4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
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>.
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)
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.
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.
26 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
28 #include <linux/module.h>
29 #include <linux/drbd.h>
32 #include <linux/file.h>
33 #include <linux/slab.h>
34 #include <linux/blkpg.h>
35 #include <linux/cpumask.h>
37 #include "drbd_protocol.h"
39 #include "drbd_state_change.h"
40 #include <asm/unaligned.h>
41 #include <linux/drbd_limits.h>
42 #include <linux/kthread.h>
44 #include <net/genetlink.h>
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);
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);
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);
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);
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);
88 #include <linux/drbd_genl_api.h>
90 #include <linux/genl_magic_func.h>
92 static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
93 static atomic_t notify_genl_seq = ATOMIC_INIT(2); /* two. */
95 DEFINE_MUTEX(notification_mutex);
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.";
100 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
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");
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)
114 if (!info || !info[0])
117 nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY);
121 err = nla_put_string(skb, T_info_text, info);
123 nla_nest_cancel(skb, nla);
126 nla_nest_end(skb, nla);
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.
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.
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)
145 struct drbd_genlmsghdr *d_in = info->userhdr;
146 const u8 cmd = info->genlhdr->cmd;
149 memset(adm_ctx, 0, sizeof(*adm_ctx));
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))
155 adm_ctx->reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
156 if (!adm_ctx->reply_skb) {
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.
165 if (!adm_ctx->reply_dh) {
170 adm_ctx->reply_dh->minor = d_in->minor;
171 adm_ctx->reply_dh->ret_code = NO_ERROR;
173 adm_ctx->volume = VOLUME_UNSPECIFIED;
174 if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
176 /* parse and validate only */
177 err = drbd_cfg_context_from_attrs(NULL, info);
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]);
189 /* and assign stuff to the adm_ctx */
190 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
192 adm_ctx->volume = nla_get_u32(nla);
193 nla = nested_attr_tb[__nla_type(T_ctx_resource_name)];
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))) {
207 adm_ctx->minor = d_in->minor;
208 adm_ctx->device = minor_to_device(d_in->minor);
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. */
214 kref_get(&adm_ctx->device->kref);
216 if (adm_ctx->resource_name) {
217 adm_ctx->resource = drbd_find_resource(adm_ctx->resource_name);
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;
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;
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;
236 if (adm_ctx->device) {
237 drbd_msg_put_info(adm_ctx->reply_skb, "no minor number expected");
238 return ERR_INVALID_REQUEST;
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;
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;
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;
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);
282 nlmsg_free(adm_ctx->reply_skb);
283 adm_ctx->reply_skb = NULL;
287 static int drbd_adm_finish(struct drbd_config_context *adm_ctx,
288 struct genl_info *info, int retcode)
290 if (adm_ctx->device) {
291 kref_put(&adm_ctx->device->kref, drbd_destroy_device);
292 adm_ctx->device = NULL;
294 if (adm_ctx->connection) {
295 kref_put(&adm_ctx->connection->kref, &drbd_destroy_connection);
296 adm_ctx->connection = NULL;
298 if (adm_ctx->resource) {
299 kref_put(&adm_ctx->resource->kref, drbd_destroy_resource);
300 adm_ctx->resource = NULL;
303 if (!adm_ctx->reply_skb)
306 adm_ctx->reply_dh->ret_code = retcode;
307 drbd_adm_send_reply(adm_ctx->reply_skb, info);
311 static void setup_khelper_env(struct drbd_connection *connection, char **envp)
315 /* FIXME: A future version will not allow this case. */
316 if (connection->my_addr_len == 0 || connection->peer_addr_len == 0)
319 switch (((struct sockaddr *)&connection->peer_addr)->sa_family) {
322 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
323 &((struct sockaddr_in6 *)&connection->peer_addr)->sin6_addr);
327 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
328 &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
332 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
333 &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
335 snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
338 int drbd_khelper(struct drbd_device *device, char *cmd)
340 char *envp[] = { "HOME=/",
342 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
343 (char[20]) { }, /* address family */
344 (char[60]) { }, /* address */
347 char *argv[] = {usermode_helper, cmd, mb, NULL };
348 struct drbd_connection *connection = first_peer_device(device)->connection;
352 if (current == connection->worker.task)
353 set_bit(CALLBACK_PENDING, &connection->flags);
355 snprintf(mb, 14, "minor-%d", device_to_minor(device));
356 setup_khelper_env(connection, envp);
358 /* The helper may take some time.
359 * write out any unsynced meta data changes now */
360 drbd_md_sync(device);
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);
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);
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);
381 if (current == connection->worker.task)
382 clear_bit(CALLBACK_PENDING, &connection->flags);
384 if (ret < 0) /* Ignore any ERRNOs we got. */
390 enum drbd_peer_state conn_khelper(struct drbd_connection *connection, char *cmd)
392 char *envp[] = { "HOME=/",
394 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
395 (char[20]) { }, /* address family */
396 (char[60]) { }, /* address */
398 char *resource_name = connection->resource->name;
399 char *argv[] = {usermode_helper, cmd, resource_name, NULL };
402 setup_khelper_env(connection, envp);
403 conn_md_sync(connection);
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);
409 ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC);
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);
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);
421 if (ret < 0) /* Ignore any ERRNOs we got. */
427 static enum drbd_fencing_p highest_fencing_policy(struct drbd_connection *connection)
429 enum drbd_fencing_p fp = FP_NOT_AVAIL;
430 struct drbd_peer_device *peer_device;
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);
448 static bool resource_is_supended(struct drbd_resource *resource)
450 return resource->susp || resource->susp_fen || resource->susp_nod;
453 bool conn_try_outdate_peer(struct drbd_connection *connection)
455 struct drbd_resource * const resource = connection->resource;
456 unsigned int connect_cnt;
457 union drbd_state mask = { };
458 union drbd_state val = { };
459 enum drbd_fencing_p fp;
463 spin_lock_irq(&resource->req_lock);
464 if (connection->cstate >= C_WF_REPORT_PARAMS) {
465 drbd_err(connection, "Expected cstate < C_WF_REPORT_PARAMS\n");
466 spin_unlock_irq(&resource->req_lock);
470 connect_cnt = connection->connect_cnt;
471 spin_unlock_irq(&resource->req_lock);
473 fp = highest_fencing_policy(connection);
476 drbd_warn(connection, "Not fencing peer, I'm not even Consistent myself.\n");
477 spin_lock_irq(&resource->req_lock);
478 if (connection->cstate < C_WF_REPORT_PARAMS) {
479 _conn_request_state(connection,
480 (union drbd_state) { { .susp_fen = 1 } },
481 (union drbd_state) { { .susp_fen = 0 } },
482 CS_VERBOSE | CS_HARD | CS_DC_SUSP);
483 /* We are no longer suspended due to the fencing policy.
484 * We may still be suspended due to the on-no-data-accessible policy.
485 * If that was OND_IO_ERROR, fail pending requests. */
486 if (!resource_is_supended(resource))
487 _tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
489 /* Else: in case we raced with a connection handshake,
490 * let the handshake figure out if we maybe can RESEND,
491 * and do not resume/fail pending requests here.
492 * Worst case is we stay suspended for now, which may be
493 * resolved by either re-establishing the replication link, or
494 * the next link failure, or eventually the administrator. */
495 spin_unlock_irq(&resource->req_lock);
503 r = conn_khelper(connection, "fence-peer");
505 switch ((r>>8) & 0xff) {
506 case P_INCONSISTENT: /* peer is inconsistent */
507 ex_to_string = "peer is inconsistent or worse";
509 val.pdsk = D_INCONSISTENT;
511 case P_OUTDATED: /* peer got outdated, or was already outdated */
512 ex_to_string = "peer was fenced";
514 val.pdsk = D_OUTDATED;
516 case P_DOWN: /* peer was down */
517 if (conn_highest_disk(connection) == D_UP_TO_DATE) {
518 /* we will(have) create(d) a new UUID anyways... */
519 ex_to_string = "peer is unreachable, assumed to be dead";
521 val.pdsk = D_OUTDATED;
523 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
526 case P_PRIMARY: /* Peer is primary, voluntarily outdate myself.
527 * This is useful when an unconnected R_SECONDARY is asked to
528 * become R_PRIMARY, but finds the other peer being active. */
529 ex_to_string = "peer is active";
530 drbd_warn(connection, "Peer is primary, outdating myself.\n");
532 val.disk = D_OUTDATED;
535 /* THINK: do we need to handle this
536 * like case 4, or more like case 5? */
537 if (fp != FP_STONITH)
538 drbd_err(connection, "fence-peer() = 7 && fencing != Stonith !!!\n");
539 ex_to_string = "peer was stonithed";
541 val.pdsk = D_OUTDATED;
544 /* The script is broken ... */
545 drbd_err(connection, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
546 return false; /* Eventually leave IO frozen */
549 drbd_info(connection, "fence-peer helper returned %d (%s)\n",
550 (r>>8) & 0xff, ex_to_string);
553 conn_request_state(connection, mask, val, CS_VERBOSE);
554 here, because we might were able to re-establish the connection in the
556 spin_lock_irq(&resource->req_lock);
557 if (connection->cstate < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &connection->flags)) {
558 if (connection->connect_cnt != connect_cnt)
559 /* In case the connection was established and droped
560 while the fence-peer handler was running, ignore it */
561 drbd_info(connection, "Ignoring fence-peer exit code\n");
563 _conn_request_state(connection, mask, val, CS_VERBOSE);
565 spin_unlock_irq(&resource->req_lock);
567 return conn_highest_pdsk(connection) <= D_OUTDATED;
570 static int _try_outdate_peer_async(void *data)
572 struct drbd_connection *connection = (struct drbd_connection *)data;
574 conn_try_outdate_peer(connection);
576 kref_put(&connection->kref, drbd_destroy_connection);
580 void conn_try_outdate_peer_async(struct drbd_connection *connection)
582 struct task_struct *opa;
584 kref_get(&connection->kref);
585 /* We may just have force_sig()'ed this thread
586 * to get it out of some blocking network function.
587 * Clear signals; otherwise kthread_run(), which internally uses
588 * wait_on_completion_killable(), will mistake our pending signal
589 * for a new fatal signal and fail. */
590 flush_signals(current);
591 opa = kthread_run(_try_outdate_peer_async, connection, "drbd_async_h");
593 drbd_err(connection, "out of mem, failed to invoke fence-peer helper\n");
594 kref_put(&connection->kref, drbd_destroy_connection);
599 drbd_set_role(struct drbd_device *const device, enum drbd_role new_role, int force)
601 struct drbd_peer_device *const peer_device = first_peer_device(device);
602 struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL;
603 const int max_tries = 4;
604 enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
608 union drbd_state mask, val;
610 if (new_role == R_PRIMARY) {
611 struct drbd_connection *connection;
613 /* Detect dead peers as soon as possible. */
616 for_each_connection(connection, device->resource)
617 request_ping(connection);
621 mutex_lock(device->state_mutex);
623 mask.i = 0; mask.role = R_MASK;
624 val.i = 0; val.role = new_role;
626 while (try++ < max_tries) {
627 rv = _drbd_request_state_holding_state_mutex(device, mask, val, CS_WAIT_COMPLETE);
629 /* in case we first succeeded to outdate,
630 * but now suddenly could establish a connection */
631 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
637 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
638 (device->state.disk < D_UP_TO_DATE &&
639 device->state.disk >= D_INCONSISTENT)) {
641 val.disk = D_UP_TO_DATE;
646 if (rv == SS_NO_UP_TO_DATE_DISK &&
647 device->state.disk == D_CONSISTENT && mask.pdsk == 0) {
648 D_ASSERT(device, device->state.pdsk == D_UNKNOWN);
650 if (conn_try_outdate_peer(connection)) {
651 val.disk = D_UP_TO_DATE;
657 if (rv == SS_NOTHING_TO_DO)
659 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
660 if (!conn_try_outdate_peer(connection) && force) {
661 drbd_warn(device, "Forced into split brain situation!\n");
663 val.pdsk = D_OUTDATED;
668 if (rv == SS_TWO_PRIMARIES) {
669 /* Maybe the peer is detected as dead very soon...
670 retry at most once more in this case. */
673 nc = rcu_dereference(connection->net_conf);
674 timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
676 schedule_timeout_interruptible(timeo);
681 if (rv < SS_SUCCESS) {
682 rv = _drbd_request_state(device, mask, val,
683 CS_VERBOSE + CS_WAIT_COMPLETE);
694 drbd_warn(device, "Forced to consider local data as UpToDate!\n");
696 /* Wait until nothing is on the fly :) */
697 wait_event(device->misc_wait, atomic_read(&device->ap_pending_cnt) == 0);
699 /* FIXME also wait for all pending P_BARRIER_ACK? */
701 if (new_role == R_SECONDARY) {
702 if (get_ldev(device)) {
703 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
707 mutex_lock(&device->resource->conf_update);
708 nc = connection->net_conf;
710 nc->discard_my_data = 0; /* without copy; single bit op is atomic */
711 mutex_unlock(&device->resource->conf_update);
713 if (get_ldev(device)) {
714 if (((device->state.conn < C_CONNECTED ||
715 device->state.pdsk <= D_FAILED)
716 && device->ldev->md.uuid[UI_BITMAP] == 0) || forced)
717 drbd_uuid_new_current(device);
719 device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
724 /* writeout of activity log covered areas of the bitmap
725 * to stable storage done in after state change already */
727 if (device->state.conn >= C_WF_REPORT_PARAMS) {
728 /* if this was forced, we should consider sync */
730 drbd_send_uuids(peer_device);
731 drbd_send_current_state(peer_device);
734 drbd_md_sync(device);
735 set_disk_ro(device->vdisk, new_role == R_SECONDARY);
736 kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
738 mutex_unlock(device->state_mutex);
742 static const char *from_attrs_err_to_txt(int err)
744 return err == -ENOMSG ? "required attribute missing" :
745 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
746 err == -EEXIST ? "can not change invariant setting" :
747 "invalid attribute value";
750 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
752 struct drbd_config_context adm_ctx;
753 struct set_role_parms parms;
755 enum drbd_ret_code retcode;
757 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
758 if (!adm_ctx.reply_skb)
760 if (retcode != NO_ERROR)
763 memset(&parms, 0, sizeof(parms));
764 if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
765 err = set_role_parms_from_attrs(&parms, info);
767 retcode = ERR_MANDATORY_TAG;
768 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
773 mutex_lock(&adm_ctx.resource->adm_mutex);
775 if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
776 retcode = drbd_set_role(adm_ctx.device, R_PRIMARY, parms.assume_uptodate);
778 retcode = drbd_set_role(adm_ctx.device, R_SECONDARY, 0);
780 mutex_unlock(&adm_ctx.resource->adm_mutex);
783 drbd_adm_finish(&adm_ctx, info, retcode);
787 /* Initializes the md.*_offset members, so we are able to find
788 * the on disk meta data.
790 * We currently have two possible layouts:
792 * |----------- md_size_sect ------------------|
793 * [ 4k superblock ][ activity log ][ Bitmap ]
795 * | bm_offset = al_offset + X |
796 * ==> bitmap sectors = md_size_sect - bm_offset
799 * |----------- md_size_sect ------------------|
800 * [data.....][ Bitmap ][ activity log ][ 4k superblock ]
802 * | bm_offset = al_offset - Y |
803 * ==> bitmap sectors = Y = al_offset - bm_offset
805 * Activity log size used to be fixed 32kB,
806 * but is about to become configurable.
808 static void drbd_md_set_sector_offsets(struct drbd_device *device,
809 struct drbd_backing_dev *bdev)
811 sector_t md_size_sect = 0;
812 unsigned int al_size_sect = bdev->md.al_size_4k * 8;
814 bdev->md.md_offset = drbd_md_ss(bdev);
816 switch (bdev->md.meta_dev_idx) {
818 /* v07 style fixed size indexed meta data */
819 bdev->md.md_size_sect = MD_128MB_SECT;
820 bdev->md.al_offset = MD_4kB_SECT;
821 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
823 case DRBD_MD_INDEX_FLEX_EXT:
824 /* just occupy the full device; unit: sectors */
825 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
826 bdev->md.al_offset = MD_4kB_SECT;
827 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
829 case DRBD_MD_INDEX_INTERNAL:
830 case DRBD_MD_INDEX_FLEX_INT:
831 /* al size is still fixed */
832 bdev->md.al_offset = -al_size_sect;
833 /* we need (slightly less than) ~ this much bitmap sectors: */
834 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
835 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
836 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
837 md_size_sect = ALIGN(md_size_sect, 8);
839 /* plus the "drbd meta data super block",
840 * and the activity log; */
841 md_size_sect += MD_4kB_SECT + al_size_sect;
843 bdev->md.md_size_sect = md_size_sect;
844 /* bitmap offset is adjusted by 'super' block size */
845 bdev->md.bm_offset = -md_size_sect + MD_4kB_SECT;
850 /* input size is expected to be in KB */
851 char *ppsize(char *buf, unsigned long long size)
853 /* Needs 9 bytes at max including trailing NUL:
854 * -1ULL ==> "16384 EB" */
855 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
857 while (size >= 10000 && base < sizeof(units)-1) {
859 size = (size >> 10) + !!(size & (1<<9));
862 sprintf(buf, "%u %cB", (unsigned)size, units[base]);
867 /* there is still a theoretical deadlock when called from receiver
868 * on an D_INCONSISTENT R_PRIMARY:
869 * remote READ does inc_ap_bio, receiver would need to receive answer
870 * packet from remote to dec_ap_bio again.
871 * receiver receive_sizes(), comes here,
872 * waits for ap_bio_cnt == 0. -> deadlock.
873 * but this cannot happen, actually, because:
874 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
875 * (not connected, or bad/no disk on peer):
876 * see drbd_fail_request_early, ap_bio_cnt is zero.
877 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
878 * peer may not initiate a resize.
880 /* Note these are not to be confused with
881 * drbd_adm_suspend_io/drbd_adm_resume_io,
882 * which are (sub) state changes triggered by admin (drbdsetup),
883 * and can be long lived.
884 * This changes an device->flag, is triggered by drbd internals,
885 * and should be short-lived. */
886 /* It needs to be a counter, since multiple threads might
887 independently suspend and resume IO. */
888 void drbd_suspend_io(struct drbd_device *device)
890 atomic_inc(&device->suspend_cnt);
891 if (drbd_suspended(device))
893 wait_event(device->misc_wait, !atomic_read(&device->ap_bio_cnt));
896 void drbd_resume_io(struct drbd_device *device)
898 if (atomic_dec_and_test(&device->suspend_cnt))
899 wake_up(&device->misc_wait);
903 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
904 * @device: DRBD device.
906 * Returns 0 on success, negative return values indicate errors.
907 * You should call drbd_md_sync() after calling this function.
909 enum determine_dev_size
910 drbd_determine_dev_size(struct drbd_device *device, enum dds_flags flags, struct resize_parms *rs) __must_hold(local)
912 struct md_offsets_and_sizes {
913 u64 last_agreed_sect;
920 u32 al_stripe_size_4k;
922 sector_t u_size, size;
923 struct drbd_md *md = &device->ldev->md;
927 int md_moved, la_size_changed;
928 enum determine_dev_size rv = DS_UNCHANGED;
930 /* We may change the on-disk offsets of our meta data below. Lock out
931 * anything that may cause meta data IO, to avoid acting on incomplete
932 * layout changes or scribbling over meta data that is in the process
935 * Move is not exactly correct, btw, currently we have all our meta
936 * data in core memory, to "move" it we just write it all out, there
938 drbd_suspend_io(device);
939 buffer = drbd_md_get_buffer(device, __func__); /* Lock meta-data IO */
941 drbd_resume_io(device);
945 /* remember current offset and sizes */
946 prev.last_agreed_sect = md->la_size_sect;
947 prev.md_offset = md->md_offset;
948 prev.al_offset = md->al_offset;
949 prev.bm_offset = md->bm_offset;
950 prev.md_size_sect = md->md_size_sect;
951 prev.al_stripes = md->al_stripes;
952 prev.al_stripe_size_4k = md->al_stripe_size_4k;
955 /* rs is non NULL if we should change the AL layout only */
956 md->al_stripes = rs->al_stripes;
957 md->al_stripe_size_4k = rs->al_stripe_size / 4;
958 md->al_size_4k = (u64)rs->al_stripes * rs->al_stripe_size / 4;
961 drbd_md_set_sector_offsets(device, device->ldev);
964 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
966 size = drbd_new_dev_size(device, device->ldev, u_size, flags & DDSF_FORCED);
968 if (size < prev.last_agreed_sect) {
969 if (rs && u_size == 0) {
970 /* Remove "rs &&" later. This check should always be active, but
971 right now the receiver expects the permissive behavior */
972 drbd_warn(device, "Implicit shrink not allowed. "
973 "Use --size=%llus for explicit shrink.\n",
974 (unsigned long long)size);
975 rv = DS_ERROR_SHRINK;
978 rv = DS_ERROR_SPACE_MD;
979 if (rv != DS_UNCHANGED)
983 if (drbd_get_capacity(device->this_bdev) != size ||
984 drbd_bm_capacity(device) != size) {
986 err = drbd_bm_resize(device, size, !(flags & DDSF_NO_RESYNC));
988 /* currently there is only one error: ENOMEM! */
989 size = drbd_bm_capacity(device);
991 drbd_err(device, "OUT OF MEMORY! "
992 "Could not allocate bitmap!\n");
994 drbd_err(device, "BM resizing failed. "
995 "Leaving size unchanged\n");
999 /* racy, see comments above. */
1000 drbd_set_my_capacity(device, size);
1001 md->la_size_sect = size;
1002 drbd_info(device, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
1003 (unsigned long long)size>>1);
1008 la_size_changed = (prev.last_agreed_sect != md->la_size_sect);
1010 md_moved = prev.md_offset != md->md_offset
1011 || prev.md_size_sect != md->md_size_sect;
1013 if (la_size_changed || md_moved || rs) {
1016 /* We do some synchronous IO below, which may take some time.
1017 * Clear the timer, to avoid scary "timer expired!" messages,
1018 * "Superblock" is written out at least twice below, anyways. */
1019 del_timer(&device->md_sync_timer);
1021 /* We won't change the "al-extents" setting, we just may need
1022 * to move the on-disk location of the activity log ringbuffer.
1023 * Lock for transaction is good enough, it may well be "dirty"
1024 * or even "starving". */
1025 wait_event(device->al_wait, lc_try_lock_for_transaction(device->act_log));
1027 /* mark current on-disk bitmap and activity log as unreliable */
1028 prev_flags = md->flags;
1029 md->flags |= MDF_FULL_SYNC | MDF_AL_DISABLED;
1030 drbd_md_write(device, buffer);
1032 drbd_al_initialize(device, buffer);
1034 drbd_info(device, "Writing the whole bitmap, %s\n",
1035 la_size_changed && md_moved ? "size changed and md moved" :
1036 la_size_changed ? "size changed" : "md moved");
1037 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
1038 drbd_bitmap_io(device, md_moved ? &drbd_bm_write_all : &drbd_bm_write,
1039 "size changed", BM_LOCKED_MASK);
1041 /* on-disk bitmap and activity log is authoritative again
1042 * (unless there was an IO error meanwhile...) */
1043 md->flags = prev_flags;
1044 drbd_md_write(device, buffer);
1047 drbd_info(device, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n",
1048 md->al_stripes, md->al_stripe_size_4k * 4);
1051 if (size > prev.last_agreed_sect)
1052 rv = prev.last_agreed_sect ? DS_GREW : DS_GREW_FROM_ZERO;
1053 if (size < prev.last_agreed_sect)
1058 /* restore previous offset and sizes */
1059 md->la_size_sect = prev.last_agreed_sect;
1060 md->md_offset = prev.md_offset;
1061 md->al_offset = prev.al_offset;
1062 md->bm_offset = prev.bm_offset;
1063 md->md_size_sect = prev.md_size_sect;
1064 md->al_stripes = prev.al_stripes;
1065 md->al_stripe_size_4k = prev.al_stripe_size_4k;
1066 md->al_size_4k = (u64)prev.al_stripes * prev.al_stripe_size_4k;
1068 lc_unlock(device->act_log);
1069 wake_up(&device->al_wait);
1070 drbd_md_put_buffer(device);
1071 drbd_resume_io(device);
1077 drbd_new_dev_size(struct drbd_device *device, struct drbd_backing_dev *bdev,
1078 sector_t u_size, int assume_peer_has_space)
1080 sector_t p_size = device->p_size; /* partner's disk size. */
1081 sector_t la_size_sect = bdev->md.la_size_sect; /* last agreed size. */
1082 sector_t m_size; /* my size */
1085 m_size = drbd_get_max_capacity(bdev);
1087 if (device->state.conn < C_CONNECTED && assume_peer_has_space) {
1088 drbd_warn(device, "Resize while not connected was forced by the user!\n");
1092 if (p_size && m_size) {
1093 size = min_t(sector_t, p_size, m_size);
1096 size = la_size_sect;
1097 if (m_size && m_size < size)
1099 if (p_size && p_size < size)
1110 drbd_err(device, "Both nodes diskless!\n");
1114 drbd_err(device, "Requested disk size is too big (%lu > %lu)\n",
1115 (unsigned long)u_size>>1, (unsigned long)size>>1);
1124 * drbd_check_al_size() - Ensures that the AL is of the right size
1125 * @device: DRBD device.
1127 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
1128 * failed, and 0 on success. You should call drbd_md_sync() after you called
1131 static int drbd_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1133 struct lru_cache *n, *t;
1134 struct lc_element *e;
1135 unsigned int in_use;
1138 if (device->act_log &&
1139 device->act_log->nr_elements == dc->al_extents)
1143 t = device->act_log;
1144 n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
1145 dc->al_extents, sizeof(struct lc_element), 0);
1148 drbd_err(device, "Cannot allocate act_log lru!\n");
1151 spin_lock_irq(&device->al_lock);
1153 for (i = 0; i < t->nr_elements; i++) {
1154 e = lc_element_by_index(t, i);
1156 drbd_err(device, "refcnt(%d)==%d\n",
1157 e->lc_number, e->refcnt);
1158 in_use += e->refcnt;
1162 device->act_log = n;
1163 spin_unlock_irq(&device->al_lock);
1165 drbd_err(device, "Activity log still in use!\n");
1171 drbd_md_mark_dirty(device); /* we changed device->act_log->nr_elemens */
1175 static void blk_queue_discard_granularity(struct request_queue *q, unsigned int granularity)
1177 q->limits.discard_granularity = granularity;
1180 static unsigned int drbd_max_discard_sectors(struct drbd_connection *connection)
1182 /* when we introduced REQ_WRITE_SAME support, we also bumped
1183 * our maximum supported batch bio size used for discards. */
1184 if (connection->agreed_features & DRBD_FF_WSAME)
1185 return DRBD_MAX_BBIO_SECTORS;
1186 /* before, with DRBD <= 8.4.6, we only allowed up to one AL_EXTENT_SIZE. */
1187 return AL_EXTENT_SIZE >> 9;
1190 static void decide_on_discard_support(struct drbd_device *device,
1191 struct request_queue *q,
1192 struct request_queue *b,
1193 bool discard_zeroes_if_aligned)
1195 /* q = drbd device queue (device->rq_queue)
1196 * b = backing device queue (device->ldev->backing_bdev->bd_disk->queue),
1197 * or NULL if diskless
1199 struct drbd_connection *connection = first_peer_device(device)->connection;
1200 bool can_do = b ? blk_queue_discard(b) : true;
1202 if (can_do && connection->cstate >= C_CONNECTED && !(connection->agreed_features & DRBD_FF_TRIM)) {
1204 drbd_info(connection, "peer DRBD too old, does not support TRIM: disabling discards\n");
1207 /* We don't care for the granularity, really.
1208 * Stacking limits below should fix it for the local
1209 * device. Whether or not it is a suitable granularity
1210 * on the remote device is not our problem, really. If
1211 * you care, you need to use devices with similar
1212 * topology on all peers. */
1213 blk_queue_discard_granularity(q, 512);
1214 q->limits.max_discard_sectors = drbd_max_discard_sectors(connection);
1215 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1216 q->limits.max_write_zeroes_sectors = drbd_max_discard_sectors(connection);
1218 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1219 blk_queue_discard_granularity(q, 0);
1220 q->limits.max_discard_sectors = 0;
1221 q->limits.max_write_zeroes_sectors = 0;
1225 static void fixup_discard_if_not_supported(struct request_queue *q)
1227 /* To avoid confusion, if this queue does not support discard, clear
1228 * max_discard_sectors, which is what lsblk -D reports to the user.
1229 * Older kernels got this wrong in "stack limits".
1231 if (!blk_queue_discard(q)) {
1232 blk_queue_max_discard_sectors(q, 0);
1233 blk_queue_discard_granularity(q, 0);
1237 static void decide_on_write_same_support(struct drbd_device *device,
1238 struct request_queue *q,
1239 struct request_queue *b, struct o_qlim *o)
1241 struct drbd_peer_device *peer_device = first_peer_device(device);
1242 struct drbd_connection *connection = peer_device->connection;
1243 bool can_do = b ? b->limits.max_write_same_sectors : true;
1245 if (can_do && connection->cstate >= C_CONNECTED && !(connection->agreed_features & DRBD_FF_WSAME)) {
1247 drbd_info(peer_device, "peer does not support WRITE_SAME\n");
1251 /* logical block size; queue_logical_block_size(NULL) is 512 */
1252 unsigned int peer_lbs = be32_to_cpu(o->logical_block_size);
1253 unsigned int me_lbs_b = queue_logical_block_size(b);
1254 unsigned int me_lbs = queue_logical_block_size(q);
1256 if (me_lbs_b != me_lbs) {
1258 "logical block size of local backend does not match (drbd:%u, backend:%u); was this a late attach?\n",
1260 /* rather disable write same than trigger some BUG_ON later in the scsi layer. */
1263 if (me_lbs_b != peer_lbs) {
1264 drbd_warn(peer_device, "logical block sizes do not match (me:%u, peer:%u); this may cause problems.\n",
1267 drbd_dbg(peer_device, "logical block size mismatch: WRITE_SAME disabled.\n");
1270 me_lbs = max(me_lbs, me_lbs_b);
1271 /* We cannot change the logical block size of an in-use queue.
1272 * We can only hope that access happens to be properly aligned.
1273 * If not, the peer will likely produce an IO error, and detach. */
1274 if (peer_lbs > me_lbs) {
1275 if (device->state.role != R_PRIMARY) {
1276 blk_queue_logical_block_size(q, peer_lbs);
1277 drbd_warn(peer_device, "logical block size set to %u\n", peer_lbs);
1279 drbd_warn(peer_device,
1280 "current Primary must NOT adjust logical block size (%u -> %u); hope for the best.\n",
1285 if (can_do && !o->write_same_capable) {
1286 /* If we introduce an open-coded write-same loop on the receiving side,
1287 * the peer would present itself as "capable". */
1288 drbd_dbg(peer_device, "WRITE_SAME disabled (peer device not capable)\n");
1293 blk_queue_max_write_same_sectors(q, can_do ? DRBD_MAX_BBIO_SECTORS : 0);
1296 static void drbd_setup_queue_param(struct drbd_device *device, struct drbd_backing_dev *bdev,
1297 unsigned int max_bio_size, struct o_qlim *o)
1299 struct request_queue * const q = device->rq_queue;
1300 unsigned int max_hw_sectors = max_bio_size >> 9;
1301 unsigned int max_segments = 0;
1302 struct request_queue *b = NULL;
1303 struct disk_conf *dc;
1304 bool discard_zeroes_if_aligned = true;
1307 b = bdev->backing_bdev->bd_disk->queue;
1309 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1311 dc = rcu_dereference(device->ldev->disk_conf);
1312 max_segments = dc->max_bio_bvecs;
1313 discard_zeroes_if_aligned = dc->discard_zeroes_if_aligned;
1316 blk_set_stacking_limits(&q->limits);
1319 blk_queue_max_hw_sectors(q, max_hw_sectors);
1320 /* This is the workaround for "bio would need to, but cannot, be split" */
1321 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1322 blk_queue_segment_boundary(q, PAGE_SIZE-1);
1323 decide_on_discard_support(device, q, b, discard_zeroes_if_aligned);
1324 decide_on_write_same_support(device, q, b, o);
1327 blk_queue_stack_limits(q, b);
1329 if (q->backing_dev_info->ra_pages !=
1330 b->backing_dev_info->ra_pages) {
1331 drbd_info(device, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1332 q->backing_dev_info->ra_pages,
1333 b->backing_dev_info->ra_pages);
1334 q->backing_dev_info->ra_pages =
1335 b->backing_dev_info->ra_pages;
1338 fixup_discard_if_not_supported(q);
1341 void drbd_reconsider_queue_parameters(struct drbd_device *device, struct drbd_backing_dev *bdev, struct o_qlim *o)
1343 unsigned int now, new, local, peer;
1345 now = queue_max_hw_sectors(device->rq_queue) << 9;
1346 local = device->local_max_bio_size; /* Eventually last known value, from volatile memory */
1347 peer = device->peer_max_bio_size; /* Eventually last known value, from meta data */
1350 local = queue_max_hw_sectors(bdev->backing_bdev->bd_disk->queue) << 9;
1351 device->local_max_bio_size = local;
1353 local = min(local, DRBD_MAX_BIO_SIZE);
1355 /* We may ignore peer limits if the peer is modern enough.
1356 Because new from 8.3.8 onwards the peer can use multiple
1357 BIOs for a single peer_request */
1358 if (device->state.conn >= C_WF_REPORT_PARAMS) {
1359 if (first_peer_device(device)->connection->agreed_pro_version < 94)
1360 peer = min(device->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1361 /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1362 else if (first_peer_device(device)->connection->agreed_pro_version == 94)
1363 peer = DRBD_MAX_SIZE_H80_PACKET;
1364 else if (first_peer_device(device)->connection->agreed_pro_version < 100)
1365 peer = DRBD_MAX_BIO_SIZE_P95; /* drbd 8.3.8 onwards, before 8.4.0 */
1367 peer = DRBD_MAX_BIO_SIZE;
1369 /* We may later detach and re-attach on a disconnected Primary.
1370 * Avoid this setting to jump back in that case.
1371 * We want to store what we know the peer DRBD can handle,
1372 * not what the peer IO backend can handle. */
1373 if (peer > device->peer_max_bio_size)
1374 device->peer_max_bio_size = peer;
1376 new = min(local, peer);
1378 if (device->state.role == R_PRIMARY && new < now)
1379 drbd_err(device, "ASSERT FAILED new < now; (%u < %u)\n", new, now);
1382 drbd_info(device, "max BIO size = %u\n", new);
1384 drbd_setup_queue_param(device, bdev, new, o);
1387 /* Starts the worker thread */
1388 static void conn_reconfig_start(struct drbd_connection *connection)
1390 drbd_thread_start(&connection->worker);
1391 drbd_flush_workqueue(&connection->sender_work);
1394 /* if still unconfigured, stops worker again. */
1395 static void conn_reconfig_done(struct drbd_connection *connection)
1398 spin_lock_irq(&connection->resource->req_lock);
1399 stop_threads = conn_all_vols_unconf(connection) &&
1400 connection->cstate == C_STANDALONE;
1401 spin_unlock_irq(&connection->resource->req_lock);
1403 /* ack_receiver thread and ack_sender workqueue are implicitly
1404 * stopped by receiver in conn_disconnect() */
1405 drbd_thread_stop(&connection->receiver);
1406 drbd_thread_stop(&connection->worker);
1410 /* Make sure IO is suspended before calling this function(). */
1411 static void drbd_suspend_al(struct drbd_device *device)
1415 if (!lc_try_lock(device->act_log)) {
1416 drbd_warn(device, "Failed to lock al in drbd_suspend_al()\n");
1420 drbd_al_shrink(device);
1421 spin_lock_irq(&device->resource->req_lock);
1422 if (device->state.conn < C_CONNECTED)
1423 s = !test_and_set_bit(AL_SUSPENDED, &device->flags);
1424 spin_unlock_irq(&device->resource->req_lock);
1425 lc_unlock(device->act_log);
1428 drbd_info(device, "Suspended AL updates\n");
1432 static bool should_set_defaults(struct genl_info *info)
1434 unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1435 return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1438 static unsigned int drbd_al_extents_max(struct drbd_backing_dev *bdev)
1440 /* This is limited by 16 bit "slot" numbers,
1441 * and by available on-disk context storage.
1443 * Also (u16)~0 is special (denotes a "free" extent).
1445 * One transaction occupies one 4kB on-disk block,
1446 * we have n such blocks in the on disk ring buffer,
1447 * the "current" transaction may fail (n-1),
1448 * and there is 919 slot numbers context information per transaction.
1450 * 72 transaction blocks amounts to more than 2**16 context slots,
1451 * so cap there first.
1453 const unsigned int max_al_nr = DRBD_AL_EXTENTS_MAX;
1454 const unsigned int sufficient_on_disk =
1455 (max_al_nr + AL_CONTEXT_PER_TRANSACTION -1)
1456 /AL_CONTEXT_PER_TRANSACTION;
1458 unsigned int al_size_4k = bdev->md.al_size_4k;
1460 if (al_size_4k > sufficient_on_disk)
1463 return (al_size_4k - 1) * AL_CONTEXT_PER_TRANSACTION;
1466 static bool write_ordering_changed(struct disk_conf *a, struct disk_conf *b)
1468 return a->disk_barrier != b->disk_barrier ||
1469 a->disk_flushes != b->disk_flushes ||
1470 a->disk_drain != b->disk_drain;
1473 static void sanitize_disk_conf(struct drbd_device *device, struct disk_conf *disk_conf,
1474 struct drbd_backing_dev *nbc)
1476 struct request_queue * const q = nbc->backing_bdev->bd_disk->queue;
1478 if (disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1479 disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1480 if (disk_conf->al_extents > drbd_al_extents_max(nbc))
1481 disk_conf->al_extents = drbd_al_extents_max(nbc);
1483 if (!blk_queue_discard(q)) {
1484 if (disk_conf->rs_discard_granularity) {
1485 disk_conf->rs_discard_granularity = 0; /* disable feature */
1486 drbd_info(device, "rs_discard_granularity feature disabled\n");
1490 if (disk_conf->rs_discard_granularity) {
1491 int orig_value = disk_conf->rs_discard_granularity;
1494 if (q->limits.discard_granularity > disk_conf->rs_discard_granularity)
1495 disk_conf->rs_discard_granularity = q->limits.discard_granularity;
1497 remainder = disk_conf->rs_discard_granularity % q->limits.discard_granularity;
1498 disk_conf->rs_discard_granularity += remainder;
1500 if (disk_conf->rs_discard_granularity > q->limits.max_discard_sectors << 9)
1501 disk_conf->rs_discard_granularity = q->limits.max_discard_sectors << 9;
1503 if (disk_conf->rs_discard_granularity != orig_value)
1504 drbd_info(device, "rs_discard_granularity changed to %d\n",
1505 disk_conf->rs_discard_granularity);
1509 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1511 struct drbd_config_context adm_ctx;
1512 enum drbd_ret_code retcode;
1513 struct drbd_device *device;
1514 struct disk_conf *new_disk_conf, *old_disk_conf;
1515 struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1518 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1519 if (!adm_ctx.reply_skb)
1521 if (retcode != NO_ERROR)
1524 device = adm_ctx.device;
1525 mutex_lock(&adm_ctx.resource->adm_mutex);
1527 /* we also need a disk
1528 * to change the options on */
1529 if (!get_ldev(device)) {
1530 retcode = ERR_NO_DISK;
1534 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1535 if (!new_disk_conf) {
1536 retcode = ERR_NOMEM;
1540 mutex_lock(&device->resource->conf_update);
1541 old_disk_conf = device->ldev->disk_conf;
1542 *new_disk_conf = *old_disk_conf;
1543 if (should_set_defaults(info))
1544 set_disk_conf_defaults(new_disk_conf);
1546 err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1547 if (err && err != -ENOMSG) {
1548 retcode = ERR_MANDATORY_TAG;
1549 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1553 if (!expect(new_disk_conf->resync_rate >= 1))
1554 new_disk_conf->resync_rate = 1;
1556 sanitize_disk_conf(device, new_disk_conf, device->ldev);
1558 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1559 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1561 fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1562 if (fifo_size != device->rs_plan_s->size) {
1563 new_plan = fifo_alloc(fifo_size);
1565 drbd_err(device, "kmalloc of fifo_buffer failed");
1566 retcode = ERR_NOMEM;
1571 drbd_suspend_io(device);
1572 wait_event(device->al_wait, lc_try_lock(device->act_log));
1573 drbd_al_shrink(device);
1574 err = drbd_check_al_size(device, new_disk_conf);
1575 lc_unlock(device->act_log);
1576 wake_up(&device->al_wait);
1577 drbd_resume_io(device);
1580 retcode = ERR_NOMEM;
1584 lock_all_resources();
1585 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1586 if (retcode == NO_ERROR) {
1587 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
1588 drbd_resync_after_changed(device);
1590 unlock_all_resources();
1592 if (retcode != NO_ERROR)
1596 old_plan = device->rs_plan_s;
1597 rcu_assign_pointer(device->rs_plan_s, new_plan);
1600 mutex_unlock(&device->resource->conf_update);
1602 if (new_disk_conf->al_updates)
1603 device->ldev->md.flags &= ~MDF_AL_DISABLED;
1605 device->ldev->md.flags |= MDF_AL_DISABLED;
1607 if (new_disk_conf->md_flushes)
1608 clear_bit(MD_NO_FUA, &device->flags);
1610 set_bit(MD_NO_FUA, &device->flags);
1612 if (write_ordering_changed(old_disk_conf, new_disk_conf))
1613 drbd_bump_write_ordering(device->resource, NULL, WO_BDEV_FLUSH);
1615 if (old_disk_conf->discard_zeroes_if_aligned != new_disk_conf->discard_zeroes_if_aligned)
1616 drbd_reconsider_queue_parameters(device, device->ldev, NULL);
1618 drbd_md_sync(device);
1620 if (device->state.conn >= C_CONNECTED) {
1621 struct drbd_peer_device *peer_device;
1623 for_each_peer_device(peer_device, device)
1624 drbd_send_sync_param(peer_device);
1628 kfree(old_disk_conf);
1630 mod_timer(&device->request_timer, jiffies + HZ);
1634 mutex_unlock(&device->resource->conf_update);
1636 kfree(new_disk_conf);
1641 mutex_unlock(&adm_ctx.resource->adm_mutex);
1643 drbd_adm_finish(&adm_ctx, info, retcode);
1647 static struct block_device *open_backing_dev(struct drbd_device *device,
1648 const char *bdev_path, void *claim_ptr, bool do_bd_link)
1650 struct block_device *bdev;
1653 bdev = blkdev_get_by_path(bdev_path,
1654 FMODE_READ | FMODE_WRITE | FMODE_EXCL, claim_ptr);
1656 drbd_err(device, "open(\"%s\") failed with %ld\n",
1657 bdev_path, PTR_ERR(bdev));
1664 err = bd_link_disk_holder(bdev, device->vdisk);
1666 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1667 drbd_err(device, "bd_link_disk_holder(\"%s\", ...) failed with %d\n",
1669 bdev = ERR_PTR(err);
1674 static int open_backing_devices(struct drbd_device *device,
1675 struct disk_conf *new_disk_conf,
1676 struct drbd_backing_dev *nbc)
1678 struct block_device *bdev;
1680 bdev = open_backing_dev(device, new_disk_conf->backing_dev, device, true);
1682 return ERR_OPEN_DISK;
1683 nbc->backing_bdev = bdev;
1686 * meta_dev_idx >= 0: external fixed size, possibly multiple
1687 * drbd sharing one meta device. TODO in that case, paranoia
1688 * check that [md_bdev, meta_dev_idx] is not yet used by some
1689 * other drbd minor! (if you use drbd.conf + drbdadm, that
1690 * should check it for you already; but if you don't, or
1691 * someone fooled it, we need to double check here)
1693 bdev = open_backing_dev(device, new_disk_conf->meta_dev,
1694 /* claim ptr: device, if claimed exclusively; shared drbd_m_holder,
1695 * if potentially shared with other drbd minors */
1696 (new_disk_conf->meta_dev_idx < 0) ? (void*)device : (void*)drbd_m_holder,
1697 /* avoid double bd_claim_by_disk() for the same (source,target) tuple,
1698 * as would happen with internal metadata. */
1699 (new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_FLEX_INT &&
1700 new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_INTERNAL));
1702 return ERR_OPEN_MD_DISK;
1703 nbc->md_bdev = bdev;
1707 static void close_backing_dev(struct drbd_device *device, struct block_device *bdev,
1713 bd_unlink_disk_holder(bdev, device->vdisk);
1714 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1717 void drbd_backing_dev_free(struct drbd_device *device, struct drbd_backing_dev *ldev)
1722 close_backing_dev(device, ldev->md_bdev, ldev->md_bdev != ldev->backing_bdev);
1723 close_backing_dev(device, ldev->backing_bdev, true);
1725 kfree(ldev->disk_conf);
1729 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1731 struct drbd_config_context adm_ctx;
1732 struct drbd_device *device;
1733 struct drbd_peer_device *peer_device;
1734 struct drbd_connection *connection;
1736 enum drbd_ret_code retcode;
1737 enum determine_dev_size dd;
1738 sector_t max_possible_sectors;
1739 sector_t min_md_device_sectors;
1740 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1741 struct disk_conf *new_disk_conf = NULL;
1742 struct lru_cache *resync_lru = NULL;
1743 struct fifo_buffer *new_plan = NULL;
1744 union drbd_state ns, os;
1745 enum drbd_state_rv rv;
1746 struct net_conf *nc;
1748 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1749 if (!adm_ctx.reply_skb)
1751 if (retcode != NO_ERROR)
1754 device = adm_ctx.device;
1755 mutex_lock(&adm_ctx.resource->adm_mutex);
1756 peer_device = first_peer_device(device);
1757 connection = peer_device->connection;
1758 conn_reconfig_start(connection);
1760 /* if you want to reconfigure, please tear down first */
1761 if (device->state.disk > D_DISKLESS) {
1762 retcode = ERR_DISK_CONFIGURED;
1765 /* It may just now have detached because of IO error. Make sure
1766 * drbd_ldev_destroy is done already, we may end up here very fast,
1767 * e.g. if someone calls attach from the on-io-error handler,
1768 * to realize a "hot spare" feature (not that I'd recommend that) */
1769 wait_event(device->misc_wait, !test_bit(GOING_DISKLESS, &device->flags));
1771 /* make sure there is no leftover from previous force-detach attempts */
1772 clear_bit(FORCE_DETACH, &device->flags);
1773 clear_bit(WAS_IO_ERROR, &device->flags);
1774 clear_bit(WAS_READ_ERROR, &device->flags);
1776 /* and no leftover from previously aborted resync or verify, either */
1777 device->rs_total = 0;
1778 device->rs_failed = 0;
1779 atomic_set(&device->rs_pending_cnt, 0);
1781 /* allocation not in the IO path, drbdsetup context */
1782 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1784 retcode = ERR_NOMEM;
1787 spin_lock_init(&nbc->md.uuid_lock);
1789 new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1790 if (!new_disk_conf) {
1791 retcode = ERR_NOMEM;
1794 nbc->disk_conf = new_disk_conf;
1796 set_disk_conf_defaults(new_disk_conf);
1797 err = disk_conf_from_attrs(new_disk_conf, info);
1799 retcode = ERR_MANDATORY_TAG;
1800 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1804 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1805 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1807 new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1809 retcode = ERR_NOMEM;
1813 if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1814 retcode = ERR_MD_IDX_INVALID;
1819 nc = rcu_dereference(connection->net_conf);
1821 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1823 retcode = ERR_STONITH_AND_PROT_A;
1829 retcode = open_backing_devices(device, new_disk_conf, nbc);
1830 if (retcode != NO_ERROR)
1833 if ((nbc->backing_bdev == nbc->md_bdev) !=
1834 (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1835 new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1836 retcode = ERR_MD_IDX_INVALID;
1840 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1841 1, 61, sizeof(struct bm_extent),
1842 offsetof(struct bm_extent, lce));
1844 retcode = ERR_NOMEM;
1848 /* Read our meta data super block early.
1849 * This also sets other on-disk offsets. */
1850 retcode = drbd_md_read(device, nbc);
1851 if (retcode != NO_ERROR)
1854 sanitize_disk_conf(device, new_disk_conf, nbc);
1856 if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1857 drbd_err(device, "max capacity %llu smaller than disk size %llu\n",
1858 (unsigned long long) drbd_get_max_capacity(nbc),
1859 (unsigned long long) new_disk_conf->disk_size);
1860 retcode = ERR_DISK_TOO_SMALL;
1864 if (new_disk_conf->meta_dev_idx < 0) {
1865 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1866 /* at least one MB, otherwise it does not make sense */
1867 min_md_device_sectors = (2<<10);
1869 max_possible_sectors = DRBD_MAX_SECTORS;
1870 min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1);
1873 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1874 retcode = ERR_MD_DISK_TOO_SMALL;
1875 drbd_warn(device, "refusing attach: md-device too small, "
1876 "at least %llu sectors needed for this meta-disk type\n",
1877 (unsigned long long) min_md_device_sectors);
1881 /* Make sure the new disk is big enough
1882 * (we may currently be R_PRIMARY with no local disk...) */
1883 if (drbd_get_max_capacity(nbc) <
1884 drbd_get_capacity(device->this_bdev)) {
1885 retcode = ERR_DISK_TOO_SMALL;
1889 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1891 if (nbc->known_size > max_possible_sectors) {
1892 drbd_warn(device, "==> truncating very big lower level device "
1893 "to currently maximum possible %llu sectors <==\n",
1894 (unsigned long long) max_possible_sectors);
1895 if (new_disk_conf->meta_dev_idx >= 0)
1896 drbd_warn(device, "==>> using internal or flexible "
1897 "meta data may help <<==\n");
1900 drbd_suspend_io(device);
1901 /* also wait for the last barrier ack. */
1902 /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1903 * We need a way to either ignore barrier acks for barriers sent before a device
1904 * was attached, or a way to wait for all pending barrier acks to come in.
1905 * As barriers are counted per resource,
1906 * we'd need to suspend io on all devices of a resource.
1908 wait_event(device->misc_wait, !atomic_read(&device->ap_pending_cnt) || drbd_suspended(device));
1909 /* and for any other previously queued work */
1910 drbd_flush_workqueue(&connection->sender_work);
1912 rv = _drbd_request_state(device, NS(disk, D_ATTACHING), CS_VERBOSE);
1913 retcode = rv; /* FIXME: Type mismatch. */
1914 drbd_resume_io(device);
1915 if (rv < SS_SUCCESS)
1918 if (!get_ldev_if_state(device, D_ATTACHING))
1919 goto force_diskless;
1921 if (!device->bitmap) {
1922 if (drbd_bm_init(device)) {
1923 retcode = ERR_NOMEM;
1924 goto force_diskless_dec;
1928 if (device->state.conn < C_CONNECTED &&
1929 device->state.role == R_PRIMARY && device->ed_uuid &&
1930 (device->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1931 drbd_err(device, "Can only attach to data with current UUID=%016llX\n",
1932 (unsigned long long)device->ed_uuid);
1933 retcode = ERR_DATA_NOT_CURRENT;
1934 goto force_diskless_dec;
1937 /* Since we are diskless, fix the activity log first... */
1938 if (drbd_check_al_size(device, new_disk_conf)) {
1939 retcode = ERR_NOMEM;
1940 goto force_diskless_dec;
1943 /* Prevent shrinking of consistent devices ! */
1944 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1945 drbd_new_dev_size(device, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1946 drbd_warn(device, "refusing to truncate a consistent device\n");
1947 retcode = ERR_DISK_TOO_SMALL;
1948 goto force_diskless_dec;
1951 lock_all_resources();
1952 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1953 if (retcode != NO_ERROR) {
1954 unlock_all_resources();
1955 goto force_diskless_dec;
1958 /* Reset the "barriers don't work" bits here, then force meta data to
1959 * be written, to ensure we determine if barriers are supported. */
1960 if (new_disk_conf->md_flushes)
1961 clear_bit(MD_NO_FUA, &device->flags);
1963 set_bit(MD_NO_FUA, &device->flags);
1965 /* Point of no return reached.
1966 * Devices and memory are no longer released by error cleanup below.
1967 * now device takes over responsibility, and the state engine should
1968 * clean it up somewhere. */
1969 D_ASSERT(device, device->ldev == NULL);
1971 device->resync = resync_lru;
1972 device->rs_plan_s = new_plan;
1975 new_disk_conf = NULL;
1978 drbd_resync_after_changed(device);
1979 drbd_bump_write_ordering(device->resource, device->ldev, WO_BDEV_FLUSH);
1980 unlock_all_resources();
1982 if (drbd_md_test_flag(device->ldev, MDF_CRASHED_PRIMARY))
1983 set_bit(CRASHED_PRIMARY, &device->flags);
1985 clear_bit(CRASHED_PRIMARY, &device->flags);
1987 if (drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
1988 !(device->state.role == R_PRIMARY && device->resource->susp_nod))
1989 set_bit(CRASHED_PRIMARY, &device->flags);
1991 device->send_cnt = 0;
1992 device->recv_cnt = 0;
1993 device->read_cnt = 0;
1994 device->writ_cnt = 0;
1996 drbd_reconsider_queue_parameters(device, device->ldev, NULL);
1998 /* If I am currently not R_PRIMARY,
1999 * but meta data primary indicator is set,
2000 * I just now recover from a hard crash,
2001 * and have been R_PRIMARY before that crash.
2003 * Now, if I had no connection before that crash
2004 * (have been degraded R_PRIMARY), chances are that
2005 * I won't find my peer now either.
2007 * In that case, and _only_ in that case,
2008 * we use the degr-wfc-timeout instead of the default,
2009 * so we can automatically recover from a crash of a
2010 * degraded but active "cluster" after a certain timeout.
2012 clear_bit(USE_DEGR_WFC_T, &device->flags);
2013 if (device->state.role != R_PRIMARY &&
2014 drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
2015 !drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND))
2016 set_bit(USE_DEGR_WFC_T, &device->flags);
2018 dd = drbd_determine_dev_size(device, 0, NULL);
2019 if (dd <= DS_ERROR) {
2020 retcode = ERR_NOMEM_BITMAP;
2021 goto force_diskless_dec;
2022 } else if (dd == DS_GREW)
2023 set_bit(RESYNC_AFTER_NEG, &device->flags);
2025 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ||
2026 (test_bit(CRASHED_PRIMARY, &device->flags) &&
2027 drbd_md_test_flag(device->ldev, MDF_AL_DISABLED))) {
2028 drbd_info(device, "Assuming that all blocks are out of sync "
2029 "(aka FullSync)\n");
2030 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
2031 "set_n_write from attaching", BM_LOCKED_MASK)) {
2032 retcode = ERR_IO_MD_DISK;
2033 goto force_diskless_dec;
2036 if (drbd_bitmap_io(device, &drbd_bm_read,
2037 "read from attaching", BM_LOCKED_MASK)) {
2038 retcode = ERR_IO_MD_DISK;
2039 goto force_diskless_dec;
2043 if (_drbd_bm_total_weight(device) == drbd_bm_bits(device))
2044 drbd_suspend_al(device); /* IO is still suspended here... */
2046 spin_lock_irq(&device->resource->req_lock);
2047 os = drbd_read_state(device);
2049 /* If MDF_CONSISTENT is not set go into inconsistent state,
2050 otherwise investigate MDF_WasUpToDate...
2051 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
2052 otherwise into D_CONSISTENT state.
2054 if (drbd_md_test_flag(device->ldev, MDF_CONSISTENT)) {
2055 if (drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE))
2056 ns.disk = D_CONSISTENT;
2058 ns.disk = D_OUTDATED;
2060 ns.disk = D_INCONSISTENT;
2063 if (drbd_md_test_flag(device->ldev, MDF_PEER_OUT_DATED))
2064 ns.pdsk = D_OUTDATED;
2067 if (ns.disk == D_CONSISTENT &&
2068 (ns.pdsk == D_OUTDATED || rcu_dereference(device->ldev->disk_conf)->fencing == FP_DONT_CARE))
2069 ns.disk = D_UP_TO_DATE;
2071 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
2072 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
2073 this point, because drbd_request_state() modifies these
2076 if (rcu_dereference(device->ldev->disk_conf)->al_updates)
2077 device->ldev->md.flags &= ~MDF_AL_DISABLED;
2079 device->ldev->md.flags |= MDF_AL_DISABLED;
2083 /* In case we are C_CONNECTED postpone any decision on the new disk
2084 state after the negotiation phase. */
2085 if (device->state.conn == C_CONNECTED) {
2086 device->new_state_tmp.i = ns.i;
2088 ns.disk = D_NEGOTIATING;
2090 /* We expect to receive up-to-date UUIDs soon.
2091 To avoid a race in receive_state, free p_uuid while
2092 holding req_lock. I.e. atomic with the state change */
2093 kfree(device->p_uuid);
2094 device->p_uuid = NULL;
2097 rv = _drbd_set_state(device, ns, CS_VERBOSE, NULL);
2098 spin_unlock_irq(&device->resource->req_lock);
2100 if (rv < SS_SUCCESS)
2101 goto force_diskless_dec;
2103 mod_timer(&device->request_timer, jiffies + HZ);
2105 if (device->state.role == R_PRIMARY)
2106 device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
2108 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
2110 drbd_md_mark_dirty(device);
2111 drbd_md_sync(device);
2113 kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
2115 conn_reconfig_done(connection);
2116 mutex_unlock(&adm_ctx.resource->adm_mutex);
2117 drbd_adm_finish(&adm_ctx, info, retcode);
2123 drbd_force_state(device, NS(disk, D_DISKLESS));
2124 drbd_md_sync(device);
2126 conn_reconfig_done(connection);
2128 close_backing_dev(device, nbc->md_bdev, nbc->md_bdev != nbc->backing_bdev);
2129 close_backing_dev(device, nbc->backing_bdev, true);
2132 kfree(new_disk_conf);
2133 lc_destroy(resync_lru);
2135 mutex_unlock(&adm_ctx.resource->adm_mutex);
2137 drbd_adm_finish(&adm_ctx, info, retcode);
2141 static int adm_detach(struct drbd_device *device, int force)
2143 enum drbd_state_rv retcode;
2148 set_bit(FORCE_DETACH, &device->flags);
2149 drbd_force_state(device, NS(disk, D_FAILED));
2150 retcode = SS_SUCCESS;
2154 drbd_suspend_io(device); /* so no-one is stuck in drbd_al_begin_io */
2155 buffer = drbd_md_get_buffer(device, __func__); /* make sure there is no in-flight meta-data IO */
2157 retcode = drbd_request_state(device, NS(disk, D_FAILED));
2158 drbd_md_put_buffer(device);
2159 } else /* already <= D_FAILED */
2160 retcode = SS_NOTHING_TO_DO;
2161 /* D_FAILED will transition to DISKLESS. */
2162 drbd_resume_io(device);
2163 ret = wait_event_interruptible(device->misc_wait,
2164 device->state.disk != D_FAILED);
2165 if ((int)retcode == (int)SS_IS_DISKLESS)
2166 retcode = SS_NOTHING_TO_DO;
2173 /* Detaching the disk is a process in multiple stages. First we need to lock
2174 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
2175 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
2176 * internal references as well.
2177 * Only then we have finally detached. */
2178 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
2180 struct drbd_config_context adm_ctx;
2181 enum drbd_ret_code retcode;
2182 struct detach_parms parms = { };
2185 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2186 if (!adm_ctx.reply_skb)
2188 if (retcode != NO_ERROR)
2191 if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
2192 err = detach_parms_from_attrs(&parms, info);
2194 retcode = ERR_MANDATORY_TAG;
2195 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2200 mutex_lock(&adm_ctx.resource->adm_mutex);
2201 retcode = adm_detach(adm_ctx.device, parms.force_detach);
2202 mutex_unlock(&adm_ctx.resource->adm_mutex);
2204 drbd_adm_finish(&adm_ctx, info, retcode);
2208 static bool conn_resync_running(struct drbd_connection *connection)
2210 struct drbd_peer_device *peer_device;
2215 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2216 struct drbd_device *device = peer_device->device;
2217 if (device->state.conn == C_SYNC_SOURCE ||
2218 device->state.conn == C_SYNC_TARGET ||
2219 device->state.conn == C_PAUSED_SYNC_S ||
2220 device->state.conn == C_PAUSED_SYNC_T) {
2230 static bool conn_ov_running(struct drbd_connection *connection)
2232 struct drbd_peer_device *peer_device;
2237 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2238 struct drbd_device *device = peer_device->device;
2239 if (device->state.conn == C_VERIFY_S ||
2240 device->state.conn == C_VERIFY_T) {
2250 static enum drbd_ret_code
2251 _check_net_options(struct drbd_connection *connection, struct net_conf *old_net_conf, struct net_conf *new_net_conf)
2253 struct drbd_peer_device *peer_device;
2256 if (old_net_conf && connection->cstate == C_WF_REPORT_PARAMS && connection->agreed_pro_version < 100) {
2257 if (new_net_conf->wire_protocol != old_net_conf->wire_protocol)
2258 return ERR_NEED_APV_100;
2260 if (new_net_conf->two_primaries != old_net_conf->two_primaries)
2261 return ERR_NEED_APV_100;
2263 if (strcmp(new_net_conf->integrity_alg, old_net_conf->integrity_alg))
2264 return ERR_NEED_APV_100;
2267 if (!new_net_conf->two_primaries &&
2268 conn_highest_role(connection) == R_PRIMARY &&
2269 conn_highest_peer(connection) == R_PRIMARY)
2270 return ERR_NEED_ALLOW_TWO_PRI;
2272 if (new_net_conf->two_primaries &&
2273 (new_net_conf->wire_protocol != DRBD_PROT_C))
2274 return ERR_NOT_PROTO_C;
2276 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2277 struct drbd_device *device = peer_device->device;
2278 if (get_ldev(device)) {
2279 enum drbd_fencing_p fp = rcu_dereference(device->ldev->disk_conf)->fencing;
2281 if (new_net_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
2282 return ERR_STONITH_AND_PROT_A;
2284 if (device->state.role == R_PRIMARY && new_net_conf->discard_my_data)
2285 return ERR_DISCARD_IMPOSSIBLE;
2288 if (new_net_conf->on_congestion != OC_BLOCK && new_net_conf->wire_protocol != DRBD_PROT_A)
2289 return ERR_CONG_NOT_PROTO_A;
2294 static enum drbd_ret_code
2295 check_net_options(struct drbd_connection *connection, struct net_conf *new_net_conf)
2297 enum drbd_ret_code rv;
2298 struct drbd_peer_device *peer_device;
2302 rv = _check_net_options(connection, rcu_dereference(connection->net_conf), new_net_conf);
2305 /* connection->peer_devices protected by genl_lock() here */
2306 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2307 struct drbd_device *device = peer_device->device;
2308 if (!device->bitmap) {
2309 if (drbd_bm_init(device))
2318 struct crypto_ahash *verify_tfm;
2319 struct crypto_ahash *csums_tfm;
2320 struct crypto_shash *cram_hmac_tfm;
2321 struct crypto_ahash *integrity_tfm;
2325 alloc_shash(struct crypto_shash **tfm, char *tfm_name, int err_alg)
2330 *tfm = crypto_alloc_shash(tfm_name, 0, 0);
2340 alloc_ahash(struct crypto_ahash **tfm, char *tfm_name, int err_alg)
2345 *tfm = crypto_alloc_ahash(tfm_name, 0, CRYPTO_ALG_ASYNC);
2354 static enum drbd_ret_code
2355 alloc_crypto(struct crypto *crypto, struct net_conf *new_net_conf)
2357 char hmac_name[CRYPTO_MAX_ALG_NAME];
2358 enum drbd_ret_code rv;
2360 rv = alloc_ahash(&crypto->csums_tfm, new_net_conf->csums_alg,
2364 rv = alloc_ahash(&crypto->verify_tfm, new_net_conf->verify_alg,
2368 rv = alloc_ahash(&crypto->integrity_tfm, new_net_conf->integrity_alg,
2372 if (new_net_conf->cram_hmac_alg[0] != 0) {
2373 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
2374 new_net_conf->cram_hmac_alg);
2376 rv = alloc_shash(&crypto->cram_hmac_tfm, hmac_name,
2383 static void free_crypto(struct crypto *crypto)
2385 crypto_free_shash(crypto->cram_hmac_tfm);
2386 crypto_free_ahash(crypto->integrity_tfm);
2387 crypto_free_ahash(crypto->csums_tfm);
2388 crypto_free_ahash(crypto->verify_tfm);
2391 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
2393 struct drbd_config_context adm_ctx;
2394 enum drbd_ret_code retcode;
2395 struct drbd_connection *connection;
2396 struct net_conf *old_net_conf, *new_net_conf = NULL;
2398 int ovr; /* online verify running */
2399 int rsr; /* re-sync running */
2400 struct crypto crypto = { };
2402 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2403 if (!adm_ctx.reply_skb)
2405 if (retcode != NO_ERROR)
2408 connection = adm_ctx.connection;
2409 mutex_lock(&adm_ctx.resource->adm_mutex);
2411 new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
2412 if (!new_net_conf) {
2413 retcode = ERR_NOMEM;
2417 conn_reconfig_start(connection);
2419 mutex_lock(&connection->data.mutex);
2420 mutex_lock(&connection->resource->conf_update);
2421 old_net_conf = connection->net_conf;
2423 if (!old_net_conf) {
2424 drbd_msg_put_info(adm_ctx.reply_skb, "net conf missing, try connect");
2425 retcode = ERR_INVALID_REQUEST;
2429 *new_net_conf = *old_net_conf;
2430 if (should_set_defaults(info))
2431 set_net_conf_defaults(new_net_conf);
2433 err = net_conf_from_attrs_for_change(new_net_conf, info);
2434 if (err && err != -ENOMSG) {
2435 retcode = ERR_MANDATORY_TAG;
2436 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2440 retcode = check_net_options(connection, new_net_conf);
2441 if (retcode != NO_ERROR)
2444 /* re-sync running */
2445 rsr = conn_resync_running(connection);
2446 if (rsr && strcmp(new_net_conf->csums_alg, old_net_conf->csums_alg)) {
2447 retcode = ERR_CSUMS_RESYNC_RUNNING;
2451 /* online verify running */
2452 ovr = conn_ov_running(connection);
2453 if (ovr && strcmp(new_net_conf->verify_alg, old_net_conf->verify_alg)) {
2454 retcode = ERR_VERIFY_RUNNING;
2458 retcode = alloc_crypto(&crypto, new_net_conf);
2459 if (retcode != NO_ERROR)
2462 rcu_assign_pointer(connection->net_conf, new_net_conf);
2465 crypto_free_ahash(connection->csums_tfm);
2466 connection->csums_tfm = crypto.csums_tfm;
2467 crypto.csums_tfm = NULL;
2470 crypto_free_ahash(connection->verify_tfm);
2471 connection->verify_tfm = crypto.verify_tfm;
2472 crypto.verify_tfm = NULL;
2475 crypto_free_ahash(connection->integrity_tfm);
2476 connection->integrity_tfm = crypto.integrity_tfm;
2477 if (connection->cstate >= C_WF_REPORT_PARAMS && connection->agreed_pro_version >= 100)
2478 /* Do this without trying to take connection->data.mutex again. */
2479 __drbd_send_protocol(connection, P_PROTOCOL_UPDATE);
2481 crypto_free_shash(connection->cram_hmac_tfm);
2482 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2484 mutex_unlock(&connection->resource->conf_update);
2485 mutex_unlock(&connection->data.mutex);
2487 kfree(old_net_conf);
2489 if (connection->cstate >= C_WF_REPORT_PARAMS) {
2490 struct drbd_peer_device *peer_device;
2493 idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
2494 drbd_send_sync_param(peer_device);
2500 mutex_unlock(&connection->resource->conf_update);
2501 mutex_unlock(&connection->data.mutex);
2502 free_crypto(&crypto);
2503 kfree(new_net_conf);
2505 conn_reconfig_done(connection);
2507 mutex_unlock(&adm_ctx.resource->adm_mutex);
2509 drbd_adm_finish(&adm_ctx, info, retcode);
2513 static void connection_to_info(struct connection_info *info,
2514 struct drbd_connection *connection)
2516 info->conn_connection_state = connection->cstate;
2517 info->conn_role = conn_highest_peer(connection);
2520 static void peer_device_to_info(struct peer_device_info *info,
2521 struct drbd_peer_device *peer_device)
2523 struct drbd_device *device = peer_device->device;
2525 info->peer_repl_state =
2526 max_t(enum drbd_conns, C_WF_REPORT_PARAMS, device->state.conn);
2527 info->peer_disk_state = device->state.pdsk;
2528 info->peer_resync_susp_user = device->state.user_isp;
2529 info->peer_resync_susp_peer = device->state.peer_isp;
2530 info->peer_resync_susp_dependency = device->state.aftr_isp;
2533 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2535 struct connection_info connection_info;
2536 enum drbd_notification_type flags;
2537 unsigned int peer_devices = 0;
2538 struct drbd_config_context adm_ctx;
2539 struct drbd_peer_device *peer_device;
2540 struct net_conf *old_net_conf, *new_net_conf = NULL;
2541 struct crypto crypto = { };
2542 struct drbd_resource *resource;
2543 struct drbd_connection *connection;
2544 enum drbd_ret_code retcode;
2548 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2550 if (!adm_ctx.reply_skb)
2552 if (retcode != NO_ERROR)
2554 if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2555 drbd_msg_put_info(adm_ctx.reply_skb, "connection endpoint(s) missing");
2556 retcode = ERR_INVALID_REQUEST;
2560 /* No need for _rcu here. All reconfiguration is
2561 * strictly serialized on genl_lock(). We are protected against
2562 * concurrent reconfiguration/addition/deletion */
2563 for_each_resource(resource, &drbd_resources) {
2564 for_each_connection(connection, resource) {
2565 if (nla_len(adm_ctx.my_addr) == connection->my_addr_len &&
2566 !memcmp(nla_data(adm_ctx.my_addr), &connection->my_addr,
2567 connection->my_addr_len)) {
2568 retcode = ERR_LOCAL_ADDR;
2572 if (nla_len(adm_ctx.peer_addr) == connection->peer_addr_len &&
2573 !memcmp(nla_data(adm_ctx.peer_addr), &connection->peer_addr,
2574 connection->peer_addr_len)) {
2575 retcode = ERR_PEER_ADDR;
2581 mutex_lock(&adm_ctx.resource->adm_mutex);
2582 connection = first_connection(adm_ctx.resource);
2583 conn_reconfig_start(connection);
2585 if (connection->cstate > C_STANDALONE) {
2586 retcode = ERR_NET_CONFIGURED;
2590 /* allocation not in the IO path, drbdsetup / netlink process context */
2591 new_net_conf = kzalloc(sizeof(*new_net_conf), GFP_KERNEL);
2592 if (!new_net_conf) {
2593 retcode = ERR_NOMEM;
2597 set_net_conf_defaults(new_net_conf);
2599 err = net_conf_from_attrs(new_net_conf, info);
2600 if (err && err != -ENOMSG) {
2601 retcode = ERR_MANDATORY_TAG;
2602 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2606 retcode = check_net_options(connection, new_net_conf);
2607 if (retcode != NO_ERROR)
2610 retcode = alloc_crypto(&crypto, new_net_conf);
2611 if (retcode != NO_ERROR)
2614 ((char *)new_net_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2616 drbd_flush_workqueue(&connection->sender_work);
2618 mutex_lock(&adm_ctx.resource->conf_update);
2619 old_net_conf = connection->net_conf;
2621 retcode = ERR_NET_CONFIGURED;
2622 mutex_unlock(&adm_ctx.resource->conf_update);
2625 rcu_assign_pointer(connection->net_conf, new_net_conf);
2627 conn_free_crypto(connection);
2628 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2629 connection->integrity_tfm = crypto.integrity_tfm;
2630 connection->csums_tfm = crypto.csums_tfm;
2631 connection->verify_tfm = crypto.verify_tfm;
2633 connection->my_addr_len = nla_len(adm_ctx.my_addr);
2634 memcpy(&connection->my_addr, nla_data(adm_ctx.my_addr), connection->my_addr_len);
2635 connection->peer_addr_len = nla_len(adm_ctx.peer_addr);
2636 memcpy(&connection->peer_addr, nla_data(adm_ctx.peer_addr), connection->peer_addr_len);
2638 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2642 connection_to_info(&connection_info, connection);
2643 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2644 mutex_lock(¬ification_mutex);
2645 notify_connection_state(NULL, 0, connection, &connection_info, NOTIFY_CREATE | flags);
2646 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2647 struct peer_device_info peer_device_info;
2649 peer_device_to_info(&peer_device_info, peer_device);
2650 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2651 notify_peer_device_state(NULL, 0, peer_device, &peer_device_info, NOTIFY_CREATE | flags);
2653 mutex_unlock(¬ification_mutex);
2654 mutex_unlock(&adm_ctx.resource->conf_update);
2657 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2658 struct drbd_device *device = peer_device->device;
2659 device->send_cnt = 0;
2660 device->recv_cnt = 0;
2664 retcode = conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2666 conn_reconfig_done(connection);
2667 mutex_unlock(&adm_ctx.resource->adm_mutex);
2668 drbd_adm_finish(&adm_ctx, info, retcode);
2672 free_crypto(&crypto);
2673 kfree(new_net_conf);
2675 conn_reconfig_done(connection);
2676 mutex_unlock(&adm_ctx.resource->adm_mutex);
2678 drbd_adm_finish(&adm_ctx, info, retcode);
2682 static enum drbd_state_rv conn_try_disconnect(struct drbd_connection *connection, bool force)
2684 enum drbd_state_rv rv;
2686 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2687 force ? CS_HARD : 0);
2690 case SS_NOTHING_TO_DO:
2692 case SS_ALREADY_STANDALONE:
2694 case SS_PRIMARY_NOP:
2695 /* Our state checking code wants to see the peer outdated. */
2696 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, pdsk, D_OUTDATED), 0);
2698 if (rv == SS_OUTDATE_WO_CONN) /* lost connection before graceful disconnect succeeded */
2699 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_VERBOSE);
2702 case SS_CW_FAILED_BY_PEER:
2703 /* The peer probably wants to see us outdated. */
2704 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING,
2705 disk, D_OUTDATED), 0);
2706 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2707 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2712 /* no special handling necessary */
2715 if (rv >= SS_SUCCESS) {
2716 enum drbd_state_rv rv2;
2717 /* No one else can reconfigure the network while I am here.
2718 * The state handling only uses drbd_thread_stop_nowait(),
2719 * we want to really wait here until the receiver is no more.
2721 drbd_thread_stop(&connection->receiver);
2723 /* Race breaker. This additional state change request may be
2724 * necessary, if this was a forced disconnect during a receiver
2725 * restart. We may have "killed" the receiver thread just
2726 * after drbd_receiver() returned. Typically, we should be
2727 * C_STANDALONE already, now, and this becomes a no-op.
2729 rv2 = conn_request_state(connection, NS(conn, C_STANDALONE),
2730 CS_VERBOSE | CS_HARD);
2731 if (rv2 < SS_SUCCESS)
2732 drbd_err(connection,
2733 "unexpected rv2=%d in conn_try_disconnect()\n",
2735 /* Unlike in DRBD 9, the state engine has generated
2736 * NOTIFY_DESTROY events before clearing connection->net_conf. */
2741 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2743 struct drbd_config_context adm_ctx;
2744 struct disconnect_parms parms;
2745 struct drbd_connection *connection;
2746 enum drbd_state_rv rv;
2747 enum drbd_ret_code retcode;
2750 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2751 if (!adm_ctx.reply_skb)
2753 if (retcode != NO_ERROR)
2756 connection = adm_ctx.connection;
2757 memset(&parms, 0, sizeof(parms));
2758 if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2759 err = disconnect_parms_from_attrs(&parms, info);
2761 retcode = ERR_MANDATORY_TAG;
2762 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2767 mutex_lock(&adm_ctx.resource->adm_mutex);
2768 rv = conn_try_disconnect(connection, parms.force_disconnect);
2769 if (rv < SS_SUCCESS)
2770 retcode = rv; /* FIXME: Type mismatch. */
2773 mutex_unlock(&adm_ctx.resource->adm_mutex);
2775 drbd_adm_finish(&adm_ctx, info, retcode);
2779 void resync_after_online_grow(struct drbd_device *device)
2781 int iass; /* I am sync source */
2783 drbd_info(device, "Resync of new storage after online grow\n");
2784 if (device->state.role != device->state.peer)
2785 iass = (device->state.role == R_PRIMARY);
2787 iass = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags);
2790 drbd_start_resync(device, C_SYNC_SOURCE);
2792 _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2795 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2797 struct drbd_config_context adm_ctx;
2798 struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2799 struct resize_parms rs;
2800 struct drbd_device *device;
2801 enum drbd_ret_code retcode;
2802 enum determine_dev_size dd;
2803 bool change_al_layout = false;
2804 enum dds_flags ddsf;
2808 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2809 if (!adm_ctx.reply_skb)
2811 if (retcode != NO_ERROR)
2814 mutex_lock(&adm_ctx.resource->adm_mutex);
2815 device = adm_ctx.device;
2816 if (!get_ldev(device)) {
2817 retcode = ERR_NO_DISK;
2821 memset(&rs, 0, sizeof(struct resize_parms));
2822 rs.al_stripes = device->ldev->md.al_stripes;
2823 rs.al_stripe_size = device->ldev->md.al_stripe_size_4k * 4;
2824 if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2825 err = resize_parms_from_attrs(&rs, info);
2827 retcode = ERR_MANDATORY_TAG;
2828 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2833 if (device->state.conn > C_CONNECTED) {
2834 retcode = ERR_RESIZE_RESYNC;
2838 if (device->state.role == R_SECONDARY &&
2839 device->state.peer == R_SECONDARY) {
2840 retcode = ERR_NO_PRIMARY;
2844 if (rs.no_resync && first_peer_device(device)->connection->agreed_pro_version < 93) {
2845 retcode = ERR_NEED_APV_93;
2850 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
2852 if (u_size != (sector_t)rs.resize_size) {
2853 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2854 if (!new_disk_conf) {
2855 retcode = ERR_NOMEM;
2860 if (device->ldev->md.al_stripes != rs.al_stripes ||
2861 device->ldev->md.al_stripe_size_4k != rs.al_stripe_size / 4) {
2862 u32 al_size_k = rs.al_stripes * rs.al_stripe_size;
2864 if (al_size_k > (16 * 1024 * 1024)) {
2865 retcode = ERR_MD_LAYOUT_TOO_BIG;
2869 if (al_size_k < MD_32kB_SECT/2) {
2870 retcode = ERR_MD_LAYOUT_TOO_SMALL;
2874 if (device->state.conn != C_CONNECTED && !rs.resize_force) {
2875 retcode = ERR_MD_LAYOUT_CONNECTED;
2879 change_al_layout = true;
2882 if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev))
2883 device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev);
2885 if (new_disk_conf) {
2886 mutex_lock(&device->resource->conf_update);
2887 old_disk_conf = device->ldev->disk_conf;
2888 *new_disk_conf = *old_disk_conf;
2889 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2890 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
2891 mutex_unlock(&device->resource->conf_update);
2893 kfree(old_disk_conf);
2894 new_disk_conf = NULL;
2897 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2898 dd = drbd_determine_dev_size(device, ddsf, change_al_layout ? &rs : NULL);
2899 drbd_md_sync(device);
2901 if (dd == DS_ERROR) {
2902 retcode = ERR_NOMEM_BITMAP;
2904 } else if (dd == DS_ERROR_SPACE_MD) {
2905 retcode = ERR_MD_LAYOUT_NO_FIT;
2907 } else if (dd == DS_ERROR_SHRINK) {
2908 retcode = ERR_IMPLICIT_SHRINK;
2912 if (device->state.conn == C_CONNECTED) {
2914 set_bit(RESIZE_PENDING, &device->flags);
2916 drbd_send_uuids(first_peer_device(device));
2917 drbd_send_sizes(first_peer_device(device), 1, ddsf);
2921 mutex_unlock(&adm_ctx.resource->adm_mutex);
2923 drbd_adm_finish(&adm_ctx, info, retcode);
2928 kfree(new_disk_conf);
2932 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2934 struct drbd_config_context adm_ctx;
2935 enum drbd_ret_code retcode;
2936 struct res_opts res_opts;
2939 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2940 if (!adm_ctx.reply_skb)
2942 if (retcode != NO_ERROR)
2945 res_opts = adm_ctx.resource->res_opts;
2946 if (should_set_defaults(info))
2947 set_res_opts_defaults(&res_opts);
2949 err = res_opts_from_attrs(&res_opts, info);
2950 if (err && err != -ENOMSG) {
2951 retcode = ERR_MANDATORY_TAG;
2952 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2956 mutex_lock(&adm_ctx.resource->adm_mutex);
2957 err = set_resource_options(adm_ctx.resource, &res_opts);
2959 retcode = ERR_INVALID_REQUEST;
2961 retcode = ERR_NOMEM;
2963 mutex_unlock(&adm_ctx.resource->adm_mutex);
2966 drbd_adm_finish(&adm_ctx, info, retcode);
2970 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2972 struct drbd_config_context adm_ctx;
2973 struct drbd_device *device;
2974 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2976 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2977 if (!adm_ctx.reply_skb)
2979 if (retcode != NO_ERROR)
2982 device = adm_ctx.device;
2983 if (!get_ldev(device)) {
2984 retcode = ERR_NO_DISK;
2988 mutex_lock(&adm_ctx.resource->adm_mutex);
2990 /* If there is still bitmap IO pending, probably because of a previous
2991 * resync just being finished, wait for it before requesting a new resync.
2992 * Also wait for it's after_state_ch(). */
2993 drbd_suspend_io(device);
2994 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
2995 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
2997 /* If we happen to be C_STANDALONE R_SECONDARY, just change to
2998 * D_INCONSISTENT, and set all bits in the bitmap. Otherwise,
2999 * try to start a resync handshake as sync target for full sync.
3001 if (device->state.conn == C_STANDALONE && device->state.role == R_SECONDARY) {
3002 retcode = drbd_request_state(device, NS(disk, D_INCONSISTENT));
3003 if (retcode >= SS_SUCCESS) {
3004 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
3005 "set_n_write from invalidate", BM_LOCKED_MASK))
3006 retcode = ERR_IO_MD_DISK;
3009 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_T));
3010 drbd_resume_io(device);
3011 mutex_unlock(&adm_ctx.resource->adm_mutex);
3014 drbd_adm_finish(&adm_ctx, info, retcode);
3018 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
3019 union drbd_state mask, union drbd_state val)
3021 struct drbd_config_context adm_ctx;
3022 enum drbd_ret_code retcode;
3024 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3025 if (!adm_ctx.reply_skb)
3027 if (retcode != NO_ERROR)
3030 mutex_lock(&adm_ctx.resource->adm_mutex);
3031 retcode = drbd_request_state(adm_ctx.device, mask, val);
3032 mutex_unlock(&adm_ctx.resource->adm_mutex);
3034 drbd_adm_finish(&adm_ctx, info, retcode);
3038 static int drbd_bmio_set_susp_al(struct drbd_device *device) __must_hold(local)
3042 rv = drbd_bmio_set_n_write(device);
3043 drbd_suspend_al(device);
3047 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
3049 struct drbd_config_context adm_ctx;
3050 int retcode; /* drbd_ret_code, drbd_state_rv */
3051 struct drbd_device *device;
3053 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3054 if (!adm_ctx.reply_skb)
3056 if (retcode != NO_ERROR)
3059 device = adm_ctx.device;
3060 if (!get_ldev(device)) {
3061 retcode = ERR_NO_DISK;
3065 mutex_lock(&adm_ctx.resource->adm_mutex);
3067 /* If there is still bitmap IO pending, probably because of a previous
3068 * resync just being finished, wait for it before requesting a new resync.
3069 * Also wait for it's after_state_ch(). */
3070 drbd_suspend_io(device);
3071 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
3072 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
3074 /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
3075 * in the bitmap. Otherwise, try to start a resync handshake
3076 * as sync source for full sync.
3078 if (device->state.conn == C_STANDALONE && device->state.role == R_PRIMARY) {
3079 /* The peer will get a resync upon connect anyways. Just make that
3080 into a full resync. */
3081 retcode = drbd_request_state(device, NS(pdsk, D_INCONSISTENT));
3082 if (retcode >= SS_SUCCESS) {
3083 if (drbd_bitmap_io(device, &drbd_bmio_set_susp_al,
3084 "set_n_write from invalidate_peer",
3085 BM_LOCKED_SET_ALLOWED))
3086 retcode = ERR_IO_MD_DISK;
3089 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_S));
3090 drbd_resume_io(device);
3091 mutex_unlock(&adm_ctx.resource->adm_mutex);
3094 drbd_adm_finish(&adm_ctx, info, retcode);
3098 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
3100 struct drbd_config_context adm_ctx;
3101 enum drbd_ret_code retcode;
3103 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3104 if (!adm_ctx.reply_skb)
3106 if (retcode != NO_ERROR)
3109 mutex_lock(&adm_ctx.resource->adm_mutex);
3110 if (drbd_request_state(adm_ctx.device, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
3111 retcode = ERR_PAUSE_IS_SET;
3112 mutex_unlock(&adm_ctx.resource->adm_mutex);
3114 drbd_adm_finish(&adm_ctx, info, retcode);
3118 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
3120 struct drbd_config_context adm_ctx;
3121 union drbd_dev_state s;
3122 enum drbd_ret_code retcode;
3124 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3125 if (!adm_ctx.reply_skb)
3127 if (retcode != NO_ERROR)
3130 mutex_lock(&adm_ctx.resource->adm_mutex);
3131 if (drbd_request_state(adm_ctx.device, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
3132 s = adm_ctx.device->state;
3133 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
3134 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
3135 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
3137 retcode = ERR_PAUSE_IS_CLEAR;
3140 mutex_unlock(&adm_ctx.resource->adm_mutex);
3142 drbd_adm_finish(&adm_ctx, info, retcode);
3146 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
3148 return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
3151 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
3153 struct drbd_config_context adm_ctx;
3154 struct drbd_device *device;
3155 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3157 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3158 if (!adm_ctx.reply_skb)
3160 if (retcode != NO_ERROR)
3163 mutex_lock(&adm_ctx.resource->adm_mutex);
3164 device = adm_ctx.device;
3165 if (test_bit(NEW_CUR_UUID, &device->flags)) {
3166 if (get_ldev_if_state(device, D_ATTACHING)) {
3167 drbd_uuid_new_current(device);
3170 /* This is effectively a multi-stage "forced down".
3171 * The NEW_CUR_UUID bit is supposedly only set, if we
3172 * lost the replication connection, and are configured
3173 * to freeze IO and wait for some fence-peer handler.
3174 * So we still don't have a replication connection.
3175 * And now we don't have a local disk either. After
3176 * resume, we will fail all pending and new IO, because
3177 * we don't have any data anymore. Which means we will
3178 * eventually be able to terminate all users of this
3179 * device, and then take it down. By bumping the
3180 * "effective" data uuid, we make sure that you really
3181 * need to tear down before you reconfigure, we will
3182 * the refuse to re-connect or re-attach (because no
3183 * matching real data uuid exists).
3186 get_random_bytes(&val, sizeof(u64));
3187 drbd_set_ed_uuid(device, val);
3188 drbd_warn(device, "Resumed without access to data; please tear down before attempting to re-configure.\n");
3190 clear_bit(NEW_CUR_UUID, &device->flags);
3192 drbd_suspend_io(device);
3193 retcode = drbd_request_state(device, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
3194 if (retcode == SS_SUCCESS) {
3195 if (device->state.conn < C_CONNECTED)
3196 tl_clear(first_peer_device(device)->connection);
3197 if (device->state.disk == D_DISKLESS || device->state.disk == D_FAILED)
3198 tl_restart(first_peer_device(device)->connection, FAIL_FROZEN_DISK_IO);
3200 drbd_resume_io(device);
3201 mutex_unlock(&adm_ctx.resource->adm_mutex);
3203 drbd_adm_finish(&adm_ctx, info, retcode);
3207 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
3209 return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
3212 static int nla_put_drbd_cfg_context(struct sk_buff *skb,
3213 struct drbd_resource *resource,
3214 struct drbd_connection *connection,
3215 struct drbd_device *device)
3218 nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
3220 goto nla_put_failure;
3222 nla_put_u32(skb, T_ctx_volume, device->vnr))
3223 goto nla_put_failure;
3224 if (nla_put_string(skb, T_ctx_resource_name, resource->name))
3225 goto nla_put_failure;
3227 if (connection->my_addr_len &&
3228 nla_put(skb, T_ctx_my_addr, connection->my_addr_len, &connection->my_addr))
3229 goto nla_put_failure;
3230 if (connection->peer_addr_len &&
3231 nla_put(skb, T_ctx_peer_addr, connection->peer_addr_len, &connection->peer_addr))
3232 goto nla_put_failure;
3234 nla_nest_end(skb, nla);
3239 nla_nest_cancel(skb, nla);
3244 * The generic netlink dump callbacks are called outside the genl_lock(), so
3245 * they cannot use the simple attribute parsing code which uses global
3248 static struct nlattr *find_cfg_context_attr(const struct nlmsghdr *nlh, int attr)
3250 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3251 const int maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
3254 nla = nla_find(nlmsg_attrdata(nlh, hdrlen), nlmsg_attrlen(nlh, hdrlen),
3255 DRBD_NLA_CFG_CONTEXT);
3258 return drbd_nla_find_nested(maxtype, nla, __nla_type(attr));
3261 static void resource_to_info(struct resource_info *, struct drbd_resource *);
3263 int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb)
3265 struct drbd_genlmsghdr *dh;
3266 struct drbd_resource *resource;
3267 struct resource_info resource_info;
3268 struct resource_statistics resource_statistics;
3273 for_each_resource_rcu(resource, &drbd_resources)
3274 if (resource == (struct drbd_resource *)cb->args[0])
3275 goto found_resource;
3276 err = 0; /* resource was probably deleted */
3279 resource = list_entry(&drbd_resources,
3280 struct drbd_resource, resources);
3283 list_for_each_entry_continue_rcu(resource, &drbd_resources, resources) {
3290 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3291 cb->nlh->nlmsg_seq, &drbd_genl_family,
3292 NLM_F_MULTI, DRBD_ADM_GET_RESOURCES);
3297 dh->ret_code = NO_ERROR;
3298 err = nla_put_drbd_cfg_context(skb, resource, NULL, NULL);
3301 err = res_opts_to_skb(skb, &resource->res_opts, !capable(CAP_SYS_ADMIN));
3304 resource_to_info(&resource_info, resource);
3305 err = resource_info_to_skb(skb, &resource_info, !capable(CAP_SYS_ADMIN));
3308 resource_statistics.res_stat_write_ordering = resource->write_ordering;
3309 err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
3312 cb->args[0] = (long)resource;
3313 genlmsg_end(skb, dh);
3323 static void device_to_statistics(struct device_statistics *s,
3324 struct drbd_device *device)
3326 memset(s, 0, sizeof(*s));
3327 s->dev_upper_blocked = !may_inc_ap_bio(device);
3328 if (get_ldev(device)) {
3329 struct drbd_md *md = &device->ldev->md;
3330 u64 *history_uuids = (u64 *)s->history_uuids;
3331 struct request_queue *q;
3334 spin_lock_irq(&md->uuid_lock);
3335 s->dev_current_uuid = md->uuid[UI_CURRENT];
3336 BUILD_BUG_ON(sizeof(s->history_uuids) < UI_HISTORY_END - UI_HISTORY_START + 1);
3337 for (n = 0; n < UI_HISTORY_END - UI_HISTORY_START + 1; n++)
3338 history_uuids[n] = md->uuid[UI_HISTORY_START + n];
3339 for (; n < HISTORY_UUIDS; n++)
3340 history_uuids[n] = 0;
3341 s->history_uuids_len = HISTORY_UUIDS;
3342 spin_unlock_irq(&md->uuid_lock);
3344 s->dev_disk_flags = md->flags;
3345 q = bdev_get_queue(device->ldev->backing_bdev);
3346 s->dev_lower_blocked =
3347 bdi_congested(q->backing_dev_info,
3348 (1 << WB_async_congested) |
3349 (1 << WB_sync_congested));
3352 s->dev_size = drbd_get_capacity(device->this_bdev);
3353 s->dev_read = device->read_cnt;
3354 s->dev_write = device->writ_cnt;
3355 s->dev_al_writes = device->al_writ_cnt;
3356 s->dev_bm_writes = device->bm_writ_cnt;
3357 s->dev_upper_pending = atomic_read(&device->ap_bio_cnt);
3358 s->dev_lower_pending = atomic_read(&device->local_cnt);
3359 s->dev_al_suspended = test_bit(AL_SUSPENDED, &device->flags);
3360 s->dev_exposed_data_uuid = device->ed_uuid;
3363 static int put_resource_in_arg0(struct netlink_callback *cb, int holder_nr)
3366 struct drbd_resource *resource =
3367 (struct drbd_resource *)cb->args[0];
3368 kref_put(&resource->kref, drbd_destroy_resource);
3374 int drbd_adm_dump_devices_done(struct netlink_callback *cb) {
3375 return put_resource_in_arg0(cb, 7);
3378 static void device_to_info(struct device_info *, struct drbd_device *);
3380 int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb)
3382 struct nlattr *resource_filter;
3383 struct drbd_resource *resource;
3384 struct drbd_device *uninitialized_var(device);
3385 int minor, err, retcode;
3386 struct drbd_genlmsghdr *dh;
3387 struct device_info device_info;
3388 struct device_statistics device_statistics;
3389 struct idr *idr_to_search;
3391 resource = (struct drbd_resource *)cb->args[0];
3392 if (!cb->args[0] && !cb->args[1]) {
3393 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3394 if (resource_filter) {
3395 retcode = ERR_RES_NOT_KNOWN;
3396 resource = drbd_find_resource(nla_data(resource_filter));
3399 cb->args[0] = (long)resource;
3404 minor = cb->args[1];
3405 idr_to_search = resource ? &resource->devices : &drbd_devices;
3406 device = idr_get_next(idr_to_search, &minor);
3411 idr_for_each_entry_continue(idr_to_search, device, minor) {
3413 goto put_result; /* only one iteration */
3416 goto out; /* no more devices */
3419 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3420 cb->nlh->nlmsg_seq, &drbd_genl_family,
3421 NLM_F_MULTI, DRBD_ADM_GET_DEVICES);
3425 dh->ret_code = retcode;
3427 if (retcode == NO_ERROR) {
3428 dh->minor = device->minor;
3429 err = nla_put_drbd_cfg_context(skb, device->resource, NULL, device);
3432 if (get_ldev(device)) {
3433 struct disk_conf *disk_conf =
3434 rcu_dereference(device->ldev->disk_conf);
3436 err = disk_conf_to_skb(skb, disk_conf, !capable(CAP_SYS_ADMIN));
3441 device_to_info(&device_info, device);
3442 err = device_info_to_skb(skb, &device_info, !capable(CAP_SYS_ADMIN));
3446 device_to_statistics(&device_statistics, device);
3447 err = device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
3450 cb->args[1] = minor + 1;
3452 genlmsg_end(skb, dh);
3462 int drbd_adm_dump_connections_done(struct netlink_callback *cb)
3464 return put_resource_in_arg0(cb, 6);
3467 enum { SINGLE_RESOURCE, ITERATE_RESOURCES };
3469 int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb)
3471 struct nlattr *resource_filter;
3472 struct drbd_resource *resource = NULL, *next_resource;
3473 struct drbd_connection *uninitialized_var(connection);
3474 int err = 0, retcode;
3475 struct drbd_genlmsghdr *dh;
3476 struct connection_info connection_info;
3477 struct connection_statistics connection_statistics;
3480 resource = (struct drbd_resource *)cb->args[0];
3482 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3483 if (resource_filter) {
3484 retcode = ERR_RES_NOT_KNOWN;
3485 resource = drbd_find_resource(nla_data(resource_filter));
3488 cb->args[0] = (long)resource;
3489 cb->args[1] = SINGLE_RESOURCE;
3493 if (list_empty(&drbd_resources))
3495 resource = list_first_entry(&drbd_resources, struct drbd_resource, resources);
3496 kref_get(&resource->kref);
3497 cb->args[0] = (long)resource;
3498 cb->args[1] = ITERATE_RESOURCES;
3503 mutex_lock(&resource->conf_update);
3506 for_each_connection_rcu(connection, resource)
3507 if (connection == (struct drbd_connection *)cb->args[2])
3508 goto found_connection;
3509 /* connection was probably deleted */
3510 goto no_more_connections;
3512 connection = list_entry(&resource->connections, struct drbd_connection, connections);
3515 list_for_each_entry_continue_rcu(connection, &resource->connections, connections) {
3516 if (!has_net_conf(connection))
3519 goto put_result; /* only one iteration */
3522 no_more_connections:
3523 if (cb->args[1] == ITERATE_RESOURCES) {
3524 for_each_resource_rcu(next_resource, &drbd_resources) {
3525 if (next_resource == resource)
3526 goto found_resource;
3528 /* resource was probably deleted */
3533 list_for_each_entry_continue_rcu(next_resource, &drbd_resources, resources) {
3534 mutex_unlock(&resource->conf_update);
3535 kref_put(&resource->kref, drbd_destroy_resource);
3536 resource = next_resource;
3537 kref_get(&resource->kref);
3538 cb->args[0] = (long)resource;
3542 goto out; /* no more resources */
3545 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3546 cb->nlh->nlmsg_seq, &drbd_genl_family,
3547 NLM_F_MULTI, DRBD_ADM_GET_CONNECTIONS);
3551 dh->ret_code = retcode;
3553 if (retcode == NO_ERROR) {
3554 struct net_conf *net_conf;
3556 err = nla_put_drbd_cfg_context(skb, resource, connection, NULL);
3559 net_conf = rcu_dereference(connection->net_conf);
3561 err = net_conf_to_skb(skb, net_conf, !capable(CAP_SYS_ADMIN));
3565 connection_to_info(&connection_info, connection);
3566 err = connection_info_to_skb(skb, &connection_info, !capable(CAP_SYS_ADMIN));
3569 connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
3570 err = connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
3573 cb->args[2] = (long)connection;
3575 genlmsg_end(skb, dh);
3581 mutex_unlock(&resource->conf_update);
3587 enum mdf_peer_flag {
3588 MDF_PEER_CONNECTED = 1 << 0,
3589 MDF_PEER_OUTDATED = 1 << 1,
3590 MDF_PEER_FENCING = 1 << 2,
3591 MDF_PEER_FULL_SYNC = 1 << 3,
3594 static void peer_device_to_statistics(struct peer_device_statistics *s,
3595 struct drbd_peer_device *peer_device)
3597 struct drbd_device *device = peer_device->device;
3599 memset(s, 0, sizeof(*s));
3600 s->peer_dev_received = device->recv_cnt;
3601 s->peer_dev_sent = device->send_cnt;
3602 s->peer_dev_pending = atomic_read(&device->ap_pending_cnt) +
3603 atomic_read(&device->rs_pending_cnt);
3604 s->peer_dev_unacked = atomic_read(&device->unacked_cnt);
3605 s->peer_dev_out_of_sync = drbd_bm_total_weight(device) << (BM_BLOCK_SHIFT - 9);
3606 s->peer_dev_resync_failed = device->rs_failed << (BM_BLOCK_SHIFT - 9);
3607 if (get_ldev(device)) {
3608 struct drbd_md *md = &device->ldev->md;
3610 spin_lock_irq(&md->uuid_lock);
3611 s->peer_dev_bitmap_uuid = md->uuid[UI_BITMAP];
3612 spin_unlock_irq(&md->uuid_lock);
3614 (drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND) ?
3615 MDF_PEER_CONNECTED : 0) +
3616 (drbd_md_test_flag(device->ldev, MDF_CONSISTENT) &&
3617 !drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE) ?
3618 MDF_PEER_OUTDATED : 0) +
3619 /* FIXME: MDF_PEER_FENCING? */
3620 (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ?
3621 MDF_PEER_FULL_SYNC : 0);
3626 int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb)
3628 return put_resource_in_arg0(cb, 9);
3631 int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb)
3633 struct nlattr *resource_filter;
3634 struct drbd_resource *resource;
3635 struct drbd_device *uninitialized_var(device);
3636 struct drbd_peer_device *peer_device = NULL;
3637 int minor, err, retcode;
3638 struct drbd_genlmsghdr *dh;
3639 struct idr *idr_to_search;
3641 resource = (struct drbd_resource *)cb->args[0];
3642 if (!cb->args[0] && !cb->args[1]) {
3643 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3644 if (resource_filter) {
3645 retcode = ERR_RES_NOT_KNOWN;
3646 resource = drbd_find_resource(nla_data(resource_filter));
3650 cb->args[0] = (long)resource;
3654 minor = cb->args[1];
3655 idr_to_search = resource ? &resource->devices : &drbd_devices;
3656 device = idr_find(idr_to_search, minor);
3661 device = idr_get_next(idr_to_search, &minor);
3668 for_each_peer_device(peer_device, device)
3669 if (peer_device == (struct drbd_peer_device *)cb->args[2])
3670 goto found_peer_device;
3671 /* peer device was probably deleted */
3674 /* Make peer_device point to the list head (not the first entry). */
3675 peer_device = list_entry(&device->peer_devices, struct drbd_peer_device, peer_devices);
3678 list_for_each_entry_continue_rcu(peer_device, &device->peer_devices, peer_devices) {
3679 if (!has_net_conf(peer_device->connection))
3682 goto put_result; /* only one iteration */
3687 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3688 cb->nlh->nlmsg_seq, &drbd_genl_family,
3689 NLM_F_MULTI, DRBD_ADM_GET_PEER_DEVICES);
3693 dh->ret_code = retcode;
3695 if (retcode == NO_ERROR) {
3696 struct peer_device_info peer_device_info;
3697 struct peer_device_statistics peer_device_statistics;
3700 err = nla_put_drbd_cfg_context(skb, device->resource, peer_device->connection, device);
3703 peer_device_to_info(&peer_device_info, peer_device);
3704 err = peer_device_info_to_skb(skb, &peer_device_info, !capable(CAP_SYS_ADMIN));
3707 peer_device_to_statistics(&peer_device_statistics, peer_device);
3708 err = peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
3711 cb->args[1] = minor;
3712 cb->args[2] = (long)peer_device;
3714 genlmsg_end(skb, dh);
3724 * Return the connection of @resource if @resource has exactly one connection.
3726 static struct drbd_connection *the_only_connection(struct drbd_resource *resource)
3728 struct list_head *connections = &resource->connections;
3730 if (list_empty(connections) || connections->next->next != connections)
3732 return list_first_entry(&resource->connections, struct drbd_connection, connections);
3735 static int nla_put_status_info(struct sk_buff *skb, struct drbd_device *device,
3736 const struct sib_info *sib)
3738 struct drbd_resource *resource = device->resource;
3739 struct state_info *si = NULL; /* for sizeof(si->member); */
3743 int exclude_sensitive;
3745 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
3746 * to. So we better exclude_sensitive information.
3748 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
3749 * in the context of the requesting user process. Exclude sensitive
3750 * information, unless current has superuser.
3752 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
3753 * relies on the current implementation of netlink_dump(), which
3754 * executes the dump callback successively from netlink_recvmsg(),
3755 * always in the context of the receiving process */
3756 exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
3758 got_ldev = get_ldev(device);
3760 /* We need to add connection name and volume number information still.
3761 * Minor number is in drbd_genlmsghdr. */
3762 if (nla_put_drbd_cfg_context(skb, resource, the_only_connection(resource), device))
3763 goto nla_put_failure;
3765 if (res_opts_to_skb(skb, &device->resource->res_opts, exclude_sensitive))
3766 goto nla_put_failure;
3770 struct disk_conf *disk_conf;
3772 disk_conf = rcu_dereference(device->ldev->disk_conf);
3773 err = disk_conf_to_skb(skb, disk_conf, exclude_sensitive);
3776 struct net_conf *nc;
3778 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3780 err = net_conf_to_skb(skb, nc, exclude_sensitive);
3784 goto nla_put_failure;
3786 nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
3788 goto nla_put_failure;
3789 if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) ||
3790 nla_put_u32(skb, T_current_state, device->state.i) ||
3791 nla_put_u64_0pad(skb, T_ed_uuid, device->ed_uuid) ||
3792 nla_put_u64_0pad(skb, T_capacity,
3793 drbd_get_capacity(device->this_bdev)) ||
3794 nla_put_u64_0pad(skb, T_send_cnt, device->send_cnt) ||
3795 nla_put_u64_0pad(skb, T_recv_cnt, device->recv_cnt) ||
3796 nla_put_u64_0pad(skb, T_read_cnt, device->read_cnt) ||
3797 nla_put_u64_0pad(skb, T_writ_cnt, device->writ_cnt) ||
3798 nla_put_u64_0pad(skb, T_al_writ_cnt, device->al_writ_cnt) ||
3799 nla_put_u64_0pad(skb, T_bm_writ_cnt, device->bm_writ_cnt) ||
3800 nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&device->ap_bio_cnt)) ||
3801 nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&device->ap_pending_cnt)) ||
3802 nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&device->rs_pending_cnt)))
3803 goto nla_put_failure;
3808 spin_lock_irq(&device->ldev->md.uuid_lock);
3809 err = nla_put(skb, T_uuids, sizeof(si->uuids), device->ldev->md.uuid);
3810 spin_unlock_irq(&device->ldev->md.uuid_lock);
3813 goto nla_put_failure;
3815 if (nla_put_u32(skb, T_disk_flags, device->ldev->md.flags) ||
3816 nla_put_u64_0pad(skb, T_bits_total, drbd_bm_bits(device)) ||
3817 nla_put_u64_0pad(skb, T_bits_oos,
3818 drbd_bm_total_weight(device)))
3819 goto nla_put_failure;
3820 if (C_SYNC_SOURCE <= device->state.conn &&
3821 C_PAUSED_SYNC_T >= device->state.conn) {
3822 if (nla_put_u64_0pad(skb, T_bits_rs_total,
3823 device->rs_total) ||
3824 nla_put_u64_0pad(skb, T_bits_rs_failed,
3826 goto nla_put_failure;
3831 switch(sib->sib_reason) {
3832 case SIB_SYNC_PROGRESS:
3833 case SIB_GET_STATUS_REPLY:
3835 case SIB_STATE_CHANGE:
3836 if (nla_put_u32(skb, T_prev_state, sib->os.i) ||
3837 nla_put_u32(skb, T_new_state, sib->ns.i))
3838 goto nla_put_failure;
3840 case SIB_HELPER_POST:
3841 if (nla_put_u32(skb, T_helper_exit_code,
3842 sib->helper_exit_code))
3843 goto nla_put_failure;
3845 case SIB_HELPER_PRE:
3846 if (nla_put_string(skb, T_helper, sib->helper_name))
3847 goto nla_put_failure;
3851 nla_nest_end(skb, nla);
3861 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
3863 struct drbd_config_context adm_ctx;
3864 enum drbd_ret_code retcode;
3867 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3868 if (!adm_ctx.reply_skb)
3870 if (retcode != NO_ERROR)
3873 err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.device, NULL);
3875 nlmsg_free(adm_ctx.reply_skb);
3879 drbd_adm_finish(&adm_ctx, info, retcode);
3883 static int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
3885 struct drbd_device *device;
3886 struct drbd_genlmsghdr *dh;
3887 struct drbd_resource *pos = (struct drbd_resource *)cb->args[0];
3888 struct drbd_resource *resource = NULL;
3889 struct drbd_resource *tmp;
3890 unsigned volume = cb->args[1];
3892 /* Open coded, deferred, iteration:
3893 * for_each_resource_safe(resource, tmp, &drbd_resources) {
3894 * connection = "first connection of resource or undefined";
3895 * idr_for_each_entry(&resource->devices, device, i) {
3899 * where resource is cb->args[0];
3900 * and i is cb->args[1];
3902 * cb->args[2] indicates if we shall loop over all resources,
3903 * or just dump all volumes of a single resource.
3905 * This may miss entries inserted after this dump started,
3906 * or entries deleted before they are reached.
3908 * We need to make sure the device won't disappear while
3909 * we are looking at it, and revalidate our iterators
3910 * on each iteration.
3913 /* synchronize with conn_create()/drbd_destroy_connection() */
3915 /* revalidate iterator position */
3916 for_each_resource_rcu(tmp, &drbd_resources) {
3918 /* first iteration */
3930 device = idr_get_next(&resource->devices, &volume);
3932 /* No more volumes to dump on this resource.
3933 * Advance resource iterator. */
3934 pos = list_entry_rcu(resource->resources.next,
3935 struct drbd_resource, resources);
3936 /* Did we dump any volume of this resource yet? */
3938 /* If we reached the end of the list,
3939 * or only a single resource dump was requested,
3941 if (&pos->resources == &drbd_resources || cb->args[2])
3949 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3950 cb->nlh->nlmsg_seq, &drbd_genl_family,
3951 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
3956 /* This is a connection without a single volume.
3957 * Suprisingly enough, it may have a network
3959 struct drbd_connection *connection;
3962 dh->ret_code = NO_ERROR;
3963 connection = the_only_connection(resource);
3964 if (nla_put_drbd_cfg_context(skb, resource, connection, NULL))
3967 struct net_conf *nc;
3969 nc = rcu_dereference(connection->net_conf);
3970 if (nc && net_conf_to_skb(skb, nc, 1) != 0)
3976 D_ASSERT(device, device->vnr == volume);
3977 D_ASSERT(device, device->resource == resource);
3979 dh->minor = device_to_minor(device);
3980 dh->ret_code = NO_ERROR;
3982 if (nla_put_status_info(skb, device, NULL)) {
3984 genlmsg_cancel(skb, dh);
3988 genlmsg_end(skb, dh);
3993 /* where to start the next iteration */
3994 cb->args[0] = (long)pos;
3995 cb->args[1] = (pos == resource) ? volume + 1 : 0;
3997 /* No more resources/volumes/minors found results in an empty skb.
3998 * Which will terminate the dump. */
4003 * Request status of all resources, or of all volumes within a single resource.
4005 * This is a dump, as the answer may not fit in a single reply skb otherwise.
4006 * Which means we cannot use the family->attrbuf or other such members, because
4007 * dump is NOT protected by the genl_lock(). During dump, we only have access
4008 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
4010 * Once things are setup properly, we call into get_one_status().
4012 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
4014 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
4016 const char *resource_name;
4017 struct drbd_resource *resource;
4020 /* Is this a followup call? */
4022 /* ... of a single resource dump,
4023 * and the resource iterator has been advanced already? */
4024 if (cb->args[2] && cb->args[2] != cb->args[0])
4025 return 0; /* DONE. */
4029 /* First call (from netlink_dump_start). We need to figure out
4030 * which resource(s) the user wants us to dump. */
4031 nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
4032 nlmsg_attrlen(cb->nlh, hdrlen),
4033 DRBD_NLA_CFG_CONTEXT);
4035 /* No explicit context given. Dump all. */
4038 maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
4039 nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
4041 return PTR_ERR(nla);
4042 /* context given, but no name present? */
4045 resource_name = nla_data(nla);
4046 if (!*resource_name)
4048 resource = drbd_find_resource(resource_name);
4052 kref_put(&resource->kref, drbd_destroy_resource); /* get_one_status() revalidates the resource */
4054 /* prime iterators, and set "filter" mode mark:
4055 * only dump this connection. */
4056 cb->args[0] = (long)resource;
4057 /* cb->args[1] = 0; passed in this way. */
4058 cb->args[2] = (long)resource;
4061 return get_one_status(skb, cb);
4064 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
4066 struct drbd_config_context adm_ctx;
4067 enum drbd_ret_code retcode;
4068 struct timeout_parms tp;
4071 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4072 if (!adm_ctx.reply_skb)
4074 if (retcode != NO_ERROR)
4078 adm_ctx.device->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
4079 test_bit(USE_DEGR_WFC_T, &adm_ctx.device->flags) ? UT_DEGRADED :
4082 err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
4084 nlmsg_free(adm_ctx.reply_skb);
4088 drbd_adm_finish(&adm_ctx, info, retcode);
4092 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
4094 struct drbd_config_context adm_ctx;
4095 struct drbd_device *device;
4096 enum drbd_ret_code retcode;
4097 struct start_ov_parms parms;
4099 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4100 if (!adm_ctx.reply_skb)
4102 if (retcode != NO_ERROR)
4105 device = adm_ctx.device;
4107 /* resume from last known position, if possible */
4108 parms.ov_start_sector = device->ov_start_sector;
4109 parms.ov_stop_sector = ULLONG_MAX;
4110 if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
4111 int err = start_ov_parms_from_attrs(&parms, info);
4113 retcode = ERR_MANDATORY_TAG;
4114 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4118 mutex_lock(&adm_ctx.resource->adm_mutex);
4120 /* w_make_ov_request expects position to be aligned */
4121 device->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1);
4122 device->ov_stop_sector = parms.ov_stop_sector;
4124 /* If there is still bitmap IO pending, e.g. previous resync or verify
4125 * just being finished, wait for it before requesting a new resync. */
4126 drbd_suspend_io(device);
4127 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
4128 retcode = drbd_request_state(device, NS(conn, C_VERIFY_S));
4129 drbd_resume_io(device);
4131 mutex_unlock(&adm_ctx.resource->adm_mutex);
4133 drbd_adm_finish(&adm_ctx, info, retcode);
4138 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
4140 struct drbd_config_context adm_ctx;
4141 struct drbd_device *device;
4142 enum drbd_ret_code retcode;
4143 int skip_initial_sync = 0;
4145 struct new_c_uuid_parms args;
4147 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4148 if (!adm_ctx.reply_skb)
4150 if (retcode != NO_ERROR)
4153 device = adm_ctx.device;
4154 memset(&args, 0, sizeof(args));
4155 if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
4156 err = new_c_uuid_parms_from_attrs(&args, info);
4158 retcode = ERR_MANDATORY_TAG;
4159 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4164 mutex_lock(&adm_ctx.resource->adm_mutex);
4165 mutex_lock(device->state_mutex); /* Protects us against serialized state changes. */
4167 if (!get_ldev(device)) {
4168 retcode = ERR_NO_DISK;
4172 /* this is "skip initial sync", assume to be clean */
4173 if (device->state.conn == C_CONNECTED &&
4174 first_peer_device(device)->connection->agreed_pro_version >= 90 &&
4175 device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
4176 drbd_info(device, "Preparing to skip initial sync\n");
4177 skip_initial_sync = 1;
4178 } else if (device->state.conn != C_STANDALONE) {
4179 retcode = ERR_CONNECTED;
4183 drbd_uuid_set(device, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
4184 drbd_uuid_new_current(device); /* New current, previous to UI_BITMAP */
4186 if (args.clear_bm) {
4187 err = drbd_bitmap_io(device, &drbd_bmio_clear_n_write,
4188 "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
4190 drbd_err(device, "Writing bitmap failed with %d\n", err);
4191 retcode = ERR_IO_MD_DISK;
4193 if (skip_initial_sync) {
4194 drbd_send_uuids_skip_initial_sync(first_peer_device(device));
4195 _drbd_uuid_set(device, UI_BITMAP, 0);
4196 drbd_print_uuids(device, "cleared bitmap UUID");
4197 spin_lock_irq(&device->resource->req_lock);
4198 _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
4200 spin_unlock_irq(&device->resource->req_lock);
4204 drbd_md_sync(device);
4208 mutex_unlock(device->state_mutex);
4209 mutex_unlock(&adm_ctx.resource->adm_mutex);
4211 drbd_adm_finish(&adm_ctx, info, retcode);
4215 static enum drbd_ret_code
4216 drbd_check_resource_name(struct drbd_config_context *adm_ctx)
4218 const char *name = adm_ctx->resource_name;
4219 if (!name || !name[0]) {
4220 drbd_msg_put_info(adm_ctx->reply_skb, "resource name missing");
4221 return ERR_MANDATORY_TAG;
4223 /* if we want to use these in sysfs/configfs/debugfs some day,
4224 * we must not allow slashes */
4225 if (strchr(name, '/')) {
4226 drbd_msg_put_info(adm_ctx->reply_skb, "invalid resource name");
4227 return ERR_INVALID_REQUEST;
4232 static void resource_to_info(struct resource_info *info,
4233 struct drbd_resource *resource)
4235 info->res_role = conn_highest_role(first_connection(resource));
4236 info->res_susp = resource->susp;
4237 info->res_susp_nod = resource->susp_nod;
4238 info->res_susp_fen = resource->susp_fen;
4241 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
4243 struct drbd_connection *connection;
4244 struct drbd_config_context adm_ctx;
4245 enum drbd_ret_code retcode;
4246 struct res_opts res_opts;
4249 retcode = drbd_adm_prepare(&adm_ctx, skb, info, 0);
4250 if (!adm_ctx.reply_skb)
4252 if (retcode != NO_ERROR)
4255 set_res_opts_defaults(&res_opts);
4256 err = res_opts_from_attrs(&res_opts, info);
4257 if (err && err != -ENOMSG) {
4258 retcode = ERR_MANDATORY_TAG;
4259 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4263 retcode = drbd_check_resource_name(&adm_ctx);
4264 if (retcode != NO_ERROR)
4267 if (adm_ctx.resource) {
4268 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
4269 retcode = ERR_INVALID_REQUEST;
4270 drbd_msg_put_info(adm_ctx.reply_skb, "resource exists");
4272 /* else: still NO_ERROR */
4276 /* not yet safe for genl_family.parallel_ops */
4277 mutex_lock(&resources_mutex);
4278 connection = conn_create(adm_ctx.resource_name, &res_opts);
4279 mutex_unlock(&resources_mutex);
4282 struct resource_info resource_info;
4284 mutex_lock(¬ification_mutex);
4285 resource_to_info(&resource_info, connection->resource);
4286 notify_resource_state(NULL, 0, connection->resource,
4287 &resource_info, NOTIFY_CREATE);
4288 mutex_unlock(¬ification_mutex);
4290 retcode = ERR_NOMEM;
4293 drbd_adm_finish(&adm_ctx, info, retcode);
4297 static void device_to_info(struct device_info *info,
4298 struct drbd_device *device)
4300 info->dev_disk_state = device->state.disk;
4304 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info)
4306 struct drbd_config_context adm_ctx;
4307 struct drbd_genlmsghdr *dh = info->userhdr;
4308 enum drbd_ret_code retcode;
4310 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4311 if (!adm_ctx.reply_skb)
4313 if (retcode != NO_ERROR)
4316 if (dh->minor > MINORMASK) {
4317 drbd_msg_put_info(adm_ctx.reply_skb, "requested minor out of range");
4318 retcode = ERR_INVALID_REQUEST;
4321 if (adm_ctx.volume > DRBD_VOLUME_MAX) {
4322 drbd_msg_put_info(adm_ctx.reply_skb, "requested volume id out of range");
4323 retcode = ERR_INVALID_REQUEST;
4327 /* drbd_adm_prepare made sure already
4328 * that first_peer_device(device)->connection and device->vnr match the request. */
4329 if (adm_ctx.device) {
4330 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
4331 retcode = ERR_MINOR_OR_VOLUME_EXISTS;
4332 /* else: still NO_ERROR */
4336 mutex_lock(&adm_ctx.resource->adm_mutex);
4337 retcode = drbd_create_device(&adm_ctx, dh->minor);
4338 if (retcode == NO_ERROR) {
4339 struct drbd_device *device;
4340 struct drbd_peer_device *peer_device;
4341 struct device_info info;
4342 unsigned int peer_devices = 0;
4343 enum drbd_notification_type flags;
4345 device = minor_to_device(dh->minor);
4346 for_each_peer_device(peer_device, device) {
4347 if (!has_net_conf(peer_device->connection))
4352 device_to_info(&info, device);
4353 mutex_lock(¬ification_mutex);
4354 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4355 notify_device_state(NULL, 0, device, &info, NOTIFY_CREATE | flags);
4356 for_each_peer_device(peer_device, device) {
4357 struct peer_device_info peer_device_info;
4359 if (!has_net_conf(peer_device->connection))
4361 peer_device_to_info(&peer_device_info, peer_device);
4362 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4363 notify_peer_device_state(NULL, 0, peer_device, &peer_device_info,
4364 NOTIFY_CREATE | flags);
4366 mutex_unlock(¬ification_mutex);
4368 mutex_unlock(&adm_ctx.resource->adm_mutex);
4370 drbd_adm_finish(&adm_ctx, info, retcode);
4374 static enum drbd_ret_code adm_del_minor(struct drbd_device *device)
4376 struct drbd_peer_device *peer_device;
4378 if (device->state.disk == D_DISKLESS &&
4379 /* no need to be device->state.conn == C_STANDALONE &&
4380 * we may want to delete a minor from a live replication group.
4382 device->state.role == R_SECONDARY) {
4383 struct drbd_connection *connection =
4384 first_connection(device->resource);
4386 _drbd_request_state(device, NS(conn, C_WF_REPORT_PARAMS),
4387 CS_VERBOSE + CS_WAIT_COMPLETE);
4389 /* If the state engine hasn't stopped the sender thread yet, we
4390 * need to flush the sender work queue before generating the
4391 * DESTROY events here. */
4392 if (get_t_state(&connection->worker) == RUNNING)
4393 drbd_flush_workqueue(&connection->sender_work);
4395 mutex_lock(¬ification_mutex);
4396 for_each_peer_device(peer_device, device) {
4397 if (!has_net_conf(peer_device->connection))
4399 notify_peer_device_state(NULL, 0, peer_device, NULL,
4400 NOTIFY_DESTROY | NOTIFY_CONTINUES);
4402 notify_device_state(NULL, 0, device, NULL, NOTIFY_DESTROY);
4403 mutex_unlock(¬ification_mutex);
4405 drbd_delete_device(device);
4408 return ERR_MINOR_CONFIGURED;
4411 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info)
4413 struct drbd_config_context adm_ctx;
4414 enum drbd_ret_code retcode;
4416 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4417 if (!adm_ctx.reply_skb)
4419 if (retcode != NO_ERROR)
4422 mutex_lock(&adm_ctx.resource->adm_mutex);
4423 retcode = adm_del_minor(adm_ctx.device);
4424 mutex_unlock(&adm_ctx.resource->adm_mutex);
4426 drbd_adm_finish(&adm_ctx, info, retcode);
4430 static int adm_del_resource(struct drbd_resource *resource)
4432 struct drbd_connection *connection;
4434 for_each_connection(connection, resource) {
4435 if (connection->cstate > C_STANDALONE)
4436 return ERR_NET_CONFIGURED;
4438 if (!idr_is_empty(&resource->devices))
4439 return ERR_RES_IN_USE;
4441 /* The state engine has stopped the sender thread, so we don't
4442 * need to flush the sender work queue before generating the
4443 * DESTROY event here. */
4444 mutex_lock(¬ification_mutex);
4445 notify_resource_state(NULL, 0, resource, NULL, NOTIFY_DESTROY);
4446 mutex_unlock(¬ification_mutex);
4448 mutex_lock(&resources_mutex);
4449 list_del_rcu(&resource->resources);
4450 mutex_unlock(&resources_mutex);
4451 /* Make sure all threads have actually stopped: state handling only
4452 * does drbd_thread_stop_nowait(). */
4453 list_for_each_entry(connection, &resource->connections, connections)
4454 drbd_thread_stop(&connection->worker);
4456 drbd_free_resource(resource);
4460 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
4462 struct drbd_config_context adm_ctx;
4463 struct drbd_resource *resource;
4464 struct drbd_connection *connection;
4465 struct drbd_device *device;
4466 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
4469 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4470 if (!adm_ctx.reply_skb)
4472 if (retcode != NO_ERROR)
4475 resource = adm_ctx.resource;
4476 mutex_lock(&resource->adm_mutex);
4478 for_each_connection(connection, resource) {
4479 struct drbd_peer_device *peer_device;
4481 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
4482 retcode = drbd_set_role(peer_device->device, R_SECONDARY, 0);
4483 if (retcode < SS_SUCCESS) {
4484 drbd_msg_put_info(adm_ctx.reply_skb, "failed to demote");
4489 retcode = conn_try_disconnect(connection, 0);
4490 if (retcode < SS_SUCCESS) {
4491 drbd_msg_put_info(adm_ctx.reply_skb, "failed to disconnect");
4497 idr_for_each_entry(&resource->devices, device, i) {
4498 retcode = adm_detach(device, 0);
4499 if (retcode < SS_SUCCESS || retcode > NO_ERROR) {
4500 drbd_msg_put_info(adm_ctx.reply_skb, "failed to detach");
4505 /* delete volumes */
4506 idr_for_each_entry(&resource->devices, device, i) {
4507 retcode = adm_del_minor(device);
4508 if (retcode != NO_ERROR) {
4509 /* "can not happen" */
4510 drbd_msg_put_info(adm_ctx.reply_skb, "failed to delete volume");
4515 retcode = adm_del_resource(resource);
4517 mutex_unlock(&resource->adm_mutex);
4519 drbd_adm_finish(&adm_ctx, info, retcode);
4523 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
4525 struct drbd_config_context adm_ctx;
4526 struct drbd_resource *resource;
4527 enum drbd_ret_code retcode;
4529 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4530 if (!adm_ctx.reply_skb)
4532 if (retcode != NO_ERROR)
4534 resource = adm_ctx.resource;
4536 mutex_lock(&resource->adm_mutex);
4537 retcode = adm_del_resource(resource);
4538 mutex_unlock(&resource->adm_mutex);
4540 drbd_adm_finish(&adm_ctx, info, retcode);
4544 void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib)
4546 struct sk_buff *msg;
4547 struct drbd_genlmsghdr *d_out;
4551 seq = atomic_inc_return(&drbd_genl_seq);
4552 msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4557 d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
4558 if (!d_out) /* cannot happen, but anyways. */
4559 goto nla_put_failure;
4560 d_out->minor = device_to_minor(device);
4561 d_out->ret_code = NO_ERROR;
4563 if (nla_put_status_info(msg, device, sib))
4564 goto nla_put_failure;
4565 genlmsg_end(msg, d_out);
4566 err = drbd_genl_multicast_events(msg, GFP_NOWAIT);
4567 /* msg has been consumed or freed in netlink_broadcast() */
4568 if (err && err != -ESRCH)
4576 drbd_err(device, "Error %d while broadcasting event. "
4577 "Event seq:%u sib_reason:%u\n",
4578 err, seq, sib->sib_reason);
4581 static int nla_put_notification_header(struct sk_buff *msg,
4582 enum drbd_notification_type type)
4584 struct drbd_notification_header nh = {
4588 return drbd_notification_header_to_skb(msg, &nh, true);
4591 void notify_resource_state(struct sk_buff *skb,
4593 struct drbd_resource *resource,
4594 struct resource_info *resource_info,
4595 enum drbd_notification_type type)
4597 struct resource_statistics resource_statistics;
4598 struct drbd_genlmsghdr *dh;
4599 bool multicast = false;
4603 seq = atomic_inc_return(¬ify_genl_seq);
4604 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4612 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_RESOURCE_STATE);
4614 goto nla_put_failure;
4616 dh->ret_code = NO_ERROR;
4617 if (nla_put_drbd_cfg_context(skb, resource, NULL, NULL) ||
4618 nla_put_notification_header(skb, type) ||
4619 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4620 resource_info_to_skb(skb, resource_info, true)))
4621 goto nla_put_failure;
4622 resource_statistics.res_stat_write_ordering = resource->write_ordering;
4623 err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
4625 goto nla_put_failure;
4626 genlmsg_end(skb, dh);
4628 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4629 /* skb has been consumed or freed in netlink_broadcast() */
4630 if (err && err != -ESRCH)
4638 drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4642 void notify_device_state(struct sk_buff *skb,
4644 struct drbd_device *device,
4645 struct device_info *device_info,
4646 enum drbd_notification_type type)
4648 struct device_statistics device_statistics;
4649 struct drbd_genlmsghdr *dh;
4650 bool multicast = false;
4654 seq = atomic_inc_return(¬ify_genl_seq);
4655 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4663 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_DEVICE_STATE);
4665 goto nla_put_failure;
4666 dh->minor = device->minor;
4667 dh->ret_code = NO_ERROR;
4668 if (nla_put_drbd_cfg_context(skb, device->resource, NULL, device) ||
4669 nla_put_notification_header(skb, type) ||
4670 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4671 device_info_to_skb(skb, device_info, true)))
4672 goto nla_put_failure;
4673 device_to_statistics(&device_statistics, device);
4674 device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
4675 genlmsg_end(skb, dh);
4677 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4678 /* skb has been consumed or freed in netlink_broadcast() */
4679 if (err && err != -ESRCH)
4687 drbd_err(device, "Error %d while broadcasting event. Event seq:%u\n",
4691 void notify_connection_state(struct sk_buff *skb,
4693 struct drbd_connection *connection,
4694 struct connection_info *connection_info,
4695 enum drbd_notification_type type)
4697 struct connection_statistics connection_statistics;
4698 struct drbd_genlmsghdr *dh;
4699 bool multicast = false;
4703 seq = atomic_inc_return(¬ify_genl_seq);
4704 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4712 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_CONNECTION_STATE);
4714 goto nla_put_failure;
4716 dh->ret_code = NO_ERROR;
4717 if (nla_put_drbd_cfg_context(skb, connection->resource, connection, NULL) ||
4718 nla_put_notification_header(skb, type) ||
4719 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4720 connection_info_to_skb(skb, connection_info, true)))
4721 goto nla_put_failure;
4722 connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
4723 connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
4724 genlmsg_end(skb, dh);
4726 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4727 /* skb has been consumed or freed in netlink_broadcast() */
4728 if (err && err != -ESRCH)
4736 drbd_err(connection, "Error %d while broadcasting event. Event seq:%u\n",
4740 void notify_peer_device_state(struct sk_buff *skb,
4742 struct drbd_peer_device *peer_device,
4743 struct peer_device_info *peer_device_info,
4744 enum drbd_notification_type type)
4746 struct peer_device_statistics peer_device_statistics;
4747 struct drbd_resource *resource = peer_device->device->resource;
4748 struct drbd_genlmsghdr *dh;
4749 bool multicast = false;
4753 seq = atomic_inc_return(¬ify_genl_seq);
4754 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4762 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_PEER_DEVICE_STATE);
4764 goto nla_put_failure;
4766 dh->ret_code = NO_ERROR;
4767 if (nla_put_drbd_cfg_context(skb, resource, peer_device->connection, peer_device->device) ||
4768 nla_put_notification_header(skb, type) ||
4769 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4770 peer_device_info_to_skb(skb, peer_device_info, true)))
4771 goto nla_put_failure;
4772 peer_device_to_statistics(&peer_device_statistics, peer_device);
4773 peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
4774 genlmsg_end(skb, dh);
4776 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4777 /* skb has been consumed or freed in netlink_broadcast() */
4778 if (err && err != -ESRCH)
4786 drbd_err(peer_device, "Error %d while broadcasting event. Event seq:%u\n",
4790 void notify_helper(enum drbd_notification_type type,
4791 struct drbd_device *device, struct drbd_connection *connection,
4792 const char *name, int status)
4794 struct drbd_resource *resource = device ? device->resource : connection->resource;
4795 struct drbd_helper_info helper_info;
4796 unsigned int seq = atomic_inc_return(¬ify_genl_seq);
4797 struct sk_buff *skb = NULL;
4798 struct drbd_genlmsghdr *dh;
4801 strlcpy(helper_info.helper_name, name, sizeof(helper_info.helper_name));
4802 helper_info.helper_name_len = min(strlen(name), sizeof(helper_info.helper_name));
4803 helper_info.helper_status = status;
4805 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4811 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_HELPER);
4814 dh->minor = device ? device->minor : -1;
4815 dh->ret_code = NO_ERROR;
4816 mutex_lock(¬ification_mutex);
4817 if (nla_put_drbd_cfg_context(skb, resource, connection, device) ||
4818 nla_put_notification_header(skb, type) ||
4819 drbd_helper_info_to_skb(skb, &helper_info, true))
4821 genlmsg_end(skb, dh);
4822 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4824 /* skb has been consumed or freed in netlink_broadcast() */
4825 if (err && err != -ESRCH)
4827 mutex_unlock(¬ification_mutex);
4831 mutex_unlock(¬ification_mutex);
4834 drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4838 static void notify_initial_state_done(struct sk_buff *skb, unsigned int seq)
4840 struct drbd_genlmsghdr *dh;
4844 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_INITIAL_STATE_DONE);
4846 goto nla_put_failure;
4848 dh->ret_code = NO_ERROR;
4849 if (nla_put_notification_header(skb, NOTIFY_EXISTS))
4850 goto nla_put_failure;
4851 genlmsg_end(skb, dh);
4856 pr_err("Error %d sending event. Event seq:%u\n", err, seq);
4859 static void free_state_changes(struct list_head *list)
4861 while (!list_empty(list)) {
4862 struct drbd_state_change *state_change =
4863 list_first_entry(list, struct drbd_state_change, list);
4864 list_del(&state_change->list);
4865 forget_state_change(state_change);
4869 static unsigned int notifications_for_state_change(struct drbd_state_change *state_change)
4872 state_change->n_connections +
4873 state_change->n_devices +
4874 state_change->n_devices * state_change->n_connections;
4877 static int get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4879 struct drbd_state_change *state_change = (struct drbd_state_change *)cb->args[0];
4880 unsigned int seq = cb->args[2];
4882 enum drbd_notification_type flags = 0;
4884 /* There is no need for taking notification_mutex here: it doesn't
4885 matter if the initial state events mix with later state chage
4886 events; we can always tell the events apart by the NOTIFY_EXISTS
4890 if (cb->args[5] == 1) {
4891 notify_initial_state_done(skb, seq);
4895 if (cb->args[4] < cb->args[3])
4896 flags |= NOTIFY_CONTINUES;
4898 notify_resource_state_change(skb, seq, state_change->resource,
4899 NOTIFY_EXISTS | flags);
4903 if (n < state_change->n_connections) {
4904 notify_connection_state_change(skb, seq, &state_change->connections[n],
4905 NOTIFY_EXISTS | flags);
4908 n -= state_change->n_connections;
4909 if (n < state_change->n_devices) {
4910 notify_device_state_change(skb, seq, &state_change->devices[n],
4911 NOTIFY_EXISTS | flags);
4914 n -= state_change->n_devices;
4915 if (n < state_change->n_devices * state_change->n_connections) {
4916 notify_peer_device_state_change(skb, seq, &state_change->peer_devices[n],
4917 NOTIFY_EXISTS | flags);
4922 if (cb->args[4] == cb->args[3]) {
4923 struct drbd_state_change *next_state_change =
4924 list_entry(state_change->list.next,
4925 struct drbd_state_change, list);
4926 cb->args[0] = (long)next_state_change;
4927 cb->args[3] = notifications_for_state_change(next_state_change);
4934 int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4936 struct drbd_resource *resource;
4939 if (cb->args[5] >= 1) {
4940 if (cb->args[5] > 1)
4941 return get_initial_state(skb, cb);
4943 struct drbd_state_change *state_change =
4944 (struct drbd_state_change *)cb->args[0];
4946 /* connect list to head */
4947 list_add(&head, &state_change->list);
4948 free_state_changes(&head);
4953 cb->args[5] = 2; /* number of iterations */
4954 mutex_lock(&resources_mutex);
4955 for_each_resource(resource, &drbd_resources) {
4956 struct drbd_state_change *state_change;
4958 state_change = remember_old_state(resource, GFP_KERNEL);
4959 if (!state_change) {
4960 if (!list_empty(&head))
4961 free_state_changes(&head);
4962 mutex_unlock(&resources_mutex);
4965 copy_old_to_new_state_change(state_change);
4966 list_add_tail(&state_change->list, &head);
4967 cb->args[5] += notifications_for_state_change(state_change);
4969 mutex_unlock(&resources_mutex);
4971 if (!list_empty(&head)) {
4972 struct drbd_state_change *state_change =
4973 list_entry(head.next, struct drbd_state_change, list);
4974 cb->args[0] = (long)state_change;
4975 cb->args[3] = notifications_for_state_change(state_change);
4976 list_del(&head); /* detach list from head */
4979 cb->args[2] = cb->nlh->nlmsg_seq;
4980 return get_initial_state(skb, cb);