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 <asm/unaligned.h>
40 #include <linux/drbd_limits.h>
41 #include <linux/kthread.h>
43 #include <net/genetlink.h>
46 // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info);
47 // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info);
49 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info);
50 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info);
52 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info);
53 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info);
54 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info);
56 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info);
57 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info);
58 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info);
59 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info);
60 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info);
61 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info);
62 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info);
63 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info);
64 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info);
65 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info);
66 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info);
67 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info);
68 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info);
69 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info);
70 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info);
71 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info);
72 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info);
73 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info);
74 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info);
75 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info);
77 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb);
79 #include <linux/drbd_genl_api.h>
81 #include <linux/genl_magic_func.h>
83 /* used blkdev_get_by_path, to claim our meta data device(s) */
84 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
86 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
88 genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
89 if (genlmsg_reply(skb, info))
90 pr_err("error sending genl reply\n");
93 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
94 * reason it could fail was no space in skb, and there are 4k available. */
95 static int drbd_msg_put_info(struct sk_buff *skb, const char *info)
100 if (!info || !info[0])
103 nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY);
107 err = nla_put_string(skb, T_info_text, info);
109 nla_nest_cancel(skb, nla);
112 nla_nest_end(skb, nla);
116 /* This would be a good candidate for a "pre_doit" hook,
117 * and per-family private info->pointers.
118 * But we need to stay compatible with older kernels.
119 * If it returns successfully, adm_ctx members are valid.
121 * At this point, we still rely on the global genl_lock().
122 * If we want to avoid that, and allow "genl_family.parallel_ops", we may need
123 * to add additional synchronization against object destruction/modification.
125 #define DRBD_ADM_NEED_MINOR 1
126 #define DRBD_ADM_NEED_RESOURCE 2
127 #define DRBD_ADM_NEED_CONNECTION 4
128 static int drbd_adm_prepare(struct drbd_config_context *adm_ctx,
129 struct sk_buff *skb, struct genl_info *info, unsigned flags)
131 struct drbd_genlmsghdr *d_in = info->userhdr;
132 const u8 cmd = info->genlhdr->cmd;
135 memset(adm_ctx, 0, sizeof(*adm_ctx));
137 /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
138 if (cmd != DRBD_ADM_GET_STATUS && !capable(CAP_NET_ADMIN))
141 adm_ctx->reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
142 if (!adm_ctx->reply_skb) {
147 adm_ctx->reply_dh = genlmsg_put_reply(adm_ctx->reply_skb,
148 info, &drbd_genl_family, 0, cmd);
149 /* put of a few bytes into a fresh skb of >= 4k will always succeed.
151 if (!adm_ctx->reply_dh) {
156 adm_ctx->reply_dh->minor = d_in->minor;
157 adm_ctx->reply_dh->ret_code = NO_ERROR;
159 adm_ctx->volume = VOLUME_UNSPECIFIED;
160 if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
162 /* parse and validate only */
163 err = drbd_cfg_context_from_attrs(NULL, info);
167 /* It was present, and valid,
168 * copy it over to the reply skb. */
169 err = nla_put_nohdr(adm_ctx->reply_skb,
170 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
171 info->attrs[DRBD_NLA_CFG_CONTEXT]);
175 /* and assign stuff to the adm_ctx */
176 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
178 adm_ctx->volume = nla_get_u32(nla);
179 nla = nested_attr_tb[__nla_type(T_ctx_resource_name)];
181 adm_ctx->resource_name = nla_data(nla);
182 adm_ctx->my_addr = nested_attr_tb[__nla_type(T_ctx_my_addr)];
183 adm_ctx->peer_addr = nested_attr_tb[__nla_type(T_ctx_peer_addr)];
184 if ((adm_ctx->my_addr &&
185 nla_len(adm_ctx->my_addr) > sizeof(adm_ctx->connection->my_addr)) ||
186 (adm_ctx->peer_addr &&
187 nla_len(adm_ctx->peer_addr) > sizeof(adm_ctx->connection->peer_addr))) {
193 adm_ctx->minor = d_in->minor;
194 adm_ctx->device = minor_to_device(d_in->minor);
196 /* We are protected by the global genl_lock().
197 * But we may explicitly drop it/retake it in drbd_adm_set_role(),
198 * so make sure this object stays around. */
200 kref_get(&adm_ctx->device->kref);
202 if (adm_ctx->resource_name) {
203 adm_ctx->resource = drbd_find_resource(adm_ctx->resource_name);
206 if (!adm_ctx->device && (flags & DRBD_ADM_NEED_MINOR)) {
207 drbd_msg_put_info(adm_ctx->reply_skb, "unknown minor");
208 return ERR_MINOR_INVALID;
210 if (!adm_ctx->resource && (flags & DRBD_ADM_NEED_RESOURCE)) {
211 drbd_msg_put_info(adm_ctx->reply_skb, "unknown resource");
212 if (adm_ctx->resource_name)
213 return ERR_RES_NOT_KNOWN;
214 return ERR_INVALID_REQUEST;
217 if (flags & DRBD_ADM_NEED_CONNECTION) {
218 if (adm_ctx->resource) {
219 drbd_msg_put_info(adm_ctx->reply_skb, "no resource name expected");
220 return ERR_INVALID_REQUEST;
222 if (adm_ctx->device) {
223 drbd_msg_put_info(adm_ctx->reply_skb, "no minor number expected");
224 return ERR_INVALID_REQUEST;
226 if (adm_ctx->my_addr && adm_ctx->peer_addr)
227 adm_ctx->connection = conn_get_by_addrs(nla_data(adm_ctx->my_addr),
228 nla_len(adm_ctx->my_addr),
229 nla_data(adm_ctx->peer_addr),
230 nla_len(adm_ctx->peer_addr));
231 if (!adm_ctx->connection) {
232 drbd_msg_put_info(adm_ctx->reply_skb, "unknown connection");
233 return ERR_INVALID_REQUEST;
237 /* some more paranoia, if the request was over-determined */
238 if (adm_ctx->device && adm_ctx->resource &&
239 adm_ctx->device->resource != adm_ctx->resource) {
240 pr_warning("request: minor=%u, resource=%s; but that minor belongs to resource %s\n",
241 adm_ctx->minor, adm_ctx->resource->name,
242 adm_ctx->device->resource->name);
243 drbd_msg_put_info(adm_ctx->reply_skb, "minor exists in different resource");
244 return ERR_INVALID_REQUEST;
246 if (adm_ctx->device &&
247 adm_ctx->volume != VOLUME_UNSPECIFIED &&
248 adm_ctx->volume != adm_ctx->device->vnr) {
249 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
250 adm_ctx->minor, adm_ctx->volume,
251 adm_ctx->device->vnr,
252 adm_ctx->device->resource->name);
253 drbd_msg_put_info(adm_ctx->reply_skb, "minor exists as different volume");
254 return ERR_INVALID_REQUEST;
257 /* still, provide adm_ctx->resource always, if possible. */
258 if (!adm_ctx->resource) {
259 adm_ctx->resource = adm_ctx->device ? adm_ctx->device->resource
260 : adm_ctx->connection ? adm_ctx->connection->resource : NULL;
261 if (adm_ctx->resource)
262 kref_get(&adm_ctx->resource->kref);
268 nlmsg_free(adm_ctx->reply_skb);
269 adm_ctx->reply_skb = NULL;
273 static int drbd_adm_finish(struct drbd_config_context *adm_ctx,
274 struct genl_info *info, int retcode)
276 if (adm_ctx->device) {
277 kref_put(&adm_ctx->device->kref, drbd_destroy_device);
278 adm_ctx->device = NULL;
280 if (adm_ctx->connection) {
281 kref_put(&adm_ctx->connection->kref, &drbd_destroy_connection);
282 adm_ctx->connection = NULL;
284 if (adm_ctx->resource) {
285 kref_put(&adm_ctx->resource->kref, drbd_destroy_resource);
286 adm_ctx->resource = NULL;
289 if (!adm_ctx->reply_skb)
292 adm_ctx->reply_dh->ret_code = retcode;
293 drbd_adm_send_reply(adm_ctx->reply_skb, info);
297 static void setup_khelper_env(struct drbd_connection *connection, char **envp)
301 /* FIXME: A future version will not allow this case. */
302 if (connection->my_addr_len == 0 || connection->peer_addr_len == 0)
305 switch (((struct sockaddr *)&connection->peer_addr)->sa_family) {
308 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
309 &((struct sockaddr_in6 *)&connection->peer_addr)->sin6_addr);
313 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
314 &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
318 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
319 &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
321 snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
324 int drbd_khelper(struct drbd_device *device, char *cmd)
326 char *envp[] = { "HOME=/",
328 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
329 (char[20]) { }, /* address family */
330 (char[60]) { }, /* address */
333 char *argv[] = {usermode_helper, cmd, mb, NULL };
334 struct drbd_connection *connection = first_peer_device(device)->connection;
338 if (current == connection->worker.task)
339 set_bit(CALLBACK_PENDING, &connection->flags);
341 snprintf(mb, 12, "minor-%d", device_to_minor(device));
342 setup_khelper_env(connection, envp);
344 /* The helper may take some time.
345 * write out any unsynced meta data changes now */
346 drbd_md_sync(device);
348 drbd_info(device, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
349 sib.sib_reason = SIB_HELPER_PRE;
350 sib.helper_name = cmd;
351 drbd_bcast_event(device, &sib);
352 ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC);
354 drbd_warn(device, "helper command: %s %s %s exit code %u (0x%x)\n",
355 usermode_helper, cmd, mb,
356 (ret >> 8) & 0xff, ret);
358 drbd_info(device, "helper command: %s %s %s exit code %u (0x%x)\n",
359 usermode_helper, cmd, mb,
360 (ret >> 8) & 0xff, ret);
361 sib.sib_reason = SIB_HELPER_POST;
362 sib.helper_exit_code = ret;
363 drbd_bcast_event(device, &sib);
365 if (current == connection->worker.task)
366 clear_bit(CALLBACK_PENDING, &connection->flags);
368 if (ret < 0) /* Ignore any ERRNOs we got. */
374 static int conn_khelper(struct drbd_connection *connection, char *cmd)
376 char *envp[] = { "HOME=/",
378 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
379 (char[20]) { }, /* address family */
380 (char[60]) { }, /* address */
382 char *resource_name = connection->resource->name;
383 char *argv[] = {usermode_helper, cmd, resource_name, NULL };
386 setup_khelper_env(connection, envp);
387 conn_md_sync(connection);
389 drbd_info(connection, "helper command: %s %s %s\n", usermode_helper, cmd, resource_name);
390 /* TODO: conn_bcast_event() ?? */
392 ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC);
394 drbd_warn(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
395 usermode_helper, cmd, resource_name,
396 (ret >> 8) & 0xff, ret);
398 drbd_info(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
399 usermode_helper, cmd, resource_name,
400 (ret >> 8) & 0xff, ret);
401 /* TODO: conn_bcast_event() ?? */
403 if (ret < 0) /* Ignore any ERRNOs we got. */
409 static enum drbd_fencing_p highest_fencing_policy(struct drbd_connection *connection)
411 enum drbd_fencing_p fp = FP_NOT_AVAIL;
412 struct drbd_peer_device *peer_device;
416 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
417 struct drbd_device *device = peer_device->device;
418 if (get_ldev_if_state(device, D_CONSISTENT)) {
419 struct disk_conf *disk_conf =
420 rcu_dereference(peer_device->device->ldev->disk_conf);
421 fp = max_t(enum drbd_fencing_p, fp, disk_conf->fencing);
427 if (fp == FP_NOT_AVAIL) {
428 /* IO Suspending works on the whole resource.
429 Do it only for one device. */
431 peer_device = idr_get_next(&connection->peer_devices, &vnr);
432 drbd_change_state(peer_device->device, CS_VERBOSE | CS_HARD, NS(susp_fen, 0));
438 bool conn_try_outdate_peer(struct drbd_connection *connection)
440 unsigned int connect_cnt;
441 union drbd_state mask = { };
442 union drbd_state val = { };
443 enum drbd_fencing_p fp;
447 spin_lock_irq(&connection->resource->req_lock);
448 if (connection->cstate >= C_WF_REPORT_PARAMS) {
449 drbd_err(connection, "Expected cstate < C_WF_REPORT_PARAMS\n");
450 spin_unlock_irq(&connection->resource->req_lock);
454 connect_cnt = connection->connect_cnt;
455 spin_unlock_irq(&connection->resource->req_lock);
457 fp = highest_fencing_policy(connection);
460 drbd_warn(connection, "Not fencing peer, I'm not even Consistent myself.\n");
467 r = conn_khelper(connection, "fence-peer");
469 switch ((r>>8) & 0xff) {
470 case 3: /* peer is inconsistent */
471 ex_to_string = "peer is inconsistent or worse";
473 val.pdsk = D_INCONSISTENT;
475 case 4: /* peer got outdated, or was already outdated */
476 ex_to_string = "peer was fenced";
478 val.pdsk = D_OUTDATED;
480 case 5: /* peer was down */
481 if (conn_highest_disk(connection) == D_UP_TO_DATE) {
482 /* we will(have) create(d) a new UUID anyways... */
483 ex_to_string = "peer is unreachable, assumed to be dead";
485 val.pdsk = D_OUTDATED;
487 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
490 case 6: /* Peer is primary, voluntarily outdate myself.
491 * This is useful when an unconnected R_SECONDARY is asked to
492 * become R_PRIMARY, but finds the other peer being active. */
493 ex_to_string = "peer is active";
494 drbd_warn(connection, "Peer is primary, outdating myself.\n");
496 val.disk = D_OUTDATED;
499 if (fp != FP_STONITH)
500 drbd_err(connection, "fence-peer() = 7 && fencing != Stonith !!!\n");
501 ex_to_string = "peer was stonithed";
503 val.pdsk = D_OUTDATED;
506 /* The script is broken ... */
507 drbd_err(connection, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
508 return false; /* Eventually leave IO frozen */
511 drbd_info(connection, "fence-peer helper returned %d (%s)\n",
512 (r>>8) & 0xff, ex_to_string);
517 conn_request_state(connection, mask, val, CS_VERBOSE);
518 here, because we might were able to re-establish the connection in the
520 spin_lock_irq(&connection->resource->req_lock);
521 if (connection->cstate < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &connection->flags)) {
522 if (connection->connect_cnt != connect_cnt)
523 /* In case the connection was established and droped
524 while the fence-peer handler was running, ignore it */
525 drbd_info(connection, "Ignoring fence-peer exit code\n");
527 _conn_request_state(connection, mask, val, CS_VERBOSE);
529 spin_unlock_irq(&connection->resource->req_lock);
531 return conn_highest_pdsk(connection) <= D_OUTDATED;
534 static int _try_outdate_peer_async(void *data)
536 struct drbd_connection *connection = (struct drbd_connection *)data;
538 conn_try_outdate_peer(connection);
540 kref_put(&connection->kref, drbd_destroy_connection);
544 void conn_try_outdate_peer_async(struct drbd_connection *connection)
546 struct task_struct *opa;
548 kref_get(&connection->kref);
549 /* We may just have force_sig()'ed this thread
550 * to get it out of some blocking network function.
551 * Clear signals; otherwise kthread_run(), which internally uses
552 * wait_on_completion_killable(), will mistake our pending signal
553 * for a new fatal signal and fail. */
554 flush_signals(current);
555 opa = kthread_run(_try_outdate_peer_async, connection, "drbd_async_h");
557 drbd_err(connection, "out of mem, failed to invoke fence-peer helper\n");
558 kref_put(&connection->kref, drbd_destroy_connection);
563 drbd_set_role(struct drbd_device *const device, enum drbd_role new_role, int force)
565 struct drbd_peer_device *const peer_device = first_peer_device(device);
566 struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL;
567 const int max_tries = 4;
568 enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
572 union drbd_state mask, val;
574 if (new_role == R_PRIMARY) {
575 struct drbd_connection *connection;
577 /* Detect dead peers as soon as possible. */
580 for_each_connection(connection, device->resource)
581 request_ping(connection);
585 mutex_lock(device->state_mutex);
587 mask.i = 0; mask.role = R_MASK;
588 val.i = 0; val.role = new_role;
590 while (try++ < max_tries) {
591 rv = _drbd_request_state_holding_state_mutex(device, mask, val, CS_WAIT_COMPLETE);
593 /* in case we first succeeded to outdate,
594 * but now suddenly could establish a connection */
595 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
601 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
602 (device->state.disk < D_UP_TO_DATE &&
603 device->state.disk >= D_INCONSISTENT)) {
605 val.disk = D_UP_TO_DATE;
610 if (rv == SS_NO_UP_TO_DATE_DISK &&
611 device->state.disk == D_CONSISTENT && mask.pdsk == 0) {
612 D_ASSERT(device, device->state.pdsk == D_UNKNOWN);
614 if (conn_try_outdate_peer(connection)) {
615 val.disk = D_UP_TO_DATE;
621 if (rv == SS_NOTHING_TO_DO)
623 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
624 if (!conn_try_outdate_peer(connection) && force) {
625 drbd_warn(device, "Forced into split brain situation!\n");
627 val.pdsk = D_OUTDATED;
632 if (rv == SS_TWO_PRIMARIES) {
633 /* Maybe the peer is detected as dead very soon...
634 retry at most once more in this case. */
637 nc = rcu_dereference(connection->net_conf);
638 timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
640 schedule_timeout_interruptible(timeo);
645 if (rv < SS_SUCCESS) {
646 rv = _drbd_request_state(device, mask, val,
647 CS_VERBOSE + CS_WAIT_COMPLETE);
658 drbd_warn(device, "Forced to consider local data as UpToDate!\n");
660 /* Wait until nothing is on the fly :) */
661 wait_event(device->misc_wait, atomic_read(&device->ap_pending_cnt) == 0);
663 /* FIXME also wait for all pending P_BARRIER_ACK? */
665 if (new_role == R_SECONDARY) {
666 if (get_ldev(device)) {
667 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
671 mutex_lock(&device->resource->conf_update);
672 nc = connection->net_conf;
674 nc->discard_my_data = 0; /* without copy; single bit op is atomic */
675 mutex_unlock(&device->resource->conf_update);
677 if (get_ldev(device)) {
678 if (((device->state.conn < C_CONNECTED ||
679 device->state.pdsk <= D_FAILED)
680 && device->ldev->md.uuid[UI_BITMAP] == 0) || forced)
681 drbd_uuid_new_current(device);
683 device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
688 /* writeout of activity log covered areas of the bitmap
689 * to stable storage done in after state change already */
691 if (device->state.conn >= C_WF_REPORT_PARAMS) {
692 /* if this was forced, we should consider sync */
694 drbd_send_uuids(peer_device);
695 drbd_send_current_state(peer_device);
698 drbd_md_sync(device);
699 set_disk_ro(device->vdisk, new_role == R_SECONDARY);
700 kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
702 mutex_unlock(device->state_mutex);
706 static const char *from_attrs_err_to_txt(int err)
708 return err == -ENOMSG ? "required attribute missing" :
709 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
710 err == -EEXIST ? "can not change invariant setting" :
711 "invalid attribute value";
714 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
716 struct drbd_config_context adm_ctx;
717 struct set_role_parms parms;
719 enum drbd_ret_code retcode;
721 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
722 if (!adm_ctx.reply_skb)
724 if (retcode != NO_ERROR)
727 memset(&parms, 0, sizeof(parms));
728 if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
729 err = set_role_parms_from_attrs(&parms, info);
731 retcode = ERR_MANDATORY_TAG;
732 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
737 mutex_lock(&adm_ctx.resource->adm_mutex);
739 if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
740 retcode = drbd_set_role(adm_ctx.device, R_PRIMARY, parms.assume_uptodate);
742 retcode = drbd_set_role(adm_ctx.device, R_SECONDARY, 0);
744 mutex_unlock(&adm_ctx.resource->adm_mutex);
747 drbd_adm_finish(&adm_ctx, info, retcode);
751 /* Initializes the md.*_offset members, so we are able to find
752 * the on disk meta data.
754 * We currently have two possible layouts:
756 * |----------- md_size_sect ------------------|
757 * [ 4k superblock ][ activity log ][ Bitmap ]
759 * | bm_offset = al_offset + X |
760 * ==> bitmap sectors = md_size_sect - bm_offset
763 * |----------- md_size_sect ------------------|
764 * [data.....][ Bitmap ][ activity log ][ 4k superblock ]
766 * | bm_offset = al_offset - Y |
767 * ==> bitmap sectors = Y = al_offset - bm_offset
769 * Activity log size used to be fixed 32kB,
770 * but is about to become configurable.
772 static void drbd_md_set_sector_offsets(struct drbd_device *device,
773 struct drbd_backing_dev *bdev)
775 sector_t md_size_sect = 0;
776 unsigned int al_size_sect = bdev->md.al_size_4k * 8;
778 bdev->md.md_offset = drbd_md_ss(bdev);
780 switch (bdev->md.meta_dev_idx) {
782 /* v07 style fixed size indexed meta data */
783 bdev->md.md_size_sect = MD_128MB_SECT;
784 bdev->md.al_offset = MD_4kB_SECT;
785 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
787 case DRBD_MD_INDEX_FLEX_EXT:
788 /* just occupy the full device; unit: sectors */
789 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
790 bdev->md.al_offset = MD_4kB_SECT;
791 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
793 case DRBD_MD_INDEX_INTERNAL:
794 case DRBD_MD_INDEX_FLEX_INT:
795 /* al size is still fixed */
796 bdev->md.al_offset = -al_size_sect;
797 /* we need (slightly less than) ~ this much bitmap sectors: */
798 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
799 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
800 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
801 md_size_sect = ALIGN(md_size_sect, 8);
803 /* plus the "drbd meta data super block",
804 * and the activity log; */
805 md_size_sect += MD_4kB_SECT + al_size_sect;
807 bdev->md.md_size_sect = md_size_sect;
808 /* bitmap offset is adjusted by 'super' block size */
809 bdev->md.bm_offset = -md_size_sect + MD_4kB_SECT;
814 /* input size is expected to be in KB */
815 char *ppsize(char *buf, unsigned long long size)
817 /* Needs 9 bytes at max including trailing NUL:
818 * -1ULL ==> "16384 EB" */
819 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
821 while (size >= 10000 && base < sizeof(units)-1) {
823 size = (size >> 10) + !!(size & (1<<9));
826 sprintf(buf, "%u %cB", (unsigned)size, units[base]);
831 /* there is still a theoretical deadlock when called from receiver
832 * on an D_INCONSISTENT R_PRIMARY:
833 * remote READ does inc_ap_bio, receiver would need to receive answer
834 * packet from remote to dec_ap_bio again.
835 * receiver receive_sizes(), comes here,
836 * waits for ap_bio_cnt == 0. -> deadlock.
837 * but this cannot happen, actually, because:
838 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
839 * (not connected, or bad/no disk on peer):
840 * see drbd_fail_request_early, ap_bio_cnt is zero.
841 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
842 * peer may not initiate a resize.
844 /* Note these are not to be confused with
845 * drbd_adm_suspend_io/drbd_adm_resume_io,
846 * which are (sub) state changes triggered by admin (drbdsetup),
847 * and can be long lived.
848 * This changes an device->flag, is triggered by drbd internals,
849 * and should be short-lived. */
850 void drbd_suspend_io(struct drbd_device *device)
852 set_bit(SUSPEND_IO, &device->flags);
853 if (drbd_suspended(device))
855 wait_event(device->misc_wait, !atomic_read(&device->ap_bio_cnt));
858 void drbd_resume_io(struct drbd_device *device)
860 clear_bit(SUSPEND_IO, &device->flags);
861 wake_up(&device->misc_wait);
865 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
866 * @device: DRBD device.
868 * Returns 0 on success, negative return values indicate errors.
869 * You should call drbd_md_sync() after calling this function.
871 enum determine_dev_size
872 drbd_determine_dev_size(struct drbd_device *device, enum dds_flags flags, struct resize_parms *rs) __must_hold(local)
874 sector_t prev_first_sect, prev_size; /* previous meta location */
875 sector_t la_size_sect, u_size;
876 struct drbd_md *md = &device->ldev->md;
877 u32 prev_al_stripe_size_4k;
883 int md_moved, la_size_changed;
884 enum determine_dev_size rv = DS_UNCHANGED;
887 * application request passes inc_ap_bio,
888 * but then cannot get an AL-reference.
889 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
892 * Suspend IO right here.
893 * still lock the act_log to not trigger ASSERTs there.
895 drbd_suspend_io(device);
896 buffer = drbd_md_get_buffer(device, __func__); /* Lock meta-data IO */
898 drbd_resume_io(device);
902 /* no wait necessary anymore, actually we could assert that */
903 wait_event(device->al_wait, lc_try_lock(device->act_log));
905 prev_first_sect = drbd_md_first_sector(device->ldev);
906 prev_size = device->ldev->md.md_size_sect;
907 la_size_sect = device->ldev->md.la_size_sect;
910 /* rs is non NULL if we should change the AL layout only */
912 prev_al_stripes = md->al_stripes;
913 prev_al_stripe_size_4k = md->al_stripe_size_4k;
915 md->al_stripes = rs->al_stripes;
916 md->al_stripe_size_4k = rs->al_stripe_size / 4;
917 md->al_size_4k = (u64)rs->al_stripes * rs->al_stripe_size / 4;
920 drbd_md_set_sector_offsets(device, device->ldev);
923 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
925 size = drbd_new_dev_size(device, device->ldev, u_size, flags & DDSF_FORCED);
927 if (size < la_size_sect) {
928 if (rs && u_size == 0) {
929 /* Remove "rs &&" later. This check should always be active, but
930 right now the receiver expects the permissive behavior */
931 drbd_warn(device, "Implicit shrink not allowed. "
932 "Use --size=%llus for explicit shrink.\n",
933 (unsigned long long)size);
934 rv = DS_ERROR_SHRINK;
937 rv = DS_ERROR_SPACE_MD;
938 if (rv != DS_UNCHANGED)
942 if (drbd_get_capacity(device->this_bdev) != size ||
943 drbd_bm_capacity(device) != size) {
945 err = drbd_bm_resize(device, size, !(flags & DDSF_NO_RESYNC));
947 /* currently there is only one error: ENOMEM! */
948 size = drbd_bm_capacity(device)>>1;
950 drbd_err(device, "OUT OF MEMORY! "
951 "Could not allocate bitmap!\n");
953 drbd_err(device, "BM resizing failed. "
954 "Leaving size unchanged at size = %lu KB\n",
955 (unsigned long)size);
959 /* racy, see comments above. */
960 drbd_set_my_capacity(device, size);
961 device->ldev->md.la_size_sect = size;
962 drbd_info(device, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
963 (unsigned long long)size>>1);
968 la_size_changed = (la_size_sect != device->ldev->md.la_size_sect);
970 md_moved = prev_first_sect != drbd_md_first_sector(device->ldev)
971 || prev_size != device->ldev->md.md_size_sect;
973 if (la_size_changed || md_moved || rs) {
976 /* We do some synchronous IO below, which may take some time.
977 * Clear the timer, to avoid scary "timer expired!" messages,
978 * "Superblock" is written out at least twice below, anyways. */
979 del_timer(&device->md_sync_timer);
980 drbd_al_shrink(device); /* All extents inactive. */
982 prev_flags = md->flags;
983 md->flags &= ~MDF_PRIMARY_IND;
984 drbd_md_write(device, buffer);
986 drbd_info(device, "Writing the whole bitmap, %s\n",
987 la_size_changed && md_moved ? "size changed and md moved" :
988 la_size_changed ? "size changed" : "md moved");
989 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
990 drbd_bitmap_io(device, md_moved ? &drbd_bm_write_all : &drbd_bm_write,
991 "size changed", BM_LOCKED_MASK);
992 drbd_initialize_al(device, buffer);
994 md->flags = prev_flags;
995 drbd_md_write(device, buffer);
998 drbd_info(device, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n",
999 md->al_stripes, md->al_stripe_size_4k * 4);
1002 if (size > la_size_sect)
1003 rv = la_size_sect ? DS_GREW : DS_GREW_FROM_ZERO;
1004 if (size < la_size_sect)
1010 md->al_stripes = prev_al_stripes;
1011 md->al_stripe_size_4k = prev_al_stripe_size_4k;
1012 md->al_size_4k = (u64)prev_al_stripes * prev_al_stripe_size_4k;
1014 drbd_md_set_sector_offsets(device, device->ldev);
1017 lc_unlock(device->act_log);
1018 wake_up(&device->al_wait);
1019 drbd_md_put_buffer(device);
1020 drbd_resume_io(device);
1026 drbd_new_dev_size(struct drbd_device *device, struct drbd_backing_dev *bdev,
1027 sector_t u_size, int assume_peer_has_space)
1029 sector_t p_size = device->p_size; /* partner's disk size. */
1030 sector_t la_size_sect = bdev->md.la_size_sect; /* last agreed size. */
1031 sector_t m_size; /* my size */
1034 m_size = drbd_get_max_capacity(bdev);
1036 if (device->state.conn < C_CONNECTED && assume_peer_has_space) {
1037 drbd_warn(device, "Resize while not connected was forced by the user!\n");
1041 if (p_size && m_size) {
1042 size = min_t(sector_t, p_size, m_size);
1045 size = la_size_sect;
1046 if (m_size && m_size < size)
1048 if (p_size && p_size < size)
1059 drbd_err(device, "Both nodes diskless!\n");
1063 drbd_err(device, "Requested disk size is too big (%lu > %lu)\n",
1064 (unsigned long)u_size>>1, (unsigned long)size>>1);
1073 * drbd_check_al_size() - Ensures that the AL is of the right size
1074 * @device: DRBD device.
1076 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
1077 * failed, and 0 on success. You should call drbd_md_sync() after you called
1080 static int drbd_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1082 struct lru_cache *n, *t;
1083 struct lc_element *e;
1084 unsigned int in_use;
1087 if (device->act_log &&
1088 device->act_log->nr_elements == dc->al_extents)
1092 t = device->act_log;
1093 n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
1094 dc->al_extents, sizeof(struct lc_element), 0);
1097 drbd_err(device, "Cannot allocate act_log lru!\n");
1100 spin_lock_irq(&device->al_lock);
1102 for (i = 0; i < t->nr_elements; i++) {
1103 e = lc_element_by_index(t, i);
1105 drbd_err(device, "refcnt(%d)==%d\n",
1106 e->lc_number, e->refcnt);
1107 in_use += e->refcnt;
1111 device->act_log = n;
1112 spin_unlock_irq(&device->al_lock);
1114 drbd_err(device, "Activity log still in use!\n");
1121 drbd_md_mark_dirty(device); /* we changed device->act_log->nr_elemens */
1125 static void drbd_setup_queue_param(struct drbd_device *device, struct drbd_backing_dev *bdev,
1126 unsigned int max_bio_size)
1128 struct request_queue * const q = device->rq_queue;
1129 unsigned int max_hw_sectors = max_bio_size >> 9;
1130 unsigned int max_segments = 0;
1131 struct request_queue *b = NULL;
1134 b = bdev->backing_bdev->bd_disk->queue;
1136 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1138 max_segments = rcu_dereference(device->ldev->disk_conf)->max_bio_bvecs;
1141 blk_set_stacking_limits(&q->limits);
1142 blk_queue_max_write_same_sectors(q, 0);
1145 blk_queue_logical_block_size(q, 512);
1146 blk_queue_max_hw_sectors(q, max_hw_sectors);
1147 /* This is the workaround for "bio would need to, but cannot, be split" */
1148 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1149 blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
1152 struct drbd_connection *connection = first_peer_device(device)->connection;
1154 if (blk_queue_discard(b) &&
1155 (connection->cstate < C_CONNECTED || connection->agreed_features & FF_TRIM)) {
1156 /* For now, don't allow more than one activity log extent worth of data
1157 * to be discarded in one go. We may need to rework drbd_al_begin_io()
1158 * to allow for even larger discard ranges */
1159 blk_queue_max_discard_sectors(q, DRBD_MAX_DISCARD_SECTORS);
1161 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1162 /* REALLY? Is stacking secdiscard "legal"? */
1163 if (blk_queue_secdiscard(b))
1164 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, q);
1166 blk_queue_max_discard_sectors(q, 0);
1167 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1168 queue_flag_clear_unlocked(QUEUE_FLAG_SECDISCARD, q);
1171 blk_queue_stack_limits(q, b);
1173 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
1174 drbd_info(device, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1175 q->backing_dev_info.ra_pages,
1176 b->backing_dev_info.ra_pages);
1177 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
1182 void drbd_reconsider_max_bio_size(struct drbd_device *device, struct drbd_backing_dev *bdev)
1184 unsigned int now, new, local, peer;
1186 now = queue_max_hw_sectors(device->rq_queue) << 9;
1187 local = device->local_max_bio_size; /* Eventually last known value, from volatile memory */
1188 peer = device->peer_max_bio_size; /* Eventually last known value, from meta data */
1191 local = queue_max_hw_sectors(bdev->backing_bdev->bd_disk->queue) << 9;
1192 device->local_max_bio_size = local;
1194 local = min(local, DRBD_MAX_BIO_SIZE);
1196 /* We may ignore peer limits if the peer is modern enough.
1197 Because new from 8.3.8 onwards the peer can use multiple
1198 BIOs for a single peer_request */
1199 if (device->state.conn >= C_WF_REPORT_PARAMS) {
1200 if (first_peer_device(device)->connection->agreed_pro_version < 94)
1201 peer = min(device->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1202 /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1203 else if (first_peer_device(device)->connection->agreed_pro_version == 94)
1204 peer = DRBD_MAX_SIZE_H80_PACKET;
1205 else if (first_peer_device(device)->connection->agreed_pro_version < 100)
1206 peer = DRBD_MAX_BIO_SIZE_P95; /* drbd 8.3.8 onwards, before 8.4.0 */
1208 peer = DRBD_MAX_BIO_SIZE;
1210 /* We may later detach and re-attach on a disconnected Primary.
1211 * Avoid this setting to jump back in that case.
1212 * We want to store what we know the peer DRBD can handle,
1213 * not what the peer IO backend can handle. */
1214 if (peer > device->peer_max_bio_size)
1215 device->peer_max_bio_size = peer;
1217 new = min(local, peer);
1219 if (device->state.role == R_PRIMARY && new < now)
1220 drbd_err(device, "ASSERT FAILED new < now; (%u < %u)\n", new, now);
1223 drbd_info(device, "max BIO size = %u\n", new);
1225 drbd_setup_queue_param(device, bdev, new);
1228 /* Starts the worker thread */
1229 static void conn_reconfig_start(struct drbd_connection *connection)
1231 drbd_thread_start(&connection->worker);
1232 drbd_flush_workqueue(&connection->sender_work);
1235 /* if still unconfigured, stops worker again. */
1236 static void conn_reconfig_done(struct drbd_connection *connection)
1239 spin_lock_irq(&connection->resource->req_lock);
1240 stop_threads = conn_all_vols_unconf(connection) &&
1241 connection->cstate == C_STANDALONE;
1242 spin_unlock_irq(&connection->resource->req_lock);
1244 /* asender is implicitly stopped by receiver
1245 * in conn_disconnect() */
1246 drbd_thread_stop(&connection->receiver);
1247 drbd_thread_stop(&connection->worker);
1251 /* Make sure IO is suspended before calling this function(). */
1252 static void drbd_suspend_al(struct drbd_device *device)
1256 if (!lc_try_lock(device->act_log)) {
1257 drbd_warn(device, "Failed to lock al in drbd_suspend_al()\n");
1261 drbd_al_shrink(device);
1262 spin_lock_irq(&device->resource->req_lock);
1263 if (device->state.conn < C_CONNECTED)
1264 s = !test_and_set_bit(AL_SUSPENDED, &device->flags);
1265 spin_unlock_irq(&device->resource->req_lock);
1266 lc_unlock(device->act_log);
1269 drbd_info(device, "Suspended AL updates\n");
1273 static bool should_set_defaults(struct genl_info *info)
1275 unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1276 return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1279 static unsigned int drbd_al_extents_max(struct drbd_backing_dev *bdev)
1281 /* This is limited by 16 bit "slot" numbers,
1282 * and by available on-disk context storage.
1284 * Also (u16)~0 is special (denotes a "free" extent).
1286 * One transaction occupies one 4kB on-disk block,
1287 * we have n such blocks in the on disk ring buffer,
1288 * the "current" transaction may fail (n-1),
1289 * and there is 919 slot numbers context information per transaction.
1291 * 72 transaction blocks amounts to more than 2**16 context slots,
1292 * so cap there first.
1294 const unsigned int max_al_nr = DRBD_AL_EXTENTS_MAX;
1295 const unsigned int sufficient_on_disk =
1296 (max_al_nr + AL_CONTEXT_PER_TRANSACTION -1)
1297 /AL_CONTEXT_PER_TRANSACTION;
1299 unsigned int al_size_4k = bdev->md.al_size_4k;
1301 if (al_size_4k > sufficient_on_disk)
1304 return (al_size_4k - 1) * AL_CONTEXT_PER_TRANSACTION;
1307 static bool write_ordering_changed(struct disk_conf *a, struct disk_conf *b)
1309 return a->disk_barrier != b->disk_barrier ||
1310 a->disk_flushes != b->disk_flushes ||
1311 a->disk_drain != b->disk_drain;
1314 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1316 struct drbd_config_context adm_ctx;
1317 enum drbd_ret_code retcode;
1318 struct drbd_device *device;
1319 struct disk_conf *new_disk_conf, *old_disk_conf;
1320 struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1323 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1324 if (!adm_ctx.reply_skb)
1326 if (retcode != NO_ERROR)
1329 device = adm_ctx.device;
1330 mutex_lock(&adm_ctx.resource->adm_mutex);
1332 /* we also need a disk
1333 * to change the options on */
1334 if (!get_ldev(device)) {
1335 retcode = ERR_NO_DISK;
1339 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1340 if (!new_disk_conf) {
1341 retcode = ERR_NOMEM;
1345 mutex_lock(&device->resource->conf_update);
1346 old_disk_conf = device->ldev->disk_conf;
1347 *new_disk_conf = *old_disk_conf;
1348 if (should_set_defaults(info))
1349 set_disk_conf_defaults(new_disk_conf);
1351 err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1352 if (err && err != -ENOMSG) {
1353 retcode = ERR_MANDATORY_TAG;
1354 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1358 if (!expect(new_disk_conf->resync_rate >= 1))
1359 new_disk_conf->resync_rate = 1;
1361 if (new_disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1362 new_disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1363 if (new_disk_conf->al_extents > drbd_al_extents_max(device->ldev))
1364 new_disk_conf->al_extents = drbd_al_extents_max(device->ldev);
1366 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1367 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1369 fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1370 if (fifo_size != device->rs_plan_s->size) {
1371 new_plan = fifo_alloc(fifo_size);
1373 drbd_err(device, "kmalloc of fifo_buffer failed");
1374 retcode = ERR_NOMEM;
1379 drbd_suspend_io(device);
1380 wait_event(device->al_wait, lc_try_lock(device->act_log));
1381 drbd_al_shrink(device);
1382 err = drbd_check_al_size(device, new_disk_conf);
1383 lc_unlock(device->act_log);
1384 wake_up(&device->al_wait);
1385 drbd_resume_io(device);
1388 retcode = ERR_NOMEM;
1392 lock_all_resources();
1393 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1394 if (retcode == NO_ERROR) {
1395 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
1396 drbd_resync_after_changed(device);
1398 unlock_all_resources();
1400 if (retcode != NO_ERROR)
1404 old_plan = device->rs_plan_s;
1405 rcu_assign_pointer(device->rs_plan_s, new_plan);
1408 mutex_unlock(&device->resource->conf_update);
1410 if (new_disk_conf->al_updates)
1411 device->ldev->md.flags &= ~MDF_AL_DISABLED;
1413 device->ldev->md.flags |= MDF_AL_DISABLED;
1415 if (new_disk_conf->md_flushes)
1416 clear_bit(MD_NO_FUA, &device->flags);
1418 set_bit(MD_NO_FUA, &device->flags);
1420 if (write_ordering_changed(old_disk_conf, new_disk_conf))
1421 drbd_bump_write_ordering(device->resource, NULL, WO_BDEV_FLUSH);
1423 drbd_md_sync(device);
1425 if (device->state.conn >= C_CONNECTED) {
1426 struct drbd_peer_device *peer_device;
1428 for_each_peer_device(peer_device, device)
1429 drbd_send_sync_param(peer_device);
1433 kfree(old_disk_conf);
1435 mod_timer(&device->request_timer, jiffies + HZ);
1439 mutex_unlock(&device->resource->conf_update);
1441 kfree(new_disk_conf);
1446 mutex_unlock(&adm_ctx.resource->adm_mutex);
1448 drbd_adm_finish(&adm_ctx, info, retcode);
1452 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1454 struct drbd_config_context adm_ctx;
1455 struct drbd_device *device;
1456 struct drbd_peer_device *peer_device;
1457 struct drbd_connection *connection;
1459 enum drbd_ret_code retcode;
1460 enum determine_dev_size dd;
1461 sector_t max_possible_sectors;
1462 sector_t min_md_device_sectors;
1463 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1464 struct disk_conf *new_disk_conf = NULL;
1465 struct block_device *bdev;
1466 struct lru_cache *resync_lru = NULL;
1467 struct fifo_buffer *new_plan = NULL;
1468 union drbd_state ns, os;
1469 enum drbd_state_rv rv;
1470 struct net_conf *nc;
1472 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1473 if (!adm_ctx.reply_skb)
1475 if (retcode != NO_ERROR)
1478 device = adm_ctx.device;
1479 mutex_lock(&adm_ctx.resource->adm_mutex);
1480 peer_device = first_peer_device(device);
1481 connection = peer_device->connection;
1482 conn_reconfig_start(connection);
1484 /* if you want to reconfigure, please tear down first */
1485 if (device->state.disk > D_DISKLESS) {
1486 retcode = ERR_DISK_CONFIGURED;
1489 /* It may just now have detached because of IO error. Make sure
1490 * drbd_ldev_destroy is done already, we may end up here very fast,
1491 * e.g. if someone calls attach from the on-io-error handler,
1492 * to realize a "hot spare" feature (not that I'd recommend that) */
1493 wait_event(device->misc_wait, !test_bit(GOING_DISKLESS, &device->flags));
1495 /* make sure there is no leftover from previous force-detach attempts */
1496 clear_bit(FORCE_DETACH, &device->flags);
1497 clear_bit(WAS_IO_ERROR, &device->flags);
1498 clear_bit(WAS_READ_ERROR, &device->flags);
1500 /* and no leftover from previously aborted resync or verify, either */
1501 device->rs_total = 0;
1502 device->rs_failed = 0;
1503 atomic_set(&device->rs_pending_cnt, 0);
1505 /* allocation not in the IO path, drbdsetup context */
1506 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1508 retcode = ERR_NOMEM;
1511 spin_lock_init(&nbc->md.uuid_lock);
1513 new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1514 if (!new_disk_conf) {
1515 retcode = ERR_NOMEM;
1518 nbc->disk_conf = new_disk_conf;
1520 set_disk_conf_defaults(new_disk_conf);
1521 err = disk_conf_from_attrs(new_disk_conf, info);
1523 retcode = ERR_MANDATORY_TAG;
1524 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1528 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1529 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1531 new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1533 retcode = ERR_NOMEM;
1537 if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1538 retcode = ERR_MD_IDX_INVALID;
1543 nc = rcu_dereference(connection->net_conf);
1545 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1547 retcode = ERR_STONITH_AND_PROT_A;
1553 bdev = blkdev_get_by_path(new_disk_conf->backing_dev,
1554 FMODE_READ | FMODE_WRITE | FMODE_EXCL, device);
1556 drbd_err(device, "open(\"%s\") failed with %ld\n", new_disk_conf->backing_dev,
1558 retcode = ERR_OPEN_DISK;
1561 nbc->backing_bdev = bdev;
1564 * meta_dev_idx >= 0: external fixed size, possibly multiple
1565 * drbd sharing one meta device. TODO in that case, paranoia
1566 * check that [md_bdev, meta_dev_idx] is not yet used by some
1567 * other drbd minor! (if you use drbd.conf + drbdadm, that
1568 * should check it for you already; but if you don't, or
1569 * someone fooled it, we need to double check here)
1571 bdev = blkdev_get_by_path(new_disk_conf->meta_dev,
1572 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1573 (new_disk_conf->meta_dev_idx < 0) ?
1574 (void *)device : (void *)drbd_m_holder);
1576 drbd_err(device, "open(\"%s\") failed with %ld\n", new_disk_conf->meta_dev,
1578 retcode = ERR_OPEN_MD_DISK;
1581 nbc->md_bdev = bdev;
1583 if ((nbc->backing_bdev == nbc->md_bdev) !=
1584 (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1585 new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1586 retcode = ERR_MD_IDX_INVALID;
1590 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1591 1, 61, sizeof(struct bm_extent),
1592 offsetof(struct bm_extent, lce));
1594 retcode = ERR_NOMEM;
1598 /* Read our meta data super block early.
1599 * This also sets other on-disk offsets. */
1600 retcode = drbd_md_read(device, nbc);
1601 if (retcode != NO_ERROR)
1604 if (new_disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1605 new_disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1606 if (new_disk_conf->al_extents > drbd_al_extents_max(nbc))
1607 new_disk_conf->al_extents = drbd_al_extents_max(nbc);
1609 if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1610 drbd_err(device, "max capacity %llu smaller than disk size %llu\n",
1611 (unsigned long long) drbd_get_max_capacity(nbc),
1612 (unsigned long long) new_disk_conf->disk_size);
1613 retcode = ERR_DISK_TOO_SMALL;
1617 if (new_disk_conf->meta_dev_idx < 0) {
1618 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1619 /* at least one MB, otherwise it does not make sense */
1620 min_md_device_sectors = (2<<10);
1622 max_possible_sectors = DRBD_MAX_SECTORS;
1623 min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1);
1626 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1627 retcode = ERR_MD_DISK_TOO_SMALL;
1628 drbd_warn(device, "refusing attach: md-device too small, "
1629 "at least %llu sectors needed for this meta-disk type\n",
1630 (unsigned long long) min_md_device_sectors);
1634 /* Make sure the new disk is big enough
1635 * (we may currently be R_PRIMARY with no local disk...) */
1636 if (drbd_get_max_capacity(nbc) <
1637 drbd_get_capacity(device->this_bdev)) {
1638 retcode = ERR_DISK_TOO_SMALL;
1642 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1644 if (nbc->known_size > max_possible_sectors) {
1645 drbd_warn(device, "==> truncating very big lower level device "
1646 "to currently maximum possible %llu sectors <==\n",
1647 (unsigned long long) max_possible_sectors);
1648 if (new_disk_conf->meta_dev_idx >= 0)
1649 drbd_warn(device, "==>> using internal or flexible "
1650 "meta data may help <<==\n");
1653 drbd_suspend_io(device);
1654 /* also wait for the last barrier ack. */
1655 /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1656 * We need a way to either ignore barrier acks for barriers sent before a device
1657 * was attached, or a way to wait for all pending barrier acks to come in.
1658 * As barriers are counted per resource,
1659 * we'd need to suspend io on all devices of a resource.
1661 wait_event(device->misc_wait, !atomic_read(&device->ap_pending_cnt) || drbd_suspended(device));
1662 /* and for any other previously queued work */
1663 drbd_flush_workqueue(&connection->sender_work);
1665 rv = _drbd_request_state(device, NS(disk, D_ATTACHING), CS_VERBOSE);
1666 retcode = rv; /* FIXME: Type mismatch. */
1667 drbd_resume_io(device);
1668 if (rv < SS_SUCCESS)
1671 if (!get_ldev_if_state(device, D_ATTACHING))
1672 goto force_diskless;
1674 if (!device->bitmap) {
1675 if (drbd_bm_init(device)) {
1676 retcode = ERR_NOMEM;
1677 goto force_diskless_dec;
1681 if (device->state.conn < C_CONNECTED &&
1682 device->state.role == R_PRIMARY && device->ed_uuid &&
1683 (device->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1684 drbd_err(device, "Can only attach to data with current UUID=%016llX\n",
1685 (unsigned long long)device->ed_uuid);
1686 retcode = ERR_DATA_NOT_CURRENT;
1687 goto force_diskless_dec;
1690 /* Since we are diskless, fix the activity log first... */
1691 if (drbd_check_al_size(device, new_disk_conf)) {
1692 retcode = ERR_NOMEM;
1693 goto force_diskless_dec;
1696 /* Prevent shrinking of consistent devices ! */
1697 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1698 drbd_new_dev_size(device, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1699 drbd_warn(device, "refusing to truncate a consistent device\n");
1700 retcode = ERR_DISK_TOO_SMALL;
1701 goto force_diskless_dec;
1704 lock_all_resources();
1705 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1706 if (retcode != NO_ERROR) {
1707 unlock_all_resources();
1708 goto force_diskless_dec;
1711 /* Reset the "barriers don't work" bits here, then force meta data to
1712 * be written, to ensure we determine if barriers are supported. */
1713 if (new_disk_conf->md_flushes)
1714 clear_bit(MD_NO_FUA, &device->flags);
1716 set_bit(MD_NO_FUA, &device->flags);
1718 /* Point of no return reached.
1719 * Devices and memory are no longer released by error cleanup below.
1720 * now device takes over responsibility, and the state engine should
1721 * clean it up somewhere. */
1722 D_ASSERT(device, device->ldev == NULL);
1724 device->resync = resync_lru;
1725 device->rs_plan_s = new_plan;
1728 new_disk_conf = NULL;
1731 drbd_resync_after_changed(device);
1732 drbd_bump_write_ordering(device->resource, device->ldev, WO_BDEV_FLUSH);
1733 unlock_all_resources();
1735 if (drbd_md_test_flag(device->ldev, MDF_CRASHED_PRIMARY))
1736 set_bit(CRASHED_PRIMARY, &device->flags);
1738 clear_bit(CRASHED_PRIMARY, &device->flags);
1740 if (drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
1741 !(device->state.role == R_PRIMARY && device->resource->susp_nod))
1742 set_bit(CRASHED_PRIMARY, &device->flags);
1744 device->send_cnt = 0;
1745 device->recv_cnt = 0;
1746 device->read_cnt = 0;
1747 device->writ_cnt = 0;
1749 drbd_reconsider_max_bio_size(device, device->ldev);
1751 /* If I am currently not R_PRIMARY,
1752 * but meta data primary indicator is set,
1753 * I just now recover from a hard crash,
1754 * and have been R_PRIMARY before that crash.
1756 * Now, if I had no connection before that crash
1757 * (have been degraded R_PRIMARY), chances are that
1758 * I won't find my peer now either.
1760 * In that case, and _only_ in that case,
1761 * we use the degr-wfc-timeout instead of the default,
1762 * so we can automatically recover from a crash of a
1763 * degraded but active "cluster" after a certain timeout.
1765 clear_bit(USE_DEGR_WFC_T, &device->flags);
1766 if (device->state.role != R_PRIMARY &&
1767 drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
1768 !drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND))
1769 set_bit(USE_DEGR_WFC_T, &device->flags);
1771 dd = drbd_determine_dev_size(device, 0, NULL);
1772 if (dd <= DS_ERROR) {
1773 retcode = ERR_NOMEM_BITMAP;
1774 goto force_diskless_dec;
1775 } else if (dd == DS_GREW)
1776 set_bit(RESYNC_AFTER_NEG, &device->flags);
1778 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ||
1779 (test_bit(CRASHED_PRIMARY, &device->flags) &&
1780 drbd_md_test_flag(device->ldev, MDF_AL_DISABLED))) {
1781 drbd_info(device, "Assuming that all blocks are out of sync "
1782 "(aka FullSync)\n");
1783 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
1784 "set_n_write from attaching", BM_LOCKED_MASK)) {
1785 retcode = ERR_IO_MD_DISK;
1786 goto force_diskless_dec;
1789 if (drbd_bitmap_io(device, &drbd_bm_read,
1790 "read from attaching", BM_LOCKED_MASK)) {
1791 retcode = ERR_IO_MD_DISK;
1792 goto force_diskless_dec;
1796 if (_drbd_bm_total_weight(device) == drbd_bm_bits(device))
1797 drbd_suspend_al(device); /* IO is still suspended here... */
1799 spin_lock_irq(&device->resource->req_lock);
1800 os = drbd_read_state(device);
1802 /* If MDF_CONSISTENT is not set go into inconsistent state,
1803 otherwise investigate MDF_WasUpToDate...
1804 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1805 otherwise into D_CONSISTENT state.
1807 if (drbd_md_test_flag(device->ldev, MDF_CONSISTENT)) {
1808 if (drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE))
1809 ns.disk = D_CONSISTENT;
1811 ns.disk = D_OUTDATED;
1813 ns.disk = D_INCONSISTENT;
1816 if (drbd_md_test_flag(device->ldev, MDF_PEER_OUT_DATED))
1817 ns.pdsk = D_OUTDATED;
1820 if (ns.disk == D_CONSISTENT &&
1821 (ns.pdsk == D_OUTDATED || rcu_dereference(device->ldev->disk_conf)->fencing == FP_DONT_CARE))
1822 ns.disk = D_UP_TO_DATE;
1824 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1825 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1826 this point, because drbd_request_state() modifies these
1829 if (rcu_dereference(device->ldev->disk_conf)->al_updates)
1830 device->ldev->md.flags &= ~MDF_AL_DISABLED;
1832 device->ldev->md.flags |= MDF_AL_DISABLED;
1836 /* In case we are C_CONNECTED postpone any decision on the new disk
1837 state after the negotiation phase. */
1838 if (device->state.conn == C_CONNECTED) {
1839 device->new_state_tmp.i = ns.i;
1841 ns.disk = D_NEGOTIATING;
1843 /* We expect to receive up-to-date UUIDs soon.
1844 To avoid a race in receive_state, free p_uuid while
1845 holding req_lock. I.e. atomic with the state change */
1846 kfree(device->p_uuid);
1847 device->p_uuid = NULL;
1850 rv = _drbd_set_state(device, ns, CS_VERBOSE, NULL);
1851 spin_unlock_irq(&device->resource->req_lock);
1853 if (rv < SS_SUCCESS)
1854 goto force_diskless_dec;
1856 mod_timer(&device->request_timer, jiffies + HZ);
1858 if (device->state.role == R_PRIMARY)
1859 device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
1861 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1863 drbd_md_mark_dirty(device);
1864 drbd_md_sync(device);
1866 kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
1868 conn_reconfig_done(connection);
1869 mutex_unlock(&adm_ctx.resource->adm_mutex);
1870 drbd_adm_finish(&adm_ctx, info, retcode);
1876 drbd_force_state(device, NS(disk, D_DISKLESS));
1877 drbd_md_sync(device);
1879 conn_reconfig_done(connection);
1881 if (nbc->backing_bdev)
1882 blkdev_put(nbc->backing_bdev,
1883 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1885 blkdev_put(nbc->md_bdev,
1886 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1889 kfree(new_disk_conf);
1890 lc_destroy(resync_lru);
1892 mutex_unlock(&adm_ctx.resource->adm_mutex);
1894 drbd_adm_finish(&adm_ctx, info, retcode);
1898 static int adm_detach(struct drbd_device *device, int force)
1900 enum drbd_state_rv retcode;
1904 set_bit(FORCE_DETACH, &device->flags);
1905 drbd_force_state(device, NS(disk, D_FAILED));
1906 retcode = SS_SUCCESS;
1910 drbd_suspend_io(device); /* so no-one is stuck in drbd_al_begin_io */
1911 drbd_md_get_buffer(device, __func__); /* make sure there is no in-flight meta-data IO */
1912 retcode = drbd_request_state(device, NS(disk, D_FAILED));
1913 drbd_md_put_buffer(device);
1914 /* D_FAILED will transition to DISKLESS. */
1915 ret = wait_event_interruptible(device->misc_wait,
1916 device->state.disk != D_FAILED);
1917 drbd_resume_io(device);
1918 if ((int)retcode == (int)SS_IS_DISKLESS)
1919 retcode = SS_NOTHING_TO_DO;
1926 /* Detaching the disk is a process in multiple stages. First we need to lock
1927 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1928 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1929 * internal references as well.
1930 * Only then we have finally detached. */
1931 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1933 struct drbd_config_context adm_ctx;
1934 enum drbd_ret_code retcode;
1935 struct detach_parms parms = { };
1938 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1939 if (!adm_ctx.reply_skb)
1941 if (retcode != NO_ERROR)
1944 if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
1945 err = detach_parms_from_attrs(&parms, info);
1947 retcode = ERR_MANDATORY_TAG;
1948 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1953 mutex_lock(&adm_ctx.resource->adm_mutex);
1954 retcode = adm_detach(adm_ctx.device, parms.force_detach);
1955 mutex_unlock(&adm_ctx.resource->adm_mutex);
1957 drbd_adm_finish(&adm_ctx, info, retcode);
1961 static bool conn_resync_running(struct drbd_connection *connection)
1963 struct drbd_peer_device *peer_device;
1968 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
1969 struct drbd_device *device = peer_device->device;
1970 if (device->state.conn == C_SYNC_SOURCE ||
1971 device->state.conn == C_SYNC_TARGET ||
1972 device->state.conn == C_PAUSED_SYNC_S ||
1973 device->state.conn == C_PAUSED_SYNC_T) {
1983 static bool conn_ov_running(struct drbd_connection *connection)
1985 struct drbd_peer_device *peer_device;
1990 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
1991 struct drbd_device *device = peer_device->device;
1992 if (device->state.conn == C_VERIFY_S ||
1993 device->state.conn == C_VERIFY_T) {
2003 static enum drbd_ret_code
2004 _check_net_options(struct drbd_connection *connection, struct net_conf *old_net_conf, struct net_conf *new_net_conf)
2006 struct drbd_peer_device *peer_device;
2009 if (old_net_conf && connection->cstate == C_WF_REPORT_PARAMS && connection->agreed_pro_version < 100) {
2010 if (new_net_conf->wire_protocol != old_net_conf->wire_protocol)
2011 return ERR_NEED_APV_100;
2013 if (new_net_conf->two_primaries != old_net_conf->two_primaries)
2014 return ERR_NEED_APV_100;
2016 if (strcmp(new_net_conf->integrity_alg, old_net_conf->integrity_alg))
2017 return ERR_NEED_APV_100;
2020 if (!new_net_conf->two_primaries &&
2021 conn_highest_role(connection) == R_PRIMARY &&
2022 conn_highest_peer(connection) == R_PRIMARY)
2023 return ERR_NEED_ALLOW_TWO_PRI;
2025 if (new_net_conf->two_primaries &&
2026 (new_net_conf->wire_protocol != DRBD_PROT_C))
2027 return ERR_NOT_PROTO_C;
2029 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2030 struct drbd_device *device = peer_device->device;
2031 if (get_ldev(device)) {
2032 enum drbd_fencing_p fp = rcu_dereference(device->ldev->disk_conf)->fencing;
2034 if (new_net_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
2035 return ERR_STONITH_AND_PROT_A;
2037 if (device->state.role == R_PRIMARY && new_net_conf->discard_my_data)
2038 return ERR_DISCARD_IMPOSSIBLE;
2041 if (new_net_conf->on_congestion != OC_BLOCK && new_net_conf->wire_protocol != DRBD_PROT_A)
2042 return ERR_CONG_NOT_PROTO_A;
2047 static enum drbd_ret_code
2048 check_net_options(struct drbd_connection *connection, struct net_conf *new_net_conf)
2050 static enum drbd_ret_code rv;
2051 struct drbd_peer_device *peer_device;
2055 rv = _check_net_options(connection, rcu_dereference(connection->net_conf), new_net_conf);
2058 /* connection->peer_devices protected by genl_lock() here */
2059 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2060 struct drbd_device *device = peer_device->device;
2061 if (!device->bitmap) {
2062 if (drbd_bm_init(device))
2071 struct crypto_hash *verify_tfm;
2072 struct crypto_hash *csums_tfm;
2073 struct crypto_hash *cram_hmac_tfm;
2074 struct crypto_hash *integrity_tfm;
2078 alloc_hash(struct crypto_hash **tfm, char *tfm_name, int err_alg)
2083 *tfm = crypto_alloc_hash(tfm_name, 0, CRYPTO_ALG_ASYNC);
2092 static enum drbd_ret_code
2093 alloc_crypto(struct crypto *crypto, struct net_conf *new_net_conf)
2095 char hmac_name[CRYPTO_MAX_ALG_NAME];
2096 enum drbd_ret_code rv;
2098 rv = alloc_hash(&crypto->csums_tfm, new_net_conf->csums_alg,
2102 rv = alloc_hash(&crypto->verify_tfm, new_net_conf->verify_alg,
2106 rv = alloc_hash(&crypto->integrity_tfm, new_net_conf->integrity_alg,
2110 if (new_net_conf->cram_hmac_alg[0] != 0) {
2111 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
2112 new_net_conf->cram_hmac_alg);
2114 rv = alloc_hash(&crypto->cram_hmac_tfm, hmac_name,
2121 static void free_crypto(struct crypto *crypto)
2123 crypto_free_hash(crypto->cram_hmac_tfm);
2124 crypto_free_hash(crypto->integrity_tfm);
2125 crypto_free_hash(crypto->csums_tfm);
2126 crypto_free_hash(crypto->verify_tfm);
2129 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
2131 struct drbd_config_context adm_ctx;
2132 enum drbd_ret_code retcode;
2133 struct drbd_connection *connection;
2134 struct net_conf *old_net_conf, *new_net_conf = NULL;
2136 int ovr; /* online verify running */
2137 int rsr; /* re-sync running */
2138 struct crypto crypto = { };
2140 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2141 if (!adm_ctx.reply_skb)
2143 if (retcode != NO_ERROR)
2146 connection = adm_ctx.connection;
2147 mutex_lock(&adm_ctx.resource->adm_mutex);
2149 new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
2150 if (!new_net_conf) {
2151 retcode = ERR_NOMEM;
2155 conn_reconfig_start(connection);
2157 mutex_lock(&connection->data.mutex);
2158 mutex_lock(&connection->resource->conf_update);
2159 old_net_conf = connection->net_conf;
2161 if (!old_net_conf) {
2162 drbd_msg_put_info(adm_ctx.reply_skb, "net conf missing, try connect");
2163 retcode = ERR_INVALID_REQUEST;
2167 *new_net_conf = *old_net_conf;
2168 if (should_set_defaults(info))
2169 set_net_conf_defaults(new_net_conf);
2171 err = net_conf_from_attrs_for_change(new_net_conf, info);
2172 if (err && err != -ENOMSG) {
2173 retcode = ERR_MANDATORY_TAG;
2174 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2178 retcode = check_net_options(connection, new_net_conf);
2179 if (retcode != NO_ERROR)
2182 /* re-sync running */
2183 rsr = conn_resync_running(connection);
2184 if (rsr && strcmp(new_net_conf->csums_alg, old_net_conf->csums_alg)) {
2185 retcode = ERR_CSUMS_RESYNC_RUNNING;
2189 /* online verify running */
2190 ovr = conn_ov_running(connection);
2191 if (ovr && strcmp(new_net_conf->verify_alg, old_net_conf->verify_alg)) {
2192 retcode = ERR_VERIFY_RUNNING;
2196 retcode = alloc_crypto(&crypto, new_net_conf);
2197 if (retcode != NO_ERROR)
2200 rcu_assign_pointer(connection->net_conf, new_net_conf);
2203 crypto_free_hash(connection->csums_tfm);
2204 connection->csums_tfm = crypto.csums_tfm;
2205 crypto.csums_tfm = NULL;
2208 crypto_free_hash(connection->verify_tfm);
2209 connection->verify_tfm = crypto.verify_tfm;
2210 crypto.verify_tfm = NULL;
2213 crypto_free_hash(connection->integrity_tfm);
2214 connection->integrity_tfm = crypto.integrity_tfm;
2215 if (connection->cstate >= C_WF_REPORT_PARAMS && connection->agreed_pro_version >= 100)
2216 /* Do this without trying to take connection->data.mutex again. */
2217 __drbd_send_protocol(connection, P_PROTOCOL_UPDATE);
2219 crypto_free_hash(connection->cram_hmac_tfm);
2220 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2222 mutex_unlock(&connection->resource->conf_update);
2223 mutex_unlock(&connection->data.mutex);
2225 kfree(old_net_conf);
2227 if (connection->cstate >= C_WF_REPORT_PARAMS) {
2228 struct drbd_peer_device *peer_device;
2231 idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
2232 drbd_send_sync_param(peer_device);
2238 mutex_unlock(&connection->resource->conf_update);
2239 mutex_unlock(&connection->data.mutex);
2240 free_crypto(&crypto);
2241 kfree(new_net_conf);
2243 conn_reconfig_done(connection);
2245 mutex_unlock(&adm_ctx.resource->adm_mutex);
2247 drbd_adm_finish(&adm_ctx, info, retcode);
2251 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2253 struct drbd_config_context adm_ctx;
2254 struct drbd_peer_device *peer_device;
2255 struct net_conf *old_net_conf, *new_net_conf = NULL;
2256 struct crypto crypto = { };
2257 struct drbd_resource *resource;
2258 struct drbd_connection *connection;
2259 enum drbd_ret_code retcode;
2263 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2265 if (!adm_ctx.reply_skb)
2267 if (retcode != NO_ERROR)
2269 if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2270 drbd_msg_put_info(adm_ctx.reply_skb, "connection endpoint(s) missing");
2271 retcode = ERR_INVALID_REQUEST;
2275 /* No need for _rcu here. All reconfiguration is
2276 * strictly serialized on genl_lock(). We are protected against
2277 * concurrent reconfiguration/addition/deletion */
2278 for_each_resource(resource, &drbd_resources) {
2279 for_each_connection(connection, resource) {
2280 if (nla_len(adm_ctx.my_addr) == connection->my_addr_len &&
2281 !memcmp(nla_data(adm_ctx.my_addr), &connection->my_addr,
2282 connection->my_addr_len)) {
2283 retcode = ERR_LOCAL_ADDR;
2287 if (nla_len(adm_ctx.peer_addr) == connection->peer_addr_len &&
2288 !memcmp(nla_data(adm_ctx.peer_addr), &connection->peer_addr,
2289 connection->peer_addr_len)) {
2290 retcode = ERR_PEER_ADDR;
2296 mutex_lock(&adm_ctx.resource->adm_mutex);
2297 connection = first_connection(adm_ctx.resource);
2298 conn_reconfig_start(connection);
2300 if (connection->cstate > C_STANDALONE) {
2301 retcode = ERR_NET_CONFIGURED;
2305 /* allocation not in the IO path, drbdsetup / netlink process context */
2306 new_net_conf = kzalloc(sizeof(*new_net_conf), GFP_KERNEL);
2307 if (!new_net_conf) {
2308 retcode = ERR_NOMEM;
2312 set_net_conf_defaults(new_net_conf);
2314 err = net_conf_from_attrs(new_net_conf, info);
2315 if (err && err != -ENOMSG) {
2316 retcode = ERR_MANDATORY_TAG;
2317 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2321 retcode = check_net_options(connection, new_net_conf);
2322 if (retcode != NO_ERROR)
2325 retcode = alloc_crypto(&crypto, new_net_conf);
2326 if (retcode != NO_ERROR)
2329 ((char *)new_net_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2331 drbd_flush_workqueue(&connection->sender_work);
2333 mutex_lock(&adm_ctx.resource->conf_update);
2334 old_net_conf = connection->net_conf;
2336 retcode = ERR_NET_CONFIGURED;
2337 mutex_unlock(&adm_ctx.resource->conf_update);
2340 rcu_assign_pointer(connection->net_conf, new_net_conf);
2342 conn_free_crypto(connection);
2343 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2344 connection->integrity_tfm = crypto.integrity_tfm;
2345 connection->csums_tfm = crypto.csums_tfm;
2346 connection->verify_tfm = crypto.verify_tfm;
2348 connection->my_addr_len = nla_len(adm_ctx.my_addr);
2349 memcpy(&connection->my_addr, nla_data(adm_ctx.my_addr), connection->my_addr_len);
2350 connection->peer_addr_len = nla_len(adm_ctx.peer_addr);
2351 memcpy(&connection->peer_addr, nla_data(adm_ctx.peer_addr), connection->peer_addr_len);
2353 mutex_unlock(&adm_ctx.resource->conf_update);
2356 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2357 struct drbd_device *device = peer_device->device;
2358 device->send_cnt = 0;
2359 device->recv_cnt = 0;
2363 retcode = conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2365 conn_reconfig_done(connection);
2366 mutex_unlock(&adm_ctx.resource->adm_mutex);
2367 drbd_adm_finish(&adm_ctx, info, retcode);
2371 free_crypto(&crypto);
2372 kfree(new_net_conf);
2374 conn_reconfig_done(connection);
2375 mutex_unlock(&adm_ctx.resource->adm_mutex);
2377 drbd_adm_finish(&adm_ctx, info, retcode);
2381 static enum drbd_state_rv conn_try_disconnect(struct drbd_connection *connection, bool force)
2383 enum drbd_state_rv rv;
2385 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2386 force ? CS_HARD : 0);
2389 case SS_NOTHING_TO_DO:
2391 case SS_ALREADY_STANDALONE:
2393 case SS_PRIMARY_NOP:
2394 /* Our state checking code wants to see the peer outdated. */
2395 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, pdsk, D_OUTDATED), 0);
2397 if (rv == SS_OUTDATE_WO_CONN) /* lost connection before graceful disconnect succeeded */
2398 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_VERBOSE);
2401 case SS_CW_FAILED_BY_PEER:
2402 /* The peer probably wants to see us outdated. */
2403 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING,
2404 disk, D_OUTDATED), 0);
2405 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2406 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2411 /* no special handling necessary */
2414 if (rv >= SS_SUCCESS) {
2415 enum drbd_state_rv rv2;
2416 /* No one else can reconfigure the network while I am here.
2417 * The state handling only uses drbd_thread_stop_nowait(),
2418 * we want to really wait here until the receiver is no more.
2420 drbd_thread_stop(&connection->receiver);
2422 /* Race breaker. This additional state change request may be
2423 * necessary, if this was a forced disconnect during a receiver
2424 * restart. We may have "killed" the receiver thread just
2425 * after drbd_receiver() returned. Typically, we should be
2426 * C_STANDALONE already, now, and this becomes a no-op.
2428 rv2 = conn_request_state(connection, NS(conn, C_STANDALONE),
2429 CS_VERBOSE | CS_HARD);
2430 if (rv2 < SS_SUCCESS)
2431 drbd_err(connection,
2432 "unexpected rv2=%d in conn_try_disconnect()\n",
2438 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2440 struct drbd_config_context adm_ctx;
2441 struct disconnect_parms parms;
2442 struct drbd_connection *connection;
2443 enum drbd_state_rv rv;
2444 enum drbd_ret_code retcode;
2447 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2448 if (!adm_ctx.reply_skb)
2450 if (retcode != NO_ERROR)
2453 connection = adm_ctx.connection;
2454 memset(&parms, 0, sizeof(parms));
2455 if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2456 err = disconnect_parms_from_attrs(&parms, info);
2458 retcode = ERR_MANDATORY_TAG;
2459 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2464 mutex_lock(&adm_ctx.resource->adm_mutex);
2465 rv = conn_try_disconnect(connection, parms.force_disconnect);
2466 if (rv < SS_SUCCESS)
2467 retcode = rv; /* FIXME: Type mismatch. */
2470 mutex_unlock(&adm_ctx.resource->adm_mutex);
2472 drbd_adm_finish(&adm_ctx, info, retcode);
2476 void resync_after_online_grow(struct drbd_device *device)
2478 int iass; /* I am sync source */
2480 drbd_info(device, "Resync of new storage after online grow\n");
2481 if (device->state.role != device->state.peer)
2482 iass = (device->state.role == R_PRIMARY);
2484 iass = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags);
2487 drbd_start_resync(device, C_SYNC_SOURCE);
2489 _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2492 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2494 struct drbd_config_context adm_ctx;
2495 struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2496 struct resize_parms rs;
2497 struct drbd_device *device;
2498 enum drbd_ret_code retcode;
2499 enum determine_dev_size dd;
2500 bool change_al_layout = false;
2501 enum dds_flags ddsf;
2505 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2506 if (!adm_ctx.reply_skb)
2508 if (retcode != NO_ERROR)
2511 mutex_lock(&adm_ctx.resource->adm_mutex);
2512 device = adm_ctx.device;
2513 if (!get_ldev(device)) {
2514 retcode = ERR_NO_DISK;
2518 memset(&rs, 0, sizeof(struct resize_parms));
2519 rs.al_stripes = device->ldev->md.al_stripes;
2520 rs.al_stripe_size = device->ldev->md.al_stripe_size_4k * 4;
2521 if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2522 err = resize_parms_from_attrs(&rs, info);
2524 retcode = ERR_MANDATORY_TAG;
2525 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2530 if (device->state.conn > C_CONNECTED) {
2531 retcode = ERR_RESIZE_RESYNC;
2535 if (device->state.role == R_SECONDARY &&
2536 device->state.peer == R_SECONDARY) {
2537 retcode = ERR_NO_PRIMARY;
2541 if (rs.no_resync && first_peer_device(device)->connection->agreed_pro_version < 93) {
2542 retcode = ERR_NEED_APV_93;
2547 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
2549 if (u_size != (sector_t)rs.resize_size) {
2550 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2551 if (!new_disk_conf) {
2552 retcode = ERR_NOMEM;
2557 if (device->ldev->md.al_stripes != rs.al_stripes ||
2558 device->ldev->md.al_stripe_size_4k != rs.al_stripe_size / 4) {
2559 u32 al_size_k = rs.al_stripes * rs.al_stripe_size;
2561 if (al_size_k > (16 * 1024 * 1024)) {
2562 retcode = ERR_MD_LAYOUT_TOO_BIG;
2566 if (al_size_k < MD_32kB_SECT/2) {
2567 retcode = ERR_MD_LAYOUT_TOO_SMALL;
2571 if (device->state.conn != C_CONNECTED && !rs.resize_force) {
2572 retcode = ERR_MD_LAYOUT_CONNECTED;
2576 change_al_layout = true;
2579 if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev))
2580 device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev);
2582 if (new_disk_conf) {
2583 mutex_lock(&device->resource->conf_update);
2584 old_disk_conf = device->ldev->disk_conf;
2585 *new_disk_conf = *old_disk_conf;
2586 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2587 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
2588 mutex_unlock(&device->resource->conf_update);
2590 kfree(old_disk_conf);
2593 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2594 dd = drbd_determine_dev_size(device, ddsf, change_al_layout ? &rs : NULL);
2595 drbd_md_sync(device);
2597 if (dd == DS_ERROR) {
2598 retcode = ERR_NOMEM_BITMAP;
2600 } else if (dd == DS_ERROR_SPACE_MD) {
2601 retcode = ERR_MD_LAYOUT_NO_FIT;
2603 } else if (dd == DS_ERROR_SHRINK) {
2604 retcode = ERR_IMPLICIT_SHRINK;
2608 if (device->state.conn == C_CONNECTED) {
2610 set_bit(RESIZE_PENDING, &device->flags);
2612 drbd_send_uuids(first_peer_device(device));
2613 drbd_send_sizes(first_peer_device(device), 1, ddsf);
2617 mutex_unlock(&adm_ctx.resource->adm_mutex);
2619 drbd_adm_finish(&adm_ctx, info, retcode);
2627 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2629 struct drbd_config_context adm_ctx;
2630 enum drbd_ret_code retcode;
2631 struct res_opts res_opts;
2634 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2635 if (!adm_ctx.reply_skb)
2637 if (retcode != NO_ERROR)
2640 res_opts = adm_ctx.resource->res_opts;
2641 if (should_set_defaults(info))
2642 set_res_opts_defaults(&res_opts);
2644 err = res_opts_from_attrs(&res_opts, info);
2645 if (err && err != -ENOMSG) {
2646 retcode = ERR_MANDATORY_TAG;
2647 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2651 mutex_lock(&adm_ctx.resource->adm_mutex);
2652 err = set_resource_options(adm_ctx.resource, &res_opts);
2654 retcode = ERR_INVALID_REQUEST;
2656 retcode = ERR_NOMEM;
2658 mutex_unlock(&adm_ctx.resource->adm_mutex);
2661 drbd_adm_finish(&adm_ctx, info, retcode);
2665 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2667 struct drbd_config_context adm_ctx;
2668 struct drbd_device *device;
2669 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2671 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2672 if (!adm_ctx.reply_skb)
2674 if (retcode != NO_ERROR)
2677 device = adm_ctx.device;
2678 if (!get_ldev(device)) {
2679 retcode = ERR_NO_DISK;
2683 mutex_lock(&adm_ctx.resource->adm_mutex);
2685 /* If there is still bitmap IO pending, probably because of a previous
2686 * resync just being finished, wait for it before requesting a new resync.
2687 * Also wait for it's after_state_ch(). */
2688 drbd_suspend_io(device);
2689 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
2690 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
2692 /* If we happen to be C_STANDALONE R_SECONDARY, just change to
2693 * D_INCONSISTENT, and set all bits in the bitmap. Otherwise,
2694 * try to start a resync handshake as sync target for full sync.
2696 if (device->state.conn == C_STANDALONE && device->state.role == R_SECONDARY) {
2697 retcode = drbd_request_state(device, NS(disk, D_INCONSISTENT));
2698 if (retcode >= SS_SUCCESS) {
2699 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
2700 "set_n_write from invalidate", BM_LOCKED_MASK))
2701 retcode = ERR_IO_MD_DISK;
2704 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_T));
2705 drbd_resume_io(device);
2706 mutex_unlock(&adm_ctx.resource->adm_mutex);
2709 drbd_adm_finish(&adm_ctx, info, retcode);
2713 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2714 union drbd_state mask, union drbd_state val)
2716 struct drbd_config_context adm_ctx;
2717 enum drbd_ret_code retcode;
2719 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2720 if (!adm_ctx.reply_skb)
2722 if (retcode != NO_ERROR)
2725 mutex_lock(&adm_ctx.resource->adm_mutex);
2726 retcode = drbd_request_state(adm_ctx.device, mask, val);
2727 mutex_unlock(&adm_ctx.resource->adm_mutex);
2729 drbd_adm_finish(&adm_ctx, info, retcode);
2733 static int drbd_bmio_set_susp_al(struct drbd_device *device) __must_hold(local)
2737 rv = drbd_bmio_set_n_write(device);
2738 drbd_suspend_al(device);
2742 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2744 struct drbd_config_context adm_ctx;
2745 int retcode; /* drbd_ret_code, drbd_state_rv */
2746 struct drbd_device *device;
2748 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2749 if (!adm_ctx.reply_skb)
2751 if (retcode != NO_ERROR)
2754 device = adm_ctx.device;
2755 if (!get_ldev(device)) {
2756 retcode = ERR_NO_DISK;
2760 mutex_lock(&adm_ctx.resource->adm_mutex);
2762 /* If there is still bitmap IO pending, probably because of a previous
2763 * resync just being finished, wait for it before requesting a new resync.
2764 * Also wait for it's after_state_ch(). */
2765 drbd_suspend_io(device);
2766 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
2767 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
2769 /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
2770 * in the bitmap. Otherwise, try to start a resync handshake
2771 * as sync source for full sync.
2773 if (device->state.conn == C_STANDALONE && device->state.role == R_PRIMARY) {
2774 /* The peer will get a resync upon connect anyways. Just make that
2775 into a full resync. */
2776 retcode = drbd_request_state(device, NS(pdsk, D_INCONSISTENT));
2777 if (retcode >= SS_SUCCESS) {
2778 if (drbd_bitmap_io(device, &drbd_bmio_set_susp_al,
2779 "set_n_write from invalidate_peer",
2780 BM_LOCKED_SET_ALLOWED))
2781 retcode = ERR_IO_MD_DISK;
2784 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_S));
2785 drbd_resume_io(device);
2786 mutex_unlock(&adm_ctx.resource->adm_mutex);
2789 drbd_adm_finish(&adm_ctx, info, retcode);
2793 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2795 struct drbd_config_context adm_ctx;
2796 enum drbd_ret_code retcode;
2798 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2799 if (!adm_ctx.reply_skb)
2801 if (retcode != NO_ERROR)
2804 mutex_lock(&adm_ctx.resource->adm_mutex);
2805 if (drbd_request_state(adm_ctx.device, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2806 retcode = ERR_PAUSE_IS_SET;
2807 mutex_unlock(&adm_ctx.resource->adm_mutex);
2809 drbd_adm_finish(&adm_ctx, info, retcode);
2813 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2815 struct drbd_config_context adm_ctx;
2816 union drbd_dev_state s;
2817 enum drbd_ret_code retcode;
2819 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2820 if (!adm_ctx.reply_skb)
2822 if (retcode != NO_ERROR)
2825 mutex_lock(&adm_ctx.resource->adm_mutex);
2826 if (drbd_request_state(adm_ctx.device, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2827 s = adm_ctx.device->state;
2828 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2829 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2830 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2832 retcode = ERR_PAUSE_IS_CLEAR;
2835 mutex_unlock(&adm_ctx.resource->adm_mutex);
2837 drbd_adm_finish(&adm_ctx, info, retcode);
2841 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2843 return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2846 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2848 struct drbd_config_context adm_ctx;
2849 struct drbd_device *device;
2850 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2852 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2853 if (!adm_ctx.reply_skb)
2855 if (retcode != NO_ERROR)
2858 mutex_lock(&adm_ctx.resource->adm_mutex);
2859 device = adm_ctx.device;
2860 if (test_bit(NEW_CUR_UUID, &device->flags)) {
2861 drbd_uuid_new_current(device);
2862 clear_bit(NEW_CUR_UUID, &device->flags);
2864 drbd_suspend_io(device);
2865 retcode = drbd_request_state(device, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2866 if (retcode == SS_SUCCESS) {
2867 if (device->state.conn < C_CONNECTED)
2868 tl_clear(first_peer_device(device)->connection);
2869 if (device->state.disk == D_DISKLESS || device->state.disk == D_FAILED)
2870 tl_restart(first_peer_device(device)->connection, FAIL_FROZEN_DISK_IO);
2872 drbd_resume_io(device);
2873 mutex_unlock(&adm_ctx.resource->adm_mutex);
2875 drbd_adm_finish(&adm_ctx, info, retcode);
2879 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2881 return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2884 static int nla_put_drbd_cfg_context(struct sk_buff *skb,
2885 struct drbd_resource *resource,
2886 struct drbd_connection *connection,
2887 struct drbd_device *device)
2890 nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2892 goto nla_put_failure;
2894 nla_put_u32(skb, T_ctx_volume, device->vnr))
2895 goto nla_put_failure;
2896 if (nla_put_string(skb, T_ctx_resource_name, resource->name))
2897 goto nla_put_failure;
2899 if (connection->my_addr_len &&
2900 nla_put(skb, T_ctx_my_addr, connection->my_addr_len, &connection->my_addr))
2901 goto nla_put_failure;
2902 if (connection->peer_addr_len &&
2903 nla_put(skb, T_ctx_peer_addr, connection->peer_addr_len, &connection->peer_addr))
2904 goto nla_put_failure;
2906 nla_nest_end(skb, nla);
2911 nla_nest_cancel(skb, nla);
2916 * Return the connection of @resource if @resource has exactly one connection.
2918 static struct drbd_connection *the_only_connection(struct drbd_resource *resource)
2920 struct list_head *connections = &resource->connections;
2922 if (list_empty(connections) || connections->next->next != connections)
2924 return list_first_entry(&resource->connections, struct drbd_connection, connections);
2927 static int nla_put_status_info(struct sk_buff *skb, struct drbd_device *device,
2928 const struct sib_info *sib)
2930 struct drbd_resource *resource = device->resource;
2931 struct state_info *si = NULL; /* for sizeof(si->member); */
2935 int exclude_sensitive;
2937 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2938 * to. So we better exclude_sensitive information.
2940 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2941 * in the context of the requesting user process. Exclude sensitive
2942 * information, unless current has superuser.
2944 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2945 * relies on the current implementation of netlink_dump(), which
2946 * executes the dump callback successively from netlink_recvmsg(),
2947 * always in the context of the receiving process */
2948 exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2950 got_ldev = get_ldev(device);
2952 /* We need to add connection name and volume number information still.
2953 * Minor number is in drbd_genlmsghdr. */
2954 if (nla_put_drbd_cfg_context(skb, resource, the_only_connection(resource), device))
2955 goto nla_put_failure;
2957 if (res_opts_to_skb(skb, &device->resource->res_opts, exclude_sensitive))
2958 goto nla_put_failure;
2962 struct disk_conf *disk_conf;
2964 disk_conf = rcu_dereference(device->ldev->disk_conf);
2965 err = disk_conf_to_skb(skb, disk_conf, exclude_sensitive);
2968 struct net_conf *nc;
2970 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
2972 err = net_conf_to_skb(skb, nc, exclude_sensitive);
2976 goto nla_put_failure;
2978 nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2980 goto nla_put_failure;
2981 if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) ||
2982 nla_put_u32(skb, T_current_state, device->state.i) ||
2983 nla_put_u64(skb, T_ed_uuid, device->ed_uuid) ||
2984 nla_put_u64(skb, T_capacity, drbd_get_capacity(device->this_bdev)) ||
2985 nla_put_u64(skb, T_send_cnt, device->send_cnt) ||
2986 nla_put_u64(skb, T_recv_cnt, device->recv_cnt) ||
2987 nla_put_u64(skb, T_read_cnt, device->read_cnt) ||
2988 nla_put_u64(skb, T_writ_cnt, device->writ_cnt) ||
2989 nla_put_u64(skb, T_al_writ_cnt, device->al_writ_cnt) ||
2990 nla_put_u64(skb, T_bm_writ_cnt, device->bm_writ_cnt) ||
2991 nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&device->ap_bio_cnt)) ||
2992 nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&device->ap_pending_cnt)) ||
2993 nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&device->rs_pending_cnt)))
2994 goto nla_put_failure;
2999 spin_lock_irq(&device->ldev->md.uuid_lock);
3000 err = nla_put(skb, T_uuids, sizeof(si->uuids), device->ldev->md.uuid);
3001 spin_unlock_irq(&device->ldev->md.uuid_lock);
3004 goto nla_put_failure;
3006 if (nla_put_u32(skb, T_disk_flags, device->ldev->md.flags) ||
3007 nla_put_u64(skb, T_bits_total, drbd_bm_bits(device)) ||
3008 nla_put_u64(skb, T_bits_oos, drbd_bm_total_weight(device)))
3009 goto nla_put_failure;
3010 if (C_SYNC_SOURCE <= device->state.conn &&
3011 C_PAUSED_SYNC_T >= device->state.conn) {
3012 if (nla_put_u64(skb, T_bits_rs_total, device->rs_total) ||
3013 nla_put_u64(skb, T_bits_rs_failed, device->rs_failed))
3014 goto nla_put_failure;
3019 switch(sib->sib_reason) {
3020 case SIB_SYNC_PROGRESS:
3021 case SIB_GET_STATUS_REPLY:
3023 case SIB_STATE_CHANGE:
3024 if (nla_put_u32(skb, T_prev_state, sib->os.i) ||
3025 nla_put_u32(skb, T_new_state, sib->ns.i))
3026 goto nla_put_failure;
3028 case SIB_HELPER_POST:
3029 if (nla_put_u32(skb, T_helper_exit_code,
3030 sib->helper_exit_code))
3031 goto nla_put_failure;
3033 case SIB_HELPER_PRE:
3034 if (nla_put_string(skb, T_helper, sib->helper_name))
3035 goto nla_put_failure;
3039 nla_nest_end(skb, nla);
3049 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
3051 struct drbd_config_context adm_ctx;
3052 enum drbd_ret_code retcode;
3055 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3056 if (!adm_ctx.reply_skb)
3058 if (retcode != NO_ERROR)
3061 err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.device, NULL);
3063 nlmsg_free(adm_ctx.reply_skb);
3067 drbd_adm_finish(&adm_ctx, info, retcode);
3071 static int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
3073 struct drbd_device *device;
3074 struct drbd_genlmsghdr *dh;
3075 struct drbd_resource *pos = (struct drbd_resource *)cb->args[0];
3076 struct drbd_resource *resource = NULL;
3077 struct drbd_resource *tmp;
3078 unsigned volume = cb->args[1];
3080 /* Open coded, deferred, iteration:
3081 * for_each_resource_safe(resource, tmp, &drbd_resources) {
3082 * connection = "first connection of resource or undefined";
3083 * idr_for_each_entry(&resource->devices, device, i) {
3087 * where resource is cb->args[0];
3088 * and i is cb->args[1];
3090 * cb->args[2] indicates if we shall loop over all resources,
3091 * or just dump all volumes of a single resource.
3093 * This may miss entries inserted after this dump started,
3094 * or entries deleted before they are reached.
3096 * We need to make sure the device won't disappear while
3097 * we are looking at it, and revalidate our iterators
3098 * on each iteration.
3101 /* synchronize with conn_create()/drbd_destroy_connection() */
3103 /* revalidate iterator position */
3104 for_each_resource_rcu(tmp, &drbd_resources) {
3106 /* first iteration */
3118 device = idr_get_next(&resource->devices, &volume);
3120 /* No more volumes to dump on this resource.
3121 * Advance resource iterator. */
3122 pos = list_entry_rcu(resource->resources.next,
3123 struct drbd_resource, resources);
3124 /* Did we dump any volume of this resource yet? */
3126 /* If we reached the end of the list,
3127 * or only a single resource dump was requested,
3129 if (&pos->resources == &drbd_resources || cb->args[2])
3137 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3138 cb->nlh->nlmsg_seq, &drbd_genl_family,
3139 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
3144 /* This is a connection without a single volume.
3145 * Suprisingly enough, it may have a network
3147 struct drbd_connection *connection;
3150 dh->ret_code = NO_ERROR;
3151 connection = the_only_connection(resource);
3152 if (nla_put_drbd_cfg_context(skb, resource, connection, NULL))
3155 struct net_conf *nc;
3157 nc = rcu_dereference(connection->net_conf);
3158 if (nc && net_conf_to_skb(skb, nc, 1) != 0)
3164 D_ASSERT(device, device->vnr == volume);
3165 D_ASSERT(device, device->resource == resource);
3167 dh->minor = device_to_minor(device);
3168 dh->ret_code = NO_ERROR;
3170 if (nla_put_status_info(skb, device, NULL)) {
3172 genlmsg_cancel(skb, dh);
3176 genlmsg_end(skb, dh);
3181 /* where to start the next iteration */
3182 cb->args[0] = (long)pos;
3183 cb->args[1] = (pos == resource) ? volume + 1 : 0;
3185 /* No more resources/volumes/minors found results in an empty skb.
3186 * Which will terminate the dump. */
3191 * Request status of all resources, or of all volumes within a single resource.
3193 * This is a dump, as the answer may not fit in a single reply skb otherwise.
3194 * Which means we cannot use the family->attrbuf or other such members, because
3195 * dump is NOT protected by the genl_lock(). During dump, we only have access
3196 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
3198 * Once things are setup properly, we call into get_one_status().
3200 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
3202 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3204 const char *resource_name;
3205 struct drbd_resource *resource;
3208 /* Is this a followup call? */
3210 /* ... of a single resource dump,
3211 * and the resource iterator has been advanced already? */
3212 if (cb->args[2] && cb->args[2] != cb->args[0])
3213 return 0; /* DONE. */
3217 /* First call (from netlink_dump_start). We need to figure out
3218 * which resource(s) the user wants us to dump. */
3219 nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
3220 nlmsg_attrlen(cb->nlh, hdrlen),
3221 DRBD_NLA_CFG_CONTEXT);
3223 /* No explicit context given. Dump all. */
3226 maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
3227 nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
3229 return PTR_ERR(nla);
3230 /* context given, but no name present? */
3233 resource_name = nla_data(nla);
3234 if (!*resource_name)
3236 resource = drbd_find_resource(resource_name);
3240 kref_put(&resource->kref, drbd_destroy_resource); /* get_one_status() revalidates the resource */
3242 /* prime iterators, and set "filter" mode mark:
3243 * only dump this connection. */
3244 cb->args[0] = (long)resource;
3245 /* cb->args[1] = 0; passed in this way. */
3246 cb->args[2] = (long)resource;
3249 return get_one_status(skb, cb);
3252 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
3254 struct drbd_config_context adm_ctx;
3255 enum drbd_ret_code retcode;
3256 struct timeout_parms tp;
3259 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3260 if (!adm_ctx.reply_skb)
3262 if (retcode != NO_ERROR)
3266 adm_ctx.device->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
3267 test_bit(USE_DEGR_WFC_T, &adm_ctx.device->flags) ? UT_DEGRADED :
3270 err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
3272 nlmsg_free(adm_ctx.reply_skb);
3276 drbd_adm_finish(&adm_ctx, info, retcode);
3280 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
3282 struct drbd_config_context adm_ctx;
3283 struct drbd_device *device;
3284 enum drbd_ret_code retcode;
3285 struct start_ov_parms parms;
3287 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3288 if (!adm_ctx.reply_skb)
3290 if (retcode != NO_ERROR)
3293 device = adm_ctx.device;
3295 /* resume from last known position, if possible */
3296 parms.ov_start_sector = device->ov_start_sector;
3297 parms.ov_stop_sector = ULLONG_MAX;
3298 if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
3299 int err = start_ov_parms_from_attrs(&parms, info);
3301 retcode = ERR_MANDATORY_TAG;
3302 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
3306 mutex_lock(&adm_ctx.resource->adm_mutex);
3308 /* w_make_ov_request expects position to be aligned */
3309 device->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1);
3310 device->ov_stop_sector = parms.ov_stop_sector;
3312 /* If there is still bitmap IO pending, e.g. previous resync or verify
3313 * just being finished, wait for it before requesting a new resync. */
3314 drbd_suspend_io(device);
3315 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
3316 retcode = drbd_request_state(device, NS(conn, C_VERIFY_S));
3317 drbd_resume_io(device);
3319 mutex_unlock(&adm_ctx.resource->adm_mutex);
3321 drbd_adm_finish(&adm_ctx, info, retcode);
3326 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
3328 struct drbd_config_context adm_ctx;
3329 struct drbd_device *device;
3330 enum drbd_ret_code retcode;
3331 int skip_initial_sync = 0;
3333 struct new_c_uuid_parms args;
3335 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3336 if (!adm_ctx.reply_skb)
3338 if (retcode != NO_ERROR)
3341 device = adm_ctx.device;
3342 memset(&args, 0, sizeof(args));
3343 if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
3344 err = new_c_uuid_parms_from_attrs(&args, info);
3346 retcode = ERR_MANDATORY_TAG;
3347 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
3352 mutex_lock(&adm_ctx.resource->adm_mutex);
3353 mutex_lock(device->state_mutex); /* Protects us against serialized state changes. */
3355 if (!get_ldev(device)) {
3356 retcode = ERR_NO_DISK;
3360 /* this is "skip initial sync", assume to be clean */
3361 if (device->state.conn == C_CONNECTED &&
3362 first_peer_device(device)->connection->agreed_pro_version >= 90 &&
3363 device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
3364 drbd_info(device, "Preparing to skip initial sync\n");
3365 skip_initial_sync = 1;
3366 } else if (device->state.conn != C_STANDALONE) {
3367 retcode = ERR_CONNECTED;
3371 drbd_uuid_set(device, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
3372 drbd_uuid_new_current(device); /* New current, previous to UI_BITMAP */
3374 if (args.clear_bm) {
3375 err = drbd_bitmap_io(device, &drbd_bmio_clear_n_write,
3376 "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
3378 drbd_err(device, "Writing bitmap failed with %d\n", err);
3379 retcode = ERR_IO_MD_DISK;
3381 if (skip_initial_sync) {
3382 drbd_send_uuids_skip_initial_sync(first_peer_device(device));
3383 _drbd_uuid_set(device, UI_BITMAP, 0);
3384 drbd_print_uuids(device, "cleared bitmap UUID");
3385 spin_lock_irq(&device->resource->req_lock);
3386 _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
3388 spin_unlock_irq(&device->resource->req_lock);
3392 drbd_md_sync(device);
3396 mutex_unlock(device->state_mutex);
3397 mutex_unlock(&adm_ctx.resource->adm_mutex);
3399 drbd_adm_finish(&adm_ctx, info, retcode);
3403 static enum drbd_ret_code
3404 drbd_check_resource_name(struct drbd_config_context *adm_ctx)
3406 const char *name = adm_ctx->resource_name;
3407 if (!name || !name[0]) {
3408 drbd_msg_put_info(adm_ctx->reply_skb, "resource name missing");
3409 return ERR_MANDATORY_TAG;
3411 /* if we want to use these in sysfs/configfs/debugfs some day,
3412 * we must not allow slashes */
3413 if (strchr(name, '/')) {
3414 drbd_msg_put_info(adm_ctx->reply_skb, "invalid resource name");
3415 return ERR_INVALID_REQUEST;
3420 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
3422 struct drbd_config_context adm_ctx;
3423 enum drbd_ret_code retcode;
3424 struct res_opts res_opts;
3427 retcode = drbd_adm_prepare(&adm_ctx, skb, info, 0);
3428 if (!adm_ctx.reply_skb)
3430 if (retcode != NO_ERROR)
3433 set_res_opts_defaults(&res_opts);
3434 err = res_opts_from_attrs(&res_opts, info);
3435 if (err && err != -ENOMSG) {
3436 retcode = ERR_MANDATORY_TAG;
3437 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
3441 retcode = drbd_check_resource_name(&adm_ctx);
3442 if (retcode != NO_ERROR)
3445 if (adm_ctx.resource) {
3446 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
3447 retcode = ERR_INVALID_REQUEST;
3448 drbd_msg_put_info(adm_ctx.reply_skb, "resource exists");
3450 /* else: still NO_ERROR */
3454 /* not yet safe for genl_family.parallel_ops */
3455 mutex_lock(&resources_mutex);
3456 if (!conn_create(adm_ctx.resource_name, &res_opts))
3457 retcode = ERR_NOMEM;
3458 mutex_unlock(&resources_mutex);
3460 drbd_adm_finish(&adm_ctx, info, retcode);
3464 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info)
3466 struct drbd_config_context adm_ctx;
3467 struct drbd_genlmsghdr *dh = info->userhdr;
3468 enum drbd_ret_code retcode;
3470 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
3471 if (!adm_ctx.reply_skb)
3473 if (retcode != NO_ERROR)
3476 if (dh->minor > MINORMASK) {
3477 drbd_msg_put_info(adm_ctx.reply_skb, "requested minor out of range");
3478 retcode = ERR_INVALID_REQUEST;
3481 if (adm_ctx.volume > DRBD_VOLUME_MAX) {
3482 drbd_msg_put_info(adm_ctx.reply_skb, "requested volume id out of range");
3483 retcode = ERR_INVALID_REQUEST;
3487 /* drbd_adm_prepare made sure already
3488 * that first_peer_device(device)->connection and device->vnr match the request. */
3489 if (adm_ctx.device) {
3490 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
3491 retcode = ERR_MINOR_OR_VOLUME_EXISTS;
3492 /* else: still NO_ERROR */
3496 mutex_lock(&adm_ctx.resource->adm_mutex);
3497 retcode = drbd_create_device(&adm_ctx, dh->minor);
3498 mutex_unlock(&adm_ctx.resource->adm_mutex);
3500 drbd_adm_finish(&adm_ctx, info, retcode);
3504 static enum drbd_ret_code adm_del_minor(struct drbd_device *device)
3506 if (device->state.disk == D_DISKLESS &&
3507 /* no need to be device->state.conn == C_STANDALONE &&
3508 * we may want to delete a minor from a live replication group.
3510 device->state.role == R_SECONDARY) {
3511 _drbd_request_state(device, NS(conn, C_WF_REPORT_PARAMS),
3512 CS_VERBOSE + CS_WAIT_COMPLETE);
3513 drbd_delete_device(device);
3516 return ERR_MINOR_CONFIGURED;
3519 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info)
3521 struct drbd_config_context adm_ctx;
3522 enum drbd_ret_code retcode;
3524 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3525 if (!adm_ctx.reply_skb)
3527 if (retcode != NO_ERROR)
3530 mutex_lock(&adm_ctx.resource->adm_mutex);
3531 retcode = adm_del_minor(adm_ctx.device);
3532 mutex_unlock(&adm_ctx.resource->adm_mutex);
3534 drbd_adm_finish(&adm_ctx, info, retcode);
3538 static int adm_del_resource(struct drbd_resource *resource)
3540 struct drbd_connection *connection;
3542 for_each_connection(connection, resource) {
3543 if (connection->cstate > C_STANDALONE)
3544 return ERR_NET_CONFIGURED;
3546 if (!idr_is_empty(&resource->devices))
3547 return ERR_RES_IN_USE;
3549 mutex_lock(&resources_mutex);
3550 list_del_rcu(&resource->resources);
3551 mutex_unlock(&resources_mutex);
3552 /* Make sure all threads have actually stopped: state handling only
3553 * does drbd_thread_stop_nowait(). */
3554 list_for_each_entry(connection, &resource->connections, connections)
3555 drbd_thread_stop(&connection->worker);
3557 drbd_free_resource(resource);
3561 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
3563 struct drbd_config_context adm_ctx;
3564 struct drbd_resource *resource;
3565 struct drbd_connection *connection;
3566 struct drbd_device *device;
3567 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3570 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
3571 if (!adm_ctx.reply_skb)
3573 if (retcode != NO_ERROR)
3576 resource = adm_ctx.resource;
3577 mutex_lock(&resource->adm_mutex);
3579 for_each_connection(connection, resource) {
3580 struct drbd_peer_device *peer_device;
3582 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
3583 retcode = drbd_set_role(peer_device->device, R_SECONDARY, 0);
3584 if (retcode < SS_SUCCESS) {
3585 drbd_msg_put_info(adm_ctx.reply_skb, "failed to demote");
3590 retcode = conn_try_disconnect(connection, 0);
3591 if (retcode < SS_SUCCESS) {
3592 drbd_msg_put_info(adm_ctx.reply_skb, "failed to disconnect");
3598 idr_for_each_entry(&resource->devices, device, i) {
3599 retcode = adm_detach(device, 0);
3600 if (retcode < SS_SUCCESS || retcode > NO_ERROR) {
3601 drbd_msg_put_info(adm_ctx.reply_skb, "failed to detach");
3606 /* delete volumes */
3607 idr_for_each_entry(&resource->devices, device, i) {
3608 retcode = adm_del_minor(device);
3609 if (retcode != NO_ERROR) {
3610 /* "can not happen" */
3611 drbd_msg_put_info(adm_ctx.reply_skb, "failed to delete volume");
3616 retcode = adm_del_resource(resource);
3618 mutex_unlock(&resource->adm_mutex);
3620 drbd_adm_finish(&adm_ctx, info, retcode);
3624 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
3626 struct drbd_config_context adm_ctx;
3627 struct drbd_resource *resource;
3628 enum drbd_ret_code retcode;
3630 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
3631 if (!adm_ctx.reply_skb)
3633 if (retcode != NO_ERROR)
3635 resource = adm_ctx.resource;
3637 mutex_lock(&resource->adm_mutex);
3638 retcode = adm_del_resource(resource);
3639 mutex_unlock(&resource->adm_mutex);
3641 drbd_adm_finish(&adm_ctx, info, retcode);
3645 void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib)
3647 static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3648 struct sk_buff *msg;
3649 struct drbd_genlmsghdr *d_out;
3653 seq = atomic_inc_return(&drbd_genl_seq);
3654 msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3659 d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3660 if (!d_out) /* cannot happen, but anyways. */
3661 goto nla_put_failure;
3662 d_out->minor = device_to_minor(device);
3663 d_out->ret_code = NO_ERROR;
3665 if (nla_put_status_info(msg, device, sib))
3666 goto nla_put_failure;
3667 genlmsg_end(msg, d_out);
3668 err = drbd_genl_multicast_events(msg, 0);
3669 /* msg has been consumed or freed in netlink_broadcast() */
3670 if (err && err != -ESRCH)
3678 drbd_err(device, "Error %d while broadcasting event. "
3679 "Event seq:%u sib_reason:%u\n",
3680 err, seq, sib->sib_reason);