| 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 | void drbd_suspend_io(struct drbd_device *device) |
| 869 | { |
| 870 | set_bit(SUSPEND_IO, &device->flags); |
| 871 | if (drbd_suspended(device)) |
| 872 | return; |
| 873 | wait_event(device->misc_wait, !atomic_read(&device->ap_bio_cnt)); |
| 874 | } |
| 875 | |
| 876 | void drbd_resume_io(struct drbd_device *device) |
| 877 | { |
| 878 | clear_bit(SUSPEND_IO, &device->flags); |
| 879 | wake_up(&device->misc_wait); |
| 880 | } |
| 881 | |
| 882 | /** |
| 883 | * drbd_determine_dev_size() - Sets the right device size obeying all constraints |
| 884 | * @device: DRBD device. |
| 885 | * |
| 886 | * Returns 0 on success, negative return values indicate errors. |
| 887 | * You should call drbd_md_sync() after calling this function. |
| 888 | */ |
| 889 | enum determine_dev_size |
| 890 | drbd_determine_dev_size(struct drbd_device *device, enum dds_flags flags, struct resize_parms *rs) __must_hold(local) |
| 891 | { |
| 892 | sector_t prev_first_sect, prev_size; /* previous meta location */ |
| 893 | sector_t la_size_sect, u_size; |
| 894 | struct drbd_md *md = &device->ldev->md; |
| 895 | u32 prev_al_stripe_size_4k; |
| 896 | u32 prev_al_stripes; |
| 897 | sector_t size; |
| 898 | char ppb[10]; |
| 899 | void *buffer; |
| 900 | |
| 901 | int md_moved, la_size_changed; |
| 902 | enum determine_dev_size rv = DS_UNCHANGED; |
| 903 | |
| 904 | /* race: |
| 905 | * application request passes inc_ap_bio, |
| 906 | * but then cannot get an AL-reference. |
| 907 | * this function later may wait on ap_bio_cnt == 0. -> deadlock. |
| 908 | * |
| 909 | * to avoid that: |
| 910 | * Suspend IO right here. |
| 911 | * still lock the act_log to not trigger ASSERTs there. |
| 912 | */ |
| 913 | drbd_suspend_io(device); |
| 914 | buffer = drbd_md_get_buffer(device, __func__); /* Lock meta-data IO */ |
| 915 | if (!buffer) { |
| 916 | drbd_resume_io(device); |
| 917 | return DS_ERROR; |
| 918 | } |
| 919 | |
| 920 | /* no wait necessary anymore, actually we could assert that */ |
| 921 | wait_event(device->al_wait, lc_try_lock(device->act_log)); |
| 922 | |
| 923 | prev_first_sect = drbd_md_first_sector(device->ldev); |
| 924 | prev_size = device->ldev->md.md_size_sect; |
| 925 | la_size_sect = device->ldev->md.la_size_sect; |
| 926 | |
| 927 | if (rs) { |
| 928 | /* rs is non NULL if we should change the AL layout only */ |
| 929 | |
| 930 | prev_al_stripes = md->al_stripes; |
| 931 | prev_al_stripe_size_4k = md->al_stripe_size_4k; |
| 932 | |
| 933 | md->al_stripes = rs->al_stripes; |
| 934 | md->al_stripe_size_4k = rs->al_stripe_size / 4; |
| 935 | md->al_size_4k = (u64)rs->al_stripes * rs->al_stripe_size / 4; |
| 936 | } |
| 937 | |
| 938 | drbd_md_set_sector_offsets(device, device->ldev); |
| 939 | |
| 940 | rcu_read_lock(); |
| 941 | u_size = rcu_dereference(device->ldev->disk_conf)->disk_size; |
| 942 | rcu_read_unlock(); |
| 943 | size = drbd_new_dev_size(device, device->ldev, u_size, flags & DDSF_FORCED); |
| 944 | |
| 945 | if (size < la_size_sect) { |
| 946 | if (rs && u_size == 0) { |
| 947 | /* Remove "rs &&" later. This check should always be active, but |
| 948 | right now the receiver expects the permissive behavior */ |
| 949 | drbd_warn(device, "Implicit shrink not allowed. " |
| 950 | "Use --size=%llus for explicit shrink.\n", |
| 951 | (unsigned long long)size); |
| 952 | rv = DS_ERROR_SHRINK; |
| 953 | } |
| 954 | if (u_size > size) |
| 955 | rv = DS_ERROR_SPACE_MD; |
| 956 | if (rv != DS_UNCHANGED) |
| 957 | goto err_out; |
| 958 | } |
| 959 | |
| 960 | if (drbd_get_capacity(device->this_bdev) != size || |
| 961 | drbd_bm_capacity(device) != size) { |
| 962 | int err; |
| 963 | err = drbd_bm_resize(device, size, !(flags & DDSF_NO_RESYNC)); |
| 964 | if (unlikely(err)) { |
| 965 | /* currently there is only one error: ENOMEM! */ |
| 966 | size = drbd_bm_capacity(device)>>1; |
| 967 | if (size == 0) { |
| 968 | drbd_err(device, "OUT OF MEMORY! " |
| 969 | "Could not allocate bitmap!\n"); |
| 970 | } else { |
| 971 | drbd_err(device, "BM resizing failed. " |
| 972 | "Leaving size unchanged at size = %lu KB\n", |
| 973 | (unsigned long)size); |
| 974 | } |
| 975 | rv = DS_ERROR; |
| 976 | } |
| 977 | /* racy, see comments above. */ |
| 978 | drbd_set_my_capacity(device, size); |
| 979 | device->ldev->md.la_size_sect = size; |
| 980 | drbd_info(device, "size = %s (%llu KB)\n", ppsize(ppb, size>>1), |
| 981 | (unsigned long long)size>>1); |
| 982 | } |
| 983 | if (rv <= DS_ERROR) |
| 984 | goto err_out; |
| 985 | |
| 986 | la_size_changed = (la_size_sect != device->ldev->md.la_size_sect); |
| 987 | |
| 988 | md_moved = prev_first_sect != drbd_md_first_sector(device->ldev) |
| 989 | || prev_size != device->ldev->md.md_size_sect; |
| 990 | |
| 991 | if (la_size_changed || md_moved || rs) { |
| 992 | u32 prev_flags; |
| 993 | |
| 994 | /* We do some synchronous IO below, which may take some time. |
| 995 | * Clear the timer, to avoid scary "timer expired!" messages, |
| 996 | * "Superblock" is written out at least twice below, anyways. */ |
| 997 | del_timer(&device->md_sync_timer); |
| 998 | drbd_al_shrink(device); /* All extents inactive. */ |
| 999 | |
| 1000 | prev_flags = md->flags; |
| 1001 | md->flags &= ~MDF_PRIMARY_IND; |
| 1002 | drbd_md_write(device, buffer); |
| 1003 | |
| 1004 | drbd_info(device, "Writing the whole bitmap, %s\n", |
| 1005 | la_size_changed && md_moved ? "size changed and md moved" : |
| 1006 | la_size_changed ? "size changed" : "md moved"); |
| 1007 | /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */ |
| 1008 | drbd_bitmap_io(device, md_moved ? &drbd_bm_write_all : &drbd_bm_write, |
| 1009 | "size changed", BM_LOCKED_MASK); |
| 1010 | drbd_initialize_al(device, buffer); |
| 1011 | |
| 1012 | md->flags = prev_flags; |
| 1013 | drbd_md_write(device, buffer); |
| 1014 | |
| 1015 | if (rs) |
| 1016 | drbd_info(device, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n", |
| 1017 | md->al_stripes, md->al_stripe_size_4k * 4); |
| 1018 | } |
| 1019 | |
| 1020 | if (size > la_size_sect) |
| 1021 | rv = la_size_sect ? DS_GREW : DS_GREW_FROM_ZERO; |
| 1022 | if (size < la_size_sect) |
| 1023 | rv = DS_SHRUNK; |
| 1024 | |
| 1025 | if (0) { |
| 1026 | err_out: |
| 1027 | if (rs) { |
| 1028 | md->al_stripes = prev_al_stripes; |
| 1029 | md->al_stripe_size_4k = prev_al_stripe_size_4k; |
| 1030 | md->al_size_4k = (u64)prev_al_stripes * prev_al_stripe_size_4k; |
| 1031 | |
| 1032 | drbd_md_set_sector_offsets(device, device->ldev); |
| 1033 | } |
| 1034 | } |
| 1035 | lc_unlock(device->act_log); |
| 1036 | wake_up(&device->al_wait); |
| 1037 | drbd_md_put_buffer(device); |
| 1038 | drbd_resume_io(device); |
| 1039 | |
| 1040 | return rv; |
| 1041 | } |
| 1042 | |
| 1043 | sector_t |
| 1044 | drbd_new_dev_size(struct drbd_device *device, struct drbd_backing_dev *bdev, |
| 1045 | sector_t u_size, int assume_peer_has_space) |
| 1046 | { |
| 1047 | sector_t p_size = device->p_size; /* partner's disk size. */ |
| 1048 | sector_t la_size_sect = bdev->md.la_size_sect; /* last agreed size. */ |
| 1049 | sector_t m_size; /* my size */ |
| 1050 | sector_t size = 0; |
| 1051 | |
| 1052 | m_size = drbd_get_max_capacity(bdev); |
| 1053 | |
| 1054 | if (device->state.conn < C_CONNECTED && assume_peer_has_space) { |
| 1055 | drbd_warn(device, "Resize while not connected was forced by the user!\n"); |
| 1056 | p_size = m_size; |
| 1057 | } |
| 1058 | |
| 1059 | if (p_size && m_size) { |
| 1060 | size = min_t(sector_t, p_size, m_size); |
| 1061 | } else { |
| 1062 | if (la_size_sect) { |
| 1063 | size = la_size_sect; |
| 1064 | if (m_size && m_size < size) |
| 1065 | size = m_size; |
| 1066 | if (p_size && p_size < size) |
| 1067 | size = p_size; |
| 1068 | } else { |
| 1069 | if (m_size) |
| 1070 | size = m_size; |
| 1071 | if (p_size) |
| 1072 | size = p_size; |
| 1073 | } |
| 1074 | } |
| 1075 | |
| 1076 | if (size == 0) |
| 1077 | drbd_err(device, "Both nodes diskless!\n"); |
| 1078 | |
| 1079 | if (u_size) { |
| 1080 | if (u_size > size) |
| 1081 | drbd_err(device, "Requested disk size is too big (%lu > %lu)\n", |
| 1082 | (unsigned long)u_size>>1, (unsigned long)size>>1); |
| 1083 | else |
| 1084 | size = u_size; |
| 1085 | } |
| 1086 | |
| 1087 | return size; |
| 1088 | } |
| 1089 | |
| 1090 | /** |
| 1091 | * drbd_check_al_size() - Ensures that the AL is of the right size |
| 1092 | * @device: DRBD device. |
| 1093 | * |
| 1094 | * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation |
| 1095 | * failed, and 0 on success. You should call drbd_md_sync() after you called |
| 1096 | * this function. |
| 1097 | */ |
| 1098 | static int drbd_check_al_size(struct drbd_device *device, struct disk_conf *dc) |
| 1099 | { |
| 1100 | struct lru_cache *n, *t; |
| 1101 | struct lc_element *e; |
| 1102 | unsigned int in_use; |
| 1103 | int i; |
| 1104 | |
| 1105 | if (device->act_log && |
| 1106 | device->act_log->nr_elements == dc->al_extents) |
| 1107 | return 0; |
| 1108 | |
| 1109 | in_use = 0; |
| 1110 | t = device->act_log; |
| 1111 | n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION, |
| 1112 | dc->al_extents, sizeof(struct lc_element), 0); |
| 1113 | |
| 1114 | if (n == NULL) { |
| 1115 | drbd_err(device, "Cannot allocate act_log lru!\n"); |
| 1116 | return -ENOMEM; |
| 1117 | } |
| 1118 | spin_lock_irq(&device->al_lock); |
| 1119 | if (t) { |
| 1120 | for (i = 0; i < t->nr_elements; i++) { |
| 1121 | e = lc_element_by_index(t, i); |
| 1122 | if (e->refcnt) |
| 1123 | drbd_err(device, "refcnt(%d)==%d\n", |
| 1124 | e->lc_number, e->refcnt); |
| 1125 | in_use += e->refcnt; |
| 1126 | } |
| 1127 | } |
| 1128 | if (!in_use) |
| 1129 | device->act_log = n; |
| 1130 | spin_unlock_irq(&device->al_lock); |
| 1131 | if (in_use) { |
| 1132 | drbd_err(device, "Activity log still in use!\n"); |
| 1133 | lc_destroy(n); |
| 1134 | return -EBUSY; |
| 1135 | } else { |
| 1136 | lc_destroy(t); |
| 1137 | } |
| 1138 | drbd_md_mark_dirty(device); /* we changed device->act_log->nr_elemens */ |
| 1139 | return 0; |
| 1140 | } |
| 1141 | |
| 1142 | static void drbd_setup_queue_param(struct drbd_device *device, struct drbd_backing_dev *bdev, |
| 1143 | unsigned int max_bio_size) |
| 1144 | { |
| 1145 | struct request_queue * const q = device->rq_queue; |
| 1146 | unsigned int max_hw_sectors = max_bio_size >> 9; |
| 1147 | unsigned int max_segments = 0; |
| 1148 | struct request_queue *b = NULL; |
| 1149 | |
| 1150 | if (bdev) { |
| 1151 | b = bdev->backing_bdev->bd_disk->queue; |
| 1152 | |
| 1153 | max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9); |
| 1154 | rcu_read_lock(); |
| 1155 | max_segments = rcu_dereference(device->ldev->disk_conf)->max_bio_bvecs; |
| 1156 | rcu_read_unlock(); |
| 1157 | |
| 1158 | blk_set_stacking_limits(&q->limits); |
| 1159 | blk_queue_max_write_same_sectors(q, 0); |
| 1160 | } |
| 1161 | |
| 1162 | blk_queue_logical_block_size(q, 512); |
| 1163 | blk_queue_max_hw_sectors(q, max_hw_sectors); |
| 1164 | /* This is the workaround for "bio would need to, but cannot, be split" */ |
| 1165 | blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS); |
| 1166 | blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1); |
| 1167 | |
| 1168 | if (b) { |
| 1169 | struct drbd_connection *connection = first_peer_device(device)->connection; |
| 1170 | |
| 1171 | if (blk_queue_discard(b) && |
| 1172 | (connection->cstate < C_CONNECTED || connection->agreed_features & FF_TRIM)) { |
| 1173 | /* For now, don't allow more than one activity log extent worth of data |
| 1174 | * to be discarded in one go. We may need to rework drbd_al_begin_io() |
| 1175 | * to allow for even larger discard ranges */ |
| 1176 | blk_queue_max_discard_sectors(q, DRBD_MAX_DISCARD_SECTORS); |
| 1177 | |
| 1178 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); |
| 1179 | /* REALLY? Is stacking secdiscard "legal"? */ |
| 1180 | if (blk_queue_secdiscard(b)) |
| 1181 | queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, q); |
| 1182 | } else { |
| 1183 | blk_queue_max_discard_sectors(q, 0); |
| 1184 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); |
| 1185 | queue_flag_clear_unlocked(QUEUE_FLAG_SECDISCARD, q); |
| 1186 | } |
| 1187 | |
| 1188 | blk_queue_stack_limits(q, b); |
| 1189 | |
| 1190 | if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) { |
| 1191 | drbd_info(device, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n", |
| 1192 | q->backing_dev_info.ra_pages, |
| 1193 | b->backing_dev_info.ra_pages); |
| 1194 | q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages; |
| 1195 | } |
| 1196 | } |
| 1197 | } |
| 1198 | |
| 1199 | void drbd_reconsider_max_bio_size(struct drbd_device *device, struct drbd_backing_dev *bdev) |
| 1200 | { |
| 1201 | unsigned int now, new, local, peer; |
| 1202 | |
| 1203 | now = queue_max_hw_sectors(device->rq_queue) << 9; |
| 1204 | local = device->local_max_bio_size; /* Eventually last known value, from volatile memory */ |
| 1205 | peer = device->peer_max_bio_size; /* Eventually last known value, from meta data */ |
| 1206 | |
| 1207 | if (bdev) { |
| 1208 | local = queue_max_hw_sectors(bdev->backing_bdev->bd_disk->queue) << 9; |
| 1209 | device->local_max_bio_size = local; |
| 1210 | } |
| 1211 | local = min(local, DRBD_MAX_BIO_SIZE); |
| 1212 | |
| 1213 | /* We may ignore peer limits if the peer is modern enough. |
| 1214 | Because new from 8.3.8 onwards the peer can use multiple |
| 1215 | BIOs for a single peer_request */ |
| 1216 | if (device->state.conn >= C_WF_REPORT_PARAMS) { |
| 1217 | if (first_peer_device(device)->connection->agreed_pro_version < 94) |
| 1218 | peer = min(device->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET); |
| 1219 | /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */ |
| 1220 | else if (first_peer_device(device)->connection->agreed_pro_version == 94) |
| 1221 | peer = DRBD_MAX_SIZE_H80_PACKET; |
| 1222 | else if (first_peer_device(device)->connection->agreed_pro_version < 100) |
| 1223 | peer = DRBD_MAX_BIO_SIZE_P95; /* drbd 8.3.8 onwards, before 8.4.0 */ |
| 1224 | else |
| 1225 | peer = DRBD_MAX_BIO_SIZE; |
| 1226 | |
| 1227 | /* We may later detach and re-attach on a disconnected Primary. |
| 1228 | * Avoid this setting to jump back in that case. |
| 1229 | * We want to store what we know the peer DRBD can handle, |
| 1230 | * not what the peer IO backend can handle. */ |
| 1231 | if (peer > device->peer_max_bio_size) |
| 1232 | device->peer_max_bio_size = peer; |
| 1233 | } |
| 1234 | new = min(local, peer); |
| 1235 | |
| 1236 | if (device->state.role == R_PRIMARY && new < now) |
| 1237 | drbd_err(device, "ASSERT FAILED new < now; (%u < %u)\n", new, now); |
| 1238 | |
| 1239 | if (new != now) |
| 1240 | drbd_info(device, "max BIO size = %u\n", new); |
| 1241 | |
| 1242 | drbd_setup_queue_param(device, bdev, new); |
| 1243 | } |
| 1244 | |
| 1245 | /* Starts the worker thread */ |
| 1246 | static void conn_reconfig_start(struct drbd_connection *connection) |
| 1247 | { |
| 1248 | drbd_thread_start(&connection->worker); |
| 1249 | drbd_flush_workqueue(&connection->sender_work); |
| 1250 | } |
| 1251 | |
| 1252 | /* if still unconfigured, stops worker again. */ |
| 1253 | static void conn_reconfig_done(struct drbd_connection *connection) |
| 1254 | { |
| 1255 | bool stop_threads; |
| 1256 | spin_lock_irq(&connection->resource->req_lock); |
| 1257 | stop_threads = conn_all_vols_unconf(connection) && |
| 1258 | connection->cstate == C_STANDALONE; |
| 1259 | spin_unlock_irq(&connection->resource->req_lock); |
| 1260 | if (stop_threads) { |
| 1261 | /* asender is implicitly stopped by receiver |
| 1262 | * in conn_disconnect() */ |
| 1263 | drbd_thread_stop(&connection->receiver); |
| 1264 | drbd_thread_stop(&connection->worker); |
| 1265 | } |
| 1266 | } |
| 1267 | |
| 1268 | /* Make sure IO is suspended before calling this function(). */ |
| 1269 | static void drbd_suspend_al(struct drbd_device *device) |
| 1270 | { |
| 1271 | int s = 0; |
| 1272 | |
| 1273 | if (!lc_try_lock(device->act_log)) { |
| 1274 | drbd_warn(device, "Failed to lock al in drbd_suspend_al()\n"); |
| 1275 | return; |
| 1276 | } |
| 1277 | |
| 1278 | drbd_al_shrink(device); |
| 1279 | spin_lock_irq(&device->resource->req_lock); |
| 1280 | if (device->state.conn < C_CONNECTED) |
| 1281 | s = !test_and_set_bit(AL_SUSPENDED, &device->flags); |
| 1282 | spin_unlock_irq(&device->resource->req_lock); |
| 1283 | lc_unlock(device->act_log); |
| 1284 | |
| 1285 | if (s) |
| 1286 | drbd_info(device, "Suspended AL updates\n"); |
| 1287 | } |
| 1288 | |
| 1289 | |
| 1290 | static bool should_set_defaults(struct genl_info *info) |
| 1291 | { |
| 1292 | unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags; |
| 1293 | return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS); |
| 1294 | } |
| 1295 | |
| 1296 | static unsigned int drbd_al_extents_max(struct drbd_backing_dev *bdev) |
| 1297 | { |
| 1298 | /* This is limited by 16 bit "slot" numbers, |
| 1299 | * and by available on-disk context storage. |
| 1300 | * |
| 1301 | * Also (u16)~0 is special (denotes a "free" extent). |
| 1302 | * |
| 1303 | * One transaction occupies one 4kB on-disk block, |
| 1304 | * we have n such blocks in the on disk ring buffer, |
| 1305 | * the "current" transaction may fail (n-1), |
| 1306 | * and there is 919 slot numbers context information per transaction. |
| 1307 | * |
| 1308 | * 72 transaction blocks amounts to more than 2**16 context slots, |
| 1309 | * so cap there first. |
| 1310 | */ |
| 1311 | const unsigned int max_al_nr = DRBD_AL_EXTENTS_MAX; |
| 1312 | const unsigned int sufficient_on_disk = |
| 1313 | (max_al_nr + AL_CONTEXT_PER_TRANSACTION -1) |
| 1314 | /AL_CONTEXT_PER_TRANSACTION; |
| 1315 | |
| 1316 | unsigned int al_size_4k = bdev->md.al_size_4k; |
| 1317 | |
| 1318 | if (al_size_4k > sufficient_on_disk) |
| 1319 | return max_al_nr; |
| 1320 | |
| 1321 | return (al_size_4k - 1) * AL_CONTEXT_PER_TRANSACTION; |
| 1322 | } |
| 1323 | |
| 1324 | static bool write_ordering_changed(struct disk_conf *a, struct disk_conf *b) |
| 1325 | { |
| 1326 | return a->disk_barrier != b->disk_barrier || |
| 1327 | a->disk_flushes != b->disk_flushes || |
| 1328 | a->disk_drain != b->disk_drain; |
| 1329 | } |
| 1330 | |
| 1331 | int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info) |
| 1332 | { |
| 1333 | struct drbd_config_context adm_ctx; |
| 1334 | enum drbd_ret_code retcode; |
| 1335 | struct drbd_device *device; |
| 1336 | struct disk_conf *new_disk_conf, *old_disk_conf; |
| 1337 | struct fifo_buffer *old_plan = NULL, *new_plan = NULL; |
| 1338 | int err, fifo_size; |
| 1339 | |
| 1340 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 1341 | if (!adm_ctx.reply_skb) |
| 1342 | return retcode; |
| 1343 | if (retcode != NO_ERROR) |
| 1344 | goto finish; |
| 1345 | |
| 1346 | device = adm_ctx.device; |
| 1347 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 1348 | |
| 1349 | /* we also need a disk |
| 1350 | * to change the options on */ |
| 1351 | if (!get_ldev(device)) { |
| 1352 | retcode = ERR_NO_DISK; |
| 1353 | goto out; |
| 1354 | } |
| 1355 | |
| 1356 | new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL); |
| 1357 | if (!new_disk_conf) { |
| 1358 | retcode = ERR_NOMEM; |
| 1359 | goto fail; |
| 1360 | } |
| 1361 | |
| 1362 | mutex_lock(&device->resource->conf_update); |
| 1363 | old_disk_conf = device->ldev->disk_conf; |
| 1364 | *new_disk_conf = *old_disk_conf; |
| 1365 | if (should_set_defaults(info)) |
| 1366 | set_disk_conf_defaults(new_disk_conf); |
| 1367 | |
| 1368 | err = disk_conf_from_attrs_for_change(new_disk_conf, info); |
| 1369 | if (err && err != -ENOMSG) { |
| 1370 | retcode = ERR_MANDATORY_TAG; |
| 1371 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 1372 | goto fail_unlock; |
| 1373 | } |
| 1374 | |
| 1375 | if (!expect(new_disk_conf->resync_rate >= 1)) |
| 1376 | new_disk_conf->resync_rate = 1; |
| 1377 | |
| 1378 | if (new_disk_conf->al_extents < DRBD_AL_EXTENTS_MIN) |
| 1379 | new_disk_conf->al_extents = DRBD_AL_EXTENTS_MIN; |
| 1380 | if (new_disk_conf->al_extents > drbd_al_extents_max(device->ldev)) |
| 1381 | new_disk_conf->al_extents = drbd_al_extents_max(device->ldev); |
| 1382 | |
| 1383 | if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX) |
| 1384 | new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX; |
| 1385 | |
| 1386 | fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ; |
| 1387 | if (fifo_size != device->rs_plan_s->size) { |
| 1388 | new_plan = fifo_alloc(fifo_size); |
| 1389 | if (!new_plan) { |
| 1390 | drbd_err(device, "kmalloc of fifo_buffer failed"); |
| 1391 | retcode = ERR_NOMEM; |
| 1392 | goto fail_unlock; |
| 1393 | } |
| 1394 | } |
| 1395 | |
| 1396 | drbd_suspend_io(device); |
| 1397 | wait_event(device->al_wait, lc_try_lock(device->act_log)); |
| 1398 | drbd_al_shrink(device); |
| 1399 | err = drbd_check_al_size(device, new_disk_conf); |
| 1400 | lc_unlock(device->act_log); |
| 1401 | wake_up(&device->al_wait); |
| 1402 | drbd_resume_io(device); |
| 1403 | |
| 1404 | if (err) { |
| 1405 | retcode = ERR_NOMEM; |
| 1406 | goto fail_unlock; |
| 1407 | } |
| 1408 | |
| 1409 | lock_all_resources(); |
| 1410 | retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after); |
| 1411 | if (retcode == NO_ERROR) { |
| 1412 | rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); |
| 1413 | drbd_resync_after_changed(device); |
| 1414 | } |
| 1415 | unlock_all_resources(); |
| 1416 | |
| 1417 | if (retcode != NO_ERROR) |
| 1418 | goto fail_unlock; |
| 1419 | |
| 1420 | if (new_plan) { |
| 1421 | old_plan = device->rs_plan_s; |
| 1422 | rcu_assign_pointer(device->rs_plan_s, new_plan); |
| 1423 | } |
| 1424 | |
| 1425 | mutex_unlock(&device->resource->conf_update); |
| 1426 | |
| 1427 | if (new_disk_conf->al_updates) |
| 1428 | device->ldev->md.flags &= ~MDF_AL_DISABLED; |
| 1429 | else |
| 1430 | device->ldev->md.flags |= MDF_AL_DISABLED; |
| 1431 | |
| 1432 | if (new_disk_conf->md_flushes) |
| 1433 | clear_bit(MD_NO_FUA, &device->flags); |
| 1434 | else |
| 1435 | set_bit(MD_NO_FUA, &device->flags); |
| 1436 | |
| 1437 | if (write_ordering_changed(old_disk_conf, new_disk_conf)) |
| 1438 | drbd_bump_write_ordering(device->resource, NULL, WO_BDEV_FLUSH); |
| 1439 | |
| 1440 | drbd_md_sync(device); |
| 1441 | |
| 1442 | if (device->state.conn >= C_CONNECTED) { |
| 1443 | struct drbd_peer_device *peer_device; |
| 1444 | |
| 1445 | for_each_peer_device(peer_device, device) |
| 1446 | drbd_send_sync_param(peer_device); |
| 1447 | } |
| 1448 | |
| 1449 | synchronize_rcu(); |
| 1450 | kfree(old_disk_conf); |
| 1451 | kfree(old_plan); |
| 1452 | mod_timer(&device->request_timer, jiffies + HZ); |
| 1453 | goto success; |
| 1454 | |
| 1455 | fail_unlock: |
| 1456 | mutex_unlock(&device->resource->conf_update); |
| 1457 | fail: |
| 1458 | kfree(new_disk_conf); |
| 1459 | kfree(new_plan); |
| 1460 | success: |
| 1461 | put_ldev(device); |
| 1462 | out: |
| 1463 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 1464 | finish: |
| 1465 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 1466 | return 0; |
| 1467 | } |
| 1468 | |
| 1469 | int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info) |
| 1470 | { |
| 1471 | struct drbd_config_context adm_ctx; |
| 1472 | struct drbd_device *device; |
| 1473 | struct drbd_peer_device *peer_device; |
| 1474 | struct drbd_connection *connection; |
| 1475 | int err; |
| 1476 | enum drbd_ret_code retcode; |
| 1477 | enum determine_dev_size dd; |
| 1478 | sector_t max_possible_sectors; |
| 1479 | sector_t min_md_device_sectors; |
| 1480 | struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */ |
| 1481 | struct disk_conf *new_disk_conf = NULL; |
| 1482 | struct block_device *bdev; |
| 1483 | struct lru_cache *resync_lru = NULL; |
| 1484 | struct fifo_buffer *new_plan = NULL; |
| 1485 | union drbd_state ns, os; |
| 1486 | enum drbd_state_rv rv; |
| 1487 | struct net_conf *nc; |
| 1488 | |
| 1489 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 1490 | if (!adm_ctx.reply_skb) |
| 1491 | return retcode; |
| 1492 | if (retcode != NO_ERROR) |
| 1493 | goto finish; |
| 1494 | |
| 1495 | device = adm_ctx.device; |
| 1496 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 1497 | peer_device = first_peer_device(device); |
| 1498 | connection = peer_device->connection; |
| 1499 | conn_reconfig_start(connection); |
| 1500 | |
| 1501 | /* if you want to reconfigure, please tear down first */ |
| 1502 | if (device->state.disk > D_DISKLESS) { |
| 1503 | retcode = ERR_DISK_CONFIGURED; |
| 1504 | goto fail; |
| 1505 | } |
| 1506 | /* It may just now have detached because of IO error. Make sure |
| 1507 | * drbd_ldev_destroy is done already, we may end up here very fast, |
| 1508 | * e.g. if someone calls attach from the on-io-error handler, |
| 1509 | * to realize a "hot spare" feature (not that I'd recommend that) */ |
| 1510 | wait_event(device->misc_wait, !test_bit(GOING_DISKLESS, &device->flags)); |
| 1511 | |
| 1512 | /* make sure there is no leftover from previous force-detach attempts */ |
| 1513 | clear_bit(FORCE_DETACH, &device->flags); |
| 1514 | clear_bit(WAS_IO_ERROR, &device->flags); |
| 1515 | clear_bit(WAS_READ_ERROR, &device->flags); |
| 1516 | |
| 1517 | /* and no leftover from previously aborted resync or verify, either */ |
| 1518 | device->rs_total = 0; |
| 1519 | device->rs_failed = 0; |
| 1520 | atomic_set(&device->rs_pending_cnt, 0); |
| 1521 | |
| 1522 | /* allocation not in the IO path, drbdsetup context */ |
| 1523 | nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL); |
| 1524 | if (!nbc) { |
| 1525 | retcode = ERR_NOMEM; |
| 1526 | goto fail; |
| 1527 | } |
| 1528 | spin_lock_init(&nbc->md.uuid_lock); |
| 1529 | |
| 1530 | new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL); |
| 1531 | if (!new_disk_conf) { |
| 1532 | retcode = ERR_NOMEM; |
| 1533 | goto fail; |
| 1534 | } |
| 1535 | nbc->disk_conf = new_disk_conf; |
| 1536 | |
| 1537 | set_disk_conf_defaults(new_disk_conf); |
| 1538 | err = disk_conf_from_attrs(new_disk_conf, info); |
| 1539 | if (err) { |
| 1540 | retcode = ERR_MANDATORY_TAG; |
| 1541 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 1542 | goto fail; |
| 1543 | } |
| 1544 | |
| 1545 | if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX) |
| 1546 | new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX; |
| 1547 | |
| 1548 | new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ); |
| 1549 | if (!new_plan) { |
| 1550 | retcode = ERR_NOMEM; |
| 1551 | goto fail; |
| 1552 | } |
| 1553 | |
| 1554 | if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) { |
| 1555 | retcode = ERR_MD_IDX_INVALID; |
| 1556 | goto fail; |
| 1557 | } |
| 1558 | |
| 1559 | rcu_read_lock(); |
| 1560 | nc = rcu_dereference(connection->net_conf); |
| 1561 | if (nc) { |
| 1562 | if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) { |
| 1563 | rcu_read_unlock(); |
| 1564 | retcode = ERR_STONITH_AND_PROT_A; |
| 1565 | goto fail; |
| 1566 | } |
| 1567 | } |
| 1568 | rcu_read_unlock(); |
| 1569 | |
| 1570 | bdev = blkdev_get_by_path(new_disk_conf->backing_dev, |
| 1571 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, device); |
| 1572 | if (IS_ERR(bdev)) { |
| 1573 | drbd_err(device, "open(\"%s\") failed with %ld\n", new_disk_conf->backing_dev, |
| 1574 | PTR_ERR(bdev)); |
| 1575 | retcode = ERR_OPEN_DISK; |
| 1576 | goto fail; |
| 1577 | } |
| 1578 | nbc->backing_bdev = bdev; |
| 1579 | |
| 1580 | /* |
| 1581 | * meta_dev_idx >= 0: external fixed size, possibly multiple |
| 1582 | * drbd sharing one meta device. TODO in that case, paranoia |
| 1583 | * check that [md_bdev, meta_dev_idx] is not yet used by some |
| 1584 | * other drbd minor! (if you use drbd.conf + drbdadm, that |
| 1585 | * should check it for you already; but if you don't, or |
| 1586 | * someone fooled it, we need to double check here) |
| 1587 | */ |
| 1588 | bdev = blkdev_get_by_path(new_disk_conf->meta_dev, |
| 1589 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, |
| 1590 | (new_disk_conf->meta_dev_idx < 0) ? |
| 1591 | (void *)device : (void *)drbd_m_holder); |
| 1592 | if (IS_ERR(bdev)) { |
| 1593 | drbd_err(device, "open(\"%s\") failed with %ld\n", new_disk_conf->meta_dev, |
| 1594 | PTR_ERR(bdev)); |
| 1595 | retcode = ERR_OPEN_MD_DISK; |
| 1596 | goto fail; |
| 1597 | } |
| 1598 | nbc->md_bdev = bdev; |
| 1599 | |
| 1600 | if ((nbc->backing_bdev == nbc->md_bdev) != |
| 1601 | (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL || |
| 1602 | new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) { |
| 1603 | retcode = ERR_MD_IDX_INVALID; |
| 1604 | goto fail; |
| 1605 | } |
| 1606 | |
| 1607 | resync_lru = lc_create("resync", drbd_bm_ext_cache, |
| 1608 | 1, 61, sizeof(struct bm_extent), |
| 1609 | offsetof(struct bm_extent, lce)); |
| 1610 | if (!resync_lru) { |
| 1611 | retcode = ERR_NOMEM; |
| 1612 | goto fail; |
| 1613 | } |
| 1614 | |
| 1615 | /* Read our meta data super block early. |
| 1616 | * This also sets other on-disk offsets. */ |
| 1617 | retcode = drbd_md_read(device, nbc); |
| 1618 | if (retcode != NO_ERROR) |
| 1619 | goto fail; |
| 1620 | |
| 1621 | if (new_disk_conf->al_extents < DRBD_AL_EXTENTS_MIN) |
| 1622 | new_disk_conf->al_extents = DRBD_AL_EXTENTS_MIN; |
| 1623 | if (new_disk_conf->al_extents > drbd_al_extents_max(nbc)) |
| 1624 | new_disk_conf->al_extents = drbd_al_extents_max(nbc); |
| 1625 | |
| 1626 | if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) { |
| 1627 | drbd_err(device, "max capacity %llu smaller than disk size %llu\n", |
| 1628 | (unsigned long long) drbd_get_max_capacity(nbc), |
| 1629 | (unsigned long long) new_disk_conf->disk_size); |
| 1630 | retcode = ERR_DISK_TOO_SMALL; |
| 1631 | goto fail; |
| 1632 | } |
| 1633 | |
| 1634 | if (new_disk_conf->meta_dev_idx < 0) { |
| 1635 | max_possible_sectors = DRBD_MAX_SECTORS_FLEX; |
| 1636 | /* at least one MB, otherwise it does not make sense */ |
| 1637 | min_md_device_sectors = (2<<10); |
| 1638 | } else { |
| 1639 | max_possible_sectors = DRBD_MAX_SECTORS; |
| 1640 | min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1); |
| 1641 | } |
| 1642 | |
| 1643 | if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) { |
| 1644 | retcode = ERR_MD_DISK_TOO_SMALL; |
| 1645 | drbd_warn(device, "refusing attach: md-device too small, " |
| 1646 | "at least %llu sectors needed for this meta-disk type\n", |
| 1647 | (unsigned long long) min_md_device_sectors); |
| 1648 | goto fail; |
| 1649 | } |
| 1650 | |
| 1651 | /* Make sure the new disk is big enough |
| 1652 | * (we may currently be R_PRIMARY with no local disk...) */ |
| 1653 | if (drbd_get_max_capacity(nbc) < |
| 1654 | drbd_get_capacity(device->this_bdev)) { |
| 1655 | retcode = ERR_DISK_TOO_SMALL; |
| 1656 | goto fail; |
| 1657 | } |
| 1658 | |
| 1659 | nbc->known_size = drbd_get_capacity(nbc->backing_bdev); |
| 1660 | |
| 1661 | if (nbc->known_size > max_possible_sectors) { |
| 1662 | drbd_warn(device, "==> truncating very big lower level device " |
| 1663 | "to currently maximum possible %llu sectors <==\n", |
| 1664 | (unsigned long long) max_possible_sectors); |
| 1665 | if (new_disk_conf->meta_dev_idx >= 0) |
| 1666 | drbd_warn(device, "==>> using internal or flexible " |
| 1667 | "meta data may help <<==\n"); |
| 1668 | } |
| 1669 | |
| 1670 | drbd_suspend_io(device); |
| 1671 | /* also wait for the last barrier ack. */ |
| 1672 | /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171 |
| 1673 | * We need a way to either ignore barrier acks for barriers sent before a device |
| 1674 | * was attached, or a way to wait for all pending barrier acks to come in. |
| 1675 | * As barriers are counted per resource, |
| 1676 | * we'd need to suspend io on all devices of a resource. |
| 1677 | */ |
| 1678 | wait_event(device->misc_wait, !atomic_read(&device->ap_pending_cnt) || drbd_suspended(device)); |
| 1679 | /* and for any other previously queued work */ |
| 1680 | drbd_flush_workqueue(&connection->sender_work); |
| 1681 | |
| 1682 | rv = _drbd_request_state(device, NS(disk, D_ATTACHING), CS_VERBOSE); |
| 1683 | retcode = rv; /* FIXME: Type mismatch. */ |
| 1684 | drbd_resume_io(device); |
| 1685 | if (rv < SS_SUCCESS) |
| 1686 | goto fail; |
| 1687 | |
| 1688 | if (!get_ldev_if_state(device, D_ATTACHING)) |
| 1689 | goto force_diskless; |
| 1690 | |
| 1691 | if (!device->bitmap) { |
| 1692 | if (drbd_bm_init(device)) { |
| 1693 | retcode = ERR_NOMEM; |
| 1694 | goto force_diskless_dec; |
| 1695 | } |
| 1696 | } |
| 1697 | |
| 1698 | if (device->state.conn < C_CONNECTED && |
| 1699 | device->state.role == R_PRIMARY && device->ed_uuid && |
| 1700 | (device->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) { |
| 1701 | drbd_err(device, "Can only attach to data with current UUID=%016llX\n", |
| 1702 | (unsigned long long)device->ed_uuid); |
| 1703 | retcode = ERR_DATA_NOT_CURRENT; |
| 1704 | goto force_diskless_dec; |
| 1705 | } |
| 1706 | |
| 1707 | /* Since we are diskless, fix the activity log first... */ |
| 1708 | if (drbd_check_al_size(device, new_disk_conf)) { |
| 1709 | retcode = ERR_NOMEM; |
| 1710 | goto force_diskless_dec; |
| 1711 | } |
| 1712 | |
| 1713 | /* Prevent shrinking of consistent devices ! */ |
| 1714 | if (drbd_md_test_flag(nbc, MDF_CONSISTENT) && |
| 1715 | drbd_new_dev_size(device, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) { |
| 1716 | drbd_warn(device, "refusing to truncate a consistent device\n"); |
| 1717 | retcode = ERR_DISK_TOO_SMALL; |
| 1718 | goto force_diskless_dec; |
| 1719 | } |
| 1720 | |
| 1721 | lock_all_resources(); |
| 1722 | retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after); |
| 1723 | if (retcode != NO_ERROR) { |
| 1724 | unlock_all_resources(); |
| 1725 | goto force_diskless_dec; |
| 1726 | } |
| 1727 | |
| 1728 | /* Reset the "barriers don't work" bits here, then force meta data to |
| 1729 | * be written, to ensure we determine if barriers are supported. */ |
| 1730 | if (new_disk_conf->md_flushes) |
| 1731 | clear_bit(MD_NO_FUA, &device->flags); |
| 1732 | else |
| 1733 | set_bit(MD_NO_FUA, &device->flags); |
| 1734 | |
| 1735 | /* Point of no return reached. |
| 1736 | * Devices and memory are no longer released by error cleanup below. |
| 1737 | * now device takes over responsibility, and the state engine should |
| 1738 | * clean it up somewhere. */ |
| 1739 | D_ASSERT(device, device->ldev == NULL); |
| 1740 | device->ldev = nbc; |
| 1741 | device->resync = resync_lru; |
| 1742 | device->rs_plan_s = new_plan; |
| 1743 | nbc = NULL; |
| 1744 | resync_lru = NULL; |
| 1745 | new_disk_conf = NULL; |
| 1746 | new_plan = NULL; |
| 1747 | |
| 1748 | drbd_resync_after_changed(device); |
| 1749 | drbd_bump_write_ordering(device->resource, device->ldev, WO_BDEV_FLUSH); |
| 1750 | unlock_all_resources(); |
| 1751 | |
| 1752 | if (drbd_md_test_flag(device->ldev, MDF_CRASHED_PRIMARY)) |
| 1753 | set_bit(CRASHED_PRIMARY, &device->flags); |
| 1754 | else |
| 1755 | clear_bit(CRASHED_PRIMARY, &device->flags); |
| 1756 | |
| 1757 | if (drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) && |
| 1758 | !(device->state.role == R_PRIMARY && device->resource->susp_nod)) |
| 1759 | set_bit(CRASHED_PRIMARY, &device->flags); |
| 1760 | |
| 1761 | device->send_cnt = 0; |
| 1762 | device->recv_cnt = 0; |
| 1763 | device->read_cnt = 0; |
| 1764 | device->writ_cnt = 0; |
| 1765 | |
| 1766 | drbd_reconsider_max_bio_size(device, device->ldev); |
| 1767 | |
| 1768 | /* If I am currently not R_PRIMARY, |
| 1769 | * but meta data primary indicator is set, |
| 1770 | * I just now recover from a hard crash, |
| 1771 | * and have been R_PRIMARY before that crash. |
| 1772 | * |
| 1773 | * Now, if I had no connection before that crash |
| 1774 | * (have been degraded R_PRIMARY), chances are that |
| 1775 | * I won't find my peer now either. |
| 1776 | * |
| 1777 | * In that case, and _only_ in that case, |
| 1778 | * we use the degr-wfc-timeout instead of the default, |
| 1779 | * so we can automatically recover from a crash of a |
| 1780 | * degraded but active "cluster" after a certain timeout. |
| 1781 | */ |
| 1782 | clear_bit(USE_DEGR_WFC_T, &device->flags); |
| 1783 | if (device->state.role != R_PRIMARY && |
| 1784 | drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) && |
| 1785 | !drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND)) |
| 1786 | set_bit(USE_DEGR_WFC_T, &device->flags); |
| 1787 | |
| 1788 | dd = drbd_determine_dev_size(device, 0, NULL); |
| 1789 | if (dd <= DS_ERROR) { |
| 1790 | retcode = ERR_NOMEM_BITMAP; |
| 1791 | goto force_diskless_dec; |
| 1792 | } else if (dd == DS_GREW) |
| 1793 | set_bit(RESYNC_AFTER_NEG, &device->flags); |
| 1794 | |
| 1795 | if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) || |
| 1796 | (test_bit(CRASHED_PRIMARY, &device->flags) && |
| 1797 | drbd_md_test_flag(device->ldev, MDF_AL_DISABLED))) { |
| 1798 | drbd_info(device, "Assuming that all blocks are out of sync " |
| 1799 | "(aka FullSync)\n"); |
| 1800 | if (drbd_bitmap_io(device, &drbd_bmio_set_n_write, |
| 1801 | "set_n_write from attaching", BM_LOCKED_MASK)) { |
| 1802 | retcode = ERR_IO_MD_DISK; |
| 1803 | goto force_diskless_dec; |
| 1804 | } |
| 1805 | } else { |
| 1806 | if (drbd_bitmap_io(device, &drbd_bm_read, |
| 1807 | "read from attaching", BM_LOCKED_MASK)) { |
| 1808 | retcode = ERR_IO_MD_DISK; |
| 1809 | goto force_diskless_dec; |
| 1810 | } |
| 1811 | } |
| 1812 | |
| 1813 | if (_drbd_bm_total_weight(device) == drbd_bm_bits(device)) |
| 1814 | drbd_suspend_al(device); /* IO is still suspended here... */ |
| 1815 | |
| 1816 | spin_lock_irq(&device->resource->req_lock); |
| 1817 | os = drbd_read_state(device); |
| 1818 | ns = os; |
| 1819 | /* If MDF_CONSISTENT is not set go into inconsistent state, |
| 1820 | otherwise investigate MDF_WasUpToDate... |
| 1821 | If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state, |
| 1822 | otherwise into D_CONSISTENT state. |
| 1823 | */ |
| 1824 | if (drbd_md_test_flag(device->ldev, MDF_CONSISTENT)) { |
| 1825 | if (drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE)) |
| 1826 | ns.disk = D_CONSISTENT; |
| 1827 | else |
| 1828 | ns.disk = D_OUTDATED; |
| 1829 | } else { |
| 1830 | ns.disk = D_INCONSISTENT; |
| 1831 | } |
| 1832 | |
| 1833 | if (drbd_md_test_flag(device->ldev, MDF_PEER_OUT_DATED)) |
| 1834 | ns.pdsk = D_OUTDATED; |
| 1835 | |
| 1836 | rcu_read_lock(); |
| 1837 | if (ns.disk == D_CONSISTENT && |
| 1838 | (ns.pdsk == D_OUTDATED || rcu_dereference(device->ldev->disk_conf)->fencing == FP_DONT_CARE)) |
| 1839 | ns.disk = D_UP_TO_DATE; |
| 1840 | |
| 1841 | /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND, |
| 1842 | MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before |
| 1843 | this point, because drbd_request_state() modifies these |
| 1844 | flags. */ |
| 1845 | |
| 1846 | if (rcu_dereference(device->ldev->disk_conf)->al_updates) |
| 1847 | device->ldev->md.flags &= ~MDF_AL_DISABLED; |
| 1848 | else |
| 1849 | device->ldev->md.flags |= MDF_AL_DISABLED; |
| 1850 | |
| 1851 | rcu_read_unlock(); |
| 1852 | |
| 1853 | /* In case we are C_CONNECTED postpone any decision on the new disk |
| 1854 | state after the negotiation phase. */ |
| 1855 | if (device->state.conn == C_CONNECTED) { |
| 1856 | device->new_state_tmp.i = ns.i; |
| 1857 | ns.i = os.i; |
| 1858 | ns.disk = D_NEGOTIATING; |
| 1859 | |
| 1860 | /* We expect to receive up-to-date UUIDs soon. |
| 1861 | To avoid a race in receive_state, free p_uuid while |
| 1862 | holding req_lock. I.e. atomic with the state change */ |
| 1863 | kfree(device->p_uuid); |
| 1864 | device->p_uuid = NULL; |
| 1865 | } |
| 1866 | |
| 1867 | rv = _drbd_set_state(device, ns, CS_VERBOSE, NULL); |
| 1868 | spin_unlock_irq(&device->resource->req_lock); |
| 1869 | |
| 1870 | if (rv < SS_SUCCESS) |
| 1871 | goto force_diskless_dec; |
| 1872 | |
| 1873 | mod_timer(&device->request_timer, jiffies + HZ); |
| 1874 | |
| 1875 | if (device->state.role == R_PRIMARY) |
| 1876 | device->ldev->md.uuid[UI_CURRENT] |= (u64)1; |
| 1877 | else |
| 1878 | device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1; |
| 1879 | |
| 1880 | drbd_md_mark_dirty(device); |
| 1881 | drbd_md_sync(device); |
| 1882 | |
| 1883 | kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE); |
| 1884 | put_ldev(device); |
| 1885 | conn_reconfig_done(connection); |
| 1886 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 1887 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 1888 | return 0; |
| 1889 | |
| 1890 | force_diskless_dec: |
| 1891 | put_ldev(device); |
| 1892 | force_diskless: |
| 1893 | drbd_force_state(device, NS(disk, D_DISKLESS)); |
| 1894 | drbd_md_sync(device); |
| 1895 | fail: |
| 1896 | conn_reconfig_done(connection); |
| 1897 | if (nbc) { |
| 1898 | if (nbc->backing_bdev) |
| 1899 | blkdev_put(nbc->backing_bdev, |
| 1900 | FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
| 1901 | if (nbc->md_bdev) |
| 1902 | blkdev_put(nbc->md_bdev, |
| 1903 | FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
| 1904 | kfree(nbc); |
| 1905 | } |
| 1906 | kfree(new_disk_conf); |
| 1907 | lc_destroy(resync_lru); |
| 1908 | kfree(new_plan); |
| 1909 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 1910 | finish: |
| 1911 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 1912 | return 0; |
| 1913 | } |
| 1914 | |
| 1915 | static int adm_detach(struct drbd_device *device, int force) |
| 1916 | { |
| 1917 | enum drbd_state_rv retcode; |
| 1918 | int ret; |
| 1919 | |
| 1920 | if (force) { |
| 1921 | set_bit(FORCE_DETACH, &device->flags); |
| 1922 | drbd_force_state(device, NS(disk, D_FAILED)); |
| 1923 | retcode = SS_SUCCESS; |
| 1924 | goto out; |
| 1925 | } |
| 1926 | |
| 1927 | drbd_suspend_io(device); /* so no-one is stuck in drbd_al_begin_io */ |
| 1928 | drbd_md_get_buffer(device, __func__); /* make sure there is no in-flight meta-data IO */ |
| 1929 | retcode = drbd_request_state(device, NS(disk, D_FAILED)); |
| 1930 | drbd_md_put_buffer(device); |
| 1931 | /* D_FAILED will transition to DISKLESS. */ |
| 1932 | ret = wait_event_interruptible(device->misc_wait, |
| 1933 | device->state.disk != D_FAILED); |
| 1934 | drbd_resume_io(device); |
| 1935 | if ((int)retcode == (int)SS_IS_DISKLESS) |
| 1936 | retcode = SS_NOTHING_TO_DO; |
| 1937 | if (ret) |
| 1938 | retcode = ERR_INTR; |
| 1939 | out: |
| 1940 | return retcode; |
| 1941 | } |
| 1942 | |
| 1943 | /* Detaching the disk is a process in multiple stages. First we need to lock |
| 1944 | * out application IO, in-flight IO, IO stuck in drbd_al_begin_io. |
| 1945 | * Then we transition to D_DISKLESS, and wait for put_ldev() to return all |
| 1946 | * internal references as well. |
| 1947 | * Only then we have finally detached. */ |
| 1948 | int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info) |
| 1949 | { |
| 1950 | struct drbd_config_context adm_ctx; |
| 1951 | enum drbd_ret_code retcode; |
| 1952 | struct detach_parms parms = { }; |
| 1953 | int err; |
| 1954 | |
| 1955 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 1956 | if (!adm_ctx.reply_skb) |
| 1957 | return retcode; |
| 1958 | if (retcode != NO_ERROR) |
| 1959 | goto out; |
| 1960 | |
| 1961 | if (info->attrs[DRBD_NLA_DETACH_PARMS]) { |
| 1962 | err = detach_parms_from_attrs(&parms, info); |
| 1963 | if (err) { |
| 1964 | retcode = ERR_MANDATORY_TAG; |
| 1965 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 1966 | goto out; |
| 1967 | } |
| 1968 | } |
| 1969 | |
| 1970 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 1971 | retcode = adm_detach(adm_ctx.device, parms.force_detach); |
| 1972 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 1973 | out: |
| 1974 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 1975 | return 0; |
| 1976 | } |
| 1977 | |
| 1978 | static bool conn_resync_running(struct drbd_connection *connection) |
| 1979 | { |
| 1980 | struct drbd_peer_device *peer_device; |
| 1981 | bool rv = false; |
| 1982 | int vnr; |
| 1983 | |
| 1984 | rcu_read_lock(); |
| 1985 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
| 1986 | struct drbd_device *device = peer_device->device; |
| 1987 | if (device->state.conn == C_SYNC_SOURCE || |
| 1988 | device->state.conn == C_SYNC_TARGET || |
| 1989 | device->state.conn == C_PAUSED_SYNC_S || |
| 1990 | device->state.conn == C_PAUSED_SYNC_T) { |
| 1991 | rv = true; |
| 1992 | break; |
| 1993 | } |
| 1994 | } |
| 1995 | rcu_read_unlock(); |
| 1996 | |
| 1997 | return rv; |
| 1998 | } |
| 1999 | |
| 2000 | static bool conn_ov_running(struct drbd_connection *connection) |
| 2001 | { |
| 2002 | struct drbd_peer_device *peer_device; |
| 2003 | bool rv = false; |
| 2004 | int vnr; |
| 2005 | |
| 2006 | rcu_read_lock(); |
| 2007 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
| 2008 | struct drbd_device *device = peer_device->device; |
| 2009 | if (device->state.conn == C_VERIFY_S || |
| 2010 | device->state.conn == C_VERIFY_T) { |
| 2011 | rv = true; |
| 2012 | break; |
| 2013 | } |
| 2014 | } |
| 2015 | rcu_read_unlock(); |
| 2016 | |
| 2017 | return rv; |
| 2018 | } |
| 2019 | |
| 2020 | static enum drbd_ret_code |
| 2021 | _check_net_options(struct drbd_connection *connection, struct net_conf *old_net_conf, struct net_conf *new_net_conf) |
| 2022 | { |
| 2023 | struct drbd_peer_device *peer_device; |
| 2024 | int i; |
| 2025 | |
| 2026 | if (old_net_conf && connection->cstate == C_WF_REPORT_PARAMS && connection->agreed_pro_version < 100) { |
| 2027 | if (new_net_conf->wire_protocol != old_net_conf->wire_protocol) |
| 2028 | return ERR_NEED_APV_100; |
| 2029 | |
| 2030 | if (new_net_conf->two_primaries != old_net_conf->two_primaries) |
| 2031 | return ERR_NEED_APV_100; |
| 2032 | |
| 2033 | if (strcmp(new_net_conf->integrity_alg, old_net_conf->integrity_alg)) |
| 2034 | return ERR_NEED_APV_100; |
| 2035 | } |
| 2036 | |
| 2037 | if (!new_net_conf->two_primaries && |
| 2038 | conn_highest_role(connection) == R_PRIMARY && |
| 2039 | conn_highest_peer(connection) == R_PRIMARY) |
| 2040 | return ERR_NEED_ALLOW_TWO_PRI; |
| 2041 | |
| 2042 | if (new_net_conf->two_primaries && |
| 2043 | (new_net_conf->wire_protocol != DRBD_PROT_C)) |
| 2044 | return ERR_NOT_PROTO_C; |
| 2045 | |
| 2046 | idr_for_each_entry(&connection->peer_devices, peer_device, i) { |
| 2047 | struct drbd_device *device = peer_device->device; |
| 2048 | if (get_ldev(device)) { |
| 2049 | enum drbd_fencing_p fp = rcu_dereference(device->ldev->disk_conf)->fencing; |
| 2050 | put_ldev(device); |
| 2051 | if (new_net_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) |
| 2052 | return ERR_STONITH_AND_PROT_A; |
| 2053 | } |
| 2054 | if (device->state.role == R_PRIMARY && new_net_conf->discard_my_data) |
| 2055 | return ERR_DISCARD_IMPOSSIBLE; |
| 2056 | } |
| 2057 | |
| 2058 | if (new_net_conf->on_congestion != OC_BLOCK && new_net_conf->wire_protocol != DRBD_PROT_A) |
| 2059 | return ERR_CONG_NOT_PROTO_A; |
| 2060 | |
| 2061 | return NO_ERROR; |
| 2062 | } |
| 2063 | |
| 2064 | static enum drbd_ret_code |
| 2065 | check_net_options(struct drbd_connection *connection, struct net_conf *new_net_conf) |
| 2066 | { |
| 2067 | static enum drbd_ret_code rv; |
| 2068 | struct drbd_peer_device *peer_device; |
| 2069 | int i; |
| 2070 | |
| 2071 | rcu_read_lock(); |
| 2072 | rv = _check_net_options(connection, rcu_dereference(connection->net_conf), new_net_conf); |
| 2073 | rcu_read_unlock(); |
| 2074 | |
| 2075 | /* connection->peer_devices protected by genl_lock() here */ |
| 2076 | idr_for_each_entry(&connection->peer_devices, peer_device, i) { |
| 2077 | struct drbd_device *device = peer_device->device; |
| 2078 | if (!device->bitmap) { |
| 2079 | if (drbd_bm_init(device)) |
| 2080 | return ERR_NOMEM; |
| 2081 | } |
| 2082 | } |
| 2083 | |
| 2084 | return rv; |
| 2085 | } |
| 2086 | |
| 2087 | struct crypto { |
| 2088 | struct crypto_hash *verify_tfm; |
| 2089 | struct crypto_hash *csums_tfm; |
| 2090 | struct crypto_hash *cram_hmac_tfm; |
| 2091 | struct crypto_hash *integrity_tfm; |
| 2092 | }; |
| 2093 | |
| 2094 | static int |
| 2095 | alloc_hash(struct crypto_hash **tfm, char *tfm_name, int err_alg) |
| 2096 | { |
| 2097 | if (!tfm_name[0]) |
| 2098 | return NO_ERROR; |
| 2099 | |
| 2100 | *tfm = crypto_alloc_hash(tfm_name, 0, CRYPTO_ALG_ASYNC); |
| 2101 | if (IS_ERR(*tfm)) { |
| 2102 | *tfm = NULL; |
| 2103 | return err_alg; |
| 2104 | } |
| 2105 | |
| 2106 | return NO_ERROR; |
| 2107 | } |
| 2108 | |
| 2109 | static enum drbd_ret_code |
| 2110 | alloc_crypto(struct crypto *crypto, struct net_conf *new_net_conf) |
| 2111 | { |
| 2112 | char hmac_name[CRYPTO_MAX_ALG_NAME]; |
| 2113 | enum drbd_ret_code rv; |
| 2114 | |
| 2115 | rv = alloc_hash(&crypto->csums_tfm, new_net_conf->csums_alg, |
| 2116 | ERR_CSUMS_ALG); |
| 2117 | if (rv != NO_ERROR) |
| 2118 | return rv; |
| 2119 | rv = alloc_hash(&crypto->verify_tfm, new_net_conf->verify_alg, |
| 2120 | ERR_VERIFY_ALG); |
| 2121 | if (rv != NO_ERROR) |
| 2122 | return rv; |
| 2123 | rv = alloc_hash(&crypto->integrity_tfm, new_net_conf->integrity_alg, |
| 2124 | ERR_INTEGRITY_ALG); |
| 2125 | if (rv != NO_ERROR) |
| 2126 | return rv; |
| 2127 | if (new_net_conf->cram_hmac_alg[0] != 0) { |
| 2128 | snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)", |
| 2129 | new_net_conf->cram_hmac_alg); |
| 2130 | |
| 2131 | rv = alloc_hash(&crypto->cram_hmac_tfm, hmac_name, |
| 2132 | ERR_AUTH_ALG); |
| 2133 | } |
| 2134 | |
| 2135 | return rv; |
| 2136 | } |
| 2137 | |
| 2138 | static void free_crypto(struct crypto *crypto) |
| 2139 | { |
| 2140 | crypto_free_hash(crypto->cram_hmac_tfm); |
| 2141 | crypto_free_hash(crypto->integrity_tfm); |
| 2142 | crypto_free_hash(crypto->csums_tfm); |
| 2143 | crypto_free_hash(crypto->verify_tfm); |
| 2144 | } |
| 2145 | |
| 2146 | int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info) |
| 2147 | { |
| 2148 | struct drbd_config_context adm_ctx; |
| 2149 | enum drbd_ret_code retcode; |
| 2150 | struct drbd_connection *connection; |
| 2151 | struct net_conf *old_net_conf, *new_net_conf = NULL; |
| 2152 | int err; |
| 2153 | int ovr; /* online verify running */ |
| 2154 | int rsr; /* re-sync running */ |
| 2155 | struct crypto crypto = { }; |
| 2156 | |
| 2157 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION); |
| 2158 | if (!adm_ctx.reply_skb) |
| 2159 | return retcode; |
| 2160 | if (retcode != NO_ERROR) |
| 2161 | goto finish; |
| 2162 | |
| 2163 | connection = adm_ctx.connection; |
| 2164 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2165 | |
| 2166 | new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL); |
| 2167 | if (!new_net_conf) { |
| 2168 | retcode = ERR_NOMEM; |
| 2169 | goto out; |
| 2170 | } |
| 2171 | |
| 2172 | conn_reconfig_start(connection); |
| 2173 | |
| 2174 | mutex_lock(&connection->data.mutex); |
| 2175 | mutex_lock(&connection->resource->conf_update); |
| 2176 | old_net_conf = connection->net_conf; |
| 2177 | |
| 2178 | if (!old_net_conf) { |
| 2179 | drbd_msg_put_info(adm_ctx.reply_skb, "net conf missing, try connect"); |
| 2180 | retcode = ERR_INVALID_REQUEST; |
| 2181 | goto fail; |
| 2182 | } |
| 2183 | |
| 2184 | *new_net_conf = *old_net_conf; |
| 2185 | if (should_set_defaults(info)) |
| 2186 | set_net_conf_defaults(new_net_conf); |
| 2187 | |
| 2188 | err = net_conf_from_attrs_for_change(new_net_conf, info); |
| 2189 | if (err && err != -ENOMSG) { |
| 2190 | retcode = ERR_MANDATORY_TAG; |
| 2191 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 2192 | goto fail; |
| 2193 | } |
| 2194 | |
| 2195 | retcode = check_net_options(connection, new_net_conf); |
| 2196 | if (retcode != NO_ERROR) |
| 2197 | goto fail; |
| 2198 | |
| 2199 | /* re-sync running */ |
| 2200 | rsr = conn_resync_running(connection); |
| 2201 | if (rsr && strcmp(new_net_conf->csums_alg, old_net_conf->csums_alg)) { |
| 2202 | retcode = ERR_CSUMS_RESYNC_RUNNING; |
| 2203 | goto fail; |
| 2204 | } |
| 2205 | |
| 2206 | /* online verify running */ |
| 2207 | ovr = conn_ov_running(connection); |
| 2208 | if (ovr && strcmp(new_net_conf->verify_alg, old_net_conf->verify_alg)) { |
| 2209 | retcode = ERR_VERIFY_RUNNING; |
| 2210 | goto fail; |
| 2211 | } |
| 2212 | |
| 2213 | retcode = alloc_crypto(&crypto, new_net_conf); |
| 2214 | if (retcode != NO_ERROR) |
| 2215 | goto fail; |
| 2216 | |
| 2217 | rcu_assign_pointer(connection->net_conf, new_net_conf); |
| 2218 | |
| 2219 | if (!rsr) { |
| 2220 | crypto_free_hash(connection->csums_tfm); |
| 2221 | connection->csums_tfm = crypto.csums_tfm; |
| 2222 | crypto.csums_tfm = NULL; |
| 2223 | } |
| 2224 | if (!ovr) { |
| 2225 | crypto_free_hash(connection->verify_tfm); |
| 2226 | connection->verify_tfm = crypto.verify_tfm; |
| 2227 | crypto.verify_tfm = NULL; |
| 2228 | } |
| 2229 | |
| 2230 | crypto_free_hash(connection->integrity_tfm); |
| 2231 | connection->integrity_tfm = crypto.integrity_tfm; |
| 2232 | if (connection->cstate >= C_WF_REPORT_PARAMS && connection->agreed_pro_version >= 100) |
| 2233 | /* Do this without trying to take connection->data.mutex again. */ |
| 2234 | __drbd_send_protocol(connection, P_PROTOCOL_UPDATE); |
| 2235 | |
| 2236 | crypto_free_hash(connection->cram_hmac_tfm); |
| 2237 | connection->cram_hmac_tfm = crypto.cram_hmac_tfm; |
| 2238 | |
| 2239 | mutex_unlock(&connection->resource->conf_update); |
| 2240 | mutex_unlock(&connection->data.mutex); |
| 2241 | synchronize_rcu(); |
| 2242 | kfree(old_net_conf); |
| 2243 | |
| 2244 | if (connection->cstate >= C_WF_REPORT_PARAMS) { |
| 2245 | struct drbd_peer_device *peer_device; |
| 2246 | int vnr; |
| 2247 | |
| 2248 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) |
| 2249 | drbd_send_sync_param(peer_device); |
| 2250 | } |
| 2251 | |
| 2252 | goto done; |
| 2253 | |
| 2254 | fail: |
| 2255 | mutex_unlock(&connection->resource->conf_update); |
| 2256 | mutex_unlock(&connection->data.mutex); |
| 2257 | free_crypto(&crypto); |
| 2258 | kfree(new_net_conf); |
| 2259 | done: |
| 2260 | conn_reconfig_done(connection); |
| 2261 | out: |
| 2262 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2263 | finish: |
| 2264 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2265 | return 0; |
| 2266 | } |
| 2267 | |
| 2268 | static void connection_to_info(struct connection_info *info, |
| 2269 | struct drbd_connection *connection) |
| 2270 | { |
| 2271 | info->conn_connection_state = connection->cstate; |
| 2272 | info->conn_role = conn_highest_peer(connection); |
| 2273 | } |
| 2274 | |
| 2275 | static void peer_device_to_info(struct peer_device_info *info, |
| 2276 | struct drbd_peer_device *peer_device) |
| 2277 | { |
| 2278 | struct drbd_device *device = peer_device->device; |
| 2279 | |
| 2280 | info->peer_repl_state = |
| 2281 | max_t(enum drbd_conns, C_WF_REPORT_PARAMS, device->state.conn); |
| 2282 | info->peer_disk_state = device->state.pdsk; |
| 2283 | info->peer_resync_susp_user = device->state.user_isp; |
| 2284 | info->peer_resync_susp_peer = device->state.peer_isp; |
| 2285 | info->peer_resync_susp_dependency = device->state.aftr_isp; |
| 2286 | } |
| 2287 | |
| 2288 | int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info) |
| 2289 | { |
| 2290 | struct connection_info connection_info; |
| 2291 | enum drbd_notification_type flags; |
| 2292 | unsigned int peer_devices = 0; |
| 2293 | struct drbd_config_context adm_ctx; |
| 2294 | struct drbd_peer_device *peer_device; |
| 2295 | struct net_conf *old_net_conf, *new_net_conf = NULL; |
| 2296 | struct crypto crypto = { }; |
| 2297 | struct drbd_resource *resource; |
| 2298 | struct drbd_connection *connection; |
| 2299 | enum drbd_ret_code retcode; |
| 2300 | int i; |
| 2301 | int err; |
| 2302 | |
| 2303 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE); |
| 2304 | |
| 2305 | if (!adm_ctx.reply_skb) |
| 2306 | return retcode; |
| 2307 | if (retcode != NO_ERROR) |
| 2308 | goto out; |
| 2309 | if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) { |
| 2310 | drbd_msg_put_info(adm_ctx.reply_skb, "connection endpoint(s) missing"); |
| 2311 | retcode = ERR_INVALID_REQUEST; |
| 2312 | goto out; |
| 2313 | } |
| 2314 | |
| 2315 | /* No need for _rcu here. All reconfiguration is |
| 2316 | * strictly serialized on genl_lock(). We are protected against |
| 2317 | * concurrent reconfiguration/addition/deletion */ |
| 2318 | for_each_resource(resource, &drbd_resources) { |
| 2319 | for_each_connection(connection, resource) { |
| 2320 | if (nla_len(adm_ctx.my_addr) == connection->my_addr_len && |
| 2321 | !memcmp(nla_data(adm_ctx.my_addr), &connection->my_addr, |
| 2322 | connection->my_addr_len)) { |
| 2323 | retcode = ERR_LOCAL_ADDR; |
| 2324 | goto out; |
| 2325 | } |
| 2326 | |
| 2327 | if (nla_len(adm_ctx.peer_addr) == connection->peer_addr_len && |
| 2328 | !memcmp(nla_data(adm_ctx.peer_addr), &connection->peer_addr, |
| 2329 | connection->peer_addr_len)) { |
| 2330 | retcode = ERR_PEER_ADDR; |
| 2331 | goto out; |
| 2332 | } |
| 2333 | } |
| 2334 | } |
| 2335 | |
| 2336 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2337 | connection = first_connection(adm_ctx.resource); |
| 2338 | conn_reconfig_start(connection); |
| 2339 | |
| 2340 | if (connection->cstate > C_STANDALONE) { |
| 2341 | retcode = ERR_NET_CONFIGURED; |
| 2342 | goto fail; |
| 2343 | } |
| 2344 | |
| 2345 | /* allocation not in the IO path, drbdsetup / netlink process context */ |
| 2346 | new_net_conf = kzalloc(sizeof(*new_net_conf), GFP_KERNEL); |
| 2347 | if (!new_net_conf) { |
| 2348 | retcode = ERR_NOMEM; |
| 2349 | goto fail; |
| 2350 | } |
| 2351 | |
| 2352 | set_net_conf_defaults(new_net_conf); |
| 2353 | |
| 2354 | err = net_conf_from_attrs(new_net_conf, info); |
| 2355 | if (err && err != -ENOMSG) { |
| 2356 | retcode = ERR_MANDATORY_TAG; |
| 2357 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 2358 | goto fail; |
| 2359 | } |
| 2360 | |
| 2361 | retcode = check_net_options(connection, new_net_conf); |
| 2362 | if (retcode != NO_ERROR) |
| 2363 | goto fail; |
| 2364 | |
| 2365 | retcode = alloc_crypto(&crypto, new_net_conf); |
| 2366 | if (retcode != NO_ERROR) |
| 2367 | goto fail; |
| 2368 | |
| 2369 | ((char *)new_net_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0; |
| 2370 | |
| 2371 | drbd_flush_workqueue(&connection->sender_work); |
| 2372 | |
| 2373 | mutex_lock(&adm_ctx.resource->conf_update); |
| 2374 | old_net_conf = connection->net_conf; |
| 2375 | if (old_net_conf) { |
| 2376 | retcode = ERR_NET_CONFIGURED; |
| 2377 | mutex_unlock(&adm_ctx.resource->conf_update); |
| 2378 | goto fail; |
| 2379 | } |
| 2380 | rcu_assign_pointer(connection->net_conf, new_net_conf); |
| 2381 | |
| 2382 | conn_free_crypto(connection); |
| 2383 | connection->cram_hmac_tfm = crypto.cram_hmac_tfm; |
| 2384 | connection->integrity_tfm = crypto.integrity_tfm; |
| 2385 | connection->csums_tfm = crypto.csums_tfm; |
| 2386 | connection->verify_tfm = crypto.verify_tfm; |
| 2387 | |
| 2388 | connection->my_addr_len = nla_len(adm_ctx.my_addr); |
| 2389 | memcpy(&connection->my_addr, nla_data(adm_ctx.my_addr), connection->my_addr_len); |
| 2390 | connection->peer_addr_len = nla_len(adm_ctx.peer_addr); |
| 2391 | memcpy(&connection->peer_addr, nla_data(adm_ctx.peer_addr), connection->peer_addr_len); |
| 2392 | |
| 2393 | idr_for_each_entry(&connection->peer_devices, peer_device, i) { |
| 2394 | peer_devices++; |
| 2395 | } |
| 2396 | |
| 2397 | connection_to_info(&connection_info, connection); |
| 2398 | flags = (peer_devices--) ? NOTIFY_CONTINUES : 0; |
| 2399 | mutex_lock(¬ification_mutex); |
| 2400 | notify_connection_state(NULL, 0, connection, &connection_info, NOTIFY_CREATE | flags); |
| 2401 | idr_for_each_entry(&connection->peer_devices, peer_device, i) { |
| 2402 | struct peer_device_info peer_device_info; |
| 2403 | |
| 2404 | peer_device_to_info(&peer_device_info, peer_device); |
| 2405 | flags = (peer_devices--) ? NOTIFY_CONTINUES : 0; |
| 2406 | notify_peer_device_state(NULL, 0, peer_device, &peer_device_info, NOTIFY_CREATE | flags); |
| 2407 | } |
| 2408 | mutex_unlock(¬ification_mutex); |
| 2409 | mutex_unlock(&adm_ctx.resource->conf_update); |
| 2410 | |
| 2411 | rcu_read_lock(); |
| 2412 | idr_for_each_entry(&connection->peer_devices, peer_device, i) { |
| 2413 | struct drbd_device *device = peer_device->device; |
| 2414 | device->send_cnt = 0; |
| 2415 | device->recv_cnt = 0; |
| 2416 | } |
| 2417 | rcu_read_unlock(); |
| 2418 | |
| 2419 | retcode = conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE); |
| 2420 | |
| 2421 | conn_reconfig_done(connection); |
| 2422 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2423 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2424 | return 0; |
| 2425 | |
| 2426 | fail: |
| 2427 | free_crypto(&crypto); |
| 2428 | kfree(new_net_conf); |
| 2429 | |
| 2430 | conn_reconfig_done(connection); |
| 2431 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2432 | out: |
| 2433 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2434 | return 0; |
| 2435 | } |
| 2436 | |
| 2437 | static enum drbd_state_rv conn_try_disconnect(struct drbd_connection *connection, bool force) |
| 2438 | { |
| 2439 | enum drbd_state_rv rv; |
| 2440 | |
| 2441 | rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), |
| 2442 | force ? CS_HARD : 0); |
| 2443 | |
| 2444 | switch (rv) { |
| 2445 | case SS_NOTHING_TO_DO: |
| 2446 | break; |
| 2447 | case SS_ALREADY_STANDALONE: |
| 2448 | return SS_SUCCESS; |
| 2449 | case SS_PRIMARY_NOP: |
| 2450 | /* Our state checking code wants to see the peer outdated. */ |
| 2451 | rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, pdsk, D_OUTDATED), 0); |
| 2452 | |
| 2453 | if (rv == SS_OUTDATE_WO_CONN) /* lost connection before graceful disconnect succeeded */ |
| 2454 | rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_VERBOSE); |
| 2455 | |
| 2456 | break; |
| 2457 | case SS_CW_FAILED_BY_PEER: |
| 2458 | /* The peer probably wants to see us outdated. */ |
| 2459 | rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, |
| 2460 | disk, D_OUTDATED), 0); |
| 2461 | if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) { |
| 2462 | rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), |
| 2463 | CS_HARD); |
| 2464 | } |
| 2465 | break; |
| 2466 | default:; |
| 2467 | /* no special handling necessary */ |
| 2468 | } |
| 2469 | |
| 2470 | if (rv >= SS_SUCCESS) { |
| 2471 | enum drbd_state_rv rv2; |
| 2472 | /* No one else can reconfigure the network while I am here. |
| 2473 | * The state handling only uses drbd_thread_stop_nowait(), |
| 2474 | * we want to really wait here until the receiver is no more. |
| 2475 | */ |
| 2476 | drbd_thread_stop(&connection->receiver); |
| 2477 | |
| 2478 | /* Race breaker. This additional state change request may be |
| 2479 | * necessary, if this was a forced disconnect during a receiver |
| 2480 | * restart. We may have "killed" the receiver thread just |
| 2481 | * after drbd_receiver() returned. Typically, we should be |
| 2482 | * C_STANDALONE already, now, and this becomes a no-op. |
| 2483 | */ |
| 2484 | rv2 = conn_request_state(connection, NS(conn, C_STANDALONE), |
| 2485 | CS_VERBOSE | CS_HARD); |
| 2486 | if (rv2 < SS_SUCCESS) |
| 2487 | drbd_err(connection, |
| 2488 | "unexpected rv2=%d in conn_try_disconnect()\n", |
| 2489 | rv2); |
| 2490 | /* Unlike in DRBD 9, the state engine has generated |
| 2491 | * NOTIFY_DESTROY events before clearing connection->net_conf. */ |
| 2492 | } |
| 2493 | return rv; |
| 2494 | } |
| 2495 | |
| 2496 | int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info) |
| 2497 | { |
| 2498 | struct drbd_config_context adm_ctx; |
| 2499 | struct disconnect_parms parms; |
| 2500 | struct drbd_connection *connection; |
| 2501 | enum drbd_state_rv rv; |
| 2502 | enum drbd_ret_code retcode; |
| 2503 | int err; |
| 2504 | |
| 2505 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION); |
| 2506 | if (!adm_ctx.reply_skb) |
| 2507 | return retcode; |
| 2508 | if (retcode != NO_ERROR) |
| 2509 | goto fail; |
| 2510 | |
| 2511 | connection = adm_ctx.connection; |
| 2512 | memset(&parms, 0, sizeof(parms)); |
| 2513 | if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) { |
| 2514 | err = disconnect_parms_from_attrs(&parms, info); |
| 2515 | if (err) { |
| 2516 | retcode = ERR_MANDATORY_TAG; |
| 2517 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 2518 | goto fail; |
| 2519 | } |
| 2520 | } |
| 2521 | |
| 2522 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2523 | rv = conn_try_disconnect(connection, parms.force_disconnect); |
| 2524 | if (rv < SS_SUCCESS) |
| 2525 | retcode = rv; /* FIXME: Type mismatch. */ |
| 2526 | else |
| 2527 | retcode = NO_ERROR; |
| 2528 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2529 | fail: |
| 2530 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2531 | return 0; |
| 2532 | } |
| 2533 | |
| 2534 | void resync_after_online_grow(struct drbd_device *device) |
| 2535 | { |
| 2536 | int iass; /* I am sync source */ |
| 2537 | |
| 2538 | drbd_info(device, "Resync of new storage after online grow\n"); |
| 2539 | if (device->state.role != device->state.peer) |
| 2540 | iass = (device->state.role == R_PRIMARY); |
| 2541 | else |
| 2542 | iass = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags); |
| 2543 | |
| 2544 | if (iass) |
| 2545 | drbd_start_resync(device, C_SYNC_SOURCE); |
| 2546 | else |
| 2547 | _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE); |
| 2548 | } |
| 2549 | |
| 2550 | int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info) |
| 2551 | { |
| 2552 | struct drbd_config_context adm_ctx; |
| 2553 | struct disk_conf *old_disk_conf, *new_disk_conf = NULL; |
| 2554 | struct resize_parms rs; |
| 2555 | struct drbd_device *device; |
| 2556 | enum drbd_ret_code retcode; |
| 2557 | enum determine_dev_size dd; |
| 2558 | bool change_al_layout = false; |
| 2559 | enum dds_flags ddsf; |
| 2560 | sector_t u_size; |
| 2561 | int err; |
| 2562 | |
| 2563 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2564 | if (!adm_ctx.reply_skb) |
| 2565 | return retcode; |
| 2566 | if (retcode != NO_ERROR) |
| 2567 | goto finish; |
| 2568 | |
| 2569 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2570 | device = adm_ctx.device; |
| 2571 | if (!get_ldev(device)) { |
| 2572 | retcode = ERR_NO_DISK; |
| 2573 | goto fail; |
| 2574 | } |
| 2575 | |
| 2576 | memset(&rs, 0, sizeof(struct resize_parms)); |
| 2577 | rs.al_stripes = device->ldev->md.al_stripes; |
| 2578 | rs.al_stripe_size = device->ldev->md.al_stripe_size_4k * 4; |
| 2579 | if (info->attrs[DRBD_NLA_RESIZE_PARMS]) { |
| 2580 | err = resize_parms_from_attrs(&rs, info); |
| 2581 | if (err) { |
| 2582 | retcode = ERR_MANDATORY_TAG; |
| 2583 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 2584 | goto fail_ldev; |
| 2585 | } |
| 2586 | } |
| 2587 | |
| 2588 | if (device->state.conn > C_CONNECTED) { |
| 2589 | retcode = ERR_RESIZE_RESYNC; |
| 2590 | goto fail_ldev; |
| 2591 | } |
| 2592 | |
| 2593 | if (device->state.role == R_SECONDARY && |
| 2594 | device->state.peer == R_SECONDARY) { |
| 2595 | retcode = ERR_NO_PRIMARY; |
| 2596 | goto fail_ldev; |
| 2597 | } |
| 2598 | |
| 2599 | if (rs.no_resync && first_peer_device(device)->connection->agreed_pro_version < 93) { |
| 2600 | retcode = ERR_NEED_APV_93; |
| 2601 | goto fail_ldev; |
| 2602 | } |
| 2603 | |
| 2604 | rcu_read_lock(); |
| 2605 | u_size = rcu_dereference(device->ldev->disk_conf)->disk_size; |
| 2606 | rcu_read_unlock(); |
| 2607 | if (u_size != (sector_t)rs.resize_size) { |
| 2608 | new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL); |
| 2609 | if (!new_disk_conf) { |
| 2610 | retcode = ERR_NOMEM; |
| 2611 | goto fail_ldev; |
| 2612 | } |
| 2613 | } |
| 2614 | |
| 2615 | if (device->ldev->md.al_stripes != rs.al_stripes || |
| 2616 | device->ldev->md.al_stripe_size_4k != rs.al_stripe_size / 4) { |
| 2617 | u32 al_size_k = rs.al_stripes * rs.al_stripe_size; |
| 2618 | |
| 2619 | if (al_size_k > (16 * 1024 * 1024)) { |
| 2620 | retcode = ERR_MD_LAYOUT_TOO_BIG; |
| 2621 | goto fail_ldev; |
| 2622 | } |
| 2623 | |
| 2624 | if (al_size_k < MD_32kB_SECT/2) { |
| 2625 | retcode = ERR_MD_LAYOUT_TOO_SMALL; |
| 2626 | goto fail_ldev; |
| 2627 | } |
| 2628 | |
| 2629 | if (device->state.conn != C_CONNECTED && !rs.resize_force) { |
| 2630 | retcode = ERR_MD_LAYOUT_CONNECTED; |
| 2631 | goto fail_ldev; |
| 2632 | } |
| 2633 | |
| 2634 | change_al_layout = true; |
| 2635 | } |
| 2636 | |
| 2637 | if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev)) |
| 2638 | device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev); |
| 2639 | |
| 2640 | if (new_disk_conf) { |
| 2641 | mutex_lock(&device->resource->conf_update); |
| 2642 | old_disk_conf = device->ldev->disk_conf; |
| 2643 | *new_disk_conf = *old_disk_conf; |
| 2644 | new_disk_conf->disk_size = (sector_t)rs.resize_size; |
| 2645 | rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf); |
| 2646 | mutex_unlock(&device->resource->conf_update); |
| 2647 | synchronize_rcu(); |
| 2648 | kfree(old_disk_conf); |
| 2649 | } |
| 2650 | |
| 2651 | ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0); |
| 2652 | dd = drbd_determine_dev_size(device, ddsf, change_al_layout ? &rs : NULL); |
| 2653 | drbd_md_sync(device); |
| 2654 | put_ldev(device); |
| 2655 | if (dd == DS_ERROR) { |
| 2656 | retcode = ERR_NOMEM_BITMAP; |
| 2657 | goto fail; |
| 2658 | } else if (dd == DS_ERROR_SPACE_MD) { |
| 2659 | retcode = ERR_MD_LAYOUT_NO_FIT; |
| 2660 | goto fail; |
| 2661 | } else if (dd == DS_ERROR_SHRINK) { |
| 2662 | retcode = ERR_IMPLICIT_SHRINK; |
| 2663 | goto fail; |
| 2664 | } |
| 2665 | |
| 2666 | if (device->state.conn == C_CONNECTED) { |
| 2667 | if (dd == DS_GREW) |
| 2668 | set_bit(RESIZE_PENDING, &device->flags); |
| 2669 | |
| 2670 | drbd_send_uuids(first_peer_device(device)); |
| 2671 | drbd_send_sizes(first_peer_device(device), 1, ddsf); |
| 2672 | } |
| 2673 | |
| 2674 | fail: |
| 2675 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2676 | finish: |
| 2677 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2678 | return 0; |
| 2679 | |
| 2680 | fail_ldev: |
| 2681 | put_ldev(device); |
| 2682 | goto fail; |
| 2683 | } |
| 2684 | |
| 2685 | int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info) |
| 2686 | { |
| 2687 | struct drbd_config_context adm_ctx; |
| 2688 | enum drbd_ret_code retcode; |
| 2689 | struct res_opts res_opts; |
| 2690 | int err; |
| 2691 | |
| 2692 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE); |
| 2693 | if (!adm_ctx.reply_skb) |
| 2694 | return retcode; |
| 2695 | if (retcode != NO_ERROR) |
| 2696 | goto fail; |
| 2697 | |
| 2698 | res_opts = adm_ctx.resource->res_opts; |
| 2699 | if (should_set_defaults(info)) |
| 2700 | set_res_opts_defaults(&res_opts); |
| 2701 | |
| 2702 | err = res_opts_from_attrs(&res_opts, info); |
| 2703 | if (err && err != -ENOMSG) { |
| 2704 | retcode = ERR_MANDATORY_TAG; |
| 2705 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 2706 | goto fail; |
| 2707 | } |
| 2708 | |
| 2709 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2710 | err = set_resource_options(adm_ctx.resource, &res_opts); |
| 2711 | if (err) { |
| 2712 | retcode = ERR_INVALID_REQUEST; |
| 2713 | if (err == -ENOMEM) |
| 2714 | retcode = ERR_NOMEM; |
| 2715 | } |
| 2716 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2717 | |
| 2718 | fail: |
| 2719 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2720 | return 0; |
| 2721 | } |
| 2722 | |
| 2723 | int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info) |
| 2724 | { |
| 2725 | struct drbd_config_context adm_ctx; |
| 2726 | struct drbd_device *device; |
| 2727 | int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */ |
| 2728 | |
| 2729 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2730 | if (!adm_ctx.reply_skb) |
| 2731 | return retcode; |
| 2732 | if (retcode != NO_ERROR) |
| 2733 | goto out; |
| 2734 | |
| 2735 | device = adm_ctx.device; |
| 2736 | if (!get_ldev(device)) { |
| 2737 | retcode = ERR_NO_DISK; |
| 2738 | goto out; |
| 2739 | } |
| 2740 | |
| 2741 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2742 | |
| 2743 | /* If there is still bitmap IO pending, probably because of a previous |
| 2744 | * resync just being finished, wait for it before requesting a new resync. |
| 2745 | * Also wait for it's after_state_ch(). */ |
| 2746 | drbd_suspend_io(device); |
| 2747 | wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags)); |
| 2748 | drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work); |
| 2749 | |
| 2750 | /* If we happen to be C_STANDALONE R_SECONDARY, just change to |
| 2751 | * D_INCONSISTENT, and set all bits in the bitmap. Otherwise, |
| 2752 | * try to start a resync handshake as sync target for full sync. |
| 2753 | */ |
| 2754 | if (device->state.conn == C_STANDALONE && device->state.role == R_SECONDARY) { |
| 2755 | retcode = drbd_request_state(device, NS(disk, D_INCONSISTENT)); |
| 2756 | if (retcode >= SS_SUCCESS) { |
| 2757 | if (drbd_bitmap_io(device, &drbd_bmio_set_n_write, |
| 2758 | "set_n_write from invalidate", BM_LOCKED_MASK)) |
| 2759 | retcode = ERR_IO_MD_DISK; |
| 2760 | } |
| 2761 | } else |
| 2762 | retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_T)); |
| 2763 | drbd_resume_io(device); |
| 2764 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2765 | put_ldev(device); |
| 2766 | out: |
| 2767 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2768 | return 0; |
| 2769 | } |
| 2770 | |
| 2771 | static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info, |
| 2772 | union drbd_state mask, union drbd_state val) |
| 2773 | { |
| 2774 | struct drbd_config_context adm_ctx; |
| 2775 | enum drbd_ret_code retcode; |
| 2776 | |
| 2777 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2778 | if (!adm_ctx.reply_skb) |
| 2779 | return retcode; |
| 2780 | if (retcode != NO_ERROR) |
| 2781 | goto out; |
| 2782 | |
| 2783 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2784 | retcode = drbd_request_state(adm_ctx.device, mask, val); |
| 2785 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2786 | out: |
| 2787 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2788 | return 0; |
| 2789 | } |
| 2790 | |
| 2791 | static int drbd_bmio_set_susp_al(struct drbd_device *device) __must_hold(local) |
| 2792 | { |
| 2793 | int rv; |
| 2794 | |
| 2795 | rv = drbd_bmio_set_n_write(device); |
| 2796 | drbd_suspend_al(device); |
| 2797 | return rv; |
| 2798 | } |
| 2799 | |
| 2800 | int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info) |
| 2801 | { |
| 2802 | struct drbd_config_context adm_ctx; |
| 2803 | int retcode; /* drbd_ret_code, drbd_state_rv */ |
| 2804 | struct drbd_device *device; |
| 2805 | |
| 2806 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2807 | if (!adm_ctx.reply_skb) |
| 2808 | return retcode; |
| 2809 | if (retcode != NO_ERROR) |
| 2810 | goto out; |
| 2811 | |
| 2812 | device = adm_ctx.device; |
| 2813 | if (!get_ldev(device)) { |
| 2814 | retcode = ERR_NO_DISK; |
| 2815 | goto out; |
| 2816 | } |
| 2817 | |
| 2818 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2819 | |
| 2820 | /* If there is still bitmap IO pending, probably because of a previous |
| 2821 | * resync just being finished, wait for it before requesting a new resync. |
| 2822 | * Also wait for it's after_state_ch(). */ |
| 2823 | drbd_suspend_io(device); |
| 2824 | wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags)); |
| 2825 | drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work); |
| 2826 | |
| 2827 | /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits |
| 2828 | * in the bitmap. Otherwise, try to start a resync handshake |
| 2829 | * as sync source for full sync. |
| 2830 | */ |
| 2831 | if (device->state.conn == C_STANDALONE && device->state.role == R_PRIMARY) { |
| 2832 | /* The peer will get a resync upon connect anyways. Just make that |
| 2833 | into a full resync. */ |
| 2834 | retcode = drbd_request_state(device, NS(pdsk, D_INCONSISTENT)); |
| 2835 | if (retcode >= SS_SUCCESS) { |
| 2836 | if (drbd_bitmap_io(device, &drbd_bmio_set_susp_al, |
| 2837 | "set_n_write from invalidate_peer", |
| 2838 | BM_LOCKED_SET_ALLOWED)) |
| 2839 | retcode = ERR_IO_MD_DISK; |
| 2840 | } |
| 2841 | } else |
| 2842 | retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_S)); |
| 2843 | drbd_resume_io(device); |
| 2844 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2845 | put_ldev(device); |
| 2846 | out: |
| 2847 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2848 | return 0; |
| 2849 | } |
| 2850 | |
| 2851 | int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info) |
| 2852 | { |
| 2853 | struct drbd_config_context adm_ctx; |
| 2854 | enum drbd_ret_code retcode; |
| 2855 | |
| 2856 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2857 | if (!adm_ctx.reply_skb) |
| 2858 | return retcode; |
| 2859 | if (retcode != NO_ERROR) |
| 2860 | goto out; |
| 2861 | |
| 2862 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2863 | if (drbd_request_state(adm_ctx.device, NS(user_isp, 1)) == SS_NOTHING_TO_DO) |
| 2864 | retcode = ERR_PAUSE_IS_SET; |
| 2865 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2866 | out: |
| 2867 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2868 | return 0; |
| 2869 | } |
| 2870 | |
| 2871 | int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info) |
| 2872 | { |
| 2873 | struct drbd_config_context adm_ctx; |
| 2874 | union drbd_dev_state s; |
| 2875 | enum drbd_ret_code retcode; |
| 2876 | |
| 2877 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2878 | if (!adm_ctx.reply_skb) |
| 2879 | return retcode; |
| 2880 | if (retcode != NO_ERROR) |
| 2881 | goto out; |
| 2882 | |
| 2883 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2884 | if (drbd_request_state(adm_ctx.device, NS(user_isp, 0)) == SS_NOTHING_TO_DO) { |
| 2885 | s = adm_ctx.device->state; |
| 2886 | if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) { |
| 2887 | retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP : |
| 2888 | s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR; |
| 2889 | } else { |
| 2890 | retcode = ERR_PAUSE_IS_CLEAR; |
| 2891 | } |
| 2892 | } |
| 2893 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2894 | out: |
| 2895 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2896 | return 0; |
| 2897 | } |
| 2898 | |
| 2899 | int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info) |
| 2900 | { |
| 2901 | return drbd_adm_simple_request_state(skb, info, NS(susp, 1)); |
| 2902 | } |
| 2903 | |
| 2904 | int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info) |
| 2905 | { |
| 2906 | struct drbd_config_context adm_ctx; |
| 2907 | struct drbd_device *device; |
| 2908 | int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */ |
| 2909 | |
| 2910 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 2911 | if (!adm_ctx.reply_skb) |
| 2912 | return retcode; |
| 2913 | if (retcode != NO_ERROR) |
| 2914 | goto out; |
| 2915 | |
| 2916 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 2917 | device = adm_ctx.device; |
| 2918 | if (test_bit(NEW_CUR_UUID, &device->flags)) { |
| 2919 | drbd_uuid_new_current(device); |
| 2920 | clear_bit(NEW_CUR_UUID, &device->flags); |
| 2921 | } |
| 2922 | drbd_suspend_io(device); |
| 2923 | retcode = drbd_request_state(device, NS3(susp, 0, susp_nod, 0, susp_fen, 0)); |
| 2924 | if (retcode == SS_SUCCESS) { |
| 2925 | if (device->state.conn < C_CONNECTED) |
| 2926 | tl_clear(first_peer_device(device)->connection); |
| 2927 | if (device->state.disk == D_DISKLESS || device->state.disk == D_FAILED) |
| 2928 | tl_restart(first_peer_device(device)->connection, FAIL_FROZEN_DISK_IO); |
| 2929 | } |
| 2930 | drbd_resume_io(device); |
| 2931 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 2932 | out: |
| 2933 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 2934 | return 0; |
| 2935 | } |
| 2936 | |
| 2937 | int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info) |
| 2938 | { |
| 2939 | return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED)); |
| 2940 | } |
| 2941 | |
| 2942 | static int nla_put_drbd_cfg_context(struct sk_buff *skb, |
| 2943 | struct drbd_resource *resource, |
| 2944 | struct drbd_connection *connection, |
| 2945 | struct drbd_device *device) |
| 2946 | { |
| 2947 | struct nlattr *nla; |
| 2948 | nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT); |
| 2949 | if (!nla) |
| 2950 | goto nla_put_failure; |
| 2951 | if (device && |
| 2952 | nla_put_u32(skb, T_ctx_volume, device->vnr)) |
| 2953 | goto nla_put_failure; |
| 2954 | if (nla_put_string(skb, T_ctx_resource_name, resource->name)) |
| 2955 | goto nla_put_failure; |
| 2956 | if (connection) { |
| 2957 | if (connection->my_addr_len && |
| 2958 | nla_put(skb, T_ctx_my_addr, connection->my_addr_len, &connection->my_addr)) |
| 2959 | goto nla_put_failure; |
| 2960 | if (connection->peer_addr_len && |
| 2961 | nla_put(skb, T_ctx_peer_addr, connection->peer_addr_len, &connection->peer_addr)) |
| 2962 | goto nla_put_failure; |
| 2963 | } |
| 2964 | nla_nest_end(skb, nla); |
| 2965 | return 0; |
| 2966 | |
| 2967 | nla_put_failure: |
| 2968 | if (nla) |
| 2969 | nla_nest_cancel(skb, nla); |
| 2970 | return -EMSGSIZE; |
| 2971 | } |
| 2972 | |
| 2973 | /* |
| 2974 | * The generic netlink dump callbacks are called outside the genl_lock(), so |
| 2975 | * they cannot use the simple attribute parsing code which uses global |
| 2976 | * attribute tables. |
| 2977 | */ |
| 2978 | static struct nlattr *find_cfg_context_attr(const struct nlmsghdr *nlh, int attr) |
| 2979 | { |
| 2980 | const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ; |
| 2981 | const int maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1; |
| 2982 | struct nlattr *nla; |
| 2983 | |
| 2984 | nla = nla_find(nlmsg_attrdata(nlh, hdrlen), nlmsg_attrlen(nlh, hdrlen), |
| 2985 | DRBD_NLA_CFG_CONTEXT); |
| 2986 | if (!nla) |
| 2987 | return NULL; |
| 2988 | return drbd_nla_find_nested(maxtype, nla, __nla_type(attr)); |
| 2989 | } |
| 2990 | |
| 2991 | static void resource_to_info(struct resource_info *, struct drbd_resource *); |
| 2992 | |
| 2993 | int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb) |
| 2994 | { |
| 2995 | struct drbd_genlmsghdr *dh; |
| 2996 | struct drbd_resource *resource; |
| 2997 | struct resource_info resource_info; |
| 2998 | struct resource_statistics resource_statistics; |
| 2999 | int err; |
| 3000 | |
| 3001 | rcu_read_lock(); |
| 3002 | if (cb->args[0]) { |
| 3003 | for_each_resource_rcu(resource, &drbd_resources) |
| 3004 | if (resource == (struct drbd_resource *)cb->args[0]) |
| 3005 | goto found_resource; |
| 3006 | err = 0; /* resource was probably deleted */ |
| 3007 | goto out; |
| 3008 | } |
| 3009 | resource = list_entry(&drbd_resources, |
| 3010 | struct drbd_resource, resources); |
| 3011 | |
| 3012 | found_resource: |
| 3013 | list_for_each_entry_continue_rcu(resource, &drbd_resources, resources) { |
| 3014 | goto put_result; |
| 3015 | } |
| 3016 | err = 0; |
| 3017 | goto out; |
| 3018 | |
| 3019 | put_result: |
| 3020 | dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, |
| 3021 | cb->nlh->nlmsg_seq, &drbd_genl_family, |
| 3022 | NLM_F_MULTI, DRBD_ADM_GET_RESOURCES); |
| 3023 | err = -ENOMEM; |
| 3024 | if (!dh) |
| 3025 | goto out; |
| 3026 | dh->minor = -1U; |
| 3027 | dh->ret_code = NO_ERROR; |
| 3028 | err = nla_put_drbd_cfg_context(skb, resource, NULL, NULL); |
| 3029 | if (err) |
| 3030 | goto out; |
| 3031 | err = res_opts_to_skb(skb, &resource->res_opts, !capable(CAP_SYS_ADMIN)); |
| 3032 | if (err) |
| 3033 | goto out; |
| 3034 | resource_to_info(&resource_info, resource); |
| 3035 | err = resource_info_to_skb(skb, &resource_info, !capable(CAP_SYS_ADMIN)); |
| 3036 | if (err) |
| 3037 | goto out; |
| 3038 | resource_statistics.res_stat_write_ordering = resource->write_ordering; |
| 3039 | err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN)); |
| 3040 | if (err) |
| 3041 | goto out; |
| 3042 | cb->args[0] = (long)resource; |
| 3043 | genlmsg_end(skb, dh); |
| 3044 | err = 0; |
| 3045 | |
| 3046 | out: |
| 3047 | rcu_read_unlock(); |
| 3048 | if (err) |
| 3049 | return err; |
| 3050 | return skb->len; |
| 3051 | } |
| 3052 | |
| 3053 | static void device_to_statistics(struct device_statistics *s, |
| 3054 | struct drbd_device *device) |
| 3055 | { |
| 3056 | memset(s, 0, sizeof(*s)); |
| 3057 | s->dev_upper_blocked = !may_inc_ap_bio(device); |
| 3058 | if (get_ldev(device)) { |
| 3059 | struct drbd_md *md = &device->ldev->md; |
| 3060 | u64 *history_uuids = (u64 *)s->history_uuids; |
| 3061 | struct request_queue *q; |
| 3062 | int n; |
| 3063 | |
| 3064 | spin_lock_irq(&md->uuid_lock); |
| 3065 | s->dev_current_uuid = md->uuid[UI_CURRENT]; |
| 3066 | BUILD_BUG_ON(sizeof(s->history_uuids) < UI_HISTORY_END - UI_HISTORY_START + 1); |
| 3067 | for (n = 0; n < UI_HISTORY_END - UI_HISTORY_START + 1; n++) |
| 3068 | history_uuids[n] = md->uuid[UI_HISTORY_START + n]; |
| 3069 | for (; n < HISTORY_UUIDS; n++) |
| 3070 | history_uuids[n] = 0; |
| 3071 | s->history_uuids_len = HISTORY_UUIDS; |
| 3072 | spin_unlock_irq(&md->uuid_lock); |
| 3073 | |
| 3074 | s->dev_disk_flags = md->flags; |
| 3075 | q = bdev_get_queue(device->ldev->backing_bdev); |
| 3076 | s->dev_lower_blocked = |
| 3077 | bdi_congested(&q->backing_dev_info, |
| 3078 | (1 << WB_async_congested) | |
| 3079 | (1 << WB_sync_congested)); |
| 3080 | put_ldev(device); |
| 3081 | } |
| 3082 | s->dev_size = drbd_get_capacity(device->this_bdev); |
| 3083 | s->dev_read = device->read_cnt; |
| 3084 | s->dev_write = device->writ_cnt; |
| 3085 | s->dev_al_writes = device->al_writ_cnt; |
| 3086 | s->dev_bm_writes = device->bm_writ_cnt; |
| 3087 | s->dev_upper_pending = atomic_read(&device->ap_bio_cnt); |
| 3088 | s->dev_lower_pending = atomic_read(&device->local_cnt); |
| 3089 | s->dev_al_suspended = test_bit(AL_SUSPENDED, &device->flags); |
| 3090 | s->dev_exposed_data_uuid = device->ed_uuid; |
| 3091 | } |
| 3092 | |
| 3093 | static int put_resource_in_arg0(struct netlink_callback *cb, int holder_nr) |
| 3094 | { |
| 3095 | if (cb->args[0]) { |
| 3096 | struct drbd_resource *resource = |
| 3097 | (struct drbd_resource *)cb->args[0]; |
| 3098 | kref_put(&resource->kref, drbd_destroy_resource); |
| 3099 | } |
| 3100 | |
| 3101 | return 0; |
| 3102 | } |
| 3103 | |
| 3104 | int drbd_adm_dump_devices_done(struct netlink_callback *cb) { |
| 3105 | return put_resource_in_arg0(cb, 7); |
| 3106 | } |
| 3107 | |
| 3108 | static void device_to_info(struct device_info *, struct drbd_device *); |
| 3109 | |
| 3110 | int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb) |
| 3111 | { |
| 3112 | struct nlattr *resource_filter; |
| 3113 | struct drbd_resource *resource; |
| 3114 | struct drbd_device *uninitialized_var(device); |
| 3115 | int minor, err, retcode; |
| 3116 | struct drbd_genlmsghdr *dh; |
| 3117 | struct device_info device_info; |
| 3118 | struct device_statistics device_statistics; |
| 3119 | struct idr *idr_to_search; |
| 3120 | |
| 3121 | resource = (struct drbd_resource *)cb->args[0]; |
| 3122 | if (!cb->args[0] && !cb->args[1]) { |
| 3123 | resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name); |
| 3124 | if (resource_filter) { |
| 3125 | retcode = ERR_RES_NOT_KNOWN; |
| 3126 | resource = drbd_find_resource(nla_data(resource_filter)); |
| 3127 | if (!resource) |
| 3128 | goto put_result; |
| 3129 | cb->args[0] = (long)resource; |
| 3130 | } |
| 3131 | } |
| 3132 | |
| 3133 | rcu_read_lock(); |
| 3134 | minor = cb->args[1]; |
| 3135 | idr_to_search = resource ? &resource->devices : &drbd_devices; |
| 3136 | device = idr_get_next(idr_to_search, &minor); |
| 3137 | if (!device) { |
| 3138 | err = 0; |
| 3139 | goto out; |
| 3140 | } |
| 3141 | idr_for_each_entry_continue(idr_to_search, device, minor) { |
| 3142 | retcode = NO_ERROR; |
| 3143 | goto put_result; /* only one iteration */ |
| 3144 | } |
| 3145 | err = 0; |
| 3146 | goto out; /* no more devices */ |
| 3147 | |
| 3148 | put_result: |
| 3149 | dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, |
| 3150 | cb->nlh->nlmsg_seq, &drbd_genl_family, |
| 3151 | NLM_F_MULTI, DRBD_ADM_GET_DEVICES); |
| 3152 | err = -ENOMEM; |
| 3153 | if (!dh) |
| 3154 | goto out; |
| 3155 | dh->ret_code = retcode; |
| 3156 | dh->minor = -1U; |
| 3157 | if (retcode == NO_ERROR) { |
| 3158 | dh->minor = device->minor; |
| 3159 | err = nla_put_drbd_cfg_context(skb, device->resource, NULL, device); |
| 3160 | if (err) |
| 3161 | goto out; |
| 3162 | if (get_ldev(device)) { |
| 3163 | struct disk_conf *disk_conf = |
| 3164 | rcu_dereference(device->ldev->disk_conf); |
| 3165 | |
| 3166 | err = disk_conf_to_skb(skb, disk_conf, !capable(CAP_SYS_ADMIN)); |
| 3167 | put_ldev(device); |
| 3168 | if (err) |
| 3169 | goto out; |
| 3170 | } |
| 3171 | device_to_info(&device_info, device); |
| 3172 | err = device_info_to_skb(skb, &device_info, !capable(CAP_SYS_ADMIN)); |
| 3173 | if (err) |
| 3174 | goto out; |
| 3175 | |
| 3176 | device_to_statistics(&device_statistics, device); |
| 3177 | err = device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN)); |
| 3178 | if (err) |
| 3179 | goto out; |
| 3180 | cb->args[1] = minor + 1; |
| 3181 | } |
| 3182 | genlmsg_end(skb, dh); |
| 3183 | err = 0; |
| 3184 | |
| 3185 | out: |
| 3186 | rcu_read_unlock(); |
| 3187 | if (err) |
| 3188 | return err; |
| 3189 | return skb->len; |
| 3190 | } |
| 3191 | |
| 3192 | int drbd_adm_dump_connections_done(struct netlink_callback *cb) |
| 3193 | { |
| 3194 | return put_resource_in_arg0(cb, 6); |
| 3195 | } |
| 3196 | |
| 3197 | enum { SINGLE_RESOURCE, ITERATE_RESOURCES }; |
| 3198 | |
| 3199 | int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb) |
| 3200 | { |
| 3201 | struct nlattr *resource_filter; |
| 3202 | struct drbd_resource *resource = NULL, *next_resource; |
| 3203 | struct drbd_connection *uninitialized_var(connection); |
| 3204 | int err = 0, retcode; |
| 3205 | struct drbd_genlmsghdr *dh; |
| 3206 | struct connection_info connection_info; |
| 3207 | struct connection_statistics connection_statistics; |
| 3208 | |
| 3209 | rcu_read_lock(); |
| 3210 | resource = (struct drbd_resource *)cb->args[0]; |
| 3211 | if (!cb->args[0]) { |
| 3212 | resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name); |
| 3213 | if (resource_filter) { |
| 3214 | retcode = ERR_RES_NOT_KNOWN; |
| 3215 | resource = drbd_find_resource(nla_data(resource_filter)); |
| 3216 | if (!resource) |
| 3217 | goto put_result; |
| 3218 | cb->args[0] = (long)resource; |
| 3219 | cb->args[1] = SINGLE_RESOURCE; |
| 3220 | } |
| 3221 | } |
| 3222 | if (!resource) { |
| 3223 | if (list_empty(&drbd_resources)) |
| 3224 | goto out; |
| 3225 | resource = list_first_entry(&drbd_resources, struct drbd_resource, resources); |
| 3226 | kref_get(&resource->kref); |
| 3227 | cb->args[0] = (long)resource; |
| 3228 | cb->args[1] = ITERATE_RESOURCES; |
| 3229 | } |
| 3230 | |
| 3231 | next_resource: |
| 3232 | rcu_read_unlock(); |
| 3233 | mutex_lock(&resource->conf_update); |
| 3234 | rcu_read_lock(); |
| 3235 | if (cb->args[2]) { |
| 3236 | for_each_connection_rcu(connection, resource) |
| 3237 | if (connection == (struct drbd_connection *)cb->args[2]) |
| 3238 | goto found_connection; |
| 3239 | /* connection was probably deleted */ |
| 3240 | goto no_more_connections; |
| 3241 | } |
| 3242 | connection = list_entry(&resource->connections, struct drbd_connection, connections); |
| 3243 | |
| 3244 | found_connection: |
| 3245 | list_for_each_entry_continue_rcu(connection, &resource->connections, connections) { |
| 3246 | if (!has_net_conf(connection)) |
| 3247 | continue; |
| 3248 | retcode = NO_ERROR; |
| 3249 | goto put_result; /* only one iteration */ |
| 3250 | } |
| 3251 | |
| 3252 | no_more_connections: |
| 3253 | if (cb->args[1] == ITERATE_RESOURCES) { |
| 3254 | for_each_resource_rcu(next_resource, &drbd_resources) { |
| 3255 | if (next_resource == resource) |
| 3256 | goto found_resource; |
| 3257 | } |
| 3258 | /* resource was probably deleted */ |
| 3259 | } |
| 3260 | goto out; |
| 3261 | |
| 3262 | found_resource: |
| 3263 | list_for_each_entry_continue_rcu(next_resource, &drbd_resources, resources) { |
| 3264 | mutex_unlock(&resource->conf_update); |
| 3265 | kref_put(&resource->kref, drbd_destroy_resource); |
| 3266 | resource = next_resource; |
| 3267 | kref_get(&resource->kref); |
| 3268 | cb->args[0] = (long)resource; |
| 3269 | cb->args[2] = 0; |
| 3270 | goto next_resource; |
| 3271 | } |
| 3272 | goto out; /* no more resources */ |
| 3273 | |
| 3274 | put_result: |
| 3275 | dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, |
| 3276 | cb->nlh->nlmsg_seq, &drbd_genl_family, |
| 3277 | NLM_F_MULTI, DRBD_ADM_GET_CONNECTIONS); |
| 3278 | err = -ENOMEM; |
| 3279 | if (!dh) |
| 3280 | goto out; |
| 3281 | dh->ret_code = retcode; |
| 3282 | dh->minor = -1U; |
| 3283 | if (retcode == NO_ERROR) { |
| 3284 | struct net_conf *net_conf; |
| 3285 | |
| 3286 | err = nla_put_drbd_cfg_context(skb, resource, connection, NULL); |
| 3287 | if (err) |
| 3288 | goto out; |
| 3289 | net_conf = rcu_dereference(connection->net_conf); |
| 3290 | if (net_conf) { |
| 3291 | err = net_conf_to_skb(skb, net_conf, !capable(CAP_SYS_ADMIN)); |
| 3292 | if (err) |
| 3293 | goto out; |
| 3294 | } |
| 3295 | connection_to_info(&connection_info, connection); |
| 3296 | err = connection_info_to_skb(skb, &connection_info, !capable(CAP_SYS_ADMIN)); |
| 3297 | if (err) |
| 3298 | goto out; |
| 3299 | connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags); |
| 3300 | err = connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN)); |
| 3301 | if (err) |
| 3302 | goto out; |
| 3303 | cb->args[2] = (long)connection; |
| 3304 | } |
| 3305 | genlmsg_end(skb, dh); |
| 3306 | err = 0; |
| 3307 | |
| 3308 | out: |
| 3309 | rcu_read_unlock(); |
| 3310 | if (resource) |
| 3311 | mutex_unlock(&resource->conf_update); |
| 3312 | if (err) |
| 3313 | return err; |
| 3314 | return skb->len; |
| 3315 | } |
| 3316 | |
| 3317 | enum mdf_peer_flag { |
| 3318 | MDF_PEER_CONNECTED = 1 << 0, |
| 3319 | MDF_PEER_OUTDATED = 1 << 1, |
| 3320 | MDF_PEER_FENCING = 1 << 2, |
| 3321 | MDF_PEER_FULL_SYNC = 1 << 3, |
| 3322 | }; |
| 3323 | |
| 3324 | static void peer_device_to_statistics(struct peer_device_statistics *s, |
| 3325 | struct drbd_peer_device *peer_device) |
| 3326 | { |
| 3327 | struct drbd_device *device = peer_device->device; |
| 3328 | |
| 3329 | memset(s, 0, sizeof(*s)); |
| 3330 | s->peer_dev_received = device->recv_cnt; |
| 3331 | s->peer_dev_sent = device->send_cnt; |
| 3332 | s->peer_dev_pending = atomic_read(&device->ap_pending_cnt) + |
| 3333 | atomic_read(&device->rs_pending_cnt); |
| 3334 | s->peer_dev_unacked = atomic_read(&device->unacked_cnt); |
| 3335 | s->peer_dev_out_of_sync = drbd_bm_total_weight(device) << (BM_BLOCK_SHIFT - 9); |
| 3336 | s->peer_dev_resync_failed = device->rs_failed << (BM_BLOCK_SHIFT - 9); |
| 3337 | if (get_ldev(device)) { |
| 3338 | struct drbd_md *md = &device->ldev->md; |
| 3339 | |
| 3340 | spin_lock_irq(&md->uuid_lock); |
| 3341 | s->peer_dev_bitmap_uuid = md->uuid[UI_BITMAP]; |
| 3342 | spin_unlock_irq(&md->uuid_lock); |
| 3343 | s->peer_dev_flags = |
| 3344 | (drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND) ? |
| 3345 | MDF_PEER_CONNECTED : 0) + |
| 3346 | (drbd_md_test_flag(device->ldev, MDF_CONSISTENT) && |
| 3347 | !drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE) ? |
| 3348 | MDF_PEER_OUTDATED : 0) + |
| 3349 | /* FIXME: MDF_PEER_FENCING? */ |
| 3350 | (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ? |
| 3351 | MDF_PEER_FULL_SYNC : 0); |
| 3352 | put_ldev(device); |
| 3353 | } |
| 3354 | } |
| 3355 | |
| 3356 | int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb) |
| 3357 | { |
| 3358 | return put_resource_in_arg0(cb, 9); |
| 3359 | } |
| 3360 | |
| 3361 | int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb) |
| 3362 | { |
| 3363 | struct nlattr *resource_filter; |
| 3364 | struct drbd_resource *resource; |
| 3365 | struct drbd_device *uninitialized_var(device); |
| 3366 | struct drbd_peer_device *peer_device = NULL; |
| 3367 | int minor, err, retcode; |
| 3368 | struct drbd_genlmsghdr *dh; |
| 3369 | struct idr *idr_to_search; |
| 3370 | |
| 3371 | resource = (struct drbd_resource *)cb->args[0]; |
| 3372 | if (!cb->args[0] && !cb->args[1]) { |
| 3373 | resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name); |
| 3374 | if (resource_filter) { |
| 3375 | retcode = ERR_RES_NOT_KNOWN; |
| 3376 | resource = drbd_find_resource(nla_data(resource_filter)); |
| 3377 | if (!resource) |
| 3378 | goto put_result; |
| 3379 | } |
| 3380 | cb->args[0] = (long)resource; |
| 3381 | } |
| 3382 | |
| 3383 | rcu_read_lock(); |
| 3384 | minor = cb->args[1]; |
| 3385 | idr_to_search = resource ? &resource->devices : &drbd_devices; |
| 3386 | device = idr_find(idr_to_search, minor); |
| 3387 | if (!device) { |
| 3388 | next_device: |
| 3389 | minor++; |
| 3390 | cb->args[2] = 0; |
| 3391 | device = idr_get_next(idr_to_search, &minor); |
| 3392 | if (!device) { |
| 3393 | err = 0; |
| 3394 | goto out; |
| 3395 | } |
| 3396 | } |
| 3397 | if (cb->args[2]) { |
| 3398 | for_each_peer_device(peer_device, device) |
| 3399 | if (peer_device == (struct drbd_peer_device *)cb->args[2]) |
| 3400 | goto found_peer_device; |
| 3401 | /* peer device was probably deleted */ |
| 3402 | goto next_device; |
| 3403 | } |
| 3404 | /* Make peer_device point to the list head (not the first entry). */ |
| 3405 | peer_device = list_entry(&device->peer_devices, struct drbd_peer_device, peer_devices); |
| 3406 | |
| 3407 | found_peer_device: |
| 3408 | list_for_each_entry_continue_rcu(peer_device, &device->peer_devices, peer_devices) { |
| 3409 | if (!has_net_conf(peer_device->connection)) |
| 3410 | continue; |
| 3411 | retcode = NO_ERROR; |
| 3412 | goto put_result; /* only one iteration */ |
| 3413 | } |
| 3414 | goto next_device; |
| 3415 | |
| 3416 | put_result: |
| 3417 | dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, |
| 3418 | cb->nlh->nlmsg_seq, &drbd_genl_family, |
| 3419 | NLM_F_MULTI, DRBD_ADM_GET_PEER_DEVICES); |
| 3420 | err = -ENOMEM; |
| 3421 | if (!dh) |
| 3422 | goto out; |
| 3423 | dh->ret_code = retcode; |
| 3424 | dh->minor = -1U; |
| 3425 | if (retcode == NO_ERROR) { |
| 3426 | struct peer_device_info peer_device_info; |
| 3427 | struct peer_device_statistics peer_device_statistics; |
| 3428 | |
| 3429 | dh->minor = minor; |
| 3430 | err = nla_put_drbd_cfg_context(skb, device->resource, peer_device->connection, device); |
| 3431 | if (err) |
| 3432 | goto out; |
| 3433 | peer_device_to_info(&peer_device_info, peer_device); |
| 3434 | err = peer_device_info_to_skb(skb, &peer_device_info, !capable(CAP_SYS_ADMIN)); |
| 3435 | if (err) |
| 3436 | goto out; |
| 3437 | peer_device_to_statistics(&peer_device_statistics, peer_device); |
| 3438 | err = peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN)); |
| 3439 | if (err) |
| 3440 | goto out; |
| 3441 | cb->args[1] = minor; |
| 3442 | cb->args[2] = (long)peer_device; |
| 3443 | } |
| 3444 | genlmsg_end(skb, dh); |
| 3445 | err = 0; |
| 3446 | |
| 3447 | out: |
| 3448 | rcu_read_unlock(); |
| 3449 | if (err) |
| 3450 | return err; |
| 3451 | return skb->len; |
| 3452 | } |
| 3453 | /* |
| 3454 | * Return the connection of @resource if @resource has exactly one connection. |
| 3455 | */ |
| 3456 | static struct drbd_connection *the_only_connection(struct drbd_resource *resource) |
| 3457 | { |
| 3458 | struct list_head *connections = &resource->connections; |
| 3459 | |
| 3460 | if (list_empty(connections) || connections->next->next != connections) |
| 3461 | return NULL; |
| 3462 | return list_first_entry(&resource->connections, struct drbd_connection, connections); |
| 3463 | } |
| 3464 | |
| 3465 | static int nla_put_status_info(struct sk_buff *skb, struct drbd_device *device, |
| 3466 | const struct sib_info *sib) |
| 3467 | { |
| 3468 | struct drbd_resource *resource = device->resource; |
| 3469 | struct state_info *si = NULL; /* for sizeof(si->member); */ |
| 3470 | struct nlattr *nla; |
| 3471 | int got_ldev; |
| 3472 | int err = 0; |
| 3473 | int exclude_sensitive; |
| 3474 | |
| 3475 | /* If sib != NULL, this is drbd_bcast_event, which anyone can listen |
| 3476 | * to. So we better exclude_sensitive information. |
| 3477 | * |
| 3478 | * If sib == NULL, this is drbd_adm_get_status, executed synchronously |
| 3479 | * in the context of the requesting user process. Exclude sensitive |
| 3480 | * information, unless current has superuser. |
| 3481 | * |
| 3482 | * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and |
| 3483 | * relies on the current implementation of netlink_dump(), which |
| 3484 | * executes the dump callback successively from netlink_recvmsg(), |
| 3485 | * always in the context of the receiving process */ |
| 3486 | exclude_sensitive = sib || !capable(CAP_SYS_ADMIN); |
| 3487 | |
| 3488 | got_ldev = get_ldev(device); |
| 3489 | |
| 3490 | /* We need to add connection name and volume number information still. |
| 3491 | * Minor number is in drbd_genlmsghdr. */ |
| 3492 | if (nla_put_drbd_cfg_context(skb, resource, the_only_connection(resource), device)) |
| 3493 | goto nla_put_failure; |
| 3494 | |
| 3495 | if (res_opts_to_skb(skb, &device->resource->res_opts, exclude_sensitive)) |
| 3496 | goto nla_put_failure; |
| 3497 | |
| 3498 | rcu_read_lock(); |
| 3499 | if (got_ldev) { |
| 3500 | struct disk_conf *disk_conf; |
| 3501 | |
| 3502 | disk_conf = rcu_dereference(device->ldev->disk_conf); |
| 3503 | err = disk_conf_to_skb(skb, disk_conf, exclude_sensitive); |
| 3504 | } |
| 3505 | if (!err) { |
| 3506 | struct net_conf *nc; |
| 3507 | |
| 3508 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
| 3509 | if (nc) |
| 3510 | err = net_conf_to_skb(skb, nc, exclude_sensitive); |
| 3511 | } |
| 3512 | rcu_read_unlock(); |
| 3513 | if (err) |
| 3514 | goto nla_put_failure; |
| 3515 | |
| 3516 | nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO); |
| 3517 | if (!nla) |
| 3518 | goto nla_put_failure; |
| 3519 | if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) || |
| 3520 | nla_put_u32(skb, T_current_state, device->state.i) || |
| 3521 | nla_put_u64(skb, T_ed_uuid, device->ed_uuid) || |
| 3522 | nla_put_u64(skb, T_capacity, drbd_get_capacity(device->this_bdev)) || |
| 3523 | nla_put_u64(skb, T_send_cnt, device->send_cnt) || |
| 3524 | nla_put_u64(skb, T_recv_cnt, device->recv_cnt) || |
| 3525 | nla_put_u64(skb, T_read_cnt, device->read_cnt) || |
| 3526 | nla_put_u64(skb, T_writ_cnt, device->writ_cnt) || |
| 3527 | nla_put_u64(skb, T_al_writ_cnt, device->al_writ_cnt) || |
| 3528 | nla_put_u64(skb, T_bm_writ_cnt, device->bm_writ_cnt) || |
| 3529 | nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&device->ap_bio_cnt)) || |
| 3530 | nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&device->ap_pending_cnt)) || |
| 3531 | nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&device->rs_pending_cnt))) |
| 3532 | goto nla_put_failure; |
| 3533 | |
| 3534 | if (got_ldev) { |
| 3535 | int err; |
| 3536 | |
| 3537 | spin_lock_irq(&device->ldev->md.uuid_lock); |
| 3538 | err = nla_put(skb, T_uuids, sizeof(si->uuids), device->ldev->md.uuid); |
| 3539 | spin_unlock_irq(&device->ldev->md.uuid_lock); |
| 3540 | |
| 3541 | if (err) |
| 3542 | goto nla_put_failure; |
| 3543 | |
| 3544 | if (nla_put_u32(skb, T_disk_flags, device->ldev->md.flags) || |
| 3545 | nla_put_u64(skb, T_bits_total, drbd_bm_bits(device)) || |
| 3546 | nla_put_u64(skb, T_bits_oos, drbd_bm_total_weight(device))) |
| 3547 | goto nla_put_failure; |
| 3548 | if (C_SYNC_SOURCE <= device->state.conn && |
| 3549 | C_PAUSED_SYNC_T >= device->state.conn) { |
| 3550 | if (nla_put_u64(skb, T_bits_rs_total, device->rs_total) || |
| 3551 | nla_put_u64(skb, T_bits_rs_failed, device->rs_failed)) |
| 3552 | goto nla_put_failure; |
| 3553 | } |
| 3554 | } |
| 3555 | |
| 3556 | if (sib) { |
| 3557 | switch(sib->sib_reason) { |
| 3558 | case SIB_SYNC_PROGRESS: |
| 3559 | case SIB_GET_STATUS_REPLY: |
| 3560 | break; |
| 3561 | case SIB_STATE_CHANGE: |
| 3562 | if (nla_put_u32(skb, T_prev_state, sib->os.i) || |
| 3563 | nla_put_u32(skb, T_new_state, sib->ns.i)) |
| 3564 | goto nla_put_failure; |
| 3565 | break; |
| 3566 | case SIB_HELPER_POST: |
| 3567 | if (nla_put_u32(skb, T_helper_exit_code, |
| 3568 | sib->helper_exit_code)) |
| 3569 | goto nla_put_failure; |
| 3570 | /* fall through */ |
| 3571 | case SIB_HELPER_PRE: |
| 3572 | if (nla_put_string(skb, T_helper, sib->helper_name)) |
| 3573 | goto nla_put_failure; |
| 3574 | break; |
| 3575 | } |
| 3576 | } |
| 3577 | nla_nest_end(skb, nla); |
| 3578 | |
| 3579 | if (0) |
| 3580 | nla_put_failure: |
| 3581 | err = -EMSGSIZE; |
| 3582 | if (got_ldev) |
| 3583 | put_ldev(device); |
| 3584 | return err; |
| 3585 | } |
| 3586 | |
| 3587 | int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info) |
| 3588 | { |
| 3589 | struct drbd_config_context adm_ctx; |
| 3590 | enum drbd_ret_code retcode; |
| 3591 | int err; |
| 3592 | |
| 3593 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 3594 | if (!adm_ctx.reply_skb) |
| 3595 | return retcode; |
| 3596 | if (retcode != NO_ERROR) |
| 3597 | goto out; |
| 3598 | |
| 3599 | err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.device, NULL); |
| 3600 | if (err) { |
| 3601 | nlmsg_free(adm_ctx.reply_skb); |
| 3602 | return err; |
| 3603 | } |
| 3604 | out: |
| 3605 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 3606 | return 0; |
| 3607 | } |
| 3608 | |
| 3609 | static int get_one_status(struct sk_buff *skb, struct netlink_callback *cb) |
| 3610 | { |
| 3611 | struct drbd_device *device; |
| 3612 | struct drbd_genlmsghdr *dh; |
| 3613 | struct drbd_resource *pos = (struct drbd_resource *)cb->args[0]; |
| 3614 | struct drbd_resource *resource = NULL; |
| 3615 | struct drbd_resource *tmp; |
| 3616 | unsigned volume = cb->args[1]; |
| 3617 | |
| 3618 | /* Open coded, deferred, iteration: |
| 3619 | * for_each_resource_safe(resource, tmp, &drbd_resources) { |
| 3620 | * connection = "first connection of resource or undefined"; |
| 3621 | * idr_for_each_entry(&resource->devices, device, i) { |
| 3622 | * ... |
| 3623 | * } |
| 3624 | * } |
| 3625 | * where resource is cb->args[0]; |
| 3626 | * and i is cb->args[1]; |
| 3627 | * |
| 3628 | * cb->args[2] indicates if we shall loop over all resources, |
| 3629 | * or just dump all volumes of a single resource. |
| 3630 | * |
| 3631 | * This may miss entries inserted after this dump started, |
| 3632 | * or entries deleted before they are reached. |
| 3633 | * |
| 3634 | * We need to make sure the device won't disappear while |
| 3635 | * we are looking at it, and revalidate our iterators |
| 3636 | * on each iteration. |
| 3637 | */ |
| 3638 | |
| 3639 | /* synchronize with conn_create()/drbd_destroy_connection() */ |
| 3640 | rcu_read_lock(); |
| 3641 | /* revalidate iterator position */ |
| 3642 | for_each_resource_rcu(tmp, &drbd_resources) { |
| 3643 | if (pos == NULL) { |
| 3644 | /* first iteration */ |
| 3645 | pos = tmp; |
| 3646 | resource = pos; |
| 3647 | break; |
| 3648 | } |
| 3649 | if (tmp == pos) { |
| 3650 | resource = pos; |
| 3651 | break; |
| 3652 | } |
| 3653 | } |
| 3654 | if (resource) { |
| 3655 | next_resource: |
| 3656 | device = idr_get_next(&resource->devices, &volume); |
| 3657 | if (!device) { |
| 3658 | /* No more volumes to dump on this resource. |
| 3659 | * Advance resource iterator. */ |
| 3660 | pos = list_entry_rcu(resource->resources.next, |
| 3661 | struct drbd_resource, resources); |
| 3662 | /* Did we dump any volume of this resource yet? */ |
| 3663 | if (volume != 0) { |
| 3664 | /* If we reached the end of the list, |
| 3665 | * or only a single resource dump was requested, |
| 3666 | * we are done. */ |
| 3667 | if (&pos->resources == &drbd_resources || cb->args[2]) |
| 3668 | goto out; |
| 3669 | volume = 0; |
| 3670 | resource = pos; |
| 3671 | goto next_resource; |
| 3672 | } |
| 3673 | } |
| 3674 | |
| 3675 | dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, |
| 3676 | cb->nlh->nlmsg_seq, &drbd_genl_family, |
| 3677 | NLM_F_MULTI, DRBD_ADM_GET_STATUS); |
| 3678 | if (!dh) |
| 3679 | goto out; |
| 3680 | |
| 3681 | if (!device) { |
| 3682 | /* This is a connection without a single volume. |
| 3683 | * Suprisingly enough, it may have a network |
| 3684 | * configuration. */ |
| 3685 | struct drbd_connection *connection; |
| 3686 | |
| 3687 | dh->minor = -1U; |
| 3688 | dh->ret_code = NO_ERROR; |
| 3689 | connection = the_only_connection(resource); |
| 3690 | if (nla_put_drbd_cfg_context(skb, resource, connection, NULL)) |
| 3691 | goto cancel; |
| 3692 | if (connection) { |
| 3693 | struct net_conf *nc; |
| 3694 | |
| 3695 | nc = rcu_dereference(connection->net_conf); |
| 3696 | if (nc && net_conf_to_skb(skb, nc, 1) != 0) |
| 3697 | goto cancel; |
| 3698 | } |
| 3699 | goto done; |
| 3700 | } |
| 3701 | |
| 3702 | D_ASSERT(device, device->vnr == volume); |
| 3703 | D_ASSERT(device, device->resource == resource); |
| 3704 | |
| 3705 | dh->minor = device_to_minor(device); |
| 3706 | dh->ret_code = NO_ERROR; |
| 3707 | |
| 3708 | if (nla_put_status_info(skb, device, NULL)) { |
| 3709 | cancel: |
| 3710 | genlmsg_cancel(skb, dh); |
| 3711 | goto out; |
| 3712 | } |
| 3713 | done: |
| 3714 | genlmsg_end(skb, dh); |
| 3715 | } |
| 3716 | |
| 3717 | out: |
| 3718 | rcu_read_unlock(); |
| 3719 | /* where to start the next iteration */ |
| 3720 | cb->args[0] = (long)pos; |
| 3721 | cb->args[1] = (pos == resource) ? volume + 1 : 0; |
| 3722 | |
| 3723 | /* No more resources/volumes/minors found results in an empty skb. |
| 3724 | * Which will terminate the dump. */ |
| 3725 | return skb->len; |
| 3726 | } |
| 3727 | |
| 3728 | /* |
| 3729 | * Request status of all resources, or of all volumes within a single resource. |
| 3730 | * |
| 3731 | * This is a dump, as the answer may not fit in a single reply skb otherwise. |
| 3732 | * Which means we cannot use the family->attrbuf or other such members, because |
| 3733 | * dump is NOT protected by the genl_lock(). During dump, we only have access |
| 3734 | * to the incoming skb, and need to opencode "parsing" of the nlattr payload. |
| 3735 | * |
| 3736 | * Once things are setup properly, we call into get_one_status(). |
| 3737 | */ |
| 3738 | int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb) |
| 3739 | { |
| 3740 | const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ; |
| 3741 | struct nlattr *nla; |
| 3742 | const char *resource_name; |
| 3743 | struct drbd_resource *resource; |
| 3744 | int maxtype; |
| 3745 | |
| 3746 | /* Is this a followup call? */ |
| 3747 | if (cb->args[0]) { |
| 3748 | /* ... of a single resource dump, |
| 3749 | * and the resource iterator has been advanced already? */ |
| 3750 | if (cb->args[2] && cb->args[2] != cb->args[0]) |
| 3751 | return 0; /* DONE. */ |
| 3752 | goto dump; |
| 3753 | } |
| 3754 | |
| 3755 | /* First call (from netlink_dump_start). We need to figure out |
| 3756 | * which resource(s) the user wants us to dump. */ |
| 3757 | nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen), |
| 3758 | nlmsg_attrlen(cb->nlh, hdrlen), |
| 3759 | DRBD_NLA_CFG_CONTEXT); |
| 3760 | |
| 3761 | /* No explicit context given. Dump all. */ |
| 3762 | if (!nla) |
| 3763 | goto dump; |
| 3764 | maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1; |
| 3765 | nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name)); |
| 3766 | if (IS_ERR(nla)) |
| 3767 | return PTR_ERR(nla); |
| 3768 | /* context given, but no name present? */ |
| 3769 | if (!nla) |
| 3770 | return -EINVAL; |
| 3771 | resource_name = nla_data(nla); |
| 3772 | if (!*resource_name) |
| 3773 | return -ENODEV; |
| 3774 | resource = drbd_find_resource(resource_name); |
| 3775 | if (!resource) |
| 3776 | return -ENODEV; |
| 3777 | |
| 3778 | kref_put(&resource->kref, drbd_destroy_resource); /* get_one_status() revalidates the resource */ |
| 3779 | |
| 3780 | /* prime iterators, and set "filter" mode mark: |
| 3781 | * only dump this connection. */ |
| 3782 | cb->args[0] = (long)resource; |
| 3783 | /* cb->args[1] = 0; passed in this way. */ |
| 3784 | cb->args[2] = (long)resource; |
| 3785 | |
| 3786 | dump: |
| 3787 | return get_one_status(skb, cb); |
| 3788 | } |
| 3789 | |
| 3790 | int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info) |
| 3791 | { |
| 3792 | struct drbd_config_context adm_ctx; |
| 3793 | enum drbd_ret_code retcode; |
| 3794 | struct timeout_parms tp; |
| 3795 | int err; |
| 3796 | |
| 3797 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 3798 | if (!adm_ctx.reply_skb) |
| 3799 | return retcode; |
| 3800 | if (retcode != NO_ERROR) |
| 3801 | goto out; |
| 3802 | |
| 3803 | tp.timeout_type = |
| 3804 | adm_ctx.device->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED : |
| 3805 | test_bit(USE_DEGR_WFC_T, &adm_ctx.device->flags) ? UT_DEGRADED : |
| 3806 | UT_DEFAULT; |
| 3807 | |
| 3808 | err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp); |
| 3809 | if (err) { |
| 3810 | nlmsg_free(adm_ctx.reply_skb); |
| 3811 | return err; |
| 3812 | } |
| 3813 | out: |
| 3814 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 3815 | return 0; |
| 3816 | } |
| 3817 | |
| 3818 | int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info) |
| 3819 | { |
| 3820 | struct drbd_config_context adm_ctx; |
| 3821 | struct drbd_device *device; |
| 3822 | enum drbd_ret_code retcode; |
| 3823 | struct start_ov_parms parms; |
| 3824 | |
| 3825 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 3826 | if (!adm_ctx.reply_skb) |
| 3827 | return retcode; |
| 3828 | if (retcode != NO_ERROR) |
| 3829 | goto out; |
| 3830 | |
| 3831 | device = adm_ctx.device; |
| 3832 | |
| 3833 | /* resume from last known position, if possible */ |
| 3834 | parms.ov_start_sector = device->ov_start_sector; |
| 3835 | parms.ov_stop_sector = ULLONG_MAX; |
| 3836 | if (info->attrs[DRBD_NLA_START_OV_PARMS]) { |
| 3837 | int err = start_ov_parms_from_attrs(&parms, info); |
| 3838 | if (err) { |
| 3839 | retcode = ERR_MANDATORY_TAG; |
| 3840 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 3841 | goto out; |
| 3842 | } |
| 3843 | } |
| 3844 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 3845 | |
| 3846 | /* w_make_ov_request expects position to be aligned */ |
| 3847 | device->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1); |
| 3848 | device->ov_stop_sector = parms.ov_stop_sector; |
| 3849 | |
| 3850 | /* If there is still bitmap IO pending, e.g. previous resync or verify |
| 3851 | * just being finished, wait for it before requesting a new resync. */ |
| 3852 | drbd_suspend_io(device); |
| 3853 | wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags)); |
| 3854 | retcode = drbd_request_state(device, NS(conn, C_VERIFY_S)); |
| 3855 | drbd_resume_io(device); |
| 3856 | |
| 3857 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 3858 | out: |
| 3859 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 3860 | return 0; |
| 3861 | } |
| 3862 | |
| 3863 | |
| 3864 | int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info) |
| 3865 | { |
| 3866 | struct drbd_config_context adm_ctx; |
| 3867 | struct drbd_device *device; |
| 3868 | enum drbd_ret_code retcode; |
| 3869 | int skip_initial_sync = 0; |
| 3870 | int err; |
| 3871 | struct new_c_uuid_parms args; |
| 3872 | |
| 3873 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 3874 | if (!adm_ctx.reply_skb) |
| 3875 | return retcode; |
| 3876 | if (retcode != NO_ERROR) |
| 3877 | goto out_nolock; |
| 3878 | |
| 3879 | device = adm_ctx.device; |
| 3880 | memset(&args, 0, sizeof(args)); |
| 3881 | if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) { |
| 3882 | err = new_c_uuid_parms_from_attrs(&args, info); |
| 3883 | if (err) { |
| 3884 | retcode = ERR_MANDATORY_TAG; |
| 3885 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 3886 | goto out_nolock; |
| 3887 | } |
| 3888 | } |
| 3889 | |
| 3890 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 3891 | mutex_lock(device->state_mutex); /* Protects us against serialized state changes. */ |
| 3892 | |
| 3893 | if (!get_ldev(device)) { |
| 3894 | retcode = ERR_NO_DISK; |
| 3895 | goto out; |
| 3896 | } |
| 3897 | |
| 3898 | /* this is "skip initial sync", assume to be clean */ |
| 3899 | if (device->state.conn == C_CONNECTED && |
| 3900 | first_peer_device(device)->connection->agreed_pro_version >= 90 && |
| 3901 | device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) { |
| 3902 | drbd_info(device, "Preparing to skip initial sync\n"); |
| 3903 | skip_initial_sync = 1; |
| 3904 | } else if (device->state.conn != C_STANDALONE) { |
| 3905 | retcode = ERR_CONNECTED; |
| 3906 | goto out_dec; |
| 3907 | } |
| 3908 | |
| 3909 | drbd_uuid_set(device, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */ |
| 3910 | drbd_uuid_new_current(device); /* New current, previous to UI_BITMAP */ |
| 3911 | |
| 3912 | if (args.clear_bm) { |
| 3913 | err = drbd_bitmap_io(device, &drbd_bmio_clear_n_write, |
| 3914 | "clear_n_write from new_c_uuid", BM_LOCKED_MASK); |
| 3915 | if (err) { |
| 3916 | drbd_err(device, "Writing bitmap failed with %d\n", err); |
| 3917 | retcode = ERR_IO_MD_DISK; |
| 3918 | } |
| 3919 | if (skip_initial_sync) { |
| 3920 | drbd_send_uuids_skip_initial_sync(first_peer_device(device)); |
| 3921 | _drbd_uuid_set(device, UI_BITMAP, 0); |
| 3922 | drbd_print_uuids(device, "cleared bitmap UUID"); |
| 3923 | spin_lock_irq(&device->resource->req_lock); |
| 3924 | _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), |
| 3925 | CS_VERBOSE, NULL); |
| 3926 | spin_unlock_irq(&device->resource->req_lock); |
| 3927 | } |
| 3928 | } |
| 3929 | |
| 3930 | drbd_md_sync(device); |
| 3931 | out_dec: |
| 3932 | put_ldev(device); |
| 3933 | out: |
| 3934 | mutex_unlock(device->state_mutex); |
| 3935 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 3936 | out_nolock: |
| 3937 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 3938 | return 0; |
| 3939 | } |
| 3940 | |
| 3941 | static enum drbd_ret_code |
| 3942 | drbd_check_resource_name(struct drbd_config_context *adm_ctx) |
| 3943 | { |
| 3944 | const char *name = adm_ctx->resource_name; |
| 3945 | if (!name || !name[0]) { |
| 3946 | drbd_msg_put_info(adm_ctx->reply_skb, "resource name missing"); |
| 3947 | return ERR_MANDATORY_TAG; |
| 3948 | } |
| 3949 | /* if we want to use these in sysfs/configfs/debugfs some day, |
| 3950 | * we must not allow slashes */ |
| 3951 | if (strchr(name, '/')) { |
| 3952 | drbd_msg_put_info(adm_ctx->reply_skb, "invalid resource name"); |
| 3953 | return ERR_INVALID_REQUEST; |
| 3954 | } |
| 3955 | return NO_ERROR; |
| 3956 | } |
| 3957 | |
| 3958 | static void resource_to_info(struct resource_info *info, |
| 3959 | struct drbd_resource *resource) |
| 3960 | { |
| 3961 | info->res_role = conn_highest_role(first_connection(resource)); |
| 3962 | info->res_susp = resource->susp; |
| 3963 | info->res_susp_nod = resource->susp_nod; |
| 3964 | info->res_susp_fen = resource->susp_fen; |
| 3965 | } |
| 3966 | |
| 3967 | int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info) |
| 3968 | { |
| 3969 | struct drbd_connection *connection; |
| 3970 | struct drbd_config_context adm_ctx; |
| 3971 | enum drbd_ret_code retcode; |
| 3972 | struct res_opts res_opts; |
| 3973 | int err; |
| 3974 | |
| 3975 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, 0); |
| 3976 | if (!adm_ctx.reply_skb) |
| 3977 | return retcode; |
| 3978 | if (retcode != NO_ERROR) |
| 3979 | goto out; |
| 3980 | |
| 3981 | set_res_opts_defaults(&res_opts); |
| 3982 | err = res_opts_from_attrs(&res_opts, info); |
| 3983 | if (err && err != -ENOMSG) { |
| 3984 | retcode = ERR_MANDATORY_TAG; |
| 3985 | drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err)); |
| 3986 | goto out; |
| 3987 | } |
| 3988 | |
| 3989 | retcode = drbd_check_resource_name(&adm_ctx); |
| 3990 | if (retcode != NO_ERROR) |
| 3991 | goto out; |
| 3992 | |
| 3993 | if (adm_ctx.resource) { |
| 3994 | if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) { |
| 3995 | retcode = ERR_INVALID_REQUEST; |
| 3996 | drbd_msg_put_info(adm_ctx.reply_skb, "resource exists"); |
| 3997 | } |
| 3998 | /* else: still NO_ERROR */ |
| 3999 | goto out; |
| 4000 | } |
| 4001 | |
| 4002 | /* not yet safe for genl_family.parallel_ops */ |
| 4003 | mutex_lock(&resources_mutex); |
| 4004 | connection = conn_create(adm_ctx.resource_name, &res_opts); |
| 4005 | mutex_unlock(&resources_mutex); |
| 4006 | |
| 4007 | if (connection) { |
| 4008 | struct resource_info resource_info; |
| 4009 | |
| 4010 | mutex_lock(¬ification_mutex); |
| 4011 | resource_to_info(&resource_info, connection->resource); |
| 4012 | notify_resource_state(NULL, 0, connection->resource, |
| 4013 | &resource_info, NOTIFY_CREATE); |
| 4014 | mutex_unlock(¬ification_mutex); |
| 4015 | } else |
| 4016 | retcode = ERR_NOMEM; |
| 4017 | |
| 4018 | out: |
| 4019 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 4020 | return 0; |
| 4021 | } |
| 4022 | |
| 4023 | static void device_to_info(struct device_info *info, |
| 4024 | struct drbd_device *device) |
| 4025 | { |
| 4026 | info->dev_disk_state = device->state.disk; |
| 4027 | } |
| 4028 | |
| 4029 | |
| 4030 | int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info) |
| 4031 | { |
| 4032 | struct drbd_config_context adm_ctx; |
| 4033 | struct drbd_genlmsghdr *dh = info->userhdr; |
| 4034 | enum drbd_ret_code retcode; |
| 4035 | |
| 4036 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE); |
| 4037 | if (!adm_ctx.reply_skb) |
| 4038 | return retcode; |
| 4039 | if (retcode != NO_ERROR) |
| 4040 | goto out; |
| 4041 | |
| 4042 | if (dh->minor > MINORMASK) { |
| 4043 | drbd_msg_put_info(adm_ctx.reply_skb, "requested minor out of range"); |
| 4044 | retcode = ERR_INVALID_REQUEST; |
| 4045 | goto out; |
| 4046 | } |
| 4047 | if (adm_ctx.volume > DRBD_VOLUME_MAX) { |
| 4048 | drbd_msg_put_info(adm_ctx.reply_skb, "requested volume id out of range"); |
| 4049 | retcode = ERR_INVALID_REQUEST; |
| 4050 | goto out; |
| 4051 | } |
| 4052 | |
| 4053 | /* drbd_adm_prepare made sure already |
| 4054 | * that first_peer_device(device)->connection and device->vnr match the request. */ |
| 4055 | if (adm_ctx.device) { |
| 4056 | if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) |
| 4057 | retcode = ERR_MINOR_OR_VOLUME_EXISTS; |
| 4058 | /* else: still NO_ERROR */ |
| 4059 | goto out; |
| 4060 | } |
| 4061 | |
| 4062 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 4063 | retcode = drbd_create_device(&adm_ctx, dh->minor); |
| 4064 | if (retcode == NO_ERROR) { |
| 4065 | struct drbd_device *device; |
| 4066 | struct drbd_peer_device *peer_device; |
| 4067 | struct device_info info; |
| 4068 | unsigned int peer_devices = 0; |
| 4069 | enum drbd_notification_type flags; |
| 4070 | |
| 4071 | device = minor_to_device(dh->minor); |
| 4072 | for_each_peer_device(peer_device, device) { |
| 4073 | if (!has_net_conf(peer_device->connection)) |
| 4074 | continue; |
| 4075 | peer_devices++; |
| 4076 | } |
| 4077 | |
| 4078 | device_to_info(&info, device); |
| 4079 | mutex_lock(¬ification_mutex); |
| 4080 | flags = (peer_devices--) ? NOTIFY_CONTINUES : 0; |
| 4081 | notify_device_state(NULL, 0, device, &info, NOTIFY_CREATE | flags); |
| 4082 | for_each_peer_device(peer_device, device) { |
| 4083 | struct peer_device_info peer_device_info; |
| 4084 | |
| 4085 | if (!has_net_conf(peer_device->connection)) |
| 4086 | continue; |
| 4087 | peer_device_to_info(&peer_device_info, peer_device); |
| 4088 | flags = (peer_devices--) ? NOTIFY_CONTINUES : 0; |
| 4089 | notify_peer_device_state(NULL, 0, peer_device, &peer_device_info, |
| 4090 | NOTIFY_CREATE | flags); |
| 4091 | } |
| 4092 | mutex_unlock(¬ification_mutex); |
| 4093 | } |
| 4094 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 4095 | out: |
| 4096 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 4097 | return 0; |
| 4098 | } |
| 4099 | |
| 4100 | static enum drbd_ret_code adm_del_minor(struct drbd_device *device) |
| 4101 | { |
| 4102 | struct drbd_peer_device *peer_device; |
| 4103 | |
| 4104 | if (device->state.disk == D_DISKLESS && |
| 4105 | /* no need to be device->state.conn == C_STANDALONE && |
| 4106 | * we may want to delete a minor from a live replication group. |
| 4107 | */ |
| 4108 | device->state.role == R_SECONDARY) { |
| 4109 | struct drbd_connection *connection = |
| 4110 | first_connection(device->resource); |
| 4111 | |
| 4112 | _drbd_request_state(device, NS(conn, C_WF_REPORT_PARAMS), |
| 4113 | CS_VERBOSE + CS_WAIT_COMPLETE); |
| 4114 | |
| 4115 | /* If the state engine hasn't stopped the sender thread yet, we |
| 4116 | * need to flush the sender work queue before generating the |
| 4117 | * DESTROY events here. */ |
| 4118 | if (get_t_state(&connection->worker) == RUNNING) |
| 4119 | drbd_flush_workqueue(&connection->sender_work); |
| 4120 | |
| 4121 | mutex_lock(¬ification_mutex); |
| 4122 | for_each_peer_device(peer_device, device) { |
| 4123 | if (!has_net_conf(peer_device->connection)) |
| 4124 | continue; |
| 4125 | notify_peer_device_state(NULL, 0, peer_device, NULL, |
| 4126 | NOTIFY_DESTROY | NOTIFY_CONTINUES); |
| 4127 | } |
| 4128 | notify_device_state(NULL, 0, device, NULL, NOTIFY_DESTROY); |
| 4129 | mutex_unlock(¬ification_mutex); |
| 4130 | |
| 4131 | drbd_delete_device(device); |
| 4132 | return NO_ERROR; |
| 4133 | } else |
| 4134 | return ERR_MINOR_CONFIGURED; |
| 4135 | } |
| 4136 | |
| 4137 | int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info) |
| 4138 | { |
| 4139 | struct drbd_config_context adm_ctx; |
| 4140 | enum drbd_ret_code retcode; |
| 4141 | |
| 4142 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR); |
| 4143 | if (!adm_ctx.reply_skb) |
| 4144 | return retcode; |
| 4145 | if (retcode != NO_ERROR) |
| 4146 | goto out; |
| 4147 | |
| 4148 | mutex_lock(&adm_ctx.resource->adm_mutex); |
| 4149 | retcode = adm_del_minor(adm_ctx.device); |
| 4150 | mutex_unlock(&adm_ctx.resource->adm_mutex); |
| 4151 | out: |
| 4152 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 4153 | return 0; |
| 4154 | } |
| 4155 | |
| 4156 | static int adm_del_resource(struct drbd_resource *resource) |
| 4157 | { |
| 4158 | struct drbd_connection *connection; |
| 4159 | |
| 4160 | for_each_connection(connection, resource) { |
| 4161 | if (connection->cstate > C_STANDALONE) |
| 4162 | return ERR_NET_CONFIGURED; |
| 4163 | } |
| 4164 | if (!idr_is_empty(&resource->devices)) |
| 4165 | return ERR_RES_IN_USE; |
| 4166 | |
| 4167 | /* The state engine has stopped the sender thread, so we don't |
| 4168 | * need to flush the sender work queue before generating the |
| 4169 | * DESTROY event here. */ |
| 4170 | mutex_lock(¬ification_mutex); |
| 4171 | notify_resource_state(NULL, 0, resource, NULL, NOTIFY_DESTROY); |
| 4172 | mutex_unlock(¬ification_mutex); |
| 4173 | |
| 4174 | mutex_lock(&resources_mutex); |
| 4175 | list_del_rcu(&resource->resources); |
| 4176 | mutex_unlock(&resources_mutex); |
| 4177 | /* Make sure all threads have actually stopped: state handling only |
| 4178 | * does drbd_thread_stop_nowait(). */ |
| 4179 | list_for_each_entry(connection, &resource->connections, connections) |
| 4180 | drbd_thread_stop(&connection->worker); |
| 4181 | synchronize_rcu(); |
| 4182 | drbd_free_resource(resource); |
| 4183 | return NO_ERROR; |
| 4184 | } |
| 4185 | |
| 4186 | int drbd_adm_down(struct sk_buff *skb, struct genl_info *info) |
| 4187 | { |
| 4188 | struct drbd_config_context adm_ctx; |
| 4189 | struct drbd_resource *resource; |
| 4190 | struct drbd_connection *connection; |
| 4191 | struct drbd_device *device; |
| 4192 | int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */ |
| 4193 | unsigned i; |
| 4194 | |
| 4195 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE); |
| 4196 | if (!adm_ctx.reply_skb) |
| 4197 | return retcode; |
| 4198 | if (retcode != NO_ERROR) |
| 4199 | goto finish; |
| 4200 | |
| 4201 | resource = adm_ctx.resource; |
| 4202 | mutex_lock(&resource->adm_mutex); |
| 4203 | /* demote */ |
| 4204 | for_each_connection(connection, resource) { |
| 4205 | struct drbd_peer_device *peer_device; |
| 4206 | |
| 4207 | idr_for_each_entry(&connection->peer_devices, peer_device, i) { |
| 4208 | retcode = drbd_set_role(peer_device->device, R_SECONDARY, 0); |
| 4209 | if (retcode < SS_SUCCESS) { |
| 4210 | drbd_msg_put_info(adm_ctx.reply_skb, "failed to demote"); |
| 4211 | goto out; |
| 4212 | } |
| 4213 | } |
| 4214 | |
| 4215 | retcode = conn_try_disconnect(connection, 0); |
| 4216 | if (retcode < SS_SUCCESS) { |
| 4217 | drbd_msg_put_info(adm_ctx.reply_skb, "failed to disconnect"); |
| 4218 | goto out; |
| 4219 | } |
| 4220 | } |
| 4221 | |
| 4222 | /* detach */ |
| 4223 | idr_for_each_entry(&resource->devices, device, i) { |
| 4224 | retcode = adm_detach(device, 0); |
| 4225 | if (retcode < SS_SUCCESS || retcode > NO_ERROR) { |
| 4226 | drbd_msg_put_info(adm_ctx.reply_skb, "failed to detach"); |
| 4227 | goto out; |
| 4228 | } |
| 4229 | } |
| 4230 | |
| 4231 | /* delete volumes */ |
| 4232 | idr_for_each_entry(&resource->devices, device, i) { |
| 4233 | retcode = adm_del_minor(device); |
| 4234 | if (retcode != NO_ERROR) { |
| 4235 | /* "can not happen" */ |
| 4236 | drbd_msg_put_info(adm_ctx.reply_skb, "failed to delete volume"); |
| 4237 | goto out; |
| 4238 | } |
| 4239 | } |
| 4240 | |
| 4241 | retcode = adm_del_resource(resource); |
| 4242 | out: |
| 4243 | mutex_unlock(&resource->adm_mutex); |
| 4244 | finish: |
| 4245 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 4246 | return 0; |
| 4247 | } |
| 4248 | |
| 4249 | int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info) |
| 4250 | { |
| 4251 | struct drbd_config_context adm_ctx; |
| 4252 | struct drbd_resource *resource; |
| 4253 | enum drbd_ret_code retcode; |
| 4254 | |
| 4255 | retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE); |
| 4256 | if (!adm_ctx.reply_skb) |
| 4257 | return retcode; |
| 4258 | if (retcode != NO_ERROR) |
| 4259 | goto finish; |
| 4260 | resource = adm_ctx.resource; |
| 4261 | |
| 4262 | mutex_lock(&resource->adm_mutex); |
| 4263 | retcode = adm_del_resource(resource); |
| 4264 | mutex_unlock(&resource->adm_mutex); |
| 4265 | finish: |
| 4266 | drbd_adm_finish(&adm_ctx, info, retcode); |
| 4267 | return 0; |
| 4268 | } |
| 4269 | |
| 4270 | void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib) |
| 4271 | { |
| 4272 | struct sk_buff *msg; |
| 4273 | struct drbd_genlmsghdr *d_out; |
| 4274 | unsigned seq; |
| 4275 | int err = -ENOMEM; |
| 4276 | |
| 4277 | seq = atomic_inc_return(&drbd_genl_seq); |
| 4278 | msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO); |
| 4279 | if (!msg) |
| 4280 | goto failed; |
| 4281 | |
| 4282 | err = -EMSGSIZE; |
| 4283 | d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT); |
| 4284 | if (!d_out) /* cannot happen, but anyways. */ |
| 4285 | goto nla_put_failure; |
| 4286 | d_out->minor = device_to_minor(device); |
| 4287 | d_out->ret_code = NO_ERROR; |
| 4288 | |
| 4289 | if (nla_put_status_info(msg, device, sib)) |
| 4290 | goto nla_put_failure; |
| 4291 | genlmsg_end(msg, d_out); |
| 4292 | err = drbd_genl_multicast_events(msg, 0); |
| 4293 | /* msg has been consumed or freed in netlink_broadcast() */ |
| 4294 | if (err && err != -ESRCH) |
| 4295 | goto failed; |
| 4296 | |
| 4297 | return; |
| 4298 | |
| 4299 | nla_put_failure: |
| 4300 | nlmsg_free(msg); |
| 4301 | failed: |
| 4302 | drbd_err(device, "Error %d while broadcasting event. " |
| 4303 | "Event seq:%u sib_reason:%u\n", |
| 4304 | err, seq, sib->sib_reason); |
| 4305 | } |
| 4306 | |
| 4307 | static int nla_put_notification_header(struct sk_buff *msg, |
| 4308 | enum drbd_notification_type type) |
| 4309 | { |
| 4310 | struct drbd_notification_header nh = { |
| 4311 | .nh_type = type, |
| 4312 | }; |
| 4313 | |
| 4314 | return drbd_notification_header_to_skb(msg, &nh, true); |
| 4315 | } |
| 4316 | |
| 4317 | void notify_resource_state(struct sk_buff *skb, |
| 4318 | unsigned int seq, |
| 4319 | struct drbd_resource *resource, |
| 4320 | struct resource_info *resource_info, |
| 4321 | enum drbd_notification_type type) |
| 4322 | { |
| 4323 | struct resource_statistics resource_statistics; |
| 4324 | struct drbd_genlmsghdr *dh; |
| 4325 | bool multicast = false; |
| 4326 | int err; |
| 4327 | |
| 4328 | if (!skb) { |
| 4329 | seq = atomic_inc_return(¬ify_genl_seq); |
| 4330 | skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO); |
| 4331 | err = -ENOMEM; |
| 4332 | if (!skb) |
| 4333 | goto failed; |
| 4334 | multicast = true; |
| 4335 | } |
| 4336 | |
| 4337 | err = -EMSGSIZE; |
| 4338 | dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_RESOURCE_STATE); |
| 4339 | if (!dh) |
| 4340 | goto nla_put_failure; |
| 4341 | dh->minor = -1U; |
| 4342 | dh->ret_code = NO_ERROR; |
| 4343 | if (nla_put_drbd_cfg_context(skb, resource, NULL, NULL) || |
| 4344 | nla_put_notification_header(skb, type) || |
| 4345 | ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY && |
| 4346 | resource_info_to_skb(skb, resource_info, true))) |
| 4347 | goto nla_put_failure; |
| 4348 | resource_statistics.res_stat_write_ordering = resource->write_ordering; |
| 4349 | err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN)); |
| 4350 | if (err) |
| 4351 | goto nla_put_failure; |
| 4352 | genlmsg_end(skb, dh); |
| 4353 | if (multicast) { |
| 4354 | err = drbd_genl_multicast_events(skb, 0); |
| 4355 | /* skb has been consumed or freed in netlink_broadcast() */ |
| 4356 | if (err && err != -ESRCH) |
| 4357 | goto failed; |
| 4358 | } |
| 4359 | return; |
| 4360 | |
| 4361 | nla_put_failure: |
| 4362 | nlmsg_free(skb); |
| 4363 | failed: |
| 4364 | drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n", |
| 4365 | err, seq); |
| 4366 | } |
| 4367 | |
| 4368 | void notify_device_state(struct sk_buff *skb, |
| 4369 | unsigned int seq, |
| 4370 | struct drbd_device *device, |
| 4371 | struct device_info *device_info, |
| 4372 | enum drbd_notification_type type) |
| 4373 | { |
| 4374 | struct device_statistics device_statistics; |
| 4375 | struct drbd_genlmsghdr *dh; |
| 4376 | bool multicast = false; |
| 4377 | int err; |
| 4378 | |
| 4379 | if (!skb) { |
| 4380 | seq = atomic_inc_return(¬ify_genl_seq); |
| 4381 | skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO); |
| 4382 | err = -ENOMEM; |
| 4383 | if (!skb) |
| 4384 | goto failed; |
| 4385 | multicast = true; |
| 4386 | } |
| 4387 | |
| 4388 | err = -EMSGSIZE; |
| 4389 | dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_DEVICE_STATE); |
| 4390 | if (!dh) |
| 4391 | goto nla_put_failure; |
| 4392 | dh->minor = device->minor; |
| 4393 | dh->ret_code = NO_ERROR; |
| 4394 | if (nla_put_drbd_cfg_context(skb, device->resource, NULL, device) || |
| 4395 | nla_put_notification_header(skb, type) || |
| 4396 | ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY && |
| 4397 | device_info_to_skb(skb, device_info, true))) |
| 4398 | goto nla_put_failure; |
| 4399 | device_to_statistics(&device_statistics, device); |
| 4400 | device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN)); |
| 4401 | genlmsg_end(skb, dh); |
| 4402 | if (multicast) { |
| 4403 | err = drbd_genl_multicast_events(skb, 0); |
| 4404 | /* skb has been consumed or freed in netlink_broadcast() */ |
| 4405 | if (err && err != -ESRCH) |
| 4406 | goto failed; |
| 4407 | } |
| 4408 | return; |
| 4409 | |
| 4410 | nla_put_failure: |
| 4411 | nlmsg_free(skb); |
| 4412 | failed: |
| 4413 | drbd_err(device, "Error %d while broadcasting event. Event seq:%u\n", |
| 4414 | err, seq); |
| 4415 | } |
| 4416 | |
| 4417 | void notify_connection_state(struct sk_buff *skb, |
| 4418 | unsigned int seq, |
| 4419 | struct drbd_connection *connection, |
| 4420 | struct connection_info *connection_info, |
| 4421 | enum drbd_notification_type type) |
| 4422 | { |
| 4423 | struct connection_statistics connection_statistics; |
| 4424 | struct drbd_genlmsghdr *dh; |
| 4425 | bool multicast = false; |
| 4426 | int err; |
| 4427 | |
| 4428 | if (!skb) { |
| 4429 | seq = atomic_inc_return(¬ify_genl_seq); |
| 4430 | skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO); |
| 4431 | err = -ENOMEM; |
| 4432 | if (!skb) |
| 4433 | goto failed; |
| 4434 | multicast = true; |
| 4435 | } |
| 4436 | |
| 4437 | err = -EMSGSIZE; |
| 4438 | dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_CONNECTION_STATE); |
| 4439 | if (!dh) |
| 4440 | goto nla_put_failure; |
| 4441 | dh->minor = -1U; |
| 4442 | dh->ret_code = NO_ERROR; |
| 4443 | if (nla_put_drbd_cfg_context(skb, connection->resource, connection, NULL) || |
| 4444 | nla_put_notification_header(skb, type) || |
| 4445 | ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY && |
| 4446 | connection_info_to_skb(skb, connection_info, true))) |
| 4447 | goto nla_put_failure; |
| 4448 | connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags); |
| 4449 | connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN)); |
| 4450 | genlmsg_end(skb, dh); |
| 4451 | if (multicast) { |
| 4452 | err = drbd_genl_multicast_events(skb, 0); |
| 4453 | /* skb has been consumed or freed in netlink_broadcast() */ |
| 4454 | if (err && err != -ESRCH) |
| 4455 | goto failed; |
| 4456 | } |
| 4457 | return; |
| 4458 | |
| 4459 | nla_put_failure: |
| 4460 | nlmsg_free(skb); |
| 4461 | failed: |
| 4462 | drbd_err(connection, "Error %d while broadcasting event. Event seq:%u\n", |
| 4463 | err, seq); |
| 4464 | } |
| 4465 | |
| 4466 | void notify_peer_device_state(struct sk_buff *skb, |
| 4467 | unsigned int seq, |
| 4468 | struct drbd_peer_device *peer_device, |
| 4469 | struct peer_device_info *peer_device_info, |
| 4470 | enum drbd_notification_type type) |
| 4471 | { |
| 4472 | struct peer_device_statistics peer_device_statistics; |
| 4473 | struct drbd_resource *resource = peer_device->device->resource; |
| 4474 | struct drbd_genlmsghdr *dh; |
| 4475 | bool multicast = false; |
| 4476 | int err; |
| 4477 | |
| 4478 | if (!skb) { |
| 4479 | seq = atomic_inc_return(¬ify_genl_seq); |
| 4480 | skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO); |
| 4481 | err = -ENOMEM; |
| 4482 | if (!skb) |
| 4483 | goto failed; |
| 4484 | multicast = true; |
| 4485 | } |
| 4486 | |
| 4487 | err = -EMSGSIZE; |
| 4488 | dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_PEER_DEVICE_STATE); |
| 4489 | if (!dh) |
| 4490 | goto nla_put_failure; |
| 4491 | dh->minor = -1U; |
| 4492 | dh->ret_code = NO_ERROR; |
| 4493 | if (nla_put_drbd_cfg_context(skb, resource, peer_device->connection, peer_device->device) || |
| 4494 | nla_put_notification_header(skb, type) || |
| 4495 | ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY && |
| 4496 | peer_device_info_to_skb(skb, peer_device_info, true))) |
| 4497 | goto nla_put_failure; |
| 4498 | peer_device_to_statistics(&peer_device_statistics, peer_device); |
| 4499 | peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN)); |
| 4500 | genlmsg_end(skb, dh); |
| 4501 | if (multicast) { |
| 4502 | err = drbd_genl_multicast_events(skb, 0); |
| 4503 | /* skb has been consumed or freed in netlink_broadcast() */ |
| 4504 | if (err && err != -ESRCH) |
| 4505 | goto failed; |
| 4506 | } |
| 4507 | return; |
| 4508 | |
| 4509 | nla_put_failure: |
| 4510 | nlmsg_free(skb); |
| 4511 | failed: |
| 4512 | drbd_err(peer_device, "Error %d while broadcasting event. Event seq:%u\n", |
| 4513 | err, seq); |
| 4514 | } |
| 4515 | |
| 4516 | void notify_helper(enum drbd_notification_type type, |
| 4517 | struct drbd_device *device, struct drbd_connection *connection, |
| 4518 | const char *name, int status) |
| 4519 | { |
| 4520 | struct drbd_resource *resource = device ? device->resource : connection->resource; |
| 4521 | struct drbd_helper_info helper_info; |
| 4522 | unsigned int seq = atomic_inc_return(¬ify_genl_seq); |
| 4523 | struct sk_buff *skb = NULL; |
| 4524 | struct drbd_genlmsghdr *dh; |
| 4525 | int err; |
| 4526 | |
| 4527 | strlcpy(helper_info.helper_name, name, sizeof(helper_info.helper_name)); |
| 4528 | helper_info.helper_name_len = min(strlen(name), sizeof(helper_info.helper_name)); |
| 4529 | helper_info.helper_status = status; |
| 4530 | |
| 4531 | skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO); |
| 4532 | err = -ENOMEM; |
| 4533 | if (!skb) |
| 4534 | goto fail; |
| 4535 | |
| 4536 | err = -EMSGSIZE; |
| 4537 | dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_HELPER); |
| 4538 | if (!dh) |
| 4539 | goto fail; |
| 4540 | dh->minor = device ? device->minor : -1; |
| 4541 | dh->ret_code = NO_ERROR; |
| 4542 | mutex_lock(¬ification_mutex); |
| 4543 | if (nla_put_drbd_cfg_context(skb, resource, connection, device) || |
| 4544 | nla_put_notification_header(skb, type) || |
| 4545 | drbd_helper_info_to_skb(skb, &helper_info, true)) |
| 4546 | goto unlock_fail; |
| 4547 | genlmsg_end(skb, dh); |
| 4548 | err = drbd_genl_multicast_events(skb, 0); |
| 4549 | skb = NULL; |
| 4550 | /* skb has been consumed or freed in netlink_broadcast() */ |
| 4551 | if (err && err != -ESRCH) |
| 4552 | goto unlock_fail; |
| 4553 | mutex_unlock(¬ification_mutex); |
| 4554 | return; |
| 4555 | |
| 4556 | unlock_fail: |
| 4557 | mutex_unlock(¬ification_mutex); |
| 4558 | fail: |
| 4559 | nlmsg_free(skb); |
| 4560 | drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n", |
| 4561 | err, seq); |
| 4562 | } |
| 4563 | |
| 4564 | static void notify_initial_state_done(struct sk_buff *skb, unsigned int seq) |
| 4565 | { |
| 4566 | struct drbd_genlmsghdr *dh; |
| 4567 | int err; |
| 4568 | |
| 4569 | err = -EMSGSIZE; |
| 4570 | dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_INITIAL_STATE_DONE); |
| 4571 | if (!dh) |
| 4572 | goto nla_put_failure; |
| 4573 | dh->minor = -1U; |
| 4574 | dh->ret_code = NO_ERROR; |
| 4575 | if (nla_put_notification_header(skb, NOTIFY_EXISTS)) |
| 4576 | goto nla_put_failure; |
| 4577 | genlmsg_end(skb, dh); |
| 4578 | return; |
| 4579 | |
| 4580 | nla_put_failure: |
| 4581 | nlmsg_free(skb); |
| 4582 | pr_err("Error %d sending event. Event seq:%u\n", err, seq); |
| 4583 | } |
| 4584 | |
| 4585 | static void free_state_changes(struct list_head *list) |
| 4586 | { |
| 4587 | while (!list_empty(list)) { |
| 4588 | struct drbd_state_change *state_change = |
| 4589 | list_first_entry(list, struct drbd_state_change, list); |
| 4590 | list_del(&state_change->list); |
| 4591 | forget_state_change(state_change); |
| 4592 | } |
| 4593 | } |
| 4594 | |
| 4595 | static unsigned int notifications_for_state_change(struct drbd_state_change *state_change) |
| 4596 | { |
| 4597 | return 1 + |
| 4598 | state_change->n_connections + |
| 4599 | state_change->n_devices + |
| 4600 | state_change->n_devices * state_change->n_connections; |
| 4601 | } |
| 4602 | |
| 4603 | static int get_initial_state(struct sk_buff *skb, struct netlink_callback *cb) |
| 4604 | { |
| 4605 | struct drbd_state_change *state_change = (struct drbd_state_change *)cb->args[0]; |
| 4606 | unsigned int seq = cb->args[2]; |
| 4607 | unsigned int n; |
| 4608 | enum drbd_notification_type flags = 0; |
| 4609 | |
| 4610 | /* There is no need for taking notification_mutex here: it doesn't |
| 4611 | matter if the initial state events mix with later state chage |
| 4612 | events; we can always tell the events apart by the NOTIFY_EXISTS |
| 4613 | flag. */ |
| 4614 | |
| 4615 | cb->args[5]--; |
| 4616 | if (cb->args[5] == 1) { |
| 4617 | notify_initial_state_done(skb, seq); |
| 4618 | goto out; |
| 4619 | } |
| 4620 | n = cb->args[4]++; |
| 4621 | if (cb->args[4] < cb->args[3]) |
| 4622 | flags |= NOTIFY_CONTINUES; |
| 4623 | if (n < 1) { |
| 4624 | notify_resource_state_change(skb, seq, state_change->resource, |
| 4625 | NOTIFY_EXISTS | flags); |
| 4626 | goto next; |
| 4627 | } |
| 4628 | n--; |
| 4629 | if (n < state_change->n_connections) { |
| 4630 | notify_connection_state_change(skb, seq, &state_change->connections[n], |
| 4631 | NOTIFY_EXISTS | flags); |
| 4632 | goto next; |
| 4633 | } |
| 4634 | n -= state_change->n_connections; |
| 4635 | if (n < state_change->n_devices) { |
| 4636 | notify_device_state_change(skb, seq, &state_change->devices[n], |
| 4637 | NOTIFY_EXISTS | flags); |
| 4638 | goto next; |
| 4639 | } |
| 4640 | n -= state_change->n_devices; |
| 4641 | if (n < state_change->n_devices * state_change->n_connections) { |
| 4642 | notify_peer_device_state_change(skb, seq, &state_change->peer_devices[n], |
| 4643 | NOTIFY_EXISTS | flags); |
| 4644 | goto next; |
| 4645 | } |
| 4646 | |
| 4647 | next: |
| 4648 | if (cb->args[4] == cb->args[3]) { |
| 4649 | struct drbd_state_change *next_state_change = |
| 4650 | list_entry(state_change->list.next, |
| 4651 | struct drbd_state_change, list); |
| 4652 | cb->args[0] = (long)next_state_change; |
| 4653 | cb->args[3] = notifications_for_state_change(next_state_change); |
| 4654 | cb->args[4] = 0; |
| 4655 | } |
| 4656 | out: |
| 4657 | return skb->len; |
| 4658 | } |
| 4659 | |
| 4660 | int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb) |
| 4661 | { |
| 4662 | struct drbd_resource *resource; |
| 4663 | LIST_HEAD(head); |
| 4664 | |
| 4665 | if (cb->args[5] >= 1) { |
| 4666 | if (cb->args[5] > 1) |
| 4667 | return get_initial_state(skb, cb); |
| 4668 | if (cb->args[0]) { |
| 4669 | struct drbd_state_change *state_change = |
| 4670 | (struct drbd_state_change *)cb->args[0]; |
| 4671 | |
| 4672 | /* connect list to head */ |
| 4673 | list_add(&head, &state_change->list); |
| 4674 | free_state_changes(&head); |
| 4675 | } |
| 4676 | return 0; |
| 4677 | } |
| 4678 | |
| 4679 | cb->args[5] = 2; /* number of iterations */ |
| 4680 | mutex_lock(&resources_mutex); |
| 4681 | for_each_resource(resource, &drbd_resources) { |
| 4682 | struct drbd_state_change *state_change; |
| 4683 | |
| 4684 | state_change = remember_old_state(resource, GFP_KERNEL); |
| 4685 | if (!state_change) { |
| 4686 | if (!list_empty(&head)) |
| 4687 | free_state_changes(&head); |
| 4688 | mutex_unlock(&resources_mutex); |
| 4689 | return -ENOMEM; |
| 4690 | } |
| 4691 | copy_old_to_new_state_change(state_change); |
| 4692 | list_add_tail(&state_change->list, &head); |
| 4693 | cb->args[5] += notifications_for_state_change(state_change); |
| 4694 | } |
| 4695 | mutex_unlock(&resources_mutex); |
| 4696 | |
| 4697 | if (!list_empty(&head)) { |
| 4698 | struct drbd_state_change *state_change = |
| 4699 | list_entry(head.next, struct drbd_state_change, list); |
| 4700 | cb->args[0] = (long)state_change; |
| 4701 | cb->args[3] = notifications_for_state_change(state_change); |
| 4702 | list_del(&head); /* detach list from head */ |
| 4703 | } |
| 4704 | |
| 4705 | cb->args[2] = cb->nlh->nlmsg_seq; |
| 4706 | return get_initial_state(skb, cb); |
| 4707 | } |