drivers/block/drbd/drbd_nl.c

   1 /*
   2    drbd_nl.c
   3
   4    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
   5
   6    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
   7    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
   8    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
   9
  10    drbd is free software; you can redistribute it and/or modify
  11    it under the terms of the GNU General Public License as published by
  12    the Free Software Foundation; either version 2, or (at your option)
  13    any later version.
  14
  15    drbd is distributed in the hope that it will be useful,
  16    but WITHOUT ANY WARRANTY; without even the implied warranty of
  17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18    GNU General Public License for more details.
  19
  20    You should have received a copy of the GNU General Public License
  21    along with drbd; see the file COPYING.  If not, write to
  22    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  23
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/drbd.h>
  28 #include <linux/in.h>
  29 #include <linux/fs.h>
  30 #include <linux/file.h>
  31 #include <linux/slab.h>
  32 #include <linux/blkpg.h>
  33 #include <linux/cpumask.h>
  34 #include "drbd_int.h"
  35 #include "drbd_req.h"
  36 #include "drbd_wrappers.h"
  37 #include <asm/unaligned.h>
  38 #include <linux/drbd_limits.h>
  39 #include <linux/kthread.h>
  40
  41 #include <net/genetlink.h>
  42
  43 /* .doit */
  44 // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info);
  45 // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info);
  46
  47 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info);
  48 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info);
  49
  50 int drbd_adm_create_connection(struct sk_buff *skb, struct genl_info *info);
  51 int drbd_adm_delete_connection(struct sk_buff *skb, struct genl_info *info);
  52 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info);
  53
  54 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info);
  55 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info);
  56 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info);
  57 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info);
  58 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info);
  59 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info);
  60 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info);
  61 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info);
  62 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info);
  63 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info);
  64 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info);
  65 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info);
  66 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info);
  67 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info);
  68 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info);
  69 int drbd_adm_syncer(struct sk_buff *skb, struct genl_info *info);
  70 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info);
  71 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info);
  72 /* .dumpit */
  73 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb);
  74
  75 #include <linux/drbd_genl_api.h>
  76 #include <linux/genl_magic_func.h>
  77
  78 /* used blkdev_get_by_path, to claim our meta data device(s) */
  79 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
  80
  81 /* Configuration is strictly serialized, because generic netlink message
  82  * processing is strictly serialized by the genl_lock().
  83  * Which means we can use one static global drbd_config_context struct.
  84  */
  85 static struct drbd_config_context {
  86         /* assigned from drbd_genlmsghdr */
  87         unsigned int minor;
  88         /* assigned from request attributes, if present */
  89         unsigned int volume;
  90 #define VOLUME_UNSPECIFIED              (-1U)
  91         /* pointer into the request skb,
  92          * limited lifetime! */
  93         char *conn_name;
  94
  95         /* reply buffer */
  96         struct sk_buff *reply_skb;
  97         /* pointer into reply buffer */
  98         struct drbd_genlmsghdr *reply_dh;
  99         /* resolved from attributes, if possible */
 100         struct drbd_conf *mdev;
 101         struct drbd_tconn *tconn;
 102 } adm_ctx;
 103
 104 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
 105 {
 106         genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
 107         if (genlmsg_reply(skb, info))
 108                 printk(KERN_ERR "drbd: error sending genl reply\n");
 109 }
 110
 111 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
 112  * reason it could fail was no space in skb, and there are 4k available. */
 113 int drbd_msg_put_info(const char *info)
 114 {
 115         struct sk_buff *skb = adm_ctx.reply_skb;
 116         struct nlattr *nla;
 117         int err = -EMSGSIZE;
 118
 119         if (!info || !info[0])
 120                 return 0;
 121
 122         nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY);
 123         if (!nla)
 124                 return err;
 125
 126         err = nla_put_string(skb, T_info_text, info);
 127         if (err) {
 128                 nla_nest_cancel(skb, nla);
 129                 return err;
 130         } else
 131                 nla_nest_end(skb, nla);
 132         return 0;
 133 }
 134
 135 /* This would be a good candidate for a "pre_doit" hook,
 136  * and per-family private info->pointers.
 137  * But we need to stay compatible with older kernels.
 138  * If it returns successfully, adm_ctx members are valid.
 139  */
 140 #define DRBD_ADM_NEED_MINOR     1
 141 #define DRBD_ADM_NEED_CONN      2
 142 static int drbd_adm_prepare(struct sk_buff *skb, struct genl_info *info,
 143                 unsigned flags)
 144 {
 145         struct drbd_genlmsghdr *d_in = info->userhdr;
 146         const u8 cmd = info->genlhdr->cmd;
 147         int err;
 148
 149         memset(&adm_ctx, 0, sizeof(adm_ctx));
 150
 151         /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
 152         if (cmd != DRBD_ADM_GET_STATUS
 153         && security_netlink_recv(skb, CAP_SYS_ADMIN))
 154                return -EPERM;
 155
 156         adm_ctx.reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
 157         if (!adm_ctx.reply_skb)
 158                 goto fail;
 159
 160         adm_ctx.reply_dh = genlmsg_put_reply(adm_ctx.reply_skb,
 161                                         info, &drbd_genl_family, 0, cmd);
 162         /* put of a few bytes into a fresh skb of >= 4k will always succeed.
 163          * but anyways */
 164         if (!adm_ctx.reply_dh)
 165                 goto fail;
 166
 167         adm_ctx.reply_dh->minor = d_in->minor;
 168         adm_ctx.reply_dh->ret_code = NO_ERROR;
 169
 170         if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
 171                 struct nlattr *nla;
 172                 /* parse and validate only */
 173                 err = drbd_cfg_context_from_attrs(NULL, info->attrs);
 174                 if (err)
 175                         goto fail;
 176
 177                 /* It was present, and valid,
 178                  * copy it over to the reply skb. */
 179                 err = nla_put_nohdr(adm_ctx.reply_skb,
 180                                 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
 181                                 info->attrs[DRBD_NLA_CFG_CONTEXT]);
 182                 if (err)
 183                         goto fail;
 184
 185                 /* and assign stuff to the global adm_ctx */
 186                 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
 187                 adm_ctx.volume = nla ? nla_get_u32(nla) : VOLUME_UNSPECIFIED;
 188                 nla = nested_attr_tb[__nla_type(T_ctx_conn_name)];
 189                 if (nla)
 190                         adm_ctx.conn_name = nla_data(nla);
 191         } else
 192                 adm_ctx.volume = VOLUME_UNSPECIFIED;
 193
 194         adm_ctx.minor = d_in->minor;
 195         adm_ctx.mdev = minor_to_mdev(d_in->minor);
 196         adm_ctx.tconn = conn_by_name(adm_ctx.conn_name);
 197
 198         if (!adm_ctx.mdev && (flags & DRBD_ADM_NEED_MINOR)) {
 199                 drbd_msg_put_info("unknown minor");
 200                 return ERR_MINOR_INVALID;
 201         }
 202         if (!adm_ctx.tconn && (flags & DRBD_ADM_NEED_CONN)) {
 203                 drbd_msg_put_info("unknown connection");
 204                 return ERR_INVALID_REQUEST;
 205         }
 206
 207         /* some more paranoia, if the request was over-determined */
 208         if (adm_ctx.mdev && adm_ctx.tconn &&
 209             adm_ctx.mdev->tconn != adm_ctx.tconn) {
 210                 pr_warning("request: minor=%u, conn=%s; but that minor belongs to connection %s\n",
 211                                 adm_ctx.minor, adm_ctx.conn_name, adm_ctx.mdev->tconn->name);
 212                 drbd_msg_put_info("minor exists in different connection");
 213                 return ERR_INVALID_REQUEST;
 214         }
 215         if (adm_ctx.mdev &&
 216             adm_ctx.volume != VOLUME_UNSPECIFIED &&
 217             adm_ctx.volume != adm_ctx.mdev->vnr) {
 218                 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
 219                                 adm_ctx.minor, adm_ctx.volume,
 220                                 adm_ctx.mdev->vnr, adm_ctx.mdev->tconn->name);
 221                 drbd_msg_put_info("minor exists as different volume");
 222                 return ERR_INVALID_REQUEST;
 223         }
 224         if (adm_ctx.mdev && !adm_ctx.tconn)
 225                 adm_ctx.tconn = adm_ctx.mdev->tconn;
 226         return NO_ERROR;
 227
 228 fail:
 229         nlmsg_free(adm_ctx.reply_skb);
 230         adm_ctx.reply_skb = NULL;
 231         return -ENOMEM;
 232 }
 233
 234 static int drbd_adm_finish(struct genl_info *info, int retcode)
 235 {
 236         struct nlattr *nla;
 237         const char *conn_name = NULL;
 238
 239         if (!adm_ctx.reply_skb)
 240                 return -ENOMEM;
 241
 242         adm_ctx.reply_dh->ret_code = retcode;
 243
 244         nla = info->attrs[DRBD_NLA_CFG_CONTEXT];
 245         if (nla) {
 246                 nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
 247                 if (nla)
 248                         conn_name = nla_data(nla);
 249         }
 250
 251         drbd_adm_send_reply(adm_ctx.reply_skb, info);
 252         return 0;
 253 }
 254
 255 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
 256 {
 257         char *envp[] = { "HOME=/",
 258                         "TERM=linux",
 259                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 260                         NULL, /* Will be set to address family */
 261                         NULL, /* Will be set to address */
 262                         NULL };
 263         char mb[12], af[20], ad[60], *afs;
 264         char *argv[] = {usermode_helper, cmd, mb, NULL };
 265         struct sib_info sib;
 266         int ret;
 267
 268         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 269
 270         if (get_net_conf(mdev->tconn)) {
 271                 switch (((struct sockaddr *)mdev->tconn->net_conf->peer_addr)->sa_family) {
 272                 case AF_INET6:
 273                         afs = "ipv6";
 274                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6",
 275                                  &((struct sockaddr_in6 *)mdev->tconn->net_conf->peer_addr)->sin6_addr);
 276                         break;
 277                 case AF_INET:
 278                         afs = "ipv4";
 279                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 280                                  &((struct sockaddr_in *)mdev->tconn->net_conf->peer_addr)->sin_addr);
 281                         break;
 282                 default:
 283                         afs = "ssocks";
 284                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 285                                  &((struct sockaddr_in *)mdev->tconn->net_conf->peer_addr)->sin_addr);
 286                 }
 287                 snprintf(af, 20, "DRBD_PEER_AF=%s", afs);
 288                 envp[3]=af;
 289                 envp[4]=ad;
 290                 put_net_conf(mdev->tconn);
 291         }
 292
 293         /* The helper may take some time.
 294          * write out any unsynced meta data changes now */
 295         drbd_md_sync(mdev);
 296
 297         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 298         sib.sib_reason = SIB_HELPER_PRE;
 299         sib.helper_name = cmd;
 300         drbd_bcast_event(mdev, &sib);
 301         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
 302         if (ret)
 303                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 304                                 usermode_helper, cmd, mb,
 305                                 (ret >> 8) & 0xff, ret);
 306         else
 307                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 308                                 usermode_helper, cmd, mb,
 309                                 (ret >> 8) & 0xff, ret);
 310         sib.sib_reason = SIB_HELPER_POST;
 311         sib.helper_exit_code = ret;
 312         drbd_bcast_event(mdev, &sib);
 313
 314         if (ret < 0) /* Ignore any ERRNOs we got. */
 315                 ret = 0;
 316
 317         return ret;
 318 }
 319
 320 enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
 321 {
 322         char *ex_to_string;
 323         int r;
 324         enum drbd_disk_state nps;
 325         enum drbd_fencing_p fp;
 326
 327         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 328
 329         if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 330                 fp = mdev->ldev->dc.fencing;
 331                 put_ldev(mdev);
 332         } else {
 333                 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n");
 334                 nps = mdev->state.pdsk;
 335                 goto out;
 336         }
 337
 338         r = drbd_khelper(mdev, "fence-peer");
 339
 340         switch ((r>>8) & 0xff) {
 341         case 3: /* peer is inconsistent */
 342                 ex_to_string = "peer is inconsistent or worse";
 343                 nps = D_INCONSISTENT;
 344                 break;
 345         case 4: /* peer got outdated, or was already outdated */
 346                 ex_to_string = "peer was fenced";
 347                 nps = D_OUTDATED;
 348                 break;
 349         case 5: /* peer was down */
 350                 if (mdev->state.disk == D_UP_TO_DATE) {
 351                         /* we will(have) create(d) a new UUID anyways... */
 352                         ex_to_string = "peer is unreachable, assumed to be dead";
 353                         nps = D_OUTDATED;
 354                 } else {
 355                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 356                         nps = mdev->state.pdsk;
 357                 }
 358                 break;
 359         case 6: /* Peer is primary, voluntarily outdate myself.
 360                  * This is useful when an unconnected R_SECONDARY is asked to
 361                  * become R_PRIMARY, but finds the other peer being active. */
 362                 ex_to_string = "peer is active";
 363                 dev_warn(DEV, "Peer is primary, outdating myself.\n");
 364                 nps = D_UNKNOWN;
 365                 _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE);
 366                 break;
 367         case 7:
 368                 if (fp != FP_STONITH)
 369                         dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n");
 370                 ex_to_string = "peer was stonithed";
 371                 nps = D_OUTDATED;
 372                 break;
 373         default:
 374                 /* The script is broken ... */
 375                 nps = D_UNKNOWN;
 376                 dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 377                 return nps;
 378         }
 379
 380         dev_info(DEV, "fence-peer helper returned %d (%s)\n",
 381                         (r>>8) & 0xff, ex_to_string);
 382
 383 out:
 384         if (mdev->state.susp_fen && nps >= D_UNKNOWN) {
 385                 /* The handler was not successful... unfreeze here, the
 386                    state engine can not unfreeze... */
 387                 _drbd_request_state(mdev, NS(susp_fen, 0), CS_VERBOSE);
 388         }
 389
 390         return nps;
 391 }
 392
 393 static int _try_outdate_peer_async(void *data)
 394 {
 395         struct drbd_conf *mdev = (struct drbd_conf *)data;
 396         enum drbd_disk_state nps;
 397         union drbd_state ns;
 398
 399         nps = drbd_try_outdate_peer(mdev);
 400
 401         /* Not using
 402            drbd_request_state(mdev, NS(pdsk, nps));
 403            here, because we might were able to re-establish the connection
 404            in the meantime. This can only partially be solved in the state's
 405            engine is_valid_state() and is_valid_state_transition()
 406            functions.
 407
 408            nps can be D_INCONSISTENT, D_OUTDATED or D_UNKNOWN.
 409            pdsk == D_INCONSISTENT while conn >= C_CONNECTED is valid,
 410            therefore we have to have the pre state change check here.
 411         */
 412         spin_lock_irq(&mdev->tconn->req_lock);
 413         ns = mdev->state;
 414         if (ns.conn < C_WF_REPORT_PARAMS) {
 415                 ns.pdsk = nps;
 416                 _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
 417         }
 418         spin_unlock_irq(&mdev->tconn->req_lock);
 419
 420         return 0;
 421 }
 422
 423 void drbd_try_outdate_peer_async(struct drbd_conf *mdev)
 424 {
 425         struct task_struct *opa;
 426
 427         opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev));
 428         if (IS_ERR(opa))
 429                 dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n");
 430 }
 431
 432 enum drbd_state_rv
 433 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 434 {
 435         const int max_tries = 4;
 436         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
 437         int try = 0;
 438         int forced = 0;
 439         union drbd_state mask, val;
 440         enum drbd_disk_state nps;
 441
 442         if (new_role == R_PRIMARY)
 443                 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
 444
 445         mutex_lock(mdev->state_mutex);
 446
 447         mask.i = 0; mask.role = R_MASK;
 448         val.i  = 0; val.role  = new_role;
 449
 450         while (try++ < max_tries) {
 451                 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 452
 453                 /* in case we first succeeded to outdate,
 454                  * but now suddenly could establish a connection */
 455                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 456                         val.pdsk = 0;
 457                         mask.pdsk = 0;
 458                         continue;
 459                 }
 460
 461                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
 462                     (mdev->state.disk < D_UP_TO_DATE &&
 463                      mdev->state.disk >= D_INCONSISTENT)) {
 464                         mask.disk = D_MASK;
 465                         val.disk  = D_UP_TO_DATE;
 466                         forced = 1;
 467                         continue;
 468                 }
 469
 470                 if (rv == SS_NO_UP_TO_DATE_DISK &&
 471                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 472                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 473                         nps = drbd_try_outdate_peer(mdev);
 474
 475                         if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
 476                                 val.disk = D_UP_TO_DATE;
 477                                 mask.disk = D_MASK;
 478                         }
 479
 480                         val.pdsk = nps;
 481                         mask.pdsk = D_MASK;
 482
 483                         continue;
 484                 }
 485
 486                 if (rv == SS_NOTHING_TO_DO)
 487                         goto out;
 488                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
 489                         nps = drbd_try_outdate_peer(mdev);
 490
 491                         if (force && nps > D_OUTDATED) {
 492                                 dev_warn(DEV, "Forced into split brain situation!\n");
 493                                 nps = D_OUTDATED;
 494                         }
 495
 496                         mask.pdsk = D_MASK;
 497                         val.pdsk  = nps;
 498
 499                         continue;
 500                 }
 501                 if (rv == SS_TWO_PRIMARIES) {
 502                         /* Maybe the peer is detected as dead very soon...
 503                            retry at most once more in this case. */
 504                         schedule_timeout_interruptible((mdev->tconn->net_conf->ping_timeo+1)*HZ/10);
 505                         if (try < max_tries)
 506                                 try = max_tries - 1;
 507                         continue;
 508                 }
 509                 if (rv < SS_SUCCESS) {
 510                         rv = _drbd_request_state(mdev, mask, val,
 511                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 512                         if (rv < SS_SUCCESS)
 513                                 goto out;
 514                 }
 515                 break;
 516         }
 517
 518         if (rv < SS_SUCCESS)
 519                 goto out;
 520
 521         if (forced)
 522                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 523
 524         /* Wait until nothing is on the fly :) */
 525         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 526
 527         if (new_role == R_SECONDARY) {
 528                 set_disk_ro(mdev->vdisk, true);
 529                 if (get_ldev(mdev)) {
 530                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 531                         put_ldev(mdev);
 532                 }
 533         } else {
 534                 if (get_net_conf(mdev->tconn)) {
 535                         mdev->tconn->net_conf->want_lose = 0;
 536                         put_net_conf(mdev->tconn);
 537                 }
 538                 set_disk_ro(mdev->vdisk, false);
 539                 if (get_ldev(mdev)) {
 540                         if (((mdev->state.conn < C_CONNECTED ||
 541                                mdev->state.pdsk <= D_FAILED)
 542                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 543                                 drbd_uuid_new_current(mdev);
 544
 545                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 546                         put_ldev(mdev);
 547                 }
 548         }
 549
 550         /* writeout of activity log covered areas of the bitmap
 551          * to stable storage done in after state change already */
 552
 553         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 554                 /* if this was forced, we should consider sync */
 555                 if (forced)
 556                         drbd_send_uuids(mdev);
 557                 drbd_send_state(mdev);
 558         }
 559
 560         drbd_md_sync(mdev);
 561
 562         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 563 out:
 564         mutex_unlock(mdev->state_mutex);
 565         return rv;
 566 }
 567
 568 static const char *from_attrs_err_to_txt(int err)
 569 {
 570         return  err == -ENOMSG ? "required attribute missing" :
 571                 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
 572                 "invalid attribute value";
 573 }
 574
 575 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
 576 {
 577         struct set_role_parms parms;
 578         int err;
 579         enum drbd_ret_code retcode;
 580
 581         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
 582         if (!adm_ctx.reply_skb)
 583                 return retcode;
 584         if (retcode != NO_ERROR)
 585                 goto out;
 586
 587         memset(&parms, 0, sizeof(parms));
 588         if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
 589                 err = set_role_parms_from_attrs(&parms, info->attrs);
 590                 if (err) {
 591                         retcode = ERR_MANDATORY_TAG;
 592                         drbd_msg_put_info(from_attrs_err_to_txt(err));
 593                         goto out;
 594                 }
 595         }
 596
 597         if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
 598                 retcode = drbd_set_role(adm_ctx.mdev, R_PRIMARY, parms.assume_uptodate);
 599         else
 600                 retcode = drbd_set_role(adm_ctx.mdev, R_SECONDARY, 0);
 601 out:
 602         drbd_adm_finish(info, retcode);
 603         return 0;
 604 }
 605
 606 /* initializes the md.*_offset members, so we are able to find
 607  * the on disk meta data */
 608 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 609                                        struct drbd_backing_dev *bdev)
 610 {
 611         sector_t md_size_sect = 0;
 612         switch (bdev->dc.meta_dev_idx) {
 613         default:
 614                 /* v07 style fixed size indexed meta data */
 615                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 616                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 617                 bdev->md.al_offset = MD_AL_OFFSET;
 618                 bdev->md.bm_offset = MD_BM_OFFSET;
 619                 break;
 620         case DRBD_MD_INDEX_FLEX_EXT:
 621                 /* just occupy the full device; unit: sectors */
 622                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 623                 bdev->md.md_offset = 0;
 624                 bdev->md.al_offset = MD_AL_OFFSET;
 625                 bdev->md.bm_offset = MD_BM_OFFSET;
 626                 break;
 627         case DRBD_MD_INDEX_INTERNAL:
 628         case DRBD_MD_INDEX_FLEX_INT:
 629                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 630                 /* al size is still fixed */
 631                 bdev->md.al_offset = -MD_AL_SECTORS;
 632                 /* we need (slightly less than) ~ this much bitmap sectors: */
 633                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 634                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 635                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 636                 md_size_sect = ALIGN(md_size_sect, 8);
 637
 638                 /* plus the "drbd meta data super block",
 639                  * and the activity log; */
 640                 md_size_sect += MD_BM_OFFSET;
 641
 642                 bdev->md.md_size_sect = md_size_sect;
 643                 /* bitmap offset is adjusted by 'super' block size */
 644                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 645                 break;
 646         }
 647 }
 648
 649 /* input size is expected to be in KB */
 650 char *ppsize(char *buf, unsigned long long size)
 651 {
 652         /* Needs 9 bytes at max including trailing NUL:
 653          * -1ULL ==> "16384 EB" */
 654         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 655         int base = 0;
 656         while (size >= 10000 && base < sizeof(units)-1) {
 657                 /* shift + round */
 658                 size = (size >> 10) + !!(size & (1<<9));
 659                 base++;
 660         }
 661         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
 662
 663         return buf;
 664 }
 665
 666 /* there is still a theoretical deadlock when called from receiver
 667  * on an D_INCONSISTENT R_PRIMARY:
 668  *  remote READ does inc_ap_bio, receiver would need to receive answer
 669  *  packet from remote to dec_ap_bio again.
 670  *  receiver receive_sizes(), comes here,
 671  *  waits for ap_bio_cnt == 0. -> deadlock.
 672  * but this cannot happen, actually, because:
 673  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 674  *  (not connected, or bad/no disk on peer):
 675  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 676  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 677  *  peer may not initiate a resize.
 678  */
 679 /* Note these are not to be confused with
 680  * drbd_adm_suspend_io/drbd_adm_resume_io,
 681  * which are (sub) state changes triggered by admin (drbdsetup),
 682  * and can be long lived.
 683  * This changes an mdev->flag, is triggered by drbd internals,
 684  * and should be short-lived. */
 685 void drbd_suspend_io(struct drbd_conf *mdev)
 686 {
 687         set_bit(SUSPEND_IO, &mdev->flags);
 688         if (is_susp(mdev->state))
 689                 return;
 690         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 691 }
 692
 693 void drbd_resume_io(struct drbd_conf *mdev)
 694 {
 695         clear_bit(SUSPEND_IO, &mdev->flags);
 696         wake_up(&mdev->misc_wait);
 697 }
 698
 699 /**
 700  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 701  * @mdev:       DRBD device.
 702  *
 703  * Returns 0 on success, negative return values indicate errors.
 704  * You should call drbd_md_sync() after calling this function.
 705  */
 706 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 707 {
 708         sector_t prev_first_sect, prev_size; /* previous meta location */
 709         sector_t la_size;
 710         sector_t size;
 711         char ppb[10];
 712
 713         int md_moved, la_size_changed;
 714         enum determine_dev_size rv = unchanged;
 715
 716         /* race:
 717          * application request passes inc_ap_bio,
 718          * but then cannot get an AL-reference.
 719          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 720          *
 721          * to avoid that:
 722          * Suspend IO right here.
 723          * still lock the act_log to not trigger ASSERTs there.
 724          */
 725         drbd_suspend_io(mdev);
 726
 727         /* no wait necessary anymore, actually we could assert that */
 728         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 729
 730         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 731         prev_size = mdev->ldev->md.md_size_sect;
 732         la_size = mdev->ldev->md.la_size_sect;
 733
 734         /* TODO: should only be some assert here, not (re)init... */
 735         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 736
 737         size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
 738
 739         if (drbd_get_capacity(mdev->this_bdev) != size ||
 740             drbd_bm_capacity(mdev) != size) {
 741                 int err;
 742                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 743                 if (unlikely(err)) {
 744                         /* currently there is only one error: ENOMEM! */
 745                         size = drbd_bm_capacity(mdev)>>1;
 746                         if (size == 0) {
 747                                 dev_err(DEV, "OUT OF MEMORY! "
 748                                     "Could not allocate bitmap!\n");
 749                         } else {
 750                                 dev_err(DEV, "BM resizing failed. "
 751                                     "Leaving size unchanged at size = %lu KB\n",
 752                                     (unsigned long)size);
 753                         }
 754                         rv = dev_size_error;
 755                 }
 756                 /* racy, see comments above. */
 757                 drbd_set_my_capacity(mdev, size);
 758                 mdev->ldev->md.la_size_sect = size;
 759                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 760                      (unsigned long long)size>>1);
 761         }
 762         if (rv == dev_size_error)
 763                 goto out;
 764
 765         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 766
 767         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 768                 || prev_size       != mdev->ldev->md.md_size_sect;
 769
 770         if (la_size_changed || md_moved) {
 771                 int err;
 772
 773                 drbd_al_shrink(mdev); /* All extents inactive. */
 774                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 775                          la_size_changed && md_moved ? "size changed and md moved" :
 776                          la_size_changed ? "size changed" : "md moved");
 777                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
 778                 err = drbd_bitmap_io(mdev, &drbd_bm_write,
 779                                 "size changed", BM_LOCKED_MASK);
 780                 if (err) {
 781                         rv = dev_size_error;
 782                         goto out;
 783                 }
 784                 drbd_md_mark_dirty(mdev);
 785         }
 786
 787         if (size > la_size)
 788                 rv = grew;
 789         if (size < la_size)
 790                 rv = shrunk;
 791 out:
 792         lc_unlock(mdev->act_log);
 793         wake_up(&mdev->al_wait);
 794         drbd_resume_io(mdev);
 795
 796         return rv;
 797 }
 798
 799 sector_t
 800 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
 801 {
 802         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 803         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 804         sector_t m_size; /* my size */
 805         sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
 806         sector_t size = 0;
 807
 808         m_size = drbd_get_max_capacity(bdev);
 809
 810         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 811                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 812                 p_size = m_size;
 813         }
 814
 815         if (p_size && m_size) {
 816                 size = min_t(sector_t, p_size, m_size);
 817         } else {
 818                 if (la_size) {
 819                         size = la_size;
 820                         if (m_size && m_size < size)
 821                                 size = m_size;
 822                         if (p_size && p_size < size)
 823                                 size = p_size;
 824                 } else {
 825                         if (m_size)
 826                                 size = m_size;
 827                         if (p_size)
 828                                 size = p_size;
 829                 }
 830         }
 831
 832         if (size == 0)
 833                 dev_err(DEV, "Both nodes diskless!\n");
 834
 835         if (u_size) {
 836                 if (u_size > size)
 837                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 838                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 839                 else
 840                         size = u_size;
 841         }
 842
 843         return size;
 844 }
 845
 846 /**
 847  * drbd_check_al_size() - Ensures that the AL is of the right size
 848  * @mdev:       DRBD device.
 849  *
 850  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 851  * failed, and 0 on success. You should call drbd_md_sync() after you called
 852  * this function.
 853  */
 854 static int drbd_check_al_size(struct drbd_conf *mdev)
 855 {
 856         struct lru_cache *n, *t;
 857         struct lc_element *e;
 858         unsigned int in_use;
 859         int i;
 860
 861         if (!expect(mdev->sync_conf.al_extents >= DRBD_AL_EXTENTS_MIN))
 862                 mdev->sync_conf.al_extents = DRBD_AL_EXTENTS_MIN;
 863
 864         if (mdev->act_log &&
 865             mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
 866                 return 0;
 867
 868         in_use = 0;
 869         t = mdev->act_log;
 870         n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
 871                 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
 872
 873         if (n == NULL) {
 874                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 875                 return -ENOMEM;
 876         }
 877         spin_lock_irq(&mdev->al_lock);
 878         if (t) {
 879                 for (i = 0; i < t->nr_elements; i++) {
 880                         e = lc_element_by_index(t, i);
 881                         if (e->refcnt)
 882                                 dev_err(DEV, "refcnt(%d)==%d\n",
 883                                     e->lc_number, e->refcnt);
 884                         in_use += e->refcnt;
 885                 }
 886         }
 887         if (!in_use)
 888                 mdev->act_log = n;
 889         spin_unlock_irq(&mdev->al_lock);
 890         if (in_use) {
 891                 dev_err(DEV, "Activity log still in use!\n");
 892                 lc_destroy(n);
 893                 return -EBUSY;
 894         } else {
 895                 if (t)
 896                         lc_destroy(t);
 897         }
 898         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 899         return 0;
 900 }
 901
 902 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
 903 {
 904         struct request_queue * const q = mdev->rq_queue;
 905         int max_hw_sectors = max_bio_size >> 9;
 906         int max_segments = 0;
 907
 908         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 909                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 910
 911                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
 912                 max_segments = mdev->ldev->dc.max_bio_bvecs;
 913                 put_ldev(mdev);
 914         }
 915
 916         blk_queue_logical_block_size(q, 512);
 917         blk_queue_max_hw_sectors(q, max_hw_sectors);
 918         /* This is the workaround for "bio would need to, but cannot, be split" */
 919         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
 920         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
 921
 922         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 923                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 924
 925                 blk_queue_stack_limits(q, b);
 926
 927                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
 928                         dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
 929                                  q->backing_dev_info.ra_pages,
 930                                  b->backing_dev_info.ra_pages);
 931                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
 932                 }
 933                 put_ldev(mdev);
 934         }
 935 }
 936
 937 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
 938 {
 939         int now, new, local, peer;
 940
 941         now = queue_max_hw_sectors(mdev->rq_queue) << 9;
 942         local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
 943         peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
 944
 945         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 946                 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
 947                 mdev->local_max_bio_size = local;
 948                 put_ldev(mdev);
 949         }
 950
 951         /* We may ignore peer limits if the peer is modern enough.
 952            Because new from 8.3.8 onwards the peer can use multiple
 953            BIOs for a single peer_request */
 954         if (mdev->state.conn >= C_CONNECTED) {
 955                 if (mdev->tconn->agreed_pro_version < 94)
 956                         peer = mdev->peer_max_bio_size;
 957                 else if (mdev->tconn->agreed_pro_version == 94)
 958                         peer = DRBD_MAX_SIZE_H80_PACKET;
 959                 else /* drbd 8.3.8 onwards */
 960                         peer = DRBD_MAX_BIO_SIZE;
 961         }
 962
 963         new = min_t(int, local, peer);
 964
 965         if (mdev->state.role == R_PRIMARY && new < now)
 966                 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
 967
 968         if (new != now)
 969                 dev_info(DEV, "max BIO size = %u\n", new);
 970
 971         drbd_setup_queue_param(mdev, new);
 972 }
 973
 974 /* serialize deconfig (worker exiting, doing cleanup)
 975  * and reconfig (drbdsetup disk, drbdsetup net)
 976  *
 977  * Wait for a potentially exiting worker, then restart it,
 978  * or start a new one.  Flush any pending work, there may still be an
 979  * after_state_change queued.
 980  */
 981 static void conn_reconfig_start(struct drbd_tconn *tconn)
 982 {
 983         wait_event(tconn->ping_wait, !test_and_set_bit(CONFIG_PENDING, &tconn->flags));
 984         wait_event(tconn->ping_wait, !test_bit(OBJECT_DYING, &tconn->flags));
 985         drbd_thread_start(&tconn->worker);
 986         conn_flush_workqueue(tconn);
 987 }
 988
 989 /* if still unconfigured, stops worker again.
 990  * if configured now, clears CONFIG_PENDING.
 991  * wakes potential waiters */
 992 static void conn_reconfig_done(struct drbd_tconn *tconn)
 993 {
 994         spin_lock_irq(&tconn->req_lock);
 995         if (conn_all_vols_unconf(tconn)) {
 996                 set_bit(OBJECT_DYING, &tconn->flags);
 997                 drbd_thread_stop_nowait(&tconn->worker);
 998         } else
 999                 clear_bit(CONFIG_PENDING, &tconn->flags);
1000         spin_unlock_irq(&tconn->req_lock);
1001         wake_up(&tconn->ping_wait);
1002 }
1003
1004 /* Make sure IO is suspended before calling this function(). */
1005 static void drbd_suspend_al(struct drbd_conf *mdev)
1006 {
1007         int s = 0;
1008
1009         if (!lc_try_lock(mdev->act_log)) {
1010                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
1011                 return;
1012         }
1013
1014         drbd_al_shrink(mdev);
1015         spin_lock_irq(&mdev->tconn->req_lock);
1016         if (mdev->state.conn < C_CONNECTED)
1017                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
1018         spin_unlock_irq(&mdev->tconn->req_lock);
1019         lc_unlock(mdev->act_log);
1020
1021         if (s)
1022                 dev_info(DEV, "Suspended AL updates\n");
1023 }
1024
1025 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1026 {
1027         struct drbd_conf *mdev;
1028         int err;
1029         enum drbd_ret_code retcode;
1030         enum determine_dev_size dd;
1031         sector_t max_possible_sectors;
1032         sector_t min_md_device_sectors;
1033         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1034         struct block_device *bdev;
1035         struct lru_cache *resync_lru = NULL;
1036         union drbd_state ns, os;
1037         enum drbd_state_rv rv;
1038         int cp_discovered = 0;
1039
1040         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1041         if (!adm_ctx.reply_skb)
1042                 return retcode;
1043         if (retcode != NO_ERROR)
1044                 goto fail;
1045
1046         mdev = adm_ctx.mdev;
1047         conn_reconfig_start(mdev->tconn);
1048
1049         /* if you want to reconfigure, please tear down first */
1050         if (mdev->state.disk > D_DISKLESS) {
1051                 retcode = ERR_DISK_CONFIGURED;
1052                 goto fail;
1053         }
1054         /* It may just now have detached because of IO error.  Make sure
1055          * drbd_ldev_destroy is done already, we may end up here very fast,
1056          * e.g. if someone calls attach from the on-io-error handler,
1057          * to realize a "hot spare" feature (not that I'd recommend that) */
1058         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1059
1060         /* allocation not in the IO path, drbdsetup context */
1061         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1062         if (!nbc) {
1063                 retcode = ERR_NOMEM;
1064                 goto fail;
1065         }
1066
1067         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
1068         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
1069         nbc->dc.fencing       = DRBD_FENCING_DEF;
1070         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
1071
1072         err = disk_conf_from_attrs(&nbc->dc, info->attrs);
1073         if (err) {
1074                 retcode = ERR_MANDATORY_TAG;
1075                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1076                 goto fail;
1077         }
1078
1079         if ((int)nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1080                 retcode = ERR_MD_IDX_INVALID;
1081                 goto fail;
1082         }
1083
1084         if (get_net_conf(mdev->tconn)) {
1085                 int prot = mdev->tconn->net_conf->wire_protocol;
1086                 put_net_conf(mdev->tconn);
1087                 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
1088                         retcode = ERR_STONITH_AND_PROT_A;
1089                         goto fail;
1090                 }
1091         }
1092
1093         bdev = blkdev_get_by_path(nbc->dc.backing_dev,
1094                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1095         if (IS_ERR(bdev)) {
1096                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
1097                         PTR_ERR(bdev));
1098                 retcode = ERR_OPEN_DISK;
1099                 goto fail;
1100         }
1101         nbc->backing_bdev = bdev;
1102
1103         /*
1104          * meta_dev_idx >= 0: external fixed size, possibly multiple
1105          * drbd sharing one meta device.  TODO in that case, paranoia
1106          * check that [md_bdev, meta_dev_idx] is not yet used by some
1107          * other drbd minor!  (if you use drbd.conf + drbdadm, that
1108          * should check it for you already; but if you don't, or
1109          * someone fooled it, we need to double check here)
1110          */
1111         bdev = blkdev_get_by_path(nbc->dc.meta_dev,
1112                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1113                                   ((int)nbc->dc.meta_dev_idx < 0) ?
1114                                   (void *)mdev : (void *)drbd_m_holder);
1115         if (IS_ERR(bdev)) {
1116                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
1117                         PTR_ERR(bdev));
1118                 retcode = ERR_OPEN_MD_DISK;
1119                 goto fail;
1120         }
1121         nbc->md_bdev = bdev;
1122
1123         if ((nbc->backing_bdev == nbc->md_bdev) !=
1124             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1125              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1126                 retcode = ERR_MD_IDX_INVALID;
1127                 goto fail;
1128         }
1129
1130         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1131                         1, 61, sizeof(struct bm_extent),
1132                         offsetof(struct bm_extent, lce));
1133         if (!resync_lru) {
1134                 retcode = ERR_NOMEM;
1135                 goto fail;
1136         }
1137
1138         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1139         drbd_md_set_sector_offsets(mdev, nbc);
1140
1141         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
1142                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1143                         (unsigned long long) drbd_get_max_capacity(nbc),
1144                         (unsigned long long) nbc->dc.disk_size);
1145                 retcode = ERR_DISK_TO_SMALL;
1146                 goto fail;
1147         }
1148
1149         if ((int)nbc->dc.meta_dev_idx < 0) {
1150                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1151                 /* at least one MB, otherwise it does not make sense */
1152                 min_md_device_sectors = (2<<10);
1153         } else {
1154                 max_possible_sectors = DRBD_MAX_SECTORS;
1155                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
1156         }
1157
1158         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1159                 retcode = ERR_MD_DISK_TO_SMALL;
1160                 dev_warn(DEV, "refusing attach: md-device too small, "
1161                      "at least %llu sectors needed for this meta-disk type\n",
1162                      (unsigned long long) min_md_device_sectors);
1163                 goto fail;
1164         }
1165
1166         /* Make sure the new disk is big enough
1167          * (we may currently be R_PRIMARY with no local disk...) */
1168         if (drbd_get_max_capacity(nbc) <
1169             drbd_get_capacity(mdev->this_bdev)) {
1170                 retcode = ERR_DISK_TO_SMALL;
1171                 goto fail;
1172         }
1173
1174         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1175
1176         if (nbc->known_size > max_possible_sectors) {
1177                 dev_warn(DEV, "==> truncating very big lower level device "
1178                         "to currently maximum possible %llu sectors <==\n",
1179                         (unsigned long long) max_possible_sectors);
1180                 if ((int)nbc->dc.meta_dev_idx >= 0)
1181                         dev_warn(DEV, "==>> using internal or flexible "
1182                                       "meta data may help <<==\n");
1183         }
1184
1185         drbd_suspend_io(mdev);
1186         /* also wait for the last barrier ack. */
1187         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state));
1188         /* and for any other previously queued work */
1189         drbd_flush_workqueue(mdev);
1190
1191         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1192         retcode = rv;  /* FIXME: Type mismatch. */
1193         drbd_resume_io(mdev);
1194         if (rv < SS_SUCCESS)
1195                 goto fail;
1196
1197         if (!get_ldev_if_state(mdev, D_ATTACHING))
1198                 goto force_diskless;
1199
1200         drbd_md_set_sector_offsets(mdev, nbc);
1201
1202         if (!mdev->bitmap) {
1203                 if (drbd_bm_init(mdev)) {
1204                         retcode = ERR_NOMEM;
1205                         goto force_diskless_dec;
1206                 }
1207         }
1208
1209         retcode = drbd_md_read(mdev, nbc);
1210         if (retcode != NO_ERROR)
1211                 goto force_diskless_dec;
1212
1213         if (mdev->state.conn < C_CONNECTED &&
1214             mdev->state.role == R_PRIMARY &&
1215             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1216                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1217                     (unsigned long long)mdev->ed_uuid);
1218                 retcode = ERR_DATA_NOT_CURRENT;
1219                 goto force_diskless_dec;
1220         }
1221
1222         /* Since we are diskless, fix the activity log first... */
1223         if (drbd_check_al_size(mdev)) {
1224                 retcode = ERR_NOMEM;
1225                 goto force_diskless_dec;
1226         }
1227
1228         /* Prevent shrinking of consistent devices ! */
1229         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1230             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1231                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1232                 retcode = ERR_DISK_TO_SMALL;
1233                 goto force_diskless_dec;
1234         }
1235
1236         if (!drbd_al_read_log(mdev, nbc)) {
1237                 retcode = ERR_IO_MD_DISK;
1238                 goto force_diskless_dec;
1239         }
1240
1241         /* Reset the "barriers don't work" bits here, then force meta data to
1242          * be written, to ensure we determine if barriers are supported. */
1243         if (nbc->dc.no_md_flush)
1244                 set_bit(MD_NO_FUA, &mdev->flags);
1245         else
1246                 clear_bit(MD_NO_FUA, &mdev->flags);
1247
1248         /* Point of no return reached.
1249          * Devices and memory are no longer released by error cleanup below.
1250          * now mdev takes over responsibility, and the state engine should
1251          * clean it up somewhere.  */
1252         D_ASSERT(mdev->ldev == NULL);
1253         mdev->ldev = nbc;
1254         mdev->resync = resync_lru;
1255         nbc = NULL;
1256         resync_lru = NULL;
1257
1258         mdev->write_ordering = WO_bdev_flush;
1259         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1260
1261         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1262                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1263         else
1264                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1265
1266         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1267             !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) {
1268                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1269                 cp_discovered = 1;
1270         }
1271
1272         mdev->send_cnt = 0;
1273         mdev->recv_cnt = 0;
1274         mdev->read_cnt = 0;
1275         mdev->writ_cnt = 0;
1276
1277         drbd_reconsider_max_bio_size(mdev);
1278
1279         /* If I am currently not R_PRIMARY,
1280          * but meta data primary indicator is set,
1281          * I just now recover from a hard crash,
1282          * and have been R_PRIMARY before that crash.
1283          *
1284          * Now, if I had no connection before that crash
1285          * (have been degraded R_PRIMARY), chances are that
1286          * I won't find my peer now either.
1287          *
1288          * In that case, and _only_ in that case,
1289          * we use the degr-wfc-timeout instead of the default,
1290          * so we can automatically recover from a crash of a
1291          * degraded but active "cluster" after a certain timeout.
1292          */
1293         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1294         if (mdev->state.role != R_PRIMARY &&
1295              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1296             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1297                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1298
1299         dd = drbd_determine_dev_size(mdev, 0);
1300         if (dd == dev_size_error) {
1301                 retcode = ERR_NOMEM_BITMAP;
1302                 goto force_diskless_dec;
1303         } else if (dd == grew)
1304                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1305
1306         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1307                 dev_info(DEV, "Assuming that all blocks are out of sync "
1308                      "(aka FullSync)\n");
1309                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1310                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1311                         retcode = ERR_IO_MD_DISK;
1312                         goto force_diskless_dec;
1313                 }
1314         } else {
1315                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1316                         "read from attaching", BM_LOCKED_MASK) < 0) {
1317                         retcode = ERR_IO_MD_DISK;
1318                         goto force_diskless_dec;
1319                 }
1320         }
1321
1322         if (cp_discovered) {
1323                 drbd_al_apply_to_bm(mdev);
1324                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1325                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1326                         retcode = ERR_IO_MD_DISK;
1327                         goto force_diskless_dec;
1328                 }
1329         }
1330
1331         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1332                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1333
1334         spin_lock_irq(&mdev->tconn->req_lock);
1335         os = mdev->state;
1336         ns.i = os.i;
1337         /* If MDF_CONSISTENT is not set go into inconsistent state,
1338            otherwise investigate MDF_WasUpToDate...
1339            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1340            otherwise into D_CONSISTENT state.
1341         */
1342         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1343                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1344                         ns.disk = D_CONSISTENT;
1345                 else
1346                         ns.disk = D_OUTDATED;
1347         } else {
1348                 ns.disk = D_INCONSISTENT;
1349         }
1350
1351         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1352                 ns.pdsk = D_OUTDATED;
1353
1354         if ( ns.disk == D_CONSISTENT &&
1355             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1356                 ns.disk = D_UP_TO_DATE;
1357
1358         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1359            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1360            this point, because drbd_request_state() modifies these
1361            flags. */
1362
1363         /* In case we are C_CONNECTED postpone any decision on the new disk
1364            state after the negotiation phase. */
1365         if (mdev->state.conn == C_CONNECTED) {
1366                 mdev->new_state_tmp.i = ns.i;
1367                 ns.i = os.i;
1368                 ns.disk = D_NEGOTIATING;
1369
1370                 /* We expect to receive up-to-date UUIDs soon.
1371                    To avoid a race in receive_state, free p_uuid while
1372                    holding req_lock. I.e. atomic with the state change */
1373                 kfree(mdev->p_uuid);
1374                 mdev->p_uuid = NULL;
1375         }
1376
1377         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1378         ns = mdev->state;
1379         spin_unlock_irq(&mdev->tconn->req_lock);
1380
1381         if (rv < SS_SUCCESS)
1382                 goto force_diskless_dec;
1383
1384         if (mdev->state.role == R_PRIMARY)
1385                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1386         else
1387                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1388
1389         drbd_md_mark_dirty(mdev);
1390         drbd_md_sync(mdev);
1391
1392         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1393         put_ldev(mdev);
1394         conn_reconfig_done(mdev->tconn);
1395         drbd_adm_finish(info, retcode);
1396         return 0;
1397
1398  force_diskless_dec:
1399         put_ldev(mdev);
1400  force_diskless:
1401         drbd_force_state(mdev, NS(disk, D_FAILED));
1402         drbd_md_sync(mdev);
1403         conn_reconfig_done(mdev->tconn);
1404  fail:
1405         if (nbc) {
1406                 if (nbc->backing_bdev)
1407                         blkdev_put(nbc->backing_bdev,
1408                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1409                 if (nbc->md_bdev)
1410                         blkdev_put(nbc->md_bdev,
1411                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1412                 kfree(nbc);
1413         }
1414         lc_destroy(resync_lru);
1415
1416         drbd_adm_finish(info, retcode);
1417         return 0;
1418 }
1419
1420 static int adm_detach(struct drbd_conf *mdev)
1421 {
1422         enum drbd_ret_code retcode;
1423         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1424         retcode = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1425         wait_event(mdev->misc_wait,
1426                         mdev->state.disk != D_DISKLESS ||
1427                         !atomic_read(&mdev->local_cnt));
1428         drbd_resume_io(mdev);
1429         return retcode;
1430 }
1431
1432 /* Detaching the disk is a process in multiple stages.  First we need to lock
1433  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1434  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1435  * internal references as well.
1436  * Only then we have finally detached. */
1437 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1438 {
1439         enum drbd_ret_code retcode;
1440
1441         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1442         if (!adm_ctx.reply_skb)
1443                 return retcode;
1444         if (retcode != NO_ERROR)
1445                 goto out;
1446
1447         retcode = adm_detach(adm_ctx.mdev);
1448 out:
1449         drbd_adm_finish(info, retcode);
1450         return 0;
1451 }
1452
1453 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
1454 {
1455         char hmac_name[CRYPTO_MAX_ALG_NAME];
1456         struct drbd_conf *mdev;
1457         struct net_conf *new_conf = NULL;
1458         struct crypto_hash *tfm = NULL;
1459         struct crypto_hash *integrity_w_tfm = NULL;
1460         struct crypto_hash *integrity_r_tfm = NULL;
1461         void *int_dig_out = NULL;
1462         void *int_dig_in = NULL;
1463         void *int_dig_vv = NULL;
1464         struct drbd_tconn *oconn;
1465         struct drbd_tconn *tconn;
1466         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1467         enum drbd_ret_code retcode;
1468         int i;
1469         int err;
1470
1471         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1472         if (!adm_ctx.reply_skb)
1473                 return retcode;
1474         if (retcode != NO_ERROR)
1475                 goto out;
1476
1477         tconn = adm_ctx.tconn;
1478         conn_reconfig_start(tconn);
1479
1480         if (tconn->cstate > C_STANDALONE) {
1481                 retcode = ERR_NET_CONFIGURED;
1482                 goto fail;
1483         }
1484
1485         /* allocation not in the IO path, cqueue thread context */
1486         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1487         if (!new_conf) {
1488                 retcode = ERR_NOMEM;
1489                 goto fail;
1490         }
1491
1492         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1493         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1494         new_conf->ping_int         = DRBD_PING_INT_DEF;
1495         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1496         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1497         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1498         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1499         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1500         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1501         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1502         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1503         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1504         new_conf->want_lose        = 0;
1505         new_conf->two_primaries    = 0;
1506         new_conf->wire_protocol    = DRBD_PROT_C;
1507         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1508         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1509         new_conf->on_congestion    = DRBD_ON_CONGESTION_DEF;
1510         new_conf->cong_extents     = DRBD_CONG_EXTENTS_DEF;
1511
1512         err = net_conf_from_attrs(new_conf, info->attrs);
1513         if (err) {
1514                 retcode = ERR_MANDATORY_TAG;
1515                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1516                 goto fail;
1517         }
1518
1519         if (new_conf->two_primaries
1520             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1521                 retcode = ERR_NOT_PROTO_C;
1522                 goto fail;
1523         }
1524
1525         idr_for_each_entry(&tconn->volumes, mdev, i) {
1526                 if (get_ldev(mdev)) {
1527                         enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1528                         put_ldev(mdev);
1529                         if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1530                                 retcode = ERR_STONITH_AND_PROT_A;
1531                                 goto fail;
1532                         }
1533                 }
1534                 if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1535                         retcode = ERR_DISCARD;
1536                         goto fail;
1537                 }
1538                 if (!mdev->bitmap) {
1539                         if(drbd_bm_init(mdev)) {
1540                                 retcode = ERR_NOMEM;
1541                                 goto fail;
1542                         }
1543                 }
1544         }
1545
1546         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1547                 retcode = ERR_CONG_NOT_PROTO_A;
1548                 goto fail;
1549         }
1550
1551         retcode = NO_ERROR;
1552
1553         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1554         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1555
1556         /* No need to take drbd_cfg_mutex here.  All reconfiguration is
1557          * strictly serialized on genl_lock(). We are protected against
1558          * concurrent reconfiguration/addition/deletion */
1559         list_for_each_entry(oconn, &drbd_tconns, all_tconn) {
1560                 if (oconn == tconn)
1561                         continue;
1562                 if (get_net_conf(oconn)) {
1563                         taken_addr = (struct sockaddr *)&oconn->net_conf->my_addr;
1564                         if (new_conf->my_addr_len == oconn->net_conf->my_addr_len &&
1565                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1566                                 retcode = ERR_LOCAL_ADDR;
1567
1568                         taken_addr = (struct sockaddr *)&oconn->net_conf->peer_addr;
1569                         if (new_conf->peer_addr_len == oconn->net_conf->peer_addr_len &&
1570                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1571                                 retcode = ERR_PEER_ADDR;
1572
1573                         put_net_conf(oconn);
1574                         if (retcode != NO_ERROR)
1575                                 goto fail;
1576                 }
1577         }
1578
1579         if (new_conf->cram_hmac_alg[0] != 0) {
1580                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1581                         new_conf->cram_hmac_alg);
1582                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1583                 if (IS_ERR(tfm)) {
1584                         tfm = NULL;
1585                         retcode = ERR_AUTH_ALG;
1586                         goto fail;
1587                 }
1588
1589                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1590                         retcode = ERR_AUTH_ALG_ND;
1591                         goto fail;
1592                 }
1593         }
1594
1595         if (new_conf->integrity_alg[0]) {
1596                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1597                 if (IS_ERR(integrity_w_tfm)) {
1598                         integrity_w_tfm = NULL;
1599                         retcode=ERR_INTEGRITY_ALG;
1600                         goto fail;
1601                 }
1602
1603                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1604                         retcode=ERR_INTEGRITY_ALG_ND;
1605                         goto fail;
1606                 }
1607
1608                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1609                 if (IS_ERR(integrity_r_tfm)) {
1610                         integrity_r_tfm = NULL;
1611                         retcode=ERR_INTEGRITY_ALG;
1612                         goto fail;
1613                 }
1614         }
1615
1616         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1617
1618         /* allocation not in the IO path, cqueue thread context */
1619         if (integrity_w_tfm) {
1620                 i = crypto_hash_digestsize(integrity_w_tfm);
1621                 int_dig_out = kmalloc(i, GFP_KERNEL);
1622                 if (!int_dig_out) {
1623                         retcode = ERR_NOMEM;
1624                         goto fail;
1625                 }
1626                 int_dig_in = kmalloc(i, GFP_KERNEL);
1627                 if (!int_dig_in) {
1628                         retcode = ERR_NOMEM;
1629                         goto fail;
1630                 }
1631                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1632                 if (!int_dig_vv) {
1633                         retcode = ERR_NOMEM;
1634                         goto fail;
1635                 }
1636         }
1637
1638         conn_flush_workqueue(tconn);
1639         spin_lock_irq(&tconn->req_lock);
1640         if (tconn->net_conf != NULL) {
1641                 retcode = ERR_NET_CONFIGURED;
1642                 spin_unlock_irq(&tconn->req_lock);
1643                 goto fail;
1644         }
1645         tconn->net_conf = new_conf;
1646
1647         crypto_free_hash(tconn->cram_hmac_tfm);
1648         tconn->cram_hmac_tfm = tfm;
1649
1650         crypto_free_hash(tconn->integrity_w_tfm);
1651         tconn->integrity_w_tfm = integrity_w_tfm;
1652
1653         crypto_free_hash(tconn->integrity_r_tfm);
1654         tconn->integrity_r_tfm = integrity_r_tfm;
1655
1656         kfree(tconn->int_dig_out);
1657         kfree(tconn->int_dig_in);
1658         kfree(tconn->int_dig_vv);
1659         tconn->int_dig_out=int_dig_out;
1660         tconn->int_dig_in=int_dig_in;
1661         tconn->int_dig_vv=int_dig_vv;
1662         retcode = _conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
1663         spin_unlock_irq(&tconn->req_lock);
1664
1665         idr_for_each_entry(&tconn->volumes, mdev, i) {
1666                 mdev->send_cnt = 0;
1667                 mdev->recv_cnt = 0;
1668                 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1669         }
1670         conn_reconfig_done(tconn);
1671         drbd_adm_finish(info, retcode);
1672         return 0;
1673
1674 fail:
1675         kfree(int_dig_out);
1676         kfree(int_dig_in);
1677         kfree(int_dig_vv);
1678         crypto_free_hash(tfm);
1679         crypto_free_hash(integrity_w_tfm);
1680         crypto_free_hash(integrity_r_tfm);
1681         kfree(new_conf);
1682
1683         conn_reconfig_done(tconn);
1684 out:
1685         drbd_adm_finish(info, retcode);
1686         return 0;
1687 }
1688
1689 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
1690 {
1691         enum drbd_state_rv rv;
1692         if (force) {
1693                 spin_lock_irq(&tconn->req_lock);
1694                 if (tconn->cstate >= C_WF_CONNECTION)
1695                         _conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
1696                 spin_unlock_irq(&tconn->req_lock);
1697                 return SS_SUCCESS;
1698         }
1699
1700         rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING), 0);
1701
1702         switch (rv) {
1703         case SS_NOTHING_TO_DO:
1704         case SS_ALREADY_STANDALONE:
1705                 return SS_SUCCESS;
1706         case SS_PRIMARY_NOP:
1707                 /* Our state checking code wants to see the peer outdated. */
1708                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
1709                                                         pdsk, D_OUTDATED), CS_VERBOSE);
1710                 break;
1711         case SS_CW_FAILED_BY_PEER:
1712                 /* The peer probably wants to see us outdated. */
1713                 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
1714                                                         disk, D_OUTDATED), 0);
1715                 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
1716                         conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
1717                         rv = SS_SUCCESS;
1718                 }
1719                 break;
1720         default:;
1721                 /* no special handling necessary */
1722         }
1723
1724         return rv;
1725 }
1726
1727 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
1728 {
1729         struct disconnect_parms parms;
1730         struct drbd_tconn *tconn;
1731         enum drbd_state_rv rv;
1732         enum drbd_ret_code retcode;
1733         int err;
1734
1735         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1736         if (!adm_ctx.reply_skb)
1737                 return retcode;
1738         if (retcode != NO_ERROR)
1739                 goto fail;
1740
1741         tconn = adm_ctx.tconn;
1742         memset(&parms, 0, sizeof(parms));
1743         if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
1744                 err = disconnect_parms_from_attrs(&parms, info->attrs);
1745                 if (err) {
1746                         retcode = ERR_MANDATORY_TAG;
1747                         drbd_msg_put_info(from_attrs_err_to_txt(err));
1748                         goto fail;
1749                 }
1750         }
1751
1752         rv = conn_try_disconnect(tconn, parms.force_disconnect);
1753         if (rv < SS_SUCCESS)
1754                 goto fail;
1755
1756         if (wait_event_interruptible(tconn->ping_wait,
1757                                      tconn->cstate != C_DISCONNECTING)) {
1758                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1759                    someone else might connect us in the mean time! */
1760                 retcode = ERR_INTR;
1761                 goto fail;
1762         }
1763
1764         retcode = NO_ERROR;
1765  fail:
1766         drbd_adm_finish(info, retcode);
1767         return 0;
1768 }
1769
1770 void resync_after_online_grow(struct drbd_conf *mdev)
1771 {
1772         int iass; /* I am sync source */
1773
1774         dev_info(DEV, "Resync of new storage after online grow\n");
1775         if (mdev->state.role != mdev->state.peer)
1776                 iass = (mdev->state.role == R_PRIMARY);
1777         else
1778                 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
1779
1780         if (iass)
1781                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1782         else
1783                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1784 }
1785
1786 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
1787 {
1788         struct resize_parms rs;
1789         struct drbd_conf *mdev;
1790         enum drbd_ret_code retcode;
1791         enum determine_dev_size dd;
1792         enum dds_flags ddsf;
1793         int err;
1794
1795         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1796         if (!adm_ctx.reply_skb)
1797                 return retcode;
1798         if (retcode != NO_ERROR)
1799                 goto fail;
1800
1801         memset(&rs, 0, sizeof(struct resize_parms));
1802         if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
1803                 err = resize_parms_from_attrs(&rs, info->attrs);
1804                 if (err) {
1805                         retcode = ERR_MANDATORY_TAG;
1806                         drbd_msg_put_info(from_attrs_err_to_txt(err));
1807                         goto fail;
1808                 }
1809         }
1810
1811         mdev = adm_ctx.mdev;
1812         if (mdev->state.conn > C_CONNECTED) {
1813                 retcode = ERR_RESIZE_RESYNC;
1814                 goto fail;
1815         }
1816
1817         if (mdev->state.role == R_SECONDARY &&
1818             mdev->state.peer == R_SECONDARY) {
1819                 retcode = ERR_NO_PRIMARY;
1820                 goto fail;
1821         }
1822
1823         if (!get_ldev(mdev)) {
1824                 retcode = ERR_NO_DISK;
1825                 goto fail;
1826         }
1827
1828         if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
1829                 retcode = ERR_NEED_APV_93;
1830                 goto fail;
1831         }
1832
1833         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1834                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1835
1836         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1837         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1838         dd = drbd_determine_dev_size(mdev, ddsf);
1839         drbd_md_sync(mdev);
1840         put_ldev(mdev);
1841         if (dd == dev_size_error) {
1842                 retcode = ERR_NOMEM_BITMAP;
1843                 goto fail;
1844         }
1845
1846         if (mdev->state.conn == C_CONNECTED) {
1847                 if (dd == grew)
1848                         set_bit(RESIZE_PENDING, &mdev->flags);
1849
1850                 drbd_send_uuids(mdev);
1851                 drbd_send_sizes(mdev, 1, ddsf);
1852         }
1853
1854  fail:
1855         drbd_adm_finish(info, retcode);
1856         return 0;
1857 }
1858
1859 int drbd_adm_syncer(struct sk_buff *skb, struct genl_info *info)
1860 {
1861         struct drbd_conf *mdev;
1862         enum drbd_ret_code retcode;
1863         int err;
1864         int ovr; /* online verify running */
1865         int rsr; /* re-sync running */
1866         struct crypto_hash *verify_tfm = NULL;
1867         struct crypto_hash *csums_tfm = NULL;
1868         struct syncer_conf sc;
1869         cpumask_var_t new_cpu_mask;
1870         int *rs_plan_s = NULL;
1871         int fifo_size;
1872
1873         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1874         if (!adm_ctx.reply_skb)
1875                 return retcode;
1876         if (retcode != NO_ERROR)
1877                 goto fail;
1878         mdev = adm_ctx.mdev;
1879
1880         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1881                 retcode = ERR_NOMEM;
1882                 drbd_msg_put_info("unable to allocate cpumask");
1883                 goto fail;
1884         }
1885
1886         if (((struct drbd_genlmsghdr*)info->userhdr)->flags
1887                         & DRBD_GENL_F_SET_DEFAULTS) {
1888                 memset(&sc, 0, sizeof(struct syncer_conf));
1889                 sc.rate       = DRBD_RATE_DEF;
1890                 sc.after      = DRBD_AFTER_DEF;
1891                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1892                 sc.on_no_data  = DRBD_ON_NO_DATA_DEF;
1893                 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1894                 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1895                 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1896                 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1897                 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1898         } else
1899                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1900
1901         err = syncer_conf_from_attrs(&sc, info->attrs);
1902         if (err) {
1903                 retcode = ERR_MANDATORY_TAG;
1904                 drbd_msg_put_info(from_attrs_err_to_txt(err));
1905                 goto fail;
1906         }
1907
1908         /* re-sync running */
1909         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1910                 mdev->state.conn == C_SYNC_TARGET ||
1911                 mdev->state.conn == C_PAUSED_SYNC_S ||
1912                 mdev->state.conn == C_PAUSED_SYNC_T );
1913
1914         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1915                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1916                 goto fail;
1917         }
1918
1919         if (!rsr && sc.csums_alg[0]) {
1920                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1921                 if (IS_ERR(csums_tfm)) {
1922                         csums_tfm = NULL;
1923                         retcode = ERR_CSUMS_ALG;
1924                         goto fail;
1925                 }
1926
1927                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1928                         retcode = ERR_CSUMS_ALG_ND;
1929                         goto fail;
1930                 }
1931         }
1932
1933         /* online verify running */
1934         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1935
1936         if (ovr) {
1937                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1938                         retcode = ERR_VERIFY_RUNNING;
1939                         goto fail;
1940                 }
1941         }
1942
1943         if (!ovr && sc.verify_alg[0]) {
1944                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1945                 if (IS_ERR(verify_tfm)) {
1946                         verify_tfm = NULL;
1947                         retcode = ERR_VERIFY_ALG;
1948                         goto fail;
1949                 }
1950
1951                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1952                         retcode = ERR_VERIFY_ALG_ND;
1953                         goto fail;
1954                 }
1955         }
1956
1957         /* silently ignore cpu mask on UP kernel */
1958         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1959                 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1960                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1961                 if (err) {
1962                         dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1963                         retcode = ERR_CPU_MASK_PARSE;
1964                         goto fail;
1965                 }
1966         }
1967
1968         if (!expect(sc.rate >= 1))
1969                 sc.rate = 1;
1970
1971         /* clip to allowed range */
1972         if (!expect(sc.al_extents >= DRBD_AL_EXTENTS_MIN))
1973                 sc.al_extents = DRBD_AL_EXTENTS_MIN;
1974         if (!expect(sc.al_extents <= DRBD_AL_EXTENTS_MAX))
1975                 sc.al_extents = DRBD_AL_EXTENTS_MAX;
1976
1977         /* most sanity checks done, try to assign the new sync-after
1978          * dependency.  need to hold the global lock in there,
1979          * to avoid a race in the dependency loop check. */
1980         retcode = drbd_alter_sa(mdev, sc.after);
1981         if (retcode != NO_ERROR)
1982                 goto fail;
1983
1984         fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1985         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1986                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1987                 if (!rs_plan_s) {
1988                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1989                         retcode = ERR_NOMEM;
1990                         goto fail;
1991                 }
1992         }
1993
1994         /* ok, assign the rest of it as well.
1995          * lock against receive_SyncParam() */
1996         spin_lock(&mdev->peer_seq_lock);
1997         mdev->sync_conf = sc;
1998
1999         if (!rsr) {
2000                 crypto_free_hash(mdev->csums_tfm);
2001                 mdev->csums_tfm = csums_tfm;
2002                 csums_tfm = NULL;
2003         }
2004
2005         if (!ovr) {
2006                 crypto_free_hash(mdev->verify_tfm);
2007                 mdev->verify_tfm = verify_tfm;
2008                 verify_tfm = NULL;
2009         }
2010
2011         if (fifo_size != mdev->rs_plan_s.size) {
2012                 kfree(mdev->rs_plan_s.values);
2013                 mdev->rs_plan_s.values = rs_plan_s;
2014                 mdev->rs_plan_s.size   = fifo_size;
2015                 mdev->rs_planed = 0;
2016                 rs_plan_s = NULL;
2017         }
2018
2019         spin_unlock(&mdev->peer_seq_lock);
2020
2021         if (get_ldev(mdev)) {
2022                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
2023                 drbd_al_shrink(mdev);
2024                 err = drbd_check_al_size(mdev);
2025                 lc_unlock(mdev->act_log);
2026                 wake_up(&mdev->al_wait);
2027
2028                 put_ldev(mdev);
2029                 drbd_md_sync(mdev);
2030
2031                 if (err) {
2032                         retcode = ERR_NOMEM;
2033                         goto fail;
2034                 }
2035         }
2036
2037         if (mdev->state.conn >= C_CONNECTED)
2038                 drbd_send_sync_param(mdev, &sc);
2039
2040         if (!cpumask_equal(mdev->tconn->cpu_mask, new_cpu_mask)) {
2041                 cpumask_copy(mdev->tconn->cpu_mask, new_cpu_mask);
2042                 drbd_calc_cpu_mask(mdev->tconn);
2043                 mdev->tconn->receiver.reset_cpu_mask = 1;
2044                 mdev->tconn->asender.reset_cpu_mask = 1;
2045                 mdev->tconn->worker.reset_cpu_mask = 1;
2046         }
2047
2048         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
2049 fail:
2050         kfree(rs_plan_s);
2051         free_cpumask_var(new_cpu_mask);
2052         crypto_free_hash(csums_tfm);
2053         crypto_free_hash(verify_tfm);
2054
2055         drbd_adm_finish(info, retcode);
2056         return 0;
2057 }
2058
2059 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2060 {
2061         struct drbd_conf *mdev;
2062         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2063
2064         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2065         if (!adm_ctx.reply_skb)
2066                 return retcode;
2067         if (retcode != NO_ERROR)
2068                 goto out;
2069
2070         mdev = adm_ctx.mdev;
2071
2072         /* If there is still bitmap IO pending, probably because of a previous
2073          * resync just being finished, wait for it before requesting a new resync. */
2074         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2075
2076         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2077
2078         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2079                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2080
2081         while (retcode == SS_NEED_CONNECTION) {
2082                 spin_lock_irq(&mdev->tconn->req_lock);
2083                 if (mdev->state.conn < C_CONNECTED)
2084                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2085                 spin_unlock_irq(&mdev->tconn->req_lock);
2086
2087                 if (retcode != SS_NEED_CONNECTION)
2088                         break;
2089
2090                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2091         }
2092
2093 out:
2094         drbd_adm_finish(info, retcode);
2095         return 0;
2096 }
2097
2098 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2099 {
2100         int rv;
2101
2102         rv = drbd_bmio_set_n_write(mdev);
2103         drbd_suspend_al(mdev);
2104         return rv;
2105 }
2106
2107 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2108                 union drbd_state mask, union drbd_state val)
2109 {
2110         enum drbd_ret_code retcode;
2111
2112         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2113         if (!adm_ctx.reply_skb)
2114                 return retcode;
2115         if (retcode != NO_ERROR)
2116                 goto out;
2117
2118         retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2119 out:
2120         drbd_adm_finish(info, retcode);
2121         return 0;
2122 }
2123
2124 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2125 {
2126         return drbd_adm_simple_request_state(skb, info, NS(conn, C_STARTING_SYNC_S));
2127 }
2128
2129 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2130 {
2131         enum drbd_ret_code retcode;
2132
2133         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2134         if (!adm_ctx.reply_skb)
2135                 return retcode;
2136         if (retcode != NO_ERROR)
2137                 goto out;
2138
2139         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2140                 retcode = ERR_PAUSE_IS_SET;
2141 out:
2142         drbd_adm_finish(info, retcode);
2143         return 0;
2144 }
2145
2146 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2147 {
2148         union drbd_state s;
2149         enum drbd_ret_code retcode;
2150
2151         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2152         if (!adm_ctx.reply_skb)
2153                 return retcode;
2154         if (retcode != NO_ERROR)
2155                 goto out;
2156
2157         if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2158                 s = adm_ctx.mdev->state;
2159                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2160                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2161                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2162                 } else {
2163                         retcode = ERR_PAUSE_IS_CLEAR;
2164                 }
2165         }
2166
2167 out:
2168         drbd_adm_finish(info, retcode);
2169         return 0;
2170 }
2171
2172 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2173 {
2174         return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2175 }
2176
2177 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2178 {
2179         struct drbd_conf *mdev;
2180         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2181
2182         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2183         if (!adm_ctx.reply_skb)
2184                 return retcode;
2185         if (retcode != NO_ERROR)
2186                 goto out;
2187
2188         mdev = adm_ctx.mdev;
2189         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2190                 drbd_uuid_new_current(mdev);
2191                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2192         }
2193         drbd_suspend_io(mdev);
2194         retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2195         if (retcode == SS_SUCCESS) {
2196                 if (mdev->state.conn < C_CONNECTED)
2197                         tl_clear(mdev->tconn);
2198                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2199                         tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2200         }
2201         drbd_resume_io(mdev);
2202
2203 out:
2204         drbd_adm_finish(info, retcode);
2205         return 0;
2206 }
2207
2208 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2209 {
2210         return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2211 }
2212
2213 int nla_put_drbd_cfg_context(struct sk_buff *skb, const char *conn_name, unsigned vnr)
2214 {
2215         struct nlattr *nla;
2216         nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2217         if (!nla)
2218                 goto nla_put_failure;
2219         if (vnr != VOLUME_UNSPECIFIED)
2220                 NLA_PUT_U32(skb, T_ctx_volume, vnr);
2221         NLA_PUT_STRING(skb, T_ctx_conn_name, conn_name);
2222         nla_nest_end(skb, nla);
2223         return 0;
2224
2225 nla_put_failure:
2226         if (nla)
2227                 nla_nest_cancel(skb, nla);
2228         return -EMSGSIZE;
2229 }
2230
2231 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2232                 const struct sib_info *sib)
2233 {
2234         struct state_info *si = NULL; /* for sizeof(si->member); */
2235         struct nlattr *nla;
2236         int got_ldev;
2237         int got_net;
2238         int err = 0;
2239         int exclude_sensitive;
2240
2241         /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2242          * to.  So we better exclude_sensitive information.
2243          *
2244          * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2245          * in the context of the requesting user process. Exclude sensitive
2246          * information, unless current has superuser.
2247          *
2248          * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2249          * relies on the current implementation of netlink_dump(), which
2250          * executes the dump callback successively from netlink_recvmsg(),
2251          * always in the context of the receiving process */
2252         exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2253
2254         got_ldev = get_ldev(mdev);
2255         got_net = get_net_conf(mdev->tconn);
2256
2257         /* We need to add connection name and volume number information still.
2258          * Minor number is in drbd_genlmsghdr. */
2259         if (nla_put_drbd_cfg_context(skb, mdev->tconn->name, mdev->vnr))
2260                 goto nla_put_failure;
2261
2262         if (got_ldev)
2263                 if (disk_conf_to_skb(skb, &mdev->ldev->dc, exclude_sensitive))
2264                         goto nla_put_failure;
2265         if (got_net)
2266                 if (net_conf_to_skb(skb, mdev->tconn->net_conf, exclude_sensitive))
2267                         goto nla_put_failure;
2268
2269         if (syncer_conf_to_skb(skb, &mdev->sync_conf, exclude_sensitive))
2270                         goto nla_put_failure;
2271
2272         nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2273         if (!nla)
2274                 goto nla_put_failure;
2275         NLA_PUT_U32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY);
2276         NLA_PUT_U32(skb, T_current_state, mdev->state.i);
2277         NLA_PUT_U64(skb, T_ed_uuid, mdev->ed_uuid);
2278         NLA_PUT_U64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev));
2279
2280         if (got_ldev) {
2281                 NLA_PUT_U32(skb, T_disk_flags, mdev->ldev->md.flags);
2282                 NLA_PUT(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2283                 NLA_PUT_U64(skb, T_bits_total, drbd_bm_bits(mdev));
2284                 NLA_PUT_U64(skb, T_bits_oos, drbd_bm_total_weight(mdev));
2285                 if (C_SYNC_SOURCE <= mdev->state.conn &&
2286                     C_PAUSED_SYNC_T >= mdev->state.conn) {
2287                         NLA_PUT_U64(skb, T_bits_rs_total, mdev->rs_total);
2288                         NLA_PUT_U64(skb, T_bits_rs_failed, mdev->rs_failed);
2289                 }
2290         }
2291
2292         if (sib) {
2293                 switch(sib->sib_reason) {
2294                 case SIB_SYNC_PROGRESS:
2295                 case SIB_GET_STATUS_REPLY:
2296                         break;
2297                 case SIB_STATE_CHANGE:
2298                         NLA_PUT_U32(skb, T_prev_state, sib->os.i);
2299                         NLA_PUT_U32(skb, T_new_state, sib->ns.i);
2300                         break;
2301                 case SIB_HELPER_POST:
2302                         NLA_PUT_U32(skb,
2303                                 T_helper_exit_code, sib->helper_exit_code);
2304                         /* fall through */
2305                 case SIB_HELPER_PRE:
2306                         NLA_PUT_STRING(skb, T_helper, sib->helper_name);
2307                         break;
2308                 }
2309         }
2310         nla_nest_end(skb, nla);
2311
2312         if (0)
2313 nla_put_failure:
2314                 err = -EMSGSIZE;
2315         if (got_ldev)
2316                 put_ldev(mdev);
2317         if (got_net)
2318                 put_net_conf(mdev->tconn);
2319         return err;
2320 }
2321
2322 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2323 {
2324         enum drbd_ret_code retcode;
2325         int err;
2326
2327         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2328         if (!adm_ctx.reply_skb)
2329                 return retcode;
2330         if (retcode != NO_ERROR)
2331                 goto out;
2332
2333         err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2334         if (err) {
2335                 nlmsg_free(adm_ctx.reply_skb);
2336                 return err;
2337         }
2338 out:
2339         drbd_adm_finish(info, retcode);
2340         return 0;
2341 }
2342
2343 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2344 {
2345         struct drbd_conf *mdev;
2346         struct drbd_genlmsghdr *dh;
2347         struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2348         struct drbd_tconn *tconn = NULL;
2349         struct drbd_tconn *tmp;
2350         unsigned volume = cb->args[1];
2351
2352         /* Open coded, deferred, iteration:
2353          * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2354          *      idr_for_each_entry(&tconn->volumes, mdev, i) {
2355          *        ...
2356          *      }
2357          * }
2358          * where tconn is cb->args[0];
2359          * and i is cb->args[1];
2360          *
2361          * This may miss entries inserted after this dump started,
2362          * or entries deleted before they are reached.
2363          *
2364          * We need to make sure the mdev won't disappear while
2365          * we are looking at it, and revalidate our iterators
2366          * on each iteration.
2367          */
2368
2369         /* synchronize with drbd_new_tconn/drbd_free_tconn */
2370         mutex_lock(&drbd_cfg_mutex);
2371         /* synchronize with drbd_delete_device */
2372         rcu_read_lock();
2373 next_tconn:
2374         /* revalidate iterator position */
2375         list_for_each_entry(tmp, &drbd_tconns, all_tconn) {
2376                 if (pos == NULL) {
2377                         /* first iteration */
2378                         pos = tmp;
2379                         tconn = pos;
2380                         break;
2381                 }
2382                 if (tmp == pos) {
2383                         tconn = pos;
2384                         break;
2385                 }
2386         }
2387         if (tconn) {
2388                 mdev = idr_get_next(&tconn->volumes, &volume);
2389                 if (!mdev) {
2390                         /* No more volumes to dump on this tconn.
2391                          * Advance tconn iterator. */
2392                         pos = list_entry(tconn->all_tconn.next,
2393                                         struct drbd_tconn, all_tconn);
2394                         /* But, did we dump any volume on this tconn yet? */
2395                         if (volume != 0) {
2396                                 tconn = NULL;
2397                                 volume = 0;
2398                                 goto next_tconn;
2399                         }
2400                 }
2401
2402                 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).pid,
2403                                 cb->nlh->nlmsg_seq, &drbd_genl_family,
2404                                 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2405                 if (!dh)
2406                         goto out;
2407
2408                 if (!mdev) {
2409                         /* this is a tconn without a single volume */
2410                         dh->minor = -1U;
2411                         dh->ret_code = NO_ERROR;
2412                         if (nla_put_drbd_cfg_context(skb, tconn->name, VOLUME_UNSPECIFIED))
2413                                 genlmsg_cancel(skb, dh);
2414                         else
2415                                 genlmsg_end(skb, dh);
2416                         goto out;
2417                 }
2418
2419                 D_ASSERT(mdev->vnr == volume);
2420                 D_ASSERT(mdev->tconn == tconn);
2421
2422                 dh->minor = mdev_to_minor(mdev);
2423                 dh->ret_code = NO_ERROR;
2424
2425                 if (nla_put_status_info(skb, mdev, NULL)) {
2426                         genlmsg_cancel(skb, dh);
2427                         goto out;
2428                 }
2429                 genlmsg_end(skb, dh);
2430         }
2431
2432 out:
2433         rcu_read_unlock();
2434         mutex_unlock(&drbd_cfg_mutex);
2435         /* where to start the next iteration */
2436         cb->args[0] = (long)pos;
2437         cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2438
2439         /* No more tconns/volumes/minors found results in an empty skb.
2440          * Which will terminate the dump. */
2441         return skb->len;
2442 }
2443
2444 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2445 {
2446         enum drbd_ret_code retcode;
2447         struct timeout_parms tp;
2448         int err;
2449
2450         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2451         if (!adm_ctx.reply_skb)
2452                 return retcode;
2453         if (retcode != NO_ERROR)
2454                 goto out;
2455
2456         tp.timeout_type =
2457                 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2458                 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2459                 UT_DEFAULT;
2460
2461         err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2462         if (err) {
2463                 nlmsg_free(adm_ctx.reply_skb);
2464                 return err;
2465         }
2466 out:
2467         drbd_adm_finish(info, retcode);
2468         return 0;
2469 }
2470
2471 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2472 {
2473         struct drbd_conf *mdev;
2474         enum drbd_ret_code retcode;
2475
2476         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2477         if (!adm_ctx.reply_skb)
2478                 return retcode;
2479         if (retcode != NO_ERROR)
2480                 goto out;
2481
2482         mdev = adm_ctx.mdev;
2483         if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2484                 /* resume from last known position, if possible */
2485                 struct start_ov_parms parms =
2486                         { .ov_start_sector = mdev->ov_start_sector };
2487                 int err = start_ov_parms_from_attrs(&parms, info->attrs);
2488                 if (err) {
2489                         retcode = ERR_MANDATORY_TAG;
2490                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2491                         goto out;
2492                 }
2493                 /* w_make_ov_request expects position to be aligned */
2494                 mdev->ov_start_sector = parms.ov_start_sector & ~BM_SECT_PER_BIT;
2495         }
2496         /* If there is still bitmap IO pending, e.g. previous resync or verify
2497          * just being finished, wait for it before requesting a new resync. */
2498         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2499         retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2500 out:
2501         drbd_adm_finish(info, retcode);
2502         return 0;
2503 }
2504
2505
2506 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
2507 {
2508         struct drbd_conf *mdev;
2509         enum drbd_ret_code retcode;
2510         int skip_initial_sync = 0;
2511         int err;
2512         struct new_c_uuid_parms args;
2513
2514         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2515         if (!adm_ctx.reply_skb)
2516                 return retcode;
2517         if (retcode != NO_ERROR)
2518                 goto out_nolock;
2519
2520         mdev = adm_ctx.mdev;
2521         memset(&args, 0, sizeof(args));
2522         if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
2523                 err = new_c_uuid_parms_from_attrs(&args, info->attrs);
2524                 if (err) {
2525                         retcode = ERR_MANDATORY_TAG;
2526                         drbd_msg_put_info(from_attrs_err_to_txt(err));
2527                         goto out_nolock;
2528                 }
2529         }
2530
2531         mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2532
2533         if (!get_ldev(mdev)) {
2534                 retcode = ERR_NO_DISK;
2535                 goto out;
2536         }
2537
2538         /* this is "skip initial sync", assume to be clean */
2539         if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2540             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2541                 dev_info(DEV, "Preparing to skip initial sync\n");
2542                 skip_initial_sync = 1;
2543         } else if (mdev->state.conn != C_STANDALONE) {
2544                 retcode = ERR_CONNECTED;
2545                 goto out_dec;
2546         }
2547
2548         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2549         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2550
2551         if (args.clear_bm) {
2552                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2553                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2554                 if (err) {
2555                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2556                         retcode = ERR_IO_MD_DISK;
2557                 }
2558                 if (skip_initial_sync) {
2559                         drbd_send_uuids_skip_initial_sync(mdev);
2560                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2561                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2562                         spin_lock_irq(&mdev->tconn->req_lock);
2563                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2564                                         CS_VERBOSE, NULL);
2565                         spin_unlock_irq(&mdev->tconn->req_lock);
2566                 }
2567         }
2568
2569         drbd_md_sync(mdev);
2570 out_dec:
2571         put_ldev(mdev);
2572 out:
2573         mutex_unlock(mdev->state_mutex);
2574 out_nolock:
2575         drbd_adm_finish(info, retcode);
2576         return 0;
2577 }
2578
2579 static enum drbd_ret_code
2580 drbd_check_conn_name(const char *name)
2581 {
2582         if (!name || !name[0]) {
2583                 drbd_msg_put_info("connection name missing");
2584                 return ERR_MANDATORY_TAG;
2585         }
2586         /* if we want to use these in sysfs/configfs/debugfs some day,
2587          * we must not allow slashes */
2588         if (strchr(name, '/')) {
2589                 drbd_msg_put_info("invalid connection name");
2590                 return ERR_INVALID_REQUEST;
2591         }
2592         return NO_ERROR;
2593 }
2594
2595 int drbd_adm_create_connection(struct sk_buff *skb, struct genl_info *info)
2596 {
2597         enum drbd_ret_code retcode;
2598
2599         retcode = drbd_adm_prepare(skb, info, 0);
2600         if (!adm_ctx.reply_skb)
2601                 return retcode;
2602         if (retcode != NO_ERROR)
2603                 goto out;
2604
2605         retcode = drbd_check_conn_name(adm_ctx.conn_name);
2606         if (retcode != NO_ERROR)
2607                 goto out;
2608
2609         if (adm_ctx.tconn) {
2610                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
2611                         retcode = ERR_INVALID_REQUEST;
2612                         drbd_msg_put_info("connection exists");
2613                 }
2614                 /* else: still NO_ERROR */
2615                 goto out;
2616         }
2617
2618         if (!drbd_new_tconn(adm_ctx.conn_name))
2619                 retcode = ERR_NOMEM;
2620 out:
2621         drbd_adm_finish(info, retcode);
2622         return 0;
2623 }
2624
2625 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
2626 {
2627         struct drbd_genlmsghdr *dh = info->userhdr;
2628         enum drbd_ret_code retcode;
2629
2630         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2631         if (!adm_ctx.reply_skb)
2632                 return retcode;
2633         if (retcode != NO_ERROR)
2634                 goto out;
2635
2636         /* FIXME drop minor_count parameter, limit to MINORMASK */
2637         if (dh->minor >= minor_count) {
2638                 drbd_msg_put_info("requested minor out of range");
2639                 retcode = ERR_INVALID_REQUEST;
2640                 goto out;
2641         }
2642         /* FIXME we need a define here */
2643         if (adm_ctx.volume >= 256) {
2644                 drbd_msg_put_info("requested volume id out of range");
2645                 retcode = ERR_INVALID_REQUEST;
2646                 goto out;
2647         }
2648
2649         /* drbd_adm_prepare made sure already
2650          * that mdev->tconn and mdev->vnr match the request. */
2651         if (adm_ctx.mdev) {
2652                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
2653                         retcode = ERR_MINOR_EXISTS;
2654                 /* else: still NO_ERROR */
2655                 goto out;
2656         }
2657
2658         retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
2659 out:
2660         drbd_adm_finish(info, retcode);
2661         return 0;
2662 }
2663
2664 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
2665 {
2666         if (mdev->state.disk == D_DISKLESS &&
2667             /* no need to be mdev->state.conn == C_STANDALONE &&
2668              * we may want to delete a minor from a live replication group.
2669              */
2670             mdev->state.role == R_SECONDARY) {
2671                 drbd_delete_device(mdev_to_minor(mdev));
2672                 return NO_ERROR;
2673         } else
2674                 return ERR_MINOR_CONFIGURED;
2675 }
2676
2677 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
2678 {
2679         enum drbd_ret_code retcode;
2680
2681         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2682         if (!adm_ctx.reply_skb)
2683                 return retcode;
2684         if (retcode != NO_ERROR)
2685                 goto out;
2686
2687         mutex_lock(&drbd_cfg_mutex);
2688         retcode = adm_delete_minor(adm_ctx.mdev);
2689         mutex_unlock(&drbd_cfg_mutex);
2690         /* if this was the last volume of this connection,
2691          * this will terminate all threads */
2692         if (retcode == NO_ERROR)
2693                 conn_reconfig_done(adm_ctx.tconn);
2694 out:
2695         drbd_adm_finish(info, retcode);
2696         return 0;
2697 }
2698
2699 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
2700 {
2701         enum drbd_ret_code retcode;
2702         enum drbd_state_rv rv;
2703         struct drbd_conf *mdev;
2704         unsigned i;
2705
2706         retcode = drbd_adm_prepare(skb, info, 0);
2707         if (!adm_ctx.reply_skb)
2708                 return retcode;
2709         if (retcode != NO_ERROR)
2710                 goto out;
2711
2712         if (!adm_ctx.tconn) {
2713                 retcode = ERR_CONN_NOT_KNOWN;
2714                 goto out;
2715         }
2716
2717         mutex_lock(&drbd_cfg_mutex);
2718         /* demote */
2719         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
2720                 retcode = drbd_set_role(mdev, R_SECONDARY, 0);
2721                 if (retcode < SS_SUCCESS) {
2722                         drbd_msg_put_info("failed to demote");
2723                         goto out_unlock;
2724                 }
2725         }
2726
2727         /* disconnect */
2728         rv = conn_try_disconnect(adm_ctx.tconn, 0);
2729         if (rv < SS_SUCCESS) {
2730                 retcode = rv; /* enum type mismatch! */
2731                 drbd_msg_put_info("failed to disconnect");
2732                 goto out_unlock;
2733         }
2734
2735         /* detach */
2736         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
2737                 rv = adm_detach(mdev);
2738                 if (rv < SS_SUCCESS) {
2739                         retcode = rv; /* enum type mismatch! */
2740                         drbd_msg_put_info("failed to detach");
2741                         goto out_unlock;
2742                 }
2743         }
2744
2745         /* delete volumes */
2746         idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
2747                 retcode = adm_delete_minor(mdev);
2748                 if (retcode != NO_ERROR) {
2749                         /* "can not happen" */
2750                         drbd_msg_put_info("failed to delete volume");
2751                         goto out_unlock;
2752                 }
2753         }
2754
2755         /* stop all threads */
2756         conn_reconfig_done(adm_ctx.tconn);
2757
2758         /* delete connection */
2759         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
2760                 drbd_free_tconn(adm_ctx.tconn);
2761                 retcode = NO_ERROR;
2762         } else {
2763                 /* "can not happen" */
2764                 retcode = ERR_CONN_IN_USE;
2765                 drbd_msg_put_info("failed to delete connection");
2766                 goto out_unlock;
2767         }
2768 out_unlock:
2769         mutex_unlock(&drbd_cfg_mutex);
2770 out:
2771         drbd_adm_finish(info, retcode);
2772         return 0;
2773 }
2774
2775 int drbd_adm_delete_connection(struct sk_buff *skb, struct genl_info *info)
2776 {
2777         enum drbd_ret_code retcode;
2778
2779         retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2780         if (!adm_ctx.reply_skb)
2781                 return retcode;
2782         if (retcode != NO_ERROR)
2783                 goto out;
2784
2785         mutex_lock(&drbd_cfg_mutex);
2786         if (conn_lowest_minor(adm_ctx.tconn) < 0) {
2787                 drbd_free_tconn(adm_ctx.tconn);
2788                 retcode = NO_ERROR;
2789         } else {
2790                 retcode = ERR_CONN_IN_USE;
2791         }
2792         mutex_unlock(&drbd_cfg_mutex);
2793
2794 out:
2795         drbd_adm_finish(info, retcode);
2796         return 0;
2797 }
2798
2799 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
2800 {
2801         static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
2802         struct sk_buff *msg;
2803         struct drbd_genlmsghdr *d_out;
2804         unsigned seq;
2805         int err = -ENOMEM;
2806
2807         seq = atomic_inc_return(&drbd_genl_seq);
2808         msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
2809         if (!msg)
2810                 goto failed;
2811
2812         err = -EMSGSIZE;
2813         d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
2814         if (!d_out) /* cannot happen, but anyways. */
2815                 goto nla_put_failure;
2816         d_out->minor = mdev_to_minor(mdev);
2817         d_out->ret_code = 0;
2818
2819         if (nla_put_status_info(msg, mdev, sib))
2820                 goto nla_put_failure;
2821         genlmsg_end(msg, d_out);
2822         err = drbd_genl_multicast_events(msg, 0);
2823         /* msg has been consumed or freed in netlink_broadcast() */
2824         if (err && err != -ESRCH)
2825                 goto failed;
2826
2827         return;
2828
2829 nla_put_failure:
2830         nlmsg_free(msg);
2831 failed:
2832         dev_err(DEV, "Error %d while broadcasting event. "
2833                         "Event seq:%u sib_reason:%u\n",
2834                         err, seq, sib->sib_reason);
2835 }