bits_oos, drbd_bm_total_weight(mdev)))
2749                         goto nla_put_failure;
2750                 if (C_SYNC_SOURCE <= mdev->state.conn &&
2751                     C_PAUSED_SYNC_T >= mdev->state.conn) {
2752                         if (nla_put_u64(skb, T_bits_rs_total, mdev->rs_total) ||
2753                             nla_put_u64(skb, T_bits_rs_failed, mdev->rs_failed))
2754                                 goto nla_put_failure;
2755                 }
2756         }
2757
2758         if (sib) {
2759                 switch(sib->sib_reason) {
2760                 case SIB_SYNC_PROGRESS:
2761                 case SIB_GET_STATUS_REPLY:
2762                         break;
2763                 case SIB_STATE_CHANGE:
2764                         if (nla_put_u32(skb, T_prev_state, sib->os.i) ||
2765                             nla_put_u32(skb, T_new_state, sib->ns.i))
2766                                 goto nla_put_failure;
2767                         break;
2768                 case SIB_HELPER_POST:
2769                         if (nla_put_u32(skb, T_helper_exit_code,
2770                                         sib->helper_exit_code))
2771                                 goto nla_put_failure;
2772                         /* fall through */
2773                 case SIB_HELPER_PRE:
2774                         if (nla_put_string(skb, T_helper, sib->helper_name))
2775                                 goto nla_put_failure;
2776                         break;
2777                 }
2778         }
2779         nla_nest_end(skb, nla);
2780
2781         if (0)
2782 nla_put_failure:
2783                 err = -EMSGSIZE;
2784         if (got_ldev)
2785                 put_ldev(mdev);
2786         return err;
2787 }
2788
2789 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2790 {
2791         enum drbd_ret_code retcode;
2792         int err;
2793
2794         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2795         if (!adm_ctx.reply_skb)
2796                 return retcode;
2797         if (retcode != NO_ERROR)
2798                 goto out;
2799
2800         err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2801         if (err) {
2802                 nlmsg_free(adm_ctx.reply_skb);
2803                 return err;
2804         }
2805 out:
2806         drbd_adm_finish(info, retcode);
2807         return 0;
2808 }
2809
2810 int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
2811 {
2812         struct drbd_conf *mdev;
2813         struct drbd_genlmsghdr *dh;
2814         struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2815         struct drbd_tconn *tconn = NULL;
2816         struct drbd_tconn *tmp;
2817         unsigned volume = cb->args[1];
2818
2819         /* Open coded, deferred, iteration:
2820          * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2821          *      idr_for_each_entry(&tconn->volumes, mdev, i) {
2822          *        ...
2823          *      }
2824          * }
2825          * where tconn is cb->args[0];
2826          * and i is cb->args[1];
2827          *
2828          * cb->args[2] indicates if we shall loop over all resources,
2829          * or just dump all volumes of a single resource.
2830          *
2831          * This may miss entries inserted after this dump started,
2832          * or entries deleted before they are reached.
2833          *
2834          * We need to make sure the mdev won't disappear while
2835          * we are looking at it, and revalidate our iterators
2836          * on each iteration.
2837          */
2838
2839         /* synchronize with conn_create()/conn_destroy() */
2840         rcu_read_lock();
2841         /* revalidate iterator position */
2842         list_for_each_entry_rcu(tmp, &drbd_tconns, all_tconn) {
2843                 if (pos == NULL) {
2844                         /* first iteration */
2845                         pos = tmp;
2846                         tconn = pos;
2847                         break;
2848                 }
2849                 if (tmp == pos) {
2850                         tconn = pos;
2851                         break;
2852                 }
2853         }
2854         if (tconn) {
2855 next_tconn:
2856                 mdev = idr_get_next(&tconn->volumes, &volume);
2857                 if (!mdev) {
2858                         /* No more volumes to dump on this tconn.
2859                          * Advance tconn iterator. */
2860                         pos = list_entry_rcu(tconn->all_tconn.next,
2861                                              struct drbd_tconn, all_tconn);
2862                         /* Did we dump any volume on this tconn yet? */
2863                         if (volume != 0) {
2864                                 /* If we reached the end of the list,
2865                                  * or only a single resource dump was requested,
2866                                  * we are done. */
2867                                 if (&pos->all_tconn == &drbd_tconns || cb->args[2])
2868                                         goto out;
2869                                 volume = 0;
2870                                 tconn = pos;
2871                                 goto next_tconn;
2872                         }
2873                 }
2874
2875                 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2876                                 cb->nlh->nlmsg_seq, &drbd_genl_family,
2877                                 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2878                 if (!dh)
2879                         goto out;
2880
2881                 if (!mdev) {
2882                         /* This is a tconn without a single volume.
2883                          * Suprisingly enough, it may have a network
2884                          * configuration. */
2885                         struct net_conf *nc;
2886                         dh->minor = -1U;
2887                         dh->ret_code = NO_ERROR;
2888                         if (nla_put_drbd_cfg_context(skb, tconn, VOLUME_UNSPECIFIED))
2889                                 goto cancel;
2890                         nc = rcu_dereference(tconn->net_conf);
2891                         if (nc && net_conf_to_skb(skb, nc, 1) != 0)
2892                                 goto cancel;
2893                         goto done;
2894                 }
2895
2896                 D_ASSERT(mdev->vnr == volume);
2897                 D_ASSERT(mdev->tconn == tconn);
2898
2899                 dh->minor = mdev_to_minor(mdev);
2900                 dh->ret_code = NO_ERROR;
2901
2902                 if (nla_put_status_info(skb, mdev, NULL)) {
2903 cancel:
2904                         genlmsg_cancel(skb, dh);
2905                         goto out;
2906                 }
2907 done:
2908                 genlmsg_end(skb, dh);
2909         }
2910
2911 out:
2912         rcu_read_unlock();
2913         /* where to start the next iteration */
2914         cb->args[0] = (long)pos;
2915         cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2916
2917         /* No more tconns/volumes/minors found results in an empty skb.
2918          * Which will terminate the dump. */
2919         return skb->len;
2920 }
2921
2922 /*
2923  * Request status of all resources, or of all volumes within a single resource.
2924  *
2925  * This is a dump, as the answer may not fit in a single reply skb otherwise.
2926  * Which means we cannot use the family->attrbuf or other such members, because
2927  * dump is NOT protected by the genl_lock().  During dump, we only have access
2928  * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
2929  *
2930  * Once things are setup properly, we call into get_one_status().
2931  */
2932 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2933 {
2934         const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
2935         struct nlattr *nla;
2936         const char *resource_name;
2937         struct drbd_tconn *tconn;
2938         int maxtype;
2939
2940         /* Is this a followup call? */
2941         if (cb->args[0]) {
2942                 /* ... of a single resource dump,
2943                  * and the resource iterator has been advanced already? */
2944                 if (cb->args[2] && cb->args[2] != cb->args[0])
2945                         return 0; /* DONE. */
2946                 goto dump;
2947         }
2948
2949         /* First call (from netlink_dump_start).  We need to figure out
2950          * which resource(s) the user wants us to dump. */
2951         nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
2952                         nlmsg_attrlen(cb->nlh, hdrlen),
2953                         DRBD_NLA_CFG_CONTEXT);
2954
2955         /* No explicit context given.  Dump all. */
2956         if (!nla)
2957                 goto dump;
2958         maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
2959         nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
2960         if (IS_ERR(nla))
2961                 return PTR_ERR(nla);
2962         /* context given, but no name present? */
2963         if (!nla)
2964                 return -EINVAL;
2965         resource_name = nla_data(nla);
2966         tconn = conn_get_by_name(resource_name);
2967
2968         if (!tconn)
2969                 return -ENODEV;
2970
2971         kref_put(&tconn->kref, &conn_destroy); /* get_one_status() (re)validates tconn by itself */
2972
2973         /* prime iterators, and set "filter" mode mark:
2974          * only dump this tconn. */
2975         cb->args[0] = (long)tconn;
2976         /* cb->args[1] = 0; passed in this way. */
2977         cb->args[2] = (long)tconn;
2978
2979 dump:
2980         return get_one_status(skb, cb);
2981 }
2982
2983 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2984 {
2985         enum drbd_ret_code retcode;
2986         struct timeout_parms tp;
2987         int err;
2988
2989         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2990         if (!adm_ctx.reply_skb)
2991                 return retcode;
2992         if (retcode != NO_ERROR)
2993                 goto out;
2994
2995         tp.timeout_type =
2996                 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2997                 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2998                 UT_DEFAULT;
2999
3000         err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
3001         if (err) {
3002                 nlmsg_free(adm_ctx.reply_skb);
3003                 return err;
3004         }
3005 out:
3006         drbd_adm_finish(info, retcode);
3007         return 0;
3008 }
3009
3010 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
3011 {
3012         struct drbd_conf *mdev;
3013         enum drbd_ret_code retcode;
3014         struct start_ov_parms parms;
3015
3016         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3017         if (!adm_ctx.reply_skb)
3018                 return retcode;
3019         if (retcode != NO_ERROR)
3020                 goto out;
3021
3022         mdev = adm_ctx.mdev;
3023
3024         /* resume from last known position, if possible */
3025         parms.ov_start_sector = mdev->ov_start_sector;
3026         parms.ov_stop_sector = ULLONG_MAX;
3027         if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
3028                 int err = start_ov_parms_from_attrs(&parms, info);
3029                 if (err) {
3030                         retcode = ERR_MANDATORY_TAG;
3031                         drbd_msg_put_info(from_attrs_err_to_txt(err));
3032                         goto out;
3033                 }
3034         }
3035         /* w_make_ov_request expects position to be aligned */
3036         mdev->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1);
3037         mdev->ov_stop_sector = parms.ov_stop_sector;
3038
3039         /* If there is still bitmap IO pending, e.g. previous resync or verify
3040          * just being finished, wait for it before requesting a new resync. */
3041         drbd_suspend_io(mdev);
3042         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
3043         retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
3044         drbd_resume_io(mdev);
3045 out:
3046         drbd_adm_finish(info, retcode);
3047         return 0;
3048 }
3049
3050
3051 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
3052 {
3053         struct drbd_conf *mdev;
3054         enum drbd_ret_code retcode;
3055         int skip_initial_sync = 0;
3056         int err;
3057         struct new_c_uuid_parms args;
3058
3059         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3060         if (!adm_ctx.reply_skb)
3061                 return retcode;
3062         if (retcode != NO_ERROR)
3063                 goto out_nolock;
3064
3065         mdev = adm_ctx.mdev;
3066         memset(&args, 0, sizeof(args));
3067         if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
3068                 err = new_c_uuid_parms_from_attrs(&args, info);
3069                 if (err) {
3070                         retcode = ERR_MANDATORY_TAG;
3071                         drbd_msg_put_info(from_attrs_err_to_txt(err));
3072                         goto out_nolock;
3073                 }
3074         }
3075
3076         mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
3077
3078         if (!get_ldev(mdev)) {
3079                 retcode = ERR_NO_DISK;
3080                 goto out;
3081         }
3082
3083         /* this is "skip initial sync", assume to be clean */
3084         if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
3085             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
3086                 dev_info(DEV, "Preparing to skip initial sync\n");
3087                 skip_initial_sync = 1;
3088         } else if (mdev->state.conn != C_STANDALONE) {
3089                 retcode = ERR_CONNECTED;
3090                 goto out_dec;
3091         }
3092
3093         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
3094         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
3095
3096         if (args.clear_bm) {
3097                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
3098                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
3099                 if (err) {
3100                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
3101                         retcode = ERR_IO_MD_DISK;
3102                 }
3103                 if (skip_initial_sync) {
3104                         drbd_send_uuids_skip_initial_sync(mdev);
3105                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
3106                         drbd_print_uuids(mdev, "cleared bitmap UUID");
3107                         spin_lock_irq(&mdev->tconn->req_lock);
3108                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
3109                                         CS_VERBOSE, NULL);
3110                         spin_unlock_irq(&mdev->tconn->req_lock);
3111                 }
3112         }
3113
3114         drbd_md_sync(mdev);
3115 out_dec:
3116         put_ldev(mdev);
3117 out:
3118         mutex_unlock(mdev->state_mutex);
3119 out_nolock:
3120         drbd_adm_finish(info, retcode);
3121         return 0;
3122 }
3123
3124 static enum drbd_ret_code
3125 drbd_check_resource_name(const char *name)
3126 {
3127         if (!name || !name[0]) {
3128                 drbd_msg_put_info("resource name missing");
3129                 return ERR_MANDATORY_TAG;
3130         }
3131         /* if we want to use these in sysfs/configfs/debugfs some day,
3132          * we must not allow slashes */
3133         if (strchr(name, '/')) {
3134                 drbd_msg_put_info("invalid resource name");
3135                 return ERR_INVALID_REQUEST;
3136         }
3137         return NO_ERROR;
3138 }
3139
3140 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
3141 {
3142         enum drbd_ret_code retcode;
3143         struct res_opts res_opts;
3144         int err;
3145
3146         retcode = drbd_adm_prepare(skb, info, 0);
3147         if (!adm_ctx.reply_skb)
3148                 return retcode;
3149         if (retcode != NO_ERROR)
3150                 goto out;
3151
3152         set_res_opts_defaults(&res_opts);
3153         err = res_opts_from_attrs(&res_opts, info);
3154         if (err && err != -ENOMSG) {
3155                 retcode = ERR_MANDATORY_TAG;
3156                 drbd_msg_put_info(from_attrs_err_to_txt(err));
3157                 goto out;
3158         }
3159
3160         retcode = drbd_check_resource_name(adm_ctx.resource_name);
3161         if (retcode != NO_ERROR)
3162                 goto out;
3163
3164         if (adm_ctx.tconn) {
3165                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
3166                         retcode = ERR_INVALID_REQUEST;
3167                         drbd_msg_put_info("resource exists");
3168                 }
3169                 /* else: still NO_ERROR */
3170                 goto out;
3171         }
3172
3173         if (!conn_create(adm_ctx.resource_name, &res_opts))
3174                 retcode = ERR_NOMEM;
3175 out:
3176         drbd_adm_finish(info, retcode);
3177         return 0;
3178 }
3179
3180 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
3181 {
3182         struct drbd_genlmsghdr *dh = info->userhdr;
3183         enum drbd_ret_code retcode;
3184
3185         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
3186         if (!adm_ctx.reply_skb)
3187                 return retcode;
3188         if (retcode != NO_ERROR)
3189                 goto out;
3190
3191         if (dh->minor > MINORMASK) {
3192                 drbd_msg_put_info("requested minor out of range");
3193                 retcode = ERR_INVALID_REQUEST;
3194                 goto out;
3195         }
3196         if (adm_ctx.volume > DRBD_VOLUME_MAX) {
3197                 drbd_msg_put_info("requested volume id out of range");
3198                 retcode = ERR_INVALID_REQUEST;
3199                 goto out;
3200         }
3201
3202         /* drbd_adm_prepare made sure already
3203          * that mdev->tconn and mdev->vnr match the request. */
3204         if (adm_ctx.mdev) {
3205                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
3206                         retcode = ERR_MINOR_EXISTS;
3207                 /* else: still NO_ERROR */
3208                 goto out;
3209         }
3210
3211         retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
3212 out:
3213         drbd_adm_finish(info, retcode);
3214         return 0;
3215 }
3216
3217 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
3218 {
3219         if (mdev->state.disk == D_DISKLESS &&
3220             /* no need to be mdev->state.conn == C_STANDALONE &&
3221              * we may want to delete a minor from a live replication group.
3222              */
3223             mdev->state.role == R_SECONDARY) {
3224                 _drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS),
3225                                     CS_VERBOSE + CS_WAIT_COMPLETE);
3226                 idr_remove(&mdev->tconn->volumes, mdev->vnr);
3227                 idr_remove(&minors, mdev_to_minor(mdev));
3228                 destroy_workqueue(mdev->submit.wq);
3229                 del_gendisk(mdev->vdisk);
3230                 synchronize_rcu();
3231                 kref_put(&mdev->kref, &drbd_minor_destroy);
3232                 return NO_ERROR;
3233         } else
3234                 return ERR_MINOR_CONFIGURED;
3235 }
3236
3237 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
3238 {
3239         enum drbd_ret_code retcode;
3240
3241         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3242         if (!adm_ctx.reply_skb)
3243                 return retcode;
3244         if (retcode != NO_ERROR)
3245                 goto out;
3246
3247         retcode = adm_delete_minor(adm_ctx.mdev);
3248 out:
3249         drbd_adm_finish(info, retcode);
3250         return 0;
3251 }
3252
3253 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
3254 {
3255         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3256         struct drbd_conf *mdev;
3257         unsigned i;
3258
3259         retcode = drbd_adm_prepare(skb, info, 0);
3260         if (!adm_ctx.reply_skb)
3261                 return retcode;
3262         if (retcode != NO_ERROR)
3263                 goto out;
3264
3265         if (!adm_ctx.tconn) {
3266                 retcode = ERR_RES_NOT_KNOWN;
3267                 goto out;
3268         }
3269
3270         /* demote */
3271         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3272                 retcode = drbd_set_role(mdev, R_SECONDARY, 0);
3273                 if (retcode < SS_SUCCESS) {
3274                         drbd_msg_put_info("failed to demote");
3275                         goto out;
3276                 }
3277         }
3278
3279         retcode = conn_try_disconnect(adm_ctx.tconn, 0);
3280         if (retcode < SS_SUCCESS) {
3281                 drbd_msg_put_info("failed to disconnect");
3282                 goto out;
3283         }
3284
3285         /* detach */
3286         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3287                 retcode = adm_detach(mdev, 0);
3288                 if (retcode < SS_SUCCESS || retcode > NO_ERROR) {
3289                         drbd_msg_put_info("failed to detach");
3290                         goto out;
3291                 }
3292         }
3293
3294         /* If we reach this, all volumes (of this tconn) are Secondary,
3295          * Disconnected, Diskless, aka Unconfigured. Make sure all threads have
3296          * actually stopped, state handling only does drbd_thread_stop_nowait(). */
3297         drbd_thread_stop(&adm_ctx.tconn->worker);
3298
3299         /* Now, nothing can fail anymore */
3300
3301         /* delete volumes */
3302         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3303                 retcode = adm_delete_minor(mdev);
3304                 if (retcode != NO_ERROR) {
3305                         /* "can not happen" */
3306                         drbd_msg_put_info("failed to delete volume");
3307                         goto out;
3308                 }
3309         }
3310
3311         /* delete connection */
3312         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3313                 list_del_rcu(&adm_ctx.tconn->all_tconn);
3314                 synchronize_rcu();
3315                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3316
3317                 retcode = NO_ERROR;
3318         } else {
3319                 /* "can not happen" */
3320                 retcode = ERR_RES_IN_USE;
3321                 drbd_msg_put_info("failed to delete connection");
3322         }
3323         goto out;
3324 out:
3325         drbd_adm_finish(info, retcode);
3326         return 0;
3327 }
3328
3329 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
3330 {
3331         enum drbd_ret_code retcode;
3332
3333         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
3334         if (!adm_ctx.reply_skb)
3335                 return retcode;
3336         if (retcode != NO_ERROR)
3337                 goto out;
3338
3339         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3340                 list_del_rcu(&adm_ctx.tconn->all_tconn);
3341                 synchronize_rcu();
3342                 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3343
3344                 retcode = NO_ERROR;
3345         } else {
3346                 retcode = ERR_RES_IN_USE;
3347         }
3348
3349         if (retcode == NO_ERROR)
3350                 drbd_thread_stop(&adm_ctx.tconn->worker);
3351 out:
3352         drbd_adm_finish(info, retcode);
3353         return 0;
3354 }
3355
3356 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
3357 {
3358         static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3359         struct sk_buff *msg;
3360         struct drbd_genlmsghdr *d_out;
3361         unsigned seq;
3362         int err = -ENOMEM;
3363
3364         if (sib->sib_reason == SIB_SYNC_PROGRESS) {
3365                 if (time_after(jiffies, mdev->rs_last_bcast + HZ))
3366                         mdev->rs_last_bcast = jiffies;
3367                 else
3368                         return;
3369         }
3370
3371         seq = atomic_inc_return(&drbd_genl_seq);
3372         msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3373         if (!msg)
3374                 goto failed;
3375
3376         err = -EMSGSIZE;
3377         d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3378         if (!d_out) /* cannot happen, but anyways. */
3379                 goto nla_put_failure;
3380         d_out->minor = mdev_to_minor(mdev);
3381         d_out->ret_code = NO_ERROR;
3382
3383         if (nla_put_status_info(msg, mdev, sib))
3384                 goto nla_put_failure;
3385         genlmsg_end(msg, d_out);
3386         err = drbd_genl_multicast_events(msg, 0);
3387         /* msg has been consumed or freed in netlink_broadcast() */
3388         if (err && err != -ESRCH)
3389                 goto failed;
3390
3391         return;
3392
3393 nla_put_failure:
3394         nlmsg_free(msg);
3395 failed:
3396         dev_err(DEV, "Error %d while broadcasting event. "
3397                         "Event seq:%u sib_reason:%u\n",
3398                         err, seq, sib->sib_reason);
3399 }