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 #include <linux/module.h>
27 #include <linux/drbd.h>
30 #include <linux/file.h>
31 #include <linux/slab.h>
32 #include <linux/blkpg.h>
33 #include <linux/cpumask.h>
35 #include "drbd_protocol.h"
37 #include <asm/unaligned.h>
38 #include <linux/drbd_limits.h>
39 #include <linux/kthread.h>
41 #include <net/genetlink.h>
44 // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info);
45 // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info);
47 int drbd_adm_new_minor(struct sk_buff
*skb
, struct genl_info
*info
);
48 int drbd_adm_del_minor(struct sk_buff
*skb
, struct genl_info
*info
);
50 int drbd_adm_new_resource(struct sk_buff
*skb
, struct genl_info
*info
);
51 int drbd_adm_del_resource(struct sk_buff
*skb
, struct genl_info
*info
);
52 int drbd_adm_down(struct sk_buff
*skb
, struct genl_info
*info
);
54 int drbd_adm_set_role(struct sk_buff
*skb
, struct genl_info
*info
);
55 int drbd_adm_attach(struct sk_buff
*skb
, struct genl_info
*info
);
56 int drbd_adm_disk_opts(struct sk_buff
*skb
, struct genl_info
*info
);
57 int drbd_adm_detach(struct sk_buff
*skb
, struct genl_info
*info
);
58 int drbd_adm_connect(struct sk_buff
*skb
, struct genl_info
*info
);
59 int drbd_adm_net_opts(struct sk_buff
*skb
, struct genl_info
*info
);
60 int drbd_adm_resize(struct sk_buff
*skb
, struct genl_info
*info
);
61 int drbd_adm_start_ov(struct sk_buff
*skb
, struct genl_info
*info
);
62 int drbd_adm_new_c_uuid(struct sk_buff
*skb
, struct genl_info
*info
);
63 int drbd_adm_disconnect(struct sk_buff
*skb
, struct genl_info
*info
);
64 int drbd_adm_invalidate(struct sk_buff
*skb
, struct genl_info
*info
);
65 int drbd_adm_invalidate_peer(struct sk_buff
*skb
, struct genl_info
*info
);
66 int drbd_adm_pause_sync(struct sk_buff
*skb
, struct genl_info
*info
);
67 int drbd_adm_resume_sync(struct sk_buff
*skb
, struct genl_info
*info
);
68 int drbd_adm_suspend_io(struct sk_buff
*skb
, struct genl_info
*info
);
69 int drbd_adm_resume_io(struct sk_buff
*skb
, struct genl_info
*info
);
70 int drbd_adm_outdate(struct sk_buff
*skb
, struct genl_info
*info
);
71 int drbd_adm_resource_opts(struct sk_buff
*skb
, struct genl_info
*info
);
72 int drbd_adm_get_status(struct sk_buff
*skb
, struct genl_info
*info
);
73 int drbd_adm_get_timeout_type(struct sk_buff
*skb
, struct genl_info
*info
);
75 int drbd_adm_get_status_all(struct sk_buff
*skb
, struct netlink_callback
*cb
);
77 #include <linux/drbd_genl_api.h>
79 #include <linux/genl_magic_func.h>
81 /* used blkdev_get_by_path, to claim our meta data device(s) */
82 static char *drbd_m_holder
= "Hands off! this is DRBD's meta data device.";
84 static void drbd_adm_send_reply(struct sk_buff
*skb
, struct genl_info
*info
)
86 genlmsg_end(skb
, genlmsg_data(nlmsg_data(nlmsg_hdr(skb
))));
87 if (genlmsg_reply(skb
, info
))
88 printk(KERN_ERR
"drbd: error sending genl reply\n");
91 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
92 * reason it could fail was no space in skb, and there are 4k available. */
93 int drbd_msg_put_info(struct sk_buff
*skb
, const char *info
)
98 if (!info
|| !info
[0])
101 nla
= nla_nest_start(skb
, DRBD_NLA_CFG_REPLY
);
105 err
= nla_put_string(skb
, T_info_text
, info
);
107 nla_nest_cancel(skb
, nla
);
110 nla_nest_end(skb
, nla
);
114 /* This would be a good candidate for a "pre_doit" hook,
115 * and per-family private info->pointers.
116 * But we need to stay compatible with older kernels.
117 * If it returns successfully, adm_ctx members are valid.
119 * At this point, we still rely on the global genl_lock().
120 * If we want to avoid that, and allow "genl_family.parallel_ops", we may need
121 * to add additional synchronization against object destruction/modification.
123 #define DRBD_ADM_NEED_MINOR 1
124 #define DRBD_ADM_NEED_RESOURCE 2
125 #define DRBD_ADM_NEED_CONNECTION 4
126 static int drbd_adm_prepare(struct drbd_config_context
*adm_ctx
,
127 struct sk_buff
*skb
, struct genl_info
*info
, unsigned flags
)
129 struct drbd_genlmsghdr
*d_in
= info
->userhdr
;
130 const u8 cmd
= info
->genlhdr
->cmd
;
133 memset(adm_ctx
, 0, sizeof(*adm_ctx
));
135 /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
136 if (cmd
!= DRBD_ADM_GET_STATUS
&& !capable(CAP_NET_ADMIN
))
139 adm_ctx
->reply_skb
= genlmsg_new(NLMSG_GOODSIZE
, GFP_KERNEL
);
140 if (!adm_ctx
->reply_skb
) {
145 adm_ctx
->reply_dh
= genlmsg_put_reply(adm_ctx
->reply_skb
,
146 info
, &drbd_genl_family
, 0, cmd
);
147 /* put of a few bytes into a fresh skb of >= 4k will always succeed.
149 if (!adm_ctx
->reply_dh
) {
154 adm_ctx
->reply_dh
->minor
= d_in
->minor
;
155 adm_ctx
->reply_dh
->ret_code
= NO_ERROR
;
157 adm_ctx
->volume
= VOLUME_UNSPECIFIED
;
158 if (info
->attrs
[DRBD_NLA_CFG_CONTEXT
]) {
160 /* parse and validate only */
161 err
= drbd_cfg_context_from_attrs(NULL
, info
);
165 /* It was present, and valid,
166 * copy it over to the reply skb. */
167 err
= nla_put_nohdr(adm_ctx
->reply_skb
,
168 info
->attrs
[DRBD_NLA_CFG_CONTEXT
]->nla_len
,
169 info
->attrs
[DRBD_NLA_CFG_CONTEXT
]);
173 /* and assign stuff to the adm_ctx */
174 nla
= nested_attr_tb
[__nla_type(T_ctx_volume
)];
176 adm_ctx
->volume
= nla_get_u32(nla
);
177 nla
= nested_attr_tb
[__nla_type(T_ctx_resource_name
)];
179 adm_ctx
->resource_name
= nla_data(nla
);
180 adm_ctx
->my_addr
= nested_attr_tb
[__nla_type(T_ctx_my_addr
)];
181 adm_ctx
->peer_addr
= nested_attr_tb
[__nla_type(T_ctx_peer_addr
)];
182 if ((adm_ctx
->my_addr
&&
183 nla_len(adm_ctx
->my_addr
) > sizeof(adm_ctx
->connection
->my_addr
)) ||
184 (adm_ctx
->peer_addr
&&
185 nla_len(adm_ctx
->peer_addr
) > sizeof(adm_ctx
->connection
->peer_addr
))) {
191 adm_ctx
->minor
= d_in
->minor
;
192 adm_ctx
->device
= minor_to_device(d_in
->minor
);
194 /* We are protected by the global genl_lock().
195 * But we may explicitly drop it/retake it in drbd_adm_set_role(),
196 * so make sure this object stays around. */
198 kref_get(&adm_ctx
->device
->kref
);
200 if (adm_ctx
->resource_name
) {
201 adm_ctx
->resource
= drbd_find_resource(adm_ctx
->resource_name
);
204 if (!adm_ctx
->device
&& (flags
& DRBD_ADM_NEED_MINOR
)) {
205 drbd_msg_put_info(adm_ctx
->reply_skb
, "unknown minor");
206 return ERR_MINOR_INVALID
;
208 if (!adm_ctx
->resource
&& (flags
& DRBD_ADM_NEED_RESOURCE
)) {
209 drbd_msg_put_info(adm_ctx
->reply_skb
, "unknown resource");
210 if (adm_ctx
->resource_name
)
211 return ERR_RES_NOT_KNOWN
;
212 return ERR_INVALID_REQUEST
;
215 if (flags
& DRBD_ADM_NEED_CONNECTION
) {
216 if (adm_ctx
->resource
) {
217 drbd_msg_put_info(adm_ctx
->reply_skb
, "no resource name expected");
218 return ERR_INVALID_REQUEST
;
220 if (adm_ctx
->device
) {
221 drbd_msg_put_info(adm_ctx
->reply_skb
, "no minor number expected");
222 return ERR_INVALID_REQUEST
;
224 if (adm_ctx
->my_addr
&& adm_ctx
->peer_addr
)
225 adm_ctx
->connection
= conn_get_by_addrs(nla_data(adm_ctx
->my_addr
),
226 nla_len(adm_ctx
->my_addr
),
227 nla_data(adm_ctx
->peer_addr
),
228 nla_len(adm_ctx
->peer_addr
));
229 if (!adm_ctx
->connection
) {
230 drbd_msg_put_info(adm_ctx
->reply_skb
, "unknown connection");
231 return ERR_INVALID_REQUEST
;
235 /* some more paranoia, if the request was over-determined */
236 if (adm_ctx
->device
&& adm_ctx
->resource
&&
237 adm_ctx
->device
->resource
!= adm_ctx
->resource
) {
238 pr_warning("request: minor=%u, resource=%s; but that minor belongs to resource %s\n",
239 adm_ctx
->minor
, adm_ctx
->resource
->name
,
240 adm_ctx
->device
->resource
->name
);
241 drbd_msg_put_info(adm_ctx
->reply_skb
, "minor exists in different resource");
242 return ERR_INVALID_REQUEST
;
244 if (adm_ctx
->device
&&
245 adm_ctx
->volume
!= VOLUME_UNSPECIFIED
&&
246 adm_ctx
->volume
!= adm_ctx
->device
->vnr
) {
247 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
248 adm_ctx
->minor
, adm_ctx
->volume
,
249 adm_ctx
->device
->vnr
,
250 adm_ctx
->device
->resource
->name
);
251 drbd_msg_put_info(adm_ctx
->reply_skb
, "minor exists as different volume");
252 return ERR_INVALID_REQUEST
;
255 /* still, provide adm_ctx->resource always, if possible. */
256 if (!adm_ctx
->resource
) {
257 adm_ctx
->resource
= adm_ctx
->device
? adm_ctx
->device
->resource
258 : adm_ctx
->connection
? adm_ctx
->connection
->resource
: NULL
;
259 if (adm_ctx
->resource
)
260 kref_get(&adm_ctx
->resource
->kref
);
266 nlmsg_free(adm_ctx
->reply_skb
);
267 adm_ctx
->reply_skb
= NULL
;
271 static int drbd_adm_finish(struct drbd_config_context
*adm_ctx
,
272 struct genl_info
*info
, int retcode
)
274 if (adm_ctx
->device
) {
275 kref_put(&adm_ctx
->device
->kref
, drbd_destroy_device
);
276 adm_ctx
->device
= NULL
;
278 if (adm_ctx
->connection
) {
279 kref_put(&adm_ctx
->connection
->kref
, &drbd_destroy_connection
);
280 adm_ctx
->connection
= NULL
;
282 if (adm_ctx
->resource
) {
283 kref_put(&adm_ctx
->resource
->kref
, drbd_destroy_resource
);
284 adm_ctx
->resource
= NULL
;
287 if (!adm_ctx
->reply_skb
)
290 adm_ctx
->reply_dh
->ret_code
= retcode
;
291 drbd_adm_send_reply(adm_ctx
->reply_skb
, info
);
295 static void setup_khelper_env(struct drbd_connection
*connection
, char **envp
)
299 /* FIXME: A future version will not allow this case. */
300 if (connection
->my_addr_len
== 0 || connection
->peer_addr_len
== 0)
303 switch (((struct sockaddr
*)&connection
->peer_addr
)->sa_family
) {
306 snprintf(envp
[4], 60, "DRBD_PEER_ADDRESS=%pI6",
307 &((struct sockaddr_in6
*)&connection
->peer_addr
)->sin6_addr
);
311 snprintf(envp
[4], 60, "DRBD_PEER_ADDRESS=%pI4",
312 &((struct sockaddr_in
*)&connection
->peer_addr
)->sin_addr
);
316 snprintf(envp
[4], 60, "DRBD_PEER_ADDRESS=%pI4",
317 &((struct sockaddr_in
*)&connection
->peer_addr
)->sin_addr
);
319 snprintf(envp
[3], 20, "DRBD_PEER_AF=%s", afs
);
322 int drbd_khelper(struct drbd_device
*device
, char *cmd
)
324 char *envp
[] = { "HOME=/",
326 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
327 (char[20]) { }, /* address family */
328 (char[60]) { }, /* address */
331 char *argv
[] = {usermode_helper
, cmd
, mb
, NULL
};
332 struct drbd_connection
*connection
= first_peer_device(device
)->connection
;
336 if (current
== connection
->worker
.task
)
337 set_bit(CALLBACK_PENDING
, &connection
->flags
);
339 snprintf(mb
, 12, "minor-%d", device_to_minor(device
));
340 setup_khelper_env(connection
, envp
);
342 /* The helper may take some time.
343 * write out any unsynced meta data changes now */
344 drbd_md_sync(device
);
346 drbd_info(device
, "helper command: %s %s %s\n", usermode_helper
, cmd
, mb
);
347 sib
.sib_reason
= SIB_HELPER_PRE
;
348 sib
.helper_name
= cmd
;
349 drbd_bcast_event(device
, &sib
);
350 ret
= call_usermodehelper(usermode_helper
, argv
, envp
, UMH_WAIT_PROC
);
352 drbd_warn(device
, "helper command: %s %s %s exit code %u (0x%x)\n",
353 usermode_helper
, cmd
, mb
,
354 (ret
>> 8) & 0xff, ret
);
356 drbd_info(device
, "helper command: %s %s %s exit code %u (0x%x)\n",
357 usermode_helper
, cmd
, mb
,
358 (ret
>> 8) & 0xff, ret
);
359 sib
.sib_reason
= SIB_HELPER_POST
;
360 sib
.helper_exit_code
= ret
;
361 drbd_bcast_event(device
, &sib
);
363 if (current
== connection
->worker
.task
)
364 clear_bit(CALLBACK_PENDING
, &connection
->flags
);
366 if (ret
< 0) /* Ignore any ERRNOs we got. */
372 static int conn_khelper(struct drbd_connection
*connection
, char *cmd
)
374 char *envp
[] = { "HOME=/",
376 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
377 (char[20]) { }, /* address family */
378 (char[60]) { }, /* address */
380 char *resource_name
= connection
->resource
->name
;
381 char *argv
[] = {usermode_helper
, cmd
, resource_name
, NULL
};
384 setup_khelper_env(connection
, envp
);
385 conn_md_sync(connection
);
387 drbd_info(connection
, "helper command: %s %s %s\n", usermode_helper
, cmd
, resource_name
);
388 /* TODO: conn_bcast_event() ?? */
390 ret
= call_usermodehelper(usermode_helper
, argv
, envp
, UMH_WAIT_PROC
);
392 drbd_warn(connection
, "helper command: %s %s %s exit code %u (0x%x)\n",
393 usermode_helper
, cmd
, resource_name
,
394 (ret
>> 8) & 0xff, ret
);
396 drbd_info(connection
, "helper command: %s %s %s exit code %u (0x%x)\n",
397 usermode_helper
, cmd
, resource_name
,
398 (ret
>> 8) & 0xff, ret
);
399 /* TODO: conn_bcast_event() ?? */
401 if (ret
< 0) /* Ignore any ERRNOs we got. */
407 static enum drbd_fencing_p
highest_fencing_policy(struct drbd_connection
*connection
)
409 enum drbd_fencing_p fp
= FP_NOT_AVAIL
;
410 struct drbd_peer_device
*peer_device
;
414 idr_for_each_entry(&connection
->peer_devices
, peer_device
, vnr
) {
415 struct drbd_device
*device
= peer_device
->device
;
416 if (get_ldev_if_state(device
, D_CONSISTENT
)) {
417 struct disk_conf
*disk_conf
=
418 rcu_dereference(peer_device
->device
->ldev
->disk_conf
);
419 fp
= max_t(enum drbd_fencing_p
, fp
, disk_conf
->fencing
);
425 if (fp
== FP_NOT_AVAIL
) {
426 /* IO Suspending works on the whole resource.
427 Do it only for one device. */
429 peer_device
= idr_get_next(&connection
->peer_devices
, &vnr
);
430 drbd_change_state(peer_device
->device
, CS_VERBOSE
| CS_HARD
, NS(susp_fen
, 0));
436 bool conn_try_outdate_peer(struct drbd_connection
*connection
)
438 unsigned int connect_cnt
;
439 union drbd_state mask
= { };
440 union drbd_state val
= { };
441 enum drbd_fencing_p fp
;
445 spin_lock_irq(&connection
->resource
->req_lock
);
446 if (connection
->cstate
>= C_WF_REPORT_PARAMS
) {
447 drbd_err(connection
, "Expected cstate < C_WF_REPORT_PARAMS\n");
448 spin_unlock_irq(&connection
->resource
->req_lock
);
452 connect_cnt
= connection
->connect_cnt
;
453 spin_unlock_irq(&connection
->resource
->req_lock
);
455 fp
= highest_fencing_policy(connection
);
458 drbd_warn(connection
, "Not fencing peer, I'm not even Consistent myself.\n");
465 r
= conn_khelper(connection
, "fence-peer");
467 switch ((r
>>8) & 0xff) {
468 case 3: /* peer is inconsistent */
469 ex_to_string
= "peer is inconsistent or worse";
471 val
.pdsk
= D_INCONSISTENT
;
473 case 4: /* peer got outdated, or was already outdated */
474 ex_to_string
= "peer was fenced";
476 val
.pdsk
= D_OUTDATED
;
478 case 5: /* peer was down */
479 if (conn_highest_disk(connection
) == D_UP_TO_DATE
) {
480 /* we will(have) create(d) a new UUID anyways... */
481 ex_to_string
= "peer is unreachable, assumed to be dead";
483 val
.pdsk
= D_OUTDATED
;
485 ex_to_string
= "peer unreachable, doing nothing since disk != UpToDate";
488 case 6: /* Peer is primary, voluntarily outdate myself.
489 * This is useful when an unconnected R_SECONDARY is asked to
490 * become R_PRIMARY, but finds the other peer being active. */
491 ex_to_string
= "peer is active";
492 drbd_warn(connection
, "Peer is primary, outdating myself.\n");
494 val
.disk
= D_OUTDATED
;
497 if (fp
!= FP_STONITH
)
498 drbd_err(connection
, "fence-peer() = 7 && fencing != Stonith !!!\n");
499 ex_to_string
= "peer was stonithed";
501 val
.pdsk
= D_OUTDATED
;
504 /* The script is broken ... */
505 drbd_err(connection
, "fence-peer helper broken, returned %d\n", (r
>>8)&0xff);
506 return false; /* Eventually leave IO frozen */
509 drbd_info(connection
, "fence-peer helper returned %d (%s)\n",
510 (r
>>8) & 0xff, ex_to_string
);
515 conn_request_state(connection, mask, val, CS_VERBOSE);
516 here, because we might were able to re-establish the connection in the
518 spin_lock_irq(&connection
->resource
->req_lock
);
519 if (connection
->cstate
< C_WF_REPORT_PARAMS
&& !test_bit(STATE_SENT
, &connection
->flags
)) {
520 if (connection
->connect_cnt
!= connect_cnt
)
521 /* In case the connection was established and droped
522 while the fence-peer handler was running, ignore it */
523 drbd_info(connection
, "Ignoring fence-peer exit code\n");
525 _conn_request_state(connection
, mask
, val
, CS_VERBOSE
);
527 spin_unlock_irq(&connection
->resource
->req_lock
);
529 return conn_highest_pdsk(connection
) <= D_OUTDATED
;
532 static int _try_outdate_peer_async(void *data
)
534 struct drbd_connection
*connection
= (struct drbd_connection
*)data
;
536 conn_try_outdate_peer(connection
);
538 kref_put(&connection
->kref
, drbd_destroy_connection
);
542 void conn_try_outdate_peer_async(struct drbd_connection
*connection
)
544 struct task_struct
*opa
;
546 kref_get(&connection
->kref
);
547 opa
= kthread_run(_try_outdate_peer_async
, connection
, "drbd_async_h");
549 drbd_err(connection
, "out of mem, failed to invoke fence-peer helper\n");
550 kref_put(&connection
->kref
, drbd_destroy_connection
);
555 drbd_set_role(struct drbd_device
*device
, enum drbd_role new_role
, int force
)
557 const int max_tries
= 4;
558 enum drbd_state_rv rv
= SS_UNKNOWN_ERROR
;
562 union drbd_state mask
, val
;
564 if (new_role
== R_PRIMARY
) {
565 struct drbd_connection
*connection
;
567 /* Detect dead peers as soon as possible. */
570 for_each_connection(connection
, device
->resource
)
571 request_ping(connection
);
575 mutex_lock(device
->state_mutex
);
577 mask
.i
= 0; mask
.role
= R_MASK
;
578 val
.i
= 0; val
.role
= new_role
;
580 while (try++ < max_tries
) {
581 rv
= _drbd_request_state(device
, mask
, val
, CS_WAIT_COMPLETE
);
583 /* in case we first succeeded to outdate,
584 * but now suddenly could establish a connection */
585 if (rv
== SS_CW_FAILED_BY_PEER
&& mask
.pdsk
!= 0) {
591 if (rv
== SS_NO_UP_TO_DATE_DISK
&& force
&&
592 (device
->state
.disk
< D_UP_TO_DATE
&&
593 device
->state
.disk
>= D_INCONSISTENT
)) {
595 val
.disk
= D_UP_TO_DATE
;
600 if (rv
== SS_NO_UP_TO_DATE_DISK
&&
601 device
->state
.disk
== D_CONSISTENT
&& mask
.pdsk
== 0) {
602 D_ASSERT(device
, device
->state
.pdsk
== D_UNKNOWN
);
604 if (conn_try_outdate_peer(first_peer_device(device
)->connection
)) {
605 val
.disk
= D_UP_TO_DATE
;
611 if (rv
== SS_NOTHING_TO_DO
)
613 if (rv
== SS_PRIMARY_NOP
&& mask
.pdsk
== 0) {
614 if (!conn_try_outdate_peer(first_peer_device(device
)->connection
) && force
) {
615 drbd_warn(device
, "Forced into split brain situation!\n");
617 val
.pdsk
= D_OUTDATED
;
622 if (rv
== SS_TWO_PRIMARIES
) {
623 /* Maybe the peer is detected as dead very soon...
624 retry at most once more in this case. */
627 nc
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
);
628 timeo
= nc
? (nc
->ping_timeo
+ 1) * HZ
/ 10 : 1;
630 schedule_timeout_interruptible(timeo
);
635 if (rv
< SS_SUCCESS
) {
636 rv
= _drbd_request_state(device
, mask
, val
,
637 CS_VERBOSE
+ CS_WAIT_COMPLETE
);
648 drbd_warn(device
, "Forced to consider local data as UpToDate!\n");
650 /* Wait until nothing is on the fly :) */
651 wait_event(device
->misc_wait
, atomic_read(&device
->ap_pending_cnt
) == 0);
653 /* FIXME also wait for all pending P_BARRIER_ACK? */
655 if (new_role
== R_SECONDARY
) {
656 set_disk_ro(device
->vdisk
, true);
657 if (get_ldev(device
)) {
658 device
->ldev
->md
.uuid
[UI_CURRENT
] &= ~(u64
)1;
662 /* Called from drbd_adm_set_role only.
663 * We are still holding the conf_update mutex. */
664 nc
= first_peer_device(device
)->connection
->net_conf
;
666 nc
->discard_my_data
= 0; /* without copy; single bit op is atomic */
668 set_disk_ro(device
->vdisk
, false);
669 if (get_ldev(device
)) {
670 if (((device
->state
.conn
< C_CONNECTED
||
671 device
->state
.pdsk
<= D_FAILED
)
672 && device
->ldev
->md
.uuid
[UI_BITMAP
] == 0) || forced
)
673 drbd_uuid_new_current(device
);
675 device
->ldev
->md
.uuid
[UI_CURRENT
] |= (u64
)1;
680 /* writeout of activity log covered areas of the bitmap
681 * to stable storage done in after state change already */
683 if (device
->state
.conn
>= C_WF_REPORT_PARAMS
) {
684 /* if this was forced, we should consider sync */
686 drbd_send_uuids(first_peer_device(device
));
687 drbd_send_current_state(first_peer_device(device
));
690 drbd_md_sync(device
);
692 kobject_uevent(&disk_to_dev(device
->vdisk
)->kobj
, KOBJ_CHANGE
);
694 mutex_unlock(device
->state_mutex
);
698 static const char *from_attrs_err_to_txt(int err
)
700 return err
== -ENOMSG
? "required attribute missing" :
701 err
== -EOPNOTSUPP
? "unknown mandatory attribute" :
702 err
== -EEXIST
? "can not change invariant setting" :
703 "invalid attribute value";
706 int drbd_adm_set_role(struct sk_buff
*skb
, struct genl_info
*info
)
708 struct drbd_config_context adm_ctx
;
709 struct set_role_parms parms
;
711 enum drbd_ret_code retcode
;
713 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
714 if (!adm_ctx
.reply_skb
)
716 if (retcode
!= NO_ERROR
)
719 memset(&parms
, 0, sizeof(parms
));
720 if (info
->attrs
[DRBD_NLA_SET_ROLE_PARMS
]) {
721 err
= set_role_parms_from_attrs(&parms
, info
);
723 retcode
= ERR_MANDATORY_TAG
;
724 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
729 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
731 if (info
->genlhdr
->cmd
== DRBD_ADM_PRIMARY
)
732 retcode
= drbd_set_role(adm_ctx
.device
, R_PRIMARY
, parms
.assume_uptodate
);
734 retcode
= drbd_set_role(adm_ctx
.device
, R_SECONDARY
, 0);
736 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
739 drbd_adm_finish(&adm_ctx
, info
, retcode
);
743 /* Initializes the md.*_offset members, so we are able to find
744 * the on disk meta data.
746 * We currently have two possible layouts:
748 * |----------- md_size_sect ------------------|
749 * [ 4k superblock ][ activity log ][ Bitmap ]
751 * | bm_offset = al_offset + X |
752 * ==> bitmap sectors = md_size_sect - bm_offset
755 * |----------- md_size_sect ------------------|
756 * [data.....][ Bitmap ][ activity log ][ 4k superblock ]
758 * | bm_offset = al_offset - Y |
759 * ==> bitmap sectors = Y = al_offset - bm_offset
761 * Activity log size used to be fixed 32kB,
762 * but is about to become configurable.
764 static void drbd_md_set_sector_offsets(struct drbd_device
*device
,
765 struct drbd_backing_dev
*bdev
)
767 sector_t md_size_sect
= 0;
768 unsigned int al_size_sect
= bdev
->md
.al_size_4k
* 8;
770 bdev
->md
.md_offset
= drbd_md_ss(bdev
);
772 switch (bdev
->md
.meta_dev_idx
) {
774 /* v07 style fixed size indexed meta data */
775 bdev
->md
.md_size_sect
= MD_128MB_SECT
;
776 bdev
->md
.al_offset
= MD_4kB_SECT
;
777 bdev
->md
.bm_offset
= MD_4kB_SECT
+ al_size_sect
;
779 case DRBD_MD_INDEX_FLEX_EXT
:
780 /* just occupy the full device; unit: sectors */
781 bdev
->md
.md_size_sect
= drbd_get_capacity(bdev
->md_bdev
);
782 bdev
->md
.al_offset
= MD_4kB_SECT
;
783 bdev
->md
.bm_offset
= MD_4kB_SECT
+ al_size_sect
;
785 case DRBD_MD_INDEX_INTERNAL
:
786 case DRBD_MD_INDEX_FLEX_INT
:
787 /* al size is still fixed */
788 bdev
->md
.al_offset
= -al_size_sect
;
789 /* we need (slightly less than) ~ this much bitmap sectors: */
790 md_size_sect
= drbd_get_capacity(bdev
->backing_bdev
);
791 md_size_sect
= ALIGN(md_size_sect
, BM_SECT_PER_EXT
);
792 md_size_sect
= BM_SECT_TO_EXT(md_size_sect
);
793 md_size_sect
= ALIGN(md_size_sect
, 8);
795 /* plus the "drbd meta data super block",
796 * and the activity log; */
797 md_size_sect
+= MD_4kB_SECT
+ al_size_sect
;
799 bdev
->md
.md_size_sect
= md_size_sect
;
800 /* bitmap offset is adjusted by 'super' block size */
801 bdev
->md
.bm_offset
= -md_size_sect
+ MD_4kB_SECT
;
806 /* input size is expected to be in KB */
807 char *ppsize(char *buf
, unsigned long long size
)
809 /* Needs 9 bytes at max including trailing NUL:
810 * -1ULL ==> "16384 EB" */
811 static char units
[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
813 while (size
>= 10000 && base
< sizeof(units
)-1) {
815 size
= (size
>> 10) + !!(size
& (1<<9));
818 sprintf(buf
, "%u %cB", (unsigned)size
, units
[base
]);
823 /* there is still a theoretical deadlock when called from receiver
824 * on an D_INCONSISTENT R_PRIMARY:
825 * remote READ does inc_ap_bio, receiver would need to receive answer
826 * packet from remote to dec_ap_bio again.
827 * receiver receive_sizes(), comes here,
828 * waits for ap_bio_cnt == 0. -> deadlock.
829 * but this cannot happen, actually, because:
830 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
831 * (not connected, or bad/no disk on peer):
832 * see drbd_fail_request_early, ap_bio_cnt is zero.
833 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
834 * peer may not initiate a resize.
836 /* Note these are not to be confused with
837 * drbd_adm_suspend_io/drbd_adm_resume_io,
838 * which are (sub) state changes triggered by admin (drbdsetup),
839 * and can be long lived.
840 * This changes an device->flag, is triggered by drbd internals,
841 * and should be short-lived. */
842 void drbd_suspend_io(struct drbd_device
*device
)
844 set_bit(SUSPEND_IO
, &device
->flags
);
845 if (drbd_suspended(device
))
847 wait_event(device
->misc_wait
, !atomic_read(&device
->ap_bio_cnt
));
850 void drbd_resume_io(struct drbd_device
*device
)
852 clear_bit(SUSPEND_IO
, &device
->flags
);
853 wake_up(&device
->misc_wait
);
857 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
858 * @device: DRBD device.
860 * Returns 0 on success, negative return values indicate errors.
861 * You should call drbd_md_sync() after calling this function.
863 enum determine_dev_size
864 drbd_determine_dev_size(struct drbd_device
*device
, enum dds_flags flags
, struct resize_parms
*rs
) __must_hold(local
)
866 sector_t prev_first_sect
, prev_size
; /* previous meta location */
867 sector_t la_size_sect
, u_size
;
868 struct drbd_md
*md
= &device
->ldev
->md
;
869 u32 prev_al_stripe_size_4k
;
875 int md_moved
, la_size_changed
;
876 enum determine_dev_size rv
= DS_UNCHANGED
;
879 * application request passes inc_ap_bio,
880 * but then cannot get an AL-reference.
881 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
884 * Suspend IO right here.
885 * still lock the act_log to not trigger ASSERTs there.
887 drbd_suspend_io(device
);
888 buffer
= drbd_md_get_buffer(device
); /* Lock meta-data IO */
890 drbd_resume_io(device
);
894 /* no wait necessary anymore, actually we could assert that */
895 wait_event(device
->al_wait
, lc_try_lock(device
->act_log
));
897 prev_first_sect
= drbd_md_first_sector(device
->ldev
);
898 prev_size
= device
->ldev
->md
.md_size_sect
;
899 la_size_sect
= device
->ldev
->md
.la_size_sect
;
902 /* rs is non NULL if we should change the AL layout only */
904 prev_al_stripes
= md
->al_stripes
;
905 prev_al_stripe_size_4k
= md
->al_stripe_size_4k
;
907 md
->al_stripes
= rs
->al_stripes
;
908 md
->al_stripe_size_4k
= rs
->al_stripe_size
/ 4;
909 md
->al_size_4k
= (u64
)rs
->al_stripes
* rs
->al_stripe_size
/ 4;
912 drbd_md_set_sector_offsets(device
, device
->ldev
);
915 u_size
= rcu_dereference(device
->ldev
->disk_conf
)->disk_size
;
917 size
= drbd_new_dev_size(device
, device
->ldev
, u_size
, flags
& DDSF_FORCED
);
919 if (size
< la_size_sect
) {
920 if (rs
&& u_size
== 0) {
921 /* Remove "rs &&" later. This check should always be active, but
922 right now the receiver expects the permissive behavior */
923 drbd_warn(device
, "Implicit shrink not allowed. "
924 "Use --size=%llus for explicit shrink.\n",
925 (unsigned long long)size
);
926 rv
= DS_ERROR_SHRINK
;
929 rv
= DS_ERROR_SPACE_MD
;
930 if (rv
!= DS_UNCHANGED
)
934 if (drbd_get_capacity(device
->this_bdev
) != size
||
935 drbd_bm_capacity(device
) != size
) {
937 err
= drbd_bm_resize(device
, size
, !(flags
& DDSF_NO_RESYNC
));
939 /* currently there is only one error: ENOMEM! */
940 size
= drbd_bm_capacity(device
)>>1;
942 drbd_err(device
, "OUT OF MEMORY! "
943 "Could not allocate bitmap!\n");
945 drbd_err(device
, "BM resizing failed. "
946 "Leaving size unchanged at size = %lu KB\n",
947 (unsigned long)size
);
951 /* racy, see comments above. */
952 drbd_set_my_capacity(device
, size
);
953 device
->ldev
->md
.la_size_sect
= size
;
954 drbd_info(device
, "size = %s (%llu KB)\n", ppsize(ppb
, size
>>1),
955 (unsigned long long)size
>>1);
960 la_size_changed
= (la_size_sect
!= device
->ldev
->md
.la_size_sect
);
962 md_moved
= prev_first_sect
!= drbd_md_first_sector(device
->ldev
)
963 || prev_size
!= device
->ldev
->md
.md_size_sect
;
965 if (la_size_changed
|| md_moved
|| rs
) {
968 drbd_al_shrink(device
); /* All extents inactive. */
970 prev_flags
= md
->flags
;
971 md
->flags
&= ~MDF_PRIMARY_IND
;
972 drbd_md_write(device
, buffer
);
974 drbd_info(device
, "Writing the whole bitmap, %s\n",
975 la_size_changed
&& md_moved
? "size changed and md moved" :
976 la_size_changed
? "size changed" : "md moved");
977 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
978 drbd_bitmap_io(device
, md_moved
? &drbd_bm_write_all
: &drbd_bm_write
,
979 "size changed", BM_LOCKED_MASK
);
980 drbd_initialize_al(device
, buffer
);
982 md
->flags
= prev_flags
;
983 drbd_md_write(device
, buffer
);
986 drbd_info(device
, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n",
987 md
->al_stripes
, md
->al_stripe_size_4k
* 4);
990 if (size
> la_size_sect
)
991 rv
= la_size_sect
? DS_GREW
: DS_GREW_FROM_ZERO
;
992 if (size
< la_size_sect
)
998 md
->al_stripes
= prev_al_stripes
;
999 md
->al_stripe_size_4k
= prev_al_stripe_size_4k
;
1000 md
->al_size_4k
= (u64
)prev_al_stripes
* prev_al_stripe_size_4k
;
1002 drbd_md_set_sector_offsets(device
, device
->ldev
);
1005 lc_unlock(device
->act_log
);
1006 wake_up(&device
->al_wait
);
1007 drbd_md_put_buffer(device
);
1008 drbd_resume_io(device
);
1014 drbd_new_dev_size(struct drbd_device
*device
, struct drbd_backing_dev
*bdev
,
1015 sector_t u_size
, int assume_peer_has_space
)
1017 sector_t p_size
= device
->p_size
; /* partner's disk size. */
1018 sector_t la_size_sect
= bdev
->md
.la_size_sect
; /* last agreed size. */
1019 sector_t m_size
; /* my size */
1022 m_size
= drbd_get_max_capacity(bdev
);
1024 if (device
->state
.conn
< C_CONNECTED
&& assume_peer_has_space
) {
1025 drbd_warn(device
, "Resize while not connected was forced by the user!\n");
1029 if (p_size
&& m_size
) {
1030 size
= min_t(sector_t
, p_size
, m_size
);
1033 size
= la_size_sect
;
1034 if (m_size
&& m_size
< size
)
1036 if (p_size
&& p_size
< size
)
1047 drbd_err(device
, "Both nodes diskless!\n");
1051 drbd_err(device
, "Requested disk size is too big (%lu > %lu)\n",
1052 (unsigned long)u_size
>>1, (unsigned long)size
>>1);
1061 * drbd_check_al_size() - Ensures that the AL is of the right size
1062 * @device: DRBD device.
1064 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
1065 * failed, and 0 on success. You should call drbd_md_sync() after you called
1068 static int drbd_check_al_size(struct drbd_device
*device
, struct disk_conf
*dc
)
1070 struct lru_cache
*n
, *t
;
1071 struct lc_element
*e
;
1072 unsigned int in_use
;
1075 if (device
->act_log
&&
1076 device
->act_log
->nr_elements
== dc
->al_extents
)
1080 t
= device
->act_log
;
1081 n
= lc_create("act_log", drbd_al_ext_cache
, AL_UPDATES_PER_TRANSACTION
,
1082 dc
->al_extents
, sizeof(struct lc_element
), 0);
1085 drbd_err(device
, "Cannot allocate act_log lru!\n");
1088 spin_lock_irq(&device
->al_lock
);
1090 for (i
= 0; i
< t
->nr_elements
; i
++) {
1091 e
= lc_element_by_index(t
, i
);
1093 drbd_err(device
, "refcnt(%d)==%d\n",
1094 e
->lc_number
, e
->refcnt
);
1095 in_use
+= e
->refcnt
;
1099 device
->act_log
= n
;
1100 spin_unlock_irq(&device
->al_lock
);
1102 drbd_err(device
, "Activity log still in use!\n");
1109 drbd_md_mark_dirty(device
); /* we changed device->act_log->nr_elemens */
1113 static void drbd_setup_queue_param(struct drbd_device
*device
, unsigned int max_bio_size
)
1115 struct request_queue
* const q
= device
->rq_queue
;
1116 unsigned int max_hw_sectors
= max_bio_size
>> 9;
1117 unsigned int max_segments
= 0;
1118 struct request_queue
*b
= NULL
;
1120 if (get_ldev_if_state(device
, D_ATTACHING
)) {
1121 b
= device
->ldev
->backing_bdev
->bd_disk
->queue
;
1123 max_hw_sectors
= min(queue_max_hw_sectors(b
), max_bio_size
>> 9);
1125 max_segments
= rcu_dereference(device
->ldev
->disk_conf
)->max_bio_bvecs
;
1128 blk_set_stacking_limits(&q
->limits
);
1129 blk_queue_max_write_same_sectors(q
, 0);
1132 blk_queue_logical_block_size(q
, 512);
1133 blk_queue_max_hw_sectors(q
, max_hw_sectors
);
1134 /* This is the workaround for "bio would need to, but cannot, be split" */
1135 blk_queue_max_segments(q
, max_segments
? max_segments
: BLK_MAX_SEGMENTS
);
1136 blk_queue_segment_boundary(q
, PAGE_CACHE_SIZE
-1);
1139 struct drbd_connection
*connection
= first_peer_device(device
)->connection
;
1141 if (blk_queue_discard(b
) &&
1142 (connection
->cstate
< C_CONNECTED
|| connection
->agreed_features
& FF_TRIM
)) {
1143 /* For now, don't allow more than one activity log extent worth of data
1144 * to be discarded in one go. We may need to rework drbd_al_begin_io()
1145 * to allow for even larger discard ranges */
1146 q
->limits
.max_discard_sectors
= DRBD_MAX_DISCARD_SECTORS
;
1148 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
1149 /* REALLY? Is stacking secdiscard "legal"? */
1150 if (blk_queue_secdiscard(b
))
1151 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD
, q
);
1153 q
->limits
.max_discard_sectors
= 0;
1154 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
1155 queue_flag_clear_unlocked(QUEUE_FLAG_SECDISCARD
, q
);
1158 blk_queue_stack_limits(q
, b
);
1160 if (q
->backing_dev_info
.ra_pages
!= b
->backing_dev_info
.ra_pages
) {
1161 drbd_info(device
, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1162 q
->backing_dev_info
.ra_pages
,
1163 b
->backing_dev_info
.ra_pages
);
1164 q
->backing_dev_info
.ra_pages
= b
->backing_dev_info
.ra_pages
;
1170 void drbd_reconsider_max_bio_size(struct drbd_device
*device
)
1172 unsigned int now
, new, local
, peer
;
1174 now
= queue_max_hw_sectors(device
->rq_queue
) << 9;
1175 local
= device
->local_max_bio_size
; /* Eventually last known value, from volatile memory */
1176 peer
= device
->peer_max_bio_size
; /* Eventually last known value, from meta data */
1178 if (get_ldev_if_state(device
, D_ATTACHING
)) {
1179 local
= queue_max_hw_sectors(device
->ldev
->backing_bdev
->bd_disk
->queue
) << 9;
1180 device
->local_max_bio_size
= local
;
1183 local
= min(local
, DRBD_MAX_BIO_SIZE
);
1185 /* We may ignore peer limits if the peer is modern enough.
1186 Because new from 8.3.8 onwards the peer can use multiple
1187 BIOs for a single peer_request */
1188 if (device
->state
.conn
>= C_WF_REPORT_PARAMS
) {
1189 if (first_peer_device(device
)->connection
->agreed_pro_version
< 94)
1190 peer
= min(device
->peer_max_bio_size
, DRBD_MAX_SIZE_H80_PACKET
);
1191 /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1192 else if (first_peer_device(device
)->connection
->agreed_pro_version
== 94)
1193 peer
= DRBD_MAX_SIZE_H80_PACKET
;
1194 else if (first_peer_device(device
)->connection
->agreed_pro_version
< 100)
1195 peer
= DRBD_MAX_BIO_SIZE_P95
; /* drbd 8.3.8 onwards, before 8.4.0 */
1197 peer
= DRBD_MAX_BIO_SIZE
;
1199 /* We may later detach and re-attach on a disconnected Primary.
1200 * Avoid this setting to jump back in that case.
1201 * We want to store what we know the peer DRBD can handle,
1202 * not what the peer IO backend can handle. */
1203 if (peer
> device
->peer_max_bio_size
)
1204 device
->peer_max_bio_size
= peer
;
1206 new = min(local
, peer
);
1208 if (device
->state
.role
== R_PRIMARY
&& new < now
)
1209 drbd_err(device
, "ASSERT FAILED new < now; (%u < %u)\n", new, now
);
1212 drbd_info(device
, "max BIO size = %u\n", new);
1214 drbd_setup_queue_param(device
, new);
1217 /* Starts the worker thread */
1218 static void conn_reconfig_start(struct drbd_connection
*connection
)
1220 drbd_thread_start(&connection
->worker
);
1221 drbd_flush_workqueue(&connection
->sender_work
);
1224 /* if still unconfigured, stops worker again. */
1225 static void conn_reconfig_done(struct drbd_connection
*connection
)
1228 spin_lock_irq(&connection
->resource
->req_lock
);
1229 stop_threads
= conn_all_vols_unconf(connection
) &&
1230 connection
->cstate
== C_STANDALONE
;
1231 spin_unlock_irq(&connection
->resource
->req_lock
);
1233 /* asender is implicitly stopped by receiver
1234 * in conn_disconnect() */
1235 drbd_thread_stop(&connection
->receiver
);
1236 drbd_thread_stop(&connection
->worker
);
1240 /* Make sure IO is suspended before calling this function(). */
1241 static void drbd_suspend_al(struct drbd_device
*device
)
1245 if (!lc_try_lock(device
->act_log
)) {
1246 drbd_warn(device
, "Failed to lock al in drbd_suspend_al()\n");
1250 drbd_al_shrink(device
);
1251 spin_lock_irq(&device
->resource
->req_lock
);
1252 if (device
->state
.conn
< C_CONNECTED
)
1253 s
= !test_and_set_bit(AL_SUSPENDED
, &device
->flags
);
1254 spin_unlock_irq(&device
->resource
->req_lock
);
1255 lc_unlock(device
->act_log
);
1258 drbd_info(device
, "Suspended AL updates\n");
1262 static bool should_set_defaults(struct genl_info
*info
)
1264 unsigned flags
= ((struct drbd_genlmsghdr
*)info
->userhdr
)->flags
;
1265 return 0 != (flags
& DRBD_GENL_F_SET_DEFAULTS
);
1268 static unsigned int drbd_al_extents_max(struct drbd_backing_dev
*bdev
)
1270 /* This is limited by 16 bit "slot" numbers,
1271 * and by available on-disk context storage.
1273 * Also (u16)~0 is special (denotes a "free" extent).
1275 * One transaction occupies one 4kB on-disk block,
1276 * we have n such blocks in the on disk ring buffer,
1277 * the "current" transaction may fail (n-1),
1278 * and there is 919 slot numbers context information per transaction.
1280 * 72 transaction blocks amounts to more than 2**16 context slots,
1281 * so cap there first.
1283 const unsigned int max_al_nr
= DRBD_AL_EXTENTS_MAX
;
1284 const unsigned int sufficient_on_disk
=
1285 (max_al_nr
+ AL_CONTEXT_PER_TRANSACTION
-1)
1286 /AL_CONTEXT_PER_TRANSACTION
;
1288 unsigned int al_size_4k
= bdev
->md
.al_size_4k
;
1290 if (al_size_4k
> sufficient_on_disk
)
1293 return (al_size_4k
- 1) * AL_CONTEXT_PER_TRANSACTION
;
1296 int drbd_adm_disk_opts(struct sk_buff
*skb
, struct genl_info
*info
)
1298 struct drbd_config_context adm_ctx
;
1299 enum drbd_ret_code retcode
;
1300 struct drbd_device
*device
;
1301 struct disk_conf
*new_disk_conf
, *old_disk_conf
;
1302 struct fifo_buffer
*old_plan
= NULL
, *new_plan
= NULL
;
1305 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
1306 if (!adm_ctx
.reply_skb
)
1308 if (retcode
!= NO_ERROR
)
1311 device
= adm_ctx
.device
;
1312 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
1314 /* we also need a disk
1315 * to change the options on */
1316 if (!get_ldev(device
)) {
1317 retcode
= ERR_NO_DISK
;
1321 new_disk_conf
= kmalloc(sizeof(struct disk_conf
), GFP_KERNEL
);
1322 if (!new_disk_conf
) {
1323 retcode
= ERR_NOMEM
;
1327 mutex_lock(&device
->resource
->conf_update
);
1328 old_disk_conf
= device
->ldev
->disk_conf
;
1329 *new_disk_conf
= *old_disk_conf
;
1330 if (should_set_defaults(info
))
1331 set_disk_conf_defaults(new_disk_conf
);
1333 err
= disk_conf_from_attrs_for_change(new_disk_conf
, info
);
1334 if (err
&& err
!= -ENOMSG
) {
1335 retcode
= ERR_MANDATORY_TAG
;
1336 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
1340 if (!expect(new_disk_conf
->resync_rate
>= 1))
1341 new_disk_conf
->resync_rate
= 1;
1343 if (new_disk_conf
->al_extents
< DRBD_AL_EXTENTS_MIN
)
1344 new_disk_conf
->al_extents
= DRBD_AL_EXTENTS_MIN
;
1345 if (new_disk_conf
->al_extents
> drbd_al_extents_max(device
->ldev
))
1346 new_disk_conf
->al_extents
= drbd_al_extents_max(device
->ldev
);
1348 if (new_disk_conf
->c_plan_ahead
> DRBD_C_PLAN_AHEAD_MAX
)
1349 new_disk_conf
->c_plan_ahead
= DRBD_C_PLAN_AHEAD_MAX
;
1351 fifo_size
= (new_disk_conf
->c_plan_ahead
* 10 * SLEEP_TIME
) / HZ
;
1352 if (fifo_size
!= device
->rs_plan_s
->size
) {
1353 new_plan
= fifo_alloc(fifo_size
);
1355 drbd_err(device
, "kmalloc of fifo_buffer failed");
1356 retcode
= ERR_NOMEM
;
1361 drbd_suspend_io(device
);
1362 wait_event(device
->al_wait
, lc_try_lock(device
->act_log
));
1363 drbd_al_shrink(device
);
1364 err
= drbd_check_al_size(device
, new_disk_conf
);
1365 lc_unlock(device
->act_log
);
1366 wake_up(&device
->al_wait
);
1367 drbd_resume_io(device
);
1370 retcode
= ERR_NOMEM
;
1374 write_lock_irq(&global_state_lock
);
1375 retcode
= drbd_resync_after_valid(device
, new_disk_conf
->resync_after
);
1376 if (retcode
== NO_ERROR
) {
1377 rcu_assign_pointer(device
->ldev
->disk_conf
, new_disk_conf
);
1378 drbd_resync_after_changed(device
);
1380 write_unlock_irq(&global_state_lock
);
1382 if (retcode
!= NO_ERROR
)
1386 old_plan
= device
->rs_plan_s
;
1387 rcu_assign_pointer(device
->rs_plan_s
, new_plan
);
1390 mutex_unlock(&device
->resource
->conf_update
);
1392 if (new_disk_conf
->al_updates
)
1393 device
->ldev
->md
.flags
&= ~MDF_AL_DISABLED
;
1395 device
->ldev
->md
.flags
|= MDF_AL_DISABLED
;
1397 if (new_disk_conf
->md_flushes
)
1398 clear_bit(MD_NO_FUA
, &device
->flags
);
1400 set_bit(MD_NO_FUA
, &device
->flags
);
1402 drbd_bump_write_ordering(first_peer_device(device
)->connection
, WO_bdev_flush
);
1404 drbd_md_sync(device
);
1406 if (device
->state
.conn
>= C_CONNECTED
) {
1407 struct drbd_peer_device
*peer_device
;
1409 for_each_peer_device(peer_device
, device
)
1410 drbd_send_sync_param(peer_device
);
1414 kfree(old_disk_conf
);
1416 mod_timer(&device
->request_timer
, jiffies
+ HZ
);
1420 mutex_unlock(&device
->resource
->conf_update
);
1422 kfree(new_disk_conf
);
1427 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
1429 drbd_adm_finish(&adm_ctx
, info
, retcode
);
1433 int drbd_adm_attach(struct sk_buff
*skb
, struct genl_info
*info
)
1435 struct drbd_config_context adm_ctx
;
1436 struct drbd_device
*device
;
1438 enum drbd_ret_code retcode
;
1439 enum determine_dev_size dd
;
1440 sector_t max_possible_sectors
;
1441 sector_t min_md_device_sectors
;
1442 struct drbd_backing_dev
*nbc
= NULL
; /* new_backing_conf */
1443 struct disk_conf
*new_disk_conf
= NULL
;
1444 struct block_device
*bdev
;
1445 struct lru_cache
*resync_lru
= NULL
;
1446 struct fifo_buffer
*new_plan
= NULL
;
1447 union drbd_state ns
, os
;
1448 enum drbd_state_rv rv
;
1449 struct net_conf
*nc
;
1451 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
1452 if (!adm_ctx
.reply_skb
)
1454 if (retcode
!= NO_ERROR
)
1457 device
= adm_ctx
.device
;
1458 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
1459 conn_reconfig_start(first_peer_device(device
)->connection
);
1461 /* if you want to reconfigure, please tear down first */
1462 if (device
->state
.disk
> D_DISKLESS
) {
1463 retcode
= ERR_DISK_CONFIGURED
;
1466 /* It may just now have detached because of IO error. Make sure
1467 * drbd_ldev_destroy is done already, we may end up here very fast,
1468 * e.g. if someone calls attach from the on-io-error handler,
1469 * to realize a "hot spare" feature (not that I'd recommend that) */
1470 wait_event(device
->misc_wait
, !atomic_read(&device
->local_cnt
));
1472 /* make sure there is no leftover from previous force-detach attempts */
1473 clear_bit(FORCE_DETACH
, &device
->flags
);
1474 clear_bit(WAS_IO_ERROR
, &device
->flags
);
1475 clear_bit(WAS_READ_ERROR
, &device
->flags
);
1477 /* and no leftover from previously aborted resync or verify, either */
1478 device
->rs_total
= 0;
1479 device
->rs_failed
= 0;
1480 atomic_set(&device
->rs_pending_cnt
, 0);
1482 /* allocation not in the IO path, drbdsetup context */
1483 nbc
= kzalloc(sizeof(struct drbd_backing_dev
), GFP_KERNEL
);
1485 retcode
= ERR_NOMEM
;
1488 spin_lock_init(&nbc
->md
.uuid_lock
);
1490 new_disk_conf
= kzalloc(sizeof(struct disk_conf
), GFP_KERNEL
);
1491 if (!new_disk_conf
) {
1492 retcode
= ERR_NOMEM
;
1495 nbc
->disk_conf
= new_disk_conf
;
1497 set_disk_conf_defaults(new_disk_conf
);
1498 err
= disk_conf_from_attrs(new_disk_conf
, info
);
1500 retcode
= ERR_MANDATORY_TAG
;
1501 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
1505 if (new_disk_conf
->c_plan_ahead
> DRBD_C_PLAN_AHEAD_MAX
)
1506 new_disk_conf
->c_plan_ahead
= DRBD_C_PLAN_AHEAD_MAX
;
1508 new_plan
= fifo_alloc((new_disk_conf
->c_plan_ahead
* 10 * SLEEP_TIME
) / HZ
);
1510 retcode
= ERR_NOMEM
;
1514 if (new_disk_conf
->meta_dev_idx
< DRBD_MD_INDEX_FLEX_INT
) {
1515 retcode
= ERR_MD_IDX_INVALID
;
1519 write_lock_irq(&global_state_lock
);
1520 retcode
= drbd_resync_after_valid(device
, new_disk_conf
->resync_after
);
1521 write_unlock_irq(&global_state_lock
);
1522 if (retcode
!= NO_ERROR
)
1526 nc
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
);
1528 if (new_disk_conf
->fencing
== FP_STONITH
&& nc
->wire_protocol
== DRBD_PROT_A
) {
1530 retcode
= ERR_STONITH_AND_PROT_A
;
1536 bdev
= blkdev_get_by_path(new_disk_conf
->backing_dev
,
1537 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, device
);
1539 drbd_err(device
, "open(\"%s\") failed with %ld\n", new_disk_conf
->backing_dev
,
1541 retcode
= ERR_OPEN_DISK
;
1544 nbc
->backing_bdev
= bdev
;
1547 * meta_dev_idx >= 0: external fixed size, possibly multiple
1548 * drbd sharing one meta device. TODO in that case, paranoia
1549 * check that [md_bdev, meta_dev_idx] is not yet used by some
1550 * other drbd minor! (if you use drbd.conf + drbdadm, that
1551 * should check it for you already; but if you don't, or
1552 * someone fooled it, we need to double check here)
1554 bdev
= blkdev_get_by_path(new_disk_conf
->meta_dev
,
1555 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
,
1556 (new_disk_conf
->meta_dev_idx
< 0) ?
1557 (void *)device
: (void *)drbd_m_holder
);
1559 drbd_err(device
, "open(\"%s\") failed with %ld\n", new_disk_conf
->meta_dev
,
1561 retcode
= ERR_OPEN_MD_DISK
;
1564 nbc
->md_bdev
= bdev
;
1566 if ((nbc
->backing_bdev
== nbc
->md_bdev
) !=
1567 (new_disk_conf
->meta_dev_idx
== DRBD_MD_INDEX_INTERNAL
||
1568 new_disk_conf
->meta_dev_idx
== DRBD_MD_INDEX_FLEX_INT
)) {
1569 retcode
= ERR_MD_IDX_INVALID
;
1573 resync_lru
= lc_create("resync", drbd_bm_ext_cache
,
1574 1, 61, sizeof(struct bm_extent
),
1575 offsetof(struct bm_extent
, lce
));
1577 retcode
= ERR_NOMEM
;
1581 /* Read our meta data super block early.
1582 * This also sets other on-disk offsets. */
1583 retcode
= drbd_md_read(device
, nbc
);
1584 if (retcode
!= NO_ERROR
)
1587 if (new_disk_conf
->al_extents
< DRBD_AL_EXTENTS_MIN
)
1588 new_disk_conf
->al_extents
= DRBD_AL_EXTENTS_MIN
;
1589 if (new_disk_conf
->al_extents
> drbd_al_extents_max(nbc
))
1590 new_disk_conf
->al_extents
= drbd_al_extents_max(nbc
);
1592 if (drbd_get_max_capacity(nbc
) < new_disk_conf
->disk_size
) {
1593 drbd_err(device
, "max capacity %llu smaller than disk size %llu\n",
1594 (unsigned long long) drbd_get_max_capacity(nbc
),
1595 (unsigned long long) new_disk_conf
->disk_size
);
1596 retcode
= ERR_DISK_TOO_SMALL
;
1600 if (new_disk_conf
->meta_dev_idx
< 0) {
1601 max_possible_sectors
= DRBD_MAX_SECTORS_FLEX
;
1602 /* at least one MB, otherwise it does not make sense */
1603 min_md_device_sectors
= (2<<10);
1605 max_possible_sectors
= DRBD_MAX_SECTORS
;
1606 min_md_device_sectors
= MD_128MB_SECT
* (new_disk_conf
->meta_dev_idx
+ 1);
1609 if (drbd_get_capacity(nbc
->md_bdev
) < min_md_device_sectors
) {
1610 retcode
= ERR_MD_DISK_TOO_SMALL
;
1611 drbd_warn(device
, "refusing attach: md-device too small, "
1612 "at least %llu sectors needed for this meta-disk type\n",
1613 (unsigned long long) min_md_device_sectors
);
1617 /* Make sure the new disk is big enough
1618 * (we may currently be R_PRIMARY with no local disk...) */
1619 if (drbd_get_max_capacity(nbc
) <
1620 drbd_get_capacity(device
->this_bdev
)) {
1621 retcode
= ERR_DISK_TOO_SMALL
;
1625 nbc
->known_size
= drbd_get_capacity(nbc
->backing_bdev
);
1627 if (nbc
->known_size
> max_possible_sectors
) {
1628 drbd_warn(device
, "==> truncating very big lower level device "
1629 "to currently maximum possible %llu sectors <==\n",
1630 (unsigned long long) max_possible_sectors
);
1631 if (new_disk_conf
->meta_dev_idx
>= 0)
1632 drbd_warn(device
, "==>> using internal or flexible "
1633 "meta data may help <<==\n");
1636 drbd_suspend_io(device
);
1637 /* also wait for the last barrier ack. */
1638 /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1639 * We need a way to either ignore barrier acks for barriers sent before a device
1640 * was attached, or a way to wait for all pending barrier acks to come in.
1641 * As barriers are counted per resource,
1642 * we'd need to suspend io on all devices of a resource.
1644 wait_event(device
->misc_wait
, !atomic_read(&device
->ap_pending_cnt
) || drbd_suspended(device
));
1645 /* and for any other previously queued work */
1646 drbd_flush_workqueue(&first_peer_device(device
)->connection
->sender_work
);
1648 rv
= _drbd_request_state(device
, NS(disk
, D_ATTACHING
), CS_VERBOSE
);
1649 retcode
= rv
; /* FIXME: Type mismatch. */
1650 drbd_resume_io(device
);
1651 if (rv
< SS_SUCCESS
)
1654 if (!get_ldev_if_state(device
, D_ATTACHING
))
1655 goto force_diskless
;
1657 if (!device
->bitmap
) {
1658 if (drbd_bm_init(device
)) {
1659 retcode
= ERR_NOMEM
;
1660 goto force_diskless_dec
;
1664 if (device
->state
.conn
< C_CONNECTED
&&
1665 device
->state
.role
== R_PRIMARY
&& device
->ed_uuid
&&
1666 (device
->ed_uuid
& ~((u64
)1)) != (nbc
->md
.uuid
[UI_CURRENT
] & ~((u64
)1))) {
1667 drbd_err(device
, "Can only attach to data with current UUID=%016llX\n",
1668 (unsigned long long)device
->ed_uuid
);
1669 retcode
= ERR_DATA_NOT_CURRENT
;
1670 goto force_diskless_dec
;
1673 /* Since we are diskless, fix the activity log first... */
1674 if (drbd_check_al_size(device
, new_disk_conf
)) {
1675 retcode
= ERR_NOMEM
;
1676 goto force_diskless_dec
;
1679 /* Prevent shrinking of consistent devices ! */
1680 if (drbd_md_test_flag(nbc
, MDF_CONSISTENT
) &&
1681 drbd_new_dev_size(device
, nbc
, nbc
->disk_conf
->disk_size
, 0) < nbc
->md
.la_size_sect
) {
1682 drbd_warn(device
, "refusing to truncate a consistent device\n");
1683 retcode
= ERR_DISK_TOO_SMALL
;
1684 goto force_diskless_dec
;
1687 /* Reset the "barriers don't work" bits here, then force meta data to
1688 * be written, to ensure we determine if barriers are supported. */
1689 if (new_disk_conf
->md_flushes
)
1690 clear_bit(MD_NO_FUA
, &device
->flags
);
1692 set_bit(MD_NO_FUA
, &device
->flags
);
1694 /* Point of no return reached.
1695 * Devices and memory are no longer released by error cleanup below.
1696 * now device takes over responsibility, and the state engine should
1697 * clean it up somewhere. */
1698 D_ASSERT(device
, device
->ldev
== NULL
);
1700 device
->resync
= resync_lru
;
1701 device
->rs_plan_s
= new_plan
;
1704 new_disk_conf
= NULL
;
1707 drbd_bump_write_ordering(first_peer_device(device
)->connection
, WO_bdev_flush
);
1709 if (drbd_md_test_flag(device
->ldev
, MDF_CRASHED_PRIMARY
))
1710 set_bit(CRASHED_PRIMARY
, &device
->flags
);
1712 clear_bit(CRASHED_PRIMARY
, &device
->flags
);
1714 if (drbd_md_test_flag(device
->ldev
, MDF_PRIMARY_IND
) &&
1715 !(device
->state
.role
== R_PRIMARY
&& device
->resource
->susp_nod
))
1716 set_bit(CRASHED_PRIMARY
, &device
->flags
);
1718 device
->send_cnt
= 0;
1719 device
->recv_cnt
= 0;
1720 device
->read_cnt
= 0;
1721 device
->writ_cnt
= 0;
1723 drbd_reconsider_max_bio_size(device
);
1725 /* If I am currently not R_PRIMARY,
1726 * but meta data primary indicator is set,
1727 * I just now recover from a hard crash,
1728 * and have been R_PRIMARY before that crash.
1730 * Now, if I had no connection before that crash
1731 * (have been degraded R_PRIMARY), chances are that
1732 * I won't find my peer now either.
1734 * In that case, and _only_ in that case,
1735 * we use the degr-wfc-timeout instead of the default,
1736 * so we can automatically recover from a crash of a
1737 * degraded but active "cluster" after a certain timeout.
1739 clear_bit(USE_DEGR_WFC_T
, &device
->flags
);
1740 if (device
->state
.role
!= R_PRIMARY
&&
1741 drbd_md_test_flag(device
->ldev
, MDF_PRIMARY_IND
) &&
1742 !drbd_md_test_flag(device
->ldev
, MDF_CONNECTED_IND
))
1743 set_bit(USE_DEGR_WFC_T
, &device
->flags
);
1745 dd
= drbd_determine_dev_size(device
, 0, NULL
);
1746 if (dd
<= DS_ERROR
) {
1747 retcode
= ERR_NOMEM_BITMAP
;
1748 goto force_diskless_dec
;
1749 } else if (dd
== DS_GREW
)
1750 set_bit(RESYNC_AFTER_NEG
, &device
->flags
);
1752 if (drbd_md_test_flag(device
->ldev
, MDF_FULL_SYNC
) ||
1753 (test_bit(CRASHED_PRIMARY
, &device
->flags
) &&
1754 drbd_md_test_flag(device
->ldev
, MDF_AL_DISABLED
))) {
1755 drbd_info(device
, "Assuming that all blocks are out of sync "
1756 "(aka FullSync)\n");
1757 if (drbd_bitmap_io(device
, &drbd_bmio_set_n_write
,
1758 "set_n_write from attaching", BM_LOCKED_MASK
)) {
1759 retcode
= ERR_IO_MD_DISK
;
1760 goto force_diskless_dec
;
1763 if (drbd_bitmap_io(device
, &drbd_bm_read
,
1764 "read from attaching", BM_LOCKED_MASK
)) {
1765 retcode
= ERR_IO_MD_DISK
;
1766 goto force_diskless_dec
;
1770 if (_drbd_bm_total_weight(device
) == drbd_bm_bits(device
))
1771 drbd_suspend_al(device
); /* IO is still suspended here... */
1773 spin_lock_irq(&device
->resource
->req_lock
);
1774 os
= drbd_read_state(device
);
1776 /* If MDF_CONSISTENT is not set go into inconsistent state,
1777 otherwise investigate MDF_WasUpToDate...
1778 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1779 otherwise into D_CONSISTENT state.
1781 if (drbd_md_test_flag(device
->ldev
, MDF_CONSISTENT
)) {
1782 if (drbd_md_test_flag(device
->ldev
, MDF_WAS_UP_TO_DATE
))
1783 ns
.disk
= D_CONSISTENT
;
1785 ns
.disk
= D_OUTDATED
;
1787 ns
.disk
= D_INCONSISTENT
;
1790 if (drbd_md_test_flag(device
->ldev
, MDF_PEER_OUT_DATED
))
1791 ns
.pdsk
= D_OUTDATED
;
1794 if (ns
.disk
== D_CONSISTENT
&&
1795 (ns
.pdsk
== D_OUTDATED
|| rcu_dereference(device
->ldev
->disk_conf
)->fencing
== FP_DONT_CARE
))
1796 ns
.disk
= D_UP_TO_DATE
;
1798 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1799 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1800 this point, because drbd_request_state() modifies these
1803 if (rcu_dereference(device
->ldev
->disk_conf
)->al_updates
)
1804 device
->ldev
->md
.flags
&= ~MDF_AL_DISABLED
;
1806 device
->ldev
->md
.flags
|= MDF_AL_DISABLED
;
1810 /* In case we are C_CONNECTED postpone any decision on the new disk
1811 state after the negotiation phase. */
1812 if (device
->state
.conn
== C_CONNECTED
) {
1813 device
->new_state_tmp
.i
= ns
.i
;
1815 ns
.disk
= D_NEGOTIATING
;
1817 /* We expect to receive up-to-date UUIDs soon.
1818 To avoid a race in receive_state, free p_uuid while
1819 holding req_lock. I.e. atomic with the state change */
1820 kfree(device
->p_uuid
);
1821 device
->p_uuid
= NULL
;
1824 rv
= _drbd_set_state(device
, ns
, CS_VERBOSE
, NULL
);
1825 spin_unlock_irq(&device
->resource
->req_lock
);
1827 if (rv
< SS_SUCCESS
)
1828 goto force_diskless_dec
;
1830 mod_timer(&device
->request_timer
, jiffies
+ HZ
);
1832 if (device
->state
.role
== R_PRIMARY
)
1833 device
->ldev
->md
.uuid
[UI_CURRENT
] |= (u64
)1;
1835 device
->ldev
->md
.uuid
[UI_CURRENT
] &= ~(u64
)1;
1837 drbd_md_mark_dirty(device
);
1838 drbd_md_sync(device
);
1840 kobject_uevent(&disk_to_dev(device
->vdisk
)->kobj
, KOBJ_CHANGE
);
1842 conn_reconfig_done(first_peer_device(device
)->connection
);
1843 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
1844 drbd_adm_finish(&adm_ctx
, info
, retcode
);
1850 drbd_force_state(device
, NS(disk
, D_DISKLESS
));
1851 drbd_md_sync(device
);
1853 conn_reconfig_done(first_peer_device(device
)->connection
);
1855 if (nbc
->backing_bdev
)
1856 blkdev_put(nbc
->backing_bdev
,
1857 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
1859 blkdev_put(nbc
->md_bdev
,
1860 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
1863 kfree(new_disk_conf
);
1864 lc_destroy(resync_lru
);
1866 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
1868 drbd_adm_finish(&adm_ctx
, info
, retcode
);
1872 static int adm_detach(struct drbd_device
*device
, int force
)
1874 enum drbd_state_rv retcode
;
1878 set_bit(FORCE_DETACH
, &device
->flags
);
1879 drbd_force_state(device
, NS(disk
, D_FAILED
));
1880 retcode
= SS_SUCCESS
;
1884 drbd_suspend_io(device
); /* so no-one is stuck in drbd_al_begin_io */
1885 drbd_md_get_buffer(device
); /* make sure there is no in-flight meta-data IO */
1886 retcode
= drbd_request_state(device
, NS(disk
, D_FAILED
));
1887 drbd_md_put_buffer(device
);
1888 /* D_FAILED will transition to DISKLESS. */
1889 ret
= wait_event_interruptible(device
->misc_wait
,
1890 device
->state
.disk
!= D_FAILED
);
1891 drbd_resume_io(device
);
1892 if ((int)retcode
== (int)SS_IS_DISKLESS
)
1893 retcode
= SS_NOTHING_TO_DO
;
1900 /* Detaching the disk is a process in multiple stages. First we need to lock
1901 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1902 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1903 * internal references as well.
1904 * Only then we have finally detached. */
1905 int drbd_adm_detach(struct sk_buff
*skb
, struct genl_info
*info
)
1907 struct drbd_config_context adm_ctx
;
1908 enum drbd_ret_code retcode
;
1909 struct detach_parms parms
= { };
1912 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
1913 if (!adm_ctx
.reply_skb
)
1915 if (retcode
!= NO_ERROR
)
1918 if (info
->attrs
[DRBD_NLA_DETACH_PARMS
]) {
1919 err
= detach_parms_from_attrs(&parms
, info
);
1921 retcode
= ERR_MANDATORY_TAG
;
1922 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
1927 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
1928 retcode
= adm_detach(adm_ctx
.device
, parms
.force_detach
);
1929 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
1931 drbd_adm_finish(&adm_ctx
, info
, retcode
);
1935 static bool conn_resync_running(struct drbd_connection
*connection
)
1937 struct drbd_peer_device
*peer_device
;
1942 idr_for_each_entry(&connection
->peer_devices
, peer_device
, vnr
) {
1943 struct drbd_device
*device
= peer_device
->device
;
1944 if (device
->state
.conn
== C_SYNC_SOURCE
||
1945 device
->state
.conn
== C_SYNC_TARGET
||
1946 device
->state
.conn
== C_PAUSED_SYNC_S
||
1947 device
->state
.conn
== C_PAUSED_SYNC_T
) {
1957 static bool conn_ov_running(struct drbd_connection
*connection
)
1959 struct drbd_peer_device
*peer_device
;
1964 idr_for_each_entry(&connection
->peer_devices
, peer_device
, vnr
) {
1965 struct drbd_device
*device
= peer_device
->device
;
1966 if (device
->state
.conn
== C_VERIFY_S
||
1967 device
->state
.conn
== C_VERIFY_T
) {
1977 static enum drbd_ret_code
1978 _check_net_options(struct drbd_connection
*connection
, struct net_conf
*old_net_conf
, struct net_conf
*new_net_conf
)
1980 struct drbd_peer_device
*peer_device
;
1983 if (old_net_conf
&& connection
->cstate
== C_WF_REPORT_PARAMS
&& connection
->agreed_pro_version
< 100) {
1984 if (new_net_conf
->wire_protocol
!= old_net_conf
->wire_protocol
)
1985 return ERR_NEED_APV_100
;
1987 if (new_net_conf
->two_primaries
!= old_net_conf
->two_primaries
)
1988 return ERR_NEED_APV_100
;
1990 if (strcmp(new_net_conf
->integrity_alg
, old_net_conf
->integrity_alg
))
1991 return ERR_NEED_APV_100
;
1994 if (!new_net_conf
->two_primaries
&&
1995 conn_highest_role(connection
) == R_PRIMARY
&&
1996 conn_highest_peer(connection
) == R_PRIMARY
)
1997 return ERR_NEED_ALLOW_TWO_PRI
;
1999 if (new_net_conf
->two_primaries
&&
2000 (new_net_conf
->wire_protocol
!= DRBD_PROT_C
))
2001 return ERR_NOT_PROTO_C
;
2003 idr_for_each_entry(&connection
->peer_devices
, peer_device
, i
) {
2004 struct drbd_device
*device
= peer_device
->device
;
2005 if (get_ldev(device
)) {
2006 enum drbd_fencing_p fp
= rcu_dereference(device
->ldev
->disk_conf
)->fencing
;
2008 if (new_net_conf
->wire_protocol
== DRBD_PROT_A
&& fp
== FP_STONITH
)
2009 return ERR_STONITH_AND_PROT_A
;
2011 if (device
->state
.role
== R_PRIMARY
&& new_net_conf
->discard_my_data
)
2012 return ERR_DISCARD_IMPOSSIBLE
;
2015 if (new_net_conf
->on_congestion
!= OC_BLOCK
&& new_net_conf
->wire_protocol
!= DRBD_PROT_A
)
2016 return ERR_CONG_NOT_PROTO_A
;
2021 static enum drbd_ret_code
2022 check_net_options(struct drbd_connection
*connection
, struct net_conf
*new_net_conf
)
2024 static enum drbd_ret_code rv
;
2025 struct drbd_peer_device
*peer_device
;
2029 rv
= _check_net_options(connection
, rcu_dereference(connection
->net_conf
), new_net_conf
);
2032 /* connection->volumes protected by genl_lock() here */
2033 idr_for_each_entry(&connection
->peer_devices
, peer_device
, i
) {
2034 struct drbd_device
*device
= peer_device
->device
;
2035 if (!device
->bitmap
) {
2036 if (drbd_bm_init(device
))
2045 struct crypto_hash
*verify_tfm
;
2046 struct crypto_hash
*csums_tfm
;
2047 struct crypto_hash
*cram_hmac_tfm
;
2048 struct crypto_hash
*integrity_tfm
;
2052 alloc_hash(struct crypto_hash
**tfm
, char *tfm_name
, int err_alg
)
2057 *tfm
= crypto_alloc_hash(tfm_name
, 0, CRYPTO_ALG_ASYNC
);
2066 static enum drbd_ret_code
2067 alloc_crypto(struct crypto
*crypto
, struct net_conf
*new_net_conf
)
2069 char hmac_name
[CRYPTO_MAX_ALG_NAME
];
2070 enum drbd_ret_code rv
;
2072 rv
= alloc_hash(&crypto
->csums_tfm
, new_net_conf
->csums_alg
,
2076 rv
= alloc_hash(&crypto
->verify_tfm
, new_net_conf
->verify_alg
,
2080 rv
= alloc_hash(&crypto
->integrity_tfm
, new_net_conf
->integrity_alg
,
2084 if (new_net_conf
->cram_hmac_alg
[0] != 0) {
2085 snprintf(hmac_name
, CRYPTO_MAX_ALG_NAME
, "hmac(%s)",
2086 new_net_conf
->cram_hmac_alg
);
2088 rv
= alloc_hash(&crypto
->cram_hmac_tfm
, hmac_name
,
2095 static void free_crypto(struct crypto
*crypto
)
2097 crypto_free_hash(crypto
->cram_hmac_tfm
);
2098 crypto_free_hash(crypto
->integrity_tfm
);
2099 crypto_free_hash(crypto
->csums_tfm
);
2100 crypto_free_hash(crypto
->verify_tfm
);
2103 int drbd_adm_net_opts(struct sk_buff
*skb
, struct genl_info
*info
)
2105 struct drbd_config_context adm_ctx
;
2106 enum drbd_ret_code retcode
;
2107 struct drbd_connection
*connection
;
2108 struct net_conf
*old_net_conf
, *new_net_conf
= NULL
;
2110 int ovr
; /* online verify running */
2111 int rsr
; /* re-sync running */
2112 struct crypto crypto
= { };
2114 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_CONNECTION
);
2115 if (!adm_ctx
.reply_skb
)
2117 if (retcode
!= NO_ERROR
)
2120 connection
= adm_ctx
.connection
;
2121 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2123 new_net_conf
= kzalloc(sizeof(struct net_conf
), GFP_KERNEL
);
2124 if (!new_net_conf
) {
2125 retcode
= ERR_NOMEM
;
2129 conn_reconfig_start(connection
);
2131 mutex_lock(&connection
->data
.mutex
);
2132 mutex_lock(&connection
->resource
->conf_update
);
2133 old_net_conf
= connection
->net_conf
;
2135 if (!old_net_conf
) {
2136 drbd_msg_put_info(adm_ctx
.reply_skb
, "net conf missing, try connect");
2137 retcode
= ERR_INVALID_REQUEST
;
2141 *new_net_conf
= *old_net_conf
;
2142 if (should_set_defaults(info
))
2143 set_net_conf_defaults(new_net_conf
);
2145 err
= net_conf_from_attrs_for_change(new_net_conf
, info
);
2146 if (err
&& err
!= -ENOMSG
) {
2147 retcode
= ERR_MANDATORY_TAG
;
2148 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2152 retcode
= check_net_options(connection
, new_net_conf
);
2153 if (retcode
!= NO_ERROR
)
2156 /* re-sync running */
2157 rsr
= conn_resync_running(connection
);
2158 if (rsr
&& strcmp(new_net_conf
->csums_alg
, old_net_conf
->csums_alg
)) {
2159 retcode
= ERR_CSUMS_RESYNC_RUNNING
;
2163 /* online verify running */
2164 ovr
= conn_ov_running(connection
);
2165 if (ovr
&& strcmp(new_net_conf
->verify_alg
, old_net_conf
->verify_alg
)) {
2166 retcode
= ERR_VERIFY_RUNNING
;
2170 retcode
= alloc_crypto(&crypto
, new_net_conf
);
2171 if (retcode
!= NO_ERROR
)
2174 rcu_assign_pointer(connection
->net_conf
, new_net_conf
);
2177 crypto_free_hash(connection
->csums_tfm
);
2178 connection
->csums_tfm
= crypto
.csums_tfm
;
2179 crypto
.csums_tfm
= NULL
;
2182 crypto_free_hash(connection
->verify_tfm
);
2183 connection
->verify_tfm
= crypto
.verify_tfm
;
2184 crypto
.verify_tfm
= NULL
;
2187 crypto_free_hash(connection
->integrity_tfm
);
2188 connection
->integrity_tfm
= crypto
.integrity_tfm
;
2189 if (connection
->cstate
>= C_WF_REPORT_PARAMS
&& connection
->agreed_pro_version
>= 100)
2190 /* Do this without trying to take connection->data.mutex again. */
2191 __drbd_send_protocol(connection
, P_PROTOCOL_UPDATE
);
2193 crypto_free_hash(connection
->cram_hmac_tfm
);
2194 connection
->cram_hmac_tfm
= crypto
.cram_hmac_tfm
;
2196 mutex_unlock(&connection
->resource
->conf_update
);
2197 mutex_unlock(&connection
->data
.mutex
);
2199 kfree(old_net_conf
);
2201 if (connection
->cstate
>= C_WF_REPORT_PARAMS
) {
2202 struct drbd_peer_device
*peer_device
;
2205 idr_for_each_entry(&connection
->peer_devices
, peer_device
, vnr
)
2206 drbd_send_sync_param(peer_device
);
2212 mutex_unlock(&connection
->resource
->conf_update
);
2213 mutex_unlock(&connection
->data
.mutex
);
2214 free_crypto(&crypto
);
2215 kfree(new_net_conf
);
2217 conn_reconfig_done(connection
);
2219 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2221 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2225 int drbd_adm_connect(struct sk_buff
*skb
, struct genl_info
*info
)
2227 struct drbd_config_context adm_ctx
;
2228 struct drbd_peer_device
*peer_device
;
2229 struct net_conf
*old_net_conf
, *new_net_conf
= NULL
;
2230 struct crypto crypto
= { };
2231 struct drbd_resource
*resource
;
2232 struct drbd_connection
*connection
;
2233 enum drbd_ret_code retcode
;
2237 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
2239 if (!adm_ctx
.reply_skb
)
2241 if (retcode
!= NO_ERROR
)
2243 if (!(adm_ctx
.my_addr
&& adm_ctx
.peer_addr
)) {
2244 drbd_msg_put_info(adm_ctx
.reply_skb
, "connection endpoint(s) missing");
2245 retcode
= ERR_INVALID_REQUEST
;
2249 /* No need for _rcu here. All reconfiguration is
2250 * strictly serialized on genl_lock(). We are protected against
2251 * concurrent reconfiguration/addition/deletion */
2252 for_each_resource(resource
, &drbd_resources
) {
2253 for_each_connection(connection
, resource
) {
2254 if (nla_len(adm_ctx
.my_addr
) == connection
->my_addr_len
&&
2255 !memcmp(nla_data(adm_ctx
.my_addr
), &connection
->my_addr
,
2256 connection
->my_addr_len
)) {
2257 retcode
= ERR_LOCAL_ADDR
;
2261 if (nla_len(adm_ctx
.peer_addr
) == connection
->peer_addr_len
&&
2262 !memcmp(nla_data(adm_ctx
.peer_addr
), &connection
->peer_addr
,
2263 connection
->peer_addr_len
)) {
2264 retcode
= ERR_PEER_ADDR
;
2270 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2271 connection
= first_connection(adm_ctx
.resource
);
2272 conn_reconfig_start(connection
);
2274 if (connection
->cstate
> C_STANDALONE
) {
2275 retcode
= ERR_NET_CONFIGURED
;
2279 /* allocation not in the IO path, drbdsetup / netlink process context */
2280 new_net_conf
= kzalloc(sizeof(*new_net_conf
), GFP_KERNEL
);
2281 if (!new_net_conf
) {
2282 retcode
= ERR_NOMEM
;
2286 set_net_conf_defaults(new_net_conf
);
2288 err
= net_conf_from_attrs(new_net_conf
, info
);
2289 if (err
&& err
!= -ENOMSG
) {
2290 retcode
= ERR_MANDATORY_TAG
;
2291 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2295 retcode
= check_net_options(connection
, new_net_conf
);
2296 if (retcode
!= NO_ERROR
)
2299 retcode
= alloc_crypto(&crypto
, new_net_conf
);
2300 if (retcode
!= NO_ERROR
)
2303 ((char *)new_net_conf
->shared_secret
)[SHARED_SECRET_MAX
-1] = 0;
2305 drbd_flush_workqueue(&connection
->sender_work
);
2307 mutex_lock(&adm_ctx
.resource
->conf_update
);
2308 old_net_conf
= connection
->net_conf
;
2310 retcode
= ERR_NET_CONFIGURED
;
2311 mutex_unlock(&adm_ctx
.resource
->conf_update
);
2314 rcu_assign_pointer(connection
->net_conf
, new_net_conf
);
2316 conn_free_crypto(connection
);
2317 connection
->cram_hmac_tfm
= crypto
.cram_hmac_tfm
;
2318 connection
->integrity_tfm
= crypto
.integrity_tfm
;
2319 connection
->csums_tfm
= crypto
.csums_tfm
;
2320 connection
->verify_tfm
= crypto
.verify_tfm
;
2322 connection
->my_addr_len
= nla_len(adm_ctx
.my_addr
);
2323 memcpy(&connection
->my_addr
, nla_data(adm_ctx
.my_addr
), connection
->my_addr_len
);
2324 connection
->peer_addr_len
= nla_len(adm_ctx
.peer_addr
);
2325 memcpy(&connection
->peer_addr
, nla_data(adm_ctx
.peer_addr
), connection
->peer_addr_len
);
2327 mutex_unlock(&adm_ctx
.resource
->conf_update
);
2330 idr_for_each_entry(&connection
->peer_devices
, peer_device
, i
) {
2331 struct drbd_device
*device
= peer_device
->device
;
2332 device
->send_cnt
= 0;
2333 device
->recv_cnt
= 0;
2337 retcode
= conn_request_state(connection
, NS(conn
, C_UNCONNECTED
), CS_VERBOSE
);
2339 conn_reconfig_done(connection
);
2340 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2341 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2345 free_crypto(&crypto
);
2346 kfree(new_net_conf
);
2348 conn_reconfig_done(connection
);
2349 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2351 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2355 static enum drbd_state_rv
conn_try_disconnect(struct drbd_connection
*connection
, bool force
)
2357 enum drbd_state_rv rv
;
2359 rv
= conn_request_state(connection
, NS(conn
, C_DISCONNECTING
),
2360 force
? CS_HARD
: 0);
2363 case SS_NOTHING_TO_DO
:
2365 case SS_ALREADY_STANDALONE
:
2367 case SS_PRIMARY_NOP
:
2368 /* Our state checking code wants to see the peer outdated. */
2369 rv
= conn_request_state(connection
, NS2(conn
, C_DISCONNECTING
, pdsk
, D_OUTDATED
), 0);
2371 if (rv
== SS_OUTDATE_WO_CONN
) /* lost connection before graceful disconnect succeeded */
2372 rv
= conn_request_state(connection
, NS(conn
, C_DISCONNECTING
), CS_VERBOSE
);
2375 case SS_CW_FAILED_BY_PEER
:
2376 /* The peer probably wants to see us outdated. */
2377 rv
= conn_request_state(connection
, NS2(conn
, C_DISCONNECTING
,
2378 disk
, D_OUTDATED
), 0);
2379 if (rv
== SS_IS_DISKLESS
|| rv
== SS_LOWER_THAN_OUTDATED
) {
2380 rv
= conn_request_state(connection
, NS(conn
, C_DISCONNECTING
),
2385 /* no special handling necessary */
2388 if (rv
>= SS_SUCCESS
) {
2389 enum drbd_state_rv rv2
;
2390 /* No one else can reconfigure the network while I am here.
2391 * The state handling only uses drbd_thread_stop_nowait(),
2392 * we want to really wait here until the receiver is no more.
2394 drbd_thread_stop(&connection
->receiver
);
2396 /* Race breaker. This additional state change request may be
2397 * necessary, if this was a forced disconnect during a receiver
2398 * restart. We may have "killed" the receiver thread just
2399 * after drbd_receiver() returned. Typically, we should be
2400 * C_STANDALONE already, now, and this becomes a no-op.
2402 rv2
= conn_request_state(connection
, NS(conn
, C_STANDALONE
),
2403 CS_VERBOSE
| CS_HARD
);
2404 if (rv2
< SS_SUCCESS
)
2405 drbd_err(connection
,
2406 "unexpected rv2=%d in conn_try_disconnect()\n",
2412 int drbd_adm_disconnect(struct sk_buff
*skb
, struct genl_info
*info
)
2414 struct drbd_config_context adm_ctx
;
2415 struct disconnect_parms parms
;
2416 struct drbd_connection
*connection
;
2417 enum drbd_state_rv rv
;
2418 enum drbd_ret_code retcode
;
2421 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_CONNECTION
);
2422 if (!adm_ctx
.reply_skb
)
2424 if (retcode
!= NO_ERROR
)
2427 connection
= adm_ctx
.connection
;
2428 memset(&parms
, 0, sizeof(parms
));
2429 if (info
->attrs
[DRBD_NLA_DISCONNECT_PARMS
]) {
2430 err
= disconnect_parms_from_attrs(&parms
, info
);
2432 retcode
= ERR_MANDATORY_TAG
;
2433 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2438 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2439 rv
= conn_try_disconnect(connection
, parms
.force_disconnect
);
2440 if (rv
< SS_SUCCESS
)
2441 retcode
= rv
; /* FIXME: Type mismatch. */
2444 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2446 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2450 void resync_after_online_grow(struct drbd_device
*device
)
2452 int iass
; /* I am sync source */
2454 drbd_info(device
, "Resync of new storage after online grow\n");
2455 if (device
->state
.role
!= device
->state
.peer
)
2456 iass
= (device
->state
.role
== R_PRIMARY
);
2458 iass
= test_bit(RESOLVE_CONFLICTS
, &first_peer_device(device
)->connection
->flags
);
2461 drbd_start_resync(device
, C_SYNC_SOURCE
);
2463 _drbd_request_state(device
, NS(conn
, C_WF_SYNC_UUID
), CS_VERBOSE
+ CS_SERIALIZE
);
2466 int drbd_adm_resize(struct sk_buff
*skb
, struct genl_info
*info
)
2468 struct drbd_config_context adm_ctx
;
2469 struct disk_conf
*old_disk_conf
, *new_disk_conf
= NULL
;
2470 struct resize_parms rs
;
2471 struct drbd_device
*device
;
2472 enum drbd_ret_code retcode
;
2473 enum determine_dev_size dd
;
2474 bool change_al_layout
= false;
2475 enum dds_flags ddsf
;
2479 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2480 if (!adm_ctx
.reply_skb
)
2482 if (retcode
!= NO_ERROR
)
2485 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2486 device
= adm_ctx
.device
;
2487 if (!get_ldev(device
)) {
2488 retcode
= ERR_NO_DISK
;
2492 memset(&rs
, 0, sizeof(struct resize_parms
));
2493 rs
.al_stripes
= device
->ldev
->md
.al_stripes
;
2494 rs
.al_stripe_size
= device
->ldev
->md
.al_stripe_size_4k
* 4;
2495 if (info
->attrs
[DRBD_NLA_RESIZE_PARMS
]) {
2496 err
= resize_parms_from_attrs(&rs
, info
);
2498 retcode
= ERR_MANDATORY_TAG
;
2499 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2504 if (device
->state
.conn
> C_CONNECTED
) {
2505 retcode
= ERR_RESIZE_RESYNC
;
2509 if (device
->state
.role
== R_SECONDARY
&&
2510 device
->state
.peer
== R_SECONDARY
) {
2511 retcode
= ERR_NO_PRIMARY
;
2515 if (rs
.no_resync
&& first_peer_device(device
)->connection
->agreed_pro_version
< 93) {
2516 retcode
= ERR_NEED_APV_93
;
2521 u_size
= rcu_dereference(device
->ldev
->disk_conf
)->disk_size
;
2523 if (u_size
!= (sector_t
)rs
.resize_size
) {
2524 new_disk_conf
= kmalloc(sizeof(struct disk_conf
), GFP_KERNEL
);
2525 if (!new_disk_conf
) {
2526 retcode
= ERR_NOMEM
;
2531 if (device
->ldev
->md
.al_stripes
!= rs
.al_stripes
||
2532 device
->ldev
->md
.al_stripe_size_4k
!= rs
.al_stripe_size
/ 4) {
2533 u32 al_size_k
= rs
.al_stripes
* rs
.al_stripe_size
;
2535 if (al_size_k
> (16 * 1024 * 1024)) {
2536 retcode
= ERR_MD_LAYOUT_TOO_BIG
;
2540 if (al_size_k
< MD_32kB_SECT
/2) {
2541 retcode
= ERR_MD_LAYOUT_TOO_SMALL
;
2545 if (device
->state
.conn
!= C_CONNECTED
&& !rs
.resize_force
) {
2546 retcode
= ERR_MD_LAYOUT_CONNECTED
;
2550 change_al_layout
= true;
2553 if (device
->ldev
->known_size
!= drbd_get_capacity(device
->ldev
->backing_bdev
))
2554 device
->ldev
->known_size
= drbd_get_capacity(device
->ldev
->backing_bdev
);
2556 if (new_disk_conf
) {
2557 mutex_lock(&device
->resource
->conf_update
);
2558 old_disk_conf
= device
->ldev
->disk_conf
;
2559 *new_disk_conf
= *old_disk_conf
;
2560 new_disk_conf
->disk_size
= (sector_t
)rs
.resize_size
;
2561 rcu_assign_pointer(device
->ldev
->disk_conf
, new_disk_conf
);
2562 mutex_unlock(&device
->resource
->conf_update
);
2564 kfree(old_disk_conf
);
2567 ddsf
= (rs
.resize_force
? DDSF_FORCED
: 0) | (rs
.no_resync
? DDSF_NO_RESYNC
: 0);
2568 dd
= drbd_determine_dev_size(device
, ddsf
, change_al_layout
? &rs
: NULL
);
2569 drbd_md_sync(device
);
2571 if (dd
== DS_ERROR
) {
2572 retcode
= ERR_NOMEM_BITMAP
;
2574 } else if (dd
== DS_ERROR_SPACE_MD
) {
2575 retcode
= ERR_MD_LAYOUT_NO_FIT
;
2577 } else if (dd
== DS_ERROR_SHRINK
) {
2578 retcode
= ERR_IMPLICIT_SHRINK
;
2582 if (device
->state
.conn
== C_CONNECTED
) {
2584 set_bit(RESIZE_PENDING
, &device
->flags
);
2586 drbd_send_uuids(first_peer_device(device
));
2587 drbd_send_sizes(first_peer_device(device
), 1, ddsf
);
2591 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2593 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2601 int drbd_adm_resource_opts(struct sk_buff
*skb
, struct genl_info
*info
)
2603 struct drbd_config_context adm_ctx
;
2604 enum drbd_ret_code retcode
;
2605 struct res_opts res_opts
;
2608 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
2609 if (!adm_ctx
.reply_skb
)
2611 if (retcode
!= NO_ERROR
)
2614 res_opts
= adm_ctx
.resource
->res_opts
;
2615 if (should_set_defaults(info
))
2616 set_res_opts_defaults(&res_opts
);
2618 err
= res_opts_from_attrs(&res_opts
, info
);
2619 if (err
&& err
!= -ENOMSG
) {
2620 retcode
= ERR_MANDATORY_TAG
;
2621 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2625 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2626 err
= set_resource_options(adm_ctx
.resource
, &res_opts
);
2628 retcode
= ERR_INVALID_REQUEST
;
2630 retcode
= ERR_NOMEM
;
2632 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2635 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2639 int drbd_adm_invalidate(struct sk_buff
*skb
, struct genl_info
*info
)
2641 struct drbd_config_context adm_ctx
;
2642 struct drbd_device
*device
;
2643 int retcode
; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2645 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2646 if (!adm_ctx
.reply_skb
)
2648 if (retcode
!= NO_ERROR
)
2651 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2652 device
= adm_ctx
.device
;
2654 /* If there is still bitmap IO pending, probably because of a previous
2655 * resync just being finished, wait for it before requesting a new resync.
2656 * Also wait for it's after_state_ch(). */
2657 drbd_suspend_io(device
);
2658 wait_event(device
->misc_wait
, !test_bit(BITMAP_IO
, &device
->flags
));
2659 drbd_flush_workqueue(&first_peer_device(device
)->connection
->sender_work
);
2661 /* If we happen to be C_STANDALONE R_SECONDARY, just change to
2662 * D_INCONSISTENT, and set all bits in the bitmap. Otherwise,
2663 * try to start a resync handshake as sync target for full sync.
2665 if (device
->state
.conn
== C_STANDALONE
&& device
->state
.role
== R_SECONDARY
) {
2666 retcode
= drbd_request_state(device
, NS(disk
, D_INCONSISTENT
));
2667 if (retcode
>= SS_SUCCESS
) {
2668 if (drbd_bitmap_io(device
, &drbd_bmio_set_n_write
,
2669 "set_n_write from invalidate", BM_LOCKED_MASK
))
2670 retcode
= ERR_IO_MD_DISK
;
2673 retcode
= drbd_request_state(device
, NS(conn
, C_STARTING_SYNC_T
));
2674 drbd_resume_io(device
);
2675 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2677 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2681 static int drbd_adm_simple_request_state(struct sk_buff
*skb
, struct genl_info
*info
,
2682 union drbd_state mask
, union drbd_state val
)
2684 struct drbd_config_context adm_ctx
;
2685 enum drbd_ret_code retcode
;
2687 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2688 if (!adm_ctx
.reply_skb
)
2690 if (retcode
!= NO_ERROR
)
2693 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2694 retcode
= drbd_request_state(adm_ctx
.device
, mask
, val
);
2695 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2697 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2701 static int drbd_bmio_set_susp_al(struct drbd_device
*device
)
2705 rv
= drbd_bmio_set_n_write(device
);
2706 drbd_suspend_al(device
);
2710 int drbd_adm_invalidate_peer(struct sk_buff
*skb
, struct genl_info
*info
)
2712 struct drbd_config_context adm_ctx
;
2713 int retcode
; /* drbd_ret_code, drbd_state_rv */
2714 struct drbd_device
*device
;
2716 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2717 if (!adm_ctx
.reply_skb
)
2719 if (retcode
!= NO_ERROR
)
2722 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2723 device
= adm_ctx
.device
;
2725 /* If there is still bitmap IO pending, probably because of a previous
2726 * resync just being finished, wait for it before requesting a new resync.
2727 * Also wait for it's after_state_ch(). */
2728 drbd_suspend_io(device
);
2729 wait_event(device
->misc_wait
, !test_bit(BITMAP_IO
, &device
->flags
));
2730 drbd_flush_workqueue(&first_peer_device(device
)->connection
->sender_work
);
2732 /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
2733 * in the bitmap. Otherwise, try to start a resync handshake
2734 * as sync source for full sync.
2736 if (device
->state
.conn
== C_STANDALONE
&& device
->state
.role
== R_PRIMARY
) {
2737 /* The peer will get a resync upon connect anyways. Just make that
2738 into a full resync. */
2739 retcode
= drbd_request_state(device
, NS(pdsk
, D_INCONSISTENT
));
2740 if (retcode
>= SS_SUCCESS
) {
2741 if (drbd_bitmap_io(device
, &drbd_bmio_set_susp_al
,
2742 "set_n_write from invalidate_peer",
2743 BM_LOCKED_SET_ALLOWED
))
2744 retcode
= ERR_IO_MD_DISK
;
2747 retcode
= drbd_request_state(device
, NS(conn
, C_STARTING_SYNC_S
));
2748 drbd_resume_io(device
);
2749 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2751 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2755 int drbd_adm_pause_sync(struct sk_buff
*skb
, struct genl_info
*info
)
2757 struct drbd_config_context adm_ctx
;
2758 enum drbd_ret_code retcode
;
2760 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2761 if (!adm_ctx
.reply_skb
)
2763 if (retcode
!= NO_ERROR
)
2766 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2767 if (drbd_request_state(adm_ctx
.device
, NS(user_isp
, 1)) == SS_NOTHING_TO_DO
)
2768 retcode
= ERR_PAUSE_IS_SET
;
2769 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2771 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2775 int drbd_adm_resume_sync(struct sk_buff
*skb
, struct genl_info
*info
)
2777 struct drbd_config_context adm_ctx
;
2778 union drbd_dev_state s
;
2779 enum drbd_ret_code retcode
;
2781 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2782 if (!adm_ctx
.reply_skb
)
2784 if (retcode
!= NO_ERROR
)
2787 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2788 if (drbd_request_state(adm_ctx
.device
, NS(user_isp
, 0)) == SS_NOTHING_TO_DO
) {
2789 s
= adm_ctx
.device
->state
;
2790 if (s
.conn
== C_PAUSED_SYNC_S
|| s
.conn
== C_PAUSED_SYNC_T
) {
2791 retcode
= s
.aftr_isp
? ERR_PIC_AFTER_DEP
:
2792 s
.peer_isp
? ERR_PIC_PEER_DEP
: ERR_PAUSE_IS_CLEAR
;
2794 retcode
= ERR_PAUSE_IS_CLEAR
;
2797 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2799 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2803 int drbd_adm_suspend_io(struct sk_buff
*skb
, struct genl_info
*info
)
2805 return drbd_adm_simple_request_state(skb
, info
, NS(susp
, 1));
2808 int drbd_adm_resume_io(struct sk_buff
*skb
, struct genl_info
*info
)
2810 struct drbd_config_context adm_ctx
;
2811 struct drbd_device
*device
;
2812 int retcode
; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2814 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2815 if (!adm_ctx
.reply_skb
)
2817 if (retcode
!= NO_ERROR
)
2820 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2821 device
= adm_ctx
.device
;
2822 if (test_bit(NEW_CUR_UUID
, &device
->flags
)) {
2823 drbd_uuid_new_current(device
);
2824 clear_bit(NEW_CUR_UUID
, &device
->flags
);
2826 drbd_suspend_io(device
);
2827 retcode
= drbd_request_state(device
, NS3(susp
, 0, susp_nod
, 0, susp_fen
, 0));
2828 if (retcode
== SS_SUCCESS
) {
2829 if (device
->state
.conn
< C_CONNECTED
)
2830 tl_clear(first_peer_device(device
)->connection
);
2831 if (device
->state
.disk
== D_DISKLESS
|| device
->state
.disk
== D_FAILED
)
2832 tl_restart(first_peer_device(device
)->connection
, FAIL_FROZEN_DISK_IO
);
2834 drbd_resume_io(device
);
2835 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2837 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2841 int drbd_adm_outdate(struct sk_buff
*skb
, struct genl_info
*info
)
2843 return drbd_adm_simple_request_state(skb
, info
, NS(disk
, D_OUTDATED
));
2846 static int nla_put_drbd_cfg_context(struct sk_buff
*skb
,
2847 struct drbd_resource
*resource
,
2848 struct drbd_connection
*connection
,
2849 struct drbd_device
*device
)
2852 nla
= nla_nest_start(skb
, DRBD_NLA_CFG_CONTEXT
);
2854 goto nla_put_failure
;
2856 nla_put_u32(skb
, T_ctx_volume
, device
->vnr
))
2857 goto nla_put_failure
;
2858 if (nla_put_string(skb
, T_ctx_resource_name
, resource
->name
))
2859 goto nla_put_failure
;
2861 if (connection
->my_addr_len
&&
2862 nla_put(skb
, T_ctx_my_addr
, connection
->my_addr_len
, &connection
->my_addr
))
2863 goto nla_put_failure
;
2864 if (connection
->peer_addr_len
&&
2865 nla_put(skb
, T_ctx_peer_addr
, connection
->peer_addr_len
, &connection
->peer_addr
))
2866 goto nla_put_failure
;
2868 nla_nest_end(skb
, nla
);
2873 nla_nest_cancel(skb
, nla
);
2878 * Return the connection of @resource if @resource has exactly one connection.
2880 static struct drbd_connection
*the_only_connection(struct drbd_resource
*resource
)
2882 struct list_head
*connections
= &resource
->connections
;
2884 if (list_empty(connections
) || connections
->next
->next
!= connections
)
2886 return list_first_entry(&resource
->connections
, struct drbd_connection
, connections
);
2889 int nla_put_status_info(struct sk_buff
*skb
, struct drbd_device
*device
,
2890 const struct sib_info
*sib
)
2892 struct drbd_resource
*resource
= device
->resource
;
2893 struct state_info
*si
= NULL
; /* for sizeof(si->member); */
2897 int exclude_sensitive
;
2899 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2900 * to. So we better exclude_sensitive information.
2902 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2903 * in the context of the requesting user process. Exclude sensitive
2904 * information, unless current has superuser.
2906 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2907 * relies on the current implementation of netlink_dump(), which
2908 * executes the dump callback successively from netlink_recvmsg(),
2909 * always in the context of the receiving process */
2910 exclude_sensitive
= sib
|| !capable(CAP_SYS_ADMIN
);
2912 got_ldev
= get_ldev(device
);
2914 /* We need to add connection name and volume number information still.
2915 * Minor number is in drbd_genlmsghdr. */
2916 if (nla_put_drbd_cfg_context(skb
, resource
, the_only_connection(resource
), device
))
2917 goto nla_put_failure
;
2919 if (res_opts_to_skb(skb
, &device
->resource
->res_opts
, exclude_sensitive
))
2920 goto nla_put_failure
;
2924 struct disk_conf
*disk_conf
;
2926 disk_conf
= rcu_dereference(device
->ldev
->disk_conf
);
2927 err
= disk_conf_to_skb(skb
, disk_conf
, exclude_sensitive
);
2930 struct net_conf
*nc
;
2932 nc
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
);
2934 err
= net_conf_to_skb(skb
, nc
, exclude_sensitive
);
2938 goto nla_put_failure
;
2940 nla
= nla_nest_start(skb
, DRBD_NLA_STATE_INFO
);
2942 goto nla_put_failure
;
2943 if (nla_put_u32(skb
, T_sib_reason
, sib
? sib
->sib_reason
: SIB_GET_STATUS_REPLY
) ||
2944 nla_put_u32(skb
, T_current_state
, device
->state
.i
) ||
2945 nla_put_u64(skb
, T_ed_uuid
, device
->ed_uuid
) ||
2946 nla_put_u64(skb
, T_capacity
, drbd_get_capacity(device
->this_bdev
)) ||
2947 nla_put_u64(skb
, T_send_cnt
, device
->send_cnt
) ||
2948 nla_put_u64(skb
, T_recv_cnt
, device
->recv_cnt
) ||
2949 nla_put_u64(skb
, T_read_cnt
, device
->read_cnt
) ||
2950 nla_put_u64(skb
, T_writ_cnt
, device
->writ_cnt
) ||
2951 nla_put_u64(skb
, T_al_writ_cnt
, device
->al_writ_cnt
) ||
2952 nla_put_u64(skb
, T_bm_writ_cnt
, device
->bm_writ_cnt
) ||
2953 nla_put_u32(skb
, T_ap_bio_cnt
, atomic_read(&device
->ap_bio_cnt
)) ||
2954 nla_put_u32(skb
, T_ap_pending_cnt
, atomic_read(&device
->ap_pending_cnt
)) ||
2955 nla_put_u32(skb
, T_rs_pending_cnt
, atomic_read(&device
->rs_pending_cnt
)))
2956 goto nla_put_failure
;
2961 spin_lock_irq(&device
->ldev
->md
.uuid_lock
);
2962 err
= nla_put(skb
, T_uuids
, sizeof(si
->uuids
), device
->ldev
->md
.uuid
);
2963 spin_unlock_irq(&device
->ldev
->md
.uuid_lock
);
2966 goto nla_put_failure
;
2968 if (nla_put_u32(skb
, T_disk_flags
, device
->ldev
->md
.flags
) ||
2969 nla_put_u64(skb
, T_bits_total
, drbd_bm_bits(device
)) ||
2970 nla_put_u64(skb
, T_bits_oos
, drbd_bm_total_weight(device
)))
2971 goto nla_put_failure
;
2972 if (C_SYNC_SOURCE
<= device
->state
.conn
&&
2973 C_PAUSED_SYNC_T
>= device
->state
.conn
) {
2974 if (nla_put_u64(skb
, T_bits_rs_total
, device
->rs_total
) ||
2975 nla_put_u64(skb
, T_bits_rs_failed
, device
->rs_failed
))
2976 goto nla_put_failure
;
2981 switch(sib
->sib_reason
) {
2982 case SIB_SYNC_PROGRESS
:
2983 case SIB_GET_STATUS_REPLY
:
2985 case SIB_STATE_CHANGE
:
2986 if (nla_put_u32(skb
, T_prev_state
, sib
->os
.i
) ||
2987 nla_put_u32(skb
, T_new_state
, sib
->ns
.i
))
2988 goto nla_put_failure
;
2990 case SIB_HELPER_POST
:
2991 if (nla_put_u32(skb
, T_helper_exit_code
,
2992 sib
->helper_exit_code
))
2993 goto nla_put_failure
;
2995 case SIB_HELPER_PRE
:
2996 if (nla_put_string(skb
, T_helper
, sib
->helper_name
))
2997 goto nla_put_failure
;
3001 nla_nest_end(skb
, nla
);
3011 int drbd_adm_get_status(struct sk_buff
*skb
, struct genl_info
*info
)
3013 struct drbd_config_context adm_ctx
;
3014 enum drbd_ret_code retcode
;
3017 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3018 if (!adm_ctx
.reply_skb
)
3020 if (retcode
!= NO_ERROR
)
3023 err
= nla_put_status_info(adm_ctx
.reply_skb
, adm_ctx
.device
, NULL
);
3025 nlmsg_free(adm_ctx
.reply_skb
);
3029 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3033 static int get_one_status(struct sk_buff
*skb
, struct netlink_callback
*cb
)
3035 struct drbd_device
*device
;
3036 struct drbd_genlmsghdr
*dh
;
3037 struct drbd_resource
*pos
= (struct drbd_resource
*)cb
->args
[0];
3038 struct drbd_resource
*resource
= NULL
;
3039 struct drbd_resource
*tmp
;
3040 unsigned volume
= cb
->args
[1];
3042 /* Open coded, deferred, iteration:
3043 * for_each_resource_safe(resource, tmp, &drbd_resources) {
3044 * connection = "first connection of resource or undefined";
3045 * idr_for_each_entry(&resource->devices, device, i) {
3049 * where resource is cb->args[0];
3050 * and i is cb->args[1];
3052 * cb->args[2] indicates if we shall loop over all resources,
3053 * or just dump all volumes of a single resource.
3055 * This may miss entries inserted after this dump started,
3056 * or entries deleted before they are reached.
3058 * We need to make sure the device won't disappear while
3059 * we are looking at it, and revalidate our iterators
3060 * on each iteration.
3063 /* synchronize with conn_create()/drbd_destroy_connection() */
3065 /* revalidate iterator position */
3066 for_each_resource_rcu(tmp
, &drbd_resources
) {
3068 /* first iteration */
3080 device
= idr_get_next(&resource
->devices
, &volume
);
3082 /* No more volumes to dump on this resource.
3083 * Advance resource iterator. */
3084 pos
= list_entry_rcu(resource
->resources
.next
,
3085 struct drbd_resource
, resources
);
3086 /* Did we dump any volume of this resource yet? */
3088 /* If we reached the end of the list,
3089 * or only a single resource dump was requested,
3091 if (&pos
->resources
== &drbd_resources
|| cb
->args
[2])
3099 dh
= genlmsg_put(skb
, NETLINK_CB(cb
->skb
).portid
,
3100 cb
->nlh
->nlmsg_seq
, &drbd_genl_family
,
3101 NLM_F_MULTI
, DRBD_ADM_GET_STATUS
);
3106 /* This is a connection without a single volume.
3107 * Suprisingly enough, it may have a network
3109 struct drbd_connection
*connection
;
3112 dh
->ret_code
= NO_ERROR
;
3113 connection
= the_only_connection(resource
);
3114 if (nla_put_drbd_cfg_context(skb
, resource
, connection
, NULL
))
3117 struct net_conf
*nc
;
3119 nc
= rcu_dereference(connection
->net_conf
);
3120 if (nc
&& net_conf_to_skb(skb
, nc
, 1) != 0)
3126 D_ASSERT(device
, device
->vnr
== volume
);
3127 D_ASSERT(device
, device
->resource
== resource
);
3129 dh
->minor
= device_to_minor(device
);
3130 dh
->ret_code
= NO_ERROR
;
3132 if (nla_put_status_info(skb
, device
, NULL
)) {
3134 genlmsg_cancel(skb
, dh
);
3138 genlmsg_end(skb
, dh
);
3143 /* where to start the next iteration */
3144 cb
->args
[0] = (long)pos
;
3145 cb
->args
[1] = (pos
== resource
) ? volume
+ 1 : 0;
3147 /* No more resources/volumes/minors found results in an empty skb.
3148 * Which will terminate the dump. */
3153 * Request status of all resources, or of all volumes within a single resource.
3155 * This is a dump, as the answer may not fit in a single reply skb otherwise.
3156 * Which means we cannot use the family->attrbuf or other such members, because
3157 * dump is NOT protected by the genl_lock(). During dump, we only have access
3158 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
3160 * Once things are setup properly, we call into get_one_status().
3162 int drbd_adm_get_status_all(struct sk_buff
*skb
, struct netlink_callback
*cb
)
3164 const unsigned hdrlen
= GENL_HDRLEN
+ GENL_MAGIC_FAMILY_HDRSZ
;
3166 const char *resource_name
;
3167 struct drbd_resource
*resource
;
3170 /* Is this a followup call? */
3172 /* ... of a single resource dump,
3173 * and the resource iterator has been advanced already? */
3174 if (cb
->args
[2] && cb
->args
[2] != cb
->args
[0])
3175 return 0; /* DONE. */
3179 /* First call (from netlink_dump_start). We need to figure out
3180 * which resource(s) the user wants us to dump. */
3181 nla
= nla_find(nlmsg_attrdata(cb
->nlh
, hdrlen
),
3182 nlmsg_attrlen(cb
->nlh
, hdrlen
),
3183 DRBD_NLA_CFG_CONTEXT
);
3185 /* No explicit context given. Dump all. */
3188 maxtype
= ARRAY_SIZE(drbd_cfg_context_nl_policy
) - 1;
3189 nla
= drbd_nla_find_nested(maxtype
, nla
, __nla_type(T_ctx_resource_name
));
3191 return PTR_ERR(nla
);
3192 /* context given, but no name present? */
3195 resource_name
= nla_data(nla
);
3196 if (!*resource_name
)
3198 resource
= drbd_find_resource(resource_name
);
3202 kref_put(&resource
->kref
, drbd_destroy_resource
); /* get_one_status() revalidates the resource */
3204 /* prime iterators, and set "filter" mode mark:
3205 * only dump this connection. */
3206 cb
->args
[0] = (long)resource
;
3207 /* cb->args[1] = 0; passed in this way. */
3208 cb
->args
[2] = (long)resource
;
3211 return get_one_status(skb
, cb
);
3214 int drbd_adm_get_timeout_type(struct sk_buff
*skb
, struct genl_info
*info
)
3216 struct drbd_config_context adm_ctx
;
3217 enum drbd_ret_code retcode
;
3218 struct timeout_parms tp
;
3221 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3222 if (!adm_ctx
.reply_skb
)
3224 if (retcode
!= NO_ERROR
)
3228 adm_ctx
.device
->state
.pdsk
== D_OUTDATED
? UT_PEER_OUTDATED
:
3229 test_bit(USE_DEGR_WFC_T
, &adm_ctx
.device
->flags
) ? UT_DEGRADED
:
3232 err
= timeout_parms_to_priv_skb(adm_ctx
.reply_skb
, &tp
);
3234 nlmsg_free(adm_ctx
.reply_skb
);
3238 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3242 int drbd_adm_start_ov(struct sk_buff
*skb
, struct genl_info
*info
)
3244 struct drbd_config_context adm_ctx
;
3245 struct drbd_device
*device
;
3246 enum drbd_ret_code retcode
;
3247 struct start_ov_parms parms
;
3249 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3250 if (!adm_ctx
.reply_skb
)
3252 if (retcode
!= NO_ERROR
)
3255 device
= adm_ctx
.device
;
3257 /* resume from last known position, if possible */
3258 parms
.ov_start_sector
= device
->ov_start_sector
;
3259 parms
.ov_stop_sector
= ULLONG_MAX
;
3260 if (info
->attrs
[DRBD_NLA_START_OV_PARMS
]) {
3261 int err
= start_ov_parms_from_attrs(&parms
, info
);
3263 retcode
= ERR_MANDATORY_TAG
;
3264 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
3268 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
3270 /* w_make_ov_request expects position to be aligned */
3271 device
->ov_start_sector
= parms
.ov_start_sector
& ~(BM_SECT_PER_BIT
-1);
3272 device
->ov_stop_sector
= parms
.ov_stop_sector
;
3274 /* If there is still bitmap IO pending, e.g. previous resync or verify
3275 * just being finished, wait for it before requesting a new resync. */
3276 drbd_suspend_io(device
);
3277 wait_event(device
->misc_wait
, !test_bit(BITMAP_IO
, &device
->flags
));
3278 retcode
= drbd_request_state(device
, NS(conn
, C_VERIFY_S
));
3279 drbd_resume_io(device
);
3281 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
3283 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3288 int drbd_adm_new_c_uuid(struct sk_buff
*skb
, struct genl_info
*info
)
3290 struct drbd_config_context adm_ctx
;
3291 struct drbd_device
*device
;
3292 enum drbd_ret_code retcode
;
3293 int skip_initial_sync
= 0;
3295 struct new_c_uuid_parms args
;
3297 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3298 if (!adm_ctx
.reply_skb
)
3300 if (retcode
!= NO_ERROR
)
3303 device
= adm_ctx
.device
;
3304 memset(&args
, 0, sizeof(args
));
3305 if (info
->attrs
[DRBD_NLA_NEW_C_UUID_PARMS
]) {
3306 err
= new_c_uuid_parms_from_attrs(&args
, info
);
3308 retcode
= ERR_MANDATORY_TAG
;
3309 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
3314 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
3315 mutex_lock(device
->state_mutex
); /* Protects us against serialized state changes. */
3317 if (!get_ldev(device
)) {
3318 retcode
= ERR_NO_DISK
;
3322 /* this is "skip initial sync", assume to be clean */
3323 if (device
->state
.conn
== C_CONNECTED
&&
3324 first_peer_device(device
)->connection
->agreed_pro_version
>= 90 &&
3325 device
->ldev
->md
.uuid
[UI_CURRENT
] == UUID_JUST_CREATED
&& args
.clear_bm
) {
3326 drbd_info(device
, "Preparing to skip initial sync\n");
3327 skip_initial_sync
= 1;
3328 } else if (device
->state
.conn
!= C_STANDALONE
) {
3329 retcode
= ERR_CONNECTED
;
3333 drbd_uuid_set(device
, UI_BITMAP
, 0); /* Rotate UI_BITMAP to History 1, etc... */
3334 drbd_uuid_new_current(device
); /* New current, previous to UI_BITMAP */
3336 if (args
.clear_bm
) {
3337 err
= drbd_bitmap_io(device
, &drbd_bmio_clear_n_write
,
3338 "clear_n_write from new_c_uuid", BM_LOCKED_MASK
);
3340 drbd_err(device
, "Writing bitmap failed with %d\n", err
);
3341 retcode
= ERR_IO_MD_DISK
;
3343 if (skip_initial_sync
) {
3344 drbd_send_uuids_skip_initial_sync(first_peer_device(device
));
3345 _drbd_uuid_set(device
, UI_BITMAP
, 0);
3346 drbd_print_uuids(device
, "cleared bitmap UUID");
3347 spin_lock_irq(&device
->resource
->req_lock
);
3348 _drbd_set_state(_NS2(device
, disk
, D_UP_TO_DATE
, pdsk
, D_UP_TO_DATE
),
3350 spin_unlock_irq(&device
->resource
->req_lock
);
3354 drbd_md_sync(device
);
3358 mutex_unlock(device
->state_mutex
);
3359 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
3361 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3365 static enum drbd_ret_code
3366 drbd_check_resource_name(struct drbd_config_context
*adm_ctx
)
3368 const char *name
= adm_ctx
->resource_name
;
3369 if (!name
|| !name
[0]) {
3370 drbd_msg_put_info(adm_ctx
->reply_skb
, "resource name missing");
3371 return ERR_MANDATORY_TAG
;
3373 /* if we want to use these in sysfs/configfs/debugfs some day,
3374 * we must not allow slashes */
3375 if (strchr(name
, '/')) {
3376 drbd_msg_put_info(adm_ctx
->reply_skb
, "invalid resource name");
3377 return ERR_INVALID_REQUEST
;
3382 int drbd_adm_new_resource(struct sk_buff
*skb
, struct genl_info
*info
)
3384 struct drbd_config_context adm_ctx
;
3385 enum drbd_ret_code retcode
;
3386 struct res_opts res_opts
;
3389 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, 0);
3390 if (!adm_ctx
.reply_skb
)
3392 if (retcode
!= NO_ERROR
)
3395 set_res_opts_defaults(&res_opts
);
3396 err
= res_opts_from_attrs(&res_opts
, info
);
3397 if (err
&& err
!= -ENOMSG
) {
3398 retcode
= ERR_MANDATORY_TAG
;
3399 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
3403 retcode
= drbd_check_resource_name(&adm_ctx
);
3404 if (retcode
!= NO_ERROR
)
3407 if (adm_ctx
.resource
) {
3408 if (info
->nlhdr
->nlmsg_flags
& NLM_F_EXCL
) {
3409 retcode
= ERR_INVALID_REQUEST
;
3410 drbd_msg_put_info(adm_ctx
.reply_skb
, "resource exists");
3412 /* else: still NO_ERROR */
3416 /* not yet safe for genl_family.parallel_ops */
3417 if (!conn_create(adm_ctx
.resource_name
, &res_opts
))
3418 retcode
= ERR_NOMEM
;
3420 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3424 int drbd_adm_new_minor(struct sk_buff
*skb
, struct genl_info
*info
)
3426 struct drbd_config_context adm_ctx
;
3427 struct drbd_genlmsghdr
*dh
= info
->userhdr
;
3428 enum drbd_ret_code retcode
;
3430 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
3431 if (!adm_ctx
.reply_skb
)
3433 if (retcode
!= NO_ERROR
)
3436 if (dh
->minor
> MINORMASK
) {
3437 drbd_msg_put_info(adm_ctx
.reply_skb
, "requested minor out of range");
3438 retcode
= ERR_INVALID_REQUEST
;
3441 if (adm_ctx
.volume
> DRBD_VOLUME_MAX
) {
3442 drbd_msg_put_info(adm_ctx
.reply_skb
, "requested volume id out of range");
3443 retcode
= ERR_INVALID_REQUEST
;
3447 /* drbd_adm_prepare made sure already
3448 * that first_peer_device(device)->connection and device->vnr match the request. */
3449 if (adm_ctx
.device
) {
3450 if (info
->nlhdr
->nlmsg_flags
& NLM_F_EXCL
)
3451 retcode
= ERR_MINOR_EXISTS
;
3452 /* else: still NO_ERROR */
3456 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
3457 retcode
= drbd_create_device(&adm_ctx
, dh
->minor
);
3458 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
3460 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3464 static enum drbd_ret_code
adm_del_minor(struct drbd_device
*device
)
3466 if (device
->state
.disk
== D_DISKLESS
&&
3467 /* no need to be device->state.conn == C_STANDALONE &&
3468 * we may want to delete a minor from a live replication group.
3470 device
->state
.role
== R_SECONDARY
) {
3471 _drbd_request_state(device
, NS(conn
, C_WF_REPORT_PARAMS
),
3472 CS_VERBOSE
+ CS_WAIT_COMPLETE
);
3473 drbd_delete_device(device
);
3476 return ERR_MINOR_CONFIGURED
;
3479 int drbd_adm_del_minor(struct sk_buff
*skb
, struct genl_info
*info
)
3481 struct drbd_config_context adm_ctx
;
3482 enum drbd_ret_code retcode
;
3484 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3485 if (!adm_ctx
.reply_skb
)
3487 if (retcode
!= NO_ERROR
)
3490 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
3491 retcode
= adm_del_minor(adm_ctx
.device
);
3492 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
3494 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3498 int drbd_adm_down(struct sk_buff
*skb
, struct genl_info
*info
)
3500 struct drbd_config_context adm_ctx
;
3501 struct drbd_resource
*resource
;
3502 struct drbd_connection
*connection
;
3503 struct drbd_device
*device
;
3504 int retcode
; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3507 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
3508 if (!adm_ctx
.reply_skb
)
3510 if (retcode
!= NO_ERROR
)
3513 resource
= adm_ctx
.resource
;
3514 mutex_lock(&resource
->adm_mutex
);
3516 for_each_connection(connection
, resource
) {
3517 struct drbd_peer_device
*peer_device
;
3519 idr_for_each_entry(&connection
->peer_devices
, peer_device
, i
) {
3520 retcode
= drbd_set_role(peer_device
->device
, R_SECONDARY
, 0);
3521 if (retcode
< SS_SUCCESS
) {
3522 drbd_msg_put_info(adm_ctx
.reply_skb
, "failed to demote");
3527 retcode
= conn_try_disconnect(connection
, 0);
3528 if (retcode
< SS_SUCCESS
) {
3529 drbd_msg_put_info(adm_ctx
.reply_skb
, "failed to disconnect");
3535 idr_for_each_entry(&resource
->devices
, device
, i
) {
3536 retcode
= adm_detach(device
, 0);
3537 if (retcode
< SS_SUCCESS
|| retcode
> NO_ERROR
) {
3538 drbd_msg_put_info(adm_ctx
.reply_skb
, "failed to detach");
3543 /* If we reach this, all volumes (of this connection) are Secondary,
3544 * Disconnected, Diskless, aka Unconfigured. Make sure all threads have
3545 * actually stopped, state handling only does drbd_thread_stop_nowait(). */
3546 for_each_connection(connection
, resource
)
3547 drbd_thread_stop(&connection
->worker
);
3549 /* Now, nothing can fail anymore */
3551 /* delete volumes */
3552 idr_for_each_entry(&resource
->devices
, device
, i
) {
3553 retcode
= adm_del_minor(device
);
3554 if (retcode
!= NO_ERROR
) {
3555 /* "can not happen" */
3556 drbd_msg_put_info(adm_ctx
.reply_skb
, "failed to delete volume");
3561 list_del_rcu(&resource
->resources
);
3563 drbd_free_resource(resource
);
3566 mutex_unlock(&resource
->adm_mutex
);
3568 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3572 int drbd_adm_del_resource(struct sk_buff
*skb
, struct genl_info
*info
)
3574 struct drbd_config_context adm_ctx
;
3575 struct drbd_resource
*resource
;
3576 struct drbd_connection
*connection
;
3577 enum drbd_ret_code retcode
;
3579 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
3580 if (!adm_ctx
.reply_skb
)
3582 if (retcode
!= NO_ERROR
)
3585 resource
= adm_ctx
.resource
;
3586 mutex_lock(&resource
->adm_mutex
);
3587 for_each_connection(connection
, resource
) {
3588 if (connection
->cstate
> C_STANDALONE
) {
3589 retcode
= ERR_NET_CONFIGURED
;
3593 if (!idr_is_empty(&resource
->devices
)) {
3594 retcode
= ERR_RES_IN_USE
;
3598 list_del_rcu(&resource
->resources
);
3599 for_each_connection(connection
, resource
)
3600 drbd_thread_stop(&connection
->worker
);
3602 drbd_free_resource(resource
);
3605 mutex_unlock(&resource
->adm_mutex
);
3607 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3611 void drbd_bcast_event(struct drbd_device
*device
, const struct sib_info
*sib
)
3613 static atomic_t drbd_genl_seq
= ATOMIC_INIT(2); /* two. */
3614 struct sk_buff
*msg
;
3615 struct drbd_genlmsghdr
*d_out
;
3619 if (sib
->sib_reason
== SIB_SYNC_PROGRESS
) {
3620 if (time_after(jiffies
, device
->rs_last_bcast
+ HZ
))
3621 device
->rs_last_bcast
= jiffies
;
3626 seq
= atomic_inc_return(&drbd_genl_seq
);
3627 msg
= genlmsg_new(NLMSG_GOODSIZE
, GFP_NOIO
);
3632 d_out
= genlmsg_put(msg
, 0, seq
, &drbd_genl_family
, 0, DRBD_EVENT
);
3633 if (!d_out
) /* cannot happen, but anyways. */
3634 goto nla_put_failure
;
3635 d_out
->minor
= device_to_minor(device
);
3636 d_out
->ret_code
= NO_ERROR
;
3638 if (nla_put_status_info(msg
, device
, sib
))
3639 goto nla_put_failure
;
3640 genlmsg_end(msg
, d_out
);
3641 err
= drbd_genl_multicast_events(msg
, 0);
3642 /* msg has been consumed or freed in netlink_broadcast() */
3643 if (err
&& err
!= -ESRCH
)
3651 drbd_err(device
, "Error %d while broadcasting event. "
3652 "Event seq:%u sib_reason:%u\n",
3653 err
, seq
, sib
->sib_reason
);