drivers/block/drbd/drbd_nl.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3    drbd_nl.c
   4
   5    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
   6
   7    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
   8    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
   9    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
  10
  11
  12  */
  13
  14 #define pr_fmt(fmt)     KBUILD_MODNAME ": " fmt
  15
  16 #include <linux/module.h>
  17 #include <linux/drbd.h>
  18 #include <linux/in.h>
  19 #include <linux/fs.h>
  20 #include <linux/file.h>
  21 #include <linux/slab.h>
  22 #include <linux/blkpg.h>
  23 #include <linux/cpumask.h>
  24 #include "drbd_int.h"
  25 #include "drbd_protocol.h"
  26 #include "drbd_req.h"
  27 #include "drbd_state_change.h"
  28 #include <asm/unaligned.h>
  29 #include <linux/drbd_limits.h>
  30 #include <linux/kthread.h>
  31
  32 #include <net/genetlink.h>
  33
  34 /* .doit */
  35 // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info);
  36 // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info);
  37
  38 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info);
  39 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info);
  40
  41 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info);
  42 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info);
  43 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info);
  44
  45 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info);
  46 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info);
  47 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info);
  48 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info);
  49 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info);
  50 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info);
  51 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info);
  52 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info);
  53 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info);
  54 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info);
  55 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info);
  56 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info);
  57 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info);
  58 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info);
  59 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info);
  60 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info);
  61 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info);
  62 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info);
  63 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info);
  64 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info);
  65 /* .dumpit */
  66 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb);
  67 int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb);
  68 int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb);
  69 int drbd_adm_dump_devices_done(struct netlink_callback *cb);
  70 int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb);
  71 int drbd_adm_dump_connections_done(struct netlink_callback *cb);
  72 int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb);
  73 int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb);
  74 int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb);
  75
  76 #include <linux/drbd_genl_api.h>
  77 #include "drbd_nla.h"
  78 #include <linux/genl_magic_func.h>
  79
  80 static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
  81 static atomic_t notify_genl_seq = ATOMIC_INIT(2); /* two. */
  82
  83 DEFINE_MUTEX(notification_mutex);
  84
  85 /* used blkdev_get_by_path, to claim our meta data device(s) */
  86 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
  87
  88 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
  89 {
  90         genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
  91         if (genlmsg_reply(skb, info))
  92                 pr_err("error sending genl reply\n");
  93 }
  94
  95 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
  96  * reason it could fail was no space in skb, and there are 4k available. */
  97 static int drbd_msg_put_info(struct sk_buff *skb, const char *info)
  98 {
  99         struct nlattr *nla;
 100         int err = -EMSGSIZE;
 101
 102         if (!info || !info[0])
 103                 return 0;
 104
 105         nla = nla_nest_start_noflag(skb, DRBD_NLA_CFG_REPLY);
 106         if (!nla)
 107                 return err;
 108
 109         err = nla_put_string(skb, T_info_text, info);
 110         if (err) {
 111                 nla_nest_cancel(skb, nla);
 112                 return err;
 113         } else
 114                 nla_nest_end(skb, nla);
 115         return 0;
 116 }
 117
 118 __printf(2, 3)
 119 static int drbd_msg_sprintf_info(struct sk_buff *skb, const char *fmt, ...)
 120 {
 121         va_list args;
 122         struct nlattr *nla, *txt;
 123         int err = -EMSGSIZE;
 124         int len;
 125
 126         nla = nla_nest_start_noflag(skb, DRBD_NLA_CFG_REPLY);
 127         if (!nla)
 128                 return err;
 129
 130         txt = nla_reserve(skb, T_info_text, 256);
 131         if (!txt) {
 132                 nla_nest_cancel(skb, nla);
 133                 return err;
 134         }
 135         va_start(args, fmt);
 136         len = vscnprintf(nla_data(txt), 256, fmt, args);
 137         va_end(args);
 138
 139         /* maybe: retry with larger reserve, if truncated */
 140         txt->nla_len = nla_attr_size(len+1);
 141         nlmsg_trim(skb, (char*)txt + NLA_ALIGN(txt->nla_len));
 142         nla_nest_end(skb, nla);
 143
 144         return 0;
 145 }
 146
 147 /* This would be a good candidate for a "pre_doit" hook,
 148  * and per-family private info->pointers.
 149  * But we need to stay compatible with older kernels.
 150  * If it returns successfully, adm_ctx members are valid.
 151  *
 152  * At this point, we still rely on the global genl_lock().
 153  * If we want to avoid that, and allow "genl_family.parallel_ops", we may need
 154  * to add additional synchronization against object destruction/modification.
 155  */
 156 #define DRBD_ADM_NEED_MINOR     1
 157 #define DRBD_ADM_NEED_RESOURCE  2
 158 #define DRBD_ADM_NEED_CONNECTION 4
 159 static int drbd_adm_prepare(struct drbd_config_context *adm_ctx,
 160         struct sk_buff *skb, struct genl_info *info, unsigned flags)
 161 {
 162         struct drbd_genlmsghdr *d_in = info->userhdr;
 163         const u8 cmd = info->genlhdr->cmd;
 164         int err;
 165
 166         memset(adm_ctx, 0, sizeof(*adm_ctx));
 167
 168         /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
 169         if (cmd != DRBD_ADM_GET_STATUS && !capable(CAP_NET_ADMIN))
 170                return -EPERM;
 171
 172         adm_ctx->reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
 173         if (!adm_ctx->reply_skb) {
 174                 err = -ENOMEM;
 175                 goto fail;
 176         }
 177
 178         adm_ctx->reply_dh = genlmsg_put_reply(adm_ctx->reply_skb,
 179                                         info, &drbd_genl_family, 0, cmd);
 180         /* put of a few bytes into a fresh skb of >= 4k will always succeed.
 181          * but anyways */
 182         if (!adm_ctx->reply_dh) {
 183                 err = -ENOMEM;
 184                 goto fail;
 185         }
 186
 187         adm_ctx->reply_dh->minor = d_in->minor;
 188         adm_ctx->reply_dh->ret_code = NO_ERROR;
 189
 190         adm_ctx->volume = VOLUME_UNSPECIFIED;
 191         if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
 192                 struct nlattr *nla;
 193                 /* parse and validate only */
 194                 err = drbd_cfg_context_from_attrs(NULL, info);
 195                 if (err)
 196                         goto fail;
 197
 198                 /* It was present, and valid,
 199                  * copy it over to the reply skb. */
 200                 err = nla_put_nohdr(adm_ctx->reply_skb,
 201                                 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
 202                                 info->attrs[DRBD_NLA_CFG_CONTEXT]);
 203                 if (err)
 204                         goto fail;
 205
 206                 /* and assign stuff to the adm_ctx */
 207                 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
 208                 if (nla)
 209                         adm_ctx->volume = nla_get_u32(nla);
 210                 nla = nested_attr_tb[__nla_type(T_ctx_resource_name)];
 211                 if (nla)
 212                         adm_ctx->resource_name = nla_data(nla);
 213                 adm_ctx->my_addr = nested_attr_tb[__nla_type(T_ctx_my_addr)];
 214                 adm_ctx->peer_addr = nested_attr_tb[__nla_type(T_ctx_peer_addr)];
 215                 if ((adm_ctx->my_addr &&
 216                      nla_len(adm_ctx->my_addr) > sizeof(adm_ctx->connection->my_addr)) ||
 217                     (adm_ctx->peer_addr &&
 218                      nla_len(adm_ctx->peer_addr) > sizeof(adm_ctx->connection->peer_addr))) {
 219                         err = -EINVAL;
 220                         goto fail;
 221                 }
 222         }
 223
 224         adm_ctx->minor = d_in->minor;
 225         adm_ctx->device = minor_to_device(d_in->minor);
 226
 227         /* We are protected by the global genl_lock().
 228          * But we may explicitly drop it/retake it in drbd_adm_set_role(),
 229          * so make sure this object stays around. */
 230         if (adm_ctx->device)
 231                 kref_get(&adm_ctx->device->kref);
 232
 233         if (adm_ctx->resource_name) {
 234                 adm_ctx->resource = drbd_find_resource(adm_ctx->resource_name);
 235         }
 236
 237         if (!adm_ctx->device && (flags & DRBD_ADM_NEED_MINOR)) {
 238                 drbd_msg_put_info(adm_ctx->reply_skb, "unknown minor");
 239                 return ERR_MINOR_INVALID;
 240         }
 241         if (!adm_ctx->resource && (flags & DRBD_ADM_NEED_RESOURCE)) {
 242                 drbd_msg_put_info(adm_ctx->reply_skb, "unknown resource");
 243                 if (adm_ctx->resource_name)
 244                         return ERR_RES_NOT_KNOWN;
 245                 return ERR_INVALID_REQUEST;
 246         }
 247
 248         if (flags & DRBD_ADM_NEED_CONNECTION) {
 249                 if (adm_ctx->resource) {
 250                         drbd_msg_put_info(adm_ctx->reply_skb, "no resource name expected");
 251                         return ERR_INVALID_REQUEST;
 252                 }
 253                 if (adm_ctx->device) {
 254                         drbd_msg_put_info(adm_ctx->reply_skb, "no minor number expected");
 255                         return ERR_INVALID_REQUEST;
 256                 }
 257                 if (adm_ctx->my_addr && adm_ctx->peer_addr)
 258                         adm_ctx->connection = conn_get_by_addrs(nla_data(adm_ctx->my_addr),
 259                                                           nla_len(adm_ctx->my_addr),
 260                                                           nla_data(adm_ctx->peer_addr),
 261                                                           nla_len(adm_ctx->peer_addr));
 262                 if (!adm_ctx->connection) {
 263                         drbd_msg_put_info(adm_ctx->reply_skb, "unknown connection");
 264                         return ERR_INVALID_REQUEST;
 265                 }
 266         }
 267
 268         /* some more paranoia, if the request was over-determined */
 269         if (adm_ctx->device && adm_ctx->resource &&
 270             adm_ctx->device->resource != adm_ctx->resource) {
 271                 pr_warn("request: minor=%u, resource=%s; but that minor belongs to resource %s\n",
 272                         adm_ctx->minor, adm_ctx->resource->name,
 273                         adm_ctx->device->resource->name);
 274                 drbd_msg_put_info(adm_ctx->reply_skb, "minor exists in different resource");
 275                 return ERR_INVALID_REQUEST;
 276         }
 277         if (adm_ctx->device &&
 278             adm_ctx->volume != VOLUME_UNSPECIFIED &&
 279             adm_ctx->volume != adm_ctx->device->vnr) {
 280                 pr_warn("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
 281                         adm_ctx->minor, adm_ctx->volume,
 282                         adm_ctx->device->vnr, adm_ctx->device->resource->name);
 283                 drbd_msg_put_info(adm_ctx->reply_skb, "minor exists as different volume");
 284                 return ERR_INVALID_REQUEST;
 285         }
 286
 287         /* still, provide adm_ctx->resource always, if possible. */
 288         if (!adm_ctx->resource) {
 289                 adm_ctx->resource = adm_ctx->device ? adm_ctx->device->resource
 290                         : adm_ctx->connection ? adm_ctx->connection->resource : NULL;
 291                 if (adm_ctx->resource)
 292                         kref_get(&adm_ctx->resource->kref);
 293         }
 294
 295         return NO_ERROR;
 296
 297 fail:
 298         nlmsg_free(adm_ctx->reply_skb);
 299         adm_ctx->reply_skb = NULL;
 300         return err;
 301 }
 302
 303 static int drbd_adm_finish(struct drbd_config_context *adm_ctx,
 304         struct genl_info *info, int retcode)
 305 {
 306         if (adm_ctx->device) {
 307                 kref_put(&adm_ctx->device->kref, drbd_destroy_device);
 308                 adm_ctx->device = NULL;
 309         }
 310         if (adm_ctx->connection) {
 311                 kref_put(&adm_ctx->connection->kref, &drbd_destroy_connection);
 312                 adm_ctx->connection = NULL;
 313         }
 314         if (adm_ctx->resource) {
 315                 kref_put(&adm_ctx->resource->kref, drbd_destroy_resource);
 316                 adm_ctx->resource = NULL;
 317         }
 318
 319         if (!adm_ctx->reply_skb)
 320                 return -ENOMEM;
 321
 322         adm_ctx->reply_dh->ret_code = retcode;
 323         drbd_adm_send_reply(adm_ctx->reply_skb, info);
 324         return 0;
 325 }
 326
 327 static void setup_khelper_env(struct drbd_connection *connection, char **envp)
 328 {
 329         char *afs;
 330
 331         /* FIXME: A future version will not allow this case. */
 332         if (connection->my_addr_len == 0 || connection->peer_addr_len == 0)
 333                 return;
 334
 335         switch (((struct sockaddr *)&connection->peer_addr)->sa_family) {
 336         case AF_INET6:
 337                 afs = "ipv6";
 338                 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
 339                          &((struct sockaddr_in6 *)&connection->peer_addr)->sin6_addr);
 340                 break;
 341         case AF_INET:
 342                 afs = "ipv4";
 343                 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
 344                          &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
 345                 break;
 346         default:
 347                 afs = "ssocks";
 348                 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
 349                          &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
 350         }
 351         snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
 352 }
 353
 354 int drbd_khelper(struct drbd_device *device, char *cmd)
 355 {
 356         char *envp[] = { "HOME=/",
 357                         "TERM=linux",
 358                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 359                          (char[20]) { }, /* address family */
 360                          (char[60]) { }, /* address */
 361                         NULL };
 362         char mb[14];
 363         char *argv[] = {drbd_usermode_helper, cmd, mb, NULL };
 364         struct drbd_connection *connection = first_peer_device(device)->connection;
 365         struct sib_info sib;
 366         int ret;
 367
 368         if (current == connection->worker.task)
 369                 set_bit(CALLBACK_PENDING, &connection->flags);
 370
 371         snprintf(mb, 14, "minor-%d", device_to_minor(device));
 372         setup_khelper_env(connection, envp);
 373
 374         /* The helper may take some time.
 375          * write out any unsynced meta data changes now */
 376         drbd_md_sync(device);
 377
 378         drbd_info(device, "helper command: %s %s %s\n", drbd_usermode_helper, cmd, mb);
 379         sib.sib_reason = SIB_HELPER_PRE;
 380         sib.helper_name = cmd;
 381         drbd_bcast_event(device, &sib);
 382         notify_helper(NOTIFY_CALL, device, connection, cmd, 0);
 383         ret = call_usermodehelper(drbd_usermode_helper, argv, envp, UMH_WAIT_PROC);
 384         if (ret)
 385                 drbd_warn(device, "helper command: %s %s %s exit code %u (0x%x)\n",
 386                                 drbd_usermode_helper, cmd, mb,
 387                                 (ret >> 8) & 0xff, ret);
 388         else
 389                 drbd_info(device, "helper command: %s %s %s exit code %u (0x%x)\n",
 390                                 drbd_usermode_helper, cmd, mb,
 391                                 (ret >> 8) & 0xff, ret);
 392         sib.sib_reason = SIB_HELPER_POST;
 393         sib.helper_exit_code = ret;
 394         drbd_bcast_event(device, &sib);
 395         notify_helper(NOTIFY_RESPONSE, device, connection, cmd, ret);
 396
 397         if (current == connection->worker.task)
 398                 clear_bit(CALLBACK_PENDING, &connection->flags);
 399
 400         if (ret < 0) /* Ignore any ERRNOs we got. */
 401                 ret = 0;
 402
 403         return ret;
 404 }
 405
 406 enum drbd_peer_state conn_khelper(struct drbd_connection *connection, char *cmd)
 407 {
 408         char *envp[] = { "HOME=/",
 409                         "TERM=linux",
 410                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 411                          (char[20]) { }, /* address family */
 412                          (char[60]) { }, /* address */
 413                         NULL };
 414         char *resource_name = connection->resource->name;
 415         char *argv[] = {drbd_usermode_helper, cmd, resource_name, NULL };
 416         int ret;
 417
 418         setup_khelper_env(connection, envp);
 419         conn_md_sync(connection);
 420
 421         drbd_info(connection, "helper command: %s %s %s\n", drbd_usermode_helper, cmd, resource_name);
 422         /* TODO: conn_bcast_event() ?? */
 423         notify_helper(NOTIFY_CALL, NULL, connection, cmd, 0);
 424
 425         ret = call_usermodehelper(drbd_usermode_helper, argv, envp, UMH_WAIT_PROC);
 426         if (ret)
 427                 drbd_warn(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
 428                           drbd_usermode_helper, cmd, resource_name,
 429                           (ret >> 8) & 0xff, ret);
 430         else
 431                 drbd_info(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
 432                           drbd_usermode_helper, cmd, resource_name,
 433                           (ret >> 8) & 0xff, ret);
 434         /* TODO: conn_bcast_event() ?? */
 435         notify_helper(NOTIFY_RESPONSE, NULL, connection, cmd, ret);
 436
 437         if (ret < 0) /* Ignore any ERRNOs we got. */
 438                 ret = 0;
 439
 440         return ret;
 441 }
 442
 443 static enum drbd_fencing_p highest_fencing_policy(struct drbd_connection *connection)
 444 {
 445         enum drbd_fencing_p fp = FP_NOT_AVAIL;
 446         struct drbd_peer_device *peer_device;
 447         int vnr;
 448
 449         rcu_read_lock();
 450         idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
 451                 struct drbd_device *device = peer_device->device;
 452                 if (get_ldev_if_state(device, D_CONSISTENT)) {
 453                         struct disk_conf *disk_conf =
 454                                 rcu_dereference(peer_device->device->ldev->disk_conf);
 455                         fp = max_t(enum drbd_fencing_p, fp, disk_conf->fencing);
 456                         put_ldev(device);
 457                 }
 458         }
 459         rcu_read_unlock();
 460
 461         return fp;
 462 }
 463
 464 static bool resource_is_supended(struct drbd_resource *resource)
 465 {
 466         return resource->susp || resource->susp_fen || resource->susp_nod;
 467 }
 468
 469 bool conn_try_outdate_peer(struct drbd_connection *connection)
 470 {
 471         struct drbd_resource * const resource = connection->resource;
 472         unsigned int connect_cnt;
 473         union drbd_state mask = { };
 474         union drbd_state val = { };
 475         enum drbd_fencing_p fp;
 476         char *ex_to_string;
 477         int r;
 478
 479         spin_lock_irq(&resource->req_lock);
 480         if (connection->cstate >= C_WF_REPORT_PARAMS) {
 481                 drbd_err(connection, "Expected cstate < C_WF_REPORT_PARAMS\n");
 482                 spin_unlock_irq(&resource->req_lock);
 483                 return false;
 484         }
 485
 486         connect_cnt = connection->connect_cnt;
 487         spin_unlock_irq(&resource->req_lock);
 488
 489         fp = highest_fencing_policy(connection);
 490         switch (fp) {
 491         case FP_NOT_AVAIL:
 492                 drbd_warn(connection, "Not fencing peer, I'm not even Consistent myself.\n");
 493                 spin_lock_irq(&resource->req_lock);
 494                 if (connection->cstate < C_WF_REPORT_PARAMS) {
 495                         _conn_request_state(connection,
 496                                             (union drbd_state) { { .susp_fen = 1 } },
 497                                             (union drbd_state) { { .susp_fen = 0 } },
 498                                             CS_VERBOSE | CS_HARD | CS_DC_SUSP);
 499                         /* We are no longer suspended due to the fencing policy.
 500                          * We may still be suspended due to the on-no-data-accessible policy.
 501                          * If that was OND_IO_ERROR, fail pending requests. */
 502                         if (!resource_is_supended(resource))
 503                                 _tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
 504                 }
 505                 /* Else: in case we raced with a connection handshake,
 506                  * let the handshake figure out if we maybe can RESEND,
 507                  * and do not resume/fail pending requests here.
 508                  * Worst case is we stay suspended for now, which may be
 509                  * resolved by either re-establishing the replication link, or
 510                  * the next link failure, or eventually the administrator.  */
 511                 spin_unlock_irq(&resource->req_lock);
 512                 return false;
 513
 514         case FP_DONT_CARE:
 515                 return true;
 516         default: ;
 517         }
 518
 519         r = conn_khelper(connection, "fence-peer");
 520
 521         switch ((r>>8) & 0xff) {
 522         case P_INCONSISTENT: /* peer is inconsistent */
 523                 ex_to_string = "peer is inconsistent or worse";
 524                 mask.pdsk = D_MASK;
 525                 val.pdsk = D_INCONSISTENT;
 526                 break;
 527         case P_OUTDATED: /* peer got outdated, or was already outdated */
 528                 ex_to_string = "peer was fenced";
 529                 mask.pdsk = D_MASK;
 530                 val.pdsk = D_OUTDATED;
 531                 break;
 532         case P_DOWN: /* peer was down */
 533                 if (conn_highest_disk(connection) == D_UP_TO_DATE) {
 534                         /* we will(have) create(d) a new UUID anyways... */
 535                         ex_to_string = "peer is unreachable, assumed to be dead";
 536                         mask.pdsk = D_MASK;
 537                         val.pdsk = D_OUTDATED;
 538                 } else {
 539                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 540                 }
 541                 break;
 542         case P_PRIMARY: /* Peer is primary, voluntarily outdate myself.
 543                  * This is useful when an unconnected R_SECONDARY is asked to
 544                  * become R_PRIMARY, but finds the other peer being active. */
 545                 ex_to_string = "peer is active";
 546                 drbd_warn(connection, "Peer is primary, outdating myself.\n");
 547                 mask.disk = D_MASK;
 548                 val.disk = D_OUTDATED;
 549                 break;
 550         case P_FENCING:
 551                 /* THINK: do we need to handle this
 552                  * like case 4, or more like case 5? */
 553                 if (fp != FP_STONITH)
 554                         drbd_err(connection, "fence-peer() = 7 && fencing != Stonith !!!\n");
 555                 ex_to_string = "peer was stonithed";
 556                 mask.pdsk = D_MASK;
 557                 val.pdsk = D_OUTDATED;
 558                 break;
 559         default:
 560                 /* The script is broken ... */
 561                 drbd_err(connection, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 562                 return false; /* Eventually leave IO frozen */
 563         }
 564
 565         drbd_info(connection, "fence-peer helper returned %d (%s)\n",
 566                   (r>>8) & 0xff, ex_to_string);
 567
 568         /* Not using
 569            conn_request_state(connection, mask, val, CS_VERBOSE);
 570            here, because we might were able to re-establish the connection in the
 571            meantime. */
 572         spin_lock_irq(&resource->req_lock);
 573         if (connection->cstate < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &connection->flags)) {
 574                 if (connection->connect_cnt != connect_cnt)
 575                         /* In case the connection was established and droped
 576                            while the fence-peer handler was running, ignore it */
 577                         drbd_info(connection, "Ignoring fence-peer exit code\n");
 578                 else
 579                         _conn_request_state(connection, mask, val, CS_VERBOSE);
 580         }
 581         spin_unlock_irq(&resource->req_lock);
 582
 583         return conn_highest_pdsk(connection) <= D_OUTDATED;
 584 }
 585
 586 static int _try_outdate_peer_async(void *data)
 587 {
 588         struct drbd_connection *connection = (struct drbd_connection *)data;
 589
 590         conn_try_outdate_peer(connection);
 591
 592         kref_put(&connection->kref, drbd_destroy_connection);
 593         return 0;
 594 }
 595
 596 void conn_try_outdate_peer_async(struct drbd_connection *connection)
 597 {
 598         struct task_struct *opa;
 599
 600         kref_get(&connection->kref);
 601         /* We may have just sent a signal to this thread
 602          * to get it out of some blocking network function.
 603          * Clear signals; otherwise kthread_run(), which internally uses
 604          * wait_on_completion_killable(), will mistake our pending signal
 605          * for a new fatal signal and fail. */
 606         flush_signals(current);
 607         opa = kthread_run(_try_outdate_peer_async, connection, "drbd_async_h");
 608         if (IS_ERR(opa)) {
 609                 drbd_err(connection, "out of mem, failed to invoke fence-peer helper\n");
 610                 kref_put(&connection->kref, drbd_destroy_connection);
 611         }
 612 }
 613
 614 enum drbd_state_rv
 615 drbd_set_role(struct drbd_device *const device, enum drbd_role new_role, int force)
 616 {
 617         struct drbd_peer_device *const peer_device = first_peer_device(device);
 618         struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL;
 619         const int max_tries = 4;
 620         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
 621         struct net_conf *nc;
 622         int try = 0;
 623         int forced = 0;
 624         union drbd_state mask, val;
 625
 626         if (new_role == R_PRIMARY) {
 627                 struct drbd_connection *connection;
 628
 629                 /* Detect dead peers as soon as possible.  */
 630
 631                 rcu_read_lock();
 632                 for_each_connection(connection, device->resource)
 633                         request_ping(connection);
 634                 rcu_read_unlock();
 635         }
 636
 637         mutex_lock(device->state_mutex);
 638
 639         mask.i = 0; mask.role = R_MASK;
 640         val.i  = 0; val.role  = new_role;
 641
 642         while (try++ < max_tries) {
 643                 rv = _drbd_request_state_holding_state_mutex(device, mask, val, CS_WAIT_COMPLETE);
 644
 645                 /* in case we first succeeded to outdate,
 646                  * but now suddenly could establish a connection */
 647                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 648                         val.pdsk = 0;
 649                         mask.pdsk = 0;
 650                         continue;
 651                 }
 652
 653                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
 654                     (device->state.disk < D_UP_TO_DATE &&
 655                      device->state.disk >= D_INCONSISTENT)) {
 656                         mask.disk = D_MASK;
 657                         val.disk  = D_UP_TO_DATE;
 658                         forced = 1;
 659                         continue;
 660                 }
 661
 662                 if (rv == SS_NO_UP_TO_DATE_DISK &&
 663                     device->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 664                         D_ASSERT(device, device->state.pdsk == D_UNKNOWN);
 665
 666                         if (conn_try_outdate_peer(connection)) {
 667                                 val.disk = D_UP_TO_DATE;
 668                                 mask.disk = D_MASK;
 669                         }
 670                         continue;
 671                 }
 672
 673                 if (rv == SS_NOTHING_TO_DO)
 674                         goto out;
 675                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
 676                         if (!conn_try_outdate_peer(connection) && force) {
 677                                 drbd_warn(device, "Forced into split brain situation!\n");
 678                                 mask.pdsk = D_MASK;
 679                                 val.pdsk  = D_OUTDATED;
 680
 681                         }
 682                         continue;
 683                 }
 684                 if (rv == SS_TWO_PRIMARIES) {
 685                         /* Maybe the peer is detected as dead very soon...
 686                            retry at most once more in this case. */
 687                         if (try < max_tries) {
 688                                 int timeo;
 689                                 try = max_tries - 1;
 690                                 rcu_read_lock();
 691                                 nc = rcu_dereference(connection->net_conf);
 692                                 timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
 693                                 rcu_read_unlock();
 694                                 schedule_timeout_interruptible(timeo);
 695                         }
 696                         continue;
 697                 }
 698                 if (rv < SS_SUCCESS) {
 699                         rv = _drbd_request_state(device, mask, val,
 700                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 701                         if (rv < SS_SUCCESS)
 702                                 goto out;
 703                 }
 704                 break;
 705         }
 706
 707         if (rv < SS_SUCCESS)
 708                 goto out;
 709
 710         if (forced)
 711                 drbd_warn(device, "Forced to consider local data as UpToDate!\n");
 712
 713         /* Wait until nothing is on the fly :) */
 714         wait_event(device->misc_wait, atomic_read(&device->ap_pending_cnt) == 0);
 715
 716         /* FIXME also wait for all pending P_BARRIER_ACK? */
 717
 718         if (new_role == R_SECONDARY) {
 719                 if (get_ldev(device)) {
 720                         device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 721                         put_ldev(device);
 722                 }
 723         } else {
 724                 mutex_lock(&device->resource->conf_update);
 725                 nc = connection->net_conf;
 726                 if (nc)
 727                         nc->discard_my_data = 0; /* without copy; single bit op is atomic */
 728                 mutex_unlock(&device->resource->conf_update);
 729
 730                 if (get_ldev(device)) {
 731                         if (((device->state.conn < C_CONNECTED ||
 732                                device->state.pdsk <= D_FAILED)
 733                               && device->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 734                                 drbd_uuid_new_current(device);
 735
 736                         device->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 737                         put_ldev(device);
 738                 }
 739         }
 740
 741         /* writeout of activity log covered areas of the bitmap
 742          * to stable storage done in after state change already */
 743
 744         if (device->state.conn >= C_WF_REPORT_PARAMS) {
 745                 /* if this was forced, we should consider sync */
 746                 if (forced)
 747                         drbd_send_uuids(peer_device);
 748                 drbd_send_current_state(peer_device);
 749         }
 750
 751         drbd_md_sync(device);
 752         set_disk_ro(device->vdisk, new_role == R_SECONDARY);
 753         kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
 754 out:
 755         mutex_unlock(device->state_mutex);
 756         return rv;
 757 }
 758
 759 static const char *from_attrs_err_to_txt(int err)
 760 {
 761         return  err == -ENOMSG ? "required attribute missing" :
 762                 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
 763                 err == -EEXIST ? "can not change invariant setting" :
 764                 "invalid attribute value";
 765 }
 766
 767 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
 768 {
 769         struct drbd_config_context adm_ctx;
 770         struct set_role_parms parms;
 771         int err;
 772         enum drbd_ret_code retcode;
 773         enum drbd_state_rv rv;
 774
 775         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
 776         if (!adm_ctx.reply_skb)
 777                 return retcode;
 778         if (retcode != NO_ERROR)
 779                 goto out;
 780
 781         memset(&parms, 0, sizeof(parms));
 782         if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
 783                 err = set_role_parms_from_attrs(&parms, info);
 784                 if (err) {
 785                         retcode = ERR_MANDATORY_TAG;
 786                         drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
 787                         goto out;
 788                 }
 789         }
 790         genl_unlock();
 791         mutex_lock(&adm_ctx.resource->adm_mutex);
 792
 793         if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
 794                 rv = drbd_set_role(adm_ctx.device, R_PRIMARY, parms.assume_uptodate);
 795         else
 796                 rv = drbd_set_role(adm_ctx.device, R_SECONDARY, 0);
 797
 798         mutex_unlock(&adm_ctx.resource->adm_mutex);
 799         genl_lock();
 800         drbd_adm_finish(&adm_ctx, info, rv);
 801         return 0;
 802 out:
 803         drbd_adm_finish(&adm_ctx, info, retcode);
 804         return 0;
 805 }
 806
 807 /* Initializes the md.*_offset members, so we are able to find
 808  * the on disk meta data.
 809  *
 810  * We currently have two possible layouts:
 811  * external:
 812  *   |----------- md_size_sect ------------------|
 813  *   [ 4k superblock ][ activity log ][  Bitmap  ]
 814  *   | al_offset == 8 |
 815  *   | bm_offset = al_offset + X      |
 816  *  ==> bitmap sectors = md_size_sect - bm_offset
 817  *
 818  * internal:
 819  *            |----------- md_size_sect ------------------|
 820  * [data.....][  Bitmap  ][ activity log ][ 4k superblock ]
 821  *                        | al_offset < 0 |
 822  *            | bm_offset = al_offset - Y |
 823  *  ==> bitmap sectors = Y = al_offset - bm_offset
 824  *
 825  *  Activity log size used to be fixed 32kB,
 826  *  but is about to become configurable.
 827  */
 828 static void drbd_md_set_sector_offsets(struct drbd_device *device,
 829                                        struct drbd_backing_dev *bdev)
 830 {
 831         sector_t md_size_sect = 0;
 832         unsigned int al_size_sect = bdev->md.al_size_4k * 8;
 833
 834         bdev->md.md_offset = drbd_md_ss(bdev);
 835
 836         switch (bdev->md.meta_dev_idx) {
 837         default:
 838                 /* v07 style fixed size indexed meta data */
 839                 bdev->md.md_size_sect = MD_128MB_SECT;
 840                 bdev->md.al_offset = MD_4kB_SECT;
 841                 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
 842                 break;
 843         case DRBD_MD_INDEX_FLEX_EXT:
 844                 /* just occupy the full device; unit: sectors */
 845                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 846                 bdev->md.al_offset = MD_4kB_SECT;
 847                 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
 848                 break;
 849         case DRBD_MD_INDEX_INTERNAL:
 850         case DRBD_MD_INDEX_FLEX_INT:
 851                 /* al size is still fixed */
 852                 bdev->md.al_offset = -al_size_sect;
 853                 /* we need (slightly less than) ~ this much bitmap sectors: */
 854                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 855                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 856                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 857                 md_size_sect = ALIGN(md_size_sect, 8);
 858
 859                 /* plus the "drbd meta data super block",
 860                  * and the activity log; */
 861                 md_size_sect += MD_4kB_SECT + al_size_sect;
 862
 863                 bdev->md.md_size_sect = md_size_sect;
 864                 /* bitmap offset is adjusted by 'super' block size */
 865                 bdev->md.bm_offset   = -md_size_sect + MD_4kB_SECT;
 866                 break;
 867         }
 868 }
 869
 870 /* input size is expected to be in KB */
 871 char *ppsize(char *buf, unsigned long long size)
 872 {
 873         /* Needs 9 bytes at max including trailing NUL:
 874          * -1ULL ==> "16384 EB" */
 875         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 876         int base = 0;
 877         while (size >= 10000 && base < sizeof(units)-1) {
 878                 /* shift + round */
 879                 size = (size >> 10) + !!(size & (1<<9));
 880                 base++;
 881         }
 882         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
 883
 884         return buf;
 885 }
 886
 887 /* there is still a theoretical deadlock when called from receiver
 888  * on an D_INCONSISTENT R_PRIMARY:
 889  *  remote READ does inc_ap_bio, receiver would need to receive answer
 890  *  packet from remote to dec_ap_bio again.
 891  *  receiver receive_sizes(), comes here,
 892  *  waits for ap_bio_cnt == 0. -> deadlock.
 893  * but this cannot happen, actually, because:
 894  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 895  *  (not connected, or bad/no disk on peer):
 896  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 897  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 898  *  peer may not initiate a resize.
 899  */
 900 /* Note these are not to be confused with
 901  * drbd_adm_suspend_io/drbd_adm_resume_io,
 902  * which are (sub) state changes triggered by admin (drbdsetup),
 903  * and can be long lived.
 904  * This changes an device->flag, is triggered by drbd internals,
 905  * and should be short-lived. */
 906 /* It needs to be a counter, since multiple threads might
 907    independently suspend and resume IO. */
 908 void drbd_suspend_io(struct drbd_device *device)
 909 {
 910         atomic_inc(&device->suspend_cnt);
 911         if (drbd_suspended(device))
 912                 return;
 913         wait_event(device->misc_wait, !atomic_read(&device->ap_bio_cnt));
 914 }
 915
 916 void drbd_resume_io(struct drbd_device *device)
 917 {
 918         if (atomic_dec_and_test(&device->suspend_cnt))
 919                 wake_up(&device->misc_wait);
 920 }
 921
 922 /*
 923  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 924  * @device:     DRBD device.
 925  *
 926  * Returns 0 on success, negative return values indicate errors.
 927  * You should call drbd_md_sync() after calling this function.
 928  */
 929 enum determine_dev_size
 930 drbd_determine_dev_size(struct drbd_device *device, enum dds_flags flags, struct resize_parms *rs) __must_hold(local)
 931 {
 932         struct md_offsets_and_sizes {
 933                 u64 last_agreed_sect;
 934                 u64 md_offset;
 935                 s32 al_offset;
 936                 s32 bm_offset;
 937                 u32 md_size_sect;
 938
 939                 u32 al_stripes;
 940                 u32 al_stripe_size_4k;
 941         } prev;
 942         sector_t u_size, size;
 943         struct drbd_md *md = &device->ldev->md;
 944         void *buffer;
 945
 946         int md_moved, la_size_changed;
 947         enum determine_dev_size rv = DS_UNCHANGED;
 948
 949         /* We may change the on-disk offsets of our meta data below.  Lock out
 950          * anything that may cause meta data IO, to avoid acting on incomplete
 951          * layout changes or scribbling over meta data that is in the process
 952          * of being moved.
 953          *
 954          * Move is not exactly correct, btw, currently we have all our meta
 955          * data in core memory, to "move" it we just write it all out, there
 956          * are no reads. */
 957         drbd_suspend_io(device);
 958         buffer = drbd_md_get_buffer(device, __func__); /* Lock meta-data IO */
 959         if (!buffer) {
 960                 drbd_resume_io(device);
 961                 return DS_ERROR;
 962         }
 963
 964         /* remember current offset and sizes */
 965         prev.last_agreed_sect = md->la_size_sect;
 966         prev.md_offset = md->md_offset;
 967         prev.al_offset = md->al_offset;
 968         prev.bm_offset = md->bm_offset;
 969         prev.md_size_sect = md->md_size_sect;
 970         prev.al_stripes = md->al_stripes;
 971         prev.al_stripe_size_4k = md->al_stripe_size_4k;
 972
 973         if (rs) {
 974                 /* rs is non NULL if we should change the AL layout only */
 975                 md->al_stripes = rs->al_stripes;
 976                 md->al_stripe_size_4k = rs->al_stripe_size / 4;
 977                 md->al_size_4k = (u64)rs->al_stripes * rs->al_stripe_size / 4;
 978         }
 979
 980         drbd_md_set_sector_offsets(device, device->ldev);
 981
 982         rcu_read_lock();
 983         u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
 984         rcu_read_unlock();
 985         size = drbd_new_dev_size(device, device->ldev, u_size, flags & DDSF_FORCED);
 986
 987         if (size < prev.last_agreed_sect) {
 988                 if (rs && u_size == 0) {
 989                         /* Remove "rs &&" later. This check should always be active, but
 990                            right now the receiver expects the permissive behavior */
 991                         drbd_warn(device, "Implicit shrink not allowed. "
 992                                  "Use --size=%llus for explicit shrink.\n",
 993                                  (unsigned long long)size);
 994                         rv = DS_ERROR_SHRINK;
 995                 }
 996                 if (u_size > size)
 997                         rv = DS_ERROR_SPACE_MD;
 998                 if (rv != DS_UNCHANGED)
 999                         goto err_out;
1000         }
1001
1002         if (get_capacity(device->vdisk) != size ||
1003             drbd_bm_capacity(device) != size) {
1004                 int err;
1005                 err = drbd_bm_resize(device, size, !(flags & DDSF_NO_RESYNC));
1006                 if (unlikely(err)) {
1007                         /* currently there is only one error: ENOMEM! */
1008                         size = drbd_bm_capacity(device);
1009                         if (size == 0) {
1010                                 drbd_err(device, "OUT OF MEMORY! "
1011                                     "Could not allocate bitmap!\n");
1012                         } else {
1013                                 drbd_err(device, "BM resizing failed. "
1014                                     "Leaving size unchanged\n");
1015                         }
1016                         rv = DS_ERROR;
1017                 }
1018                 /* racy, see comments above. */
1019                 drbd_set_my_capacity(device, size);
1020                 md->la_size_sect = size;
1021         }
1022         if (rv <= DS_ERROR)
1023                 goto err_out;
1024
1025         la_size_changed = (prev.last_agreed_sect != md->la_size_sect);
1026
1027         md_moved = prev.md_offset    != md->md_offset
1028                 || prev.md_size_sect != md->md_size_sect;
1029
1030         if (la_size_changed || md_moved || rs) {
1031                 u32 prev_flags;
1032
1033                 /* We do some synchronous IO below, which may take some time.
1034                  * Clear the timer, to avoid scary "timer expired!" messages,
1035                  * "Superblock" is written out at least twice below, anyways. */
1036                 del_timer(&device->md_sync_timer);
1037
1038                 /* We won't change the "al-extents" setting, we just may need
1039                  * to move the on-disk location of the activity log ringbuffer.
1040                  * Lock for transaction is good enough, it may well be "dirty"
1041                  * or even "starving". */
1042                 wait_event(device->al_wait, lc_try_lock_for_transaction(device->act_log));
1043
1044                 /* mark current on-disk bitmap and activity log as unreliable */
1045                 prev_flags = md->flags;
1046                 md->flags |= MDF_FULL_SYNC | MDF_AL_DISABLED;
1047                 drbd_md_write(device, buffer);
1048
1049                 drbd_al_initialize(device, buffer);
1050
1051                 drbd_info(device, "Writing the whole bitmap, %s\n",
1052                          la_size_changed && md_moved ? "size changed and md moved" :
1053                          la_size_changed ? "size changed" : "md moved");
1054                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
1055                 drbd_bitmap_io(device, md_moved ? &drbd_bm_write_all : &drbd_bm_write,
1056                                "size changed", BM_LOCKED_MASK);
1057
1058                 /* on-disk bitmap and activity log is authoritative again
1059                  * (unless there was an IO error meanwhile...) */
1060                 md->flags = prev_flags;
1061                 drbd_md_write(device, buffer);
1062
1063                 if (rs)
1064                         drbd_info(device, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n",
1065                                   md->al_stripes, md->al_stripe_size_4k * 4);
1066         }
1067
1068         if (size > prev.last_agreed_sect)
1069                 rv = prev.last_agreed_sect ? DS_GREW : DS_GREW_FROM_ZERO;
1070         if (size < prev.last_agreed_sect)
1071                 rv = DS_SHRUNK;
1072
1073         if (0) {
1074         err_out:
1075                 /* restore previous offset and sizes */
1076                 md->la_size_sect = prev.last_agreed_sect;
1077                 md->md_offset = prev.md_offset;
1078                 md->al_offset = prev.al_offset;
1079                 md->bm_offset = prev.bm_offset;
1080                 md->md_size_sect = prev.md_size_sect;
1081                 md->al_stripes = prev.al_stripes;
1082                 md->al_stripe_size_4k = prev.al_stripe_size_4k;
1083                 md->al_size_4k = (u64)prev.al_stripes * prev.al_stripe_size_4k;
1084         }
1085         lc_unlock(device->act_log);
1086         wake_up(&device->al_wait);
1087         drbd_md_put_buffer(device);
1088         drbd_resume_io(device);
1089
1090         return rv;
1091 }
1092
1093 sector_t
1094 drbd_new_dev_size(struct drbd_device *device, struct drbd_backing_dev *bdev,
1095                   sector_t u_size, int assume_peer_has_space)
1096 {
1097         sector_t p_size = device->p_size;   /* partner's disk size. */
1098         sector_t la_size_sect = bdev->md.la_size_sect; /* last agreed size. */
1099         sector_t m_size; /* my size */
1100         sector_t size = 0;
1101
1102         m_size = drbd_get_max_capacity(bdev);
1103
1104         if (device->state.conn < C_CONNECTED && assume_peer_has_space) {
1105                 drbd_warn(device, "Resize while not connected was forced by the user!\n");
1106                 p_size = m_size;
1107         }
1108
1109         if (p_size && m_size) {
1110                 size = min_t(sector_t, p_size, m_size);
1111         } else {
1112                 if (la_size_sect) {
1113                         size = la_size_sect;
1114                         if (m_size && m_size < size)
1115                                 size = m_size;
1116                         if (p_size && p_size < size)
1117                                 size = p_size;
1118                 } else {
1119                         if (m_size)
1120                                 size = m_size;
1121                         if (p_size)
1122                                 size = p_size;
1123                 }
1124         }
1125
1126         if (size == 0)
1127                 drbd_err(device, "Both nodes diskless!\n");
1128
1129         if (u_size) {
1130                 if (u_size > size)
1131                         drbd_err(device, "Requested disk size is too big (%lu > %lu)\n",
1132                             (unsigned long)u_size>>1, (unsigned long)size>>1);
1133                 else
1134                         size = u_size;
1135         }
1136
1137         return size;
1138 }
1139
1140 /*
1141  * drbd_check_al_size() - Ensures that the AL is of the right size
1142  * @device:     DRBD device.
1143  *
1144  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
1145  * failed, and 0 on success. You should call drbd_md_sync() after you called
1146  * this function.
1147  */
1148 static int drbd_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1149 {
1150         struct lru_cache *n, *t;
1151         struct lc_element *e;
1152         unsigned int in_use;
1153         int i;
1154
1155         if (device->act_log &&
1156             device->act_log->nr_elements == dc->al_extents)
1157                 return 0;
1158
1159         in_use = 0;
1160         t = device->act_log;
1161         n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
1162                 dc->al_extents, sizeof(struct lc_element), 0);
1163
1164         if (n == NULL) {
1165                 drbd_err(device, "Cannot allocate act_log lru!\n");
1166                 return -ENOMEM;
1167         }
1168         spin_lock_irq(&device->al_lock);
1169         if (t) {
1170                 for (i = 0; i < t->nr_elements; i++) {
1171                         e = lc_element_by_index(t, i);
1172                         if (e->refcnt)
1173                                 drbd_err(device, "refcnt(%d)==%d\n",
1174                                     e->lc_number, e->refcnt);
1175                         in_use += e->refcnt;
1176                 }
1177         }
1178         if (!in_use)
1179                 device->act_log = n;
1180         spin_unlock_irq(&device->al_lock);
1181         if (in_use) {
1182                 drbd_err(device, "Activity log still in use!\n");
1183                 lc_destroy(n);
1184                 return -EBUSY;
1185         } else {
1186                 lc_destroy(t);
1187         }
1188         drbd_md_mark_dirty(device); /* we changed device->act_log->nr_elemens */
1189         return 0;
1190 }
1191
1192 static void blk_queue_discard_granularity(struct request_queue *q, unsigned int granularity)
1193 {
1194         q->limits.discard_granularity = granularity;
1195 }
1196
1197 static unsigned int drbd_max_discard_sectors(struct drbd_connection *connection)
1198 {
1199         /* when we introduced REQ_WRITE_SAME support, we also bumped
1200          * our maximum supported batch bio size used for discards. */
1201         if (connection->agreed_features & DRBD_FF_WSAME)
1202                 return DRBD_MAX_BBIO_SECTORS;
1203         /* before, with DRBD <= 8.4.6, we only allowed up to one AL_EXTENT_SIZE. */
1204         return AL_EXTENT_SIZE >> 9;
1205 }
1206
1207 static void decide_on_discard_support(struct drbd_device *device,
1208                 struct drbd_backing_dev *bdev)
1209 {
1210         struct drbd_connection *connection =
1211                 first_peer_device(device)->connection;
1212         struct request_queue *q = device->rq_queue;
1213         unsigned int max_discard_sectors;
1214
1215         if (bdev && !bdev_max_discard_sectors(bdev->backing_bdev))
1216                 goto not_supported;
1217
1218         if (connection->cstate >= C_CONNECTED &&
1219             !(connection->agreed_features & DRBD_FF_TRIM)) {
1220                 drbd_info(connection,
1221                         "peer DRBD too old, does not support TRIM: disabling discards\n");
1222                 goto not_supported;
1223         }
1224
1225         /*
1226          * We don't care for the granularity, really.
1227          *
1228          * Stacking limits below should fix it for the local device.  Whether or
1229          * not it is a suitable granularity on the remote device is not our
1230          * problem, really. If you care, you need to use devices with similar
1231          * topology on all peers.
1232          */
1233         blk_queue_discard_granularity(q, 512);
1234         max_discard_sectors = drbd_max_discard_sectors(connection);
1235         blk_queue_max_discard_sectors(q, max_discard_sectors);
1236         blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
1237         return;
1238
1239 not_supported:
1240         blk_queue_discard_granularity(q, 0);
1241         blk_queue_max_discard_sectors(q, 0);
1242 }
1243
1244 static void fixup_write_zeroes(struct drbd_device *device, struct request_queue *q)
1245 {
1246         /* Fixup max_write_zeroes_sectors after blk_stack_limits():
1247          * if we can handle "zeroes" efficiently on the protocol,
1248          * we want to do that, even if our backend does not announce
1249          * max_write_zeroes_sectors itself. */
1250         struct drbd_connection *connection = first_peer_device(device)->connection;
1251         /* If the peer announces WZEROES support, use it.  Otherwise, rather
1252          * send explicit zeroes than rely on some discard-zeroes-data magic. */
1253         if (connection->agreed_features & DRBD_FF_WZEROES)
1254                 q->limits.max_write_zeroes_sectors = DRBD_MAX_BBIO_SECTORS;
1255         else
1256                 q->limits.max_write_zeroes_sectors = 0;
1257 }
1258
1259 static void fixup_discard_support(struct drbd_device *device, struct request_queue *q)
1260 {
1261         unsigned int max_discard = device->rq_queue->limits.max_discard_sectors;
1262         unsigned int discard_granularity =
1263                 device->rq_queue->limits.discard_granularity >> SECTOR_SHIFT;
1264
1265         if (discard_granularity > max_discard) {
1266                 blk_queue_discard_granularity(q, 0);
1267                 blk_queue_max_discard_sectors(q, 0);
1268         }
1269 }
1270
1271 static void drbd_setup_queue_param(struct drbd_device *device, struct drbd_backing_dev *bdev,
1272                                    unsigned int max_bio_size, struct o_qlim *o)
1273 {
1274         struct request_queue * const q = device->rq_queue;
1275         unsigned int max_hw_sectors = max_bio_size >> 9;
1276         unsigned int max_segments = 0;
1277         struct request_queue *b = NULL;
1278         struct disk_conf *dc;
1279
1280         if (bdev) {
1281                 b = bdev->backing_bdev->bd_disk->queue;
1282
1283                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1284                 rcu_read_lock();
1285                 dc = rcu_dereference(device->ldev->disk_conf);
1286                 max_segments = dc->max_bio_bvecs;
1287                 rcu_read_unlock();
1288
1289                 blk_set_stacking_limits(&q->limits);
1290         }
1291
1292         blk_queue_max_hw_sectors(q, max_hw_sectors);
1293         /* This is the workaround for "bio would need to, but cannot, be split" */
1294         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1295         blk_queue_segment_boundary(q, PAGE_SIZE-1);
1296         decide_on_discard_support(device, bdev);
1297
1298         if (b) {
1299                 blk_stack_limits(&q->limits, &b->limits, 0);
1300                 disk_update_readahead(device->vdisk);
1301         }
1302         fixup_write_zeroes(device, q);
1303         fixup_discard_support(device, q);
1304 }
1305
1306 void drbd_reconsider_queue_parameters(struct drbd_device *device, struct drbd_backing_dev *bdev, struct o_qlim *o)
1307 {
1308         unsigned int now, new, local, peer;
1309
1310         now = queue_max_hw_sectors(device->rq_queue) << 9;
1311         local = device->local_max_bio_size; /* Eventually last known value, from volatile memory */
1312         peer = device->peer_max_bio_size; /* Eventually last known value, from meta data */
1313
1314         if (bdev) {
1315                 local = queue_max_hw_sectors(bdev->backing_bdev->bd_disk->queue) << 9;
1316                 device->local_max_bio_size = local;
1317         }
1318         local = min(local, DRBD_MAX_BIO_SIZE);
1319
1320         /* We may ignore peer limits if the peer is modern enough.
1321            Because new from 8.3.8 onwards the peer can use multiple
1322            BIOs for a single peer_request */
1323         if (device->state.conn >= C_WF_REPORT_PARAMS) {
1324                 if (first_peer_device(device)->connection->agreed_pro_version < 94)
1325                         peer = min(device->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1326                         /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1327                 else if (first_peer_device(device)->connection->agreed_pro_version == 94)
1328                         peer = DRBD_MAX_SIZE_H80_PACKET;
1329                 else if (first_peer_device(device)->connection->agreed_pro_version < 100)
1330                         peer = DRBD_MAX_BIO_SIZE_P95;  /* drbd 8.3.8 onwards, before 8.4.0 */
1331                 else
1332                         peer = DRBD_MAX_BIO_SIZE;
1333
1334                 /* We may later detach and re-attach on a disconnected Primary.
1335                  * Avoid this setting to jump back in that case.
1336                  * We want to store what we know the peer DRBD can handle,
1337                  * not what the peer IO backend can handle. */
1338                 if (peer > device->peer_max_bio_size)
1339                         device->peer_max_bio_size = peer;
1340         }
1341         new = min(local, peer);
1342
1343         if (device->state.role == R_PRIMARY && new < now)
1344                 drbd_err(device, "ASSERT FAILED new < now; (%u < %u)\n", new, now);
1345
1346         if (new != now)
1347                 drbd_info(device, "max BIO size = %u\n", new);
1348
1349         drbd_setup_queue_param(device, bdev, new, o);
1350 }
1351
1352 /* Starts the worker thread */
1353 static void conn_reconfig_start(struct drbd_connection *connection)
1354 {
1355         drbd_thread_start(&connection->worker);
1356         drbd_flush_workqueue(&connection->sender_work);
1357 }
1358
1359 /* if still unconfigured, stops worker again. */
1360 static void conn_reconfig_done(struct drbd_connection *connection)
1361 {
1362         bool stop_threads;
1363         spin_lock_irq(&connection->resource->req_lock);
1364         stop_threads = conn_all_vols_unconf(connection) &&
1365                 connection->cstate == C_STANDALONE;
1366         spin_unlock_irq(&connection->resource->req_lock);
1367         if (stop_threads) {
1368                 /* ack_receiver thread and ack_sender workqueue are implicitly
1369                  * stopped by receiver in conn_disconnect() */
1370                 drbd_thread_stop(&connection->receiver);
1371                 drbd_thread_stop(&connection->worker);
1372         }
1373 }
1374
1375 /* Make sure IO is suspended before calling this function(). */
1376 static void drbd_suspend_al(struct drbd_device *device)
1377 {
1378         int s = 0;
1379
1380         if (!lc_try_lock(device->act_log)) {
1381                 drbd_warn(device, "Failed to lock al in drbd_suspend_al()\n");
1382                 return;
1383         }
1384
1385         drbd_al_shrink(device);
1386         spin_lock_irq(&device->resource->req_lock);
1387         if (device->state.conn < C_CONNECTED)
1388                 s = !test_and_set_bit(AL_SUSPENDED, &device->flags);
1389         spin_unlock_irq(&device->resource->req_lock);
1390         lc_unlock(device->act_log);
1391
1392         if (s)
1393                 drbd_info(device, "Suspended AL updates\n");
1394 }
1395
1396
1397 static bool should_set_defaults(struct genl_info *info)
1398 {
1399         unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1400         return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1401 }
1402
1403 static unsigned int drbd_al_extents_max(struct drbd_backing_dev *bdev)
1404 {
1405         /* This is limited by 16 bit "slot" numbers,
1406          * and by available on-disk context storage.
1407          *
1408          * Also (u16)~0 is special (denotes a "free" extent).
1409          *
1410          * One transaction occupies one 4kB on-disk block,
1411          * we have n such blocks in the on disk ring buffer,
1412          * the "current" transaction may fail (n-1),
1413          * and there is 919 slot numbers context information per transaction.
1414          *
1415          * 72 transaction blocks amounts to more than 2**16 context slots,
1416          * so cap there first.
1417          */
1418         const unsigned int max_al_nr = DRBD_AL_EXTENTS_MAX;
1419         const unsigned int sufficient_on_disk =
1420                 (max_al_nr + AL_CONTEXT_PER_TRANSACTION -1)
1421                 /AL_CONTEXT_PER_TRANSACTION;
1422
1423         unsigned int al_size_4k = bdev->md.al_size_4k;
1424
1425         if (al_size_4k > sufficient_on_disk)
1426                 return max_al_nr;
1427
1428         return (al_size_4k - 1) * AL_CONTEXT_PER_TRANSACTION;
1429 }
1430
1431 static bool write_ordering_changed(struct disk_conf *a, struct disk_conf *b)
1432 {
1433         return  a->disk_barrier != b->disk_barrier ||
1434                 a->disk_flushes != b->disk_flushes ||
1435                 a->disk_drain != b->disk_drain;
1436 }
1437
1438 static void sanitize_disk_conf(struct drbd_device *device, struct disk_conf *disk_conf,
1439                                struct drbd_backing_dev *nbc)
1440 {
1441         struct block_device *bdev = nbc->backing_bdev;
1442
1443         if (disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1444                 disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1445         if (disk_conf->al_extents > drbd_al_extents_max(nbc))
1446                 disk_conf->al_extents = drbd_al_extents_max(nbc);
1447
1448         if (!bdev_max_discard_sectors(bdev)) {
1449                 if (disk_conf->rs_discard_granularity) {
1450                         disk_conf->rs_discard_granularity = 0; /* disable feature */
1451                         drbd_info(device, "rs_discard_granularity feature disabled\n");
1452                 }
1453         }
1454
1455         if (disk_conf->rs_discard_granularity) {
1456                 int orig_value = disk_conf->rs_discard_granularity;
1457                 sector_t discard_size = bdev_max_discard_sectors(bdev) << 9;
1458                 unsigned int discard_granularity = bdev_discard_granularity(bdev);
1459                 int remainder;
1460
1461                 if (discard_granularity > disk_conf->rs_discard_granularity)
1462                         disk_conf->rs_discard_granularity = discard_granularity;
1463
1464                 remainder = disk_conf->rs_discard_granularity %
1465                                 discard_granularity;
1466                 disk_conf->rs_discard_granularity += remainder;
1467
1468                 if (disk_conf->rs_discard_granularity > discard_size)
1469                         disk_conf->rs_discard_granularity = discard_size;
1470
1471                 if (disk_conf->rs_discard_granularity != orig_value)
1472                         drbd_info(device, "rs_discard_granularity changed to %d\n",
1473                                   disk_conf->rs_discard_granularity);
1474         }
1475 }
1476
1477 static int disk_opts_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1478 {
1479         int err = -EBUSY;
1480
1481         if (device->act_log &&
1482             device->act_log->nr_elements == dc->al_extents)
1483                 return 0;
1484
1485         drbd_suspend_io(device);
1486         /* If IO completion is currently blocked, we would likely wait
1487          * "forever" for the activity log to become unused. So we don't. */
1488         if (atomic_read(&device->ap_bio_cnt))
1489                 goto out;
1490
1491         wait_event(device->al_wait, lc_try_lock(device->act_log));
1492         drbd_al_shrink(device);
1493         err = drbd_check_al_size(device, dc);
1494         lc_unlock(device->act_log);
1495         wake_up(&device->al_wait);
1496 out:
1497         drbd_resume_io(device);
1498         return err;
1499 }
1500
1501 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1502 {
1503         struct drbd_config_context adm_ctx;
1504         enum drbd_ret_code retcode;
1505         struct drbd_device *device;
1506         struct disk_conf *new_disk_conf, *old_disk_conf;
1507         struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1508         int err;
1509         unsigned int fifo_size;
1510
1511         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1512         if (!adm_ctx.reply_skb)
1513                 return retcode;
1514         if (retcode != NO_ERROR)
1515                 goto finish;
1516
1517         device = adm_ctx.device;
1518         mutex_lock(&adm_ctx.resource->adm_mutex);
1519
1520         /* we also need a disk
1521          * to change the options on */
1522         if (!get_ldev(device)) {
1523                 retcode = ERR_NO_DISK;
1524                 goto out;
1525         }
1526
1527         new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1528         if (!new_disk_conf) {
1529                 retcode = ERR_NOMEM;
1530                 goto fail;
1531         }
1532
1533         mutex_lock(&device->resource->conf_update);
1534         old_disk_conf = device->ldev->disk_conf;
1535         *new_disk_conf = *old_disk_conf;
1536         if (should_set_defaults(info))
1537                 set_disk_conf_defaults(new_disk_conf);
1538
1539         err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1540         if (err && err != -ENOMSG) {
1541                 retcode = ERR_MANDATORY_TAG;
1542                 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1543                 goto fail_unlock;
1544         }
1545
1546         if (!expect(device, new_disk_conf->resync_rate >= 1))
1547                 new_disk_conf->resync_rate = 1;
1548
1549         sanitize_disk_conf(device, new_disk_conf, device->ldev);
1550
1551         if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1552                 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1553
1554         fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1555         if (fifo_size != device->rs_plan_s->size) {
1556                 new_plan = fifo_alloc(fifo_size);
1557                 if (!new_plan) {
1558                         drbd_err(device, "kmalloc of fifo_buffer failed");
1559                         retcode = ERR_NOMEM;
1560                         goto fail_unlock;
1561                 }
1562         }
1563
1564         err = disk_opts_check_al_size(device, new_disk_conf);
1565         if (err) {
1566                 /* Could be just "busy". Ignore?
1567                  * Introduce dedicated error code? */
1568                 drbd_msg_put_info(adm_ctx.reply_skb,
1569                         "Try again without changing current al-extents setting");
1570                 retcode = ERR_NOMEM;
1571                 goto fail_unlock;
1572         }
1573
1574         lock_all_resources();
1575         retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1576         if (retcode == NO_ERROR) {
1577                 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
1578                 drbd_resync_after_changed(device);
1579         }
1580         unlock_all_resources();
1581
1582         if (retcode != NO_ERROR)
1583                 goto fail_unlock;
1584
1585         if (new_plan) {
1586                 old_plan = device->rs_plan_s;
1587                 rcu_assign_pointer(device->rs_plan_s, new_plan);
1588         }
1589
1590         mutex_unlock(&device->resource->conf_update);
1591
1592         if (new_disk_conf->al_updates)
1593                 device->ldev->md.flags &= ~MDF_AL_DISABLED;
1594         else
1595                 device->ldev->md.flags |= MDF_AL_DISABLED;
1596
1597         if (new_disk_conf->md_flushes)
1598                 clear_bit(MD_NO_FUA, &device->flags);
1599         else
1600                 set_bit(MD_NO_FUA, &device->flags);
1601
1602         if (write_ordering_changed(old_disk_conf, new_disk_conf))
1603                 drbd_bump_write_ordering(device->resource, NULL, WO_BDEV_FLUSH);
1604
1605         if (old_disk_conf->discard_zeroes_if_aligned !=
1606             new_disk_conf->discard_zeroes_if_aligned)
1607                 drbd_reconsider_queue_parameters(device, device->ldev, NULL);
1608
1609         drbd_md_sync(device);
1610
1611         if (device->state.conn >= C_CONNECTED) {
1612                 struct drbd_peer_device *peer_device;
1613
1614                 for_each_peer_device(peer_device, device)
1615                         drbd_send_sync_param(peer_device);
1616         }
1617
1618         kvfree_rcu(old_disk_conf);
1619         kfree(old_plan);
1620         mod_timer(&device->request_timer, jiffies + HZ);
1621         goto success;
1622
1623 fail_unlock:
1624         mutex_unlock(&device->resource->conf_update);
1625  fail:
1626         kfree(new_disk_conf);
1627         kfree(new_plan);
1628 success:
1629         put_ldev(device);
1630  out:
1631         mutex_unlock(&adm_ctx.resource->adm_mutex);
1632  finish:
1633         drbd_adm_finish(&adm_ctx, info, retcode);
1634         return 0;
1635 }
1636
1637 static struct block_device *open_backing_dev(struct drbd_device *device,
1638                 const char *bdev_path, void *claim_ptr, bool do_bd_link)
1639 {
1640         struct block_device *bdev;
1641         int err = 0;
1642
1643         bdev = blkdev_get_by_path(bdev_path,
1644                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, claim_ptr);
1645         if (IS_ERR(bdev)) {
1646                 drbd_err(device, "open(\"%s\") failed with %ld\n",
1647                                 bdev_path, PTR_ERR(bdev));
1648                 return bdev;
1649         }
1650
1651         if (!do_bd_link)
1652                 return bdev;
1653
1654         err = bd_link_disk_holder(bdev, device->vdisk);
1655         if (err) {
1656                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1657                 drbd_err(device, "bd_link_disk_holder(\"%s\", ...) failed with %d\n",
1658                                 bdev_path, err);
1659                 bdev = ERR_PTR(err);
1660         }
1661         return bdev;
1662 }
1663
1664 static int open_backing_devices(struct drbd_device *device,
1665                 struct disk_conf *new_disk_conf,
1666                 struct drbd_backing_dev *nbc)
1667 {
1668         struct block_device *bdev;
1669
1670         bdev = open_backing_dev(device, new_disk_conf->backing_dev, device, true);
1671         if (IS_ERR(bdev))
1672                 return ERR_OPEN_DISK;
1673         nbc->backing_bdev = bdev;
1674
1675         /*
1676          * meta_dev_idx >= 0: external fixed size, possibly multiple
1677          * drbd sharing one meta device.  TODO in that case, paranoia
1678          * check that [md_bdev, meta_dev_idx] is not yet used by some
1679          * other drbd minor!  (if you use drbd.conf + drbdadm, that
1680          * should check it for you already; but if you don't, or
1681          * someone fooled it, we need to double check here)
1682          */
1683         bdev = open_backing_dev(device, new_disk_conf->meta_dev,
1684                 /* claim ptr: device, if claimed exclusively; shared drbd_m_holder,
1685                  * if potentially shared with other drbd minors */
1686                         (new_disk_conf->meta_dev_idx < 0) ? (void*)device : (void*)drbd_m_holder,
1687                 /* avoid double bd_claim_by_disk() for the same (source,target) tuple,
1688                  * as would happen with internal metadata. */
1689                         (new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_FLEX_INT &&
1690                          new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_INTERNAL));
1691         if (IS_ERR(bdev))
1692                 return ERR_OPEN_MD_DISK;
1693         nbc->md_bdev = bdev;
1694         return NO_ERROR;
1695 }
1696
1697 static void close_backing_dev(struct drbd_device *device, struct block_device *bdev,
1698         bool do_bd_unlink)
1699 {
1700         if (!bdev)
1701                 return;
1702         if (do_bd_unlink)
1703                 bd_unlink_disk_holder(bdev, device->vdisk);
1704         blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1705 }
1706
1707 void drbd_backing_dev_free(struct drbd_device *device, struct drbd_backing_dev *ldev)
1708 {
1709         if (ldev == NULL)
1710                 return;
1711
1712         close_backing_dev(device, ldev->md_bdev, ldev->md_bdev != ldev->backing_bdev);
1713         close_backing_dev(device, ldev->backing_bdev, true);
1714
1715         kfree(ldev->disk_conf);
1716         kfree(ldev);
1717 }
1718
1719 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1720 {
1721         struct drbd_config_context adm_ctx;
1722         struct drbd_device *device;
1723         struct drbd_peer_device *peer_device;
1724         struct drbd_connection *connection;
1725         int err;
1726         enum drbd_ret_code retcode;
1727         enum determine_dev_size dd;
1728         sector_t max_possible_sectors;
1729         sector_t min_md_device_sectors;
1730         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1731         struct disk_conf *new_disk_conf = NULL;
1732         struct lru_cache *resync_lru = NULL;
1733         struct fifo_buffer *new_plan = NULL;
1734         union drbd_state ns, os;
1735         enum drbd_state_rv rv;
1736         struct net_conf *nc;
1737
1738         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1739         if (!adm_ctx.reply_skb)
1740                 return retcode;
1741         if (retcode != NO_ERROR)
1742                 goto finish;
1743
1744         device = adm_ctx.device;
1745         mutex_lock(&adm_ctx.resource->adm_mutex);
1746         peer_device = first_peer_device(device);
1747         connection = peer_device->connection;
1748         conn_reconfig_start(connection);
1749
1750         /* if you want to reconfigure, please tear down first */
1751         if (device->state.disk > D_DISKLESS) {
1752                 retcode = ERR_DISK_CONFIGURED;
1753                 goto fail;
1754         }
1755         /* It may just now have detached because of IO error.  Make sure
1756          * drbd_ldev_destroy is done already, we may end up here very fast,
1757          * e.g. if someone calls attach from the on-io-error handler,
1758          * to realize a "hot spare" feature (not that I'd recommend that) */
1759         wait_event(device->misc_wait, !test_bit(GOING_DISKLESS, &device->flags));
1760
1761         /* make sure there is no leftover from previous force-detach attempts */
1762         clear_bit(FORCE_DETACH, &device->flags);
1763         clear_bit(WAS_IO_ERROR, &device->flags);
1764         clear_bit(WAS_READ_ERROR, &device->flags);
1765
1766         /* and no leftover from previously aborted resync or verify, either */
1767         device->rs_total = 0;
1768         device->rs_failed = 0;
1769         atomic_set(&device->rs_pending_cnt, 0);
1770
1771         /* allocation not in the IO path, drbdsetup context */
1772         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1773         if (!nbc) {
1774                 retcode = ERR_NOMEM;
1775                 goto fail;
1776         }
1777         spin_lock_init(&nbc->md.uuid_lock);
1778
1779         new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1780         if (!new_disk_conf) {
1781                 retcode = ERR_NOMEM;
1782                 goto fail;
1783         }
1784         nbc->disk_conf = new_disk_conf;
1785
1786         set_disk_conf_defaults(new_disk_conf);
1787         err = disk_conf_from_attrs(new_disk_conf, info);
1788         if (err) {
1789                 retcode = ERR_MANDATORY_TAG;
1790                 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1791                 goto fail;
1792         }
1793
1794         if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1795                 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1796
1797         new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1798         if (!new_plan) {
1799                 retcode = ERR_NOMEM;
1800                 goto fail;
1801         }
1802
1803         if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1804                 retcode = ERR_MD_IDX_INVALID;
1805                 goto fail;
1806         }
1807
1808         rcu_read_lock();
1809         nc = rcu_dereference(connection->net_conf);
1810         if (nc) {
1811                 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1812                         rcu_read_unlock();
1813                         retcode = ERR_STONITH_AND_PROT_A;
1814                         goto fail;
1815                 }
1816         }
1817         rcu_read_unlock();
1818
1819         retcode = open_backing_devices(device, new_disk_conf, nbc);
1820         if (retcode != NO_ERROR)
1821                 goto fail;
1822
1823         if ((nbc->backing_bdev == nbc->md_bdev) !=
1824             (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1825              new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1826                 retcode = ERR_MD_IDX_INVALID;
1827                 goto fail;
1828         }
1829
1830         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1831                         1, 61, sizeof(struct bm_extent),
1832                         offsetof(struct bm_extent, lce));
1833         if (!resync_lru) {
1834                 retcode = ERR_NOMEM;
1835                 goto fail;
1836         }
1837
1838         /* Read our meta data super block early.
1839          * This also sets other on-disk offsets. */
1840         retcode = drbd_md_read(device, nbc);
1841         if (retcode != NO_ERROR)
1842                 goto fail;
1843
1844         sanitize_disk_conf(device, new_disk_conf, nbc);
1845
1846         if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1847                 drbd_err(device, "max capacity %llu smaller than disk size %llu\n",
1848                         (unsigned long long) drbd_get_max_capacity(nbc),
1849                         (unsigned long long) new_disk_conf->disk_size);
1850                 retcode = ERR_DISK_TOO_SMALL;
1851                 goto fail;
1852         }
1853
1854         if (new_disk_conf->meta_dev_idx < 0) {
1855                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1856                 /* at least one MB, otherwise it does not make sense */
1857                 min_md_device_sectors = (2<<10);
1858         } else {
1859                 max_possible_sectors = DRBD_MAX_SECTORS;
1860                 min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1);
1861         }
1862
1863         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1864                 retcode = ERR_MD_DISK_TOO_SMALL;
1865                 drbd_warn(device, "refusing attach: md-device too small, "
1866                      "at least %llu sectors needed for this meta-disk type\n",
1867                      (unsigned long long) min_md_device_sectors);
1868                 goto fail;
1869         }
1870
1871         /* Make sure the new disk is big enough
1872          * (we may currently be R_PRIMARY with no local disk...) */
1873         if (drbd_get_max_capacity(nbc) < get_capacity(device->vdisk)) {
1874                 retcode = ERR_DISK_TOO_SMALL;
1875                 goto fail;
1876         }
1877
1878         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1879
1880         if (nbc->known_size > max_possible_sectors) {
1881                 drbd_warn(device, "==> truncating very big lower level device "
1882                         "to currently maximum possible %llu sectors <==\n",
1883                         (unsigned long long) max_possible_sectors);
1884                 if (new_disk_conf->meta_dev_idx >= 0)
1885                         drbd_warn(device, "==>> using internal or flexible "
1886                                       "meta data may help <<==\n");
1887         }
1888
1889         drbd_suspend_io(device);
1890         /* also wait for the last barrier ack. */
1891         /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1892          * We need a way to either ignore barrier acks for barriers sent before a device
1893          * was attached, or a way to wait for all pending barrier acks to come in.
1894          * As barriers are counted per resource,
1895          * we'd need to suspend io on all devices of a resource.
1896          */
1897         wait_event(device->misc_wait, !atomic_read(&device->ap_pending_cnt) || drbd_suspended(device));
1898         /* and for any other previously queued work */
1899         drbd_flush_workqueue(&connection->sender_work);
1900
1901         rv = _drbd_request_state(device, NS(disk, D_ATTACHING), CS_VERBOSE);
1902         retcode = (enum drbd_ret_code)rv;
1903         drbd_resume_io(device);
1904         if (rv < SS_SUCCESS)
1905                 goto fail;
1906
1907         if (!get_ldev_if_state(device, D_ATTACHING))
1908                 goto force_diskless;
1909
1910         if (!device->bitmap) {
1911                 if (drbd_bm_init(device)) {
1912                         retcode = ERR_NOMEM;
1913                         goto force_diskless_dec;
1914                 }
1915         }
1916
1917         if (device->state.pdsk != D_UP_TO_DATE && device->ed_uuid &&
1918             (device->state.role == R_PRIMARY || device->state.peer == R_PRIMARY) &&
1919             (device->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1920                 drbd_err(device, "Can only attach to data with current UUID=%016llX\n",
1921                     (unsigned long long)device->ed_uuid);
1922                 retcode = ERR_DATA_NOT_CURRENT;
1923                 goto force_diskless_dec;
1924         }
1925
1926         /* Since we are diskless, fix the activity log first... */
1927         if (drbd_check_al_size(device, new_disk_conf)) {
1928                 retcode = ERR_NOMEM;
1929                 goto force_diskless_dec;
1930         }
1931
1932         /* Prevent shrinking of consistent devices ! */
1933         {
1934         unsigned long long nsz = drbd_new_dev_size(device, nbc, nbc->disk_conf->disk_size, 0);
1935         unsigned long long eff = nbc->md.la_size_sect;
1936         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) && nsz < eff) {
1937                 if (nsz == nbc->disk_conf->disk_size) {
1938                         drbd_warn(device, "truncating a consistent device during attach (%llu < %llu)\n", nsz, eff);
1939                 } else {
1940                         drbd_warn(device, "refusing to truncate a consistent device (%llu < %llu)\n", nsz, eff);
1941                         drbd_msg_sprintf_info(adm_ctx.reply_skb,
1942                                 "To-be-attached device has last effective > current size, and is consistent\n"
1943                                 "(%llu > %llu sectors). Refusing to attach.", eff, nsz);
1944                         retcode = ERR_IMPLICIT_SHRINK;
1945                         goto force_diskless_dec;
1946                 }
1947         }
1948         }
1949
1950         lock_all_resources();
1951         retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1952         if (retcode != NO_ERROR) {
1953                 unlock_all_resources();
1954                 goto force_diskless_dec;
1955         }
1956
1957         /* Reset the "barriers don't work" bits here, then force meta data to
1958          * be written, to ensure we determine if barriers are supported. */
1959         if (new_disk_conf->md_flushes)
1960                 clear_bit(MD_NO_FUA, &device->flags);
1961         else
1962                 set_bit(MD_NO_FUA, &device->flags);
1963
1964         /* Point of no return reached.
1965          * Devices and memory are no longer released by error cleanup below.
1966          * now device takes over responsibility, and the state engine should
1967          * clean it up somewhere.  */
1968         D_ASSERT(device, device->ldev == NULL);
1969         device->ldev = nbc;
1970         device->resync = resync_lru;
1971         device->rs_plan_s = new_plan;
1972         nbc = NULL;
1973         resync_lru = NULL;
1974         new_disk_conf = NULL;
1975         new_plan = NULL;
1976
1977         drbd_resync_after_changed(device);
1978         drbd_bump_write_ordering(device->resource, device->ldev, WO_BDEV_FLUSH);
1979         unlock_all_resources();
1980
1981         if (drbd_md_test_flag(device->ldev, MDF_CRASHED_PRIMARY))
1982                 set_bit(CRASHED_PRIMARY, &device->flags);
1983         else
1984                 clear_bit(CRASHED_PRIMARY, &device->flags);
1985
1986         if (drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
1987             !(device->state.role == R_PRIMARY && device->resource->susp_nod))
1988                 set_bit(CRASHED_PRIMARY, &device->flags);
1989
1990         device->send_cnt = 0;
1991         device->recv_cnt = 0;
1992         device->read_cnt = 0;
1993         device->writ_cnt = 0;
1994
1995         drbd_reconsider_queue_parameters(device, device->ldev, NULL);
1996
1997         /* If I am currently not R_PRIMARY,
1998          * but meta data primary indicator is set,
1999          * I just now recover from a hard crash,
2000          * and have been R_PRIMARY before that crash.
2001          *
2002          * Now, if I had no connection before that crash
2003          * (have been degraded R_PRIMARY), chances are that
2004          * I won't find my peer now either.
2005          *
2006          * In that case, and _only_ in that case,
2007          * we use the degr-wfc-timeout instead of the default,
2008          * so we can automatically recover from a crash of a
2009          * degraded but active "cluster" after a certain timeout.
2010          */
2011         clear_bit(USE_DEGR_WFC_T, &device->flags);
2012         if (device->state.role != R_PRIMARY &&
2013              drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
2014             !drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND))
2015                 set_bit(USE_DEGR_WFC_T, &device->flags);
2016
2017         dd = drbd_determine_dev_size(device, 0, NULL);
2018         if (dd <= DS_ERROR) {
2019                 retcode = ERR_NOMEM_BITMAP;
2020                 goto force_diskless_dec;
2021         } else if (dd == DS_GREW)
2022                 set_bit(RESYNC_AFTER_NEG, &device->flags);
2023
2024         if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ||
2025             (test_bit(CRASHED_PRIMARY, &device->flags) &&
2026              drbd_md_test_flag(device->ldev, MDF_AL_DISABLED))) {
2027                 drbd_info(device, "Assuming that all blocks are out of sync "
2028                      "(aka FullSync)\n");
2029                 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
2030                         "set_n_write from attaching", BM_LOCKED_MASK)) {
2031                         retcode = ERR_IO_MD_DISK;
2032                         goto force_diskless_dec;
2033                 }
2034         } else {
2035                 if (drbd_bitmap_io(device, &drbd_bm_read,
2036                         "read from attaching", BM_LOCKED_MASK)) {
2037                         retcode = ERR_IO_MD_DISK;
2038                         goto force_diskless_dec;
2039                 }
2040         }
2041
2042         if (_drbd_bm_total_weight(device) == drbd_bm_bits(device))
2043                 drbd_suspend_al(device); /* IO is still suspended here... */
2044
2045         spin_lock_irq(&device->resource->req_lock);
2046         os = drbd_read_state(device);
2047         ns = os;
2048         /* If MDF_CONSISTENT is not set go into inconsistent state,
2049            otherwise investigate MDF_WasUpToDate...
2050            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
2051            otherwise into D_CONSISTENT state.
2052         */
2053         if (drbd_md_test_flag(device->ldev, MDF_CONSISTENT)) {
2054                 if (drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE))
2055                         ns.disk = D_CONSISTENT;
2056                 else
2057                         ns.disk = D_OUTDATED;
2058         } else {
2059                 ns.disk = D_INCONSISTENT;
2060         }
2061
2062         if (drbd_md_test_flag(device->ldev, MDF_PEER_OUT_DATED))
2063                 ns.pdsk = D_OUTDATED;
2064
2065         rcu_read_lock();
2066         if (ns.disk == D_CONSISTENT &&
2067             (ns.pdsk == D_OUTDATED || rcu_dereference(device->ldev->disk_conf)->fencing == FP_DONT_CARE))
2068                 ns.disk = D_UP_TO_DATE;
2069
2070         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
2071            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
2072            this point, because drbd_request_state() modifies these
2073            flags. */
2074
2075         if (rcu_dereference(device->ldev->disk_conf)->al_updates)
2076                 device->ldev->md.flags &= ~MDF_AL_DISABLED;
2077         else
2078                 device->ldev->md.flags |= MDF_AL_DISABLED;
2079
2080         rcu_read_unlock();
2081
2082         /* In case we are C_CONNECTED postpone any decision on the new disk
2083            state after the negotiation phase. */
2084         if (device->state.conn == C_CONNECTED) {
2085                 device->new_state_tmp.i = ns.i;
2086                 ns.i = os.i;
2087                 ns.disk = D_NEGOTIATING;
2088
2089                 /* We expect to receive up-to-date UUIDs soon.
2090                    To avoid a race in receive_state, free p_uuid while
2091                    holding req_lock. I.e. atomic with the state change */
2092                 kfree(device->p_uuid);
2093                 device->p_uuid = NULL;
2094         }
2095
2096         rv = _drbd_set_state(device, ns, CS_VERBOSE, NULL);
2097         spin_unlock_irq(&device->resource->req_lock);
2098
2099         if (rv < SS_SUCCESS)
2100                 goto force_diskless_dec;
2101
2102         mod_timer(&device->request_timer, jiffies + HZ);
2103
2104         if (device->state.role == R_PRIMARY)
2105                 device->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
2106         else
2107                 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
2108
2109         drbd_md_mark_dirty(device);
2110         drbd_md_sync(device);
2111
2112         kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
2113         put_ldev(device);
2114         conn_reconfig_done(connection);
2115         mutex_unlock(&adm_ctx.resource->adm_mutex);
2116         drbd_adm_finish(&adm_ctx, info, retcode);
2117         return 0;
2118
2119  force_diskless_dec:
2120         put_ldev(device);
2121  force_diskless:
2122         drbd_force_state(device, NS(disk, D_DISKLESS));
2123         drbd_md_sync(device);
2124  fail:
2125         conn_reconfig_done(connection);
2126         if (nbc) {
2127                 close_backing_dev(device, nbc->md_bdev, nbc->md_bdev != nbc->backing_bdev);
2128                 close_backing_dev(device, nbc->backing_bdev, true);
2129                 kfree(nbc);
2130         }
2131         kfree(new_disk_conf);
2132         lc_destroy(resync_lru);
2133         kfree(new_plan);
2134         mutex_unlock(&adm_ctx.resource->adm_mutex);
2135  finish:
2136         drbd_adm_finish(&adm_ctx, info, retcode);
2137         return 0;
2138 }
2139
2140 static int adm_detach(struct drbd_device *device, int force)
2141 {
2142         if (force) {
2143                 set_bit(FORCE_DETACH, &device->flags);
2144                 drbd_force_state(device, NS(disk, D_FAILED));
2145                 return SS_SUCCESS;
2146         }
2147
2148         return drbd_request_detach_interruptible(device);
2149 }
2150
2151 /* Detaching the disk is a process in multiple stages.  First we need to lock
2152  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
2153  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
2154  * internal references as well.
2155  * Only then we have finally detached. */
2156 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
2157 {
2158         struct drbd_config_context adm_ctx;
2159         enum drbd_ret_code retcode;
2160         struct detach_parms parms = { };
2161         int err;
2162
2163         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2164         if (!adm_ctx.reply_skb)
2165                 return retcode;
2166         if (retcode != NO_ERROR)
2167                 goto out;
2168
2169         if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
2170                 err = detach_parms_from_attrs(&parms, info);
2171                 if (err) {
2172                         retcode = ERR_MANDATORY_TAG;
2173                         drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2174                         goto out;
2175                 }
2176         }
2177
2178         mutex_lock(&adm_ctx.resource->adm_mutex);
2179         retcode = adm_detach(adm_ctx.device, parms.force_detach);
2180         mutex_unlock(&adm_ctx.resource->adm_mutex);
2181 out:
2182         drbd_adm_finish(&adm_ctx, info, retcode);
2183         return 0;
2184 }
2185
2186 static bool conn_resync_running(struct drbd_connection *connection)
2187 {
2188         struct drbd_peer_device *peer_device;
2189         bool rv = false;
2190         int vnr;
2191
2192         rcu_read_lock();
2193         idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2194                 struct drbd_device *device = peer_device->device;
2195                 if (device->state.conn == C_SYNC_SOURCE ||
2196                     device->state.conn == C_SYNC_TARGET ||
2197                     device->state.conn == C_PAUSED_SYNC_S ||
2198                     device->state.conn == C_PAUSED_SYNC_T) {
2199                         rv = true;
2200                         break;
2201                 }
2202         }
2203         rcu_read_unlock();
2204
2205         return rv;
2206 }
2207
2208 static bool conn_ov_running(struct drbd_connection *connection)
2209 {
2210         struct drbd_peer_device *peer_device;
2211         bool rv = false;
2212         int vnr;
2213
2214         rcu_read_lock();
2215         idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2216                 struct drbd_device *device = peer_device->device;
2217                 if (device->state.conn == C_VERIFY_S ||
2218                     device->state.conn == C_VERIFY_T) {
2219                         rv = true;
2220                         break;
2221                 }
2222         }
2223         rcu_read_unlock();
2224
2225         return rv;
2226 }
2227
2228 static enum drbd_ret_code
2229 _check_net_options(struct drbd_connection *connection, struct net_conf *old_net_conf, struct net_conf *new_net_conf)
2230 {
2231         struct drbd_peer_device *peer_device;
2232         int i;
2233
2234         if (old_net_conf && connection->cstate == C_WF_REPORT_PARAMS && connection->agreed_pro_version < 100) {
2235                 if (new_net_conf->wire_protocol != old_net_conf->wire_protocol)
2236                         return ERR_NEED_APV_100;
2237
2238                 if (new_net_conf->two_primaries != old_net_conf->two_primaries)
2239                         return ERR_NEED_APV_100;
2240
2241                 if (strcmp(new_net_conf->integrity_alg, old_net_conf->integrity_alg))
2242                         return ERR_NEED_APV_100;
2243         }
2244
2245         if (!new_net_conf->two_primaries &&
2246             conn_highest_role(connection) == R_PRIMARY &&
2247             conn_highest_peer(connection) == R_PRIMARY)
2248                 return ERR_NEED_ALLOW_TWO_PRI;
2249
2250         if (new_net_conf->two_primaries &&
2251             (new_net_conf->wire_protocol != DRBD_PROT_C))
2252                 return ERR_NOT_PROTO_C;
2253
2254         idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2255                 struct drbd_device *device = peer_device->device;
2256                 if (get_ldev(device)) {
2257                         enum drbd_fencing_p fp = rcu_dereference(device->ldev->disk_conf)->fencing;
2258                         put_ldev(device);
2259                         if (new_net_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
2260                                 return ERR_STONITH_AND_PROT_A;
2261                 }
2262                 if (device->state.role == R_PRIMARY && new_net_conf->discard_my_data)
2263                         return ERR_DISCARD_IMPOSSIBLE;
2264         }
2265
2266         if (new_net_conf->on_congestion != OC_BLOCK && new_net_conf->wire_protocol != DRBD_PROT_A)
2267                 return ERR_CONG_NOT_PROTO_A;
2268
2269         return NO_ERROR;
2270 }
2271
2272 static enum drbd_ret_code
2273 check_net_options(struct drbd_connection *connection, struct net_conf *new_net_conf)
2274 {
2275         enum drbd_ret_code rv;
2276         struct drbd_peer_device *peer_device;
2277         int i;
2278
2279         rcu_read_lock();
2280         rv = _check_net_options(connection, rcu_dereference(connection->net_conf), new_net_conf);
2281         rcu_read_unlock();
2282
2283         /* connection->peer_devices protected by genl_lock() here */
2284         idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2285                 struct drbd_device *device = peer_device->device;
2286                 if (!device->bitmap) {
2287                         if (drbd_bm_init(device))
2288                                 return ERR_NOMEM;
2289                 }
2290         }
2291
2292         return rv;
2293 }
2294
2295 struct crypto {
2296         struct crypto_shash *verify_tfm;
2297         struct crypto_shash *csums_tfm;
2298         struct crypto_shash *cram_hmac_tfm;
2299         struct crypto_shash *integrity_tfm;
2300 };
2301
2302 static int
2303 alloc_shash(struct crypto_shash **tfm, char *tfm_name, int err_alg)
2304 {
2305         if (!tfm_name[0])
2306                 return NO_ERROR;
2307
2308         *tfm = crypto_alloc_shash(tfm_name, 0, 0);
2309         if (IS_ERR(*tfm)) {
2310                 *tfm = NULL;
2311                 return err_alg;
2312         }
2313
2314         return NO_ERROR;
2315 }
2316
2317 static enum drbd_ret_code
2318 alloc_crypto(struct crypto *crypto, struct net_conf *new_net_conf)
2319 {
2320         char hmac_name[CRYPTO_MAX_ALG_NAME];
2321         enum drbd_ret_code rv;
2322
2323         rv = alloc_shash(&crypto->csums_tfm, new_net_conf->csums_alg,
2324                          ERR_CSUMS_ALG);
2325         if (rv != NO_ERROR)
2326                 return rv;
2327         rv = alloc_shash(&crypto->verify_tfm, new_net_conf->verify_alg,
2328                          ERR_VERIFY_ALG);
2329         if (rv != NO_ERROR)
2330                 return rv;
2331         rv = alloc_shash(&crypto->integrity_tfm, new_net_conf->integrity_alg,
2332                          ERR_INTEGRITY_ALG);
2333         if (rv != NO_ERROR)
2334                 return rv;
2335         if (new_net_conf->cram_hmac_alg[0] != 0) {
2336                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
2337                          new_net_conf->cram_hmac_alg);
2338
2339                 rv = alloc_shash(&crypto->cram_hmac_tfm, hmac_name,
2340                                  ERR_AUTH_ALG);
2341         }
2342
2343         return rv;
2344 }
2345
2346 static void free_crypto(struct crypto *crypto)
2347 {
2348         crypto_free_shash(crypto->cram_hmac_tfm);
2349         crypto_free_shash(crypto->integrity_tfm);
2350         crypto_free_shash(crypto->csums_tfm);
2351         crypto_free_shash(crypto->verify_tfm);
2352 }
2353
2354 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
2355 {
2356         struct drbd_config_context adm_ctx;
2357         enum drbd_ret_code retcode;
2358         struct drbd_connection *connection;
2359         struct net_conf *old_net_conf, *new_net_conf = NULL;
2360         int err;
2361         int ovr; /* online verify running */
2362         int rsr; /* re-sync running */
2363         struct crypto crypto = { };
2364
2365         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2366         if (!adm_ctx.reply_skb)
2367                 return retcode;
2368         if (retcode != NO_ERROR)
2369                 goto finish;
2370
2371         connection = adm_ctx.connection;
2372         mutex_lock(&adm_ctx.resource->adm_mutex);
2373
2374         new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
2375         if (!new_net_conf) {
2376                 retcode = ERR_NOMEM;
2377                 goto out;
2378         }
2379
2380         conn_reconfig_start(connection);
2381
2382         mutex_lock(&connection->data.mutex);
2383         mutex_lock(&connection->resource->conf_update);
2384         old_net_conf = connection->net_conf;
2385
2386         if (!old_net_conf) {
2387                 drbd_msg_put_info(adm_ctx.reply_skb, "net conf missing, try connect");
2388                 retcode = ERR_INVALID_REQUEST;
2389                 goto fail;
2390         }
2391
2392         *new_net_conf = *old_net_conf;
2393         if (should_set_defaults(info))
2394                 set_net_conf_defaults(new_net_conf);
2395
2396         err = net_conf_from_attrs_for_change(new_net_conf, info);
2397         if (err && err != -ENOMSG) {
2398                 retcode = ERR_MANDATORY_TAG;
2399                 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2400                 goto fail;
2401         }
2402
2403         retcode = check_net_options(connection, new_net_conf);
2404         if (retcode != NO_ERROR)
2405                 goto fail;
2406
2407         /* re-sync running */
2408         rsr = conn_resync_running(connection);
2409         if (rsr && strcmp(new_net_conf->csums_alg, old_net_conf->csums_alg)) {
2410                 retcode = ERR_CSUMS_RESYNC_RUNNING;
2411                 goto fail;
2412         }
2413
2414         /* online verify running */
2415         ovr = conn_ov_running(connection);
2416         if (ovr && strcmp(new_net_conf->verify_alg, old_net_conf->verify_alg)) {
2417                 retcode = ERR_VERIFY_RUNNING;
2418                 goto fail;
2419         }
2420
2421         retcode = alloc_crypto(&crypto, new_net_conf);
2422         if (retcode != NO_ERROR)
2423                 goto fail;
2424
2425         rcu_assign_pointer(connection->net_conf, new_net_conf);
2426
2427         if (!rsr) {
2428                 crypto_free_shash(connection->csums_tfm);
2429                 connection->csums_tfm = crypto.csums_tfm;
2430                 crypto.csums_tfm = NULL;
2431         }
2432         if (!ovr) {
2433                 crypto_free_shash(connection->verify_tfm);
2434                 connection->verify_tfm = crypto.verify_tfm;
2435                 crypto.verify_tfm = NULL;
2436         }
2437
2438         crypto_free_shash(connection->integrity_tfm);
2439         connection->integrity_tfm = crypto.integrity_tfm;
2440         if (connection->cstate >= C_WF_REPORT_PARAMS && connection->agreed_pro_version >= 100)
2441                 /* Do this without trying to take connection->data.mutex again.  */
2442                 __drbd_send_protocol(connection, P_PROTOCOL_UPDATE);
2443
2444         crypto_free_shash(connection->cram_hmac_tfm);
2445         connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2446
2447         mutex_unlock(&connection->resource->conf_update);
2448         mutex_unlock(&connection->data.mutex);
2449         kvfree_rcu(old_net_conf);
2450
2451         if (connection->cstate >= C_WF_REPORT_PARAMS) {
2452                 struct drbd_peer_device *peer_device;
2453                 int vnr;
2454
2455                 idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
2456                         drbd_send_sync_param(peer_device);
2457         }
2458
2459         goto done;
2460
2461  fail:
2462         mutex_unlock(&connection->resource->conf_update);
2463         mutex_unlock(&connection->data.mutex);
2464         free_crypto(&crypto);
2465         kfree(new_net_conf);
2466  done:
2467         conn_reconfig_done(connection);
2468  out:
2469         mutex_unlock(&adm_ctx.resource->adm_mutex);
2470  finish:
2471         drbd_adm_finish(&adm_ctx, info, retcode);
2472         return 0;
2473 }
2474
2475 static void connection_to_info(struct connection_info *info,
2476                                struct drbd_connection *connection)
2477 {
2478         info->conn_connection_state = connection->cstate;
2479         info->conn_role = conn_highest_peer(connection);
2480 }
2481
2482 static void peer_device_to_info(struct peer_device_info *info,
2483                                 struct drbd_peer_device *peer_device)
2484 {
2485         struct drbd_device *device = peer_device->device;
2486
2487         info->peer_repl_state =
2488                 max_t(enum drbd_conns, C_WF_REPORT_PARAMS, device->state.conn);
2489         info->peer_disk_state = device->state.pdsk;
2490         info->peer_resync_susp_user = device->state.user_isp;
2491         info->peer_resync_susp_peer = device->state.peer_isp;
2492         info->peer_resync_susp_dependency = device->state.aftr_isp;
2493 }
2494
2495 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2496 {
2497         struct connection_info connection_info;
2498         enum drbd_notification_type flags;
2499         unsigned int peer_devices = 0;
2500         struct drbd_config_context adm_ctx;
2501         struct drbd_peer_device *peer_device;
2502         struct net_conf *old_net_conf, *new_net_conf = NULL;
2503         struct crypto crypto = { };
2504         struct drbd_resource *resource;
2505         struct drbd_connection *connection;
2506         enum drbd_ret_code retcode;
2507         enum drbd_state_rv rv;
2508         int i;
2509         int err;
2510
2511         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2512
2513         if (!adm_ctx.reply_skb)
2514                 return retcode;
2515         if (retcode != NO_ERROR)
2516                 goto out;
2517         if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2518                 drbd_msg_put_info(adm_ctx.reply_skb, "connection endpoint(s) missing");
2519                 retcode = ERR_INVALID_REQUEST;
2520                 goto out;
2521         }
2522
2523         /* No need for _rcu here. All reconfiguration is
2524          * strictly serialized on genl_lock(). We are protected against
2525          * concurrent reconfiguration/addition/deletion */
2526         for_each_resource(resource, &drbd_resources) {
2527                 for_each_connection(connection, resource) {
2528                         if (nla_len(adm_ctx.my_addr) == connection->my_addr_len &&
2529                             !memcmp(nla_data(adm_ctx.my_addr), &connection->my_addr,
2530                                     connection->my_addr_len)) {
2531                                 retcode = ERR_LOCAL_ADDR;
2532                                 goto out;
2533                         }
2534
2535                         if (nla_len(adm_ctx.peer_addr) == connection->peer_addr_len &&
2536                             !memcmp(nla_data(adm_ctx.peer_addr), &connection->peer_addr,
2537                                     connection->peer_addr_len)) {
2538                                 retcode = ERR_PEER_ADDR;
2539                                 goto out;
2540                         }
2541                 }
2542         }
2543
2544         mutex_lock(&adm_ctx.resource->adm_mutex);
2545         connection = first_connection(adm_ctx.resource);
2546         conn_reconfig_start(connection);
2547
2548         if (connection->cstate > C_STANDALONE) {
2549                 retcode = ERR_NET_CONFIGURED;
2550                 goto fail;
2551         }
2552
2553         /* allocation not in the IO path, drbdsetup / netlink process context */
2554         new_net_conf = kzalloc(sizeof(*new_net_conf), GFP_KERNEL);
2555         if (!new_net_conf) {
2556                 retcode = ERR_NOMEM;
2557                 goto fail;
2558         }
2559
2560         set_net_conf_defaults(new_net_conf);
2561
2562         err = net_conf_from_attrs(new_net_conf, info);
2563         if (err && err != -ENOMSG) {
2564                 retcode = ERR_MANDATORY_TAG;
2565                 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2566                 goto fail;
2567         }
2568
2569         retcode = check_net_options(connection, new_net_conf);
2570         if (retcode != NO_ERROR)
2571                 goto fail;
2572
2573         retcode = alloc_crypto(&crypto, new_net_conf);
2574         if (retcode != NO_ERROR)
2575                 goto fail;
2576
2577         ((char *)new_net_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2578
2579         drbd_flush_workqueue(&connection->sender_work);
2580
2581         mutex_lock(&adm_ctx.resource->conf_update);
2582         old_net_conf = connection->net_conf;
2583         if (old_net_conf) {
2584                 retcode = ERR_NET_CONFIGURED;
2585                 mutex_unlock(&adm_ctx.resource->conf_update);
2586                 goto fail;
2587         }
2588         rcu_assign_pointer(connection->net_conf, new_net_conf);
2589
2590         conn_free_crypto(connection);
2591         connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2592         connection->integrity_tfm = crypto.integrity_tfm;
2593         connection->csums_tfm = crypto.csums_tfm;
2594         connection->verify_tfm = crypto.verify_tfm;
2595
2596         connection->my_addr_len = nla_len(adm_ctx.my_addr);
2597         memcpy(&connection->my_addr, nla_data(adm_ctx.my_addr), connection->my_addr_len);
2598         connection->peer_addr_len = nla_len(adm_ctx.peer_addr);
2599         memcpy(&connection->peer_addr, nla_data(adm_ctx.peer_addr), connection->peer_addr_len);
2600
2601         idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2602                 peer_devices++;
2603         }
2604
2605         connection_to_info(&connection_info, connection);
2606         flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2607         mutex_lock(&notification_mutex);
2608         notify_connection_state(NULL, 0, connection, &connection_info, NOTIFY_CREATE | flags);
2609         idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2610                 struct peer_device_info peer_device_info;
2611
2612                 peer_device_to_info(&peer_device_info, peer_device);
2613                 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2614                 notify_peer_device_state(NULL, 0, peer_device, &peer_device_info, NOTIFY_CREATE | flags);
2615         }
2616         mutex_unlock(&notification_mutex);
2617         mutex_unlock(&adm_ctx.resource->conf_update);
2618
2619         rcu_read_lock();
2620         idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2621                 struct drbd_device *device = peer_device->device;
2622                 device->send_cnt = 0;
2623                 device->recv_cnt = 0;
2624         }
2625         rcu_read_unlock();
2626
2627         rv = conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2628
2629         conn_reconfig_done(connection);
2630         mutex_unlock(&adm_ctx.resource->adm_mutex);
2631         drbd_adm_finish(&adm_ctx, info, rv);
2632         return 0;
2633
2634 fail:
2635         free_crypto(&crypto);
2636         kfree(new_net_conf);
2637
2638         conn_reconfig_done(connection);
2639         mutex_unlock(&adm_ctx.resource->adm_mutex);
2640 out:
2641         drbd_adm_finish(&adm_ctx, info, retcode);
2642         return 0;
2643 }
2644
2645 static enum drbd_state_rv conn_try_disconnect(struct drbd_connection *connection, bool force)
2646 {
2647         enum drbd_conns cstate;
2648         enum drbd_state_rv rv;
2649
2650 repeat:
2651         rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2652                         force ? CS_HARD : 0);
2653
2654         switch (rv) {
2655         case SS_NOTHING_TO_DO:
2656                 break;
2657         case SS_ALREADY_STANDALONE:
2658                 return SS_SUCCESS;
2659         case SS_PRIMARY_NOP:
2660                 /* Our state checking code wants to see the peer outdated. */
2661                 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, pdsk, D_OUTDATED), 0);
2662
2663                 if (rv == SS_OUTDATE_WO_CONN) /* lost connection before graceful disconnect succeeded */
2664                         rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_VERBOSE);
2665
2666                 break;
2667         case SS_CW_FAILED_BY_PEER:
2668                 spin_lock_irq(&connection->resource->req_lock);
2669                 cstate = connection->cstate;
2670                 spin_unlock_irq(&connection->resource->req_lock);
2671                 if (cstate <= C_WF_CONNECTION)
2672                         goto repeat;
2673                 /* The peer probably wants to see us outdated. */
2674                 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING,
2675                                                         disk, D_OUTDATED), 0);
2676                 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2677                         rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2678                                         CS_HARD);
2679                 }
2680                 break;
2681         default:;
2682                 /* no special handling necessary */
2683         }
2684
2685         if (rv >= SS_SUCCESS) {
2686                 enum drbd_state_rv rv2;
2687                 /* No one else can reconfigure the network while I am here.
2688                  * The state handling only uses drbd_thread_stop_nowait(),
2689                  * we want to really wait here until the receiver is no more.
2690                  */
2691                 drbd_thread_stop(&connection->receiver);
2692
2693                 /* Race breaker.  This additional state change request may be
2694                  * necessary, if this was a forced disconnect during a receiver
2695                  * restart.  We may have "killed" the receiver thread just
2696                  * after drbd_receiver() returned.  Typically, we should be
2697                  * C_STANDALONE already, now, and this becomes a no-op.
2698                  */
2699                 rv2 = conn_request_state(connection, NS(conn, C_STANDALONE),
2700                                 CS_VERBOSE | CS_HARD);
2701                 if (rv2 < SS_SUCCESS)
2702                         drbd_err(connection,
2703                                 "unexpected rv2=%d in conn_try_disconnect()\n",
2704                                 rv2);
2705                 /* Unlike in DRBD 9, the state engine has generated
2706                  * NOTIFY_DESTROY events before clearing connection->net_conf. */
2707         }
2708         return rv;
2709 }
2710
2711 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2712 {
2713         struct drbd_config_context adm_ctx;
2714         struct disconnect_parms parms;
2715         struct drbd_connection *connection;
2716         enum drbd_state_rv rv;
2717         enum drbd_ret_code retcode;
2718         int err;
2719
2720         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2721         if (!adm_ctx.reply_skb)
2722                 return retcode;
2723         if (retcode != NO_ERROR)
2724                 goto fail;
2725
2726         connection = adm_ctx.connection;
2727         memset(&parms, 0, sizeof(parms));
2728         if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2729                 err = disconnect_parms_from_attrs(&parms, info);
2730                 if (err) {
2731                         retcode = ERR_MANDATORY_TAG;
2732                         drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2733                         goto fail;
2734                 }
2735         }
2736
2737         mutex_lock(&adm_ctx.resource->adm_mutex);
2738         rv = conn_try_disconnect(connection, parms.force_disconnect);
2739         mutex_unlock(&adm_ctx.resource->adm_mutex);
2740         if (rv < SS_SUCCESS) {
2741                 drbd_adm_finish(&adm_ctx, info, rv);
2742                 return 0;
2743         }
2744         retcode = NO_ERROR;
2745  fail:
2746         drbd_adm_finish(&adm_ctx, info, retcode);
2747         return 0;
2748 }
2749
2750 void resync_after_online_grow(struct drbd_device *device)
2751 {
2752         int iass; /* I am sync source */
2753
2754         drbd_info(device, "Resync of new storage after online grow\n");
2755         if (device->state.role != device->state.peer)
2756                 iass = (device->state.role == R_PRIMARY);
2757         else
2758                 iass = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags);
2759
2760         if (iass)
2761                 drbd_start_resync(device, C_SYNC_SOURCE);
2762         else
2763                 _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2764 }
2765
2766 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2767 {
2768         struct drbd_config_context adm_ctx;
2769         struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2770         struct resize_parms rs;
2771         struct drbd_device *device;
2772         enum drbd_ret_code retcode;
2773         enum determine_dev_size dd;
2774         bool change_al_layout = false;
2775         enum dds_flags ddsf;
2776         sector_t u_size;
2777         int err;
2778
2779         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2780         if (!adm_ctx.reply_skb)
2781                 return retcode;
2782         if (retcode != NO_ERROR)
2783                 goto finish;
2784
2785         mutex_lock(&adm_ctx.resource->adm_mutex);
2786         device = adm_ctx.device;
2787         if (!get_ldev(device)) {
2788                 retcode = ERR_NO_DISK;
2789                 goto fail;
2790         }
2791
2792         memset(&rs, 0, sizeof(struct resize_parms));
2793         rs.al_stripes = device->ldev->md.al_stripes;
2794         rs.al_stripe_size = device->ldev->md.al_stripe_size_4k * 4;
2795         if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2796                 err = resize_parms_from_attrs(&rs, info);
2797                 if (err) {
2798                         retcode = ERR_MANDATORY_TAG;
2799                         drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2800                         goto fail_ldev;
2801                 }
2802         }
2803
2804         if (device->state.conn > C_CONNECTED) {
2805                 retcode = ERR_RESIZE_RESYNC;
2806                 goto fail_ldev;
2807         }
2808
2809         if (device->state.role == R_SECONDARY &&
2810             device->state.peer == R_SECONDARY) {
2811                 retcode = ERR_NO_PRIMARY;
2812                 goto fail_ldev;
2813         }
2814
2815         if (rs.no_resync && first_peer_device(device)->connection->agreed_pro_version < 93) {
2816                 retcode = ERR_NEED_APV_93;
2817                 goto fail_ldev;
2818         }
2819
2820         rcu_read_lock();
2821         u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
2822         rcu_read_unlock();
2823         if (u_size != (sector_t)rs.resize_size) {
2824                 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2825                 if (!new_disk_conf) {
2826                         retcode = ERR_NOMEM;
2827                         goto fail_ldev;
2828                 }
2829         }
2830
2831         if (device->ldev->md.al_stripes != rs.al_stripes ||
2832             device->ldev->md.al_stripe_size_4k != rs.al_stripe_size / 4) {
2833                 u32 al_size_k = rs.al_stripes * rs.al_stripe_size;
2834
2835                 if (al_size_k > (16 * 1024 * 1024)) {
2836                         retcode = ERR_MD_LAYOUT_TOO_BIG;
2837                         goto fail_ldev;
2838                 }
2839
2840                 if (al_size_k < MD_32kB_SECT/2) {
2841                         retcode = ERR_MD_LAYOUT_TOO_SMALL;
2842                         goto fail_ldev;
2843                 }
2844
2845                 if (device->state.conn != C_CONNECTED && !rs.resize_force) {
2846                         retcode = ERR_MD_LAYOUT_CONNECTED;
2847                         goto fail_ldev;
2848                 }
2849
2850                 change_al_layout = true;
2851         }
2852
2853         if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev))
2854                 device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev);
2855
2856         if (new_disk_conf) {
2857                 mutex_lock(&device->resource->conf_update);
2858                 old_disk_conf = device->ldev->disk_conf;
2859                 *new_disk_conf = *old_disk_conf;
2860                 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2861                 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
2862                 mutex_unlock(&device->resource->conf_update);
2863                 kvfree_rcu(old_disk_conf);
2864                 new_disk_conf = NULL;
2865         }
2866
2867         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2868         dd = drbd_determine_dev_size(device, ddsf, change_al_layout ? &rs : NULL);
2869         drbd_md_sync(device);
2870         put_ldev(device);
2871         if (dd == DS_ERROR) {
2872                 retcode = ERR_NOMEM_BITMAP;
2873                 goto fail;
2874         } else if (dd == DS_ERROR_SPACE_MD) {
2875                 retcode = ERR_MD_LAYOUT_NO_FIT;
2876                 goto fail;
2877         } else if (dd == DS_ERROR_SHRINK) {
2878                 retcode = ERR_IMPLICIT_SHRINK;
2879                 goto fail;
2880         }
2881
2882         if (device->state.conn == C_CONNECTED) {
2883                 if (dd == DS_GREW)
2884                         set_bit(RESIZE_PENDING, &device->flags);
2885
2886                 drbd_send_uuids(first_peer_device(device));
2887                 drbd_send_sizes(first_peer_device(device), 1, ddsf);
2888         }
2889
2890  fail:
2891         mutex_unlock(&adm_ctx.resource->adm_mutex);
2892  finish:
2893         drbd_adm_finish(&adm_ctx, info, retcode);
2894         return 0;
2895
2896  fail_ldev:
2897         put_ldev(device);
2898         kfree(new_disk_conf);
2899         goto fail;
2900 }
2901
2902 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2903 {
2904         struct drbd_config_context adm_ctx;
2905         enum drbd_ret_code retcode;
2906         struct res_opts res_opts;
2907         int err;
2908
2909         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2910         if (!adm_ctx.reply_skb)
2911                 return retcode;
2912         if (retcode != NO_ERROR)
2913                 goto fail;
2914
2915         res_opts = adm_ctx.resource->res_opts;
2916         if (should_set_defaults(info))
2917                 set_res_opts_defaults(&res_opts);
2918
2919         err = res_opts_from_attrs(&res_opts, info);
2920         if (err && err != -ENOMSG) {
2921                 retcode = ERR_MANDATORY_TAG;
2922                 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2923                 goto fail;
2924         }
2925
2926         mutex_lock(&adm_ctx.resource->adm_mutex);
2927         err = set_resource_options(adm_ctx.resource, &res_opts);
2928         if (err) {
2929                 retcode = ERR_INVALID_REQUEST;
2930                 if (err == -ENOMEM)
2931                         retcode = ERR_NOMEM;
2932         }
2933         mutex_unlock(&adm_ctx.resource->adm_mutex);
2934
2935 fail:
2936         drbd_adm_finish(&adm_ctx, info, retcode);
2937         return 0;
2938 }
2939
2940 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2941 {
2942         struct drbd_config_context adm_ctx;
2943         struct drbd_device *device;
2944         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2945
2946         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2947         if (!adm_ctx.reply_skb)
2948                 return retcode;
2949         if (retcode != NO_ERROR)
2950                 goto out;
2951
2952         device = adm_ctx.device;
2953         if (!get_ldev(device)) {
2954                 retcode = ERR_NO_DISK;
2955                 goto out;
2956         }
2957
2958         mutex_lock(&adm_ctx.resource->adm_mutex);
2959
2960         /* If there is still bitmap IO pending, probably because of a previous
2961          * resync just being finished, wait for it before requesting a new resync.
2962          * Also wait for it's after_state_ch(). */
2963         drbd_suspend_io(device);
2964         wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
2965         drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
2966
2967         /* If we happen to be C_STANDALONE R_SECONDARY, just change to
2968          * D_INCONSISTENT, and set all bits in the bitmap.  Otherwise,
2969          * try to start a resync handshake as sync target for full sync.
2970          */
2971         if (device->state.conn == C_STANDALONE && device->state.role == R_SECONDARY) {
2972                 retcode = drbd_request_state(device, NS(disk, D_INCONSISTENT));
2973                 if (retcode >= SS_SUCCESS) {
2974                         if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
2975                                 "set_n_write from invalidate", BM_LOCKED_MASK))
2976                                 retcode = ERR_IO_MD_DISK;
2977                 }
2978         } else
2979                 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_T));
2980         drbd_resume_io(device);
2981         mutex_unlock(&adm_ctx.resource->adm_mutex);
2982         put_ldev(device);
2983 out:
2984         drbd_adm_finish(&adm_ctx, info, retcode);
2985         return 0;
2986 }
2987
2988 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2989                 union drbd_state mask, union drbd_state val)
2990 {
2991         struct drbd_config_context adm_ctx;
2992         enum drbd_ret_code retcode;
2993
2994         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2995         if (!adm_ctx.reply_skb)
2996                 return retcode;
2997         if (retcode != NO_ERROR)
2998                 goto out;
2999
3000         mutex_lock(&adm_ctx.resource->adm_mutex);
3001         retcode = drbd_request_state(adm_ctx.device, mask, val);
3002         mutex_unlock(&adm_ctx.resource->adm_mutex);
3003 out:
3004         drbd_adm_finish(&adm_ctx, info, retcode);
3005         return 0;
3006 }
3007
3008 static int drbd_bmio_set_susp_al(struct drbd_device *device) __must_hold(local)
3009 {
3010         int rv;
3011
3012         rv = drbd_bmio_set_n_write(device);
3013         drbd_suspend_al(device);
3014         return rv;
3015 }
3016
3017 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
3018 {
3019         struct drbd_config_context adm_ctx;
3020         int retcode; /* drbd_ret_code, drbd_state_rv */
3021         struct drbd_device *device;
3022
3023         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3024         if (!adm_ctx.reply_skb)
3025                 return retcode;
3026         if (retcode != NO_ERROR)
3027                 goto out;
3028
3029         device = adm_ctx.device;
3030         if (!get_ldev(device)) {
3031                 retcode = ERR_NO_DISK;
3032                 goto out;
3033         }
3034
3035         mutex_lock(&adm_ctx.resource->adm_mutex);
3036
3037         /* If there is still bitmap IO pending, probably because of a previous
3038          * resync just being finished, wait for it before requesting a new resync.
3039          * Also wait for it's after_state_ch(). */
3040         drbd_suspend_io(device);
3041         wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
3042         drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
3043
3044         /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
3045          * in the bitmap.  Otherwise, try to start a resync handshake
3046          * as sync source for full sync.
3047          */
3048         if (device->state.conn == C_STANDALONE && device->state.role == R_PRIMARY) {
3049                 /* The peer will get a resync upon connect anyways. Just make that
3050                    into a full resync. */
3051                 retcode = drbd_request_state(device, NS(pdsk, D_INCONSISTENT));
3052                 if (retcode >= SS_SUCCESS) {
3053                         if (drbd_bitmap_io(device, &drbd_bmio_set_susp_al,
3054                                 "set_n_write from invalidate_peer",
3055                                 BM_LOCKED_SET_ALLOWED))
3056                                 retcode = ERR_IO_MD_DISK;
3057                 }
3058         } else
3059                 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_S));
3060         drbd_resume_io(device);
3061         mutex_unlock(&adm_ctx.resource->adm_mutex);
3062         put_ldev(device);
3063 out:
3064         drbd_adm_finish(&adm_ctx, info, retcode);
3065         return 0;
3066 }
3067
3068 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
3069 {
3070         struct drbd_config_context adm_ctx;
3071         enum drbd_ret_code retcode;
3072
3073         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3074         if (!adm_ctx.reply_skb)
3075                 return retcode;
3076         if (retcode != NO_ERROR)
3077                 goto out;
3078
3079         mutex_lock(&adm_ctx.resource->adm_mutex);
3080         if (drbd_request_state(adm_ctx.device, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
3081                 retcode = ERR_PAUSE_IS_SET;
3082         mutex_unlock(&adm_ctx.resource->adm_mutex);
3083 out:
3084         drbd_adm_finish(&adm_ctx, info, retcode);
3085         return 0;
3086 }
3087
3088 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
3089 {
3090         struct drbd_config_context adm_ctx;
3091         union drbd_dev_state s;
3092         enum drbd_ret_code retcode;
3093
3094         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3095         if (!adm_ctx.reply_skb)
3096                 return retcode;
3097         if (retcode != NO_ERROR)
3098                 goto out;
3099
3100         mutex_lock(&adm_ctx.resource->adm_mutex);
3101         if (drbd_request_state(adm_ctx.device, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
3102                 s = adm_ctx.device->state;
3103                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
3104                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
3105                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
3106                 } else {
3107                         retcode = ERR_PAUSE_IS_CLEAR;
3108                 }
3109         }
3110         mutex_unlock(&adm_ctx.resource->adm_mutex);
3111 out:
3112         drbd_adm_finish(&adm_ctx, info, retcode);
3113         return 0;
3114 }
3115
3116 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
3117 {
3118         return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
3119 }
3120
3121 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
3122 {
3123         struct drbd_config_context adm_ctx;
3124         struct drbd_device *device;
3125         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3126
3127         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3128         if (!adm_ctx.reply_skb)
3129                 return retcode;
3130         if (retcode != NO_ERROR)
3131                 goto out;
3132
3133         mutex_lock(&adm_ctx.resource->adm_mutex);
3134         device = adm_ctx.device;
3135         if (test_bit(NEW_CUR_UUID, &device->flags)) {
3136                 if (get_ldev_if_state(device, D_ATTACHING)) {
3137                         drbd_uuid_new_current(device);
3138                         put_ldev(device);
3139                 } else {
3140                         /* This is effectively a multi-stage "forced down".
3141                          * The NEW_CUR_UUID bit is supposedly only set, if we
3142                          * lost the replication connection, and are configured
3143                          * to freeze IO and wait for some fence-peer handler.
3144                          * So we still don't have a replication connection.
3145                          * And now we don't have a local disk either.  After
3146                          * resume, we will fail all pending and new IO, because
3147                          * we don't have any data anymore.  Which means we will
3148                          * eventually be able to terminate all users of this
3149                          * device, and then take it down.  By bumping the
3150                          * "effective" data uuid, we make sure that you really
3151                          * need to tear down before you reconfigure, we will
3152                          * the refuse to re-connect or re-attach (because no
3153                          * matching real data uuid exists).
3154                          */
3155                         u64 val;
3156                         get_random_bytes(&val, sizeof(u64));
3157                         drbd_set_ed_uuid(device, val);
3158                         drbd_warn(device, "Resumed without access to data; please tear down before attempting to re-configure.\n");
3159                 }
3160                 clear_bit(NEW_CUR_UUID, &device->flags);
3161         }
3162         drbd_suspend_io(device);
3163         retcode = drbd_request_state(device, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
3164         if (retcode == SS_SUCCESS) {
3165                 if (device->state.conn < C_CONNECTED)
3166                         tl_clear(first_peer_device(device)->connection);
3167                 if (device->state.disk == D_DISKLESS || device->state.disk == D_FAILED)
3168                         tl_restart(first_peer_device(device)->connection, FAIL_FROZEN_DISK_IO);
3169         }
3170         drbd_resume_io(device);
3171         mutex_unlock(&adm_ctx.resource->adm_mutex);
3172 out:
3173         drbd_adm_finish(&adm_ctx, info, retcode);
3174         return 0;
3175 }
3176
3177 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
3178 {
3179         return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
3180 }
3181
3182 static int nla_put_drbd_cfg_context(struct sk_buff *skb,
3183                                     struct drbd_resource *resource,
3184                                     struct drbd_connection *connection,
3185                                     struct drbd_device *device)
3186 {
3187         struct nlattr *nla;
3188         nla = nla_nest_start_noflag(skb, DRBD_NLA_CFG_CONTEXT);
3189         if (!nla)
3190                 goto nla_put_failure;
3191         if (device &&
3192             nla_put_u32(skb, T_ctx_volume, device->vnr))
3193                 goto nla_put_failure;
3194         if (nla_put_string(skb, T_ctx_resource_name, resource->name))
3195                 goto nla_put_failure;
3196         if (connection) {
3197                 if (connection->my_addr_len &&
3198                     nla_put(skb, T_ctx_my_addr, connection->my_addr_len, &connection->my_addr))
3199                         goto nla_put_failure;
3200                 if (connection->peer_addr_len &&
3201                     nla_put(skb, T_ctx_peer_addr, connection->peer_addr_len, &connection->peer_addr))
3202                         goto nla_put_failure;
3203         }
3204         nla_nest_end(skb, nla);
3205         return 0;
3206
3207 nla_put_failure:
3208         if (nla)
3209                 nla_nest_cancel(skb, nla);
3210         return -EMSGSIZE;
3211 }
3212
3213 /*
3214  * The generic netlink dump callbacks are called outside the genl_lock(), so
3215  * they cannot use the simple attribute parsing code which uses global
3216  * attribute tables.
3217  */
3218 static struct nlattr *find_cfg_context_attr(const struct nlmsghdr *nlh, int attr)
3219 {
3220         const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3221         const int maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
3222         struct nlattr *nla;
3223
3224         nla = nla_find(nlmsg_attrdata(nlh, hdrlen), nlmsg_attrlen(nlh, hdrlen),
3225                        DRBD_NLA_CFG_CONTEXT);
3226         if (!nla)
3227                 return NULL;
3228         return drbd_nla_find_nested(maxtype, nla, __nla_type(attr));
3229 }
3230
3231 static void resource_to_info(struct resource_info *, struct drbd_resource *);
3232
3233 int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb)
3234 {
3235         struct drbd_genlmsghdr *dh;
3236         struct drbd_resource *resource;
3237         struct resource_info resource_info;
3238         struct resource_statistics resource_statistics;
3239         int err;
3240
3241         rcu_read_lock();
3242         if (cb->args[0]) {
3243                 for_each_resource_rcu(resource, &drbd_resources)
3244                         if (resource == (struct drbd_resource *)cb->args[0])
3245                                 goto found_resource;
3246                 err = 0;  /* resource was probably deleted */
3247                 goto out;
3248         }
3249         resource = list_entry(&drbd_resources,
3250                               struct drbd_resource, resources);
3251
3252 found_resource:
3253         list_for_each_entry_continue_rcu(resource, &drbd_resources, resources) {
3254                 goto put_result;
3255         }
3256         err = 0;
3257         goto out;
3258
3259 put_result:
3260         dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3261                         cb->nlh->nlmsg_seq, &drbd_genl_family,
3262                         NLM_F_MULTI, DRBD_ADM_GET_RESOURCES);
3263         err = -ENOMEM;
3264         if (!dh)
3265                 goto out;
3266         dh->minor = -1U;
3267         dh->ret_code = NO_ERROR;
3268         err = nla_put_drbd_cfg_context(skb, resource, NULL, NULL);
3269         if (err)
3270                 goto out;
3271         err = res_opts_to_skb(skb, &resource->res_opts, !capable(CAP_SYS_ADMIN));
3272         if (err)
3273                 goto out;
3274         resource_to_info(&resource_info, resource);
3275         err = resource_info_to_skb(skb, &resource_info, !capable(CAP_SYS_ADMIN));
3276         if (err)
3277                 goto out;
3278         resource_statistics.res_stat_write_ordering = resource->write_ordering;
3279         err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
3280         if (err)
3281                 goto out;
3282         cb->args[0] = (long)resource;
3283         genlmsg_end(skb, dh);
3284         err = 0;
3285
3286 out:
3287         rcu_read_unlock();
3288         if (err)
3289                 return err;
3290         return skb->len;
3291 }
3292
3293 static void device_to_statistics(struct device_statistics *s,
3294                                  struct drbd_device *device)
3295 {
3296         memset(s, 0, sizeof(*s));
3297         s->dev_upper_blocked = !may_inc_ap_bio(device);
3298         if (get_ldev(device)) {
3299                 struct drbd_md *md = &device->ldev->md;
3300                 u64 *history_uuids = (u64 *)s->history_uuids;
3301                 int n;
3302
3303                 spin_lock_irq(&md->uuid_lock);
3304                 s->dev_current_uuid = md->uuid[UI_CURRENT];
3305                 BUILD_BUG_ON(sizeof(s->history_uuids) < UI_HISTORY_END - UI_HISTORY_START + 1);
3306                 for (n = 0; n < UI_HISTORY_END - UI_HISTORY_START + 1; n++)
3307                         history_uuids[n] = md->uuid[UI_HISTORY_START + n];
3308                 for (; n < HISTORY_UUIDS; n++)
3309                         history_uuids[n] = 0;
3310                 s->history_uuids_len = HISTORY_UUIDS;
3311                 spin_unlock_irq(&md->uuid_lock);
3312
3313                 s->dev_disk_flags = md->flags;
3314                 put_ldev(device);
3315         }
3316         s->dev_size = get_capacity(device->vdisk);
3317         s->dev_read = device->read_cnt;
3318         s->dev_write = device->writ_cnt;
3319         s->dev_al_writes = device->al_writ_cnt;
3320         s->dev_bm_writes = device->bm_writ_cnt;
3321         s->dev_upper_pending = atomic_read(&device->ap_bio_cnt);
3322         s->dev_lower_pending = atomic_read(&device->local_cnt);
3323         s->dev_al_suspended = test_bit(AL_SUSPENDED, &device->flags);
3324         s->dev_exposed_data_uuid = device->ed_uuid;
3325 }
3326
3327 static int put_resource_in_arg0(struct netlink_callback *cb, int holder_nr)
3328 {
3329         if (cb->args[0]) {
3330                 struct drbd_resource *resource =
3331                         (struct drbd_resource *)cb->args[0];
3332                 kref_put(&resource->kref, drbd_destroy_resource);
3333         }
3334
3335         return 0;
3336 }
3337
3338 int drbd_adm_dump_devices_done(struct netlink_callback *cb) {
3339         return put_resource_in_arg0(cb, 7);
3340 }
3341
3342 static void device_to_info(struct device_info *, struct drbd_device *);
3343
3344 int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb)
3345 {
3346         struct nlattr *resource_filter;
3347         struct drbd_resource *resource;
3348         struct drbd_device *device;
3349         int minor, err, retcode;
3350         struct drbd_genlmsghdr *dh;
3351         struct device_info device_info;
3352         struct device_statistics device_statistics;
3353         struct idr *idr_to_search;
3354
3355         resource = (struct drbd_resource *)cb->args[0];
3356         if (!cb->args[0] && !cb->args[1]) {
3357                 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3358                 if (resource_filter) {
3359                         retcode = ERR_RES_NOT_KNOWN;
3360                         resource = drbd_find_resource(nla_data(resource_filter));
3361                         if (!resource)
3362                                 goto put_result;
3363                         cb->args[0] = (long)resource;
3364                 }
3365         }
3366
3367         rcu_read_lock();
3368         minor = cb->args[1];
3369         idr_to_search = resource ? &resource->devices : &drbd_devices;
3370         device = idr_get_next(idr_to_search, &minor);
3371         if (!device) {
3372                 err = 0;
3373                 goto out;
3374         }
3375         idr_for_each_entry_continue(idr_to_search, device, minor) {
3376                 retcode = NO_ERROR;
3377                 goto put_result;  /* only one iteration */
3378         }
3379         err = 0;
3380         goto out;  /* no more devices */
3381
3382 put_result:
3383         dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3384                         cb->nlh->nlmsg_seq, &drbd_genl_family,
3385                         NLM_F_MULTI, DRBD_ADM_GET_DEVICES);
3386         err = -ENOMEM;
3387         if (!dh)
3388                 goto out;
3389         dh->ret_code = retcode;
3390         dh->minor = -1U;
3391         if (retcode == NO_ERROR) {
3392                 dh->minor = device->minor;
3393                 err = nla_put_drbd_cfg_context(skb, device->resource, NULL, device);
3394                 if (err)
3395                         goto out;
3396                 if (get_ldev(device)) {
3397                         struct disk_conf *disk_conf =
3398                                 rcu_dereference(device->ldev->disk_conf);
3399
3400                         err = disk_conf_to_skb(skb, disk_conf, !capable(CAP_SYS_ADMIN));
3401                         put_ldev(device);
3402                         if (err)
3403                                 goto out;
3404                 }
3405                 device_to_info(&device_info, device);
3406                 err = device_info_to_skb(skb, &device_info, !capable(CAP_SYS_ADMIN));
3407                 if (err)
3408                         goto out;
3409
3410                 device_to_statistics(&device_statistics, device);
3411                 err = device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
3412                 if (err)
3413                         goto out;
3414                 cb->args[1] = minor + 1;
3415         }
3416         genlmsg_end(skb, dh);
3417         err = 0;
3418
3419 out:
3420         rcu_read_unlock();
3421         if (err)
3422                 return err;
3423         return skb->len;
3424 }
3425
3426 int drbd_adm_dump_connections_done(struct netlink_callback *cb)
3427 {
3428         return put_resource_in_arg0(cb, 6);
3429 }
3430
3431 enum { SINGLE_RESOURCE, ITERATE_RESOURCES };
3432
3433 int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb)
3434 {
3435         struct nlattr *resource_filter;
3436         struct drbd_resource *resource = NULL, *next_resource;
3437         struct drbd_connection *connection;
3438         int err = 0, retcode;
3439         struct drbd_genlmsghdr *dh;
3440         struct connection_info connection_info;
3441         struct connection_statistics connection_statistics;
3442
3443         rcu_read_lock();
3444         resource = (struct drbd_resource *)cb->args[0];
3445         if (!cb->args[0]) {
3446                 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3447                 if (resource_filter) {
3448                         retcode = ERR_RES_NOT_KNOWN;
3449                         resource = drbd_find_resource(nla_data(resource_filter));
3450                         if (!resource)
3451                                 goto put_result;
3452                         cb->args[0] = (long)resource;
3453                         cb->args[1] = SINGLE_RESOURCE;
3454                 }
3455         }
3456         if (!resource) {
3457                 if (list_empty(&drbd_resources))
3458                         goto out;
3459                 resource = list_first_entry(&drbd_resources, struct drbd_resource, resources);
3460                 kref_get(&resource->kref);
3461                 cb->args[0] = (long)resource;
3462                 cb->args[1] = ITERATE_RESOURCES;
3463         }
3464
3465     next_resource:
3466         rcu_read_unlock();
3467         mutex_lock(&resource->conf_update);
3468         rcu_read_lock();
3469         if (cb->args[2]) {
3470                 for_each_connection_rcu(connection, resource)
3471                         if (connection == (struct drbd_connection *)cb->args[2])
3472                                 goto found_connection;
3473                 /* connection was probably deleted */
3474                 goto no_more_connections;
3475         }
3476         connection = list_entry(&resource->connections, struct drbd_connection, connections);
3477
3478 found_connection:
3479         list_for_each_entry_continue_rcu(connection, &resource->connections, connections) {
3480                 if (!has_net_conf(connection))
3481                         continue;
3482                 retcode = NO_ERROR;
3483                 goto put_result;  /* only one iteration */
3484         }
3485
3486 no_more_connections:
3487         if (cb->args[1] == ITERATE_RESOURCES) {
3488                 for_each_resource_rcu(next_resource, &drbd_resources) {
3489                         if (next_resource == resource)
3490                                 goto found_resource;
3491                 }
3492                 /* resource was probably deleted */
3493         }
3494         goto out;
3495
3496 found_resource:
3497         list_for_each_entry_continue_rcu(next_resource, &drbd_resources, resources) {
3498                 mutex_unlock(&resource->conf_update);
3499                 kref_put(&resource->kref, drbd_destroy_resource);
3500                 resource = next_resource;
3501                 kref_get(&resource->kref);
3502                 cb->args[0] = (long)resource;
3503                 cb->args[2] = 0;
3504                 goto next_resource;
3505         }
3506         goto out;  /* no more resources */
3507
3508 put_result:
3509         dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3510                         cb->nlh->nlmsg_seq, &drbd_genl_family,
3511                         NLM_F_MULTI, DRBD_ADM_GET_CONNECTIONS);
3512         err = -ENOMEM;
3513         if (!dh)
3514                 goto out;
3515         dh->ret_code = retcode;
3516         dh->minor = -1U;
3517         if (retcode == NO_ERROR) {
3518                 struct net_conf *net_conf;
3519
3520                 err = nla_put_drbd_cfg_context(skb, resource, connection, NULL);
3521                 if (err)
3522                         goto out;
3523                 net_conf = rcu_dereference(connection->net_conf);
3524                 if (net_conf) {
3525                         err = net_conf_to_skb(skb, net_conf, !capable(CAP_SYS_ADMIN));
3526                         if (err)
3527                                 goto out;
3528                 }
3529                 connection_to_info(&connection_info, connection);
3530                 err = connection_info_to_skb(skb, &connection_info, !capable(CAP_SYS_ADMIN));
3531                 if (err)
3532                         goto out;
3533                 connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
3534                 err = connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
3535                 if (err)
3536                         goto out;
3537                 cb->args[2] = (long)connection;
3538         }
3539         genlmsg_end(skb, dh);
3540         err = 0;
3541
3542 out:
3543         rcu_read_unlock();
3544         if (resource)
3545                 mutex_unlock(&resource->conf_update);
3546         if (err)
3547                 return err;
3548         return skb->len;
3549 }
3550
3551 enum mdf_peer_flag {
3552         MDF_PEER_CONNECTED =    1 << 0,
3553         MDF_PEER_OUTDATED =     1 << 1,
3554         MDF_PEER_FENCING =      1 << 2,
3555         MDF_PEER_FULL_SYNC =    1 << 3,
3556 };
3557
3558 static void peer_device_to_statistics(struct peer_device_statistics *s,
3559                                       struct drbd_peer_device *peer_device)
3560 {
3561         struct drbd_device *device = peer_device->device;
3562
3563         memset(s, 0, sizeof(*s));
3564         s->peer_dev_received = device->recv_cnt;
3565         s->peer_dev_sent = device->send_cnt;
3566         s->peer_dev_pending = atomic_read(&device->ap_pending_cnt) +
3567                               atomic_read(&device->rs_pending_cnt);
3568         s->peer_dev_unacked = atomic_read(&device->unacked_cnt);
3569         s->peer_dev_out_of_sync = drbd_bm_total_weight(device) << (BM_BLOCK_SHIFT - 9);
3570         s->peer_dev_resync_failed = device->rs_failed << (BM_BLOCK_SHIFT - 9);
3571         if (get_ldev(device)) {
3572                 struct drbd_md *md = &device->ldev->md;
3573
3574                 spin_lock_irq(&md->uuid_lock);
3575                 s->peer_dev_bitmap_uuid = md->uuid[UI_BITMAP];
3576                 spin_unlock_irq(&md->uuid_lock);
3577                 s->peer_dev_flags =
3578                         (drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND) ?
3579                                 MDF_PEER_CONNECTED : 0) +
3580                         (drbd_md_test_flag(device->ldev, MDF_CONSISTENT) &&
3581                          !drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE) ?
3582                                 MDF_PEER_OUTDATED : 0) +
3583                         /* FIXME: MDF_PEER_FENCING? */
3584                         (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ?
3585                                 MDF_PEER_FULL_SYNC : 0);
3586                 put_ldev(device);
3587         }
3588 }
3589
3590 int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb)
3591 {
3592         return put_resource_in_arg0(cb, 9);
3593 }
3594
3595 int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb)
3596 {
3597         struct nlattr *resource_filter;
3598         struct drbd_resource *resource;
3599         struct drbd_device *device;
3600         struct drbd_peer_device *peer_device = NULL;
3601         int minor, err, retcode;
3602         struct drbd_genlmsghdr *dh;
3603         struct idr *idr_to_search;
3604
3605         resource = (struct drbd_resource *)cb->args[0];
3606         if (!cb->args[0] && !cb->args[1]) {
3607                 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3608                 if (resource_filter) {
3609                         retcode = ERR_RES_NOT_KNOWN;
3610                         resource = drbd_find_resource(nla_data(resource_filter));
3611                         if (!resource)
3612                                 goto put_result;
3613                 }
3614                 cb->args[0] = (long)resource;
3615         }
3616
3617         rcu_read_lock();
3618         minor = cb->args[1];
3619         idr_to_search = resource ? &resource->devices : &drbd_devices;
3620         device = idr_find(idr_to_search, minor);
3621         if (!device) {
3622 next_device:
3623                 minor++;
3624                 cb->args[2] = 0;
3625                 device = idr_get_next(idr_to_search, &minor);
3626                 if (!device) {
3627                         err = 0;
3628                         goto out;
3629                 }
3630         }
3631         if (cb->args[2]) {
3632                 for_each_peer_device(peer_device, device)
3633                         if (peer_device == (struct drbd_peer_device *)cb->args[2])
3634                                 goto found_peer_device;
3635                 /* peer device was probably deleted */
3636                 goto next_device;
3637         }
3638         /* Make peer_device point to the list head (not the first entry). */
3639         peer_device = list_entry(&device->peer_devices, struct drbd_peer_device, peer_devices);
3640
3641 found_peer_device:
3642         list_for_each_entry_continue_rcu(peer_device, &device->peer_devices, peer_devices) {
3643                 if (!has_net_conf(peer_device->connection))
3644                         continue;
3645                 retcode = NO_ERROR;
3646                 goto put_result;  /* only one iteration */
3647         }
3648         goto next_device;
3649
3650 put_result:
3651         dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3652                         cb->nlh->nlmsg_seq, &drbd_genl_family,
3653                         NLM_F_MULTI, DRBD_ADM_GET_PEER_DEVICES);
3654         err = -ENOMEM;
3655         if (!dh)
3656                 goto out;
3657         dh->ret_code = retcode;
3658         dh->minor = -1U;
3659         if (retcode == NO_ERROR) {
3660                 struct peer_device_info peer_device_info;
3661                 struct peer_device_statistics peer_device_statistics;
3662
3663                 dh->minor = minor;
3664                 err = nla_put_drbd_cfg_context(skb, device->resource, peer_device->connection, device);
3665                 if (err)
3666                         goto out;
3667                 peer_device_to_info(&peer_device_info, peer_device);
3668                 err = peer_device_info_to_skb(skb, &peer_device_info, !capable(CAP_SYS_ADMIN));
3669                 if (err)
3670                         goto out;
3671                 peer_device_to_statistics(&peer_device_statistics, peer_device);
3672                 err = peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
3673                 if (err)
3674                         goto out;
3675                 cb->args[1] = minor;
3676                 cb->args[2] = (long)peer_device;
3677         }
3678         genlmsg_end(skb, dh);
3679         err = 0;
3680
3681 out:
3682         rcu_read_unlock();
3683         if (err)
3684                 return err;
3685         return skb->len;
3686 }
3687 /*
3688  * Return the connection of @resource if @resource has exactly one connection.
3689  */
3690 static struct drbd_connection *the_only_connection(struct drbd_resource *resource)
3691 {
3692         struct list_head *connections = &resource->connections;
3693
3694         if (list_empty(connections) || connections->next->next != connections)
3695                 return NULL;
3696         return list_first_entry(&resource->connections, struct drbd_connection, connections);
3697 }
3698
3699 static int nla_put_status_info(struct sk_buff *skb, struct drbd_device *device,
3700                 const struct sib_info *sib)
3701 {
3702         struct drbd_resource *resource = device->resource;
3703         struct state_info *si = NULL; /* for sizeof(si->member); */
3704         struct nlattr *nla;
3705         int got_ldev;
3706         int err = 0;
3707         int exclude_sensitive;
3708
3709         /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
3710          * to.  So we better exclude_sensitive information.
3711          *
3712          * If sib == NULL, this is drbd_adm_get_status, executed synchronously
3713          * in the context of the requesting user process. Exclude sensitive
3714          * information, unless current has superuser.
3715          *
3716          * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
3717          * relies on the current implementation of netlink_dump(), which
3718          * executes the dump callback successively from netlink_recvmsg(),
3719          * always in the context of the receiving process */
3720         exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
3721
3722         got_ldev = get_ldev(device);
3723
3724         /* We need to add connection name and volume number information still.
3725          * Minor number is in drbd_genlmsghdr. */
3726         if (nla_put_drbd_cfg_context(skb, resource, the_only_connection(resource), device))
3727                 goto nla_put_failure;
3728
3729         if (res_opts_to_skb(skb, &device->resource->res_opts, exclude_sensitive))
3730                 goto nla_put_failure;
3731
3732         rcu_read_lock();
3733         if (got_ldev) {
3734                 struct disk_conf *disk_conf;
3735
3736                 disk_conf = rcu_dereference(device->ldev->disk_conf);
3737                 err = disk_conf_to_skb(skb, disk_conf, exclude_sensitive);
3738         }
3739         if (!err) {
3740                 struct net_conf *nc;
3741
3742                 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3743                 if (nc)
3744                         err = net_conf_to_skb(skb, nc, exclude_sensitive);
3745         }
3746         rcu_read_unlock();
3747         if (err)
3748                 goto nla_put_failure;
3749
3750         nla = nla_nest_start_noflag(skb, DRBD_NLA_STATE_INFO);
3751         if (!nla)
3752                 goto nla_put_failure;
3753         if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) ||
3754             nla_put_u32(skb, T_current_state, device->state.i) ||
3755             nla_put_u64_0pad(skb, T_ed_uuid, device->ed_uuid) ||
3756             nla_put_u64_0pad(skb, T_capacity, get_capacity(device->vdisk)) ||
3757             nla_put_u64_0pad(skb, T_send_cnt, device->send_cnt) ||
3758             nla_put_u64_0pad(skb, T_recv_cnt, device->recv_cnt) ||
3759             nla_put_u64_0pad(skb, T_read_cnt, device->read_cnt) ||
3760             nla_put_u64_0pad(skb, T_writ_cnt, device->writ_cnt) ||
3761             nla_put_u64_0pad(skb, T_al_writ_cnt, device->al_writ_cnt) ||
3762             nla_put_u64_0pad(skb, T_bm_writ_cnt, device->bm_writ_cnt) ||
3763             nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&device->ap_bio_cnt)) ||
3764             nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&device->ap_pending_cnt)) ||
3765             nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&device->rs_pending_cnt)))
3766                 goto nla_put_failure;
3767
3768         if (got_ldev) {
3769                 int err;
3770
3771                 spin_lock_irq(&device->ldev->md.uuid_lock);
3772                 err = nla_put(skb, T_uuids, sizeof(si->uuids), device->ldev->md.uuid);
3773                 spin_unlock_irq(&device->ldev->md.uuid_lock);
3774
3775                 if (err)
3776                         goto nla_put_failure;
3777
3778                 if (nla_put_u32(skb, T_disk_flags, device->ldev->md.flags) ||
3779                     nla_put_u64_0pad(skb, T_bits_total, drbd_bm_bits(device)) ||
3780                     nla_put_u64_0pad(skb, T_bits_oos,
3781                                      drbd_bm_total_weight(device)))
3782                         goto nla_put_failure;
3783                 if (C_SYNC_SOURCE <= device->state.conn &&
3784                     C_PAUSED_SYNC_T >= device->state.conn) {
3785                         if (nla_put_u64_0pad(skb, T_bits_rs_total,
3786                                              device->rs_total) ||
3787                             nla_put_u64_0pad(skb, T_bits_rs_failed,
3788                                              device->rs_failed))
3789                                 goto nla_put_failure;
3790                 }
3791         }
3792
3793         if (sib) {
3794                 switch(sib->sib_reason) {
3795                 case SIB_SYNC_PROGRESS:
3796                 case SIB_GET_STATUS_REPLY:
3797                         break;
3798                 case SIB_STATE_CHANGE:
3799                         if (nla_put_u32(skb, T_prev_state, sib->os.i) ||
3800                             nla_put_u32(skb, T_new_state, sib->ns.i))
3801                                 goto nla_put_failure;
3802                         break;
3803                 case SIB_HELPER_POST:
3804                         if (nla_put_u32(skb, T_helper_exit_code,
3805                                         sib->helper_exit_code))
3806                                 goto nla_put_failure;
3807                         fallthrough;
3808                 case SIB_HELPER_PRE:
3809                         if (nla_put_string(skb, T_helper, sib->helper_name))
3810                                 goto nla_put_failure;
3811                         break;
3812                 }
3813         }
3814         nla_nest_end(skb, nla);
3815
3816         if (0)
3817 nla_put_failure:
3818                 err = -EMSGSIZE;
3819         if (got_ldev)
3820                 put_ldev(device);
3821         return err;
3822 }
3823
3824 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
3825 {
3826         struct drbd_config_context adm_ctx;
3827         enum drbd_ret_code retcode;
3828         int err;
3829
3830         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3831         if (!adm_ctx.reply_skb)
3832                 return retcode;
3833         if (retcode != NO_ERROR)
3834                 goto out;
3835
3836         err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.device, NULL);
3837         if (err) {
3838                 nlmsg_free(adm_ctx.reply_skb);
3839                 return err;
3840         }
3841 out:
3842         drbd_adm_finish(&adm_ctx, info, retcode);
3843         return 0;
3844 }
3845
3846 static int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
3847 {
3848         struct drbd_device *device;
3849         struct drbd_genlmsghdr *dh;
3850         struct drbd_resource *pos = (struct drbd_resource *)cb->args[0];
3851         struct drbd_resource *resource = NULL;
3852         struct drbd_resource *tmp;
3853         unsigned volume = cb->args[1];
3854
3855         /* Open coded, deferred, iteration:
3856          * for_each_resource_safe(resource, tmp, &drbd_resources) {
3857          *      connection = "first connection of resource or undefined";
3858          *      idr_for_each_entry(&resource->devices, device, i) {
3859          *        ...
3860          *      }
3861          * }
3862          * where resource is cb->args[0];
3863          * and i is cb->args[1];
3864          *
3865          * cb->args[2] indicates if we shall loop over all resources,
3866          * or just dump all volumes of a single resource.
3867          *
3868          * This may miss entries inserted after this dump started,
3869          * or entries deleted before they are reached.
3870          *
3871          * We need to make sure the device won't disappear while
3872          * we are looking at it, and revalidate our iterators
3873          * on each iteration.
3874          */
3875
3876         /* synchronize with conn_create()/drbd_destroy_connection() */
3877         rcu_read_lock();
3878         /* revalidate iterator position */
3879         for_each_resource_rcu(tmp, &drbd_resources) {
3880                 if (pos == NULL) {
3881                         /* first iteration */
3882                         pos = tmp;
3883                         resource = pos;
3884                         break;
3885                 }
3886                 if (tmp == pos) {
3887                         resource = pos;
3888                         break;
3889                 }
3890         }
3891         if (resource) {
3892 next_resource:
3893                 device = idr_get_next(&resource->devices, &volume);
3894                 if (!device) {
3895                         /* No more volumes to dump on this resource.
3896                          * Advance resource iterator. */
3897                         pos = list_entry_rcu(resource->resources.next,
3898                                              struct drbd_resource, resources);
3899                         /* Did we dump any volume of this resource yet? */
3900                         if (volume != 0) {
3901                                 /* If we reached the end of the list,
3902                                  * or only a single resource dump was requested,
3903                                  * we are done. */
3904                                 if (&pos->resources == &drbd_resources || cb->args[2])
3905                                         goto out;
3906                                 volume = 0;
3907                                 resource = pos;
3908                                 goto next_resource;
3909                         }
3910                 }
3911
3912                 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3913                                 cb->nlh->nlmsg_seq, &drbd_genl_family,
3914                                 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
3915                 if (!dh)
3916                         goto out;
3917
3918                 if (!device) {
3919                         /* This is a connection without a single volume.
3920                          * Suprisingly enough, it may have a network
3921                          * configuration. */
3922                         struct drbd_connection *connection;
3923
3924                         dh->minor = -1U;
3925                         dh->ret_code = NO_ERROR;
3926                         connection = the_only_connection(resource);
3927                         if (nla_put_drbd_cfg_context(skb, resource, connection, NULL))
3928                                 goto cancel;
3929                         if (connection) {
3930                                 struct net_conf *nc;
3931
3932                                 nc = rcu_dereference(connection->net_conf);
3933                                 if (nc && net_conf_to_skb(skb, nc, 1) != 0)
3934                                         goto cancel;
3935                         }
3936                         goto done;
3937                 }
3938
3939                 D_ASSERT(device, device->vnr == volume);
3940                 D_ASSERT(device, device->resource == resource);
3941
3942                 dh->minor = device_to_minor(device);
3943                 dh->ret_code = NO_ERROR;
3944
3945                 if (nla_put_status_info(skb, device, NULL)) {
3946 cancel:
3947                         genlmsg_cancel(skb, dh);
3948                         goto out;
3949                 }
3950 done:
3951                 genlmsg_end(skb, dh);
3952         }
3953
3954 out:
3955         rcu_read_unlock();
3956         /* where to start the next iteration */
3957         cb->args[0] = (long)pos;
3958         cb->args[1] = (pos == resource) ? volume + 1 : 0;
3959
3960         /* No more resources/volumes/minors found results in an empty skb.
3961          * Which will terminate the dump. */
3962         return skb->len;
3963 }
3964
3965 /*
3966  * Request status of all resources, or of all volumes within a single resource.
3967  *
3968  * This is a dump, as the answer may not fit in a single reply skb otherwise.
3969  * Which means we cannot use the family->attrbuf or other such members, because
3970  * dump is NOT protected by the genl_lock().  During dump, we only have access
3971  * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
3972  *
3973  * Once things are setup properly, we call into get_one_status().
3974  */
3975 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
3976 {
3977         const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3978         struct nlattr *nla;
3979         const char *resource_name;
3980         struct drbd_resource *resource;
3981         int maxtype;
3982
3983         /* Is this a followup call? */
3984         if (cb->args[0]) {
3985                 /* ... of a single resource dump,
3986                  * and the resource iterator has been advanced already? */
3987                 if (cb->args[2] && cb->args[2] != cb->args[0])
3988                         return 0; /* DONE. */
3989                 goto dump;
3990         }
3991
3992         /* First call (from netlink_dump_start).  We need to figure out
3993          * which resource(s) the user wants us to dump. */
3994         nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
3995                         nlmsg_attrlen(cb->nlh, hdrlen),
3996                         DRBD_NLA_CFG_CONTEXT);
3997
3998         /* No explicit context given.  Dump all. */
3999         if (!nla)
4000                 goto dump;
4001         maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
4002         nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
4003         if (IS_ERR(nla))
4004                 return PTR_ERR(nla);
4005         /* context given, but no name present? */
4006         if (!nla)
4007                 return -EINVAL;
4008         resource_name = nla_data(nla);
4009         if (!*resource_name)
4010                 return -ENODEV;
4011         resource = drbd_find_resource(resource_name);
4012         if (!resource)
4013                 return -ENODEV;
4014
4015         kref_put(&resource->kref, drbd_destroy_resource); /* get_one_status() revalidates the resource */
4016
4017         /* prime iterators, and set "filter" mode mark:
4018          * only dump this connection. */
4019         cb->args[0] = (long)resource;
4020         /* cb->args[1] = 0; passed in this way. */
4021         cb->args[2] = (long)resource;
4022
4023 dump:
4024         return get_one_status(skb, cb);
4025 }
4026
4027 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
4028 {
4029         struct drbd_config_context adm_ctx;
4030         enum drbd_ret_code retcode;
4031         struct timeout_parms tp;
4032         int err;
4033
4034         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4035         if (!adm_ctx.reply_skb)
4036                 return retcode;
4037         if (retcode != NO_ERROR)
4038                 goto out;
4039
4040         tp.timeout_type =
4041                 adm_ctx.device->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
4042                 test_bit(USE_DEGR_WFC_T, &adm_ctx.device->flags) ? UT_DEGRADED :
4043                 UT_DEFAULT;
4044
4045         err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
4046         if (err) {
4047                 nlmsg_free(adm_ctx.reply_skb);
4048                 return err;
4049         }
4050 out:
4051         drbd_adm_finish(&adm_ctx, info, retcode);
4052         return 0;
4053 }
4054
4055 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
4056 {
4057         struct drbd_config_context adm_ctx;
4058         struct drbd_device *device;
4059         enum drbd_ret_code retcode;
4060         struct start_ov_parms parms;
4061
4062         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4063         if (!adm_ctx.reply_skb)
4064                 return retcode;
4065         if (retcode != NO_ERROR)
4066                 goto out;
4067
4068         device = adm_ctx.device;
4069
4070         /* resume from last known position, if possible */
4071         parms.ov_start_sector = device->ov_start_sector;
4072         parms.ov_stop_sector = ULLONG_MAX;
4073         if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
4074                 int err = start_ov_parms_from_attrs(&parms, info);
4075                 if (err) {
4076                         retcode = ERR_MANDATORY_TAG;
4077                         drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4078                         goto out;
4079                 }
4080         }
4081         mutex_lock(&adm_ctx.resource->adm_mutex);
4082
4083         /* w_make_ov_request expects position to be aligned */
4084         device->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1);
4085         device->ov_stop_sector = parms.ov_stop_sector;
4086
4087         /* If there is still bitmap IO pending, e.g. previous resync or verify
4088          * just being finished, wait for it before requesting a new resync. */
4089         drbd_suspend_io(device);
4090         wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
4091         retcode = drbd_request_state(device, NS(conn, C_VERIFY_S));
4092         drbd_resume_io(device);
4093
4094         mutex_unlock(&adm_ctx.resource->adm_mutex);
4095 out:
4096         drbd_adm_finish(&adm_ctx, info, retcode);
4097         return 0;
4098 }
4099
4100
4101 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
4102 {
4103         struct drbd_config_context adm_ctx;
4104         struct drbd_device *device;
4105         enum drbd_ret_code retcode;
4106         int skip_initial_sync = 0;
4107         int err;
4108         struct new_c_uuid_parms args;
4109
4110         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4111         if (!adm_ctx.reply_skb)
4112                 return retcode;
4113         if (retcode != NO_ERROR)
4114                 goto out_nolock;
4115
4116         device = adm_ctx.device;
4117         memset(&args, 0, sizeof(args));
4118         if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
4119                 err = new_c_uuid_parms_from_attrs(&args, info);
4120                 if (err) {
4121                         retcode = ERR_MANDATORY_TAG;
4122                         drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4123                         goto out_nolock;
4124                 }
4125         }
4126
4127         mutex_lock(&adm_ctx.resource->adm_mutex);
4128         mutex_lock(device->state_mutex); /* Protects us against serialized state changes. */
4129
4130         if (!get_ldev(device)) {
4131                 retcode = ERR_NO_DISK;
4132                 goto out;
4133         }
4134
4135         /* this is "skip initial sync", assume to be clean */
4136         if (device->state.conn == C_CONNECTED &&
4137             first_peer_device(device)->connection->agreed_pro_version >= 90 &&
4138             device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
4139                 drbd_info(device, "Preparing to skip initial sync\n");
4140                 skip_initial_sync = 1;
4141         } else if (device->state.conn != C_STANDALONE) {
4142                 retcode = ERR_CONNECTED;
4143                 goto out_dec;
4144         }
4145
4146         drbd_uuid_set(device, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
4147         drbd_uuid_new_current(device); /* New current, previous to UI_BITMAP */
4148
4149         if (args.clear_bm) {
4150                 err = drbd_bitmap_io(device, &drbd_bmio_clear_n_write,
4151                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
4152                 if (err) {
4153                         drbd_err(device, "Writing bitmap failed with %d\n", err);
4154                         retcode = ERR_IO_MD_DISK;
4155                 }
4156                 if (skip_initial_sync) {
4157                         drbd_send_uuids_skip_initial_sync(first_peer_device(device));
4158                         _drbd_uuid_set(device, UI_BITMAP, 0);
4159                         drbd_print_uuids(device, "cleared bitmap UUID");
4160                         spin_lock_irq(&device->resource->req_lock);
4161                         _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
4162                                         CS_VERBOSE, NULL);
4163                         spin_unlock_irq(&device->resource->req_lock);
4164                 }
4165         }
4166
4167         drbd_md_sync(device);
4168 out_dec:
4169         put_ldev(device);
4170 out:
4171         mutex_unlock(device->state_mutex);
4172         mutex_unlock(&adm_ctx.resource->adm_mutex);
4173 out_nolock:
4174         drbd_adm_finish(&adm_ctx, info, retcode);
4175         return 0;
4176 }
4177
4178 static enum drbd_ret_code
4179 drbd_check_resource_name(struct drbd_config_context *adm_ctx)
4180 {
4181         const char *name = adm_ctx->resource_name;
4182         if (!name || !name[0]) {
4183                 drbd_msg_put_info(adm_ctx->reply_skb, "resource name missing");
4184                 return ERR_MANDATORY_TAG;
4185         }
4186         /* if we want to use these in sysfs/configfs/debugfs some day,
4187          * we must not allow slashes */
4188         if (strchr(name, '/')) {
4189                 drbd_msg_put_info(adm_ctx->reply_skb, "invalid resource name");
4190                 return ERR_INVALID_REQUEST;
4191         }
4192         return NO_ERROR;
4193 }
4194
4195 static void resource_to_info(struct resource_info *info,
4196                              struct drbd_resource *resource)
4197 {
4198         info->res_role = conn_highest_role(first_connection(resource));
4199         info->res_susp = resource->susp;
4200         info->res_susp_nod = resource->susp_nod;
4201         info->res_susp_fen = resource->susp_fen;
4202 }
4203
4204 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
4205 {
4206         struct drbd_connection *connection;
4207         struct drbd_config_context adm_ctx;
4208         enum drbd_ret_code retcode;
4209         struct res_opts res_opts;
4210         int err;
4211
4212         retcode = drbd_adm_prepare(&adm_ctx, skb, info, 0);
4213         if (!adm_ctx.reply_skb)
4214                 return retcode;
4215         if (retcode != NO_ERROR)
4216                 goto out;
4217
4218         set_res_opts_defaults(&res_opts);
4219         err = res_opts_from_attrs(&res_opts, info);
4220         if (err && err != -ENOMSG) {
4221                 retcode = ERR_MANDATORY_TAG;
4222                 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4223                 goto out;
4224         }
4225
4226         retcode = drbd_check_resource_name(&adm_ctx);
4227         if (retcode != NO_ERROR)
4228                 goto out;
4229
4230         if (adm_ctx.resource) {
4231                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
4232                         retcode = ERR_INVALID_REQUEST;
4233                         drbd_msg_put_info(adm_ctx.reply_skb, "resource exists");
4234                 }
4235                 /* else: still NO_ERROR */
4236                 goto out;
4237         }
4238
4239         /* not yet safe for genl_family.parallel_ops */
4240         mutex_lock(&resources_mutex);
4241         connection = conn_create(adm_ctx.resource_name, &res_opts);
4242         mutex_unlock(&resources_mutex);
4243
4244         if (connection) {
4245                 struct resource_info resource_info;
4246
4247                 mutex_lock(&notification_mutex);
4248                 resource_to_info(&resource_info, connection->resource);
4249                 notify_resource_state(NULL, 0, connection->resource,
4250                                       &resource_info, NOTIFY_CREATE);
4251                 mutex_unlock(&notification_mutex);
4252         } else
4253                 retcode = ERR_NOMEM;
4254
4255 out:
4256         drbd_adm_finish(&adm_ctx, info, retcode);
4257         return 0;
4258 }
4259
4260 static void device_to_info(struct device_info *info,
4261                            struct drbd_device *device)
4262 {
4263         info->dev_disk_state = device->state.disk;
4264 }
4265
4266
4267 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info)
4268 {
4269         struct drbd_config_context adm_ctx;
4270         struct drbd_genlmsghdr *dh = info->userhdr;
4271         enum drbd_ret_code retcode;
4272
4273         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4274         if (!adm_ctx.reply_skb)
4275                 return retcode;
4276         if (retcode != NO_ERROR)
4277                 goto out;
4278
4279         if (dh->minor > MINORMASK) {
4280                 drbd_msg_put_info(adm_ctx.reply_skb, "requested minor out of range");
4281                 retcode = ERR_INVALID_REQUEST;
4282                 goto out;
4283         }
4284         if (adm_ctx.volume > DRBD_VOLUME_MAX) {
4285                 drbd_msg_put_info(adm_ctx.reply_skb, "requested volume id out of range");
4286                 retcode = ERR_INVALID_REQUEST;
4287                 goto out;
4288         }
4289
4290         /* drbd_adm_prepare made sure already
4291          * that first_peer_device(device)->connection and device->vnr match the request. */
4292         if (adm_ctx.device) {
4293                 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
4294                         retcode = ERR_MINOR_OR_VOLUME_EXISTS;
4295                 /* else: still NO_ERROR */
4296                 goto out;
4297         }
4298
4299         mutex_lock(&adm_ctx.resource->adm_mutex);
4300         retcode = drbd_create_device(&adm_ctx, dh->minor);
4301         if (retcode == NO_ERROR) {
4302                 struct drbd_device *device;
4303                 struct drbd_peer_device *peer_device;
4304                 struct device_info info;
4305                 unsigned int peer_devices = 0;
4306                 enum drbd_notification_type flags;
4307
4308                 device = minor_to_device(dh->minor);
4309                 for_each_peer_device(peer_device, device) {
4310                         if (!has_net_conf(peer_device->connection))
4311                                 continue;
4312                         peer_devices++;
4313                 }
4314
4315                 device_to_info(&info, device);
4316                 mutex_lock(&notification_mutex);
4317                 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4318                 notify_device_state(NULL, 0, device, &info, NOTIFY_CREATE | flags);
4319                 for_each_peer_device(peer_device, device) {
4320                         struct peer_device_info peer_device_info;
4321
4322                         if (!has_net_conf(peer_device->connection))
4323                                 continue;
4324                         peer_device_to_info(&peer_device_info, peer_device);
4325                         flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4326                         notify_peer_device_state(NULL, 0, peer_device, &peer_device_info,
4327                                                  NOTIFY_CREATE | flags);
4328                 }
4329                 mutex_unlock(&notification_mutex);
4330         }
4331         mutex_unlock(&adm_ctx.resource->adm_mutex);
4332 out:
4333         drbd_adm_finish(&adm_ctx, info, retcode);
4334         return 0;
4335 }
4336
4337 static enum drbd_ret_code adm_del_minor(struct drbd_device *device)
4338 {
4339         struct drbd_peer_device *peer_device;
4340
4341         if (device->state.disk == D_DISKLESS &&
4342             /* no need to be device->state.conn == C_STANDALONE &&
4343              * we may want to delete a minor from a live replication group.
4344              */
4345             device->state.role == R_SECONDARY) {
4346                 struct drbd_connection *connection =
4347                         first_connection(device->resource);
4348
4349                 _drbd_request_state(device, NS(conn, C_WF_REPORT_PARAMS),
4350                                     CS_VERBOSE + CS_WAIT_COMPLETE);
4351
4352                 /* If the state engine hasn't stopped the sender thread yet, we
4353                  * need to flush the sender work queue before generating the
4354                  * DESTROY events here. */
4355                 if (get_t_state(&connection->worker) == RUNNING)
4356                         drbd_flush_workqueue(&connection->sender_work);
4357
4358                 mutex_lock(&notification_mutex);
4359                 for_each_peer_device(peer_device, device) {
4360                         if (!has_net_conf(peer_device->connection))
4361                                 continue;
4362                         notify_peer_device_state(NULL, 0, peer_device, NULL,
4363                                                  NOTIFY_DESTROY | NOTIFY_CONTINUES);
4364                 }
4365                 notify_device_state(NULL, 0, device, NULL, NOTIFY_DESTROY);
4366                 mutex_unlock(&notification_mutex);
4367
4368                 drbd_delete_device(device);
4369                 return NO_ERROR;
4370         } else
4371                 return ERR_MINOR_CONFIGURED;
4372 }
4373
4374 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info)
4375 {
4376         struct drbd_config_context adm_ctx;
4377         enum drbd_ret_code retcode;
4378
4379         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4380         if (!adm_ctx.reply_skb)
4381                 return retcode;
4382         if (retcode != NO_ERROR)
4383                 goto out;
4384
4385         mutex_lock(&adm_ctx.resource->adm_mutex);
4386         retcode = adm_del_minor(adm_ctx.device);
4387         mutex_unlock(&adm_ctx.resource->adm_mutex);
4388 out:
4389         drbd_adm_finish(&adm_ctx, info, retcode);
4390         return 0;
4391 }
4392
4393 static int adm_del_resource(struct drbd_resource *resource)
4394 {
4395         struct drbd_connection *connection;
4396
4397         for_each_connection(connection, resource) {
4398                 if (connection->cstate > C_STANDALONE)
4399                         return ERR_NET_CONFIGURED;
4400         }
4401         if (!idr_is_empty(&resource->devices))
4402                 return ERR_RES_IN_USE;
4403
4404         /* The state engine has stopped the sender thread, so we don't
4405          * need to flush the sender work queue before generating the
4406          * DESTROY event here. */
4407         mutex_lock(&notification_mutex);
4408         notify_resource_state(NULL, 0, resource, NULL, NOTIFY_DESTROY);
4409         mutex_unlock(&notification_mutex);
4410
4411         mutex_lock(&resources_mutex);
4412         list_del_rcu(&resource->resources);
4413         mutex_unlock(&resources_mutex);
4414         /* Make sure all threads have actually stopped: state handling only
4415          * does drbd_thread_stop_nowait(). */
4416         list_for_each_entry(connection, &resource->connections, connections)
4417                 drbd_thread_stop(&connection->worker);
4418         synchronize_rcu();
4419         drbd_free_resource(resource);
4420         return NO_ERROR;
4421 }
4422
4423 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
4424 {
4425         struct drbd_config_context adm_ctx;
4426         struct drbd_resource *resource;
4427         struct drbd_connection *connection;
4428         struct drbd_device *device;
4429         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
4430         unsigned i;
4431
4432         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4433         if (!adm_ctx.reply_skb)
4434                 return retcode;
4435         if (retcode != NO_ERROR)
4436                 goto finish;
4437
4438         resource = adm_ctx.resource;
4439         mutex_lock(&resource->adm_mutex);
4440         /* demote */
4441         for_each_connection(connection, resource) {
4442                 struct drbd_peer_device *peer_device;
4443
4444                 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
4445                         retcode = drbd_set_role(peer_device->device, R_SECONDARY, 0);
4446                         if (retcode < SS_SUCCESS) {
4447                                 drbd_msg_put_info(adm_ctx.reply_skb, "failed to demote");
4448                                 goto out;
4449                         }
4450                 }
4451
4452                 retcode = conn_try_disconnect(connection, 0);
4453                 if (retcode < SS_SUCCESS) {
4454                         drbd_msg_put_info(adm_ctx.reply_skb, "failed to disconnect");
4455                         goto out;
4456                 }
4457         }
4458
4459         /* detach */
4460         idr_for_each_entry(&resource->devices, device, i) {
4461                 retcode = adm_detach(device, 0);
4462                 if (retcode < SS_SUCCESS || retcode > NO_ERROR) {
4463                         drbd_msg_put_info(adm_ctx.reply_skb, "failed to detach");
4464                         goto out;
4465                 }
4466         }
4467
4468         /* delete volumes */
4469         idr_for_each_entry(&resource->devices, device, i) {
4470                 retcode = adm_del_minor(device);
4471                 if (retcode != NO_ERROR) {
4472                         /* "can not happen" */
4473                         drbd_msg_put_info(adm_ctx.reply_skb, "failed to delete volume");
4474                         goto out;
4475                 }
4476         }
4477
4478         retcode = adm_del_resource(resource);
4479 out:
4480         mutex_unlock(&resource->adm_mutex);
4481 finish:
4482         drbd_adm_finish(&adm_ctx, info, retcode);
4483         return 0;
4484 }
4485
4486 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
4487 {
4488         struct drbd_config_context adm_ctx;
4489         struct drbd_resource *resource;
4490         enum drbd_ret_code retcode;
4491
4492         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4493         if (!adm_ctx.reply_skb)
4494                 return retcode;
4495         if (retcode != NO_ERROR)
4496                 goto finish;
4497         resource = adm_ctx.resource;
4498
4499         mutex_lock(&resource->adm_mutex);
4500         retcode = adm_del_resource(resource);
4501         mutex_unlock(&resource->adm_mutex);
4502 finish:
4503         drbd_adm_finish(&adm_ctx, info, retcode);
4504         return 0;
4505 }
4506
4507 void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib)
4508 {
4509         struct sk_buff *msg;
4510         struct drbd_genlmsghdr *d_out;
4511         unsigned seq;
4512         int err = -ENOMEM;
4513
4514         seq = atomic_inc_return(&drbd_genl_seq);
4515         msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4516         if (!msg)
4517                 goto failed;
4518
4519         err = -EMSGSIZE;
4520         d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
4521         if (!d_out) /* cannot happen, but anyways. */
4522                 goto nla_put_failure;
4523         d_out->minor = device_to_minor(device);
4524         d_out->ret_code = NO_ERROR;
4525
4526         if (nla_put_status_info(msg, device, sib))
4527                 goto nla_put_failure;
4528         genlmsg_end(msg, d_out);
4529         err = drbd_genl_multicast_events(msg, GFP_NOWAIT);
4530         /* msg has been consumed or freed in netlink_broadcast() */
4531         if (err && err != -ESRCH)
4532                 goto failed;
4533
4534         return;
4535
4536 nla_put_failure:
4537         nlmsg_free(msg);
4538 failed:
4539         drbd_err(device, "Error %d while broadcasting event. "
4540                         "Event seq:%u sib_reason:%u\n",
4541                         err, seq, sib->sib_reason);
4542 }
4543
4544 static int nla_put_notification_header(struct sk_buff *msg,
4545                                        enum drbd_notification_type type)
4546 {
4547         struct drbd_notification_header nh = {
4548                 .nh_type = type,
4549         };
4550
4551         return drbd_notification_header_to_skb(msg, &nh, true);
4552 }
4553
4554 int notify_resource_state(struct sk_buff *skb,
4555                            unsigned int seq,
4556                            struct drbd_resource *resource,
4557                            struct resource_info *resource_info,
4558                            enum drbd_notification_type type)
4559 {
4560         struct resource_statistics resource_statistics;
4561         struct drbd_genlmsghdr *dh;
4562         bool multicast = false;
4563         int err;
4564
4565         if (!skb) {
4566                 seq = atomic_inc_return(&notify_genl_seq);
4567                 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4568                 err = -ENOMEM;
4569                 if (!skb)
4570                         goto failed;
4571                 multicast = true;
4572         }
4573
4574         err = -EMSGSIZE;
4575         dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_RESOURCE_STATE);
4576         if (!dh)
4577                 goto nla_put_failure;
4578         dh->minor = -1U;
4579         dh->ret_code = NO_ERROR;
4580         if (nla_put_drbd_cfg_context(skb, resource, NULL, NULL) ||
4581             nla_put_notification_header(skb, type) ||
4582             ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4583              resource_info_to_skb(skb, resource_info, true)))
4584                 goto nla_put_failure;
4585         resource_statistics.res_stat_write_ordering = resource->write_ordering;
4586         err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
4587         if (err)
4588                 goto nla_put_failure;
4589         genlmsg_end(skb, dh);
4590         if (multicast) {
4591                 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4592                 /* skb has been consumed or freed in netlink_broadcast() */
4593                 if (err && err != -ESRCH)
4594                         goto failed;
4595         }
4596         return 0;
4597
4598 nla_put_failure:
4599         nlmsg_free(skb);
4600 failed:
4601         drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4602                         err, seq);
4603         return err;
4604 }
4605
4606 int notify_device_state(struct sk_buff *skb,
4607                          unsigned int seq,
4608                          struct drbd_device *device,
4609                          struct device_info *device_info,
4610                          enum drbd_notification_type type)
4611 {
4612         struct device_statistics device_statistics;
4613         struct drbd_genlmsghdr *dh;
4614         bool multicast = false;
4615         int err;
4616
4617         if (!skb) {
4618                 seq = atomic_inc_return(&notify_genl_seq);
4619                 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4620                 err = -ENOMEM;
4621                 if (!skb)
4622                         goto failed;
4623                 multicast = true;
4624         }
4625
4626         err = -EMSGSIZE;
4627         dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_DEVICE_STATE);
4628         if (!dh)
4629                 goto nla_put_failure;
4630         dh->minor = device->minor;
4631         dh->ret_code = NO_ERROR;
4632         if (nla_put_drbd_cfg_context(skb, device->resource, NULL, device) ||
4633             nla_put_notification_header(skb, type) ||
4634             ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4635              device_info_to_skb(skb, device_info, true)))
4636                 goto nla_put_failure;
4637         device_to_statistics(&device_statistics, device);
4638         device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
4639         genlmsg_end(skb, dh);
4640         if (multicast) {
4641                 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4642                 /* skb has been consumed or freed in netlink_broadcast() */
4643                 if (err && err != -ESRCH)
4644                         goto failed;
4645         }
4646         return 0;
4647
4648 nla_put_failure:
4649         nlmsg_free(skb);
4650 failed:
4651         drbd_err(device, "Error %d while broadcasting event. Event seq:%u\n",
4652                  err, seq);
4653         return err;
4654 }
4655
4656 int notify_connection_state(struct sk_buff *skb,
4657                              unsigned int seq,
4658                              struct drbd_connection *connection,
4659                              struct connection_info *connection_info,
4660                              enum drbd_notification_type type)
4661 {
4662         struct connection_statistics connection_statistics;
4663         struct drbd_genlmsghdr *dh;
4664         bool multicast = false;
4665         int err;
4666
4667         if (!skb) {
4668                 seq = atomic_inc_return(&notify_genl_seq);
4669                 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4670                 err = -ENOMEM;
4671                 if (!skb)
4672                         goto failed;
4673                 multicast = true;
4674         }
4675
4676         err = -EMSGSIZE;
4677         dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_CONNECTION_STATE);
4678         if (!dh)
4679                 goto nla_put_failure;
4680         dh->minor = -1U;
4681         dh->ret_code = NO_ERROR;
4682         if (nla_put_drbd_cfg_context(skb, connection->resource, connection, NULL) ||
4683             nla_put_notification_header(skb, type) ||
4684             ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4685              connection_info_to_skb(skb, connection_info, true)))
4686                 goto nla_put_failure;
4687         connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
4688         connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
4689         genlmsg_end(skb, dh);
4690         if (multicast) {
4691                 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4692                 /* skb has been consumed or freed in netlink_broadcast() */
4693                 if (err && err != -ESRCH)
4694                         goto failed;
4695         }
4696         return 0;
4697
4698 nla_put_failure:
4699         nlmsg_free(skb);
4700 failed:
4701         drbd_err(connection, "Error %d while broadcasting event. Event seq:%u\n",
4702                  err, seq);
4703         return err;
4704 }
4705
4706 int notify_peer_device_state(struct sk_buff *skb,
4707                               unsigned int seq,
4708                               struct drbd_peer_device *peer_device,
4709                               struct peer_device_info *peer_device_info,
4710                               enum drbd_notification_type type)
4711 {
4712         struct peer_device_statistics peer_device_statistics;
4713         struct drbd_resource *resource = peer_device->device->resource;
4714         struct drbd_genlmsghdr *dh;
4715         bool multicast = false;
4716         int err;
4717
4718         if (!skb) {
4719                 seq = atomic_inc_return(&notify_genl_seq);
4720                 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4721                 err = -ENOMEM;
4722                 if (!skb)
4723                         goto failed;
4724                 multicast = true;
4725         }
4726
4727         err = -EMSGSIZE;
4728         dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_PEER_DEVICE_STATE);
4729         if (!dh)
4730                 goto nla_put_failure;
4731         dh->minor = -1U;
4732         dh->ret_code = NO_ERROR;
4733         if (nla_put_drbd_cfg_context(skb, resource, peer_device->connection, peer_device->device) ||
4734             nla_put_notification_header(skb, type) ||
4735             ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4736              peer_device_info_to_skb(skb, peer_device_info, true)))
4737                 goto nla_put_failure;
4738         peer_device_to_statistics(&peer_device_statistics, peer_device);
4739         peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
4740         genlmsg_end(skb, dh);
4741         if (multicast) {
4742                 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4743                 /* skb has been consumed or freed in netlink_broadcast() */
4744                 if (err && err != -ESRCH)
4745                         goto failed;
4746         }
4747         return 0;
4748
4749 nla_put_failure:
4750         nlmsg_free(skb);
4751 failed:
4752         drbd_err(peer_device, "Error %d while broadcasting event. Event seq:%u\n",
4753                  err, seq);
4754         return err;
4755 }
4756
4757 void notify_helper(enum drbd_notification_type type,
4758                    struct drbd_device *device, struct drbd_connection *connection,
4759                    const char *name, int status)
4760 {
4761         struct drbd_resource *resource = device ? device->resource : connection->resource;
4762         struct drbd_helper_info helper_info;
4763         unsigned int seq = atomic_inc_return(&notify_genl_seq);
4764         struct sk_buff *skb = NULL;
4765         struct drbd_genlmsghdr *dh;
4766         int err;
4767
4768         strscpy(helper_info.helper_name, name, sizeof(helper_info.helper_name));
4769         helper_info.helper_name_len = min(strlen(name), sizeof(helper_info.helper_name));
4770         helper_info.helper_status = status;
4771
4772         skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4773         err = -ENOMEM;
4774         if (!skb)
4775                 goto fail;
4776
4777         err = -EMSGSIZE;
4778         dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_HELPER);
4779         if (!dh)
4780                 goto fail;
4781         dh->minor = device ? device->minor : -1;
4782         dh->ret_code = NO_ERROR;
4783         mutex_lock(&notification_mutex);
4784         if (nla_put_drbd_cfg_context(skb, resource, connection, device) ||
4785             nla_put_notification_header(skb, type) ||
4786             drbd_helper_info_to_skb(skb, &helper_info, true))
4787                 goto unlock_fail;
4788         genlmsg_end(skb, dh);
4789         err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4790         skb = NULL;
4791         /* skb has been consumed or freed in netlink_broadcast() */
4792         if (err && err != -ESRCH)
4793                 goto unlock_fail;
4794         mutex_unlock(&notification_mutex);
4795         return;
4796
4797 unlock_fail:
4798         mutex_unlock(&notification_mutex);
4799 fail:
4800         nlmsg_free(skb);
4801         drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4802                  err, seq);
4803 }
4804
4805 static int notify_initial_state_done(struct sk_buff *skb, unsigned int seq)
4806 {
4807         struct drbd_genlmsghdr *dh;
4808         int err;
4809
4810         err = -EMSGSIZE;
4811         dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_INITIAL_STATE_DONE);
4812         if (!dh)
4813                 goto nla_put_failure;
4814         dh->minor = -1U;
4815         dh->ret_code = NO_ERROR;
4816         if (nla_put_notification_header(skb, NOTIFY_EXISTS))
4817                 goto nla_put_failure;
4818         genlmsg_end(skb, dh);
4819         return 0;
4820
4821 nla_put_failure:
4822         nlmsg_free(skb);
4823         pr_err("Error %d sending event. Event seq:%u\n", err, seq);
4824         return err;
4825 }
4826
4827 static void free_state_changes(struct list_head *list)
4828 {
4829         while (!list_empty(list)) {
4830                 struct drbd_state_change *state_change =
4831                         list_first_entry(list, struct drbd_state_change, list);
4832                 list_del(&state_change->list);
4833                 forget_state_change(state_change);
4834         }
4835 }
4836
4837 static unsigned int notifications_for_state_change(struct drbd_state_change *state_change)
4838 {
4839         return 1 +
4840                state_change->n_connections +
4841                state_change->n_devices +
4842                state_change->n_devices * state_change->n_connections;
4843 }
4844
4845 static int get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4846 {
4847         struct drbd_state_change *state_change = (struct drbd_state_change *)cb->args[0];
4848         unsigned int seq = cb->args[2];
4849         unsigned int n;
4850         enum drbd_notification_type flags = 0;
4851         int err = 0;
4852
4853         /* There is no need for taking notification_mutex here: it doesn't
4854            matter if the initial state events mix with later state chage
4855            events; we can always tell the events apart by the NOTIFY_EXISTS
4856            flag. */
4857
4858         cb->args[5]--;
4859         if (cb->args[5] == 1) {
4860                 err = notify_initial_state_done(skb, seq);
4861                 goto out;
4862         }
4863         n = cb->args[4]++;
4864         if (cb->args[4] < cb->args[3])
4865                 flags |= NOTIFY_CONTINUES;
4866         if (n < 1) {
4867                 err = notify_resource_state_change(skb, seq, state_change->resource,
4868                                              NOTIFY_EXISTS | flags);
4869                 goto next;
4870         }
4871         n--;
4872         if (n < state_change->n_connections) {
4873                 err = notify_connection_state_change(skb, seq, &state_change->connections[n],
4874                                                NOTIFY_EXISTS | flags);
4875                 goto next;
4876         }
4877         n -= state_change->n_connections;
4878         if (n < state_change->n_devices) {
4879                 err = notify_device_state_change(skb, seq, &state_change->devices[n],
4880                                            NOTIFY_EXISTS | flags);
4881                 goto next;
4882         }
4883         n -= state_change->n_devices;
4884         if (n < state_change->n_devices * state_change->n_connections) {
4885                 err = notify_peer_device_state_change(skb, seq, &state_change->peer_devices[n],
4886                                                 NOTIFY_EXISTS | flags);
4887                 goto next;
4888         }
4889
4890 next:
4891         if (cb->args[4] == cb->args[3]) {
4892                 struct drbd_state_change *next_state_change =
4893                         list_entry(state_change->list.next,
4894                                    struct drbd_state_change, list);
4895                 cb->args[0] = (long)next_state_change;
4896                 cb->args[3] = notifications_for_state_change(next_state_change);
4897                 cb->args[4] = 0;
4898         }
4899 out:
4900         if (err)
4901                 return err;
4902         else
4903                 return skb->len;
4904 }
4905
4906 int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4907 {
4908         struct drbd_resource *resource;
4909         LIST_HEAD(head);
4910
4911         if (cb->args[5] >= 1) {
4912                 if (cb->args[5] > 1)
4913                         return get_initial_state(skb, cb);
4914                 if (cb->args[0]) {
4915                         struct drbd_state_change *state_change =
4916                                 (struct drbd_state_change *)cb->args[0];
4917
4918                         /* connect list to head */
4919                         list_add(&head, &state_change->list);
4920                         free_state_changes(&head);
4921                 }
4922                 return 0;
4923         }
4924
4925         cb->args[5] = 2;  /* number of iterations */
4926         mutex_lock(&resources_mutex);
4927         for_each_resource(resource, &drbd_resources) {
4928                 struct drbd_state_change *state_change;
4929
4930                 state_change = remember_old_state(resource, GFP_KERNEL);
4931                 if (!state_change) {
4932                         if (!list_empty(&head))
4933                                 free_state_changes(&head);
4934                         mutex_unlock(&resources_mutex);
4935                         return -ENOMEM;
4936                 }
4937                 copy_old_to_new_state_change(state_change);
4938                 list_add_tail(&state_change->list, &head);
4939                 cb->args[5] += notifications_for_state_change(state_change);
4940         }
4941         mutex_unlock(&resources_mutex);
4942
4943         if (!list_empty(&head)) {
4944                 struct drbd_state_change *state_change =
4945                         list_entry(head.next, struct drbd_state_change, list);
4946                 cb->args[0] = (long)state_change;
4947                 cb->args[3] = notifications_for_state_change(state_change);
4948                 list_del(&head);  /* detach list from head */
4949         }
4950
4951         cb->args[2] = cb->nlh->nlmsg_seq;
4952         return get_initial_state(skb, cb);
4953 }