| 1 | /* |
| 2 | drbd_nl.c |
| 3 | |
| 4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. |
| 5 | |
| 6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. |
| 7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. |
| 8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. |
| 9 | |
| 10 | drbd is free software; you can redistribute it and/or modify |
| 11 | it under the terms of the GNU General Public License as published by |
| 12 | the Free Software Foundation; either version 2, or (at your option) |
| 13 | any later version. |
| 14 | |
| 15 | drbd is distributed in the hope that it will be useful, |
| 16 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 18 | GNU General Public License for more details. |
| 19 | |
| 20 | You should have received a copy of the GNU General Public License |
| 21 | along with drbd; see the file COPYING. If not, write to |
| 22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. |
| 23 | |
| 24 | */ |
| 25 | |
| 26 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 27 | |
| 28 | #include <linux/module.h> |
| 29 | #include <linux/drbd.h> |
| 30 | #include <linux/in.h> |
| 31 | #include <linux/fs.h> |
| 32 | #include <linux/file.h> |
| 33 | #include <linux/slab.h> |
| 34 | #include <linux/blkpg.h> |
| 35 | #include <linux/cpumask.h> |
| 36 | #include "drbd_int.h" |
| 37 | #include "drbd_protocol.h" |
| 38 | #include "drbd_req.h" |
| 39 | #include "drbd_state_change.h" |
| 40 | #include <asm/unaligned.h> |
| 41 | #include <linux/drbd_limits.h> |
| 42 | #include <linux/kthread.h> |
| 43 | |
| 44 | #include <net/genetlink.h> |
| 45 | |
| 46 | /* .doit */ |
| 47 | // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info); |
| 48 | // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info); |
| 49 | |
| 50 | int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info); |
| 51 | int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info); |
| 52 | |
| 53 | int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info); |
| 54 | int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info); |
| 55 | int drbd_adm_down(struct sk_buff *skb, struct genl_info *info); |
| 56 | |
| 57 | int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info); |
| 58 | int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info); |
| 59 | int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info); |
| 60 | int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info); |
| 61 | int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info); |
| 62 | int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info); |
| 63 | int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info); |
| 64 | int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info); |
| 65 | int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info); |
| 66 | int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info); |
| 67 | int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info); |
| 68 | int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info); |
| 69 | int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info); |
| 70 | int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info); |
| 71 | int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info); |
| 72 | int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info); |
| 73 | int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info); |
| 74 | int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info); |
| 75 | int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info); |
| 76 | int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info); |
| 77 | /* .dumpit */ |
| 78 | int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb); |
| 79 | int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb); |
| 80 | int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb); |
| 81 | int drbd_adm_dump_devices_done(struct netlink_callback *cb); |
| 82 | int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb); |
| 83 | int drbd_adm_dump_connections_done(struct netlink_callback *cb); |
| 84 | int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb); |
| 85 | int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb); |
| 86 | int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb); |
| 87 | |
| 88 | #include <linux/drbd_genl_api.h> |
| 89 | #include "drbd_nla.h" |
| 90 | #include <linux/genl_magic_func.h> |
| 91 | |
| 92 | static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */ |
| 93 | static atomic_t notify_genl_seq = ATOMIC_INIT(2); /* two. */ |
| 94 | |
| 95 | DEFINE_MUTEX(notification_mutex); |
| 96 | |
| 97 | /* used blkdev_get_by_path, to claim our meta data device(s) */ |
| 98 | static char *drbd_m_holder = "Hands off! this is DRBD's meta data device."; |
| 99 | |
| 100 | static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info) |
| 101 | { |
| 102 | genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb)))); |
| 103 | if (genlmsg_reply(skb, info)) |
| 104 | pr_err("error sending genl reply\n"); |
| 105 | } |
| 106 | |
| 107 | /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only |
| 108 | * reason it could fail was no space in skb, and there are 4k available. */ |
| 109 | static int drbd_msg_put_info(struct sk_buff *skb, const char *info) |
| 110 | { |
| 111 | struct nlattr *nla; |
| 112 | int err = -EMSGSIZE; |
| 113 | |
| 114 | if (!info || !info[0]) |
| 115 | return 0; |
| 116 | |
| 117 | nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY); |
| 118 | if (!nla) |
| 119 | return err; |
| 120 | |
| 121 | err = nla_put_string(skb, T_info_text, info); |
| 122 | if (err) { |
| 123 | nla_nest_cancel(skb, nla); |
| 124 | return err; |
| 125 | } else |
| 126 | nla_nest_end(skb, nla); |
| 127 | return 0; |
| 128 | } |
| 129 | |
| 130 | /* This would be a good candidate for a "pre_doit" hook, |
| 131 | * and per-family private info->pointers. |
| 132 | * But we need to stay compatible with older kernels. |
| 133 | * If it returns successfully, adm_ctx members are valid. |
| 134 | * |
| 135 | * At this point, we still rely on the global genl_lock(). |
| 136 | * If we want to avoid that, and allow "genl_family.parallel_ops", we may need |
| 137 | * to add additional synchronization against object destruction/modification. |
| 138 | */ |
| 139 | #define DRBD_ADM_NEED_MINOR 1 |
| 140 | #define DRBD_ADM_NEED_RESOURCE 2 |
| 141 | #define DRBD_ADM_NEED_CONNECTION 4 |
| 142 | static int drbd_adm_prepare(struct drbd_config_context *adm_ctx, |
| 143 | struct sk_buff *skb, struct genl_info *info, unsigned flags) |
| 144 | { |
| 145 | struct drbd_genlmsghdr *d_in = info->userhdr; |
| 146 | const u8 cmd = info->genlhdr->cmd; |
| 147 | int err; |
| 148 | |
| 149 | memset(adm_ctx, 0, sizeof(*adm_ctx)); |
| 150 | |
| 151 | /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */ |
| 152 | if (cmd != DRBD_ADM_GET_STATUS && !capable(CAP_NET_ADMIN)) |
| 153 | return -EPERM; |
| 154 | |
| 155 | adm_ctx->reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL); |
| 156 | if (!adm_ctx->reply_skb) { |
| 157 | err = -ENOMEM; |
| 158 | goto fail; |
| 159 | } |
| 160 | |
| 161 | adm_ctx->reply_dh = genlmsg_put_reply(adm_ctx->reply_skb, |
| 162 | info, &drbd_genl_family, 0, cmd); |
| 163 | /* put of a few bytes into a fresh skb of >= 4k will always succeed. |
| 164 | * but anyways */ |
| 165 | if (!adm_ctx->reply_dh) { |
| 166 | err = -ENOMEM; |
| 167 | goto fail; |
| 168 | } |
| 169 | |
| 170 | adm_ctx->reply_dh->minor = d_in->minor; |
| 171 | adm_ctx->reply_dh->ret_code = NO_ERROR; |
| 172 | |
| 173 | adm_ctx->volume = VOLUME_UNSPECIFIED; |
| 174 | if (info->attrs[DRBD_NLA_CFG_CONTEXT]) { |
| 175 | struct nlattr *nla; |
| 176 | /* parse and validate only */ |
| 177 | err = drbd_cfg_context_from_attrs(NULL, info); |
| 178 | if (err) |
| 179 | goto fail; |
| 180 | |
| 181 | /* It was present, and valid, |
| 182 | * copy it over to the reply skb. */ |
| 183 | err = nla_put_nohdr(adm_ctx->reply_skb, |
| 184 | info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len, |
| 185 | info->attrs[DRBD_NLA_CFG_CONTEXT]); |
| 186 | if (err) |
| 187 | goto fail; |
| 188 | |
| 189 | /* and assign stuff to the adm_ctx */ |
| 190 | nla = nested_attr_tb[__nla_type(T_ctx_volume)]; |
| 191 | if (nla) |
| 192 | adm_ctx->volume = nla_get_u32(nla); |
| 193 | nla = nested_attr_tb[__nla_type(T_ctx_resource_name)]; |
| 194 | if (nla) |
| 195 | adm_ctx->resource_name = nla_data(nla); |
| 196 | adm_ctx->my_addr = nested_attr_tb[__nla_type(T_ctx_my_addr)]; |
| 197 | adm_ctx->peer_addr = nested_attr_tb[__nla_type(T_ctx_peer_addr)]; |
| 198 | if ((adm_ctx->my_addr && |
| 199 | nla_len(adm_ctx->my_addr) > sizeof(adm_ctx->connection->my_addr)) || |
| 200 | (adm_ctx->peer_addr && |
| 201 | nla_len(adm_ctx->peer_addr) > sizeof(adm_ctx->connection->peer_addr))) { |
| 202 | err = -EINVAL; |
| 203 | goto fail; |
| 204 | } |
| 205 | } |
| 206 | |
| 207 | adm_ctx->minor = d_in->minor; |
| 208 | adm_ctx->device = minor_to_device(d_in->minor); |
| 209 | |
| 210 | /* We are protected by the global genl_lock(). |
| 211 | * But we may explicitly drop it/retake it in drbd_adm_set_role(), |
| 212 | * so make sure this object stays around. */ |
| 213 | if (adm_ctx->device) |
| 214 | kref_get(&adm_ctx->device->kref); |
| 215 | |
| 216 | if (adm_ctx->resource_name) { |
| 217 | adm_ctx->resource = drbd_find_resource(adm_ctx->resource_name); |
| 218 | } |
| 219 | |
| 220 | if (!adm_ctx->device && (flags & DRBD_ADM_NEED_MINOR)) { |
| 221 | drbd_msg_put_info(adm_ctx->reply_skb, "unknown minor"); |
| 222 | return ERR_MINOR_INVALID; |
| 223 | } |
| 224 | if (!adm_ctx->resource && (flags & DRBD_ADM_NEED_RESOURCE)) { |
| 225 | drbd_msg_put_info(adm_ctx->reply_skb, "unknown resource"); |
| 226 | if (adm_ctx->resource_name) |
| 227 | return ERR_RES_NOT_KNOWN; |
| 228 | return ERR_INVALID_REQUEST; |
| 229 | } |
| 230 | |
| 231 | if (flags & DRBD_ADM_NEED_CONNECTION) { |
| 232 | if (adm_ctx->resource) { |
| 233 | drbd_msg_put_info(adm_ctx->reply_skb, "no resource name expected"); |
| 234 | return ERR_INVALID_REQUEST; |
| 235 | } |
| 236 | if (adm_ctx->device) { |
| 237 | drbd_msg_put_info(adm_ctx->reply_skb, "no minor number expected"); |
| 238 | return ERR_INVALID_REQUEST; |
| 239 | } |
| 240 | if (adm_ctx->my_addr && adm_ctx->peer_addr) |
| 241 | adm_ctx->connection = conn_get_by_addrs(nla_data(adm_ctx->my_addr), |
| 242 | nla_len(adm_ctx->my_addr), |
| 243 | nla_data(adm_ctx->peer_addr), |
| 244 | nla_len(adm_ctx->peer_addr)); |
| 245 | if (!adm_ctx->connection) { |
| 246 | drbd_msg_put_info(adm_ctx->reply_skb, "unknown connection"); |
| 247 | return ERR_INVALID_REQUEST; |
| 248 | } |
| 249 | } |
| 250 | |
| 251 | /* some more paranoia, if the request was over-determined */ |
| 252 | if (adm_ctx->device && adm_ctx->resource && |
| 253 | adm_ctx->device->resource != adm_ctx->resource) { |
| 254 | pr_warning("request: minor=%u, resource=%s; but that minor belongs to resource %s\n", |
| 255 | adm_ctx->minor, adm_ctx->resource->name, |
| 256 | adm_ctx->device->resource->name); |
| 257 | drbd_msg_put_info(adm_ctx->reply_skb, "minor exists in different resource"); |
| 258 | return ERR_INVALID_REQUEST; |
| 259 | } |
| 260 | if (adm_ctx->device && |
| 261 | adm_ctx->volume != VOLUME_UNSPECIFIED && |
| 262 | adm_ctx->volume != adm_ctx->device->vnr) { |
| 263 | pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n", |
| 264 | adm_ctx->minor, adm_ctx->volume, |
| 265 | adm_ctx->device->vnr, |
| 266 | adm_ctx->device->resource->name); |
| 267 | drbd_msg_put_info(adm_ctx->reply_skb, "minor exists as different volume"); |
| 268 | return ERR_INVALID_REQUEST; |
| 269 | } |
| 270 | |
| 271 | /* still, provide adm_ctx->resource always, if possible. */ |
| 272 | if (!adm_ctx->resource) { |
| 273 | adm_ctx->resource = adm_ctx->device ? adm_ctx->device->resource |
| 274 | : adm_ctx->connection ? adm_ctx->connection->resource : NULL; |
| 275 | if (adm_ctx->resource) |
| 276 | kref_get(&adm_ctx->resource->kref); |
| 277 | } |
| 278 | |
| 279 | return NO_ERROR; |
| 280 | |
| 281 | fail: |
| 282 | nlmsg_free(adm_ctx->reply_skb); |
| 283 | adm_ctx->reply_skb = NULL; |
| 284 | return err; |
| 285 | } |
| 286 | |
| 287 | static int drbd_adm_finish(struct drbd_config_context *adm_ctx, |
| 288 | struct genl_info *info, int retcode) |
| 289 | { |
| 290 | if (adm_ctx->device) { |
| 291 | kref_put(&adm_ctx->device->kref, drbd_destroy_device); |
| 292 | adm_ctx->device = NULL; |
| 293 | } |
| 294 | if (adm_ctx->connection) { |
| 295 | kref_put(&adm_ctx->connection->kref, &drbd_destroy_connection); |
| 296 | adm_ctx->connection = NULL; |
| 297 | } |
| 298 | if (adm_ctx->resource) { |
| 299 | kref_put(&adm_ctx->resource->kref, drbd_destroy_resource); |
| 300 | adm_ctx->resource = NULL; |
| 301 | } |
| 302 | |
| 303 | if (!adm_ctx->reply_skb) |
| 304 | return -ENOMEM; |
| 305 | |
| 306 | adm_ctx->reply_dh->ret_code = retcode; |
| 307 | drbd_adm_send_reply(adm_ctx->reply_skb, info); |
| 308 | return 0; |
| 309 | } |
| 310 | |
| 311 | static void setup_khelper_env(struct drbd_connection *connection, char **envp) |
| 312 | { |
| 313 | char *afs; |
| 314 | |
| 315 | /* FIXME: A future version will not allow this case. */ |
| 316 | if (connection->my_addr_len == 0 || connection->peer_addr_len == 0) |
| 317 | return; |
| 318 | |
| 319 | switch (((struct sockaddr *)&connection->peer_addr)->sa_family) { |
| 320 | case AF_INET6: |
| 321 | afs = "ipv6"; |
| 322 | snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6", |
| 323 | &((struct sockaddr_in6 *)&connection->peer_addr)->sin6_addr); |
| 324 | break; |
| 325 | case AF_INET: |
| 326 | afs = "ipv4"; |
| 327 | snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4", |
| 328 | &((struct sockaddr_in *)&connection->peer_addr)->sin_addr); |
| 329 | break; |
| 330 | default: |
| 331 | afs = "ssocks"; |
| 332 | snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4", |
| 333 | &((struct sockaddr_in *)&connection->peer_addr)->sin_addr); |
| 334 | } |
| 335 | snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs); |
| 336 | } |
| 337 | |
| 338 | int drbd_khelper(struct drbd_device *device, char *cmd) |
| 339 | { |
| 340 | char *envp[] = { "HOME=/", |
| 341 | "TERM=linux", |
| 342 | "PATH=/sbin:/usr/sbin:/bin:/usr/bin", |
| 343 | (char[20]) { }, /* address family */ |
| 344 | (char[60]) { }, /* address */ |
| 345 | NULL }; |
| 346 | char mb[12]; |
| 347 | char *argv[] = {usermode_helper, cmd, mb, NULL }; |
| 348 | struct drbd_connection *connection = first_peer_device(device)->connection; |
| 349 | struct sib_info sib; |
| 350 | int ret; |
| 351 | |
| 352 | if (current == connection->worker.task) |
| 353 | set_bit(CALLBACK_PENDING, &connection->flags); |
| 354 | |
| 355 | snprintf(mb, 12, "minor-%d", device_to_minor(device)); |
| 356 | setup_khelper_env(connection, envp); |
| 357 | |
| 358 | /* The helper may take some time. |
| 359 | * write out any unsynced meta data changes now */ |
| 360 | drbd_md_sync(device); |
| 361 | |
| 362 | drbd_info(device, "helper command: %s %s %s\n", usermode_helper, cmd, mb); |
| 363 | sib.sib_reason = SIB_HELPER_PRE; |
| 364 | sib.helper_name = cmd; |
| 365 | drbd_bcast_event(device, &sib); |
| 366 | notify_helper(NOTIFY_CALL, device, connection, cmd, 0); |
| 367 | ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC); |
| 368 | if (ret) |
| 369 | drbd_warn(device, "helper command: %s %s %s exit code %u (0x%x)\n", |
| 370 | usermode_helper, cmd, mb, |
| 371 | (ret >> 8) & 0xff, ret); |
| 372 | else |
| 373 | drbd_info(device, "helper command: %s %s %s exit code %u (0x%x)\n", |
| 374 | usermode_helper, cmd, mb, |
| 375 | (ret >> 8) & 0xff, ret); |
| 376 | sib.sib_reason = SIB_HELPER_POST; |
| 377 | sib.helper_exit_code = ret; |
| 378 | drbd_bcast_event(device, &sib); |
| 379 | notify_helper(NOTIFY_RESPONSE, device, connection, cmd, ret); |
| 380 | |
| 381 | if (current == connection->worker.task) |
| 382 | clear_bit(CALLBACK_PENDING, &connection->flags); |
| 383 | |
| 384 | if (ret < 0) /* Ignore any ERRNOs we got. */ |
| 385 | ret = 0; |
| 386 | |
| 387 | return ret; |
| 388 | } |
| 389 | |
| 390 | static int conn_khelper(struct drbd_connection *connection, char *cmd) |
| 391 | { |
| 392 | char *envp[] = { "HOME=/", |
| 393 | "TERM=linux", |
| 394 | "PATH=/sbin:/usr/sbin:/bin:/usr/bin", |
| 395 | (char[20]) { }, /* address family */ |
| 396 | (char[60]) { }, /* address */ |
| 397 | NULL }; |
| 398 | char *resource_name = connection->resource->name; |
| 399 | char *argv[] = {usermode_helper, cmd, resource_name, NULL }; |
| 400 | int ret; |
| 401 | |
| 402 | setup_khelper_env(connection, envp); |
| 403 | conn_md_sync(connection); |
| 404 | |
| 405 | drbd_info(connection, "helper command: %s %s %s\n", usermode_helper, cmd, resource_name); |
| 406 | /* TODO: conn_bcast_event() ?? */ |
| 407 | notify_helper(NOTIFY_CALL, NULL, connection, cmd, 0); |
| 408 | |
| 409 | ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC); |
| 410 | if (ret) |
| 411 | drbd_warn(connection, "helper command: %s %s %s exit code %u (0x%x)\n", |
| 412 | usermode_helper, cmd, resource_name, |
| 413 | (ret >> 8) & 0xff, ret); |
| 414 | else |
| 415 | drbd_info(connection, "helper command: %s %s %s exit code %u (0x%x)\n", |
| 416 | usermode_helper, cmd, resource_name, |
| 417 | (ret >> 8) & 0xff, ret); |
| 418 | /* TODO: conn_bcast_event() ?? */ |
| 419 | notify_helper(NOTIFY_RESPONSE, NULL, connection, cmd, ret); |
| 420 | |
| 421 | if (ret < 0) /* Ignore any ERRNOs we got. */ |
| 422 | ret = 0; |
| 423 | |
| 424 | return ret; |
| 425 | } |
| 426 | |
| 427 | static enum drbd_fencing_p highest_fencing_policy(struct drbd_connection *connection) |
| 428 | { |
| 429 | enum drbd_fencing_p fp = FP_NOT_AVAIL; |
| 430 | struct drbd_peer_device *peer_device; |
| 431 | int vnr; |
| 432 | |
| 433 | rcu_read_lock(); |
| 434 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
| 435 | struct drbd_device *device = peer_device->device; |
| 436 | if (get_ldev_if_state(device, D_CONSISTENT)) { |
| 437 | struct disk_conf *disk_conf = |
| 438 | rcu_dereference(peer_device->device->ldev->disk_conf); |
| 439 | fp = max_t(enum drbd_fencing_p, fp, disk_conf->fencing); |
| 440 | put_ldev(device); |
| 441 | } |
| 442 | } |
| 443 | rcu_read_unlock(); |
| 444 | |
| 445 | if (fp == FP_NOT_AVAIL) { |
| 446 | /* IO Suspending works on the whole resource. |
| 447 | Do it only for one device. */ |
| 448 | vnr = 0; |
| 449 | peer_device = idr_get_next(&connection->peer_devices, &vnr); |
| 450 | drbd_change_state(peer_device->device, CS_VERBOSE | CS_HARD, NS(susp_fen, 0)); |
| 451 | } |
| 452 | |
| 453 | return fp; |
| 454 | } |
| 455 | |
| 456 | bool conn_try_outdate_peer(struct drbd_connection *connection) |
| 457 | { |
| 458 | unsigned int connect_cnt; |
| 459 | union drbd_state mask = { }; |
| 460 | union drbd_state val = { }; |
| 461 | enum drbd_fencing_p fp; |
| 462 | char *ex_to_string; |
| 463 | int r; |
| 464 | |
| 465 | spin_lock_irq(&connection->resource->req_lock); |
| 466 | if (connection->cstate >= C_WF_REPORT_PARAMS) { |
| 467 | drbd_err(connection, "Expected cstate < C_WF_REPORT_PARAMS\n"); |
| 468 | spin_unlock_irq(&connection->resource->req_lock); |
| 469 | return false; |
| 470 | } |
| 471 | |
| 472 | connect_cnt = connection->connect_cnt; |
| 473 | spin_unlock_irq(&connection->resource->req_lock); |
| 474 | |
| 475 | fp = highest_fencing_policy(connection); |
| 476 | switch (fp) { |
| 477 | case FP_NOT_AVAIL: |
| 478 | drbd_warn(connection, "Not fencing peer, I'm not even Consistent myself.\n"); |
| 479 | goto out; |
| 480 | case FP_DONT_CARE: |
| 481 | return true; |
| 482 | default: ; |
| 483 | } |
| 484 | |
| 485 | r = conn_khelper(connection, "fence-peer"); |
| 486 | |
| 487 | switch ((r>>8) & 0xff) { |
| 488 | case 3: /* peer is inconsistent */ |
| 489 | ex_to_string = "peer is inconsistent or worse"; |
| 490 | mask.pdsk = D_MASK; |
| 491 | val.pdsk = D_INCONSISTENT; |
| 492 | break; |
| 493 | case 4: /* peer got outdated, or was already outdated */ |
| 494 | ex_to_string = "peer was fenced"; |
| 495 | mask.pdsk = D_MASK; |
| 496 | val.pdsk = D_OUTDATED; |
| 497 | break; |
| 498 | case 5: /* peer was down */ |
| 499 | if (conn_highest_disk(connection) == D_UP_TO_DATE) { |
| 500 | /* we will(have) create(d) a new UUID anyways... */ |
| 501 | ex_to_string = "peer is unreachable, assumed to be dead"; |
| 502 | mask.pdsk = D_MASK; |
| 503 | val.pdsk = D_OUTDATED; |
| 504 | } else { |
| 505 | ex_to_string = "peer unreachable, doing nothing since disk != UpToDate"; |
| 506 | } |
| 507 | break; |
| 508 | case 6: /* Peer is primary, voluntarily outdate myself. |
| 509 | * This is useful when an unconnected R_SECONDARY is asked to |
| 510 | * become R_PRIMARY, but finds the other peer being active. */ |
| 511 | ex_to_string = "peer is active"; |
| 512 | drbd_warn(connection, "Peer is primary, outdating myself.\n"); |
| 513 | mask.disk = D_MASK; |
| 514 | val.disk = D_OUTDATED; |
| 515 | break; |
| 516 | case 7: |
| 517 | if (fp != FP_STONITH) |
| 518 | drbd_err(connection, "fence-peer() = 7 && fencing != Stonith !!!\n"); |
| 519 | ex_to_string = "peer was stonithed"; |
| 520 | mask.pdsk = D_MASK; |
| 521 | val.pdsk = D_OUTDATED; |
| 522 | break; |
| 523 | default: |
| 524 | /* The script is broken ... */ |
| 525 | drbd_err(connection, "fence-peer helper broken, returned %d\n", (r>>8)&0xff); |
| 526 | return false; /* Eventually leave IO frozen */ |
| 527 | } |
| 528 | |
| 529 | drbd_info(connection, "fence-peer helper returned %d (%s)\n", |
| 530 | (r>>8) & 0xff, ex_to_string); |
| 531 | |
| 532 | out: |
| 533 | |
| 534 | /* Not using |
| 535 | conn_request_state(connection, mask, val, CS_VERBOSE); |
| 536 | here, because we might were able to re-establish the connection in the |
| 537 | meantime. */ |
| 538 | spin_lock_irq(&connection->resource->req_lock); |
| 539 | if (connection->cstate < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &connection->flags)) { |
| 540 | if (connection->connect_cnt != connect_cnt) |
| 541 | /* In case the connection was established and droped |
| 542 | while the fence-peer handler was running, ignore it */ |
| 543 | drbd_info(connection, "Ignoring fence-peer exit code\n"); |
| 544 | else |
| 545 | _conn_request_state(connection, mask, val, CS_VERBOSE); |
| 546 | } |
| 547 | spin_unlock_irq(&connection->resource->req_lock); |
| 548 | |
| 549 | return conn_highest_pdsk(connection) <= D_OUTDATED; |
| 550 | } |
| 551 | |
| 552 | static int _try_outdate_peer_async(void *data) |
| 553 | { |
| 554 | struct drbd_connection *connection = (struct drbd_connection *)data; |
| 555 | |
| 556 | conn_try_outdate_peer(connection); |
| 557 | |
| 558 | kref_put(&connection->kref, drbd_destroy_connection); |
| 559 | return 0; |
| 560 | } |
| 561 | |
| 562 | void conn_try_outdate_peer_async(struct drbd_connection *connection) |
| 563 | { |
| 564 | struct task_struct *opa; |
| 565 | |
| 566 | kref_get(&connection->kref); |
| 567 | /* We may just have force_sig()'ed this thread |
| 568 | * to get it out of some blocking network function. |
| 569 | * Clear signals; otherwise kthread_run(), which internally uses |
| 570 | * wait_on_completion_killable(), will mistake our pending signal |
| 571 | * for a new fatal signal and fail. */ |
| 572 | flush_signals(current); |
| 573 | opa = kthread_run(_try_outdate_peer_async, connection, "drbd_async_h"); |
| 574 | if (IS_ERR(opa)) { |
| 575 | drbd_err(connection, "out of mem, failed to invoke fence-peer helper\n"); |
| 576 | kref_put(&connection->kref, drbd_destroy_connection); |
| 577 | } |
| 578 | } |
| 579 | |
| 580 | enum drbd_state_rv |
| 581 | drbd_set_role(struct drbd_device *const device, enum drbd_role new_role, int force) |
| 582 | { |
| 583 | struct drbd_peer_device *const peer_device = first_peer_device(device); |
| 584 | struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL; |
| 585 | const int max_tries = 4; |
| 586 | enum drbd_state_rv rv = SS_UNKNOWN_ERROR; |
| 587 | struct net_conf *nc; |
| 588 | int try = 0; |
| 589 | int forced = 0; |
| 590 | union drbd_state mask, val; |
| 591 | |
| 592 | if (new_role == R_PRIMARY) { |
| 593 | struct drbd_connection *connection; |
| 594 | |
| 595 | /* Detect dead peers as soon as possible. */ |
| 596 | |
| 597 | rcu_read_lock(); |
| 598 | for_each_connection(connection, device->resource) |
| 599 | request_ping(connection); |
| 600 | rcu_read_unlock(); |
| 601 | } |
| 602 | |
| 603 | mutex_lock(device->state_mutex); |
| 604 | |
| 605 | mask.i = 0; mask.role = R_MASK; |
| 606 | val.i = 0; val.role = new_role; |
| 607 | |
| 608 | while (try++ < max_tries) { |
| 609 | rv = _drbd_request_state_holding_state_mutex(device, mask, val, CS_WAIT_COMPLETE); |
| 610 | |
| 611 | /* in case we first succeeded to outdate, |
| 612 | * but now suddenly could establish a connection */ |
| 613 | if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) { |
| 614 | val.pdsk = 0; |
| 615 | mask.pdsk = 0; |
| 616 | continue; |
| 617 | } |
| 618 | |
| 619 | if (rv == SS_NO_UP_TO_DATE_DISK && force && |
| 620 | (device->state.disk < D_UP_TO_DATE && |
| 621 | device->state.disk >= D_INCONSISTENT)) { |
| 622 | mask.disk = D_MASK; |
| 623 | val.disk = D_UP_TO_DATE; |
| 624 | forced = 1; |
| 625 | continue; |
| 626 | } |
| 627 | |
| 628 | if (rv == SS_NO_UP_TO_DATE_DISK && |
| 629 | device->state.disk == D_CONSISTENT && mask.pdsk == 0) { |
| 630 | D_ASSERT(device, device->state.pdsk == D_UNKNOWN); |
| 631 | |
| 632 | if (conn_try_outdate_peer(connection)) { |
| 633 | val.disk = D_UP_TO_DATE; |
| 634 | mask.disk = D_MASK; |
| 635 | } |
| 636 | continue; |
| 637 | } |
| 638 | |
| 639 | if (rv == SS_NOTHING_TO_DO) |
| 640 | goto out; |
| 641 | if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) { |
| 642 | if (!conn_try_outdate_peer(connection) && force) { |
| 643 | drbd_warn(device, "Forced into split brain situation!\n"); |
| 644 | mask.pdsk = D_MASK; |
| 645 | val.pdsk = D_OUTDATED; |
| 646 | |
| 647 | } |
| 648 | continue; |
| 649 | } |
| 650 | if (rv == SS_TWO_PRIMARIES) { |
| 651 | /* Maybe the peer is detected as dead very soon... |
| 652 | retry at most once more in this case. */ |
| 653 | int timeo; |
| 654 | rcu_read_lock(); |
| 655 | nc = rcu_dereference(connection->net_conf); |
| 656 | timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1; |
| 657 | rcu_read_unlock(); |
| 658 | schedule_timeout_interruptible(timeo); |
| 659 | if (try < max_tries) |
| 660 | try = max_tries - 1; |
| 661 | continue; |
| 662 | } |
| 663 | if (rv < SS_SUCCESS) { |
| 664 | rv = _drbd_request_state(device, mask, val, |
| 665 | CS_VERBOSE + CS_WAIT_COMPLETE); |
| 666 | if (rv < SS_SUCCESS) |
| 667 | goto out; |
| 668 | } |
| 669 | break; |
| 670 | } |
| 671 | |
| 672 | if (rv < SS_SUCCESS) |
| 673 | goto out; |
| 674 | |
| 675 | if (forced) |
| 676 | drbd_warn(device, "Forced to consider local data as UpToDate!\n"); |
| 677 | |
| 678 | /* Wait until nothing is on the fly :) */ |
| 679 | wait_event(device->misc_wait, atomic_read(&device->ap_pending_cnt) == 0); |
| 680 | |
| 681 | /* FIXME also wait for all pending P_BARRIER_ACK? */ |
| 682 | |
| 683 | if (new_role == R_SECONDARY) { |
| 684 | if (get_ldev(device)) { |
| 685 | device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1; |
| 686 | put_ldev(device); |
| 687 | } |
| 688 | } else { |
| 689 | mutex_lock(&device->resource->conf_update); |
| 690 | nc = connection->net_conf; |
| 691 | if (nc) |
| 692 | nc->discard_my_data = 0; /* without copy; single bit op is atomic */ |
| 693 | mutex_unlock(&device->resource->conf_update); |
| 694 | |
| 695 | if (get_ldev(device)) { |
| 696 | if (((device->state.conn < C_CONNECTED || |
| 697 | device->state.pdsk <= D_FAILED) |
| 698 | && device->ldev->md.uuid[UI_BITMAP] == 0) || forced) |
| 699 | drbd_uuid_new_current(device); |
| 700 | |
| 701 | device->ldev->md.uuid[UI_CURRENT] |= (u64)1; |
| 702 | put_ldev(device); |
| 703 | } |
| 704 | } |
| 705 | |
| 706 | /* writeout of activity log covered areas of the bitmap |
| 707 | * to stable storage done in after state change already */ |
| 708 | |
| 709 | if (device->state.conn >= C_WF_REPORT_PARAMS) { |
| 710 | /* if this was forced, we should consider sync */ |
| 711 | if (forced) |
| 712 | drbd_send_uuids(peer_device); |
| 713 | drbd_send_current_state(peer_device); |
| 714 | } |
| 715 | |
| 716 | drbd_md_sync(device); |
| 717 | set_disk_ro(device->vdisk, new_role == R_SECONDARY); |
| 718 | kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE); |
| 719 | out: |
| 720 | mutex_unlock(device->state_mutex); |
| 721 | return rv; |
| 722 | } |
| 723 | |
| 724 | static const char *from_attrs_err_to_txt(int err) |
| 725 | { |
| 726 | return err == -ENOMSG ? "required attribute missing" : |
| 727 | err == -EOPNOTSUPP ? "unknown mandatory attribute" : |
| 728 | err == -EEXIST ? "can not change invariant setting" : |
| 729 | "invalid attribute value"; |
| 730 | } |
| 731 | |
| 732 | int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info) |
| 733 | { |
| 734 | struct drbd_config_context adm_ctx; |
| 735 | struct set_role_parms parms; |
| 736 | int err; |
| 737 | enum drbd_ret_code retcode; |
| 738 | |
| 739 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 740 | if (!adm_ctx.reply_skb) |
| 741 | return retcode; |
| 742 | if (retcode != NO_ERROR) |
| 743 | goto out; |
| 744 | |
| 745 | memset(&parms, 0, sizeof(parms)); |
| 746 | if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) { |
| 747 | err = set_role_parms_from_attrs(&parms, info); |
| 748 | if (err) { |
| 749 | retcode = ERR_MANDATORY_TAG; |
| 750 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 751 | goto out; |
| 752 | } |
| 753 | } |
| 754 | genl_unlock(); |
| 755 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 756 | |
| 757 | if (info->genlhdr->cmd == DRBD_ADM_PRIMARY) |
| 758 | retcode = drbd_set_role(adm_ctx.device, R_PRIMARY, parms.assume_uptodate); |
| 759 | else |
| 760 | retcode = drbd_set_role(adm_ctx.device, R_SECONDARY, 0); |
| 761 | |
| 762 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 763 | genl_lock(); |
| 764 | out: |
| 765 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 766 | return 0; |
| 767 | } |
| 768 | |
| 769 | /* Initializes the md.*_offset members, so we are able to find |
| 770 | * the on disk meta data. |
| 771 | * |
| 772 | * We currently have two possible layouts: |
| 773 | * external: |
| 774 | * |----------- md_size_sect ------------------| |
| 775 | * [ 4k superblock ][ activity log ][ Bitmap ] |
| 776 | * | al_offset == 8 | |
| 777 | * | bm_offset = al_offset + X | |
| 778 | * ==> bitmap sectors = md_size_sect - bm_offset |
| 779 | * |
| 780 | * internal: |
| 781 | * |----------- md_size_sect ------------------| |
| 782 | * [data.....][ Bitmap ][ activity log ][ 4k superblock ] |
| 783 | * | al_offset < 0 | |
| 784 | * | bm_offset = al_offset - Y | |
| 785 | * ==> bitmap sectors = Y = al_offset - bm_offset |
| 786 | * |
| 787 | * Activity log size used to be fixed 32kB, |
| 788 | * but is about to become configurable. |
| 789 | */ |
| 790 | static void drbd_md_set_sector_offsets(struct drbd_device *device, |
| 791 | struct drbd_backing_dev *bdev) |
| 792 | { |
| 793 | sector_t md_size_sect = 0; |
| 794 | unsigned int al_size_sect = bdev->md.al_size_4k * 8; |
| 795 | |
| 796 | bdev->md.md_offset = drbd_md_ss(bdev); |
| 797 | |
| 798 | switch (bdev->md.meta_dev_idx) { |
| 799 | default: |
| 800 | /* v07 style fixed size indexed meta data */ |
| 801 | bdev->md.md_size_sect = MD_128MB_SECT; |
| 802 | bdev->md.al_offset = MD_4kB_SECT; |
| 803 | bdev->md.bm_offset = MD_4kB_SECT + al_size_sect; |
| 804 | break; |
| 805 | case DRBD_MD_INDEX_FLEX_EXT: |
| 806 | /* just occupy the full device; unit: sectors */ |
| 807 | bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev); |
| 808 | bdev->md.al_offset = MD_4kB_SECT; |
| 809 | bdev->md.bm_offset = MD_4kB_SECT + al_size_sect; |
| 810 | break; |
| 811 | case DRBD_MD_INDEX_INTERNAL: |
| 812 | case DRBD_MD_INDEX_FLEX_INT: |
| 813 | /* al size is still fixed */ |
| 814 | bdev->md.al_offset = -al_size_sect; |
| 815 | /* we need (slightly less than) ~ this much bitmap sectors: */ |
| 816 | md_size_sect = drbd_get_capacity(bdev->backing_bdev); |
| 817 | md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT); |
| 818 | md_size_sect = BM_SECT_TO_EXT(md_size_sect); |
| 819 | md_size_sect = ALIGN(md_size_sect, 8); |
| 820 | |
| 821 | /* plus the "drbd meta data super block", |
| 822 | * and the activity log; */ |
| 823 | md_size_sect += MD_4kB_SECT + al_size_sect; |
| 824 | |
| 825 | bdev->md.md_size_sect = md_size_sect; |
| 826 | /* bitmap offset is adjusted by 'super' block size */ |
| 827 | bdev->md.bm_offset = -md_size_sect + MD_4kB_SECT; |
| 828 | break; |
| 829 | } |
| 830 | } |
| 831 | |
| 832 | /* input size is expected to be in KB */ |
| 833 | char *ppsize(char *buf, unsigned long long size) |
| 834 | { |
| 835 | /* Needs 9 bytes at max including trailing NUL: |
| 836 | * -1ULL ==> "16384 EB" */ |
| 837 | static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' }; |
| 838 | int base = 0; |
| 839 | while (size >= 10000 && base < sizeof(units)-1) { |
| 840 | /* shift + round */ |
| 841 | size = (size >> 10) + !!(size & (1<<9)); |
| 842 | base++; |
| 843 | } |
| 844 | sprintf(buf, "%u %cB", (unsigned)size, units[base]); |
| 845 | |
| 846 | return buf; |
| 847 | } |
| 848 | |
| 849 | /* there is still a theoretical deadlock when called from receiver |
| 850 | * on an D_INCONSISTENT R_PRIMARY: |
| 851 | * remote READ does inc_ap_bio, receiver would need to receive answer |
| 852 | * packet from remote to dec_ap_bio again. |
| 853 | * receiver receive_sizes(), comes here, |
| 854 | * waits for ap_bio_cnt == 0. -> deadlock. |
| 855 | * but this cannot happen, actually, because: |
| 856 | * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable |
| 857 | * (not connected, or bad/no disk on peer): |
| 858 | * see drbd_fail_request_early, ap_bio_cnt is zero. |
| 859 | * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET: |
| 860 | * peer may not initiate a resize. |
| 861 | */ |
| 862 | /* Note these are not to be confused with |
| 863 | * drbd_adm_suspend_io/drbd_adm_resume_io, |
| 864 | * which are (sub) state changes triggered by admin (drbdsetup), |
| 865 | * and can be long lived. |
| 866 | * This changes an device->flag, is triggered by drbd internals, |
| 867 | * and should be short-lived. */ |
| 868 | /* It needs to be a counter, since multiple threads might |
| 869 | independently suspend and resume IO. */ |
| 870 | void drbd_suspend_io(struct drbd_device *device) |
| 871 | { |
| 872 | atomic_inc(&device->suspend_cnt); |
| 873 | if (drbd_suspended(device)) |
| 874 | return; |
| 875 | wait_event(device->misc_wait, !atomic_read(&device->ap_bio_cnt)); |
| 876 | } |
| 877 | |
| 878 | void drbd_resume_io(struct drbd_device *device) |
| 879 | { |
| 880 | if (atomic_dec_and_test(&device->suspend_cnt)) |
| 881 | wake_up(&device->misc_wait); |
| 882 | } |
| 883 | |
| 884 | /** |
| 885 | * drbd_determine_dev_size() - Sets the right device size obeying all constraints |
| 886 | * @device: DRBD device. |
| 887 | * |
| 888 | * Returns 0 on success, negative return values indicate errors. |
| 889 | * You should call drbd_md_sync() after calling this function. |
| 890 | */ |
| 891 | enum determine_dev_size |
| 892 | drbd_determine_dev_size(struct drbd_device *device, enum dds_flags flags, struct resize_parms *rs) __must_hold(local) |
| 893 | { |
| 894 | struct md_offsets_and_sizes { |
| 895 | u64 last_agreed_sect; |
| 896 | u64 md_offset; |
| 897 | s32 al_offset; |
| 898 | s32 bm_offset; |
| 899 | u32 md_size_sect; |
| 900 | |
| 901 | u32 al_stripes; |
| 902 | u32 al_stripe_size_4k; |
| 903 | } prev; |
| 904 | sector_t u_size, size; |
| 905 | struct drbd_md *md = &device->ldev->md; |
| 906 | char ppb[10]; |
| 907 | void *buffer; |
| 908 | |
| 909 | int md_moved, la_size_changed; |
| 910 | enum determine_dev_size rv = DS_UNCHANGED; |
| 911 | |
| 912 | /* We may change the on-disk offsets of our meta data below. Lock out |
| 913 | * anything that may cause meta data IO, to avoid acting on incomplete |
| 914 | * layout changes or scribbling over meta data that is in the process |
| 915 | * of being moved. |
| 916 | * |
| 917 | * Move is not exactly correct, btw, currently we have all our meta |
| 918 | * data in core memory, to "move" it we just write it all out, there |
| 919 | * are no reads. */ |
| 920 | drbd_suspend_io(device); |
| 921 | buffer = drbd_md_get_buffer(device, __func__); /* Lock meta-data IO */ |
| 922 | if (!buffer) { |
| 923 | drbd_resume_io(device); |
| 924 | return DS_ERROR; |
| 925 | } |
| 926 | |
| 927 | /* remember current offset and sizes */ |
| 928 | prev.last_agreed_sect = md->la_size_sect; |
| 929 | prev.md_offset = md->md_offset; |
| 930 | prev.al_offset = md->al_offset; |
| 931 | prev.bm_offset = md->bm_offset; |
| 932 | prev.md_size_sect = md->md_size_sect; |
| 933 | prev.al_stripes = md->al_stripes; |
| 934 | prev.al_stripe_size_4k = md->al_stripe_size_4k; |
| 935 | |
| 936 | if (rs) { |
| 937 | /* rs is non NULL if we should change the AL layout only */ |
| 938 | md->al_stripes = rs->al_stripes; |
| 939 | md->al_stripe_size_4k = rs->al_stripe_size / 4; |
| 940 | md->al_size_4k = (u64)rs->al_stripes * rs->al_stripe_size / 4; |
| 941 | } |
| 942 | |
| 943 | drbd_md_set_sector_offsets(device, device->ldev); |
| 944 | |
| 945 | rcu_read_lock(); |
| 946 | u_size = rcu_dereference(device->ldev->disk_conf)->disk_size; |
| 947 | rcu_read_unlock(); |
| 948 | size = drbd_new_dev_size(device, device->ldev, u_size, flags & DDSF_FORCED); |
| 949 | |
| 950 | if (size < prev.last_agreed_sect) { |
| 951 | if (rs && u_size == 0) { |
| 952 | /* Remove "rs &&" later. This check should always be active, but |
| 953 | right now the receiver expects the permissive behavior */ |
| 954 | drbd_warn(device, "Implicit shrink not allowed. " |
| 955 | "Use --size=%llus for explicit shrink.\n", |
| 956 | (unsigned long long)size); |
| 957 | rv = DS_ERROR_SHRINK; |
| 958 | } |
| 959 | if (u_size > size) |
| 960 | rv = DS_ERROR_SPACE_MD; |
| 961 | if (rv != DS_UNCHANGED) |
| 962 | goto err_out; |
| 963 | } |
| 964 | |
| 965 | if (drbd_get_capacity(device->this_bdev) != size || |
| 966 | drbd_bm_capacity(device) != size) { |
| 967 | int err; |
| 968 | err = drbd_bm_resize(device, size, !(flags & DDSF_NO_RESYNC)); |
| 969 | if (unlikely(err)) { |
| 970 | /* currently there is only one error: ENOMEM! */ |
| 971 | size = drbd_bm_capacity(device); |
| 972 | if (size == 0) { |
| 973 | drbd_err(device, "OUT OF MEMORY! " |
| 974 | "Could not allocate bitmap!\n"); |
| 975 | } else { |
| 976 | drbd_err(device, "BM resizing failed. " |
| 977 | "Leaving size unchanged\n"); |
| 978 | } |
| 979 | rv = DS_ERROR; |
| 980 | } |
| 981 | /* racy, see comments above. */ |
| 982 | drbd_set_my_capacity(device, size); |
| 983 | md->la_size_sect = size; |
| 984 | drbd_info(device, "size = %s (%llu KB)\n", ppsize(ppb, size>>1), |
| 985 | (unsigned long long)size>>1); |
| 986 | } |
| 987 | if (rv <= DS_ERROR) |
| 988 | goto err_out; |
| 989 | |
| 990 | la_size_changed = (prev.last_agreed_sect != md->la_size_sect); |
| 991 | |
| 992 | md_moved = prev.md_offset != md->md_offset |
| 993 | || prev.md_size_sect != md->md_size_sect; |
| 994 | |
| 995 | if (la_size_changed || md_moved || rs) { |
| 996 | u32 prev_flags; |
| 997 | |
| 998 | /* We do some synchronous IO below, which may take some time. |
| 999 | * Clear the timer, to avoid scary "timer expired!" messages, |
| 1000 | * "Superblock" is written out at least twice below, anyways. */ |
| 1001 | del_timer(&device->md_sync_timer); |
| 1002 | |
| 1003 | /* We won't change the "al-extents" setting, we just may need |
| 1004 | * to move the on-disk location of the activity log ringbuffer. |
| 1005 | * Lock for transaction is good enough, it may well be "dirty" |
| 1006 | * or even "starving". */ |
| 1007 | wait_event(device->al_wait, lc_try_lock_for_transaction(device->act_log)); |
| 1008 | |
| 1009 | /* mark current on-disk bitmap and activity log as unreliable */ |
| 1010 | prev_flags = md->flags; |
| 1011 | md->flags |= MDF_FULL_SYNC | MDF_AL_DISABLED; |
| 1012 | drbd_md_write(device, buffer); |
| 1013 | |
| 1014 | drbd_al_initialize(device, buffer); |
| 1015 | |
| 1016 | drbd_info(device, "Writing the whole bitmap, %s\n", |
| 1017 | la_size_changed && md_moved ? "size changed and md moved" : |
| 1018 | la_size_changed ? "size changed" : "md moved"); |
| 1019 | /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */ |
| 1020 | drbd_bitmap_io(device, md_moved ? &drbd_bm_write_all : &drbd_bm_write, |
| 1021 | "size changed", BM_LOCKED_MASK); |
| 1022 | |
| 1023 | /* on-disk bitmap and activity log is authoritative again |
| 1024 | * (unless there was an IO error meanwhile...) */ |
| 1025 | md->flags = prev_flags; |
| 1026 | drbd_md_write(device, buffer); |
| 1027 | |
| 1028 | if (rs) |
| 1029 | drbd_info(device, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n", |
| 1030 | md->al_stripes, md->al_stripe_size_4k * 4); |
| 1031 | } |
| 1032 | |
| 1033 | if (size > prev.last_agreed_sect) |
| 1034 | rv = prev.last_agreed_sect ? DS_GREW : DS_GREW_FROM_ZERO; |
| 1035 | if (size < prev.last_agreed_sect) |
| 1036 | rv = DS_SHRUNK; |
| 1037 | |
| 1038 | if (0) { |
| 1039 | err_out: |
| 1040 | /* restore previous offset and sizes */ |
| 1041 | md->la_size_sect = prev.last_agreed_sect; |
| 1042 | md->md_offset = prev.md_offset; |
| 1043 | md->al_offset = prev.al_offset; |
| 1044 | md->bm_offset = prev.bm_offset; |
| 1045 | md->md_size_sect = prev.md_size_sect; |
| 1046 | md->al_stripes = prev.al_stripes; |
| 1047 | md->al_stripe_size_4k = prev.al_stripe_size_4k; |
| 1048 | md->al_size_4k = (u64)prev.al_stripes * prev.al_stripe_size_4k; |
| 1049 | } |
| 1050 | lc_unlock(device->act_log); |
| 1051 | wake_up(&device->al_wait); |
| 1052 | drbd_md_put_buffer(device); |
| 1053 | drbd_resume_io(device); |
| 1054 | |
| 1055 | return rv; |
| 1056 | } |
| 1057 | |
| 1058 | sector_t |
| 1059 | drbd_new_dev_size(struct drbd_device *device, struct drbd_backing_dev *bdev, |
| 1060 | sector_t u_size, int assume_peer_has_space) |
| 1061 | { |
| 1062 | sector_t p_size = device->p_size; /* partner's disk size. */ |
| 1063 | sector_t la_size_sect = bdev->md.la_size_sect; /* last agreed size. */ |
| 1064 | sector_t m_size; /* my size */ |
| 1065 | sector_t size = 0; |
| 1066 | |
| 1067 | m_size = drbd_get_max_capacity(bdev); |
| 1068 | |
| 1069 | if (device->state.conn < C_CONNECTED && assume_peer_has_space) { |
| 1070 | drbd_warn(device, "Resize while not connected was forced by the user!\n"); |
| 1071 | p_size = m_size; |
| 1072 | } |
| 1073 | |
| 1074 | if (p_size && m_size) { |
| 1075 | size = min_t(sector_t, p_size, m_size); |
| 1076 | } else { |
| 1077 | if (la_size_sect) { |
| 1078 | size = la_size_sect; |
| 1079 | if (m_size && m_size < size) |
| 1080 | size = m_size; |
| 1081 | if (p_size && p_size < size) |
| 1082 | size = p_size; |
| 1083 | } else { |
| 1084 | if (m_size) |
| 1085 | size = m_size; |
| 1086 | if (p_size) |
| 1087 | size = p_size; |
| 1088 | } |
| 1089 | } |
| 1090 | |
| 1091 | if (size == 0) |
| 1092 | drbd_err(device, "Both nodes diskless!\n"); |
| 1093 | |
| 1094 | if (u_size) { |
| 1095 | if (u_size > size) |
| 1096 | drbd_err(device, "Requested disk size is too big (%lu > %lu)\n", |
| 1097 | (unsigned long)u_size>>1, (unsigned long)size>>1); |
| 1098 | else |
| 1099 | size = u_size; |
| 1100 | } |
| 1101 | |
| 1102 | return size; |
| 1103 | } |
| 1104 | |
| 1105 | /** |
| 1106 | * drbd_check_al_size() - Ensures that the AL is of the right size |
| 1107 | * @device: DRBD device. |
| 1108 | * |
| 1109 | * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation |
| 1110 | * failed, and 0 on success. You should call drbd_md_sync() after you called |
| 1111 | * this function. |
| 1112 | */ |
| 1113 | static int drbd_check_al_size(struct drbd_device *device, struct disk_conf *dc) |
| 1114 | { |
| 1115 | struct lru_cache *n, *t; |
| 1116 | struct lc_element *e; |
| 1117 | unsigned int in_use; |
| 1118 | int i; |
| 1119 | |
| 1120 | if (device->act_log && |
| 1121 | device->act_log->nr_elements == dc->al_extents) |
| 1122 | return 0; |
| 1123 | |
| 1124 | in_use = 0; |
| 1125 | t = device->act_log; |
| 1126 | n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION, |
| 1127 | dc->al_extents, sizeof(struct lc_element), 0); |
| 1128 | |
| 1129 | if (n == NULL) { |
| 1130 | drbd_err(device, "Cannot allocate act_log lru!\n"); |
| 1131 | return -ENOMEM; |
| 1132 | } |
| 1133 | spin_lock_irq(&device->al_lock); |
| 1134 | if (t) { |
| 1135 | for (i = 0; i < t->nr_elements; i++) { |
| 1136 | e = lc_element_by_index(t, i); |
| 1137 | if (e->refcnt) |
| 1138 | drbd_err(device, "refcnt(%d)==%d\n", |
| 1139 | e->lc_number, e->refcnt); |
| 1140 | in_use += e->refcnt; |
| 1141 | } |
| 1142 | } |
| 1143 | if (!in_use) |
| 1144 | device->act_log = n; |
| 1145 | spin_unlock_irq(&device->al_lock); |
| 1146 | if (in_use) { |
| 1147 | drbd_err(device, "Activity log still in use!\n"); |
| 1148 | lc_destroy(n); |
| 1149 | return -EBUSY; |
| 1150 | } else { |
| 1151 | lc_destroy(t); |
| 1152 | } |
| 1153 | drbd_md_mark_dirty(device); /* we changed device->act_log->nr_elemens */ |
| 1154 | return 0; |
| 1155 | } |
| 1156 | |
| 1157 | static void drbd_setup_queue_param(struct drbd_device *device, struct drbd_backing_dev *bdev, |
| 1158 | unsigned int max_bio_size) |
| 1159 | { |
| 1160 | struct request_queue * const q = device->rq_queue; |
| 1161 | unsigned int max_hw_sectors = max_bio_size >> 9; |
| 1162 | unsigned int max_segments = 0; |
| 1163 | struct request_queue *b = NULL; |
| 1164 | |
| 1165 | if (bdev) { |
| 1166 | b = bdev->backing_bdev->bd_disk->queue; |
| 1167 | |
| 1168 | max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9); |
| 1169 | rcu_read_lock(); |
| 1170 | max_segments = rcu_dereference(device->ldev->disk_conf)->max_bio_bvecs; |
| 1171 | rcu_read_unlock(); |
| 1172 | |
| 1173 | blk_set_stacking_limits(&q->limits); |
| 1174 | blk_queue_max_write_same_sectors(q, 0); |
| 1175 | } |
| 1176 | |
| 1177 | blk_queue_logical_block_size(q, 512); |
| 1178 | blk_queue_max_hw_sectors(q, max_hw_sectors); |
| 1179 | /* This is the workaround for "bio would need to, but cannot, be split" */ |
| 1180 | blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS); |
| 1181 | blk_queue_segment_boundary(q, PAGE_SIZE-1); |
| 1182 | |
| 1183 | if (b) { |
| 1184 | struct drbd_connection *connection = first_peer_device(device)->connection; |
| 1185 | |
| 1186 | blk_queue_max_discard_sectors(q, DRBD_MAX_DISCARD_SECTORS); |
| 1187 | |
| 1188 | if (blk_queue_discard(b) && |
| 1189 | (connection->cstate < C_CONNECTED || connection->agreed_features & FF_TRIM)) { |
| 1190 | /* We don't care, stacking below should fix it for the local device. |
| 1191 | * Whether or not it is a suitable granularity on the remote device |
| 1192 | * is not our problem, really. If you care, you need to |
| 1193 | * use devices with similar topology on all peers. */ |
| 1194 | q->limits.discard_granularity = 512; |
| 1195 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); |
| 1196 | } else { |
| 1197 | blk_queue_max_discard_sectors(q, 0); |
| 1198 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); |
| 1199 | q->limits.discard_granularity = 0; |
| 1200 | } |
| 1201 | |
| 1202 | blk_queue_stack_limits(q, b); |
| 1203 | |
| 1204 | if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) { |
| 1205 | drbd_info(device, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n", |
| 1206 | q->backing_dev_info.ra_pages, |
| 1207 | b->backing_dev_info.ra_pages); |
| 1208 | q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages; |
| 1209 | } |
| 1210 | } |
| 1211 | /* To avoid confusion, if this queue does not support discard, clear |
| 1212 | * max_discard_sectors, which is what lsblk -D reports to the user. */ |
| 1213 | if (!blk_queue_discard(q)) { |
| 1214 | blk_queue_max_discard_sectors(q, 0); |
| 1215 | q->limits.discard_granularity = 0; |
| 1216 | } |
| 1217 | } |
| 1218 | |
| 1219 | void drbd_reconsider_max_bio_size(struct drbd_device *device, struct drbd_backing_dev *bdev) |
| 1220 | { |
| 1221 | unsigned int now, new, local, peer; |
| 1222 | |
| 1223 | now = queue_max_hw_sectors(device->rq_queue) << 9; |
| 1224 | local = device->local_max_bio_size; /* Eventually last known value, from volatile memory */ |
| 1225 | peer = device->peer_max_bio_size; /* Eventually last known value, from meta data */ |
| 1226 | |
| 1227 | if (bdev) { |
| 1228 | local = queue_max_hw_sectors(bdev->backing_bdev->bd_disk->queue) << 9; |
| 1229 | device->local_max_bio_size = local; |
| 1230 | } |
| 1231 | local = min(local, DRBD_MAX_BIO_SIZE); |
| 1232 | |
| 1233 | /* We may ignore peer limits if the peer is modern enough. |
| 1234 | Because new from 8.3.8 onwards the peer can use multiple |
| 1235 | BIOs for a single peer_request */ |
| 1236 | if (device->state.conn >= C_WF_REPORT_PARAMS) { |
| 1237 | if (first_peer_device(device)->connection->agreed_pro_version < 94) |
| 1238 | peer = min(device->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET); |
| 1239 | /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */ |
| 1240 | else if (first_peer_device(device)->connection->agreed_pro_version == 94) |
| 1241 | peer = DRBD_MAX_SIZE_H80_PACKET; |
| 1242 | else if (first_peer_device(device)->connection->agreed_pro_version < 100) |
| 1243 | peer = DRBD_MAX_BIO_SIZE_P95; /* drbd 8.3.8 onwards, before 8.4.0 */ |
| 1244 | else |
| 1245 | peer = DRBD_MAX_BIO_SIZE; |
| 1246 | |
| 1247 | /* We may later detach and re-attach on a disconnected Primary. |
| 1248 | * Avoid this setting to jump back in that case. |
| 1249 | * We want to store what we know the peer DRBD can handle, |
| 1250 | * not what the peer IO backend can handle. */ |
| 1251 | if (peer > device->peer_max_bio_size) |
| 1252 | device->peer_max_bio_size = peer; |
| 1253 | } |
| 1254 | new = min(local, peer); |
| 1255 | |
| 1256 | if (device->state.role == R_PRIMARY && new < now) |
| 1257 | drbd_err(device, "ASSERT FAILED new < now; (%u < %u)\n", new, now); |
| 1258 | |
| 1259 | if (new != now) |
| 1260 | drbd_info(device, "max BIO size = %u\n", new); |
| 1261 | |
| 1262 | drbd_setup_queue_param(device, bdev, new); |
| 1263 | } |
| 1264 | |
| 1265 | /* Starts the worker thread */ |
| 1266 | static void conn_reconfig_start(struct drbd_connection *connection) |
| 1267 | { |
| 1268 | drbd_thread_start(&connection->worker); |
| 1269 | drbd_flush_workqueue(&connection->sender_work); |
| 1270 | } |
| 1271 | |
| 1272 | /* if still unconfigured, stops worker again. */ |
| 1273 | static void conn_reconfig_done(struct drbd_connection *connection) |
| 1274 | { |
| 1275 | bool stop_threads; |
| 1276 | spin_lock_irq(&connection->resource->req_lock); |
| 1277 | stop_threads = conn_all_vols_unconf(connection) && |
| 1278 | connection->cstate == C_STANDALONE; |
| 1279 | spin_unlock_irq(&connection->resource->req_lock); |
| 1280 | if (stop_threads) { |
| 1281 | /* ack_receiver thread and ack_sender workqueue are implicitly |
| 1282 | * stopped by receiver in conn_disconnect() */ |
| 1283 | drbd_thread_stop(&connection->receiver); |
| 1284 | drbd_thread_stop(&connection->worker); |
| 1285 | } |
| 1286 | } |
| 1287 | |
| 1288 | /* Make sure IO is suspended before calling this function(). */ |
| 1289 | static void drbd_suspend_al(struct drbd_device *device) |
| 1290 | { |
| 1291 | int s = 0; |
| 1292 | |
| 1293 | if (!lc_try_lock(device->act_log)) { |
| 1294 | drbd_warn(device, "Failed to lock al in drbd_suspend_al()\n"); |
| 1295 | return; |
| 1296 | } |
| 1297 | |
| 1298 | drbd_al_shrink(device); |
| 1299 | spin_lock_irq(&device->resource->req_lock); |
| 1300 | if (device->state.conn < C_CONNECTED) |
| 1301 | s = !test_and_set_bit(AL_SUSPENDED, &device->flags); |
| 1302 | spin_unlock_irq(&device->resource->req_lock); |
| 1303 | lc_unlock(device->act_log); |
| 1304 | |
| 1305 | if (s) |
| 1306 | drbd_info(device, "Suspended AL updates\n"); |
| 1307 | } |
| 1308 | |
| 1309 | |
| 1310 | static bool should_set_defaults(struct genl_info *info) |
| 1311 | { |
| 1312 | unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags; |
| 1313 | return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS); |
| 1314 | } |
| 1315 | |
| 1316 | static unsigned int drbd_al_extents_max(struct drbd_backing_dev *bdev) |
| 1317 | { |
| 1318 | /* This is limited by 16 bit "slot" numbers, |
| 1319 | * and by available on-disk context storage. |
| 1320 | * |
| 1321 | * Also (u16)~0 is special (denotes a "free" extent). |
| 1322 | * |
| 1323 | * One transaction occupies one 4kB on-disk block, |
| 1324 | * we have n such blocks in the on disk ring buffer, |
| 1325 | * the "current" transaction may fail (n-1), |
| 1326 | * and there is 919 slot numbers context information per transaction. |
| 1327 | * |
| 1328 | * 72 transaction blocks amounts to more than 2**16 context slots, |
| 1329 | * so cap there first. |
| 1330 | */ |
| 1331 | const unsigned int max_al_nr = DRBD_AL_EXTENTS_MAX; |
| 1332 | const unsigned int sufficient_on_disk = |
| 1333 | (max_al_nr + AL_CONTEXT_PER_TRANSACTION -1) |
| 1334 | /AL_CONTEXT_PER_TRANSACTION; |
| 1335 | |
| 1336 | unsigned int al_size_4k = bdev->md.al_size_4k; |
| 1337 | |
| 1338 | if (al_size_4k > sufficient_on_disk) |
| 1339 | return max_al_nr; |
| 1340 | |
| 1341 | return (al_size_4k - 1) * AL_CONTEXT_PER_TRANSACTION; |
| 1342 | } |
| 1343 | |
| 1344 | static bool write_ordering_changed(struct disk_conf *a, struct disk_conf *b) |
| 1345 | { |
| 1346 | return a->disk_barrier != b->disk_barrier || |
| 1347 | a->disk_flushes != b->disk_flushes || |
| 1348 | a->disk_drain != b->disk_drain; |
| 1349 | } |
| 1350 | |
| 1351 | int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info) |
| 1352 | { |
| 1353 | struct drbd_config_context adm_ctx; |
| 1354 | enum drbd_ret_code retcode; |
| 1355 | struct drbd_device *device; |
| 1356 | struct disk_conf *new_disk_conf, *old_disk_conf; |
| 1357 | struct fifo_buffer *old_plan = NULL, *new_plan = NULL; |
| 1358 | int err, fifo_size; |
| 1359 | |
| 1360 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 1361 | if (!adm_ctx.reply_skb) |
| 1362 | return retcode; |
| 1363 | if (retcode != NO_ERROR) |
| 1364 | goto finish; |
| 1365 | |
| 1366 | device = adm_ctx.device; |
| 1367 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 1368 | |
| 1369 | /* we also need a disk |
| 1370 | * to change the options on */ |
| 1371 | if (!get_ldev(device)) { |
| 1372 | retcode = ERR_NO_DISK; |
| 1373 | goto out; |
| 1374 | } |
| 1375 | |
| 1376 | new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL); |
| 1377 | if (!new_disk_conf) { |
| 1378 | retcode = ERR_NOMEM; |
| 1379 | goto fail; |
| 1380 | } |
| 1381 | |
| 1382 | mutex_lock(&device->resource->conf_update); |
| 1383 | old_disk_conf = device->ldev->disk_conf; |
| 1384 | *new_disk_conf = *old_disk_conf; |
| 1385 | if (should_set_defaults(info)) |
| 1386 | set_disk_conf_defaults(new_disk_conf); |
| 1387 | |
| 1388 | err = disk_conf_from_attrs_for_change(new_disk_conf, info); |
| 1389 | if (err && err != -ENOMSG) { |
| 1390 | retcode = ERR_MANDATORY_TAG; |
| 1391 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 1392 | goto fail_unlock; |
| 1393 | } |
| 1394 | |
| 1395 | if (!expect(new_disk_conf->resync_rate >= 1)) |
| 1396 | new_disk_conf->resync_rate = 1; |
| 1397 | |
| 1398 | if (new_disk_conf->al_extents < DRBD_AL_EXTENTS_MIN) |
| 1399 | new_disk_conf->al_extents = DRBD_AL_EXTENTS_MIN; |
| 1400 | if (new_disk_conf->al_extents > drbd_al_extents_max(device->ldev)) |
| 1401 | new_disk_conf->al_extents = drbd_al_extents_max(device->ldev); |
| 1402 | |
| 1403 | if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX) |
| 1404 | new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX; |
| 1405 | |
| 1406 | fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ; |
| 1407 | if (fifo_size != device->rs_plan_s->size) { |
| 1408 | new_plan = fifo_alloc(fifo_size); |
| 1409 | if (!new_plan) { |
| 1410 | drbd_err(device, "kmalloc of fifo_buffer failed"); |
| 1411 | retcode = ERR_NOMEM; |
| 1412 | goto fail_unlock; |
| 1413 | } |
| 1414 | } |
| 1415 | |
| 1416 | drbd_suspend_io(device); |
| 1417 | wait_event(device->al_wait, lc_try_lock(device->act_log)); |
| 1418 | drbd_al_shrink(device); |
| 1419 | err = drbd_check_al_size(device, new_disk_conf); |
| 1420 | lc_unlock(device->act_log); |
| 1421 | wake_up(&device->al_wait); |
| 1422 | drbd_resume_io(device); |
| 1423 | |
| 1424 | if (err) { |
| 1425 | retcode = ERR_NOMEM; |
| 1426 | goto fail_unlock; |
| 1427 | } |
| 1428 | |
| 1429 | lock_all_resources(); |
| 1430 | retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after); |
| 1431 | if (retcode == NO_ERROR) { |
| 1432 | rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); |
| 1433 | drbd_resync_after_changed(device); |
| 1434 | } |
| 1435 | unlock_all_resources(); |
| 1436 | |
| 1437 | if (retcode != NO_ERROR) |
| 1438 | goto fail_unlock; |
| 1439 | |
| 1440 | if (new_plan) { |
| 1441 | old_plan = device->rs_plan_s; |
| 1442 | rcu_assign_pointer(device->rs_plan_s, new_plan); |
| 1443 | } |
| 1444 | |
| 1445 | mutex_unlock(&device->resource->conf_update); |
| 1446 | |
| 1447 | if (new_disk_conf->al_updates) |
| 1448 | device->ldev->md.flags &= ~MDF_AL_DISABLED; |
| 1449 | else |
| 1450 | device->ldev->md.flags |= MDF_AL_DISABLED; |
| 1451 | |
| 1452 | if (new_disk_conf->md_flushes) |
| 1453 | clear_bit(MD_NO_FUA, &device->flags); |
| 1454 | else |
| 1455 | set_bit(MD_NO_FUA, &device->flags); |
| 1456 | |
| 1457 | if (write_ordering_changed(old_disk_conf, new_disk_conf)) |
| 1458 | drbd_bump_write_ordering(device->resource, NULL, WO_BDEV_FLUSH); |
| 1459 | |
| 1460 | drbd_md_sync(device); |
| 1461 | |
| 1462 | if (device->state.conn >= C_CONNECTED) { |
| 1463 | struct drbd_peer_device *peer_device; |
| 1464 | |
| 1465 | for_each_peer_device(peer_device, device) |
| 1466 | drbd_send_sync_param(peer_device); |
| 1467 | } |
| 1468 | |
| 1469 | synchronize_rcu(); |
| 1470 | kfree(old_disk_conf); |
| 1471 | kfree(old_plan); |
| 1472 | mod_timer(&device->request_timer, jiffies + HZ); |
| 1473 | goto success; |
| 1474 | |
| 1475 | fail_unlock: |
| 1476 | mutex_unlock(&device->resource->conf_update); |
| 1477 | fail: |
| 1478 | kfree(new_disk_conf); |
| 1479 | kfree(new_plan); |
| 1480 | success: |
| 1481 | put_ldev(device); |
| 1482 | out: |
| 1483 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 1484 | finish: |
| 1485 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 1486 | return 0; |
| 1487 | } |
| 1488 | |
| 1489 | static struct block_device *open_backing_dev(struct drbd_device *device, |
| 1490 | const char *bdev_path, void *claim_ptr, bool do_bd_link) |
| 1491 | { |
| 1492 | struct block_device *bdev; |
| 1493 | int err = 0; |
| 1494 | |
| 1495 | bdev = blkdev_get_by_path(bdev_path, |
| 1496 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, claim_ptr); |
| 1497 | if (IS_ERR(bdev)) { |
| 1498 | drbd_err(device, "open(\"%s\") failed with %ld\n", |
| 1499 | bdev_path, PTR_ERR(bdev)); |
| 1500 | return bdev; |
| 1501 | } |
| 1502 | |
| 1503 | if (!do_bd_link) |
| 1504 | return bdev; |
| 1505 | |
| 1506 | err = bd_link_disk_holder(bdev, device->vdisk); |
| 1507 | if (err) { |
| 1508 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
| 1509 | drbd_err(device, "bd_link_disk_holder(\"%s\", ...) failed with %d\n", |
| 1510 | bdev_path, err); |
| 1511 | bdev = ERR_PTR(err); |
| 1512 | } |
| 1513 | return bdev; |
| 1514 | } |
| 1515 | |
| 1516 | static int open_backing_devices(struct drbd_device *device, |
| 1517 | struct disk_conf *new_disk_conf, |
| 1518 | struct drbd_backing_dev *nbc) |
| 1519 | { |
| 1520 | struct block_device *bdev; |
| 1521 | |
| 1522 | bdev = open_backing_dev(device, new_disk_conf->backing_dev, device, true); |
| 1523 | if (IS_ERR(bdev)) |
| 1524 | return ERR_OPEN_DISK; |
| 1525 | nbc->backing_bdev = bdev; |
| 1526 | |
| 1527 | /* |
| 1528 | * meta_dev_idx >= 0: external fixed size, possibly multiple |
| 1529 | * drbd sharing one meta device. TODO in that case, paranoia |
| 1530 | * check that [md_bdev, meta_dev_idx] is not yet used by some |
| 1531 | * other drbd minor! (if you use drbd.conf + drbdadm, that |
| 1532 | * should check it for you already; but if you don't, or |
| 1533 | * someone fooled it, we need to double check here) |
| 1534 | */ |
| 1535 | bdev = open_backing_dev(device, new_disk_conf->meta_dev, |
| 1536 | /* claim ptr: device, if claimed exclusively; shared drbd_m_holder, |
| 1537 | * if potentially shared with other drbd minors */ |
| 1538 | (new_disk_conf->meta_dev_idx < 0) ? (void*)device : (void*)drbd_m_holder, |
| 1539 | /* avoid double bd_claim_by_disk() for the same (source,target) tuple, |
| 1540 | * as would happen with internal metadata. */ |
| 1541 | (new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_FLEX_INT && |
| 1542 | new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_INTERNAL)); |
| 1543 | if (IS_ERR(bdev)) |
| 1544 | return ERR_OPEN_MD_DISK; |
| 1545 | nbc->md_bdev = bdev; |
| 1546 | return NO_ERROR; |
| 1547 | } |
| 1548 | |
| 1549 | static void close_backing_dev(struct drbd_device *device, struct block_device *bdev, |
| 1550 | bool do_bd_unlink) |
| 1551 | { |
| 1552 | if (!bdev) |
| 1553 | return; |
| 1554 | if (do_bd_unlink) |
| 1555 | bd_unlink_disk_holder(bdev, device->vdisk); |
| 1556 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
| 1557 | } |
| 1558 | |
| 1559 | void drbd_backing_dev_free(struct drbd_device *device, struct drbd_backing_dev *ldev) |
| 1560 | { |
| 1561 | if (ldev == NULL) |
| 1562 | return; |
| 1563 | |
| 1564 | close_backing_dev(device, ldev->md_bdev, ldev->md_bdev != ldev->backing_bdev); |
| 1565 | close_backing_dev(device, ldev->backing_bdev, true); |
| 1566 | |
| 1567 | kfree(ldev->disk_conf); |
| 1568 | kfree(ldev); |
| 1569 | } |
| 1570 | |
| 1571 | int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info) |
| 1572 | { |
| 1573 | struct drbd_config_context adm_ctx; |
| 1574 | struct drbd_device *device; |
| 1575 | struct drbd_peer_device *peer_device; |
| 1576 | struct drbd_connection *connection; |
| 1577 | int err; |
| 1578 | enum drbd_ret_code retcode; |
| 1579 | enum determine_dev_size dd; |
| 1580 | sector_t max_possible_sectors; |
| 1581 | sector_t min_md_device_sectors; |
| 1582 | struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */ |
| 1583 | struct disk_conf *new_disk_conf = NULL; |
| 1584 | struct lru_cache *resync_lru = NULL; |
| 1585 | struct fifo_buffer *new_plan = NULL; |
| 1586 | union drbd_state ns, os; |
| 1587 | enum drbd_state_rv rv; |
| 1588 | struct net_conf *nc; |
| 1589 | |
| 1590 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 1591 | if (!adm_ctx.reply_skb) |
| 1592 | return retcode; |
| 1593 | if (retcode != NO_ERROR) |
| 1594 | goto finish; |
| 1595 | |
| 1596 | device = adm_ctx.device; |
| 1597 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 1598 | peer_device = first_peer_device(device); |
| 1599 | connection = peer_device->connection; |
| 1600 | conn_reconfig_start(connection); |
| 1601 | |
| 1602 | /* if you want to reconfigure, please tear down first */ |
| 1603 | if (device->state.disk > D_DISKLESS) { |
| 1604 | retcode = ERR_DISK_CONFIGURED; |
| 1605 | goto fail; |
| 1606 | } |
| 1607 | /* It may just now have detached because of IO error. Make sure |
| 1608 | * drbd_ldev_destroy is done already, we may end up here very fast, |
| 1609 | * e.g. if someone calls attach from the on-io-error handler, |
| 1610 | * to realize a "hot spare" feature (not that I'd recommend that) */ |
| 1611 | wait_event(device->misc_wait, !test_bit(GOING_DISKLESS, &device->flags)); |
| 1612 | |
| 1613 | /* make sure there is no leftover from previous force-detach attempts */ |
| 1614 | clear_bit(FORCE_DETACH, &device->flags); |
| 1615 | clear_bit(WAS_IO_ERROR, &device->flags); |
| 1616 | clear_bit(WAS_READ_ERROR, &device->flags); |
| 1617 | |
| 1618 | /* and no leftover from previously aborted resync or verify, either */ |
| 1619 | device->rs_total = 0; |
| 1620 | device->rs_failed = 0; |
| 1621 | atomic_set(&device->rs_pending_cnt, 0); |
| 1622 | |
| 1623 | /* allocation not in the IO path, drbdsetup context */ |
| 1624 | nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL); |
| 1625 | if (!nbc) { |
| 1626 | retcode = ERR_NOMEM; |
| 1627 | goto fail; |
| 1628 | } |
| 1629 | spin_lock_init(&nbc->md.uuid_lock); |
| 1630 | |
| 1631 | new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); |
| 1632 | if (!new_disk_conf) { |
| 1633 | retcode = ERR_NOMEM; |
| 1634 | goto fail; |
| 1635 | } |
| 1636 | nbc->disk_conf = new_disk_conf; |
| 1637 | |
| 1638 | set_disk_conf_defaults(new_disk_conf); |
| 1639 | err = disk_conf_from_attrs(new_disk_conf, info); |
| 1640 | if (err) { |
| 1641 | retcode = ERR_MANDATORY_TAG; |
| 1642 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 1643 | goto fail; |
| 1644 | } |
| 1645 | |
| 1646 | if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX) |
| 1647 | new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX; |
| 1648 | |
| 1649 | new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ); |
| 1650 | if (!new_plan) { |
| 1651 | retcode = ERR_NOMEM; |
| 1652 | goto fail; |
| 1653 | } |
| 1654 | |
| 1655 | if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) { |
| 1656 | retcode = ERR_MD_IDX_INVALID; |
| 1657 | goto fail; |
| 1658 | } |
| 1659 | |
| 1660 | rcu_read_lock(); |
| 1661 | nc = rcu_dereference(connection->net_conf); |
| 1662 | if (nc) { |
| 1663 | if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) { |
| 1664 | rcu_read_unlock(); |
| 1665 | retcode = ERR_STONITH_AND_PROT_A; |
| 1666 | goto fail; |
| 1667 | } |
| 1668 | } |
| 1669 | rcu_read_unlock(); |
| 1670 | |
| 1671 | retcode = open_backing_devices(device, new_disk_conf, nbc); |
| 1672 | if (retcode != NO_ERROR) |
| 1673 | goto fail; |
| 1674 | |
| 1675 | if ((nbc->backing_bdev == nbc->md_bdev) != |
| 1676 | (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL || |
| 1677 | new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) { |
| 1678 | retcode = ERR_MD_IDX_INVALID; |
| 1679 | goto fail; |
| 1680 | } |
| 1681 | |
| 1682 | resync_lru = lc_create("resync", drbd_bm_ext_cache, |
| 1683 | 1, 61, sizeof(struct bm_extent), |
| 1684 | offsetof(struct bm_extent, lce)); |
| 1685 | if (!resync_lru) { |
| 1686 | retcode = ERR_NOMEM; |
| 1687 | goto fail; |
| 1688 | } |
| 1689 | |
| 1690 | /* Read our meta data super block early. |
| 1691 | * This also sets other on-disk offsets. */ |
| 1692 | retcode = drbd_md_read(device, nbc); |
| 1693 | if (retcode != NO_ERROR) |
| 1694 | goto fail; |
| 1695 | |
| 1696 | if (new_disk_conf->al_extents < DRBD_AL_EXTENTS_MIN) |
| 1697 | new_disk_conf->al_extents = DRBD_AL_EXTENTS_MIN; |
| 1698 | if (new_disk_conf->al_extents > drbd_al_extents_max(nbc)) |
| 1699 | new_disk_conf->al_extents = drbd_al_extents_max(nbc); |
| 1700 | |
| 1701 | if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) { |
| 1702 | drbd_err(device, "max capacity %llu smaller than disk size %llu\n", |
| 1703 | (unsigned long long) drbd_get_max_capacity(nbc), |
| 1704 | (unsigned long long) new_disk_conf->disk_size); |
| 1705 | retcode = ERR_DISK_TOO_SMALL; |
| 1706 | goto fail; |
| 1707 | } |
| 1708 | |
| 1709 | if (new_disk_conf->meta_dev_idx < 0) { |
| 1710 | max_possible_sectors = DRBD_MAX_SECTORS_FLEX; |
| 1711 | /* at least one MB, otherwise it does not make sense */ |
| 1712 | min_md_device_sectors = (2<<10); |
| 1713 | } else { |
| 1714 | max_possible_sectors = DRBD_MAX_SECTORS; |
| 1715 | min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1); |
| 1716 | } |
| 1717 | |
| 1718 | if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) { |
| 1719 | retcode = ERR_MD_DISK_TOO_SMALL; |
| 1720 | drbd_warn(device, "refusing attach: md-device too small, " |
| 1721 | "at least %llu sectors needed for this meta-disk type\n", |
| 1722 | (unsigned long long) min_md_device_sectors); |
| 1723 | goto fail; |
| 1724 | } |
| 1725 | |
| 1726 | /* Make sure the new disk is big enough |
| 1727 | * (we may currently be R_PRIMARY with no local disk...) */ |
| 1728 | if (drbd_get_max_capacity(nbc) < |
| 1729 | drbd_get_capacity(device->this_bdev)) { |
| 1730 | retcode = ERR_DISK_TOO_SMALL; |
| 1731 | goto fail; |
| 1732 | } |
| 1733 | |
| 1734 | nbc->known_size = drbd_get_capacity(nbc->backing_bdev); |
| 1735 | |
| 1736 | if (nbc->known_size > max_possible_sectors) { |
| 1737 | drbd_warn(device, "==> truncating very big lower level device " |
| 1738 | "to currently maximum possible %llu sectors <==\n", |
| 1739 | (unsigned long long) max_possible_sectors); |
| 1740 | if (new_disk_conf->meta_dev_idx >= 0) |
| 1741 | drbd_warn(device, "==>> using internal or flexible " |
| 1742 | "meta data may help <<==\n"); |
| 1743 | } |
| 1744 | |
| 1745 | drbd_suspend_io(device); |
| 1746 | /* also wait for the last barrier ack. */ |
| 1747 | /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171 |
| 1748 | * We need a way to either ignore barrier acks for barriers sent before a device |
| 1749 | * was attached, or a way to wait for all pending barrier acks to come in. |
| 1750 | * As barriers are counted per resource, |
| 1751 | * we'd need to suspend io on all devices of a resource. |
| 1752 | */ |
| 1753 | wait_event(device->misc_wait, !atomic_read(&device->ap_pending_cnt) || drbd_suspended(device)); |
| 1754 | /* and for any other previously queued work */ |
| 1755 | drbd_flush_workqueue(&connection->sender_work); |
| 1756 | |
| 1757 | rv = _drbd_request_state(device, NS(disk, D_ATTACHING), CS_VERBOSE); |
| 1758 | retcode = rv; /* FIXME: Type mismatch. */ |
| 1759 | drbd_resume_io(device); |
| 1760 | if (rv < SS_SUCCESS) |
| 1761 | goto fail; |
| 1762 | |
| 1763 | if (!get_ldev_if_state(device, D_ATTACHING)) |
| 1764 | goto force_diskless; |
| 1765 | |
| 1766 | if (!device->bitmap) { |
| 1767 | if (drbd_bm_init(device)) { |
| 1768 | retcode = ERR_NOMEM; |
| 1769 | goto force_diskless_dec; |
| 1770 | } |
| 1771 | } |
| 1772 | |
| 1773 | if (device->state.conn < C_CONNECTED && |
| 1774 | device->state.role == R_PRIMARY && device->ed_uuid && |
| 1775 | (device->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) { |
| 1776 | drbd_err(device, "Can only attach to data with current UUID=%016llX\n", |
| 1777 | (unsigned long long)device->ed_uuid); |
| 1778 | retcode = ERR_DATA_NOT_CURRENT; |
| 1779 | goto force_diskless_dec; |
| 1780 | } |
| 1781 | |
| 1782 | /* Since we are diskless, fix the activity log first... */ |
| 1783 | if (drbd_check_al_size(device, new_disk_conf)) { |
| 1784 | retcode = ERR_NOMEM; |
| 1785 | goto force_diskless_dec; |
| 1786 | } |
| 1787 | |
| 1788 | /* Prevent shrinking of consistent devices ! */ |
| 1789 | if (drbd_md_test_flag(nbc, MDF_CONSISTENT) && |
| 1790 | drbd_new_dev_size(device, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) { |
| 1791 | drbd_warn(device, "refusing to truncate a consistent device\n"); |
| 1792 | retcode = ERR_DISK_TOO_SMALL; |
| 1793 | goto force_diskless_dec; |
| 1794 | } |
| 1795 | |
| 1796 | lock_all_resources(); |
| 1797 | retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after); |
| 1798 | if (retcode != NO_ERROR) { |
| 1799 | unlock_all_resources(); |
| 1800 | goto force_diskless_dec; |
| 1801 | } |
| 1802 | |
| 1803 | /* Reset the "barriers don't work" bits here, then force meta data to |
| 1804 | * be written, to ensure we determine if barriers are supported. */ |
| 1805 | if (new_disk_conf->md_flushes) |
| 1806 | clear_bit(MD_NO_FUA, &device->flags); |
| 1807 | else |
| 1808 | set_bit(MD_NO_FUA, &device->flags); |
| 1809 | |
| 1810 | /* Point of no return reached. |
| 1811 | * Devices and memory are no longer released by error cleanup below. |
| 1812 | * now device takes over responsibility, and the state engine should |
| 1813 | * clean it up somewhere. */ |
| 1814 | D_ASSERT(device, device->ldev == NULL); |
| 1815 | device->ldev = nbc; |
| 1816 | device->resync = resync_lru; |
| 1817 | device->rs_plan_s = new_plan; |
| 1818 | nbc = NULL; |
| 1819 | resync_lru = NULL; |
| 1820 | new_disk_conf = NULL; |
| 1821 | new_plan = NULL; |
| 1822 | |
| 1823 | drbd_resync_after_changed(device); |
| 1824 | drbd_bump_write_ordering(device->resource, device->ldev, WO_BDEV_FLUSH); |
| 1825 | unlock_all_resources(); |
| 1826 | |
| 1827 | if (drbd_md_test_flag(device->ldev, MDF_CRASHED_PRIMARY)) |
| 1828 | set_bit(CRASHED_PRIMARY, &device->flags); |
| 1829 | else |
| 1830 | clear_bit(CRASHED_PRIMARY, &device->flags); |
| 1831 | |
| 1832 | if (drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) && |
| 1833 | !(device->state.role == R_PRIMARY && device->resource->susp_nod)) |
| 1834 | set_bit(CRASHED_PRIMARY, &device->flags); |
| 1835 | |
| 1836 | device->send_cnt = 0; |
| 1837 | device->recv_cnt = 0; |
| 1838 | device->read_cnt = 0; |
| 1839 | device->writ_cnt = 0; |
| 1840 | |
| 1841 | drbd_reconsider_max_bio_size(device, device->ldev); |
| 1842 | |
| 1843 | /* If I am currently not R_PRIMARY, |
| 1844 | * but meta data primary indicator is set, |
| 1845 | * I just now recover from a hard crash, |
| 1846 | * and have been R_PRIMARY before that crash. |
| 1847 | * |
| 1848 | * Now, if I had no connection before that crash |
| 1849 | * (have been degraded R_PRIMARY), chances are that |
| 1850 | * I won't find my peer now either. |
| 1851 | * |
| 1852 | * In that case, and _only_ in that case, |
| 1853 | * we use the degr-wfc-timeout instead of the default, |
| 1854 | * so we can automatically recover from a crash of a |
| 1855 | * degraded but active "cluster" after a certain timeout. |
| 1856 | */ |
| 1857 | clear_bit(USE_DEGR_WFC_T, &device->flags); |
| 1858 | if (device->state.role != R_PRIMARY && |
| 1859 | drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) && |
| 1860 | !drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND)) |
| 1861 | set_bit(USE_DEGR_WFC_T, &device->flags); |
| 1862 | |
| 1863 | dd = drbd_determine_dev_size(device, 0, NULL); |
| 1864 | if (dd <= DS_ERROR) { |
| 1865 | retcode = ERR_NOMEM_BITMAP; |
| 1866 | goto force_diskless_dec; |
| 1867 | } else if (dd == DS_GREW) |
| 1868 | set_bit(RESYNC_AFTER_NEG, &device->flags); |
| 1869 | |
| 1870 | if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) || |
| 1871 | (test_bit(CRASHED_PRIMARY, &device->flags) && |
| 1872 | drbd_md_test_flag(device->ldev, MDF_AL_DISABLED))) { |
| 1873 | drbd_info(device, "Assuming that all blocks are out of sync " |
| 1874 | "(aka FullSync)\n"); |
| 1875 | if (drbd_bitmap_io(device, &drbd_bmio_set_n_write, |
| 1876 | "set_n_write from attaching", BM_LOCKED_MASK)) { |
| 1877 | retcode = ERR_IO_MD_DISK; |
| 1878 | goto force_diskless_dec; |
| 1879 | } |
| 1880 | } else { |
| 1881 | if (drbd_bitmap_io(device, &drbd_bm_read, |
| 1882 | "read from attaching", BM_LOCKED_MASK)) { |
| 1883 | retcode = ERR_IO_MD_DISK; |
| 1884 | goto force_diskless_dec; |
| 1885 | } |
| 1886 | } |
| 1887 | |
| 1888 | if (_drbd_bm_total_weight(device) == drbd_bm_bits(device)) |
| 1889 | drbd_suspend_al(device); /* IO is still suspended here... */ |
| 1890 | |
| 1891 | spin_lock_irq(&device->resource->req_lock); |
| 1892 | os = drbd_read_state(device); |
| 1893 | ns = os; |
| 1894 | /* If MDF_CONSISTENT is not set go into inconsistent state, |
| 1895 | otherwise investigate MDF_WasUpToDate... |
| 1896 | If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state, |
| 1897 | otherwise into D_CONSISTENT state. |
| 1898 | */ |
| 1899 | if (drbd_md_test_flag(device->ldev, MDF_CONSISTENT)) { |
| 1900 | if (drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE)) |
| 1901 | ns.disk = D_CONSISTENT; |
| 1902 | else |
| 1903 | ns.disk = D_OUTDATED; |
| 1904 | } else { |
| 1905 | ns.disk = D_INCONSISTENT; |
| 1906 | } |
| 1907 | |
| 1908 | if (drbd_md_test_flag(device->ldev, MDF_PEER_OUT_DATED)) |
| 1909 | ns.pdsk = D_OUTDATED; |
| 1910 | |
| 1911 | rcu_read_lock(); |
| 1912 | if (ns.disk == D_CONSISTENT && |
| 1913 | (ns.pdsk == D_OUTDATED || rcu_dereference(device->ldev->disk_conf)->fencing == FP_DONT_CARE)) |
| 1914 | ns.disk = D_UP_TO_DATE; |
| 1915 | |
| 1916 | /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND, |
| 1917 | MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before |
| 1918 | this point, because drbd_request_state() modifies these |
| 1919 | flags. */ |
| 1920 | |
| 1921 | if (rcu_dereference(device->ldev->disk_conf)->al_updates) |
| 1922 | device->ldev->md.flags &= ~MDF_AL_DISABLED; |
| 1923 | else |
| 1924 | device->ldev->md.flags |= MDF_AL_DISABLED; |
| 1925 | |
| 1926 | rcu_read_unlock(); |
| 1927 | |
| 1928 | /* In case we are C_CONNECTED postpone any decision on the new disk |
| 1929 | state after the negotiation phase. */ |
| 1930 | if (device->state.conn == C_CONNECTED) { |
| 1931 | device->new_state_tmp.i = ns.i; |
| 1932 | ns.i = os.i; |
| 1933 | ns.disk = D_NEGOTIATING; |
| 1934 | |
| 1935 | /* We expect to receive up-to-date UUIDs soon. |
| 1936 | To avoid a race in receive_state, free p_uuid while |
| 1937 | holding req_lock. I.e. atomic with the state change */ |
| 1938 | kfree(device->p_uuid); |
| 1939 | device->p_uuid = NULL; |
| 1940 | } |
| 1941 | |
| 1942 | rv = _drbd_set_state(device, ns, CS_VERBOSE, NULL); |
| 1943 | spin_unlock_irq(&device->resource->req_lock); |
| 1944 | |
| 1945 | if (rv < SS_SUCCESS) |
| 1946 | goto force_diskless_dec; |
| 1947 | |
| 1948 | mod_timer(&device->request_timer, jiffies + HZ); |
| 1949 | |
| 1950 | if (device->state.role == R_PRIMARY) |
| 1951 | device->ldev->md.uuid[UI_CURRENT] |= (u64)1; |
| 1952 | else |
| 1953 | device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1; |
| 1954 | |
| 1955 | drbd_md_mark_dirty(device); |
| 1956 | drbd_md_sync(device); |
| 1957 | |
| 1958 | kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE); |
| 1959 | put_ldev(device); |
| 1960 | conn_reconfig_done(connection); |
| 1961 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 1962 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 1963 | return 0; |
| 1964 | |
| 1965 | force_diskless_dec: |
| 1966 | put_ldev(device); |
| 1967 | force_diskless: |
| 1968 | drbd_force_state(device, NS(disk, D_DISKLESS)); |
| 1969 | drbd_md_sync(device); |
| 1970 | fail: |
| 1971 | conn_reconfig_done(connection); |
| 1972 | if (nbc) { |
| 1973 | close_backing_dev(device, nbc->md_bdev, nbc->md_bdev != nbc->backing_bdev); |
| 1974 | close_backing_dev(device, nbc->backing_bdev, true); |
| 1975 | kfree(nbc); |
| 1976 | } |
| 1977 | kfree(new_disk_conf); |
| 1978 | lc_destroy(resync_lru); |
| 1979 | kfree(new_plan); |
| 1980 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 1981 | finish: |
| 1982 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 1983 | return 0; |
| 1984 | } |
| 1985 | |
| 1986 | static int adm_detach(struct drbd_device *device, int force) |
| 1987 | { |
| 1988 | enum drbd_state_rv retcode; |
| 1989 | void *buffer; |
| 1990 | int ret; |
| 1991 | |
| 1992 | if (force) { |
| 1993 | set_bit(FORCE_DETACH, &device->flags); |
| 1994 | drbd_force_state(device, NS(disk, D_FAILED)); |
| 1995 | retcode = SS_SUCCESS; |
| 1996 | goto out; |
| 1997 | } |
| 1998 | |
| 1999 | drbd_suspend_io(device); /* so no-one is stuck in drbd_al_begin_io */ |
| 2000 | buffer = drbd_md_get_buffer(device, __func__); /* make sure there is no in-flight meta-data IO */ |
| 2001 | if (buffer) { |
| 2002 | retcode = drbd_request_state(device, NS(disk, D_FAILED)); |
| 2003 | drbd_md_put_buffer(device); |
| 2004 | } else /* already <= D_FAILED */ |
| 2005 | retcode = SS_NOTHING_TO_DO; |
| 2006 | /* D_FAILED will transition to DISKLESS. */ |
| 2007 | drbd_resume_io(device); |
| 2008 | ret = wait_event_interruptible(device->misc_wait, |
| 2009 | device->state.disk != D_FAILED); |
| 2010 | if ((int)retcode == (int)SS_IS_DISKLESS) |
| 2011 | retcode = SS_NOTHING_TO_DO; |
| 2012 | if (ret) |
| 2013 | retcode = ERR_INTR; |
| 2014 | out: |
| 2015 | return retcode; |
| 2016 | } |
| 2017 | |
| 2018 | /* Detaching the disk is a process in multiple stages. First we need to lock |
| 2019 | * out application IO, in-flight IO, IO stuck in drbd_al_begin_io. |
| 2020 | * Then we transition to D_DISKLESS, and wait for put_ldev() to return all |
| 2021 | * internal references as well. |
| 2022 | * Only then we have finally detached. */ |
| 2023 | int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info) |
| 2024 | { |
| 2025 | struct drbd_config_context adm_ctx; |
| 2026 | enum drbd_ret_code retcode; |
| 2027 | struct detach_parms parms = { }; |
| 2028 | int err; |
| 2029 | |
| 2030 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2031 | if (!adm_ctx.reply_skb) |
| 2032 | return retcode; |
| 2033 | if (retcode != NO_ERROR) |
| 2034 | goto out; |
| 2035 | |
| 2036 | if (info->attrs[DRBD_NLA_DETACH_PARMS]) { |
| 2037 | err = detach_parms_from_attrs(&parms, info); |
| 2038 | if (err) { |
| 2039 | retcode = ERR_MANDATORY_TAG; |
| 2040 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 2041 | goto out; |
| 2042 | } |
| 2043 | } |
| 2044 | |
| 2045 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2046 | retcode = adm_detach(adm_ctx.device, parms.force_detach); |
| 2047 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2048 | out: |
| 2049 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2050 | return 0; |
| 2051 | } |
| 2052 | |
| 2053 | static bool conn_resync_running(struct drbd_connection *connection) |
| 2054 | { |
| 2055 | struct drbd_peer_device *peer_device; |
| 2056 | bool rv = false; |
| 2057 | int vnr; |
| 2058 | |
| 2059 | rcu_read_lock(); |
| 2060 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
| 2061 | struct drbd_device *device = peer_device->device; |
| 2062 | if (device->state.conn == C_SYNC_SOURCE || |
| 2063 | device->state.conn == C_SYNC_TARGET || |
| 2064 | device->state.conn == C_PAUSED_SYNC_S || |
| 2065 | device->state.conn == C_PAUSED_SYNC_T) { |
| 2066 | rv = true; |
| 2067 | break; |
| 2068 | } |
| 2069 | } |
| 2070 | rcu_read_unlock(); |
| 2071 | |
| 2072 | return rv; |
| 2073 | } |
| 2074 | |
| 2075 | static bool conn_ov_running(struct drbd_connection *connection) |
| 2076 | { |
| 2077 | struct drbd_peer_device *peer_device; |
| 2078 | bool rv = false; |
| 2079 | int vnr; |
| 2080 | |
| 2081 | rcu_read_lock(); |
| 2082 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
| 2083 | struct drbd_device *device = peer_device->device; |
| 2084 | if (device->state.conn == C_VERIFY_S || |
| 2085 | device->state.conn == C_VERIFY_T) { |
| 2086 | rv = true; |
| 2087 | break; |
| 2088 | } |
| 2089 | } |
| 2090 | rcu_read_unlock(); |
| 2091 | |
| 2092 | return rv; |
| 2093 | } |
| 2094 | |
| 2095 | static enum drbd_ret_code |
| 2096 | _check_net_options(struct drbd_connection *connection, struct net_conf *old_net_conf, struct net_conf *new_net_conf) |
| 2097 | { |
| 2098 | struct drbd_peer_device *peer_device; |
| 2099 | int i; |
| 2100 | |
| 2101 | if (old_net_conf && connection->cstate == C_WF_REPORT_PARAMS && connection->agreed_pro_version < 100) { |
| 2102 | if (new_net_conf->wire_protocol != old_net_conf->wire_protocol) |
| 2103 | return ERR_NEED_APV_100; |
| 2104 | |
| 2105 | if (new_net_conf->two_primaries != old_net_conf->two_primaries) |
| 2106 | return ERR_NEED_APV_100; |
| 2107 | |
| 2108 | if (strcmp(new_net_conf->integrity_alg, old_net_conf->integrity_alg)) |
| 2109 | return ERR_NEED_APV_100; |
| 2110 | } |
| 2111 | |
| 2112 | if (!new_net_conf->two_primaries && |
| 2113 | conn_highest_role(connection) == R_PRIMARY && |
| 2114 | conn_highest_peer(connection) == R_PRIMARY) |
| 2115 | return ERR_NEED_ALLOW_TWO_PRI; |
| 2116 | |
| 2117 | if (new_net_conf->two_primaries && |
| 2118 | (new_net_conf->wire_protocol != DRBD_PROT_C)) |
| 2119 | return ERR_NOT_PROTO_C; |
| 2120 | |
| 2121 | idr_for_each_entry(&connection->peer_devices, peer_device, i) { |
| 2122 | struct drbd_device *device = peer_device->device; |
| 2123 | if (get_ldev(device)) { |
| 2124 | enum drbd_fencing_p fp = rcu_dereference(device->ldev->disk_conf)->fencing; |
| 2125 | put_ldev(device); |
| 2126 | if (new_net_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) |
| 2127 | return ERR_STONITH_AND_PROT_A; |
| 2128 | } |
| 2129 | if (device->state.role == R_PRIMARY && new_net_conf->discard_my_data) |
| 2130 | return ERR_DISCARD_IMPOSSIBLE; |
| 2131 | } |
| 2132 | |
| 2133 | if (new_net_conf->on_congestion != OC_BLOCK && new_net_conf->wire_protocol != DRBD_PROT_A) |
| 2134 | return ERR_CONG_NOT_PROTO_A; |
| 2135 | |
| 2136 | return NO_ERROR; |
| 2137 | } |
| 2138 | |
| 2139 | static enum drbd_ret_code |
| 2140 | check_net_options(struct drbd_connection *connection, struct net_conf *new_net_conf) |
| 2141 | { |
| 2142 | static enum drbd_ret_code rv; |
| 2143 | struct drbd_peer_device *peer_device; |
| 2144 | int i; |
| 2145 | |
| 2146 | rcu_read_lock(); |
| 2147 | rv = _check_net_options(connection, rcu_dereference(connection->net_conf), new_net_conf); |
| 2148 | rcu_read_unlock(); |
| 2149 | |
| 2150 | /* connection->peer_devices protected by genl_lock() here */ |
| 2151 | idr_for_each_entry(&connection->peer_devices, peer_device, i) { |
| 2152 | struct drbd_device *device = peer_device->device; |
| 2153 | if (!device->bitmap) { |
| 2154 | if (drbd_bm_init(device)) |
| 2155 | return ERR_NOMEM; |
| 2156 | } |
| 2157 | } |
| 2158 | |
| 2159 | return rv; |
| 2160 | } |
| 2161 | |
| 2162 | struct crypto { |
| 2163 | struct crypto_ahash *verify_tfm; |
| 2164 | struct crypto_ahash *csums_tfm; |
| 2165 | struct crypto_shash *cram_hmac_tfm; |
| 2166 | struct crypto_ahash *integrity_tfm; |
| 2167 | }; |
| 2168 | |
| 2169 | static int |
| 2170 | alloc_shash(struct crypto_shash **tfm, char *tfm_name, int err_alg) |
| 2171 | { |
| 2172 | if (!tfm_name[0]) |
| 2173 | return NO_ERROR; |
| 2174 | |
| 2175 | *tfm = crypto_alloc_shash(tfm_name, 0, 0); |
| 2176 | if (IS_ERR(*tfm)) { |
| 2177 | *tfm = NULL; |
| 2178 | return err_alg; |
| 2179 | } |
| 2180 | |
| 2181 | return NO_ERROR; |
| 2182 | } |
| 2183 | |
| 2184 | static int |
| 2185 | alloc_ahash(struct crypto_ahash **tfm, char *tfm_name, int err_alg) |
| 2186 | { |
| 2187 | if (!tfm_name[0]) |
| 2188 | return NO_ERROR; |
| 2189 | |
| 2190 | *tfm = crypto_alloc_ahash(tfm_name, 0, CRYPTO_ALG_ASYNC); |
| 2191 | if (IS_ERR(*tfm)) { |
| 2192 | *tfm = NULL; |
| 2193 | return err_alg; |
| 2194 | } |
| 2195 | |
| 2196 | return NO_ERROR; |
| 2197 | } |
| 2198 | |
| 2199 | static enum drbd_ret_code |
| 2200 | alloc_crypto(struct crypto *crypto, struct net_conf *new_net_conf) |
| 2201 | { |
| 2202 | char hmac_name[CRYPTO_MAX_ALG_NAME]; |
| 2203 | enum drbd_ret_code rv; |
| 2204 | |
| 2205 | rv = alloc_ahash(&crypto->csums_tfm, new_net_conf->csums_alg, |
| 2206 | ERR_CSUMS_ALG); |
| 2207 | if (rv != NO_ERROR) |
| 2208 | return rv; |
| 2209 | rv = alloc_ahash(&crypto->verify_tfm, new_net_conf->verify_alg, |
| 2210 | ERR_VERIFY_ALG); |
| 2211 | if (rv != NO_ERROR) |
| 2212 | return rv; |
| 2213 | rv = alloc_ahash(&crypto->integrity_tfm, new_net_conf->integrity_alg, |
| 2214 | ERR_INTEGRITY_ALG); |
| 2215 | if (rv != NO_ERROR) |
| 2216 | return rv; |
| 2217 | if (new_net_conf->cram_hmac_alg[0] != 0) { |
| 2218 | snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)", |
| 2219 | new_net_conf->cram_hmac_alg); |
| 2220 | |
| 2221 | rv = alloc_shash(&crypto->cram_hmac_tfm, hmac_name, |
| 2222 | ERR_AUTH_ALG); |
| 2223 | } |
| 2224 | |
| 2225 | return rv; |
| 2226 | } |
| 2227 | |
| 2228 | static void free_crypto(struct crypto *crypto) |
| 2229 | { |
| 2230 | crypto_free_shash(crypto->cram_hmac_tfm); |
| 2231 | crypto_free_ahash(crypto->integrity_tfm); |
| 2232 | crypto_free_ahash(crypto->csums_tfm); |
| 2233 | crypto_free_ahash(crypto->verify_tfm); |
| 2234 | } |
| 2235 | |
| 2236 | int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info) |
| 2237 | { |
| 2238 | struct drbd_config_context adm_ctx; |
| 2239 | enum drbd_ret_code retcode; |
| 2240 | struct drbd_connection *connection; |
| 2241 | struct net_conf *old_net_conf, *new_net_conf = NULL; |
| 2242 | int err; |
| 2243 | int ovr; /* online verify running */ |
| 2244 | int rsr; /* re-sync running */ |
| 2245 | struct crypto crypto = { }; |
| 2246 | |
| 2247 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION); |
| 2248 | if (!adm_ctx.reply_skb) |
| 2249 | return retcode; |
| 2250 | if (retcode != NO_ERROR) |
| 2251 | goto finish; |
| 2252 | |
| 2253 | connection = adm_ctx.connection; |
| 2254 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2255 | |
| 2256 | new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL); |
| 2257 | if (!new_net_conf) { |
| 2258 | retcode = ERR_NOMEM; |
| 2259 | goto out; |
| 2260 | } |
| 2261 | |
| 2262 | conn_reconfig_start(connection); |
| 2263 | |
| 2264 | mutex_lock(&connection->data.mutex); |
| 2265 | mutex_lock(&connection->resource->conf_update); |
| 2266 | old_net_conf = connection->net_conf; |
| 2267 | |
| 2268 | if (!old_net_conf) { |
| 2269 | drbd_msg_put_info(adm_ctx.reply_skb, "net conf missing, try connect"); |
| 2270 | retcode = ERR_INVALID_REQUEST; |
| 2271 | goto fail; |
| 2272 | } |
| 2273 | |
| 2274 | *new_net_conf = *old_net_conf; |
| 2275 | if (should_set_defaults(info)) |
| 2276 | set_net_conf_defaults(new_net_conf); |
| 2277 | |
| 2278 | err = net_conf_from_attrs_for_change(new_net_conf, info); |
| 2279 | if (err && err != -ENOMSG) { |
| 2280 | retcode = ERR_MANDATORY_TAG; |
| 2281 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 2282 | goto fail; |
| 2283 | } |
| 2284 | |
| 2285 | retcode = check_net_options(connection, new_net_conf); |
| 2286 | if (retcode != NO_ERROR) |
| 2287 | goto fail; |
| 2288 | |
| 2289 | /* re-sync running */ |
| 2290 | rsr = conn_resync_running(connection); |
| 2291 | if (rsr && strcmp(new_net_conf->csums_alg, old_net_conf->csums_alg)) { |
| 2292 | retcode = ERR_CSUMS_RESYNC_RUNNING; |
| 2293 | goto fail; |
| 2294 | } |
| 2295 | |
| 2296 | /* online verify running */ |
| 2297 | ovr = conn_ov_running(connection); |
| 2298 | if (ovr && strcmp(new_net_conf->verify_alg, old_net_conf->verify_alg)) { |
| 2299 | retcode = ERR_VERIFY_RUNNING; |
| 2300 | goto fail; |
| 2301 | } |
| 2302 | |
| 2303 | retcode = alloc_crypto(&crypto, new_net_conf); |
| 2304 | if (retcode != NO_ERROR) |
| 2305 | goto fail; |
| 2306 | |
| 2307 | rcu_assign_pointer(connection->net_conf, new_net_conf); |
| 2308 | |
| 2309 | if (!rsr) { |
| 2310 | crypto_free_ahash(connection->csums_tfm); |
| 2311 | connection->csums_tfm = crypto.csums_tfm; |
| 2312 | crypto.csums_tfm = NULL; |
| 2313 | } |
| 2314 | if (!ovr) { |
| 2315 | crypto_free_ahash(connection->verify_tfm); |
| 2316 | connection->verify_tfm = crypto.verify_tfm; |
| 2317 | crypto.verify_tfm = NULL; |
| 2318 | } |
| 2319 | |
| 2320 | crypto_free_ahash(connection->integrity_tfm); |
| 2321 | connection->integrity_tfm = crypto.integrity_tfm; |
| 2322 | if (connection->cstate >= C_WF_REPORT_PARAMS && connection->agreed_pro_version >= 100) |
| 2323 | /* Do this without trying to take connection->data.mutex again. */ |
| 2324 | __drbd_send_protocol(connection, P_PROTOCOL_UPDATE); |
| 2325 | |
| 2326 | crypto_free_shash(connection->cram_hmac_tfm); |
| 2327 | connection->cram_hmac_tfm = crypto.cram_hmac_tfm; |
| 2328 | |
| 2329 | mutex_unlock(&connection->resource->conf_update); |
| 2330 | mutex_unlock(&connection->data.mutex); |
| 2331 | synchronize_rcu(); |
| 2332 | kfree(old_net_conf); |
| 2333 | |
| 2334 | if (connection->cstate >= C_WF_REPORT_PARAMS) { |
| 2335 | struct drbd_peer_device *peer_device; |
| 2336 | int vnr; |
| 2337 | |
| 2338 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) |
| 2339 | drbd_send_sync_param(peer_device); |
| 2340 | } |
| 2341 | |
| 2342 | goto done; |
| 2343 | |
| 2344 | fail: |
| 2345 | mutex_unlock(&connection->resource->conf_update); |
| 2346 | mutex_unlock(&connection->data.mutex); |
| 2347 | free_crypto(&crypto); |
| 2348 | kfree(new_net_conf); |
| 2349 | done: |
| 2350 | conn_reconfig_done(connection); |
| 2351 | out: |
| 2352 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2353 | finish: |
| 2354 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2355 | return 0; |
| 2356 | } |
| 2357 | |
| 2358 | static void connection_to_info(struct connection_info *info, |
| 2359 | struct drbd_connection *connection) |
| 2360 | { |
| 2361 | info->conn_connection_state = connection->cstate; |
| 2362 | info->conn_role = conn_highest_peer(connection); |
| 2363 | } |
| 2364 | |
| 2365 | static void peer_device_to_info(struct peer_device_info *info, |
| 2366 | struct drbd_peer_device *peer_device) |
| 2367 | { |
| 2368 | struct drbd_device *device = peer_device->device; |
| 2369 | |
| 2370 | info->peer_repl_state = |
| 2371 | max_t(enum drbd_conns, C_WF_REPORT_PARAMS, device->state.conn); |
| 2372 | info->peer_disk_state = device->state.pdsk; |
| 2373 | info->peer_resync_susp_user = device->state.user_isp; |
| 2374 | info->peer_resync_susp_peer = device->state.peer_isp; |
| 2375 | info->peer_resync_susp_dependency = device->state.aftr_isp; |
| 2376 | } |
| 2377 | |
| 2378 | int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info) |
| 2379 | { |
| 2380 | struct connection_info connection_info; |
| 2381 | enum drbd_notification_type flags; |
| 2382 | unsigned int peer_devices = 0; |
| 2383 | struct drbd_config_context adm_ctx; |
| 2384 | struct drbd_peer_device *peer_device; |
| 2385 | struct net_conf *old_net_conf, *new_net_conf = NULL; |
| 2386 | struct crypto crypto = { }; |
| 2387 | struct drbd_resource *resource; |
| 2388 | struct drbd_connection *connection; |
| 2389 | enum drbd_ret_code retcode; |
| 2390 | int i; |
| 2391 | int err; |
| 2392 | |
| 2393 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE); |
| 2394 | |
| 2395 | if (!adm_ctx.reply_skb) |
| 2396 | return retcode; |
| 2397 | if (retcode != NO_ERROR) |
| 2398 | goto out; |
| 2399 | if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) { |
| 2400 | drbd_msg_put_info(adm_ctx.reply_skb, "connection endpoint(s) missing"); |
| 2401 | retcode = ERR_INVALID_REQUEST; |
| 2402 | goto out; |
| 2403 | } |
| 2404 | |
| 2405 | /* No need for _rcu here. All reconfiguration is |
| 2406 | * strictly serialized on genl_lock(). We are protected against |
| 2407 | * concurrent reconfiguration/addition/deletion */ |
| 2408 | for_each_resource(resource, &drbd_resources) { |
| 2409 | for_each_connection(connection, resource) { |
| 2410 | if (nla_len(adm_ctx.my_addr) == connection->my_addr_len && |
| 2411 | !memcmp(nla_data(adm_ctx.my_addr), &connection->my_addr, |
| 2412 | connection->my_addr_len)) { |
| 2413 | retcode = ERR_LOCAL_ADDR; |
| 2414 | goto out; |
| 2415 | } |
| 2416 | |
| 2417 | if (nla_len(adm_ctx.peer_addr) == connection->peer_addr_len && |
| 2418 | !memcmp(nla_data(adm_ctx.peer_addr), &connection->peer_addr, |
| 2419 | connection->peer_addr_len)) { |
| 2420 | retcode = ERR_PEER_ADDR; |
| 2421 | goto out; |
| 2422 | } |
| 2423 | } |
| 2424 | } |
| 2425 | |
| 2426 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2427 | connection = first_connection(adm_ctx.resource); |
| 2428 | conn_reconfig_start(connection); |
| 2429 | |
| 2430 | if (connection->cstate > C_STANDALONE) { |
| 2431 | retcode = ERR_NET_CONFIGURED; |
| 2432 | goto fail; |
| 2433 | } |
| 2434 | |
| 2435 | /* allocation not in the IO path, drbdsetup / netlink process context */ |
| 2436 | new_net_conf = kzalloc(sizeof(*new_net_conf), GFP_KERNEL); |
| 2437 | if (!new_net_conf) { |
| 2438 | retcode = ERR_NOMEM; |
| 2439 | goto fail; |
| 2440 | } |
| 2441 | |
| 2442 | set_net_conf_defaults(new_net_conf); |
| 2443 | |
| 2444 | err = net_conf_from_attrs(new_net_conf, info); |
| 2445 | if (err && err != -ENOMSG) { |
| 2446 | retcode = ERR_MANDATORY_TAG; |
| 2447 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 2448 | goto fail; |
| 2449 | } |
| 2450 | |
| 2451 | retcode = check_net_options(connection, new_net_conf); |
| 2452 | if (retcode != NO_ERROR) |
| 2453 | goto fail; |
| 2454 | |
| 2455 | retcode = alloc_crypto(&crypto, new_net_conf); |
| 2456 | if (retcode != NO_ERROR) |
| 2457 | goto fail; |
| 2458 | |
| 2459 | ((char *)new_net_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0; |
| 2460 | |
| 2461 | drbd_flush_workqueue(&connection->sender_work); |
| 2462 | |
| 2463 | mutex_lock(&adm_ctx.resource->conf_update); |
| 2464 | old_net_conf = connection->net_conf; |
| 2465 | if (old_net_conf) { |
| 2466 | retcode = ERR_NET_CONFIGURED; |
| 2467 | mutex_unlock(&adm_ctx.resource->conf_update); |
| 2468 | goto fail; |
| 2469 | } |
| 2470 | rcu_assign_pointer(connection->net_conf, new_net_conf); |
| 2471 | |
| 2472 | conn_free_crypto(connection); |
| 2473 | connection->cram_hmac_tfm = crypto.cram_hmac_tfm; |
| 2474 | connection->integrity_tfm = crypto.integrity_tfm; |
| 2475 | connection->csums_tfm = crypto.csums_tfm; |
| 2476 | connection->verify_tfm = crypto.verify_tfm; |
| 2477 | |
| 2478 | connection->my_addr_len = nla_len(adm_ctx.my_addr); |
| 2479 | memcpy(&connection->my_addr, nla_data(adm_ctx.my_addr), connection->my_addr_len); |
| 2480 | connection->peer_addr_len = nla_len(adm_ctx.peer_addr); |
| 2481 | memcpy(&connection->peer_addr, nla_data(adm_ctx.peer_addr), connection->peer_addr_len); |
| 2482 | |
| 2483 | idr_for_each_entry(&connection->peer_devices, peer_device, i) { |
| 2484 | peer_devices++; |
| 2485 | } |
| 2486 | |
| 2487 | connection_to_info(&connection_info, connection); |
| 2488 | flags = (peer_devices--) ? NOTIFY_CONTINUES : 0; |
| 2489 | mutex_lock(¬ification_mutex); |
| 2490 | notify_connection_state(NULL, 0, connection, &connection_info, NOTIFY_CREATE | flags); |
| 2491 | idr_for_each_entry(&connection->peer_devices, peer_device, i) { |
| 2492 | struct peer_device_info peer_device_info; |
| 2493 | |
| 2494 | peer_device_to_info(&peer_device_info, peer_device); |
| 2495 | flags = (peer_devices--) ? NOTIFY_CONTINUES : 0; |
| 2496 | notify_peer_device_state(NULL, 0, peer_device, &peer_device_info, NOTIFY_CREATE | flags); |
| 2497 | } |
| 2498 | mutex_unlock(¬ification_mutex); |
| 2499 | mutex_unlock(&adm_ctx.resource->conf_update); |
| 2500 | |
| 2501 | rcu_read_lock(); |
| 2502 | idr_for_each_entry(&connection->peer_devices, peer_device, i) { |
| 2503 | struct drbd_device *device = peer_device->device; |
| 2504 | device->send_cnt = 0; |
| 2505 | device->recv_cnt = 0; |
| 2506 | } |
| 2507 | rcu_read_unlock(); |
| 2508 | |
| 2509 | retcode = conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE); |
| 2510 | |
| 2511 | conn_reconfig_done(connection); |
| 2512 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2513 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2514 | return 0; |
| 2515 | |
| 2516 | fail: |
| 2517 | free_crypto(&crypto); |
| 2518 | kfree(new_net_conf); |
| 2519 | |
| 2520 | conn_reconfig_done(connection); |
| 2521 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2522 | out: |
| 2523 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2524 | return 0; |
| 2525 | } |
| 2526 | |
| 2527 | static enum drbd_state_rv conn_try_disconnect(struct drbd_connection *connection, bool force) |
| 2528 | { |
| 2529 | enum drbd_state_rv rv; |
| 2530 | |
| 2531 | rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), |
| 2532 | force ? CS_HARD : 0); |
| 2533 | |
| 2534 | switch (rv) { |
| 2535 | case SS_NOTHING_TO_DO: |
| 2536 | break; |
| 2537 | case SS_ALREADY_STANDALONE: |
| 2538 | return SS_SUCCESS; |
| 2539 | case SS_PRIMARY_NOP: |
| 2540 | /* Our state checking code wants to see the peer outdated. */ |
| 2541 | rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, pdsk, D_OUTDATED), 0); |
| 2542 | |
| 2543 | if (rv == SS_OUTDATE_WO_CONN) /* lost connection before graceful disconnect succeeded */ |
| 2544 | rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_VERBOSE); |
| 2545 | |
| 2546 | break; |
| 2547 | case SS_CW_FAILED_BY_PEER: |
| 2548 | /* The peer probably wants to see us outdated. */ |
| 2549 | rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, |
| 2550 | disk, D_OUTDATED), 0); |
| 2551 | if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) { |
| 2552 | rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), |
| 2553 | CS_HARD); |
| 2554 | } |
| 2555 | break; |
| 2556 | default:; |
| 2557 | /* no special handling necessary */ |
| 2558 | } |
| 2559 | |
| 2560 | if (rv >= SS_SUCCESS) { |
| 2561 | enum drbd_state_rv rv2; |
| 2562 | /* No one else can reconfigure the network while I am here. |
| 2563 | * The state handling only uses drbd_thread_stop_nowait(), |
| 2564 | * we want to really wait here until the receiver is no more. |
| 2565 | */ |
| 2566 | drbd_thread_stop(&connection->receiver); |
| 2567 | |
| 2568 | /* Race breaker. This additional state change request may be |
| 2569 | * necessary, if this was a forced disconnect during a receiver |
| 2570 | * restart. We may have "killed" the receiver thread just |
| 2571 | * after drbd_receiver() returned. Typically, we should be |
| 2572 | * C_STANDALONE already, now, and this becomes a no-op. |
| 2573 | */ |
| 2574 | rv2 = conn_request_state(connection, NS(conn, C_STANDALONE), |
| 2575 | CS_VERBOSE | CS_HARD); |
| 2576 | if (rv2 < SS_SUCCESS) |
| 2577 | drbd_err(connection, |
| 2578 | "unexpected rv2=%d in conn_try_disconnect()\n", |
| 2579 | rv2); |
| 2580 | /* Unlike in DRBD 9, the state engine has generated |
| 2581 | * NOTIFY_DESTROY events before clearing connection->net_conf. */ |
| 2582 | } |
| 2583 | return rv; |
| 2584 | } |
| 2585 | |
| 2586 | int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info) |
| 2587 | { |
| 2588 | struct drbd_config_context adm_ctx; |
| 2589 | struct disconnect_parms parms; |
| 2590 | struct drbd_connection *connection; |
| 2591 | enum drbd_state_rv rv; |
| 2592 | enum drbd_ret_code retcode; |
| 2593 | int err; |
| 2594 | |
| 2595 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION); |
| 2596 | if (!adm_ctx.reply_skb) |
| 2597 | return retcode; |
| 2598 | if (retcode != NO_ERROR) |
| 2599 | goto fail; |
| 2600 | |
| 2601 | connection = adm_ctx.connection; |
| 2602 | memset(&parms, 0, sizeof(parms)); |
| 2603 | if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) { |
| 2604 | err = disconnect_parms_from_attrs(&parms, info); |
| 2605 | if (err) { |
| 2606 | retcode = ERR_MANDATORY_TAG; |
| 2607 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 2608 | goto fail; |
| 2609 | } |
| 2610 | } |
| 2611 | |
| 2612 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2613 | rv = conn_try_disconnect(connection, parms.force_disconnect); |
| 2614 | if (rv < SS_SUCCESS) |
| 2615 | retcode = rv; /* FIXME: Type mismatch. */ |
| 2616 | else |
| 2617 | retcode = NO_ERROR; |
| 2618 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2619 | fail: |
| 2620 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2621 | return 0; |
| 2622 | } |
| 2623 | |
| 2624 | void resync_after_online_grow(struct drbd_device *device) |
| 2625 | { |
| 2626 | int iass; /* I am sync source */ |
| 2627 | |
| 2628 | drbd_info(device, "Resync of new storage after online grow\n"); |
| 2629 | if (device->state.role != device->state.peer) |
| 2630 | iass = (device->state.role == R_PRIMARY); |
| 2631 | else |
| 2632 | iass = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags); |
| 2633 | |
| 2634 | if (iass) |
| 2635 | drbd_start_resync(device, C_SYNC_SOURCE); |
| 2636 | else |
| 2637 | _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE); |
| 2638 | } |
| 2639 | |
| 2640 | int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info) |
| 2641 | { |
| 2642 | struct drbd_config_context adm_ctx; |
| 2643 | struct disk_conf *old_disk_conf, *new_disk_conf = NULL; |
| 2644 | struct resize_parms rs; |
| 2645 | struct drbd_device *device; |
| 2646 | enum drbd_ret_code retcode; |
| 2647 | enum determine_dev_size dd; |
| 2648 | bool change_al_layout = false; |
| 2649 | enum dds_flags ddsf; |
| 2650 | sector_t u_size; |
| 2651 | int err; |
| 2652 | |
| 2653 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2654 | if (!adm_ctx.reply_skb) |
| 2655 | return retcode; |
| 2656 | if (retcode != NO_ERROR) |
| 2657 | goto finish; |
| 2658 | |
| 2659 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2660 | device = adm_ctx.device; |
| 2661 | if (!get_ldev(device)) { |
| 2662 | retcode = ERR_NO_DISK; |
| 2663 | goto fail; |
| 2664 | } |
| 2665 | |
| 2666 | memset(&rs, 0, sizeof(struct resize_parms)); |
| 2667 | rs.al_stripes = device->ldev->md.al_stripes; |
| 2668 | rs.al_stripe_size = device->ldev->md.al_stripe_size_4k * 4; |
| 2669 | if (info->attrs[DRBD_NLA_RESIZE_PARMS]) { |
| 2670 | err = resize_parms_from_attrs(&rs, info); |
| 2671 | if (err) { |
| 2672 | retcode = ERR_MANDATORY_TAG; |
| 2673 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 2674 | goto fail_ldev; |
| 2675 | } |
| 2676 | } |
| 2677 | |
| 2678 | if (device->state.conn > C_CONNECTED) { |
| 2679 | retcode = ERR_RESIZE_RESYNC; |
| 2680 | goto fail_ldev; |
| 2681 | } |
| 2682 | |
| 2683 | if (device->state.role == R_SECONDARY && |
| 2684 | device->state.peer == R_SECONDARY) { |
| 2685 | retcode = ERR_NO_PRIMARY; |
| 2686 | goto fail_ldev; |
| 2687 | } |
| 2688 | |
| 2689 | if (rs.no_resync && first_peer_device(device)->connection->agreed_pro_version < 93) { |
| 2690 | retcode = ERR_NEED_APV_93; |
| 2691 | goto fail_ldev; |
| 2692 | } |
| 2693 | |
| 2694 | rcu_read_lock(); |
| 2695 | u_size = rcu_dereference(device->ldev->disk_conf)->disk_size; |
| 2696 | rcu_read_unlock(); |
| 2697 | if (u_size != (sector_t)rs.resize_size) { |
| 2698 | new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL); |
| 2699 | if (!new_disk_conf) { |
| 2700 | retcode = ERR_NOMEM; |
| 2701 | goto fail_ldev; |
| 2702 | } |
| 2703 | } |
| 2704 | |
| 2705 | if (device->ldev->md.al_stripes != rs.al_stripes || |
| 2706 | device->ldev->md.al_stripe_size_4k != rs.al_stripe_size / 4) { |
| 2707 | u32 al_size_k = rs.al_stripes * rs.al_stripe_size; |
| 2708 | |
| 2709 | if (al_size_k > (16 * 1024 * 1024)) { |
| 2710 | retcode = ERR_MD_LAYOUT_TOO_BIG; |
| 2711 | goto fail_ldev; |
| 2712 | } |
| 2713 | |
| 2714 | if (al_size_k < MD_32kB_SECT/2) { |
| 2715 | retcode = ERR_MD_LAYOUT_TOO_SMALL; |
| 2716 | goto fail_ldev; |
| 2717 | } |
| 2718 | |
| 2719 | if (device->state.conn != C_CONNECTED && !rs.resize_force) { |
| 2720 | retcode = ERR_MD_LAYOUT_CONNECTED; |
| 2721 | goto fail_ldev; |
| 2722 | } |
| 2723 | |
| 2724 | change_al_layout = true; |
| 2725 | } |
| 2726 | |
| 2727 | if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev)) |
| 2728 | device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev); |
| 2729 | |
| 2730 | if (new_disk_conf) { |
| 2731 | mutex_lock(&device->resource->conf_update); |
| 2732 | old_disk_conf = device->ldev->disk_conf; |
| 2733 | *new_disk_conf = *old_disk_conf; |
| 2734 | new_disk_conf->disk_size = (sector_t)rs.resize_size; |
| 2735 | rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); |
| 2736 | mutex_unlock(&device->resource->conf_update); |
| 2737 | synchronize_rcu(); |
| 2738 | kfree(old_disk_conf); |
| 2739 | new_disk_conf = NULL; |
| 2740 | } |
| 2741 | |
| 2742 | ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0); |
| 2743 | dd = drbd_determine_dev_size(device, ddsf, change_al_layout ? &rs : NULL); |
| 2744 | drbd_md_sync(device); |
| 2745 | put_ldev(device); |
| 2746 | if (dd == DS_ERROR) { |
| 2747 | retcode = ERR_NOMEM_BITMAP; |
| 2748 | goto fail; |
| 2749 | } else if (dd == DS_ERROR_SPACE_MD) { |
| 2750 | retcode = ERR_MD_LAYOUT_NO_FIT; |
| 2751 | goto fail; |
| 2752 | } else if (dd == DS_ERROR_SHRINK) { |
| 2753 | retcode = ERR_IMPLICIT_SHRINK; |
| 2754 | goto fail; |
| 2755 | } |
| 2756 | |
| 2757 | if (device->state.conn == C_CONNECTED) { |
| 2758 | if (dd == DS_GREW) |
| 2759 | set_bit(RESIZE_PENDING, &device->flags); |
| 2760 | |
| 2761 | drbd_send_uuids(first_peer_device(device)); |
| 2762 | drbd_send_sizes(first_peer_device(device), 1, ddsf); |
| 2763 | } |
| 2764 | |
| 2765 | fail: |
| 2766 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2767 | finish: |
| 2768 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2769 | return 0; |
| 2770 | |
| 2771 | fail_ldev: |
| 2772 | put_ldev(device); |
| 2773 | kfree(new_disk_conf); |
| 2774 | goto fail; |
| 2775 | } |
| 2776 | |
| 2777 | int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info) |
| 2778 | { |
| 2779 | struct drbd_config_context adm_ctx; |
| 2780 | enum drbd_ret_code retcode; |
| 2781 | struct res_opts res_opts; |
| 2782 | int err; |
| 2783 | |
| 2784 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE); |
| 2785 | if (!adm_ctx.reply_skb) |
| 2786 | return retcode; |
| 2787 | if (retcode != NO_ERROR) |
| 2788 | goto fail; |
| 2789 | |
| 2790 | res_opts = adm_ctx.resource->res_opts; |
| 2791 | if (should_set_defaults(info)) |
| 2792 | set_res_opts_defaults(&res_opts); |
| 2793 | |
| 2794 | err = res_opts_from_attrs(&res_opts, info); |
| 2795 | if (err && err != -ENOMSG) { |
| 2796 | retcode = ERR_MANDATORY_TAG; |
| 2797 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 2798 | goto fail; |
| 2799 | } |
| 2800 | |
| 2801 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2802 | err = set_resource_options(adm_ctx.resource, &res_opts); |
| 2803 | if (err) { |
| 2804 | retcode = ERR_INVALID_REQUEST; |
| 2805 | if (err == -ENOMEM) |
| 2806 | retcode = ERR_NOMEM; |
| 2807 | } |
| 2808 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2809 | |
| 2810 | fail: |
| 2811 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2812 | return 0; |
| 2813 | } |
| 2814 | |
| 2815 | int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info) |
| 2816 | { |
| 2817 | struct drbd_config_context adm_ctx; |
| 2818 | struct drbd_device *device; |
| 2819 | int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */ |
| 2820 | |
| 2821 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2822 | if (!adm_ctx.reply_skb) |
| 2823 | return retcode; |
| 2824 | if (retcode != NO_ERROR) |
| 2825 | goto out; |
| 2826 | |
| 2827 | device = adm_ctx.device; |
| 2828 | if (!get_ldev(device)) { |
| 2829 | retcode = ERR_NO_DISK; |
| 2830 | goto out; |
| 2831 | } |
| 2832 | |
| 2833 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2834 | |
| 2835 | /* If there is still bitmap IO pending, probably because of a previous |
| 2836 | * resync just being finished, wait for it before requesting a new resync. |
| 2837 | * Also wait for it's after_state_ch(). */ |
| 2838 | drbd_suspend_io(device); |
| 2839 | wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags)); |
| 2840 | drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work); |
| 2841 | |
| 2842 | /* If we happen to be C_STANDALONE R_SECONDARY, just change to |
| 2843 | * D_INCONSISTENT, and set all bits in the bitmap. Otherwise, |
| 2844 | * try to start a resync handshake as sync target for full sync. |
| 2845 | */ |
| 2846 | if (device->state.conn == C_STANDALONE && device->state.role == R_SECONDARY) { |
| 2847 | retcode = drbd_request_state(device, NS(disk, D_INCONSISTENT)); |
| 2848 | if (retcode >= SS_SUCCESS) { |
| 2849 | if (drbd_bitmap_io(device, &drbd_bmio_set_n_write, |
| 2850 | "set_n_write from invalidate", BM_LOCKED_MASK)) |
| 2851 | retcode = ERR_IO_MD_DISK; |
| 2852 | } |
| 2853 | } else |
| 2854 | retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_T)); |
| 2855 | drbd_resume_io(device); |
| 2856 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2857 | put_ldev(device); |
| 2858 | out: |
| 2859 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2860 | return 0; |
| 2861 | } |
| 2862 | |
| 2863 | static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info, |
| 2864 | union drbd_state mask, union drbd_state val) |
| 2865 | { |
| 2866 | struct drbd_config_context adm_ctx; |
| 2867 | enum drbd_ret_code retcode; |
| 2868 | |
| 2869 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2870 | if (!adm_ctx.reply_skb) |
| 2871 | return retcode; |
| 2872 | if (retcode != NO_ERROR) |
| 2873 | goto out; |
| 2874 | |
| 2875 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2876 | retcode = drbd_request_state(adm_ctx.device, mask, val); |
| 2877 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2878 | out: |
| 2879 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2880 | return 0; |
| 2881 | } |
| 2882 | |
| 2883 | static int drbd_bmio_set_susp_al(struct drbd_device *device) __must_hold(local) |
| 2884 | { |
| 2885 | int rv; |
| 2886 | |
| 2887 | rv = drbd_bmio_set_n_write(device); |
| 2888 | drbd_suspend_al(device); |
| 2889 | return rv; |
| 2890 | } |
| 2891 | |
| 2892 | int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info) |
| 2893 | { |
| 2894 | struct drbd_config_context adm_ctx; |
| 2895 | int retcode; /* drbd_ret_code, drbd_state_rv */ |
| 2896 | struct drbd_device *device; |
| 2897 | |
| 2898 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2899 | if (!adm_ctx.reply_skb) |
| 2900 | return retcode; |
| 2901 | if (retcode != NO_ERROR) |
| 2902 | goto out; |
| 2903 | |
| 2904 | device = adm_ctx.device; |
| 2905 | if (!get_ldev(device)) { |
| 2906 | retcode = ERR_NO_DISK; |
| 2907 | goto out; |
| 2908 | } |
| 2909 | |
| 2910 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2911 | |
| 2912 | /* If there is still bitmap IO pending, probably because of a previous |
| 2913 | * resync just being finished, wait for it before requesting a new resync. |
| 2914 | * Also wait for it's after_state_ch(). */ |
| 2915 | drbd_suspend_io(device); |
| 2916 | wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags)); |
| 2917 | drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work); |
| 2918 | |
| 2919 | /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits |
| 2920 | * in the bitmap. Otherwise, try to start a resync handshake |
| 2921 | * as sync source for full sync. |
| 2922 | */ |
| 2923 | if (device->state.conn == C_STANDALONE && device->state.role == R_PRIMARY) { |
| 2924 | /* The peer will get a resync upon connect anyways. Just make that |
| 2925 | into a full resync. */ |
| 2926 | retcode = drbd_request_state(device, NS(pdsk, D_INCONSISTENT)); |
| 2927 | if (retcode >= SS_SUCCESS) { |
| 2928 | if (drbd_bitmap_io(device, &drbd_bmio_set_susp_al, |
| 2929 | "set_n_write from invalidate_peer", |
| 2930 | BM_LOCKED_SET_ALLOWED)) |
| 2931 | retcode = ERR_IO_MD_DISK; |
| 2932 | } |
| 2933 | } else |
| 2934 | retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_S)); |
| 2935 | drbd_resume_io(device); |
| 2936 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2937 | put_ldev(device); |
| 2938 | out: |
| 2939 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2940 | return 0; |
| 2941 | } |
| 2942 | |
| 2943 | int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info) |
| 2944 | { |
| 2945 | struct drbd_config_context adm_ctx; |
| 2946 | enum drbd_ret_code retcode; |
| 2947 | |
| 2948 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2949 | if (!adm_ctx.reply_skb) |
| 2950 | return retcode; |
| 2951 | if (retcode != NO_ERROR) |
| 2952 | goto out; |
| 2953 | |
| 2954 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2955 | if (drbd_request_state(adm_ctx.device, NS(user_isp, 1)) == SS_NOTHING_TO_DO) |
| 2956 | retcode = ERR_PAUSE_IS_SET; |
| 2957 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2958 | out: |
| 2959 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2960 | return 0; |
| 2961 | } |
| 2962 | |
| 2963 | int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info) |
| 2964 | { |
| 2965 | struct drbd_config_context adm_ctx; |
| 2966 | union drbd_dev_state s; |
| 2967 | enum drbd_ret_code retcode; |
| 2968 | |
| 2969 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2970 | if (!adm_ctx.reply_skb) |
| 2971 | return retcode; |
| 2972 | if (retcode != NO_ERROR) |
| 2973 | goto out; |
| 2974 | |
| 2975 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2976 | if (drbd_request_state(adm_ctx.device, NS(user_isp, 0)) == SS_NOTHING_TO_DO) { |
| 2977 | s = adm_ctx.device->state; |
| 2978 | if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) { |
| 2979 | retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP : |
| 2980 | s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR; |
| 2981 | } else { |
| 2982 | retcode = ERR_PAUSE_IS_CLEAR; |
| 2983 | } |
| 2984 | } |
| 2985 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2986 | out: |
| 2987 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2988 | return 0; |
| 2989 | } |
| 2990 | |
| 2991 | int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info) |
| 2992 | { |
| 2993 | return drbd_adm_simple_request_state(skb, info, NS(susp, 1)); |
| 2994 | } |
| 2995 | |
| 2996 | int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info) |
| 2997 | { |
| 2998 | struct drbd_config_context adm_ctx; |
| 2999 | struct drbd_device *device; |
| 3000 | int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */ |
| 3001 | |
| 3002 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 3003 | if (!adm_ctx.reply_skb) |
| 3004 | return retcode; |
| 3005 | if (retcode != NO_ERROR) |
| 3006 | goto out; |
| 3007 | |
| 3008 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 3009 | device = adm_ctx.device; |
| 3010 | if (test_bit(NEW_CUR_UUID, &device->flags)) { |
| 3011 | if (get_ldev_if_state(device, D_ATTACHING)) { |
| 3012 | drbd_uuid_new_current(device); |
| 3013 | put_ldev(device); |
| 3014 | } else { |
| 3015 | /* This is effectively a multi-stage "forced down". |
| 3016 | * The NEW_CUR_UUID bit is supposedly only set, if we |
| 3017 | * lost the replication connection, and are configured |
| 3018 | * to freeze IO and wait for some fence-peer handler. |
| 3019 | * So we still don't have a replication connection. |
| 3020 | * And now we don't have a local disk either. After |
| 3021 | * resume, we will fail all pending and new IO, because |
| 3022 | * we don't have any data anymore. Which means we will |
| 3023 | * eventually be able to terminate all users of this |
| 3024 | * device, and then take it down. By bumping the |
| 3025 | * "effective" data uuid, we make sure that you really |
| 3026 | * need to tear down before you reconfigure, we will |
| 3027 | * the refuse to re-connect or re-attach (because no |
| 3028 | * matching real data uuid exists). |
| 3029 | */ |
| 3030 | u64 val; |
| 3031 | get_random_bytes(&val, sizeof(u64)); |
| 3032 | drbd_set_ed_uuid(device, val); |
| 3033 | drbd_warn(device, "Resumed without access to data; please tear down before attempting to re-configure.\n"); |
| 3034 | } |
| 3035 | clear_bit(NEW_CUR_UUID, &device->flags); |
| 3036 | } |
| 3037 | drbd_suspend_io(device); |
| 3038 | retcode = drbd_request_state(device, NS3(susp, 0, susp_nod, 0, susp_fen, 0)); |
| 3039 | if (retcode == SS_SUCCESS) { |
| 3040 | if (device->state.conn < C_CONNECTED) |
| 3041 | tl_clear(first_peer_device(device)->connection); |
| 3042 | if (device->state.disk == D_DISKLESS || device->state.disk == D_FAILED) |
| 3043 | tl_restart(first_peer_device(device)->connection, FAIL_FROZEN_DISK_IO); |
| 3044 | } |
| 3045 | drbd_resume_io(device); |
| 3046 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 3047 | out: |
| 3048 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 3049 | return 0; |
| 3050 | } |
| 3051 | |
| 3052 | int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info) |
| 3053 | { |
| 3054 | return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED)); |
| 3055 | } |
| 3056 | |
| 3057 | static int nla_put_drbd_cfg_context(struct sk_buff *skb, |
| 3058 | struct drbd_resource *resource, |
| 3059 | struct drbd_connection *connection, |
| 3060 | struct drbd_device *device) |
| 3061 | { |
| 3062 | struct nlattr *nla; |
| 3063 | nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT); |
| 3064 | if (!nla) |
| 3065 | goto nla_put_failure; |
| 3066 | if (device && |
| 3067 | nla_put_u32(skb, T_ctx_volume, device->vnr)) |
| 3068 | goto nla_put_failure; |
| 3069 | if (nla_put_string(skb, T_ctx_resource_name, resource->name)) |
| 3070 | goto nla_put_failure; |
| 3071 | if (connection) { |
| 3072 | if (connection->my_addr_len && |
| 3073 | nla_put(skb, T_ctx_my_addr, connection->my_addr_len, &connection->my_addr)) |
| 3074 | goto nla_put_failure; |
| 3075 | if (connection->peer_addr_len && |
| 3076 | nla_put(skb, T_ctx_peer_addr, connection->peer_addr_len, &connection->peer_addr)) |
| 3077 | goto nla_put_failure; |
| 3078 | } |
| 3079 | nla_nest_end(skb, nla); |
| 3080 | return 0; |
| 3081 | |
| 3082 | nla_put_failure: |
| 3083 | if (nla) |
| 3084 | nla_nest_cancel(skb, nla); |
| 3085 | return -EMSGSIZE; |
| 3086 | } |
| 3087 | |
| 3088 | /* |
| 3089 | * The generic netlink dump callbacks are called outside the genl_lock(), so |
| 3090 | * they cannot use the simple attribute parsing code which uses global |
| 3091 | * attribute tables. |
| 3092 | */ |
| 3093 | static struct nlattr *find_cfg_context_attr(const struct nlmsghdr *nlh, int attr) |
| 3094 | { |
| 3095 | const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ; |
| 3096 | const int maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1; |
| 3097 | struct nlattr *nla; |
| 3098 | |
| 3099 | nla = nla_find(nlmsg_attrdata(nlh, hdrlen), nlmsg_attrlen(nlh, hdrlen), |
| 3100 | DRBD_NLA_CFG_CONTEXT); |
| 3101 | if (!nla) |
| 3102 | return NULL; |
| 3103 | return drbd_nla_find_nested(maxtype, nla, __nla_type(attr)); |
| 3104 | } |
| 3105 | |
| 3106 | static void resource_to_info(struct resource_info *, struct drbd_resource *); |
| 3107 | |
| 3108 | int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb) |
| 3109 | { |
| 3110 | struct drbd_genlmsghdr *dh; |
| 3111 | struct drbd_resource *resource; |
| 3112 | struct resource_info resource_info; |
| 3113 | struct resource_statistics resource_statistics; |
| 3114 | int err; |
| 3115 | |
| 3116 | rcu_read_lock(); |
| 3117 | if (cb->args[0]) { |
| 3118 | for_each_resource_rcu(resource, &drbd_resources) |
| 3119 | if (resource == (struct drbd_resource *)cb->args[0]) |
| 3120 | goto found_resource; |
| 3121 | err = 0; /* resource was probably deleted */ |
| 3122 | goto out; |
| 3123 | } |
| 3124 | resource = list_entry(&drbd_resources, |
| 3125 | struct drbd_resource, resources); |
| 3126 | |
| 3127 | found_resource: |
| 3128 | list_for_each_entry_continue_rcu(resource, &drbd_resources, resources) { |
| 3129 | goto put_result; |
| 3130 | } |
| 3131 | err = 0; |
| 3132 | goto out; |
| 3133 | |
| 3134 | put_result: |
| 3135 | dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, |
| 3136 | cb->nlh->nlmsg_seq, &drbd_genl_family, |
| 3137 | NLM_F_MULTI, DRBD_ADM_GET_RESOURCES); |
| 3138 | err = -ENOMEM; |
| 3139 | if (!dh) |
| 3140 | goto out; |
| 3141 | dh->minor = -1U; |
| 3142 | dh->ret_code = NO_ERROR; |
| 3143 | err = nla_put_drbd_cfg_context(skb, resource, NULL, NULL); |
| 3144 | if (err) |
| 3145 | goto out; |
| 3146 | err = res_opts_to_skb(skb, &resource->res_opts, !capable(CAP_SYS_ADMIN)); |
| 3147 | if (err) |
| 3148 | goto out; |
| 3149 | resource_to_info(&resource_info, resource); |
| 3150 | err = resource_info_to_skb(skb, &resource_info, !capable(CAP_SYS_ADMIN)); |
| 3151 | if (err) |
| 3152 | goto out; |
| 3153 | resource_statistics.res_stat_write_ordering = resource->write_ordering; |
| 3154 | err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN)); |
| 3155 | if (err) |
| 3156 | goto out; |
| 3157 | cb->args[0] = (long)resource; |
| 3158 | genlmsg_end(skb, dh); |
| 3159 | err = 0; |
| 3160 | |
| 3161 | out: |
| 3162 | rcu_read_unlock(); |
| 3163 | if (err) |
| 3164 | return err; |
| 3165 | return skb->len; |
| 3166 | } |
| 3167 | |
| 3168 | static void device_to_statistics(struct device_statistics *s, |
| 3169 | struct drbd_device *device) |
| 3170 | { |
| 3171 | memset(s, 0, sizeof(*s)); |
| 3172 | s->dev_upper_blocked = !may_inc_ap_bio(device); |
| 3173 | if (get_ldev(device)) { |
| 3174 | struct drbd_md *md = &device->ldev->md; |
| 3175 | u64 *history_uuids = (u64 *)s->history_uuids; |
| 3176 | struct request_queue *q; |
| 3177 | int n; |
| 3178 | |
| 3179 | spin_lock_irq(&md->uuid_lock); |
| 3180 | s->dev_current_uuid = md->uuid[UI_CURRENT]; |
| 3181 | BUILD_BUG_ON(sizeof(s->history_uuids) < UI_HISTORY_END - UI_HISTORY_START + 1); |
| 3182 | for (n = 0; n < UI_HISTORY_END - UI_HISTORY_START + 1; n++) |
| 3183 | history_uuids[n] = md->uuid[UI_HISTORY_START + n]; |
| 3184 | for (; n < HISTORY_UUIDS; n++) |
| 3185 | history_uuids[n] = 0; |
| 3186 | s->history_uuids_len = HISTORY_UUIDS; |
| 3187 | spin_unlock_irq(&md->uuid_lock); |
| 3188 | |
| 3189 | s->dev_disk_flags = md->flags; |
| 3190 | q = bdev_get_queue(device->ldev->backing_bdev); |
| 3191 | s->dev_lower_blocked = |
| 3192 | bdi_congested(&q->backing_dev_info, |
| 3193 | (1 << WB_async_congested) | |
| 3194 | (1 << WB_sync_congested)); |
| 3195 | put_ldev(device); |
| 3196 | } |
| 3197 | s->dev_size = drbd_get_capacity(device->this_bdev); |
| 3198 | s->dev_read = device->read_cnt; |
| 3199 | s->dev_write = device->writ_cnt; |
| 3200 | s->dev_al_writes = device->al_writ_cnt; |
| 3201 | s->dev_bm_writes = device->bm_writ_cnt; |
| 3202 | s->dev_upper_pending = atomic_read(&device->ap_bio_cnt); |
| 3203 | s->dev_lower_pending = atomic_read(&device->local_cnt); |
| 3204 | s->dev_al_suspended = test_bit(AL_SUSPENDED, &device->flags); |
| 3205 | s->dev_exposed_data_uuid = device->ed_uuid; |
| 3206 | } |
| 3207 | |
| 3208 | static int put_resource_in_arg0(struct netlink_callback *cb, int holder_nr) |
| 3209 | { |
| 3210 | if (cb->args[0]) { |
| 3211 | struct drbd_resource *resource = |
| 3212 | (struct drbd_resource *)cb->args[0]; |
| 3213 | kref_put(&resource->kref, drbd_destroy_resource); |
| 3214 | } |
| 3215 | |
| 3216 | return 0; |
| 3217 | } |
| 3218 | |
| 3219 | int drbd_adm_dump_devices_done(struct netlink_callback *cb) { |
| 3220 | return put_resource_in_arg0(cb, 7); |
| 3221 | } |
| 3222 | |
| 3223 | static void device_to_info(struct device_info *, struct drbd_device *); |
| 3224 | |
| 3225 | int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb) |
| 3226 | { |
| 3227 | struct nlattr *resource_filter; |
| 3228 | struct drbd_resource *resource; |
| 3229 | struct drbd_device *uninitialized_var(device); |
| 3230 | int minor, err, retcode; |
| 3231 | struct drbd_genlmsghdr *dh; |
| 3232 | struct device_info device_info; |
| 3233 | struct device_statistics device_statistics; |
| 3234 | struct idr *idr_to_search; |
| 3235 | |
| 3236 | resource = (struct drbd_resource *)cb->args[0]; |
| 3237 | if (!cb->args[0] && !cb->args[1]) { |
| 3238 | resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name); |
| 3239 | if (resource_filter) { |
| 3240 | retcode = ERR_RES_NOT_KNOWN; |
| 3241 | resource = drbd_find_resource(nla_data(resource_filter)); |
| 3242 | if (!resource) |
| 3243 | goto put_result; |
| 3244 | cb->args[0] = (long)resource; |
| 3245 | } |
| 3246 | } |
| 3247 | |
| 3248 | rcu_read_lock(); |
| 3249 | minor = cb->args[1]; |
| 3250 | idr_to_search = resource ? &resource->devices : &drbd_devices; |
| 3251 | device = idr_get_next(idr_to_search, &minor); |
| 3252 | if (!device) { |
| 3253 | err = 0; |
| 3254 | goto out; |
| 3255 | } |
| 3256 | idr_for_each_entry_continue(idr_to_search, device, minor) { |
| 3257 | retcode = NO_ERROR; |
| 3258 | goto put_result; /* only one iteration */ |
| 3259 | } |
| 3260 | err = 0; |
| 3261 | goto out; /* no more devices */ |
| 3262 | |
| 3263 | put_result: |
| 3264 | dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, |
| 3265 | cb->nlh->nlmsg_seq, &drbd_genl_family, |
| 3266 | NLM_F_MULTI, DRBD_ADM_GET_DEVICES); |
| 3267 | err = -ENOMEM; |
| 3268 | if (!dh) |
| 3269 | goto out; |
| 3270 | dh->ret_code = retcode; |
| 3271 | dh->minor = -1U; |
| 3272 | if (retcode == NO_ERROR) { |
| 3273 | dh->minor = device->minor; |
| 3274 | err = nla_put_drbd_cfg_context(skb, device->resource, NULL, device); |
| 3275 | if (err) |
| 3276 | goto out; |
| 3277 | if (get_ldev(device)) { |
| 3278 | struct disk_conf *disk_conf = |
| 3279 | rcu_dereference(device->ldev->disk_conf); |
| 3280 | |
| 3281 | err = disk_conf_to_skb(skb, disk_conf, !capable(CAP_SYS_ADMIN)); |
| 3282 | put_ldev(device); |
| 3283 | if (err) |
| 3284 | goto out; |
| 3285 | } |
| 3286 | device_to_info(&device_info, device); |
| 3287 | err = device_info_to_skb(skb, &device_info, !capable(CAP_SYS_ADMIN)); |
| 3288 | if (err) |
| 3289 | goto out; |
| 3290 | |
| 3291 | device_to_statistics(&device_statistics, device); |
| 3292 | err = device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN)); |
| 3293 | if (err) |
| 3294 | goto out; |
| 3295 | cb->args[1] = minor + 1; |
| 3296 | } |
| 3297 | genlmsg_end(skb, dh); |
| 3298 | err = 0; |
| 3299 | |
| 3300 | out: |
| 3301 | rcu_read_unlock(); |
| 3302 | if (err) |
| 3303 | return err; |
| 3304 | return skb->len; |
| 3305 | } |
| 3306 | |
| 3307 | int drbd_adm_dump_connections_done(struct netlink_callback *cb) |
| 3308 | { |
| 3309 | return put_resource_in_arg0(cb, 6); |
| 3310 | } |
| 3311 | |
| 3312 | enum { SINGLE_RESOURCE, ITERATE_RESOURCES }; |
| 3313 | |
| 3314 | int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb) |
| 3315 | { |
| 3316 | struct nlattr *resource_filter; |
| 3317 | struct drbd_resource *resource = NULL, *next_resource; |
| 3318 | struct drbd_connection *uninitialized_var(connection); |
| 3319 | int err = 0, retcode; |
| 3320 | struct drbd_genlmsghdr *dh; |
| 3321 | struct connection_info connection_info; |
| 3322 | struct connection_statistics connection_statistics; |
| 3323 | |
| 3324 | rcu_read_lock(); |
| 3325 | resource = (struct drbd_resource *)cb->args[0]; |
| 3326 | if (!cb->args[0]) { |
| 3327 | resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name); |
| 3328 | if (resource_filter) { |
| 3329 | retcode = ERR_RES_NOT_KNOWN; |
| 3330 | resource = drbd_find_resource(nla_data(resource_filter)); |
| 3331 | if (!resource) |
| 3332 | goto put_result; |
| 3333 | cb->args[0] = (long)resource; |
| 3334 | cb->args[1] = SINGLE_RESOURCE; |
| 3335 | } |
| 3336 | } |
| 3337 | if (!resource) { |
| 3338 | if (list_empty(&drbd_resources)) |
| 3339 | goto out; |
| 3340 | resource = list_first_entry(&drbd_resources, struct drbd_resource, resources); |
| 3341 | kref_get(&resource->kref); |
| 3342 | cb->args[0] = (long)resource; |
| 3343 | cb->args[1] = ITERATE_RESOURCES; |
| 3344 | } |
| 3345 | |
| 3346 | next_resource: |
| 3347 | rcu_read_unlock(); |
| 3348 | mutex_lock(&resource->conf_update); |
| 3349 | rcu_read_lock(); |
| 3350 | if (cb->args[2]) { |
| 3351 | for_each_connection_rcu(connection, resource) |
| 3352 | if (connection == (struct drbd_connection *)cb->args[2]) |
| 3353 | goto found_connection; |
| 3354 | /* connection was probably deleted */ |
| 3355 | goto no_more_connections; |
| 3356 | } |
| 3357 | connection = list_entry(&resource->connections, struct drbd_connection, connections); |
| 3358 | |
| 3359 | found_connection: |
| 3360 | list_for_each_entry_continue_rcu(connection, &resource->connections, connections) { |
| 3361 | if (!has_net_conf(connection)) |
| 3362 | continue; |
| 3363 | retcode = NO_ERROR; |
| 3364 | goto put_result; /* only one iteration */ |
| 3365 | } |
| 3366 | |
| 3367 | no_more_connections: |
| 3368 | if (cb->args[1] == ITERATE_RESOURCES) { |
| 3369 | for_each_resource_rcu(next_resource, &drbd_resources) { |
| 3370 | if (next_resource == resource) |
| 3371 | goto found_resource; |
| 3372 | } |
| 3373 | /* resource was probably deleted */ |
| 3374 | } |
| 3375 | goto out; |
| 3376 | |
| 3377 | found_resource: |
| 3378 | list_for_each_entry_continue_rcu(next_resource, &drbd_resources, resources) { |
| 3379 | mutex_unlock(&resource->conf_update); |
| 3380 | kref_put(&resource->kref, drbd_destroy_resource); |
| 3381 | resource = next_resource; |
| 3382 | kref_get(&resource->kref); |
| 3383 | cb->args[0] = (long)resource; |
| 3384 | cb->args[2] = 0; |
| 3385 | goto next_resource; |
| 3386 | } |
| 3387 | goto out; /* no more resources */ |
| 3388 | |
| 3389 | put_result: |
| 3390 | dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, |
| 3391 | cb->nlh->nlmsg_seq, &drbd_genl_family, |
| 3392 | NLM_F_MULTI, DRBD_ADM_GET_CONNECTIONS); |
| 3393 | err = -ENOMEM; |
| 3394 | if (!dh) |
| 3395 | goto out; |
| 3396 | dh->ret_code = retcode; |
| 3397 | dh->minor = -1U; |
| 3398 | if (retcode == NO_ERROR) { |
| 3399 | struct net_conf *net_conf; |
| 3400 | |
| 3401 | err = nla_put_drbd_cfg_context(skb, resource, connection, NULL); |
| 3402 | if (err) |
| 3403 | goto out; |
| 3404 | net_conf = rcu_dereference(connection->net_conf); |
| 3405 | if (net_conf) { |
| 3406 | err = net_conf_to_skb(skb, net_conf, !capable(CAP_SYS_ADMIN)); |
| 3407 | if (err) |
| 3408 | goto out; |
| 3409 | } |
| 3410 | connection_to_info(&connection_info, connection); |
| 3411 | err = connection_info_to_skb(skb, &connection_info, !capable(CAP_SYS_ADMIN)); |
| 3412 | if (err) |
| 3413 | goto out; |
| 3414 | connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags); |
| 3415 | err = connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN)); |
| 3416 | if (err) |
| 3417 | goto out; |
| 3418 | cb->args[2] = (long)connection; |
| 3419 | } |
| 3420 | genlmsg_end(skb, dh); |
| 3421 | err = 0; |
| 3422 | |
| 3423 | out: |
| 3424 | rcu_read_unlock(); |
| 3425 | if (resource) |
| 3426 | mutex_unlock(&resource->conf_update); |
| 3427 | if (err) |
| 3428 | return err; |
| 3429 | return skb->len; |
| 3430 | } |
| 3431 | |
| 3432 | enum mdf_peer_flag { |
| 3433 | MDF_PEER_CONNECTED = 1 << 0, |
| 3434 | MDF_PEER_OUTDATED = 1 << 1, |
| 3435 | MDF_PEER_FENCING = 1 << 2, |
| 3436 | MDF_PEER_FULL_SYNC = 1 << 3, |
| 3437 | }; |
| 3438 | |
| 3439 | static void peer_device_to_statistics(struct peer_device_statistics *s, |
| 3440 | struct drbd_peer_device *peer_device) |
| 3441 | { |
| 3442 | struct drbd_device *device = peer_device->device; |
| 3443 | |
| 3444 | memset(s, 0, sizeof(*s)); |
| 3445 | s->peer_dev_received = device->recv_cnt; |
| 3446 | s->peer_dev_sent = device->send_cnt; |
| 3447 | s->peer_dev_pending = atomic_read(&device->ap_pending_cnt) + |
| 3448 | atomic_read(&device->rs_pending_cnt); |
| 3449 | s->peer_dev_unacked = atomic_read(&device->unacked_cnt); |
| 3450 | s->peer_dev_out_of_sync = drbd_bm_total_weight(device) << (BM_BLOCK_SHIFT - 9); |
| 3451 | s->peer_dev_resync_failed = device->rs_failed << (BM_BLOCK_SHIFT - 9); |
| 3452 | if (get_ldev(device)) { |
| 3453 | struct drbd_md *md = &device->ldev->md; |
| 3454 | |
| 3455 | spin_lock_irq(&md->uuid_lock); |
| 3456 | s->peer_dev_bitmap_uuid = md->uuid[UI_BITMAP]; |
| 3457 | spin_unlock_irq(&md->uuid_lock); |
| 3458 | s->peer_dev_flags = |
| 3459 | (drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND) ? |
| 3460 | MDF_PEER_CONNECTED : 0) + |
| 3461 | (drbd_md_test_flag(device->ldev, MDF_CONSISTENT) && |
| 3462 | !drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE) ? |
| 3463 | MDF_PEER_OUTDATED : 0) + |
| 3464 | /* FIXME: MDF_PEER_FENCING? */ |
| 3465 | (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ? |
| 3466 | MDF_PEER_FULL_SYNC : 0); |
| 3467 | put_ldev(device); |
| 3468 | } |
| 3469 | } |
| 3470 | |
| 3471 | int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb) |
| 3472 | { |
| 3473 | return put_resource_in_arg0(cb, 9); |
| 3474 | } |
| 3475 | |
| 3476 | int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb) |
| 3477 | { |
| 3478 | struct nlattr *resource_filter; |
| 3479 | struct drbd_resource *resource; |
| 3480 | struct drbd_device *uninitialized_var(device); |
| 3481 | struct drbd_peer_device *peer_device = NULL; |
| 3482 | int minor, err, retcode; |
| 3483 | struct drbd_genlmsghdr *dh; |
| 3484 | struct idr *idr_to_search; |
| 3485 | |
| 3486 | resource = (struct drbd_resource *)cb->args[0]; |
| 3487 | if (!cb->args[0] && !cb->args[1]) { |
| 3488 | resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name); |
| 3489 | if (resource_filter) { |
| 3490 | retcode = ERR_RES_NOT_KNOWN; |
| 3491 | resource = drbd_find_resource(nla_data(resource_filter)); |
| 3492 | if (!resource) |
| 3493 | goto put_result; |
| 3494 | } |
| 3495 | cb->args[0] = (long)resource; |
| 3496 | } |
| 3497 | |
| 3498 | rcu_read_lock(); |
| 3499 | minor = cb->args[1]; |
| 3500 | idr_to_search = resource ? &resource->devices : &drbd_devices; |
| 3501 | device = idr_find(idr_to_search, minor); |
| 3502 | if (!device) { |
| 3503 | next_device: |
| 3504 | minor++; |
| 3505 | cb->args[2] = 0; |
| 3506 | device = idr_get_next(idr_to_search, &minor); |
| 3507 | if (!device) { |
| 3508 | err = 0; |
| 3509 | goto out; |
| 3510 | } |
| 3511 | } |
| 3512 | if (cb->args[2]) { |
| 3513 | for_each_peer_device(peer_device, device) |
| 3514 | if (peer_device == (struct drbd_peer_device *)cb->args[2]) |
| 3515 | goto found_peer_device; |
| 3516 | /* peer device was probably deleted */ |
| 3517 | goto next_device; |
| 3518 | } |
| 3519 | /* Make peer_device point to the list head (not the first entry). */ |
| 3520 | peer_device = list_entry(&device->peer_devices, struct drbd_peer_device, peer_devices); |
| 3521 | |
| 3522 | found_peer_device: |
| 3523 | list_for_each_entry_continue_rcu(peer_device, &device->peer_devices, peer_devices) { |
| 3524 | if (!has_net_conf(peer_device->connection)) |
| 3525 | continue; |
| 3526 | retcode = NO_ERROR; |
| 3527 | goto put_result; /* only one iteration */ |
| 3528 | } |
| 3529 | goto next_device; |
| 3530 | |
| 3531 | put_result: |
| 3532 | dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, |
| 3533 | cb->nlh->nlmsg_seq, &drbd_genl_family, |
| 3534 | NLM_F_MULTI, DRBD_ADM_GET_PEER_DEVICES); |
| 3535 | err = -ENOMEM; |
| 3536 | if (!dh) |
| 3537 | goto out; |
| 3538 | dh->ret_code = retcode; |
| 3539 | dh->minor = -1U; |
| 3540 | if (retcode == NO_ERROR) { |
| 3541 | struct peer_device_info peer_device_info; |
| 3542 | struct peer_device_statistics peer_device_statistics; |
| 3543 | |
| 3544 | dh->minor = minor; |
| 3545 | err = nla_put_drbd_cfg_context(skb, device->resource, peer_device->connection, device); |
| 3546 | if (err) |
| 3547 | goto out; |
| 3548 | peer_device_to_info(&peer_device_info, peer_device); |
| 3549 | err = peer_device_info_to_skb(skb, &peer_device_info, !capable(CAP_SYS_ADMIN)); |
| 3550 | if (err) |
| 3551 | goto out; |
| 3552 | peer_device_to_statistics(&peer_device_statistics, peer_device); |
| 3553 | err = peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN)); |
| 3554 | if (err) |
| 3555 | goto out; |
| 3556 | cb->args[1] = minor; |
| 3557 | cb->args[2] = (long)peer_device; |
| 3558 | } |
| 3559 | genlmsg_end(skb, dh); |
| 3560 | err = 0; |
| 3561 | |
| 3562 | out: |
| 3563 | rcu_read_unlock(); |
| 3564 | if (err) |
| 3565 | return err; |
| 3566 | return skb->len; |
| 3567 | } |
| 3568 | /* |
| 3569 | * Return the connection of @resource if @resource has exactly one connection. |
| 3570 | */ |
| 3571 | static struct drbd_connection *the_only_connection(struct drbd_resource *resource) |
| 3572 | { |
| 3573 | struct list_head *connections = &resource->connections; |
| 3574 | |
| 3575 | if (list_empty(connections) || connections->next->next != connections) |
| 3576 | return NULL; |
| 3577 | return list_first_entry(&resource->connections, struct drbd_connection, connections); |
| 3578 | } |
| 3579 | |
| 3580 | static int nla_put_status_info(struct sk_buff *skb, struct drbd_device *device, |
| 3581 | const struct sib_info *sib) |
| 3582 | { |
| 3583 | struct drbd_resource *resource = device->resource; |
| 3584 | struct state_info *si = NULL; /* for sizeof(si->member); */ |
| 3585 | struct nlattr *nla; |
| 3586 | int got_ldev; |
| 3587 | int err = 0; |
| 3588 | int exclude_sensitive; |
| 3589 | |
| 3590 | /* If sib != NULL, this is drbd_bcast_event, which anyone can listen |
| 3591 | * to. So we better exclude_sensitive information. |
| 3592 | * |
| 3593 | * If sib == NULL, this is drbd_adm_get_status, executed synchronously |
| 3594 | * in the context of the requesting user process. Exclude sensitive |
| 3595 | * information, unless current has superuser. |
| 3596 | * |
| 3597 | * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and |
| 3598 | * relies on the current implementation of netlink_dump(), which |
| 3599 | * executes the dump callback successively from netlink_recvmsg(), |
| 3600 | * always in the context of the receiving process */ |
| 3601 | exclude_sensitive = sib || !capable(CAP_SYS_ADMIN); |
| 3602 | |
| 3603 | got_ldev = get_ldev(device); |
| 3604 | |
| 3605 | /* We need to add connection name and volume number information still. |
| 3606 | * Minor number is in drbd_genlmsghdr. */ |
| 3607 | if (nla_put_drbd_cfg_context(skb, resource, the_only_connection(resource), device)) |
| 3608 | goto nla_put_failure; |
| 3609 | |
| 3610 | if (res_opts_to_skb(skb, &device->resource->res_opts, exclude_sensitive)) |
| 3611 | goto nla_put_failure; |
| 3612 | |
| 3613 | rcu_read_lock(); |
| 3614 | if (got_ldev) { |
| 3615 | struct disk_conf *disk_conf; |
| 3616 | |
| 3617 | disk_conf = rcu_dereference(device->ldev->disk_conf); |
| 3618 | err = disk_conf_to_skb(skb, disk_conf, exclude_sensitive); |
| 3619 | } |
| 3620 | if (!err) { |
| 3621 | struct net_conf *nc; |
| 3622 | |
| 3623 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
| 3624 | if (nc) |
| 3625 | err = net_conf_to_skb(skb, nc, exclude_sensitive); |
| 3626 | } |
| 3627 | rcu_read_unlock(); |
| 3628 | if (err) |
| 3629 | goto nla_put_failure; |
| 3630 | |
| 3631 | nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO); |
| 3632 | if (!nla) |
| 3633 | goto nla_put_failure; |
| 3634 | if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) || |
| 3635 | nla_put_u32(skb, T_current_state, device->state.i) || |
| 3636 | nla_put_u64(skb, T_ed_uuid, device->ed_uuid) || |
| 3637 | nla_put_u64(skb, T_capacity, drbd_get_capacity(device->this_bdev)) || |
| 3638 | nla_put_u64(skb, T_send_cnt, device->send_cnt) || |
| 3639 | nla_put_u64(skb, T_recv_cnt, device->recv_cnt) || |
| 3640 | nla_put_u64(skb, T_read_cnt, device->read_cnt) || |
| 3641 | nla_put_u64(skb, T_writ_cnt, device->writ_cnt) || |
| 3642 | nla_put_u64(skb, T_al_writ_cnt, device->al_writ_cnt) || |
| 3643 | nla_put_u64(skb, T_bm_writ_cnt, device->bm_writ_cnt) || |
| 3644 | nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&device->ap_bio_cnt)) || |
| 3645 | nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&device->ap_pending_cnt)) || |
| 3646 | nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&device->rs_pending_cnt))) |
| 3647 | goto nla_put_failure; |
| 3648 | |
| 3649 | if (got_ldev) { |
| 3650 | int err; |
| 3651 | |
| 3652 | spin_lock_irq(&device->ldev->md.uuid_lock); |
| 3653 | err = nla_put(skb, T_uuids, sizeof(si->uuids), device->ldev->md.uuid); |
| 3654 | spin_unlock_irq(&device->ldev->md.uuid_lock); |
| 3655 | |
| 3656 | if (err) |
| 3657 | goto nla_put_failure; |
| 3658 | |
| 3659 | if (nla_put_u32(skb, T_disk_flags, device->ldev->md.flags) || |
| 3660 | nla_put_u64(skb, T_bits_total, drbd_bm_bits(device)) || |
| 3661 | nla_put_u64(skb, T_bits_oos, drbd_bm_total_weight(device))) |
| 3662 | goto nla_put_failure; |
| 3663 | if (C_SYNC_SOURCE <= device->state.conn && |
| 3664 | C_PAUSED_SYNC_T >= device->state.conn) { |
| 3665 | if (nla_put_u64(skb, T_bits_rs_total, device->rs_total) || |
| 3666 | nla_put_u64(skb, T_bits_rs_failed, device->rs_failed)) |
| 3667 | goto nla_put_failure; |
| 3668 | } |
| 3669 | } |
| 3670 | |
| 3671 | if (sib) { |
| 3672 | switch(sib->sib_reason) { |
| 3673 | case SIB_SYNC_PROGRESS: |
| 3674 | case SIB_GET_STATUS_REPLY: |
| 3675 | break; |
| 3676 | case SIB_STATE_CHANGE: |
| 3677 | if (nla_put_u32(skb, T_prev_state, sib->os.i) || |
| 3678 | nla_put_u32(skb, T_new_state, sib->ns.i)) |
| 3679 | goto nla_put_failure; |
| 3680 | break; |
| 3681 | case SIB_HELPER_POST: |
| 3682 | if (nla_put_u32(skb, T_helper_exit_code, |
| 3683 | sib->helper_exit_code)) |
| 3684 | goto nla_put_failure; |
| 3685 | /* fall through */ |
| 3686 | case SIB_HELPER_PRE: |
| 3687 | if (nla_put_string(skb, T_helper, sib->helper_name)) |
| 3688 | goto nla_put_failure; |
| 3689 | break; |
| 3690 | } |
| 3691 | } |
| 3692 | nla_nest_end(skb, nla); |
| 3693 | |
| 3694 | if (0) |
| 3695 | nla_put_failure: |
| 3696 | err = -EMSGSIZE; |
| 3697 | if (got_ldev) |
| 3698 | put_ldev(device); |
| 3699 | return err; |
| 3700 | } |
| 3701 | |
| 3702 | int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info) |
| 3703 | { |
| 3704 | struct drbd_config_context adm_ctx; |
| 3705 | enum drbd_ret_code retcode; |
| 3706 | int err; |
| 3707 | |
| 3708 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 3709 | if (!adm_ctx.reply_skb) |
| 3710 | return retcode; |
| 3711 | if (retcode != NO_ERROR) |
| 3712 | goto out; |
| 3713 | |
| 3714 | err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.device, NULL); |
| 3715 | if (err) { |
| 3716 | nlmsg_free(adm_ctx.reply_skb); |
| 3717 | return err; |
| 3718 | } |
| 3719 | out: |
| 3720 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 3721 | return 0; |
| 3722 | } |
| 3723 | |
| 3724 | static int get_one_status(struct sk_buff *skb, struct netlink_callback *cb) |
| 3725 | { |
| 3726 | struct drbd_device *device; |
| 3727 | struct drbd_genlmsghdr *dh; |
| 3728 | struct drbd_resource *pos = (struct drbd_resource *)cb->args[0]; |
| 3729 | struct drbd_resource *resource = NULL; |
| 3730 | struct drbd_resource *tmp; |
| 3731 | unsigned volume = cb->args[1]; |
| 3732 | |
| 3733 | /* Open coded, deferred, iteration: |
| 3734 | * for_each_resource_safe(resource, tmp, &drbd_resources) { |
| 3735 | * connection = "first connection of resource or undefined"; |
| 3736 | * idr_for_each_entry(&resource->devices, device, i) { |
| 3737 | * ... |
| 3738 | * } |
| 3739 | * } |
| 3740 | * where resource is cb->args[0]; |
| 3741 | * and i is cb->args[1]; |
| 3742 | * |
| 3743 | * cb->args[2] indicates if we shall loop over all resources, |
| 3744 | * or just dump all volumes of a single resource. |
| 3745 | * |
| 3746 | * This may miss entries inserted after this dump started, |
| 3747 | * or entries deleted before they are reached. |
| 3748 | * |
| 3749 | * We need to make sure the device won't disappear while |
| 3750 | * we are looking at it, and revalidate our iterators |
| 3751 | * on each iteration. |
| 3752 | */ |
| 3753 | |
| 3754 | /* synchronize with conn_create()/drbd_destroy_connection() */ |
| 3755 | rcu_read_lock(); |
| 3756 | /* revalidate iterator position */ |
| 3757 | for_each_resource_rcu(tmp, &drbd_resources) { |
| 3758 | if (pos == NULL) { |
| 3759 | /* first iteration */ |
| 3760 | pos = tmp; |
| 3761 | resource = pos; |
| 3762 | break; |
| 3763 | } |
| 3764 | if (tmp == pos) { |
| 3765 | resource = pos; |
| 3766 | break; |
| 3767 | } |
| 3768 | } |
| 3769 | if (resource) { |
| 3770 | next_resource: |
| 3771 | device = idr_get_next(&resource->devices, &volume); |
| 3772 | if (!device) { |
| 3773 | /* No more volumes to dump on this resource. |
| 3774 | * Advance resource iterator. */ |
| 3775 | pos = list_entry_rcu(resource->resources.next, |
| 3776 | struct drbd_resource, resources); |
| 3777 | /* Did we dump any volume of this resource yet? */ |
| 3778 | if (volume != 0) { |
| 3779 | /* If we reached the end of the list, |
| 3780 | * or only a single resource dump was requested, |
| 3781 | * we are done. */ |
| 3782 | if (&pos->resources == &drbd_resources || cb->args[2]) |
| 3783 | goto out; |
| 3784 | volume = 0; |
| 3785 | resource = pos; |
| 3786 | goto next_resource; |
| 3787 | } |
| 3788 | } |
| 3789 | |
| 3790 | dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, |
| 3791 | cb->nlh->nlmsg_seq, &drbd_genl_family, |
| 3792 | NLM_F_MULTI, DRBD_ADM_GET_STATUS); |
| 3793 | if (!dh) |
| 3794 | goto out; |
| 3795 | |
| 3796 | if (!device) { |
| 3797 | /* This is a connection without a single volume. |
| 3798 | * Suprisingly enough, it may have a network |
| 3799 | * configuration. */ |
| 3800 | struct drbd_connection *connection; |
| 3801 | |
| 3802 | dh->minor = -1U; |
| 3803 | dh->ret_code = NO_ERROR; |
| 3804 | connection = the_only_connection(resource); |
| 3805 | if (nla_put_drbd_cfg_context(skb, resource, connection, NULL)) |
| 3806 | goto cancel; |
| 3807 | if (connection) { |
| 3808 | struct net_conf *nc; |
| 3809 | |
| 3810 | nc = rcu_dereference(connection->net_conf); |
| 3811 | if (nc && net_conf_to_skb(skb, nc, 1) != 0) |
| 3812 | goto cancel; |
| 3813 | } |
| 3814 | goto done; |
| 3815 | } |
| 3816 | |
| 3817 | D_ASSERT(device, device->vnr == volume); |
| 3818 | D_ASSERT(device, device->resource == resource); |
| 3819 | |
| 3820 | dh->minor = device_to_minor(device); |
| 3821 | dh->ret_code = NO_ERROR; |
| 3822 | |
| 3823 | if (nla_put_status_info(skb, device, NULL)) { |
| 3824 | cancel: |
| 3825 | genlmsg_cancel(skb, dh); |
| 3826 | goto out; |
| 3827 | } |
| 3828 | done: |
| 3829 | genlmsg_end(skb, dh); |
| 3830 | } |
| 3831 | |
| 3832 | out: |
| 3833 | rcu_read_unlock(); |
| 3834 | /* where to start the next iteration */ |
| 3835 | cb->args[0] = (long)pos; |
| 3836 | cb->args[1] = (pos == resource) ? volume + 1 : 0; |
| 3837 | |
| 3838 | /* No more resources/volumes/minors found results in an empty skb. |
| 3839 | * Which will terminate the dump. */ |
| 3840 | return skb->len; |
| 3841 | } |
| 3842 | |
| 3843 | /* |
| 3844 | * Request status of all resources, or of all volumes within a single resource. |
| 3845 | * |
| 3846 | * This is a dump, as the answer may not fit in a single reply skb otherwise. |
| 3847 | * Which means we cannot use the family->attrbuf or other such members, because |
| 3848 | * dump is NOT protected by the genl_lock(). During dump, we only have access |
| 3849 | * to the incoming skb, and need to opencode "parsing" of the nlattr payload. |
| 3850 | * |
| 3851 | * Once things are setup properly, we call into get_one_status(). |
| 3852 | */ |
| 3853 | int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb) |
| 3854 | { |
| 3855 | const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ; |
| 3856 | struct nlattr *nla; |
| 3857 | const char *resource_name; |
| 3858 | struct drbd_resource *resource; |
| 3859 | int maxtype; |
| 3860 | |
| 3861 | /* Is this a followup call? */ |
| 3862 | if (cb->args[0]) { |
| 3863 | /* ... of a single resource dump, |
| 3864 | * and the resource iterator has been advanced already? */ |
| 3865 | if (cb->args[2] && cb->args[2] != cb->args[0]) |
| 3866 | return 0; /* DONE. */ |
| 3867 | goto dump; |
| 3868 | } |
| 3869 | |
| 3870 | /* First call (from netlink_dump_start). We need to figure out |
| 3871 | * which resource(s) the user wants us to dump. */ |
| 3872 | nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen), |
| 3873 | nlmsg_attrlen(cb->nlh, hdrlen), |
| 3874 | DRBD_NLA_CFG_CONTEXT); |
| 3875 | |
| 3876 | /* No explicit context given. Dump all. */ |
| 3877 | if (!nla) |
| 3878 | goto dump; |
| 3879 | maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1; |
| 3880 | nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name)); |
| 3881 | if (IS_ERR(nla)) |
| 3882 | return PTR_ERR(nla); |
| 3883 | /* context given, but no name present? */ |
| 3884 | if (!nla) |
| 3885 | return -EINVAL; |
| 3886 | resource_name = nla_data(nla); |
| 3887 | if (!*resource_name) |
| 3888 | return -ENODEV; |
| 3889 | resource = drbd_find_resource(resource_name); |
| 3890 | if (!resource) |
| 3891 | return -ENODEV; |
| 3892 | |
| 3893 | kref_put(&resource->kref, drbd_destroy_resource); /* get_one_status() revalidates the resource */ |
| 3894 | |
| 3895 | /* prime iterators, and set "filter" mode mark: |
| 3896 | * only dump this connection. */ |
| 3897 | cb->args[0] = (long)resource; |
| 3898 | /* cb->args[1] = 0; passed in this way. */ |
| 3899 | cb->args[2] = (long)resource; |
| 3900 | |
| 3901 | dump: |
| 3902 | return get_one_status(skb, cb); |
| 3903 | } |
| 3904 | |
| 3905 | int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info) |
| 3906 | { |
| 3907 | struct drbd_config_context adm_ctx; |
| 3908 | enum drbd_ret_code retcode; |
| 3909 | struct timeout_parms tp; |
| 3910 | int err; |
| 3911 | |
| 3912 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 3913 | if (!adm_ctx.reply_skb) |
| 3914 | return retcode; |
| 3915 | if (retcode != NO_ERROR) |
| 3916 | goto out; |
| 3917 | |
| 3918 | tp.timeout_type = |
| 3919 | adm_ctx.device->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED : |
| 3920 | test_bit(USE_DEGR_WFC_T, &adm_ctx.device->flags) ? UT_DEGRADED : |
| 3921 | UT_DEFAULT; |
| 3922 | |
| 3923 | err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp); |
| 3924 | if (err) { |
| 3925 | nlmsg_free(adm_ctx.reply_skb); |
| 3926 | return err; |
| 3927 | } |
| 3928 | out: |
| 3929 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 3930 | return 0; |
| 3931 | } |
| 3932 | |
| 3933 | int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info) |
| 3934 | { |
| 3935 | struct drbd_config_context adm_ctx; |
| 3936 | struct drbd_device *device; |
| 3937 | enum drbd_ret_code retcode; |
| 3938 | struct start_ov_parms parms; |
| 3939 | |
| 3940 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 3941 | if (!adm_ctx.reply_skb) |
| 3942 | return retcode; |
| 3943 | if (retcode != NO_ERROR) |
| 3944 | goto out; |
| 3945 | |
| 3946 | device = adm_ctx.device; |
| 3947 | |
| 3948 | /* resume from last known position, if possible */ |
| 3949 | parms.ov_start_sector = device->ov_start_sector; |
| 3950 | parms.ov_stop_sector = ULLONG_MAX; |
| 3951 | if (info->attrs[DRBD_NLA_START_OV_PARMS]) { |
| 3952 | int err = start_ov_parms_from_attrs(&parms, info); |
| 3953 | if (err) { |
| 3954 | retcode = ERR_MANDATORY_TAG; |
| 3955 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 3956 | goto out; |
| 3957 | } |
| 3958 | } |
| 3959 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 3960 | |
| 3961 | /* w_make_ov_request expects position to be aligned */ |
| 3962 | device->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1); |
| 3963 | device->ov_stop_sector = parms.ov_stop_sector; |
| 3964 | |
| 3965 | /* If there is still bitmap IO pending, e.g. previous resync or verify |
| 3966 | * just being finished, wait for it before requesting a new resync. */ |
| 3967 | drbd_suspend_io(device); |
| 3968 | wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags)); |
| 3969 | retcode = drbd_request_state(device, NS(conn, C_VERIFY_S)); |
| 3970 | drbd_resume_io(device); |
| 3971 | |
| 3972 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 3973 | out: |
| 3974 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 3975 | return 0; |
| 3976 | } |
| 3977 | |
| 3978 | |
| 3979 | int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info) |
| 3980 | { |
| 3981 | struct drbd_config_context adm_ctx; |
| 3982 | struct drbd_device *device; |
| 3983 | enum drbd_ret_code retcode; |
| 3984 | int skip_initial_sync = 0; |
| 3985 | int err; |
| 3986 | struct new_c_uuid_parms args; |
| 3987 | |
| 3988 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 3989 | if (!adm_ctx.reply_skb) |
| 3990 | return retcode; |
| 3991 | if (retcode != NO_ERROR) |
| 3992 | goto out_nolock; |
| 3993 | |
| 3994 | device = adm_ctx.device; |
| 3995 | memset(&args, 0, sizeof(args)); |
| 3996 | if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) { |
| 3997 | err = new_c_uuid_parms_from_attrs(&args, info); |
| 3998 | if (err) { |
| 3999 | retcode = ERR_MANDATORY_TAG; |
| 4000 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 4001 | goto out_nolock; |
| 4002 | } |
| 4003 | } |
| 4004 | |
| 4005 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 4006 | mutex_lock(device->state_mutex); /* Protects us against serialized state changes. */ |
| 4007 | |
| 4008 | if (!get_ldev(device)) { |
| 4009 | retcode = ERR_NO_DISK; |
| 4010 | goto out; |
| 4011 | } |
| 4012 | |
| 4013 | /* this is "skip initial sync", assume to be clean */ |
| 4014 | if (device->state.conn == C_CONNECTED && |
| 4015 | first_peer_device(device)->connection->agreed_pro_version >= 90 && |
| 4016 | device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) { |
| 4017 | drbd_info(device, "Preparing to skip initial sync\n"); |
| 4018 | skip_initial_sync = 1; |
| 4019 | } else if (device->state.conn != C_STANDALONE) { |
| 4020 | retcode = ERR_CONNECTED; |
| 4021 | goto out_dec; |
| 4022 | } |
| 4023 | |
| 4024 | drbd_uuid_set(device, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */ |
| 4025 | drbd_uuid_new_current(device); /* New current, previous to UI_BITMAP */ |
| 4026 | |
| 4027 | if (args.clear_bm) { |
| 4028 | err = drbd_bitmap_io(device, &drbd_bmio_clear_n_write, |
| 4029 | "clear_n_write from new_c_uuid", BM_LOCKED_MASK); |
| 4030 | if (err) { |
| 4031 | drbd_err(device, "Writing bitmap failed with %d\n", err); |
| 4032 | retcode = ERR_IO_MD_DISK; |
| 4033 | } |
| 4034 | if (skip_initial_sync) { |
| 4035 | drbd_send_uuids_skip_initial_sync(first_peer_device(device)); |
| 4036 | _drbd_uuid_set(device, UI_BITMAP, 0); |
| 4037 | drbd_print_uuids(device, "cleared bitmap UUID"); |
| 4038 | spin_lock_irq(&device->resource->req_lock); |
| 4039 | _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), |
| 4040 | CS_VERBOSE, NULL); |
| 4041 | spin_unlock_irq(&device->resource->req_lock); |
| 4042 | } |
| 4043 | } |
| 4044 | |
| 4045 | drbd_md_sync(device); |
| 4046 | out_dec: |
| 4047 | put_ldev(device); |
| 4048 | out: |
| 4049 | mutex_unlock(device->state_mutex); |
| 4050 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 4051 | out_nolock: |
| 4052 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 4053 | return 0; |
| 4054 | } |
| 4055 | |
| 4056 | static enum drbd_ret_code |
| 4057 | drbd_check_resource_name(struct drbd_config_context *adm_ctx) |
| 4058 | { |
| 4059 | const char *name = adm_ctx->resource_name; |
| 4060 | if (!name || !name[0]) { |
| 4061 | drbd_msg_put_info(adm_ctx->reply_skb, "resource name missing"); |
| 4062 | return ERR_MANDATORY_TAG; |
| 4063 | } |
| 4064 | /* if we want to use these in sysfs/configfs/debugfs some day, |
| 4065 | * we must not allow slashes */ |
| 4066 | if (strchr(name, '/')) { |
| 4067 | drbd_msg_put_info(adm_ctx->reply_skb, "invalid resource name"); |
| 4068 | return ERR_INVALID_REQUEST; |
| 4069 | } |
| 4070 | return NO_ERROR; |
| 4071 | } |
| 4072 | |
| 4073 | static void resource_to_info(struct resource_info *info, |
| 4074 | struct drbd_resource *resource) |
| 4075 | { |
| 4076 | info->res_role = conn_highest_role(first_connection(resource)); |
| 4077 | info->res_susp = resource->susp; |
| 4078 | info->res_susp_nod = resource->susp_nod; |
| 4079 | info->res_susp_fen = resource->susp_fen; |
| 4080 | } |
| 4081 | |
| 4082 | int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info) |
| 4083 | { |
| 4084 | struct drbd_connection *connection; |
| 4085 | struct drbd_config_context adm_ctx; |
| 4086 | enum drbd_ret_code retcode; |
| 4087 | struct res_opts res_opts; |
| 4088 | int err; |
| 4089 | |
| 4090 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, 0); |
| 4091 | if (!adm_ctx.reply_skb) |
| 4092 | return retcode; |
| 4093 | if (retcode != NO_ERROR) |
| 4094 | goto out; |
| 4095 | |
| 4096 | set_res_opts_defaults(&res_opts); |
| 4097 | err = res_opts_from_attrs(&res_opts, info); |
| 4098 | if (err && err != -ENOMSG) { |
| 4099 | retcode = ERR_MANDATORY_TAG; |
| 4100 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 4101 | goto out; |
| 4102 | } |
| 4103 | |
| 4104 | retcode = drbd_check_resource_name(&adm_ctx); |
| 4105 | if (retcode != NO_ERROR) |
| 4106 | goto out; |
| 4107 | |
| 4108 | if (adm_ctx.resource) { |
| 4109 | if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) { |
| 4110 | retcode = ERR_INVALID_REQUEST; |
| 4111 | drbd_msg_put_info(adm_ctx.reply_skb, "resource exists"); |
| 4112 | } |
| 4113 | /* else: still NO_ERROR */ |
| 4114 | goto out; |
| 4115 | } |
| 4116 | |
| 4117 | /* not yet safe for genl_family.parallel_ops */ |
| 4118 | mutex_lock(&resources_mutex); |
| 4119 | connection = conn_create(adm_ctx.resource_name, &res_opts); |
| 4120 | mutex_unlock(&resources_mutex); |
| 4121 | |
| 4122 | if (connection) { |
| 4123 | struct resource_info resource_info; |
| 4124 | |
| 4125 | mutex_lock(¬ification_mutex); |
| 4126 | resource_to_info(&resource_info, connection->resource); |
| 4127 | notify_resource_state(NULL, 0, connection->resource, |
| 4128 | &resource_info, NOTIFY_CREATE); |
| 4129 | mutex_unlock(¬ification_mutex); |
| 4130 | } else |
| 4131 | retcode = ERR_NOMEM; |
| 4132 | |
| 4133 | out: |
| 4134 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 4135 | return 0; |
| 4136 | } |
| 4137 | |
| 4138 | static void device_to_info(struct device_info *info, |
| 4139 | struct drbd_device *device) |
| 4140 | { |
| 4141 | info->dev_disk_state = device->state.disk; |
| 4142 | } |
| 4143 | |
| 4144 | |
| 4145 | int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info) |
| 4146 | { |
| 4147 | struct drbd_config_context adm_ctx; |
| 4148 | struct drbd_genlmsghdr *dh = info->userhdr; |
| 4149 | enum drbd_ret_code retcode; |
| 4150 | |
| 4151 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE); |
| 4152 | if (!adm_ctx.reply_skb) |
| 4153 | return retcode; |
| 4154 | if (retcode != NO_ERROR) |
| 4155 | goto out; |
| 4156 | |
| 4157 | if (dh->minor > MINORMASK) { |
| 4158 | drbd_msg_put_info(adm_ctx.reply_skb, "requested minor out of range"); |
| 4159 | retcode = ERR_INVALID_REQUEST; |
| 4160 | goto out; |
| 4161 | } |
| 4162 | if (adm_ctx.volume > DRBD_VOLUME_MAX) { |
| 4163 | drbd_msg_put_info(adm_ctx.reply_skb, "requested volume id out of range"); |
| 4164 | retcode = ERR_INVALID_REQUEST; |
| 4165 | goto out; |
| 4166 | } |
| 4167 | |
| 4168 | /* drbd_adm_prepare made sure already |
| 4169 | * that first_peer_device(device)->connection and device->vnr match the request. */ |
| 4170 | if (adm_ctx.device) { |
| 4171 | if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) |
| 4172 | retcode = ERR_MINOR_OR_VOLUME_EXISTS; |
| 4173 | /* else: still NO_ERROR */ |
| 4174 | goto out; |
| 4175 | } |
| 4176 | |
| 4177 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 4178 | retcode = drbd_create_device(&adm_ctx, dh->minor); |
| 4179 | if (retcode == NO_ERROR) { |
| 4180 | struct drbd_device *device; |
| 4181 | struct drbd_peer_device *peer_device; |
| 4182 | struct device_info info; |
| 4183 | unsigned int peer_devices = 0; |
| 4184 | enum drbd_notification_type flags; |
| 4185 | |
| 4186 | device = minor_to_device(dh->minor); |
| 4187 | for_each_peer_device(peer_device, device) { |
| 4188 | if (!has_net_conf(peer_device->connection)) |
| 4189 | continue; |
| 4190 | peer_devices++; |
| 4191 | } |
| 4192 | |
| 4193 | device_to_info(&info, device); |
| 4194 | mutex_lock(¬ification_mutex); |
| 4195 | flags = (peer_devices--) ? NOTIFY_CONTINUES : 0; |
| 4196 | notify_device_state(NULL, 0, device, &info, NOTIFY_CREATE | flags); |
| 4197 | for_each_peer_device(peer_device, device) { |
| 4198 | struct peer_device_info peer_device_info; |
| 4199 | |
| 4200 | if (!has_net_conf(peer_device->connection)) |
| 4201 | continue; |
| 4202 | peer_device_to_info(&peer_device_info, peer_device); |
| 4203 | flags = (peer_devices--) ? NOTIFY_CONTINUES : 0; |
| 4204 | notify_peer_device_state(NULL, 0, peer_device, &peer_device_info, |
| 4205 | NOTIFY_CREATE | flags); |
| 4206 | } |
| 4207 | mutex_unlock(¬ification_mutex); |
| 4208 | } |
| 4209 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 4210 | out: |
| 4211 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 4212 | return 0; |
| 4213 | } |
| 4214 | |
| 4215 | static enum drbd_ret_code adm_del_minor(struct drbd_device *device) |
| 4216 | { |
| 4217 | struct drbd_peer_device *peer_device; |
| 4218 | |
| 4219 | if (device->state.disk == D_DISKLESS && |
| 4220 | /* no need to be device->state.conn == C_STANDALONE && |
| 4221 | * we may want to delete a minor from a live replication group. |
| 4222 | */ |
| 4223 | device->state.role == R_SECONDARY) { |
| 4224 | struct drbd_connection *connection = |
| 4225 | first_connection(device->resource); |
| 4226 | |
| 4227 | _drbd_request_state(device, NS(conn, C_WF_REPORT_PARAMS), |
| 4228 | CS_VERBOSE + CS_WAIT_COMPLETE); |
| 4229 | |
| 4230 | /* If the state engine hasn't stopped the sender thread yet, we |
| 4231 | * need to flush the sender work queue before generating the |
| 4232 | * DESTROY events here. */ |
| 4233 | if (get_t_state(&connection->worker) == RUNNING) |
| 4234 | drbd_flush_workqueue(&connection->sender_work); |
| 4235 | |
| 4236 | mutex_lock(¬ification_mutex); |
| 4237 | for_each_peer_device(peer_device, device) { |
| 4238 | if (!has_net_conf(peer_device->connection)) |
| 4239 | continue; |
| 4240 | notify_peer_device_state(NULL, 0, peer_device, NULL, |
| 4241 | NOTIFY_DESTROY | NOTIFY_CONTINUES); |
| 4242 | } |
| 4243 | notify_device_state(NULL, 0, device, NULL, NOTIFY_DESTROY); |
| 4244 | mutex_unlock(¬ification_mutex); |
| 4245 | |
| 4246 | drbd_delete_device(device); |
| 4247 | return NO_ERROR; |
| 4248 | } else |
| 4249 | return ERR_MINOR_CONFIGURED; |
| 4250 | } |
| 4251 | |
| 4252 | int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info) |
| 4253 | { |
| 4254 | struct drbd_config_context adm_ctx; |
| 4255 | enum drbd_ret_code retcode; |
| 4256 | |
| 4257 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 4258 | if (!adm_ctx.reply_skb) |
| 4259 | return retcode; |
| 4260 | if (retcode != NO_ERROR) |
| 4261 | goto out; |
| 4262 | |
| 4263 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 4264 | retcode = adm_del_minor(adm_ctx.device); |
| 4265 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 4266 | out: |
| 4267 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 4268 | return 0; |
| 4269 | } |
| 4270 | |
| 4271 | static int adm_del_resource(struct drbd_resource *resource) |
| 4272 | { |
| 4273 | struct drbd_connection *connection; |
| 4274 | |
| 4275 | for_each_connection(connection, resource) { |
| 4276 | if (connection->cstate > C_STANDALONE) |
| 4277 | return ERR_NET_CONFIGURED; |
| 4278 | } |
| 4279 | if (!idr_is_empty(&resource->devices)) |
| 4280 | return ERR_RES_IN_USE; |
| 4281 | |
| 4282 | /* The state engine has stopped the sender thread, so we don't |
| 4283 | * need to flush the sender work queue before generating the |
| 4284 | * DESTROY event here. */ |
| 4285 | mutex_lock(¬ification_mutex); |
| 4286 | notify_resource_state(NULL, 0, resource, NULL, NOTIFY_DESTROY); |
| 4287 | mutex_unlock(¬ification_mutex); |
| 4288 | |
| 4289 | mutex_lock(&resources_mutex); |
| 4290 | list_del_rcu(&resource->resources); |
| 4291 | mutex_unlock(&resources_mutex); |
| 4292 | /* Make sure all threads have actually stopped: state handling only |
| 4293 | * does drbd_thread_stop_nowait(). */ |
| 4294 | list_for_each_entry(connection, &resource->connections, connections) |
| 4295 | drbd_thread_stop(&connection->worker); |
| 4296 | synchronize_rcu(); |
| 4297 | drbd_free_resource(resource); |
| 4298 | return NO_ERROR; |
| 4299 | } |
| 4300 | |
| 4301 | int drbd_adm_down(struct sk_buff *skb, struct genl_info *info) |
| 4302 | { |
| 4303 | struct drbd_config_context adm_ctx; |
| 4304 | struct drbd_resource *resource; |
| 4305 | struct drbd_connection *connection; |
| 4306 | struct drbd_device *device; |
| 4307 | int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */ |
| 4308 | unsigned i; |
| 4309 | |
| 4310 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE); |
| 4311 | if (!adm_ctx.reply_skb) |
| 4312 | return retcode; |
| 4313 | if (retcode != NO_ERROR) |
| 4314 | goto finish; |
| 4315 | |
| 4316 | resource = adm_ctx.resource; |
| 4317 | mutex_lock(&resource->adm_mutex); |
| 4318 | /* demote */ |
| 4319 | for_each_connection(connection, resource) { |
| 4320 | struct drbd_peer_device *peer_device; |
| 4321 | |
| 4322 | idr_for_each_entry(&connection->peer_devices, peer_device, i) { |
| 4323 | retcode = drbd_set_role(peer_device->device, R_SECONDARY, 0); |
| 4324 | if (retcode < SS_SUCCESS) { |
| 4325 | drbd_msg_put_info(adm_ctx.reply_skb, "failed to demote"); |
| 4326 | goto out; |
| 4327 | } |
| 4328 | } |
| 4329 | |
| 4330 | retcode = conn_try_disconnect(connection, 0); |
| 4331 | if (retcode < SS_SUCCESS) { |
| 4332 | drbd_msg_put_info(adm_ctx.reply_skb, "failed to disconnect"); |
| 4333 | goto out; |
| 4334 | } |
| 4335 | } |
| 4336 | |
| 4337 | /* detach */ |
| 4338 | idr_for_each_entry(&resource->devices, device, i) { |
| 4339 | retcode = adm_detach(device, 0); |
| 4340 | if (retcode < SS_SUCCESS || retcode > NO_ERROR) { |
| 4341 | drbd_msg_put_info(adm_ctx.reply_skb, "failed to detach"); |
| 4342 | goto out; |
| 4343 | } |
| 4344 | } |
| 4345 | |
| 4346 | /* delete volumes */ |
| 4347 | idr_for_each_entry(&resource->devices, device, i) { |
| 4348 | retcode = adm_del_minor(device); |
| 4349 | if (retcode != NO_ERROR) { |
| 4350 | /* "can not happen" */ |
| 4351 | drbd_msg_put_info(adm_ctx.reply_skb, "failed to delete volume"); |
| 4352 | goto out; |
| 4353 | } |
| 4354 | } |
| 4355 | |
| 4356 | retcode = adm_del_resource(resource); |
| 4357 | out: |
| 4358 | mutex_unlock(&resource->adm_mutex); |
| 4359 | finish: |
| 4360 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 4361 | return 0; |
| 4362 | } |
| 4363 | |
| 4364 | int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info) |
| 4365 | { |
| 4366 | struct drbd_config_context adm_ctx; |
| 4367 | struct drbd_resource *resource; |
| 4368 | enum drbd_ret_code retcode; |
| 4369 | |
| 4370 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE); |
| 4371 | if (!adm_ctx.reply_skb) |
| 4372 | return retcode; |
| 4373 | if (retcode != NO_ERROR) |
| 4374 | goto finish; |
| 4375 | resource = adm_ctx.resource; |
| 4376 | |
| 4377 | mutex_lock(&resource->adm_mutex); |
| 4378 | retcode = adm_del_resource(resource); |
| 4379 | mutex_unlock(&resource->adm_mutex); |
| 4380 | finish: |
| 4381 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 4382 | return 0; |
| 4383 | } |
| 4384 | |
| 4385 | void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib) |
| 4386 | { |
| 4387 | struct sk_buff *msg; |
| 4388 | struct drbd_genlmsghdr *d_out; |
| 4389 | unsigned seq; |
| 4390 | int err = -ENOMEM; |
| 4391 | |
| 4392 | seq = atomic_inc_return(&drbd_genl_seq); |
| 4393 | msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO); |
| 4394 | if (!msg) |
| 4395 | goto failed; |
| 4396 | |
| 4397 | err = -EMSGSIZE; |
| 4398 | d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT); |
| 4399 | if (!d_out) /* cannot happen, but anyways. */ |
| 4400 | goto nla_put_failure; |
| 4401 | d_out->minor = device_to_minor(device); |
| 4402 | d_out->ret_code = NO_ERROR; |
| 4403 | |
| 4404 | if (nla_put_status_info(msg, device, sib)) |
| 4405 | goto nla_put_failure; |
| 4406 | genlmsg_end(msg, d_out); |
| 4407 | err = drbd_genl_multicast_events(msg, GFP_NOWAIT); |
| 4408 | /* msg has been consumed or freed in netlink_broadcast() */ |
| 4409 | if (err && err != -ESRCH) |
| 4410 | goto failed; |
| 4411 | |
| 4412 | return; |
| 4413 | |
| 4414 | nla_put_failure: |
| 4415 | nlmsg_free(msg); |
| 4416 | failed: |
| 4417 | drbd_err(device, "Error %d while broadcasting event. " |
| 4418 | "Event seq:%u sib_reason:%u\n", |
| 4419 | err, seq, sib->sib_reason); |
| 4420 | } |
| 4421 | |
| 4422 | static int nla_put_notification_header(struct sk_buff *msg, |
| 4423 | enum drbd_notification_type type) |
| 4424 | { |
| 4425 | struct drbd_notification_header nh = { |
| 4426 | .nh_type = type, |
| 4427 | }; |
| 4428 | |
| 4429 | return drbd_notification_header_to_skb(msg, &nh, true); |
| 4430 | } |
| 4431 | |
| 4432 | void notify_resource_state(struct sk_buff *skb, |
| 4433 | unsigned int seq, |
| 4434 | struct drbd_resource *resource, |
| 4435 | struct resource_info *resource_info, |
| 4436 | enum drbd_notification_type type) |
| 4437 | { |
| 4438 | struct resource_statistics resource_statistics; |
| 4439 | struct drbd_genlmsghdr *dh; |
| 4440 | bool multicast = false; |
| 4441 | int err; |
| 4442 | |
| 4443 | if (!skb) { |
| 4444 | seq = atomic_inc_return(¬ify_genl_seq); |
| 4445 | skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO); |
| 4446 | err = -ENOMEM; |
| 4447 | if (!skb) |
| 4448 | goto failed; |
| 4449 | multicast = true; |
| 4450 | } |
| 4451 | |
| 4452 | err = -EMSGSIZE; |
| 4453 | dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_RESOURCE_STATE); |
| 4454 | if (!dh) |
| 4455 | goto nla_put_failure; |
| 4456 | dh->minor = -1U; |
| 4457 | dh->ret_code = NO_ERROR; |
| 4458 | if (nla_put_drbd_cfg_context(skb, resource, NULL, NULL) || |
| 4459 | nla_put_notification_header(skb, type) || |
| 4460 | ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY && |
| 4461 | resource_info_to_skb(skb, resource_info, true))) |
| 4462 | goto nla_put_failure; |
| 4463 | resource_statistics.res_stat_write_ordering = resource->write_ordering; |
| 4464 | err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN)); |
| 4465 | if (err) |
| 4466 | goto nla_put_failure; |
| 4467 | genlmsg_end(skb, dh); |
| 4468 | if (multicast) { |
| 4469 | err = drbd_genl_multicast_events(skb, GFP_NOWAIT); |
| 4470 | /* skb has been consumed or freed in netlink_broadcast() */ |
| 4471 | if (err && err != -ESRCH) |
| 4472 | goto failed; |
| 4473 | } |
| 4474 | return; |
| 4475 | |
| 4476 | nla_put_failure: |
| 4477 | nlmsg_free(skb); |
| 4478 | failed: |
| 4479 | drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n", |
| 4480 | err, seq); |
| 4481 | } |
| 4482 | |
| 4483 | void notify_device_state(struct sk_buff *skb, |
| 4484 | unsigned int seq, |
| 4485 | struct drbd_device *device, |
| 4486 | struct device_info *device_info, |
| 4487 | enum drbd_notification_type type) |
| 4488 | { |
| 4489 | struct device_statistics device_statistics; |
| 4490 | struct drbd_genlmsghdr *dh; |
| 4491 | bool multicast = false; |
| 4492 | int err; |
| 4493 | |
| 4494 | if (!skb) { |
| 4495 | seq = atomic_inc_return(¬ify_genl_seq); |
| 4496 | skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO); |
| 4497 | err = -ENOMEM; |
| 4498 | if (!skb) |
| 4499 | goto failed; |
| 4500 | multicast = true; |
| 4501 | } |
| 4502 | |
| 4503 | err = -EMSGSIZE; |
| 4504 | dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_DEVICE_STATE); |
| 4505 | if (!dh) |
| 4506 | goto nla_put_failure; |
| 4507 | dh->minor = device->minor; |
| 4508 | dh->ret_code = NO_ERROR; |
| 4509 | if (nla_put_drbd_cfg_context(skb, device->resource, NULL, device) || |
| 4510 | nla_put_notification_header(skb, type) || |
| 4511 | ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY && |
| 4512 | device_info_to_skb(skb, device_info, true))) |
| 4513 | goto nla_put_failure; |
| 4514 | device_to_statistics(&device_statistics, device); |
| 4515 | device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN)); |
| 4516 | genlmsg_end(skb, dh); |
| 4517 | if (multicast) { |
| 4518 | err = drbd_genl_multicast_events(skb, GFP_NOWAIT); |
| 4519 | /* skb has been consumed or freed in netlink_broadcast() */ |
| 4520 | if (err && err != -ESRCH) |
| 4521 | goto failed; |
| 4522 | } |
| 4523 | return; |
| 4524 | |
| 4525 | nla_put_failure: |
| 4526 | nlmsg_free(skb); |
| 4527 | failed: |
| 4528 | drbd_err(device, "Error %d while broadcasting event. Event seq:%u\n", |
| 4529 | err, seq); |
| 4530 | } |
| 4531 | |
| 4532 | void notify_connection_state(struct sk_buff *skb, |
| 4533 | unsigned int seq, |
| 4534 | struct drbd_connection *connection, |
| 4535 | struct connection_info *connection_info, |
| 4536 | enum drbd_notification_type type) |
| 4537 | { |
| 4538 | struct connection_statistics connection_statistics; |
| 4539 | struct drbd_genlmsghdr *dh; |
| 4540 | bool multicast = false; |
| 4541 | int err; |
| 4542 | |
| 4543 | if (!skb) { |
| 4544 | seq = atomic_inc_return(¬ify_genl_seq); |
| 4545 | skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO); |
| 4546 | err = -ENOMEM; |
| 4547 | if (!skb) |
| 4548 | goto failed; |
| 4549 | multicast = true; |
| 4550 | } |
| 4551 | |
| 4552 | err = -EMSGSIZE; |
| 4553 | dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_CONNECTION_STATE); |
| 4554 | if (!dh) |
| 4555 | goto nla_put_failure; |
| 4556 | dh->minor = -1U; |
| 4557 | dh->ret_code = NO_ERROR; |
| 4558 | if (nla_put_drbd_cfg_context(skb, connection->resource, connection, NULL) || |
| 4559 | nla_put_notification_header(skb, type) || |
| 4560 | ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY && |
| 4561 | connection_info_to_skb(skb, connection_info, true))) |
| 4562 | goto nla_put_failure; |
| 4563 | connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags); |
| 4564 | connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN)); |
| 4565 | genlmsg_end(skb, dh); |
| 4566 | if (multicast) { |
| 4567 | err = drbd_genl_multicast_events(skb, GFP_NOWAIT); |
| 4568 | /* skb has been consumed or freed in netlink_broadcast() */ |
| 4569 | if (err && err != -ESRCH) |
| 4570 | goto failed; |
| 4571 | } |
| 4572 | return; |
| 4573 | |
| 4574 | nla_put_failure: |
| 4575 | nlmsg_free(skb); |
| 4576 | failed: |
| 4577 | drbd_err(connection, "Error %d while broadcasting event. Event seq:%u\n", |
| 4578 | err, seq); |
| 4579 | } |
| 4580 | |
| 4581 | void notify_peer_device_state(struct sk_buff *skb, |
| 4582 | unsigned int seq, |
| 4583 | struct drbd_peer_device *peer_device, |
| 4584 | struct peer_device_info *peer_device_info, |
| 4585 | enum drbd_notification_type type) |
| 4586 | { |
| 4587 | struct peer_device_statistics peer_device_statistics; |
| 4588 | struct drbd_resource *resource = peer_device->device->resource; |
| 4589 | struct drbd_genlmsghdr *dh; |
| 4590 | bool multicast = false; |
| 4591 | int err; |
| 4592 | |
| 4593 | if (!skb) { |
| 4594 | seq = atomic_inc_return(¬ify_genl_seq); |
| 4595 | skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO); |
| 4596 | err = -ENOMEM; |
| 4597 | if (!skb) |
| 4598 | goto failed; |
| 4599 | multicast = true; |
| 4600 | } |
| 4601 | |
| 4602 | err = -EMSGSIZE; |
| 4603 | dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_PEER_DEVICE_STATE); |
| 4604 | if (!dh) |
| 4605 | goto nla_put_failure; |
| 4606 | dh->minor = -1U; |
| 4607 | dh->ret_code = NO_ERROR; |
| 4608 | if (nla_put_drbd_cfg_context(skb, resource, peer_device->connection, peer_device->device) || |
| 4609 | nla_put_notification_header(skb, type) || |
| 4610 | ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY && |
| 4611 | peer_device_info_to_skb(skb, peer_device_info, true))) |
| 4612 | goto nla_put_failure; |
| 4613 | peer_device_to_statistics(&peer_device_statistics, peer_device); |
| 4614 | peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN)); |
| 4615 | genlmsg_end(skb, dh); |
| 4616 | if (multicast) { |
| 4617 | err = drbd_genl_multicast_events(skb, GFP_NOWAIT); |
| 4618 | /* skb has been consumed or freed in netlink_broadcast() */ |
| 4619 | if (err && err != -ESRCH) |
| 4620 | goto failed; |
| 4621 | } |
| 4622 | return; |
| 4623 | |
| 4624 | nla_put_failure: |
| 4625 | nlmsg_free(skb); |
| 4626 | failed: |
| 4627 | drbd_err(peer_device, "Error %d while broadcasting event. Event seq:%u\n", |
| 4628 | err, seq); |
| 4629 | } |
| 4630 | |
| 4631 | void notify_helper(enum drbd_notification_type type, |
| 4632 | struct drbd_device *device, struct drbd_connection *connection, |
| 4633 | const char *name, int status) |
| 4634 | { |
| 4635 | struct drbd_resource *resource = device ? device->resource : connection->resource; |
| 4636 | struct drbd_helper_info helper_info; |
| 4637 | unsigned int seq = atomic_inc_return(¬ify_genl_seq); |
| 4638 | struct sk_buff *skb = NULL; |
| 4639 | struct drbd_genlmsghdr *dh; |
| 4640 | int err; |
| 4641 | |
| 4642 | strlcpy(helper_info.helper_name, name, sizeof(helper_info.helper_name)); |
| 4643 | helper_info.helper_name_len = min(strlen(name), sizeof(helper_info.helper_name)); |
| 4644 | helper_info.helper_status = status; |
| 4645 | |
| 4646 | skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO); |
| 4647 | err = -ENOMEM; |
| 4648 | if (!skb) |
| 4649 | goto fail; |
| 4650 | |
| 4651 | err = -EMSGSIZE; |
| 4652 | dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_HELPER); |
| 4653 | if (!dh) |
| 4654 | goto fail; |
| 4655 | dh->minor = device ? device->minor : -1; |
| 4656 | dh->ret_code = NO_ERROR; |
| 4657 | mutex_lock(¬ification_mutex); |
| 4658 | if (nla_put_drbd_cfg_context(skb, resource, connection, device) || |
| 4659 | nla_put_notification_header(skb, type) || |
| 4660 | drbd_helper_info_to_skb(skb, &helper_info, true)) |
| 4661 | goto unlock_fail; |
| 4662 | genlmsg_end(skb, dh); |
| 4663 | err = drbd_genl_multicast_events(skb, GFP_NOWAIT); |
| 4664 | skb = NULL; |
| 4665 | /* skb has been consumed or freed in netlink_broadcast() */ |
| 4666 | if (err && err != -ESRCH) |
| 4667 | goto unlock_fail; |
| 4668 | mutex_unlock(¬ification_mutex); |
| 4669 | return; |
| 4670 | |
| 4671 | unlock_fail: |
| 4672 | mutex_unlock(¬ification_mutex); |
| 4673 | fail: |
| 4674 | nlmsg_free(skb); |
| 4675 | drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n", |
| 4676 | err, seq); |
| 4677 | } |
| 4678 | |
| 4679 | static void notify_initial_state_done(struct sk_buff *skb, unsigned int seq) |
| 4680 | { |
| 4681 | struct drbd_genlmsghdr *dh; |
| 4682 | int err; |
| 4683 | |
| 4684 | err = -EMSGSIZE; |
| 4685 | dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_INITIAL_STATE_DONE); |
| 4686 | if (!dh) |
| 4687 | goto nla_put_failure; |
| 4688 | dh->minor = -1U; |
| 4689 | dh->ret_code = NO_ERROR; |
| 4690 | if (nla_put_notification_header(skb, NOTIFY_EXISTS)) |
| 4691 | goto nla_put_failure; |
| 4692 | genlmsg_end(skb, dh); |
| 4693 | return; |
| 4694 | |
| 4695 | nla_put_failure: |
| 4696 | nlmsg_free(skb); |
| 4697 | pr_err("Error %d sending event. Event seq:%u\n", err, seq); |
| 4698 | } |
| 4699 | |
| 4700 | static void free_state_changes(struct list_head *list) |
| 4701 | { |
| 4702 | while (!list_empty(list)) { |
| 4703 | struct drbd_state_change *state_change = |
| 4704 | list_first_entry(list, struct drbd_state_change, list); |
| 4705 | list_del(&state_change->list); |
| 4706 | forget_state_change(state_change); |
| 4707 | } |
| 4708 | } |
| 4709 | |
| 4710 | static unsigned int notifications_for_state_change(struct drbd_state_change *state_change) |
| 4711 | { |
| 4712 | return 1 + |
| 4713 | state_change->n_connections + |
| 4714 | state_change->n_devices + |
| 4715 | state_change->n_devices * state_change->n_connections; |
| 4716 | } |
| 4717 | |
| 4718 | static int get_initial_state(struct sk_buff *skb, struct netlink_callback *cb) |
| 4719 | { |
| 4720 | struct drbd_state_change *state_change = (struct drbd_state_change *)cb->args[0]; |
| 4721 | unsigned int seq = cb->args[2]; |
| 4722 | unsigned int n; |
| 4723 | enum drbd_notification_type flags = 0; |
| 4724 | |
| 4725 | /* There is no need for taking notification_mutex here: it doesn't |
| 4726 | matter if the initial state events mix with later state chage |
| 4727 | events; we can always tell the events apart by the NOTIFY_EXISTS |
| 4728 | flag. */ |
| 4729 | |
| 4730 | cb->args[5]--; |
| 4731 | if (cb->args[5] == 1) { |
| 4732 | notify_initial_state_done(skb, seq); |
| 4733 | goto out; |
| 4734 | } |
| 4735 | n = cb->args[4]++; |
| 4736 | if (cb->args[4] < cb->args[3]) |
| 4737 | flags |= NOTIFY_CONTINUES; |
| 4738 | if (n < 1) { |
| 4739 | notify_resource_state_change(skb, seq, state_change->resource, |
| 4740 | NOTIFY_EXISTS | flags); |
| 4741 | goto next; |
| 4742 | } |
| 4743 | n--; |
| 4744 | if (n < state_change->n_connections) { |
| 4745 | notify_connection_state_change(skb, seq, &state_change->connections[n], |
| 4746 | NOTIFY_EXISTS | flags); |
| 4747 | goto next; |
| 4748 | } |
| 4749 | n -= state_change->n_connections; |
| 4750 | if (n < state_change->n_devices) { |
| 4751 | notify_device_state_change(skb, seq, &state_change->devices[n], |
| 4752 | NOTIFY_EXISTS | flags); |
| 4753 | goto next; |
| 4754 | } |
| 4755 | n -= state_change->n_devices; |
| 4756 | if (n < state_change->n_devices * state_change->n_connections) { |
| 4757 | notify_peer_device_state_change(skb, seq, &state_change->peer_devices[n], |
| 4758 | NOTIFY_EXISTS | flags); |
| 4759 | goto next; |
| 4760 | } |
| 4761 | |
| 4762 | next: |
| 4763 | if (cb->args[4] == cb->args[3]) { |
| 4764 | struct drbd_state_change *next_state_change = |
| 4765 | list_entry(state_change->list.next, |
| 4766 | struct drbd_state_change, list); |
| 4767 | cb->args[0] = (long)next_state_change; |
| 4768 | cb->args[3] = notifications_for_state_change(next_state_change); |
| 4769 | cb->args[4] = 0; |
| 4770 | } |
| 4771 | out: |
| 4772 | return skb->len; |
| 4773 | } |
| 4774 | |
| 4775 | int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb) |
| 4776 | { |
| 4777 | struct drbd_resource *resource; |
| 4778 | LIST_HEAD(head); |
| 4779 | |
| 4780 | if (cb->args[5] >= 1) { |
| 4781 | if (cb->args[5] > 1) |
| 4782 | return get_initial_state(skb, cb); |
| 4783 | if (cb->args[0]) { |
| 4784 | struct drbd_state_change *state_change = |
| 4785 | (struct drbd_state_change *)cb->args[0]; |
| 4786 | |
| 4787 | /* connect list to head */ |
| 4788 | list_add(&head, &state_change->list); |
| 4789 | free_state_changes(&head); |
| 4790 | } |
| 4791 | return 0; |
| 4792 | } |
| 4793 | |
| 4794 | cb->args[5] = 2; /* number of iterations */ |
| 4795 | mutex_lock(&resources_mutex); |
| 4796 | for_each_resource(resource, &drbd_resources) { |
| 4797 | struct drbd_state_change *state_change; |
| 4798 | |
| 4799 | state_change = remember_old_state(resource, GFP_KERNEL); |
| 4800 | if (!state_change) { |
| 4801 | if (!list_empty(&head)) |
| 4802 | free_state_changes(&head); |
| 4803 | mutex_unlock(&resources_mutex); |
| 4804 | return -ENOMEM; |
| 4805 | } |
| 4806 | copy_old_to_new_state_change(state_change); |
| 4807 | list_add_tail(&state_change->list, &head); |
| 4808 | cb->args[5] += notifications_for_state_change(state_change); |
| 4809 | } |
| 4810 | mutex_unlock(&resources_mutex); |
| 4811 | |
| 4812 | if (!list_empty(&head)) { |
| 4813 | struct drbd_state_change *state_change = |
| 4814 | list_entry(head.next, struct drbd_state_change, list); |
| 4815 | cb->args[0] = (long)state_change; |
| 4816 | cb->args[3] = notifications_for_state_change(state_change); |
| 4817 | list_del(&head); /* detach list from head */ |
| 4818 | } |
| 4819 | |
| 4820 | cb->args[2] = cb->nlh->nlmsg_seq; |
| 4821 | return get_initial_state(skb, cb); |
| 4822 | } |