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 if (connection
->cstate
>= C_WF_REPORT_PARAMS
) {
446 drbd_err(connection
, "Expected cstate < C_WF_REPORT_PARAMS\n");
450 spin_lock_irq(&connection
->resource
->req_lock
);
451 connect_cnt
= connection
->connect_cnt
;
452 spin_unlock_irq(&connection
->resource
->req_lock
);
454 fp
= highest_fencing_policy(connection
);
457 drbd_warn(connection
, "Not fencing peer, I'm not even Consistent myself.\n");
464 r
= conn_khelper(connection
, "fence-peer");
466 switch ((r
>>8) & 0xff) {
467 case 3: /* peer is inconsistent */
468 ex_to_string
= "peer is inconsistent or worse";
470 val
.pdsk
= D_INCONSISTENT
;
472 case 4: /* peer got outdated, or was already outdated */
473 ex_to_string
= "peer was fenced";
475 val
.pdsk
= D_OUTDATED
;
477 case 5: /* peer was down */
478 if (conn_highest_disk(connection
) == D_UP_TO_DATE
) {
479 /* we will(have) create(d) a new UUID anyways... */
480 ex_to_string
= "peer is unreachable, assumed to be dead";
482 val
.pdsk
= D_OUTDATED
;
484 ex_to_string
= "peer unreachable, doing nothing since disk != UpToDate";
487 case 6: /* Peer is primary, voluntarily outdate myself.
488 * This is useful when an unconnected R_SECONDARY is asked to
489 * become R_PRIMARY, but finds the other peer being active. */
490 ex_to_string
= "peer is active";
491 drbd_warn(connection
, "Peer is primary, outdating myself.\n");
493 val
.disk
= D_OUTDATED
;
496 if (fp
!= FP_STONITH
)
497 drbd_err(connection
, "fence-peer() = 7 && fencing != Stonith !!!\n");
498 ex_to_string
= "peer was stonithed";
500 val
.pdsk
= D_OUTDATED
;
503 /* The script is broken ... */
504 drbd_err(connection
, "fence-peer helper broken, returned %d\n", (r
>>8)&0xff);
505 return false; /* Eventually leave IO frozen */
508 drbd_info(connection
, "fence-peer helper returned %d (%s)\n",
509 (r
>>8) & 0xff, ex_to_string
);
514 conn_request_state(connection, mask, val, CS_VERBOSE);
515 here, because we might were able to re-establish the connection in the
517 spin_lock_irq(&connection
->resource
->req_lock
);
518 if (connection
->cstate
< C_WF_REPORT_PARAMS
&& !test_bit(STATE_SENT
, &connection
->flags
)) {
519 if (connection
->connect_cnt
!= connect_cnt
)
520 /* In case the connection was established and droped
521 while the fence-peer handler was running, ignore it */
522 drbd_info(connection
, "Ignoring fence-peer exit code\n");
524 _conn_request_state(connection
, mask
, val
, CS_VERBOSE
);
526 spin_unlock_irq(&connection
->resource
->req_lock
);
528 return conn_highest_pdsk(connection
) <= D_OUTDATED
;
531 static int _try_outdate_peer_async(void *data
)
533 struct drbd_connection
*connection
= (struct drbd_connection
*)data
;
535 conn_try_outdate_peer(connection
);
537 kref_put(&connection
->kref
, drbd_destroy_connection
);
541 void conn_try_outdate_peer_async(struct drbd_connection
*connection
)
543 struct task_struct
*opa
;
545 kref_get(&connection
->kref
);
546 opa
= kthread_run(_try_outdate_peer_async
, connection
, "drbd_async_h");
548 drbd_err(connection
, "out of mem, failed to invoke fence-peer helper\n");
549 kref_put(&connection
->kref
, drbd_destroy_connection
);
554 drbd_set_role(struct drbd_device
*device
, enum drbd_role new_role
, int force
)
556 const int max_tries
= 4;
557 enum drbd_state_rv rv
= SS_UNKNOWN_ERROR
;
561 union drbd_state mask
, val
;
563 if (new_role
== R_PRIMARY
) {
564 struct drbd_connection
*connection
;
566 /* Detect dead peers as soon as possible. */
569 for_each_connection(connection
, device
->resource
)
570 request_ping(connection
);
574 mutex_lock(device
->state_mutex
);
576 mask
.i
= 0; mask
.role
= R_MASK
;
577 val
.i
= 0; val
.role
= new_role
;
579 while (try++ < max_tries
) {
580 rv
= _drbd_request_state(device
, mask
, val
, CS_WAIT_COMPLETE
);
582 /* in case we first succeeded to outdate,
583 * but now suddenly could establish a connection */
584 if (rv
== SS_CW_FAILED_BY_PEER
&& mask
.pdsk
!= 0) {
590 if (rv
== SS_NO_UP_TO_DATE_DISK
&& force
&&
591 (device
->state
.disk
< D_UP_TO_DATE
&&
592 device
->state
.disk
>= D_INCONSISTENT
)) {
594 val
.disk
= D_UP_TO_DATE
;
599 if (rv
== SS_NO_UP_TO_DATE_DISK
&&
600 device
->state
.disk
== D_CONSISTENT
&& mask
.pdsk
== 0) {
601 D_ASSERT(device
, device
->state
.pdsk
== D_UNKNOWN
);
603 if (conn_try_outdate_peer(first_peer_device(device
)->connection
)) {
604 val
.disk
= D_UP_TO_DATE
;
610 if (rv
== SS_NOTHING_TO_DO
)
612 if (rv
== SS_PRIMARY_NOP
&& mask
.pdsk
== 0) {
613 if (!conn_try_outdate_peer(first_peer_device(device
)->connection
) && force
) {
614 drbd_warn(device
, "Forced into split brain situation!\n");
616 val
.pdsk
= D_OUTDATED
;
621 if (rv
== SS_TWO_PRIMARIES
) {
622 /* Maybe the peer is detected as dead very soon...
623 retry at most once more in this case. */
626 nc
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
);
627 timeo
= nc
? (nc
->ping_timeo
+ 1) * HZ
/ 10 : 1;
629 schedule_timeout_interruptible(timeo
);
634 if (rv
< SS_SUCCESS
) {
635 rv
= _drbd_request_state(device
, mask
, val
,
636 CS_VERBOSE
+ CS_WAIT_COMPLETE
);
647 drbd_warn(device
, "Forced to consider local data as UpToDate!\n");
649 /* Wait until nothing is on the fly :) */
650 wait_event(device
->misc_wait
, atomic_read(&device
->ap_pending_cnt
) == 0);
652 /* FIXME also wait for all pending P_BARRIER_ACK? */
654 if (new_role
== R_SECONDARY
) {
655 set_disk_ro(device
->vdisk
, true);
656 if (get_ldev(device
)) {
657 device
->ldev
->md
.uuid
[UI_CURRENT
] &= ~(u64
)1;
661 /* Called from drbd_adm_set_role only.
662 * We are still holding the conf_update mutex. */
663 nc
= first_peer_device(device
)->connection
->net_conf
;
665 nc
->discard_my_data
= 0; /* without copy; single bit op is atomic */
667 set_disk_ro(device
->vdisk
, false);
668 if (get_ldev(device
)) {
669 if (((device
->state
.conn
< C_CONNECTED
||
670 device
->state
.pdsk
<= D_FAILED
)
671 && device
->ldev
->md
.uuid
[UI_BITMAP
] == 0) || forced
)
672 drbd_uuid_new_current(device
);
674 device
->ldev
->md
.uuid
[UI_CURRENT
] |= (u64
)1;
679 /* writeout of activity log covered areas of the bitmap
680 * to stable storage done in after state change already */
682 if (device
->state
.conn
>= C_WF_REPORT_PARAMS
) {
683 /* if this was forced, we should consider sync */
685 drbd_send_uuids(first_peer_device(device
));
686 drbd_send_current_state(first_peer_device(device
));
689 drbd_md_sync(device
);
691 kobject_uevent(&disk_to_dev(device
->vdisk
)->kobj
, KOBJ_CHANGE
);
693 mutex_unlock(device
->state_mutex
);
697 static const char *from_attrs_err_to_txt(int err
)
699 return err
== -ENOMSG
? "required attribute missing" :
700 err
== -EOPNOTSUPP
? "unknown mandatory attribute" :
701 err
== -EEXIST
? "can not change invariant setting" :
702 "invalid attribute value";
705 int drbd_adm_set_role(struct sk_buff
*skb
, struct genl_info
*info
)
707 struct drbd_config_context adm_ctx
;
708 struct set_role_parms parms
;
710 enum drbd_ret_code retcode
;
712 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
713 if (!adm_ctx
.reply_skb
)
715 if (retcode
!= NO_ERROR
)
718 memset(&parms
, 0, sizeof(parms
));
719 if (info
->attrs
[DRBD_NLA_SET_ROLE_PARMS
]) {
720 err
= set_role_parms_from_attrs(&parms
, info
);
722 retcode
= ERR_MANDATORY_TAG
;
723 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
728 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
730 if (info
->genlhdr
->cmd
== DRBD_ADM_PRIMARY
)
731 retcode
= drbd_set_role(adm_ctx
.device
, R_PRIMARY
, parms
.assume_uptodate
);
733 retcode
= drbd_set_role(adm_ctx
.device
, R_SECONDARY
, 0);
735 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
738 drbd_adm_finish(&adm_ctx
, info
, retcode
);
742 /* Initializes the md.*_offset members, so we are able to find
743 * the on disk meta data.
745 * We currently have two possible layouts:
747 * |----------- md_size_sect ------------------|
748 * [ 4k superblock ][ activity log ][ Bitmap ]
750 * | bm_offset = al_offset + X |
751 * ==> bitmap sectors = md_size_sect - bm_offset
754 * |----------- md_size_sect ------------------|
755 * [data.....][ Bitmap ][ activity log ][ 4k superblock ]
757 * | bm_offset = al_offset - Y |
758 * ==> bitmap sectors = Y = al_offset - bm_offset
760 * Activity log size used to be fixed 32kB,
761 * but is about to become configurable.
763 static void drbd_md_set_sector_offsets(struct drbd_device
*device
,
764 struct drbd_backing_dev
*bdev
)
766 sector_t md_size_sect
= 0;
767 unsigned int al_size_sect
= bdev
->md
.al_size_4k
* 8;
769 bdev
->md
.md_offset
= drbd_md_ss(bdev
);
771 switch (bdev
->md
.meta_dev_idx
) {
773 /* v07 style fixed size indexed meta data */
774 bdev
->md
.md_size_sect
= MD_128MB_SECT
;
775 bdev
->md
.al_offset
= MD_4kB_SECT
;
776 bdev
->md
.bm_offset
= MD_4kB_SECT
+ al_size_sect
;
778 case DRBD_MD_INDEX_FLEX_EXT
:
779 /* just occupy the full device; unit: sectors */
780 bdev
->md
.md_size_sect
= drbd_get_capacity(bdev
->md_bdev
);
781 bdev
->md
.al_offset
= MD_4kB_SECT
;
782 bdev
->md
.bm_offset
= MD_4kB_SECT
+ al_size_sect
;
784 case DRBD_MD_INDEX_INTERNAL
:
785 case DRBD_MD_INDEX_FLEX_INT
:
786 /* al size is still fixed */
787 bdev
->md
.al_offset
= -al_size_sect
;
788 /* we need (slightly less than) ~ this much bitmap sectors: */
789 md_size_sect
= drbd_get_capacity(bdev
->backing_bdev
);
790 md_size_sect
= ALIGN(md_size_sect
, BM_SECT_PER_EXT
);
791 md_size_sect
= BM_SECT_TO_EXT(md_size_sect
);
792 md_size_sect
= ALIGN(md_size_sect
, 8);
794 /* plus the "drbd meta data super block",
795 * and the activity log; */
796 md_size_sect
+= MD_4kB_SECT
+ al_size_sect
;
798 bdev
->md
.md_size_sect
= md_size_sect
;
799 /* bitmap offset is adjusted by 'super' block size */
800 bdev
->md
.bm_offset
= -md_size_sect
+ MD_4kB_SECT
;
805 /* input size is expected to be in KB */
806 char *ppsize(char *buf
, unsigned long long size
)
808 /* Needs 9 bytes at max including trailing NUL:
809 * -1ULL ==> "16384 EB" */
810 static char units
[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
812 while (size
>= 10000 && base
< sizeof(units
)-1) {
814 size
= (size
>> 10) + !!(size
& (1<<9));
817 sprintf(buf
, "%u %cB", (unsigned)size
, units
[base
]);
822 /* there is still a theoretical deadlock when called from receiver
823 * on an D_INCONSISTENT R_PRIMARY:
824 * remote READ does inc_ap_bio, receiver would need to receive answer
825 * packet from remote to dec_ap_bio again.
826 * receiver receive_sizes(), comes here,
827 * waits for ap_bio_cnt == 0. -> deadlock.
828 * but this cannot happen, actually, because:
829 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
830 * (not connected, or bad/no disk on peer):
831 * see drbd_fail_request_early, ap_bio_cnt is zero.
832 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
833 * peer may not initiate a resize.
835 /* Note these are not to be confused with
836 * drbd_adm_suspend_io/drbd_adm_resume_io,
837 * which are (sub) state changes triggered by admin (drbdsetup),
838 * and can be long lived.
839 * This changes an device->flag, is triggered by drbd internals,
840 * and should be short-lived. */
841 void drbd_suspend_io(struct drbd_device
*device
)
843 set_bit(SUSPEND_IO
, &device
->flags
);
844 if (drbd_suspended(device
))
846 wait_event(device
->misc_wait
, !atomic_read(&device
->ap_bio_cnt
));
849 void drbd_resume_io(struct drbd_device
*device
)
851 clear_bit(SUSPEND_IO
, &device
->flags
);
852 wake_up(&device
->misc_wait
);
856 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
857 * @device: DRBD device.
859 * Returns 0 on success, negative return values indicate errors.
860 * You should call drbd_md_sync() after calling this function.
862 enum determine_dev_size
863 drbd_determine_dev_size(struct drbd_device
*device
, enum dds_flags flags
, struct resize_parms
*rs
) __must_hold(local
)
865 sector_t prev_first_sect
, prev_size
; /* previous meta location */
866 sector_t la_size_sect
, u_size
;
867 struct drbd_md
*md
= &device
->ldev
->md
;
868 u32 prev_al_stripe_size_4k
;
874 int md_moved
, la_size_changed
;
875 enum determine_dev_size rv
= DS_UNCHANGED
;
878 * application request passes inc_ap_bio,
879 * but then cannot get an AL-reference.
880 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
883 * Suspend IO right here.
884 * still lock the act_log to not trigger ASSERTs there.
886 drbd_suspend_io(device
);
887 buffer
= drbd_md_get_buffer(device
); /* Lock meta-data IO */
889 drbd_resume_io(device
);
893 /* no wait necessary anymore, actually we could assert that */
894 wait_event(device
->al_wait
, lc_try_lock(device
->act_log
));
896 prev_first_sect
= drbd_md_first_sector(device
->ldev
);
897 prev_size
= device
->ldev
->md
.md_size_sect
;
898 la_size_sect
= device
->ldev
->md
.la_size_sect
;
901 /* rs is non NULL if we should change the AL layout only */
903 prev_al_stripes
= md
->al_stripes
;
904 prev_al_stripe_size_4k
= md
->al_stripe_size_4k
;
906 md
->al_stripes
= rs
->al_stripes
;
907 md
->al_stripe_size_4k
= rs
->al_stripe_size
/ 4;
908 md
->al_size_4k
= (u64
)rs
->al_stripes
* rs
->al_stripe_size
/ 4;
911 drbd_md_set_sector_offsets(device
, device
->ldev
);
914 u_size
= rcu_dereference(device
->ldev
->disk_conf
)->disk_size
;
916 size
= drbd_new_dev_size(device
, device
->ldev
, u_size
, flags
& DDSF_FORCED
);
918 if (size
< la_size_sect
) {
919 if (rs
&& u_size
== 0) {
920 /* Remove "rs &&" later. This check should always be active, but
921 right now the receiver expects the permissive behavior */
922 drbd_warn(device
, "Implicit shrink not allowed. "
923 "Use --size=%llus for explicit shrink.\n",
924 (unsigned long long)size
);
925 rv
= DS_ERROR_SHRINK
;
928 rv
= DS_ERROR_SPACE_MD
;
929 if (rv
!= DS_UNCHANGED
)
933 if (drbd_get_capacity(device
->this_bdev
) != size
||
934 drbd_bm_capacity(device
) != size
) {
936 err
= drbd_bm_resize(device
, size
, !(flags
& DDSF_NO_RESYNC
));
938 /* currently there is only one error: ENOMEM! */
939 size
= drbd_bm_capacity(device
)>>1;
941 drbd_err(device
, "OUT OF MEMORY! "
942 "Could not allocate bitmap!\n");
944 drbd_err(device
, "BM resizing failed. "
945 "Leaving size unchanged at size = %lu KB\n",
946 (unsigned long)size
);
950 /* racy, see comments above. */
951 drbd_set_my_capacity(device
, size
);
952 device
->ldev
->md
.la_size_sect
= size
;
953 drbd_info(device
, "size = %s (%llu KB)\n", ppsize(ppb
, size
>>1),
954 (unsigned long long)size
>>1);
959 la_size_changed
= (la_size_sect
!= device
->ldev
->md
.la_size_sect
);
961 md_moved
= prev_first_sect
!= drbd_md_first_sector(device
->ldev
)
962 || prev_size
!= device
->ldev
->md
.md_size_sect
;
964 if (la_size_changed
|| md_moved
|| rs
) {
967 drbd_al_shrink(device
); /* All extents inactive. */
969 prev_flags
= md
->flags
;
970 md
->flags
&= ~MDF_PRIMARY_IND
;
971 drbd_md_write(device
, buffer
);
973 drbd_info(device
, "Writing the whole bitmap, %s\n",
974 la_size_changed
&& md_moved
? "size changed and md moved" :
975 la_size_changed
? "size changed" : "md moved");
976 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
977 drbd_bitmap_io(device
, md_moved
? &drbd_bm_write_all
: &drbd_bm_write
,
978 "size changed", BM_LOCKED_MASK
);
979 drbd_initialize_al(device
, buffer
);
981 md
->flags
= prev_flags
;
982 drbd_md_write(device
, buffer
);
985 drbd_info(device
, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n",
986 md
->al_stripes
, md
->al_stripe_size_4k
* 4);
989 if (size
> la_size_sect
)
990 rv
= la_size_sect
? DS_GREW
: DS_GREW_FROM_ZERO
;
991 if (size
< la_size_sect
)
997 md
->al_stripes
= prev_al_stripes
;
998 md
->al_stripe_size_4k
= prev_al_stripe_size_4k
;
999 md
->al_size_4k
= (u64
)prev_al_stripes
* prev_al_stripe_size_4k
;
1001 drbd_md_set_sector_offsets(device
, device
->ldev
);
1004 lc_unlock(device
->act_log
);
1005 wake_up(&device
->al_wait
);
1006 drbd_md_put_buffer(device
);
1007 drbd_resume_io(device
);
1013 drbd_new_dev_size(struct drbd_device
*device
, struct drbd_backing_dev
*bdev
,
1014 sector_t u_size
, int assume_peer_has_space
)
1016 sector_t p_size
= device
->p_size
; /* partner's disk size. */
1017 sector_t la_size_sect
= bdev
->md
.la_size_sect
; /* last agreed size. */
1018 sector_t m_size
; /* my size */
1021 m_size
= drbd_get_max_capacity(bdev
);
1023 if (device
->state
.conn
< C_CONNECTED
&& assume_peer_has_space
) {
1024 drbd_warn(device
, "Resize while not connected was forced by the user!\n");
1028 if (p_size
&& m_size
) {
1029 size
= min_t(sector_t
, p_size
, m_size
);
1032 size
= la_size_sect
;
1033 if (m_size
&& m_size
< size
)
1035 if (p_size
&& p_size
< size
)
1046 drbd_err(device
, "Both nodes diskless!\n");
1050 drbd_err(device
, "Requested disk size is too big (%lu > %lu)\n",
1051 (unsigned long)u_size
>>1, (unsigned long)size
>>1);
1060 * drbd_check_al_size() - Ensures that the AL is of the right size
1061 * @device: DRBD device.
1063 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
1064 * failed, and 0 on success. You should call drbd_md_sync() after you called
1067 static int drbd_check_al_size(struct drbd_device
*device
, struct disk_conf
*dc
)
1069 struct lru_cache
*n
, *t
;
1070 struct lc_element
*e
;
1071 unsigned int in_use
;
1074 if (device
->act_log
&&
1075 device
->act_log
->nr_elements
== dc
->al_extents
)
1079 t
= device
->act_log
;
1080 n
= lc_create("act_log", drbd_al_ext_cache
, AL_UPDATES_PER_TRANSACTION
,
1081 dc
->al_extents
, sizeof(struct lc_element
), 0);
1084 drbd_err(device
, "Cannot allocate act_log lru!\n");
1087 spin_lock_irq(&device
->al_lock
);
1089 for (i
= 0; i
< t
->nr_elements
; i
++) {
1090 e
= lc_element_by_index(t
, i
);
1092 drbd_err(device
, "refcnt(%d)==%d\n",
1093 e
->lc_number
, e
->refcnt
);
1094 in_use
+= e
->refcnt
;
1098 device
->act_log
= n
;
1099 spin_unlock_irq(&device
->al_lock
);
1101 drbd_err(device
, "Activity log still in use!\n");
1108 drbd_md_mark_dirty(device
); /* we changed device->act_log->nr_elemens */
1112 static void drbd_setup_queue_param(struct drbd_device
*device
, unsigned int max_bio_size
)
1114 struct request_queue
* const q
= device
->rq_queue
;
1115 unsigned int max_hw_sectors
= max_bio_size
>> 9;
1116 unsigned int max_segments
= 0;
1118 if (get_ldev_if_state(device
, D_ATTACHING
)) {
1119 struct request_queue
* const b
= device
->ldev
->backing_bdev
->bd_disk
->queue
;
1121 max_hw_sectors
= min(queue_max_hw_sectors(b
), max_bio_size
>> 9);
1123 max_segments
= rcu_dereference(device
->ldev
->disk_conf
)->max_bio_bvecs
;
1128 blk_queue_logical_block_size(q
, 512);
1129 blk_queue_max_hw_sectors(q
, max_hw_sectors
);
1130 /* This is the workaround for "bio would need to, but cannot, be split" */
1131 blk_queue_max_segments(q
, max_segments
? max_segments
: BLK_MAX_SEGMENTS
);
1132 blk_queue_segment_boundary(q
, PAGE_CACHE_SIZE
-1);
1134 if (get_ldev_if_state(device
, D_ATTACHING
)) {
1135 struct request_queue
* const b
= device
->ldev
->backing_bdev
->bd_disk
->queue
;
1137 blk_queue_stack_limits(q
, b
);
1139 if (q
->backing_dev_info
.ra_pages
!= b
->backing_dev_info
.ra_pages
) {
1140 drbd_info(device
, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1141 q
->backing_dev_info
.ra_pages
,
1142 b
->backing_dev_info
.ra_pages
);
1143 q
->backing_dev_info
.ra_pages
= b
->backing_dev_info
.ra_pages
;
1149 void drbd_reconsider_max_bio_size(struct drbd_device
*device
)
1151 unsigned int now
, new, local
, peer
;
1153 now
= queue_max_hw_sectors(device
->rq_queue
) << 9;
1154 local
= device
->local_max_bio_size
; /* Eventually last known value, from volatile memory */
1155 peer
= device
->peer_max_bio_size
; /* Eventually last known value, from meta data */
1157 if (get_ldev_if_state(device
, D_ATTACHING
)) {
1158 local
= queue_max_hw_sectors(device
->ldev
->backing_bdev
->bd_disk
->queue
) << 9;
1159 device
->local_max_bio_size
= local
;
1162 local
= min(local
, DRBD_MAX_BIO_SIZE
);
1164 /* We may ignore peer limits if the peer is modern enough.
1165 Because new from 8.3.8 onwards the peer can use multiple
1166 BIOs for a single peer_request */
1167 if (device
->state
.conn
>= C_WF_REPORT_PARAMS
) {
1168 if (first_peer_device(device
)->connection
->agreed_pro_version
< 94)
1169 peer
= min(device
->peer_max_bio_size
, DRBD_MAX_SIZE_H80_PACKET
);
1170 /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1171 else if (first_peer_device(device
)->connection
->agreed_pro_version
== 94)
1172 peer
= DRBD_MAX_SIZE_H80_PACKET
;
1173 else if (first_peer_device(device
)->connection
->agreed_pro_version
< 100)
1174 peer
= DRBD_MAX_BIO_SIZE_P95
; /* drbd 8.3.8 onwards, before 8.4.0 */
1176 peer
= DRBD_MAX_BIO_SIZE
;
1179 new = min(local
, peer
);
1181 if (device
->state
.role
== R_PRIMARY
&& new < now
)
1182 drbd_err(device
, "ASSERT FAILED new < now; (%u < %u)\n", new, now
);
1185 drbd_info(device
, "max BIO size = %u\n", new);
1187 drbd_setup_queue_param(device
, new);
1190 /* Starts the worker thread */
1191 static void conn_reconfig_start(struct drbd_connection
*connection
)
1193 drbd_thread_start(&connection
->worker
);
1194 drbd_flush_workqueue(&connection
->sender_work
);
1197 /* if still unconfigured, stops worker again. */
1198 static void conn_reconfig_done(struct drbd_connection
*connection
)
1201 spin_lock_irq(&connection
->resource
->req_lock
);
1202 stop_threads
= conn_all_vols_unconf(connection
) &&
1203 connection
->cstate
== C_STANDALONE
;
1204 spin_unlock_irq(&connection
->resource
->req_lock
);
1206 /* asender is implicitly stopped by receiver
1207 * in conn_disconnect() */
1208 drbd_thread_stop(&connection
->receiver
);
1209 drbd_thread_stop(&connection
->worker
);
1213 /* Make sure IO is suspended before calling this function(). */
1214 static void drbd_suspend_al(struct drbd_device
*device
)
1218 if (!lc_try_lock(device
->act_log
)) {
1219 drbd_warn(device
, "Failed to lock al in drbd_suspend_al()\n");
1223 drbd_al_shrink(device
);
1224 spin_lock_irq(&device
->resource
->req_lock
);
1225 if (device
->state
.conn
< C_CONNECTED
)
1226 s
= !test_and_set_bit(AL_SUSPENDED
, &device
->flags
);
1227 spin_unlock_irq(&device
->resource
->req_lock
);
1228 lc_unlock(device
->act_log
);
1231 drbd_info(device
, "Suspended AL updates\n");
1235 static bool should_set_defaults(struct genl_info
*info
)
1237 unsigned flags
= ((struct drbd_genlmsghdr
*)info
->userhdr
)->flags
;
1238 return 0 != (flags
& DRBD_GENL_F_SET_DEFAULTS
);
1241 static unsigned int drbd_al_extents_max(struct drbd_backing_dev
*bdev
)
1243 /* This is limited by 16 bit "slot" numbers,
1244 * and by available on-disk context storage.
1246 * Also (u16)~0 is special (denotes a "free" extent).
1248 * One transaction occupies one 4kB on-disk block,
1249 * we have n such blocks in the on disk ring buffer,
1250 * the "current" transaction may fail (n-1),
1251 * and there is 919 slot numbers context information per transaction.
1253 * 72 transaction blocks amounts to more than 2**16 context slots,
1254 * so cap there first.
1256 const unsigned int max_al_nr
= DRBD_AL_EXTENTS_MAX
;
1257 const unsigned int sufficient_on_disk
=
1258 (max_al_nr
+ AL_CONTEXT_PER_TRANSACTION
-1)
1259 /AL_CONTEXT_PER_TRANSACTION
;
1261 unsigned int al_size_4k
= bdev
->md
.al_size_4k
;
1263 if (al_size_4k
> sufficient_on_disk
)
1266 return (al_size_4k
- 1) * AL_CONTEXT_PER_TRANSACTION
;
1269 int drbd_adm_disk_opts(struct sk_buff
*skb
, struct genl_info
*info
)
1271 struct drbd_config_context adm_ctx
;
1272 enum drbd_ret_code retcode
;
1273 struct drbd_device
*device
;
1274 struct disk_conf
*new_disk_conf
, *old_disk_conf
;
1275 struct fifo_buffer
*old_plan
= NULL
, *new_plan
= NULL
;
1278 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
1279 if (!adm_ctx
.reply_skb
)
1281 if (retcode
!= NO_ERROR
)
1284 device
= adm_ctx
.device
;
1285 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
1287 /* we also need a disk
1288 * to change the options on */
1289 if (!get_ldev(device
)) {
1290 retcode
= ERR_NO_DISK
;
1294 new_disk_conf
= kmalloc(sizeof(struct disk_conf
), GFP_KERNEL
);
1295 if (!new_disk_conf
) {
1296 retcode
= ERR_NOMEM
;
1300 mutex_lock(&device
->resource
->conf_update
);
1301 old_disk_conf
= device
->ldev
->disk_conf
;
1302 *new_disk_conf
= *old_disk_conf
;
1303 if (should_set_defaults(info
))
1304 set_disk_conf_defaults(new_disk_conf
);
1306 err
= disk_conf_from_attrs_for_change(new_disk_conf
, info
);
1307 if (err
&& err
!= -ENOMSG
) {
1308 retcode
= ERR_MANDATORY_TAG
;
1309 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
1313 if (!expect(new_disk_conf
->resync_rate
>= 1))
1314 new_disk_conf
->resync_rate
= 1;
1316 if (new_disk_conf
->al_extents
< DRBD_AL_EXTENTS_MIN
)
1317 new_disk_conf
->al_extents
= DRBD_AL_EXTENTS_MIN
;
1318 if (new_disk_conf
->al_extents
> drbd_al_extents_max(device
->ldev
))
1319 new_disk_conf
->al_extents
= drbd_al_extents_max(device
->ldev
);
1321 if (new_disk_conf
->c_plan_ahead
> DRBD_C_PLAN_AHEAD_MAX
)
1322 new_disk_conf
->c_plan_ahead
= DRBD_C_PLAN_AHEAD_MAX
;
1324 fifo_size
= (new_disk_conf
->c_plan_ahead
* 10 * SLEEP_TIME
) / HZ
;
1325 if (fifo_size
!= device
->rs_plan_s
->size
) {
1326 new_plan
= fifo_alloc(fifo_size
);
1328 drbd_err(device
, "kmalloc of fifo_buffer failed");
1329 retcode
= ERR_NOMEM
;
1334 drbd_suspend_io(device
);
1335 wait_event(device
->al_wait
, lc_try_lock(device
->act_log
));
1336 drbd_al_shrink(device
);
1337 err
= drbd_check_al_size(device
, new_disk_conf
);
1338 lc_unlock(device
->act_log
);
1339 wake_up(&device
->al_wait
);
1340 drbd_resume_io(device
);
1343 retcode
= ERR_NOMEM
;
1347 write_lock_irq(&global_state_lock
);
1348 retcode
= drbd_resync_after_valid(device
, new_disk_conf
->resync_after
);
1349 if (retcode
== NO_ERROR
) {
1350 rcu_assign_pointer(device
->ldev
->disk_conf
, new_disk_conf
);
1351 drbd_resync_after_changed(device
);
1353 write_unlock_irq(&global_state_lock
);
1355 if (retcode
!= NO_ERROR
)
1359 old_plan
= device
->rs_plan_s
;
1360 rcu_assign_pointer(device
->rs_plan_s
, new_plan
);
1363 mutex_unlock(&device
->resource
->conf_update
);
1365 if (new_disk_conf
->al_updates
)
1366 device
->ldev
->md
.flags
&= ~MDF_AL_DISABLED
;
1368 device
->ldev
->md
.flags
|= MDF_AL_DISABLED
;
1370 if (new_disk_conf
->md_flushes
)
1371 clear_bit(MD_NO_FUA
, &device
->flags
);
1373 set_bit(MD_NO_FUA
, &device
->flags
);
1375 drbd_bump_write_ordering(first_peer_device(device
)->connection
, WO_bdev_flush
);
1377 drbd_md_sync(device
);
1379 if (device
->state
.conn
>= C_CONNECTED
) {
1380 struct drbd_peer_device
*peer_device
;
1382 for_each_peer_device(peer_device
, device
)
1383 drbd_send_sync_param(peer_device
);
1387 kfree(old_disk_conf
);
1389 mod_timer(&device
->request_timer
, jiffies
+ HZ
);
1393 mutex_unlock(&device
->resource
->conf_update
);
1395 kfree(new_disk_conf
);
1400 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
1402 drbd_adm_finish(&adm_ctx
, info
, retcode
);
1406 int drbd_adm_attach(struct sk_buff
*skb
, struct genl_info
*info
)
1408 struct drbd_config_context adm_ctx
;
1409 struct drbd_device
*device
;
1411 enum drbd_ret_code retcode
;
1412 enum determine_dev_size dd
;
1413 sector_t max_possible_sectors
;
1414 sector_t min_md_device_sectors
;
1415 struct drbd_backing_dev
*nbc
= NULL
; /* new_backing_conf */
1416 struct disk_conf
*new_disk_conf
= NULL
;
1417 struct block_device
*bdev
;
1418 struct lru_cache
*resync_lru
= NULL
;
1419 struct fifo_buffer
*new_plan
= NULL
;
1420 union drbd_state ns
, os
;
1421 enum drbd_state_rv rv
;
1422 struct net_conf
*nc
;
1424 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
1425 if (!adm_ctx
.reply_skb
)
1427 if (retcode
!= NO_ERROR
)
1430 device
= adm_ctx
.device
;
1431 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
1432 conn_reconfig_start(first_peer_device(device
)->connection
);
1434 /* if you want to reconfigure, please tear down first */
1435 if (device
->state
.disk
> D_DISKLESS
) {
1436 retcode
= ERR_DISK_CONFIGURED
;
1439 /* It may just now have detached because of IO error. Make sure
1440 * drbd_ldev_destroy is done already, we may end up here very fast,
1441 * e.g. if someone calls attach from the on-io-error handler,
1442 * to realize a "hot spare" feature (not that I'd recommend that) */
1443 wait_event(device
->misc_wait
, !atomic_read(&device
->local_cnt
));
1445 /* make sure there is no leftover from previous force-detach attempts */
1446 clear_bit(FORCE_DETACH
, &device
->flags
);
1447 clear_bit(WAS_IO_ERROR
, &device
->flags
);
1448 clear_bit(WAS_READ_ERROR
, &device
->flags
);
1450 /* and no leftover from previously aborted resync or verify, either */
1451 device
->rs_total
= 0;
1452 device
->rs_failed
= 0;
1453 atomic_set(&device
->rs_pending_cnt
, 0);
1455 /* allocation not in the IO path, drbdsetup context */
1456 nbc
= kzalloc(sizeof(struct drbd_backing_dev
), GFP_KERNEL
);
1458 retcode
= ERR_NOMEM
;
1461 spin_lock_init(&nbc
->md
.uuid_lock
);
1463 new_disk_conf
= kzalloc(sizeof(struct disk_conf
), GFP_KERNEL
);
1464 if (!new_disk_conf
) {
1465 retcode
= ERR_NOMEM
;
1468 nbc
->disk_conf
= new_disk_conf
;
1470 set_disk_conf_defaults(new_disk_conf
);
1471 err
= disk_conf_from_attrs(new_disk_conf
, info
);
1473 retcode
= ERR_MANDATORY_TAG
;
1474 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
1478 if (new_disk_conf
->c_plan_ahead
> DRBD_C_PLAN_AHEAD_MAX
)
1479 new_disk_conf
->c_plan_ahead
= DRBD_C_PLAN_AHEAD_MAX
;
1481 new_plan
= fifo_alloc((new_disk_conf
->c_plan_ahead
* 10 * SLEEP_TIME
) / HZ
);
1483 retcode
= ERR_NOMEM
;
1487 if (new_disk_conf
->meta_dev_idx
< DRBD_MD_INDEX_FLEX_INT
) {
1488 retcode
= ERR_MD_IDX_INVALID
;
1492 write_lock_irq(&global_state_lock
);
1493 retcode
= drbd_resync_after_valid(device
, new_disk_conf
->resync_after
);
1494 write_unlock_irq(&global_state_lock
);
1495 if (retcode
!= NO_ERROR
)
1499 nc
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
);
1501 if (new_disk_conf
->fencing
== FP_STONITH
&& nc
->wire_protocol
== DRBD_PROT_A
) {
1503 retcode
= ERR_STONITH_AND_PROT_A
;
1509 bdev
= blkdev_get_by_path(new_disk_conf
->backing_dev
,
1510 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, device
);
1512 drbd_err(device
, "open(\"%s\") failed with %ld\n", new_disk_conf
->backing_dev
,
1514 retcode
= ERR_OPEN_DISK
;
1517 nbc
->backing_bdev
= bdev
;
1520 * meta_dev_idx >= 0: external fixed size, possibly multiple
1521 * drbd sharing one meta device. TODO in that case, paranoia
1522 * check that [md_bdev, meta_dev_idx] is not yet used by some
1523 * other drbd minor! (if you use drbd.conf + drbdadm, that
1524 * should check it for you already; but if you don't, or
1525 * someone fooled it, we need to double check here)
1527 bdev
= blkdev_get_by_path(new_disk_conf
->meta_dev
,
1528 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
,
1529 (new_disk_conf
->meta_dev_idx
< 0) ?
1530 (void *)device
: (void *)drbd_m_holder
);
1532 drbd_err(device
, "open(\"%s\") failed with %ld\n", new_disk_conf
->meta_dev
,
1534 retcode
= ERR_OPEN_MD_DISK
;
1537 nbc
->md_bdev
= bdev
;
1539 if ((nbc
->backing_bdev
== nbc
->md_bdev
) !=
1540 (new_disk_conf
->meta_dev_idx
== DRBD_MD_INDEX_INTERNAL
||
1541 new_disk_conf
->meta_dev_idx
== DRBD_MD_INDEX_FLEX_INT
)) {
1542 retcode
= ERR_MD_IDX_INVALID
;
1546 resync_lru
= lc_create("resync", drbd_bm_ext_cache
,
1547 1, 61, sizeof(struct bm_extent
),
1548 offsetof(struct bm_extent
, lce
));
1550 retcode
= ERR_NOMEM
;
1554 /* Read our meta data super block early.
1555 * This also sets other on-disk offsets. */
1556 retcode
= drbd_md_read(device
, nbc
);
1557 if (retcode
!= NO_ERROR
)
1560 if (new_disk_conf
->al_extents
< DRBD_AL_EXTENTS_MIN
)
1561 new_disk_conf
->al_extents
= DRBD_AL_EXTENTS_MIN
;
1562 if (new_disk_conf
->al_extents
> drbd_al_extents_max(nbc
))
1563 new_disk_conf
->al_extents
= drbd_al_extents_max(nbc
);
1565 if (drbd_get_max_capacity(nbc
) < new_disk_conf
->disk_size
) {
1566 drbd_err(device
, "max capacity %llu smaller than disk size %llu\n",
1567 (unsigned long long) drbd_get_max_capacity(nbc
),
1568 (unsigned long long) new_disk_conf
->disk_size
);
1569 retcode
= ERR_DISK_TOO_SMALL
;
1573 if (new_disk_conf
->meta_dev_idx
< 0) {
1574 max_possible_sectors
= DRBD_MAX_SECTORS_FLEX
;
1575 /* at least one MB, otherwise it does not make sense */
1576 min_md_device_sectors
= (2<<10);
1578 max_possible_sectors
= DRBD_MAX_SECTORS
;
1579 min_md_device_sectors
= MD_128MB_SECT
* (new_disk_conf
->meta_dev_idx
+ 1);
1582 if (drbd_get_capacity(nbc
->md_bdev
) < min_md_device_sectors
) {
1583 retcode
= ERR_MD_DISK_TOO_SMALL
;
1584 drbd_warn(device
, "refusing attach: md-device too small, "
1585 "at least %llu sectors needed for this meta-disk type\n",
1586 (unsigned long long) min_md_device_sectors
);
1590 /* Make sure the new disk is big enough
1591 * (we may currently be R_PRIMARY with no local disk...) */
1592 if (drbd_get_max_capacity(nbc
) <
1593 drbd_get_capacity(device
->this_bdev
)) {
1594 retcode
= ERR_DISK_TOO_SMALL
;
1598 nbc
->known_size
= drbd_get_capacity(nbc
->backing_bdev
);
1600 if (nbc
->known_size
> max_possible_sectors
) {
1601 drbd_warn(device
, "==> truncating very big lower level device "
1602 "to currently maximum possible %llu sectors <==\n",
1603 (unsigned long long) max_possible_sectors
);
1604 if (new_disk_conf
->meta_dev_idx
>= 0)
1605 drbd_warn(device
, "==>> using internal or flexible "
1606 "meta data may help <<==\n");
1609 drbd_suspend_io(device
);
1610 /* also wait for the last barrier ack. */
1611 /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1612 * We need a way to either ignore barrier acks for barriers sent before a device
1613 * was attached, or a way to wait for all pending barrier acks to come in.
1614 * As barriers are counted per resource,
1615 * we'd need to suspend io on all devices of a resource.
1617 wait_event(device
->misc_wait
, !atomic_read(&device
->ap_pending_cnt
) || drbd_suspended(device
));
1618 /* and for any other previously queued work */
1619 drbd_flush_workqueue(&first_peer_device(device
)->connection
->sender_work
);
1621 rv
= _drbd_request_state(device
, NS(disk
, D_ATTACHING
), CS_VERBOSE
);
1622 retcode
= rv
; /* FIXME: Type mismatch. */
1623 drbd_resume_io(device
);
1624 if (rv
< SS_SUCCESS
)
1627 if (!get_ldev_if_state(device
, D_ATTACHING
))
1628 goto force_diskless
;
1630 if (!device
->bitmap
) {
1631 if (drbd_bm_init(device
)) {
1632 retcode
= ERR_NOMEM
;
1633 goto force_diskless_dec
;
1637 if (device
->state
.conn
< C_CONNECTED
&&
1638 device
->state
.role
== R_PRIMARY
&&
1639 (device
->ed_uuid
& ~((u64
)1)) != (nbc
->md
.uuid
[UI_CURRENT
] & ~((u64
)1))) {
1640 drbd_err(device
, "Can only attach to data with current UUID=%016llX\n",
1641 (unsigned long long)device
->ed_uuid
);
1642 retcode
= ERR_DATA_NOT_CURRENT
;
1643 goto force_diskless_dec
;
1646 /* Since we are diskless, fix the activity log first... */
1647 if (drbd_check_al_size(device
, new_disk_conf
)) {
1648 retcode
= ERR_NOMEM
;
1649 goto force_diskless_dec
;
1652 /* Prevent shrinking of consistent devices ! */
1653 if (drbd_md_test_flag(nbc
, MDF_CONSISTENT
) &&
1654 drbd_new_dev_size(device
, nbc
, nbc
->disk_conf
->disk_size
, 0) < nbc
->md
.la_size_sect
) {
1655 drbd_warn(device
, "refusing to truncate a consistent device\n");
1656 retcode
= ERR_DISK_TOO_SMALL
;
1657 goto force_diskless_dec
;
1660 /* Reset the "barriers don't work" bits here, then force meta data to
1661 * be written, to ensure we determine if barriers are supported. */
1662 if (new_disk_conf
->md_flushes
)
1663 clear_bit(MD_NO_FUA
, &device
->flags
);
1665 set_bit(MD_NO_FUA
, &device
->flags
);
1667 /* Point of no return reached.
1668 * Devices and memory are no longer released by error cleanup below.
1669 * now device takes over responsibility, and the state engine should
1670 * clean it up somewhere. */
1671 D_ASSERT(device
, device
->ldev
== NULL
);
1673 device
->resync
= resync_lru
;
1674 device
->rs_plan_s
= new_plan
;
1677 new_disk_conf
= NULL
;
1680 drbd_bump_write_ordering(first_peer_device(device
)->connection
, WO_bdev_flush
);
1682 if (drbd_md_test_flag(device
->ldev
, MDF_CRASHED_PRIMARY
))
1683 set_bit(CRASHED_PRIMARY
, &device
->flags
);
1685 clear_bit(CRASHED_PRIMARY
, &device
->flags
);
1687 if (drbd_md_test_flag(device
->ldev
, MDF_PRIMARY_IND
) &&
1688 !(device
->state
.role
== R_PRIMARY
&& device
->resource
->susp_nod
))
1689 set_bit(CRASHED_PRIMARY
, &device
->flags
);
1691 device
->send_cnt
= 0;
1692 device
->recv_cnt
= 0;
1693 device
->read_cnt
= 0;
1694 device
->writ_cnt
= 0;
1696 drbd_reconsider_max_bio_size(device
);
1698 /* If I am currently not R_PRIMARY,
1699 * but meta data primary indicator is set,
1700 * I just now recover from a hard crash,
1701 * and have been R_PRIMARY before that crash.
1703 * Now, if I had no connection before that crash
1704 * (have been degraded R_PRIMARY), chances are that
1705 * I won't find my peer now either.
1707 * In that case, and _only_ in that case,
1708 * we use the degr-wfc-timeout instead of the default,
1709 * so we can automatically recover from a crash of a
1710 * degraded but active "cluster" after a certain timeout.
1712 clear_bit(USE_DEGR_WFC_T
, &device
->flags
);
1713 if (device
->state
.role
!= R_PRIMARY
&&
1714 drbd_md_test_flag(device
->ldev
, MDF_PRIMARY_IND
) &&
1715 !drbd_md_test_flag(device
->ldev
, MDF_CONNECTED_IND
))
1716 set_bit(USE_DEGR_WFC_T
, &device
->flags
);
1718 dd
= drbd_determine_dev_size(device
, 0, NULL
);
1719 if (dd
<= DS_ERROR
) {
1720 retcode
= ERR_NOMEM_BITMAP
;
1721 goto force_diskless_dec
;
1722 } else if (dd
== DS_GREW
)
1723 set_bit(RESYNC_AFTER_NEG
, &device
->flags
);
1725 if (drbd_md_test_flag(device
->ldev
, MDF_FULL_SYNC
) ||
1726 (test_bit(CRASHED_PRIMARY
, &device
->flags
) &&
1727 drbd_md_test_flag(device
->ldev
, MDF_AL_DISABLED
))) {
1728 drbd_info(device
, "Assuming that all blocks are out of sync "
1729 "(aka FullSync)\n");
1730 if (drbd_bitmap_io(device
, &drbd_bmio_set_n_write
,
1731 "set_n_write from attaching", BM_LOCKED_MASK
)) {
1732 retcode
= ERR_IO_MD_DISK
;
1733 goto force_diskless_dec
;
1736 if (drbd_bitmap_io(device
, &drbd_bm_read
,
1737 "read from attaching", BM_LOCKED_MASK
)) {
1738 retcode
= ERR_IO_MD_DISK
;
1739 goto force_diskless_dec
;
1743 if (_drbd_bm_total_weight(device
) == drbd_bm_bits(device
))
1744 drbd_suspend_al(device
); /* IO is still suspended here... */
1746 spin_lock_irq(&device
->resource
->req_lock
);
1747 os
= drbd_read_state(device
);
1749 /* If MDF_CONSISTENT is not set go into inconsistent state,
1750 otherwise investigate MDF_WasUpToDate...
1751 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1752 otherwise into D_CONSISTENT state.
1754 if (drbd_md_test_flag(device
->ldev
, MDF_CONSISTENT
)) {
1755 if (drbd_md_test_flag(device
->ldev
, MDF_WAS_UP_TO_DATE
))
1756 ns
.disk
= D_CONSISTENT
;
1758 ns
.disk
= D_OUTDATED
;
1760 ns
.disk
= D_INCONSISTENT
;
1763 if (drbd_md_test_flag(device
->ldev
, MDF_PEER_OUT_DATED
))
1764 ns
.pdsk
= D_OUTDATED
;
1767 if (ns
.disk
== D_CONSISTENT
&&
1768 (ns
.pdsk
== D_OUTDATED
|| rcu_dereference(device
->ldev
->disk_conf
)->fencing
== FP_DONT_CARE
))
1769 ns
.disk
= D_UP_TO_DATE
;
1771 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1772 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1773 this point, because drbd_request_state() modifies these
1776 if (rcu_dereference(device
->ldev
->disk_conf
)->al_updates
)
1777 device
->ldev
->md
.flags
&= ~MDF_AL_DISABLED
;
1779 device
->ldev
->md
.flags
|= MDF_AL_DISABLED
;
1783 /* In case we are C_CONNECTED postpone any decision on the new disk
1784 state after the negotiation phase. */
1785 if (device
->state
.conn
== C_CONNECTED
) {
1786 device
->new_state_tmp
.i
= ns
.i
;
1788 ns
.disk
= D_NEGOTIATING
;
1790 /* We expect to receive up-to-date UUIDs soon.
1791 To avoid a race in receive_state, free p_uuid while
1792 holding req_lock. I.e. atomic with the state change */
1793 kfree(device
->p_uuid
);
1794 device
->p_uuid
= NULL
;
1797 rv
= _drbd_set_state(device
, ns
, CS_VERBOSE
, NULL
);
1798 spin_unlock_irq(&device
->resource
->req_lock
);
1800 if (rv
< SS_SUCCESS
)
1801 goto force_diskless_dec
;
1803 mod_timer(&device
->request_timer
, jiffies
+ HZ
);
1805 if (device
->state
.role
== R_PRIMARY
)
1806 device
->ldev
->md
.uuid
[UI_CURRENT
] |= (u64
)1;
1808 device
->ldev
->md
.uuid
[UI_CURRENT
] &= ~(u64
)1;
1810 drbd_md_mark_dirty(device
);
1811 drbd_md_sync(device
);
1813 kobject_uevent(&disk_to_dev(device
->vdisk
)->kobj
, KOBJ_CHANGE
);
1815 conn_reconfig_done(first_peer_device(device
)->connection
);
1816 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
1817 drbd_adm_finish(&adm_ctx
, info
, retcode
);
1823 drbd_force_state(device
, NS(disk
, D_DISKLESS
));
1824 drbd_md_sync(device
);
1826 conn_reconfig_done(first_peer_device(device
)->connection
);
1828 if (nbc
->backing_bdev
)
1829 blkdev_put(nbc
->backing_bdev
,
1830 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
1832 blkdev_put(nbc
->md_bdev
,
1833 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
1836 kfree(new_disk_conf
);
1837 lc_destroy(resync_lru
);
1839 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
1841 drbd_adm_finish(&adm_ctx
, info
, retcode
);
1845 static int adm_detach(struct drbd_device
*device
, int force
)
1847 enum drbd_state_rv retcode
;
1851 set_bit(FORCE_DETACH
, &device
->flags
);
1852 drbd_force_state(device
, NS(disk
, D_FAILED
));
1853 retcode
= SS_SUCCESS
;
1857 drbd_suspend_io(device
); /* so no-one is stuck in drbd_al_begin_io */
1858 drbd_md_get_buffer(device
); /* make sure there is no in-flight meta-data IO */
1859 retcode
= drbd_request_state(device
, NS(disk
, D_FAILED
));
1860 drbd_md_put_buffer(device
);
1861 /* D_FAILED will transition to DISKLESS. */
1862 ret
= wait_event_interruptible(device
->misc_wait
,
1863 device
->state
.disk
!= D_FAILED
);
1864 drbd_resume_io(device
);
1865 if ((int)retcode
== (int)SS_IS_DISKLESS
)
1866 retcode
= SS_NOTHING_TO_DO
;
1873 /* Detaching the disk is a process in multiple stages. First we need to lock
1874 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1875 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1876 * internal references as well.
1877 * Only then we have finally detached. */
1878 int drbd_adm_detach(struct sk_buff
*skb
, struct genl_info
*info
)
1880 struct drbd_config_context adm_ctx
;
1881 enum drbd_ret_code retcode
;
1882 struct detach_parms parms
= { };
1885 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
1886 if (!adm_ctx
.reply_skb
)
1888 if (retcode
!= NO_ERROR
)
1891 if (info
->attrs
[DRBD_NLA_DETACH_PARMS
]) {
1892 err
= detach_parms_from_attrs(&parms
, info
);
1894 retcode
= ERR_MANDATORY_TAG
;
1895 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
1900 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
1901 retcode
= adm_detach(adm_ctx
.device
, parms
.force_detach
);
1902 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
1904 drbd_adm_finish(&adm_ctx
, info
, retcode
);
1908 static bool conn_resync_running(struct drbd_connection
*connection
)
1910 struct drbd_peer_device
*peer_device
;
1915 idr_for_each_entry(&connection
->peer_devices
, peer_device
, vnr
) {
1916 struct drbd_device
*device
= peer_device
->device
;
1917 if (device
->state
.conn
== C_SYNC_SOURCE
||
1918 device
->state
.conn
== C_SYNC_TARGET
||
1919 device
->state
.conn
== C_PAUSED_SYNC_S
||
1920 device
->state
.conn
== C_PAUSED_SYNC_T
) {
1930 static bool conn_ov_running(struct drbd_connection
*connection
)
1932 struct drbd_peer_device
*peer_device
;
1937 idr_for_each_entry(&connection
->peer_devices
, peer_device
, vnr
) {
1938 struct drbd_device
*device
= peer_device
->device
;
1939 if (device
->state
.conn
== C_VERIFY_S
||
1940 device
->state
.conn
== C_VERIFY_T
) {
1950 static enum drbd_ret_code
1951 _check_net_options(struct drbd_connection
*connection
, struct net_conf
*old_net_conf
, struct net_conf
*new_net_conf
)
1953 struct drbd_peer_device
*peer_device
;
1956 if (old_net_conf
&& connection
->cstate
== C_WF_REPORT_PARAMS
&& connection
->agreed_pro_version
< 100) {
1957 if (new_net_conf
->wire_protocol
!= old_net_conf
->wire_protocol
)
1958 return ERR_NEED_APV_100
;
1960 if (new_net_conf
->two_primaries
!= old_net_conf
->two_primaries
)
1961 return ERR_NEED_APV_100
;
1963 if (strcmp(new_net_conf
->integrity_alg
, old_net_conf
->integrity_alg
))
1964 return ERR_NEED_APV_100
;
1967 if (!new_net_conf
->two_primaries
&&
1968 conn_highest_role(connection
) == R_PRIMARY
&&
1969 conn_highest_peer(connection
) == R_PRIMARY
)
1970 return ERR_NEED_ALLOW_TWO_PRI
;
1972 if (new_net_conf
->two_primaries
&&
1973 (new_net_conf
->wire_protocol
!= DRBD_PROT_C
))
1974 return ERR_NOT_PROTO_C
;
1976 idr_for_each_entry(&connection
->peer_devices
, peer_device
, i
) {
1977 struct drbd_device
*device
= peer_device
->device
;
1978 if (get_ldev(device
)) {
1979 enum drbd_fencing_p fp
= rcu_dereference(device
->ldev
->disk_conf
)->fencing
;
1981 if (new_net_conf
->wire_protocol
== DRBD_PROT_A
&& fp
== FP_STONITH
)
1982 return ERR_STONITH_AND_PROT_A
;
1984 if (device
->state
.role
== R_PRIMARY
&& new_net_conf
->discard_my_data
)
1985 return ERR_DISCARD_IMPOSSIBLE
;
1988 if (new_net_conf
->on_congestion
!= OC_BLOCK
&& new_net_conf
->wire_protocol
!= DRBD_PROT_A
)
1989 return ERR_CONG_NOT_PROTO_A
;
1994 static enum drbd_ret_code
1995 check_net_options(struct drbd_connection
*connection
, struct net_conf
*new_net_conf
)
1997 static enum drbd_ret_code rv
;
1998 struct drbd_peer_device
*peer_device
;
2002 rv
= _check_net_options(connection
, rcu_dereference(connection
->net_conf
), new_net_conf
);
2005 /* connection->volumes protected by genl_lock() here */
2006 idr_for_each_entry(&connection
->peer_devices
, peer_device
, i
) {
2007 struct drbd_device
*device
= peer_device
->device
;
2008 if (!device
->bitmap
) {
2009 if (drbd_bm_init(device
))
2018 struct crypto_hash
*verify_tfm
;
2019 struct crypto_hash
*csums_tfm
;
2020 struct crypto_hash
*cram_hmac_tfm
;
2021 struct crypto_hash
*integrity_tfm
;
2025 alloc_hash(struct crypto_hash
**tfm
, char *tfm_name
, int err_alg
)
2030 *tfm
= crypto_alloc_hash(tfm_name
, 0, CRYPTO_ALG_ASYNC
);
2039 static enum drbd_ret_code
2040 alloc_crypto(struct crypto
*crypto
, struct net_conf
*new_net_conf
)
2042 char hmac_name
[CRYPTO_MAX_ALG_NAME
];
2043 enum drbd_ret_code rv
;
2045 rv
= alloc_hash(&crypto
->csums_tfm
, new_net_conf
->csums_alg
,
2049 rv
= alloc_hash(&crypto
->verify_tfm
, new_net_conf
->verify_alg
,
2053 rv
= alloc_hash(&crypto
->integrity_tfm
, new_net_conf
->integrity_alg
,
2057 if (new_net_conf
->cram_hmac_alg
[0] != 0) {
2058 snprintf(hmac_name
, CRYPTO_MAX_ALG_NAME
, "hmac(%s)",
2059 new_net_conf
->cram_hmac_alg
);
2061 rv
= alloc_hash(&crypto
->cram_hmac_tfm
, hmac_name
,
2068 static void free_crypto(struct crypto
*crypto
)
2070 crypto_free_hash(crypto
->cram_hmac_tfm
);
2071 crypto_free_hash(crypto
->integrity_tfm
);
2072 crypto_free_hash(crypto
->csums_tfm
);
2073 crypto_free_hash(crypto
->verify_tfm
);
2076 int drbd_adm_net_opts(struct sk_buff
*skb
, struct genl_info
*info
)
2078 struct drbd_config_context adm_ctx
;
2079 enum drbd_ret_code retcode
;
2080 struct drbd_connection
*connection
;
2081 struct net_conf
*old_net_conf
, *new_net_conf
= NULL
;
2083 int ovr
; /* online verify running */
2084 int rsr
; /* re-sync running */
2085 struct crypto crypto
= { };
2087 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_CONNECTION
);
2088 if (!adm_ctx
.reply_skb
)
2090 if (retcode
!= NO_ERROR
)
2093 connection
= adm_ctx
.connection
;
2094 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2096 new_net_conf
= kzalloc(sizeof(struct net_conf
), GFP_KERNEL
);
2097 if (!new_net_conf
) {
2098 retcode
= ERR_NOMEM
;
2102 conn_reconfig_start(connection
);
2104 mutex_lock(&connection
->data
.mutex
);
2105 mutex_lock(&connection
->resource
->conf_update
);
2106 old_net_conf
= connection
->net_conf
;
2108 if (!old_net_conf
) {
2109 drbd_msg_put_info(adm_ctx
.reply_skb
, "net conf missing, try connect");
2110 retcode
= ERR_INVALID_REQUEST
;
2114 *new_net_conf
= *old_net_conf
;
2115 if (should_set_defaults(info
))
2116 set_net_conf_defaults(new_net_conf
);
2118 err
= net_conf_from_attrs_for_change(new_net_conf
, info
);
2119 if (err
&& err
!= -ENOMSG
) {
2120 retcode
= ERR_MANDATORY_TAG
;
2121 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2125 retcode
= check_net_options(connection
, new_net_conf
);
2126 if (retcode
!= NO_ERROR
)
2129 /* re-sync running */
2130 rsr
= conn_resync_running(connection
);
2131 if (rsr
&& strcmp(new_net_conf
->csums_alg
, old_net_conf
->csums_alg
)) {
2132 retcode
= ERR_CSUMS_RESYNC_RUNNING
;
2136 /* online verify running */
2137 ovr
= conn_ov_running(connection
);
2138 if (ovr
&& strcmp(new_net_conf
->verify_alg
, old_net_conf
->verify_alg
)) {
2139 retcode
= ERR_VERIFY_RUNNING
;
2143 retcode
= alloc_crypto(&crypto
, new_net_conf
);
2144 if (retcode
!= NO_ERROR
)
2147 rcu_assign_pointer(connection
->net_conf
, new_net_conf
);
2150 crypto_free_hash(connection
->csums_tfm
);
2151 connection
->csums_tfm
= crypto
.csums_tfm
;
2152 crypto
.csums_tfm
= NULL
;
2155 crypto_free_hash(connection
->verify_tfm
);
2156 connection
->verify_tfm
= crypto
.verify_tfm
;
2157 crypto
.verify_tfm
= NULL
;
2160 crypto_free_hash(connection
->integrity_tfm
);
2161 connection
->integrity_tfm
= crypto
.integrity_tfm
;
2162 if (connection
->cstate
>= C_WF_REPORT_PARAMS
&& connection
->agreed_pro_version
>= 100)
2163 /* Do this without trying to take connection->data.mutex again. */
2164 __drbd_send_protocol(connection
, P_PROTOCOL_UPDATE
);
2166 crypto_free_hash(connection
->cram_hmac_tfm
);
2167 connection
->cram_hmac_tfm
= crypto
.cram_hmac_tfm
;
2169 mutex_unlock(&connection
->resource
->conf_update
);
2170 mutex_unlock(&connection
->data
.mutex
);
2172 kfree(old_net_conf
);
2174 if (connection
->cstate
>= C_WF_REPORT_PARAMS
) {
2175 struct drbd_peer_device
*peer_device
;
2178 idr_for_each_entry(&connection
->peer_devices
, peer_device
, vnr
)
2179 drbd_send_sync_param(peer_device
);
2185 mutex_unlock(&connection
->resource
->conf_update
);
2186 mutex_unlock(&connection
->data
.mutex
);
2187 free_crypto(&crypto
);
2188 kfree(new_net_conf
);
2190 conn_reconfig_done(connection
);
2192 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2194 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2198 int drbd_adm_connect(struct sk_buff
*skb
, struct genl_info
*info
)
2200 struct drbd_config_context adm_ctx
;
2201 struct drbd_peer_device
*peer_device
;
2202 struct net_conf
*old_net_conf
, *new_net_conf
= NULL
;
2203 struct crypto crypto
= { };
2204 struct drbd_resource
*resource
;
2205 struct drbd_connection
*connection
;
2206 enum drbd_ret_code retcode
;
2210 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
2212 if (!adm_ctx
.reply_skb
)
2214 if (retcode
!= NO_ERROR
)
2216 if (!(adm_ctx
.my_addr
&& adm_ctx
.peer_addr
)) {
2217 drbd_msg_put_info(adm_ctx
.reply_skb
, "connection endpoint(s) missing");
2218 retcode
= ERR_INVALID_REQUEST
;
2222 /* No need for _rcu here. All reconfiguration is
2223 * strictly serialized on genl_lock(). We are protected against
2224 * concurrent reconfiguration/addition/deletion */
2225 for_each_resource(resource
, &drbd_resources
) {
2226 for_each_connection(connection
, resource
) {
2227 if (nla_len(adm_ctx
.my_addr
) == connection
->my_addr_len
&&
2228 !memcmp(nla_data(adm_ctx
.my_addr
), &connection
->my_addr
,
2229 connection
->my_addr_len
)) {
2230 retcode
= ERR_LOCAL_ADDR
;
2234 if (nla_len(adm_ctx
.peer_addr
) == connection
->peer_addr_len
&&
2235 !memcmp(nla_data(adm_ctx
.peer_addr
), &connection
->peer_addr
,
2236 connection
->peer_addr_len
)) {
2237 retcode
= ERR_PEER_ADDR
;
2243 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2244 connection
= first_connection(adm_ctx
.resource
);
2245 conn_reconfig_start(connection
);
2247 if (connection
->cstate
> C_STANDALONE
) {
2248 retcode
= ERR_NET_CONFIGURED
;
2252 /* allocation not in the IO path, drbdsetup / netlink process context */
2253 new_net_conf
= kzalloc(sizeof(*new_net_conf
), GFP_KERNEL
);
2254 if (!new_net_conf
) {
2255 retcode
= ERR_NOMEM
;
2259 set_net_conf_defaults(new_net_conf
);
2261 err
= net_conf_from_attrs(new_net_conf
, info
);
2262 if (err
&& err
!= -ENOMSG
) {
2263 retcode
= ERR_MANDATORY_TAG
;
2264 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2268 retcode
= check_net_options(connection
, new_net_conf
);
2269 if (retcode
!= NO_ERROR
)
2272 retcode
= alloc_crypto(&crypto
, new_net_conf
);
2273 if (retcode
!= NO_ERROR
)
2276 ((char *)new_net_conf
->shared_secret
)[SHARED_SECRET_MAX
-1] = 0;
2278 drbd_flush_workqueue(&connection
->sender_work
);
2280 mutex_lock(&adm_ctx
.resource
->conf_update
);
2281 old_net_conf
= connection
->net_conf
;
2283 retcode
= ERR_NET_CONFIGURED
;
2284 mutex_unlock(&adm_ctx
.resource
->conf_update
);
2287 rcu_assign_pointer(connection
->net_conf
, new_net_conf
);
2289 conn_free_crypto(connection
);
2290 connection
->cram_hmac_tfm
= crypto
.cram_hmac_tfm
;
2291 connection
->integrity_tfm
= crypto
.integrity_tfm
;
2292 connection
->csums_tfm
= crypto
.csums_tfm
;
2293 connection
->verify_tfm
= crypto
.verify_tfm
;
2295 connection
->my_addr_len
= nla_len(adm_ctx
.my_addr
);
2296 memcpy(&connection
->my_addr
, nla_data(adm_ctx
.my_addr
), connection
->my_addr_len
);
2297 connection
->peer_addr_len
= nla_len(adm_ctx
.peer_addr
);
2298 memcpy(&connection
->peer_addr
, nla_data(adm_ctx
.peer_addr
), connection
->peer_addr_len
);
2300 mutex_unlock(&adm_ctx
.resource
->conf_update
);
2303 idr_for_each_entry(&connection
->peer_devices
, peer_device
, i
) {
2304 struct drbd_device
*device
= peer_device
->device
;
2305 device
->send_cnt
= 0;
2306 device
->recv_cnt
= 0;
2310 retcode
= conn_request_state(connection
, NS(conn
, C_UNCONNECTED
), CS_VERBOSE
);
2312 conn_reconfig_done(connection
);
2313 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2314 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2318 free_crypto(&crypto
);
2319 kfree(new_net_conf
);
2321 conn_reconfig_done(connection
);
2322 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2324 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2328 static enum drbd_state_rv
conn_try_disconnect(struct drbd_connection
*connection
, bool force
)
2330 enum drbd_state_rv rv
;
2332 rv
= conn_request_state(connection
, NS(conn
, C_DISCONNECTING
),
2333 force
? CS_HARD
: 0);
2336 case SS_NOTHING_TO_DO
:
2338 case SS_ALREADY_STANDALONE
:
2340 case SS_PRIMARY_NOP
:
2341 /* Our state checking code wants to see the peer outdated. */
2342 rv
= conn_request_state(connection
, NS2(conn
, C_DISCONNECTING
, pdsk
, D_OUTDATED
), 0);
2344 if (rv
== SS_OUTDATE_WO_CONN
) /* lost connection before graceful disconnect succeeded */
2345 rv
= conn_request_state(connection
, NS(conn
, C_DISCONNECTING
), CS_VERBOSE
);
2348 case SS_CW_FAILED_BY_PEER
:
2349 /* The peer probably wants to see us outdated. */
2350 rv
= conn_request_state(connection
, NS2(conn
, C_DISCONNECTING
,
2351 disk
, D_OUTDATED
), 0);
2352 if (rv
== SS_IS_DISKLESS
|| rv
== SS_LOWER_THAN_OUTDATED
) {
2353 rv
= conn_request_state(connection
, NS(conn
, C_DISCONNECTING
),
2358 /* no special handling necessary */
2361 if (rv
>= SS_SUCCESS
) {
2362 enum drbd_state_rv rv2
;
2363 /* No one else can reconfigure the network while I am here.
2364 * The state handling only uses drbd_thread_stop_nowait(),
2365 * we want to really wait here until the receiver is no more.
2367 drbd_thread_stop(&connection
->receiver
);
2369 /* Race breaker. This additional state change request may be
2370 * necessary, if this was a forced disconnect during a receiver
2371 * restart. We may have "killed" the receiver thread just
2372 * after drbd_receiver() returned. Typically, we should be
2373 * C_STANDALONE already, now, and this becomes a no-op.
2375 rv2
= conn_request_state(connection
, NS(conn
, C_STANDALONE
),
2376 CS_VERBOSE
| CS_HARD
);
2377 if (rv2
< SS_SUCCESS
)
2378 drbd_err(connection
,
2379 "unexpected rv2=%d in conn_try_disconnect()\n",
2385 int drbd_adm_disconnect(struct sk_buff
*skb
, struct genl_info
*info
)
2387 struct drbd_config_context adm_ctx
;
2388 struct disconnect_parms parms
;
2389 struct drbd_connection
*connection
;
2390 enum drbd_state_rv rv
;
2391 enum drbd_ret_code retcode
;
2394 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_CONNECTION
);
2395 if (!adm_ctx
.reply_skb
)
2397 if (retcode
!= NO_ERROR
)
2400 connection
= adm_ctx
.connection
;
2401 memset(&parms
, 0, sizeof(parms
));
2402 if (info
->attrs
[DRBD_NLA_DISCONNECT_PARMS
]) {
2403 err
= disconnect_parms_from_attrs(&parms
, info
);
2405 retcode
= ERR_MANDATORY_TAG
;
2406 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2411 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2412 rv
= conn_try_disconnect(connection
, parms
.force_disconnect
);
2413 if (rv
< SS_SUCCESS
)
2414 retcode
= rv
; /* FIXME: Type mismatch. */
2417 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2419 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2423 void resync_after_online_grow(struct drbd_device
*device
)
2425 int iass
; /* I am sync source */
2427 drbd_info(device
, "Resync of new storage after online grow\n");
2428 if (device
->state
.role
!= device
->state
.peer
)
2429 iass
= (device
->state
.role
== R_PRIMARY
);
2431 iass
= test_bit(RESOLVE_CONFLICTS
, &first_peer_device(device
)->connection
->flags
);
2434 drbd_start_resync(device
, C_SYNC_SOURCE
);
2436 _drbd_request_state(device
, NS(conn
, C_WF_SYNC_UUID
), CS_VERBOSE
+ CS_SERIALIZE
);
2439 int drbd_adm_resize(struct sk_buff
*skb
, struct genl_info
*info
)
2441 struct drbd_config_context adm_ctx
;
2442 struct disk_conf
*old_disk_conf
, *new_disk_conf
= NULL
;
2443 struct resize_parms rs
;
2444 struct drbd_device
*device
;
2445 enum drbd_ret_code retcode
;
2446 enum determine_dev_size dd
;
2447 bool change_al_layout
= false;
2448 enum dds_flags ddsf
;
2452 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2453 if (!adm_ctx
.reply_skb
)
2455 if (retcode
!= NO_ERROR
)
2458 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2459 device
= adm_ctx
.device
;
2460 if (!get_ldev(device
)) {
2461 retcode
= ERR_NO_DISK
;
2465 memset(&rs
, 0, sizeof(struct resize_parms
));
2466 rs
.al_stripes
= device
->ldev
->md
.al_stripes
;
2467 rs
.al_stripe_size
= device
->ldev
->md
.al_stripe_size_4k
* 4;
2468 if (info
->attrs
[DRBD_NLA_RESIZE_PARMS
]) {
2469 err
= resize_parms_from_attrs(&rs
, info
);
2471 retcode
= ERR_MANDATORY_TAG
;
2472 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2477 if (device
->state
.conn
> C_CONNECTED
) {
2478 retcode
= ERR_RESIZE_RESYNC
;
2482 if (device
->state
.role
== R_SECONDARY
&&
2483 device
->state
.peer
== R_SECONDARY
) {
2484 retcode
= ERR_NO_PRIMARY
;
2488 if (rs
.no_resync
&& first_peer_device(device
)->connection
->agreed_pro_version
< 93) {
2489 retcode
= ERR_NEED_APV_93
;
2494 u_size
= rcu_dereference(device
->ldev
->disk_conf
)->disk_size
;
2496 if (u_size
!= (sector_t
)rs
.resize_size
) {
2497 new_disk_conf
= kmalloc(sizeof(struct disk_conf
), GFP_KERNEL
);
2498 if (!new_disk_conf
) {
2499 retcode
= ERR_NOMEM
;
2504 if (device
->ldev
->md
.al_stripes
!= rs
.al_stripes
||
2505 device
->ldev
->md
.al_stripe_size_4k
!= rs
.al_stripe_size
/ 4) {
2506 u32 al_size_k
= rs
.al_stripes
* rs
.al_stripe_size
;
2508 if (al_size_k
> (16 * 1024 * 1024)) {
2509 retcode
= ERR_MD_LAYOUT_TOO_BIG
;
2513 if (al_size_k
< MD_32kB_SECT
/2) {
2514 retcode
= ERR_MD_LAYOUT_TOO_SMALL
;
2518 if (device
->state
.conn
!= C_CONNECTED
&& !rs
.resize_force
) {
2519 retcode
= ERR_MD_LAYOUT_CONNECTED
;
2523 change_al_layout
= true;
2526 if (device
->ldev
->known_size
!= drbd_get_capacity(device
->ldev
->backing_bdev
))
2527 device
->ldev
->known_size
= drbd_get_capacity(device
->ldev
->backing_bdev
);
2529 if (new_disk_conf
) {
2530 mutex_lock(&device
->resource
->conf_update
);
2531 old_disk_conf
= device
->ldev
->disk_conf
;
2532 *new_disk_conf
= *old_disk_conf
;
2533 new_disk_conf
->disk_size
= (sector_t
)rs
.resize_size
;
2534 rcu_assign_pointer(device
->ldev
->disk_conf
, new_disk_conf
);
2535 mutex_unlock(&device
->resource
->conf_update
);
2537 kfree(old_disk_conf
);
2540 ddsf
= (rs
.resize_force
? DDSF_FORCED
: 0) | (rs
.no_resync
? DDSF_NO_RESYNC
: 0);
2541 dd
= drbd_determine_dev_size(device
, ddsf
, change_al_layout
? &rs
: NULL
);
2542 drbd_md_sync(device
);
2544 if (dd
== DS_ERROR
) {
2545 retcode
= ERR_NOMEM_BITMAP
;
2547 } else if (dd
== DS_ERROR_SPACE_MD
) {
2548 retcode
= ERR_MD_LAYOUT_NO_FIT
;
2550 } else if (dd
== DS_ERROR_SHRINK
) {
2551 retcode
= ERR_IMPLICIT_SHRINK
;
2555 if (device
->state
.conn
== C_CONNECTED
) {
2557 set_bit(RESIZE_PENDING
, &device
->flags
);
2559 drbd_send_uuids(first_peer_device(device
));
2560 drbd_send_sizes(first_peer_device(device
), 1, ddsf
);
2564 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2566 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2574 int drbd_adm_resource_opts(struct sk_buff
*skb
, struct genl_info
*info
)
2576 struct drbd_config_context adm_ctx
;
2577 enum drbd_ret_code retcode
;
2578 struct res_opts res_opts
;
2581 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
2582 if (!adm_ctx
.reply_skb
)
2584 if (retcode
!= NO_ERROR
)
2587 res_opts
= adm_ctx
.resource
->res_opts
;
2588 if (should_set_defaults(info
))
2589 set_res_opts_defaults(&res_opts
);
2591 err
= res_opts_from_attrs(&res_opts
, info
);
2592 if (err
&& err
!= -ENOMSG
) {
2593 retcode
= ERR_MANDATORY_TAG
;
2594 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2598 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2599 err
= set_resource_options(adm_ctx
.resource
, &res_opts
);
2601 retcode
= ERR_INVALID_REQUEST
;
2603 retcode
= ERR_NOMEM
;
2605 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2608 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2612 int drbd_adm_invalidate(struct sk_buff
*skb
, struct genl_info
*info
)
2614 struct drbd_config_context adm_ctx
;
2615 struct drbd_device
*device
;
2616 int retcode
; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2618 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2619 if (!adm_ctx
.reply_skb
)
2621 if (retcode
!= NO_ERROR
)
2624 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2625 device
= adm_ctx
.device
;
2627 /* If there is still bitmap IO pending, probably because of a previous
2628 * resync just being finished, wait for it before requesting a new resync.
2629 * Also wait for it's after_state_ch(). */
2630 drbd_suspend_io(device
);
2631 wait_event(device
->misc_wait
, !test_bit(BITMAP_IO
, &device
->flags
));
2632 drbd_flush_workqueue(&first_peer_device(device
)->connection
->sender_work
);
2634 /* If we happen to be C_STANDALONE R_SECONDARY, just change to
2635 * D_INCONSISTENT, and set all bits in the bitmap. Otherwise,
2636 * try to start a resync handshake as sync target for full sync.
2638 if (device
->state
.conn
== C_STANDALONE
&& device
->state
.role
== R_SECONDARY
) {
2639 retcode
= drbd_request_state(device
, NS(disk
, D_INCONSISTENT
));
2640 if (retcode
>= SS_SUCCESS
) {
2641 if (drbd_bitmap_io(device
, &drbd_bmio_set_n_write
,
2642 "set_n_write from invalidate", BM_LOCKED_MASK
))
2643 retcode
= ERR_IO_MD_DISK
;
2646 retcode
= drbd_request_state(device
, NS(conn
, C_STARTING_SYNC_T
));
2647 drbd_resume_io(device
);
2648 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2650 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2654 static int drbd_adm_simple_request_state(struct sk_buff
*skb
, struct genl_info
*info
,
2655 union drbd_state mask
, union drbd_state val
)
2657 struct drbd_config_context adm_ctx
;
2658 enum drbd_ret_code retcode
;
2660 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2661 if (!adm_ctx
.reply_skb
)
2663 if (retcode
!= NO_ERROR
)
2666 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2667 retcode
= drbd_request_state(adm_ctx
.device
, mask
, val
);
2668 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2670 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2674 static int drbd_bmio_set_susp_al(struct drbd_device
*device
)
2678 rv
= drbd_bmio_set_n_write(device
);
2679 drbd_suspend_al(device
);
2683 int drbd_adm_invalidate_peer(struct sk_buff
*skb
, struct genl_info
*info
)
2685 struct drbd_config_context adm_ctx
;
2686 int retcode
; /* drbd_ret_code, drbd_state_rv */
2687 struct drbd_device
*device
;
2689 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2690 if (!adm_ctx
.reply_skb
)
2692 if (retcode
!= NO_ERROR
)
2695 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2696 device
= adm_ctx
.device
;
2698 /* If there is still bitmap IO pending, probably because of a previous
2699 * resync just being finished, wait for it before requesting a new resync.
2700 * Also wait for it's after_state_ch(). */
2701 drbd_suspend_io(device
);
2702 wait_event(device
->misc_wait
, !test_bit(BITMAP_IO
, &device
->flags
));
2703 drbd_flush_workqueue(&first_peer_device(device
)->connection
->sender_work
);
2705 /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
2706 * in the bitmap. Otherwise, try to start a resync handshake
2707 * as sync source for full sync.
2709 if (device
->state
.conn
== C_STANDALONE
&& device
->state
.role
== R_PRIMARY
) {
2710 /* The peer will get a resync upon connect anyways. Just make that
2711 into a full resync. */
2712 retcode
= drbd_request_state(device
, NS(pdsk
, D_INCONSISTENT
));
2713 if (retcode
>= SS_SUCCESS
) {
2714 if (drbd_bitmap_io(device
, &drbd_bmio_set_susp_al
,
2715 "set_n_write from invalidate_peer",
2716 BM_LOCKED_SET_ALLOWED
))
2717 retcode
= ERR_IO_MD_DISK
;
2720 retcode
= drbd_request_state(device
, NS(conn
, C_STARTING_SYNC_S
));
2721 drbd_resume_io(device
);
2722 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2724 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2728 int drbd_adm_pause_sync(struct sk_buff
*skb
, struct genl_info
*info
)
2730 struct drbd_config_context adm_ctx
;
2731 enum drbd_ret_code retcode
;
2733 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2734 if (!adm_ctx
.reply_skb
)
2736 if (retcode
!= NO_ERROR
)
2739 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2740 if (drbd_request_state(adm_ctx
.device
, NS(user_isp
, 1)) == SS_NOTHING_TO_DO
)
2741 retcode
= ERR_PAUSE_IS_SET
;
2742 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2744 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2748 int drbd_adm_resume_sync(struct sk_buff
*skb
, struct genl_info
*info
)
2750 struct drbd_config_context adm_ctx
;
2751 union drbd_dev_state s
;
2752 enum drbd_ret_code retcode
;
2754 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2755 if (!adm_ctx
.reply_skb
)
2757 if (retcode
!= NO_ERROR
)
2760 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2761 if (drbd_request_state(adm_ctx
.device
, NS(user_isp
, 0)) == SS_NOTHING_TO_DO
) {
2762 s
= adm_ctx
.device
->state
;
2763 if (s
.conn
== C_PAUSED_SYNC_S
|| s
.conn
== C_PAUSED_SYNC_T
) {
2764 retcode
= s
.aftr_isp
? ERR_PIC_AFTER_DEP
:
2765 s
.peer_isp
? ERR_PIC_PEER_DEP
: ERR_PAUSE_IS_CLEAR
;
2767 retcode
= ERR_PAUSE_IS_CLEAR
;
2770 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2772 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2776 int drbd_adm_suspend_io(struct sk_buff
*skb
, struct genl_info
*info
)
2778 return drbd_adm_simple_request_state(skb
, info
, NS(susp
, 1));
2781 int drbd_adm_resume_io(struct sk_buff
*skb
, struct genl_info
*info
)
2783 struct drbd_config_context adm_ctx
;
2784 struct drbd_device
*device
;
2785 int retcode
; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2787 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2788 if (!adm_ctx
.reply_skb
)
2790 if (retcode
!= NO_ERROR
)
2793 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2794 device
= adm_ctx
.device
;
2795 if (test_bit(NEW_CUR_UUID
, &device
->flags
)) {
2796 drbd_uuid_new_current(device
);
2797 clear_bit(NEW_CUR_UUID
, &device
->flags
);
2799 drbd_suspend_io(device
);
2800 retcode
= drbd_request_state(device
, NS3(susp
, 0, susp_nod
, 0, susp_fen
, 0));
2801 if (retcode
== SS_SUCCESS
) {
2802 if (device
->state
.conn
< C_CONNECTED
)
2803 tl_clear(first_peer_device(device
)->connection
);
2804 if (device
->state
.disk
== D_DISKLESS
|| device
->state
.disk
== D_FAILED
)
2805 tl_restart(first_peer_device(device
)->connection
, FAIL_FROZEN_DISK_IO
);
2807 drbd_resume_io(device
);
2808 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2810 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2814 int drbd_adm_outdate(struct sk_buff
*skb
, struct genl_info
*info
)
2816 return drbd_adm_simple_request_state(skb
, info
, NS(disk
, D_OUTDATED
));
2819 static int nla_put_drbd_cfg_context(struct sk_buff
*skb
,
2820 struct drbd_resource
*resource
,
2821 struct drbd_connection
*connection
,
2822 struct drbd_device
*device
)
2825 nla
= nla_nest_start(skb
, DRBD_NLA_CFG_CONTEXT
);
2827 goto nla_put_failure
;
2829 nla_put_u32(skb
, T_ctx_volume
, device
->vnr
))
2830 goto nla_put_failure
;
2831 if (nla_put_string(skb
, T_ctx_resource_name
, resource
->name
))
2832 goto nla_put_failure
;
2834 if (connection
->my_addr_len
&&
2835 nla_put(skb
, T_ctx_my_addr
, connection
->my_addr_len
, &connection
->my_addr
))
2836 goto nla_put_failure
;
2837 if (connection
->peer_addr_len
&&
2838 nla_put(skb
, T_ctx_peer_addr
, connection
->peer_addr_len
, &connection
->peer_addr
))
2839 goto nla_put_failure
;
2841 nla_nest_end(skb
, nla
);
2846 nla_nest_cancel(skb
, nla
);
2851 * Return the connection of @resource if @resource has exactly one connection.
2853 static struct drbd_connection
*the_only_connection(struct drbd_resource
*resource
)
2855 struct list_head
*connections
= &resource
->connections
;
2857 if (list_empty(connections
) || connections
->next
->next
!= connections
)
2859 return list_first_entry(&resource
->connections
, struct drbd_connection
, connections
);
2862 int nla_put_status_info(struct sk_buff
*skb
, struct drbd_device
*device
,
2863 const struct sib_info
*sib
)
2865 struct drbd_resource
*resource
= device
->resource
;
2866 struct state_info
*si
= NULL
; /* for sizeof(si->member); */
2870 int exclude_sensitive
;
2872 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2873 * to. So we better exclude_sensitive information.
2875 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2876 * in the context of the requesting user process. Exclude sensitive
2877 * information, unless current has superuser.
2879 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2880 * relies on the current implementation of netlink_dump(), which
2881 * executes the dump callback successively from netlink_recvmsg(),
2882 * always in the context of the receiving process */
2883 exclude_sensitive
= sib
|| !capable(CAP_SYS_ADMIN
);
2885 got_ldev
= get_ldev(device
);
2887 /* We need to add connection name and volume number information still.
2888 * Minor number is in drbd_genlmsghdr. */
2889 if (nla_put_drbd_cfg_context(skb
, resource
, the_only_connection(resource
), device
))
2890 goto nla_put_failure
;
2892 if (res_opts_to_skb(skb
, &device
->resource
->res_opts
, exclude_sensitive
))
2893 goto nla_put_failure
;
2897 struct disk_conf
*disk_conf
;
2899 disk_conf
= rcu_dereference(device
->ldev
->disk_conf
);
2900 err
= disk_conf_to_skb(skb
, disk_conf
, exclude_sensitive
);
2903 struct net_conf
*nc
;
2905 nc
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
);
2907 err
= net_conf_to_skb(skb
, nc
, exclude_sensitive
);
2911 goto nla_put_failure
;
2913 nla
= nla_nest_start(skb
, DRBD_NLA_STATE_INFO
);
2915 goto nla_put_failure
;
2916 if (nla_put_u32(skb
, T_sib_reason
, sib
? sib
->sib_reason
: SIB_GET_STATUS_REPLY
) ||
2917 nla_put_u32(skb
, T_current_state
, device
->state
.i
) ||
2918 nla_put_u64(skb
, T_ed_uuid
, device
->ed_uuid
) ||
2919 nla_put_u64(skb
, T_capacity
, drbd_get_capacity(device
->this_bdev
)) ||
2920 nla_put_u64(skb
, T_send_cnt
, device
->send_cnt
) ||
2921 nla_put_u64(skb
, T_recv_cnt
, device
->recv_cnt
) ||
2922 nla_put_u64(skb
, T_read_cnt
, device
->read_cnt
) ||
2923 nla_put_u64(skb
, T_writ_cnt
, device
->writ_cnt
) ||
2924 nla_put_u64(skb
, T_al_writ_cnt
, device
->al_writ_cnt
) ||
2925 nla_put_u64(skb
, T_bm_writ_cnt
, device
->bm_writ_cnt
) ||
2926 nla_put_u32(skb
, T_ap_bio_cnt
, atomic_read(&device
->ap_bio_cnt
)) ||
2927 nla_put_u32(skb
, T_ap_pending_cnt
, atomic_read(&device
->ap_pending_cnt
)) ||
2928 nla_put_u32(skb
, T_rs_pending_cnt
, atomic_read(&device
->rs_pending_cnt
)))
2929 goto nla_put_failure
;
2934 spin_lock_irq(&device
->ldev
->md
.uuid_lock
);
2935 err
= nla_put(skb
, T_uuids
, sizeof(si
->uuids
), device
->ldev
->md
.uuid
);
2936 spin_unlock_irq(&device
->ldev
->md
.uuid_lock
);
2939 goto nla_put_failure
;
2941 if (nla_put_u32(skb
, T_disk_flags
, device
->ldev
->md
.flags
) ||
2942 nla_put_u64(skb
, T_bits_total
, drbd_bm_bits(device
)) ||
2943 nla_put_u64(skb
, T_bits_oos
, drbd_bm_total_weight(device
)))
2944 goto nla_put_failure
;
2945 if (C_SYNC_SOURCE
<= device
->state
.conn
&&
2946 C_PAUSED_SYNC_T
>= device
->state
.conn
) {
2947 if (nla_put_u64(skb
, T_bits_rs_total
, device
->rs_total
) ||
2948 nla_put_u64(skb
, T_bits_rs_failed
, device
->rs_failed
))
2949 goto nla_put_failure
;
2954 switch(sib
->sib_reason
) {
2955 case SIB_SYNC_PROGRESS
:
2956 case SIB_GET_STATUS_REPLY
:
2958 case SIB_STATE_CHANGE
:
2959 if (nla_put_u32(skb
, T_prev_state
, sib
->os
.i
) ||
2960 nla_put_u32(skb
, T_new_state
, sib
->ns
.i
))
2961 goto nla_put_failure
;
2963 case SIB_HELPER_POST
:
2964 if (nla_put_u32(skb
, T_helper_exit_code
,
2965 sib
->helper_exit_code
))
2966 goto nla_put_failure
;
2968 case SIB_HELPER_PRE
:
2969 if (nla_put_string(skb
, T_helper
, sib
->helper_name
))
2970 goto nla_put_failure
;
2974 nla_nest_end(skb
, nla
);
2984 int drbd_adm_get_status(struct sk_buff
*skb
, struct genl_info
*info
)
2986 struct drbd_config_context adm_ctx
;
2987 enum drbd_ret_code retcode
;
2990 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2991 if (!adm_ctx
.reply_skb
)
2993 if (retcode
!= NO_ERROR
)
2996 err
= nla_put_status_info(adm_ctx
.reply_skb
, adm_ctx
.device
, NULL
);
2998 nlmsg_free(adm_ctx
.reply_skb
);
3002 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3006 static int get_one_status(struct sk_buff
*skb
, struct netlink_callback
*cb
)
3008 struct drbd_device
*device
;
3009 struct drbd_genlmsghdr
*dh
;
3010 struct drbd_resource
*pos
= (struct drbd_resource
*)cb
->args
[0];
3011 struct drbd_resource
*resource
= NULL
;
3012 struct drbd_resource
*tmp
;
3013 unsigned volume
= cb
->args
[1];
3015 /* Open coded, deferred, iteration:
3016 * for_each_resource_safe(resource, tmp, &drbd_resources) {
3017 * connection = "first connection of resource or undefined";
3018 * idr_for_each_entry(&resource->devices, device, i) {
3022 * where resource is cb->args[0];
3023 * and i is cb->args[1];
3025 * cb->args[2] indicates if we shall loop over all resources,
3026 * or just dump all volumes of a single resource.
3028 * This may miss entries inserted after this dump started,
3029 * or entries deleted before they are reached.
3031 * We need to make sure the device won't disappear while
3032 * we are looking at it, and revalidate our iterators
3033 * on each iteration.
3036 /* synchronize with conn_create()/drbd_destroy_connection() */
3038 /* revalidate iterator position */
3039 for_each_resource_rcu(tmp
, &drbd_resources
) {
3041 /* first iteration */
3053 device
= idr_get_next(&resource
->devices
, &volume
);
3055 /* No more volumes to dump on this resource.
3056 * Advance resource iterator. */
3057 pos
= list_entry_rcu(resource
->resources
.next
,
3058 struct drbd_resource
, resources
);
3059 /* Did we dump any volume of this resource yet? */
3061 /* If we reached the end of the list,
3062 * or only a single resource dump was requested,
3064 if (&pos
->resources
== &drbd_resources
|| cb
->args
[2])
3072 dh
= genlmsg_put(skb
, NETLINK_CB(cb
->skb
).portid
,
3073 cb
->nlh
->nlmsg_seq
, &drbd_genl_family
,
3074 NLM_F_MULTI
, DRBD_ADM_GET_STATUS
);
3079 /* This is a connection without a single volume.
3080 * Suprisingly enough, it may have a network
3082 struct drbd_connection
*connection
;
3085 dh
->ret_code
= NO_ERROR
;
3086 connection
= the_only_connection(resource
);
3087 if (nla_put_drbd_cfg_context(skb
, resource
, connection
, NULL
))
3090 struct net_conf
*nc
;
3092 nc
= rcu_dereference(connection
->net_conf
);
3093 if (nc
&& net_conf_to_skb(skb
, nc
, 1) != 0)
3099 D_ASSERT(device
, device
->vnr
== volume
);
3100 D_ASSERT(device
, device
->resource
== resource
);
3102 dh
->minor
= device_to_minor(device
);
3103 dh
->ret_code
= NO_ERROR
;
3105 if (nla_put_status_info(skb
, device
, NULL
)) {
3107 genlmsg_cancel(skb
, dh
);
3111 genlmsg_end(skb
, dh
);
3116 /* where to start the next iteration */
3117 cb
->args
[0] = (long)pos
;
3118 cb
->args
[1] = (pos
== resource
) ? volume
+ 1 : 0;
3120 /* No more resources/volumes/minors found results in an empty skb.
3121 * Which will terminate the dump. */
3126 * Request status of all resources, or of all volumes within a single resource.
3128 * This is a dump, as the answer may not fit in a single reply skb otherwise.
3129 * Which means we cannot use the family->attrbuf or other such members, because
3130 * dump is NOT protected by the genl_lock(). During dump, we only have access
3131 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
3133 * Once things are setup properly, we call into get_one_status().
3135 int drbd_adm_get_status_all(struct sk_buff
*skb
, struct netlink_callback
*cb
)
3137 const unsigned hdrlen
= GENL_HDRLEN
+ GENL_MAGIC_FAMILY_HDRSZ
;
3139 const char *resource_name
;
3140 struct drbd_resource
*resource
;
3143 /* Is this a followup call? */
3145 /* ... of a single resource dump,
3146 * and the resource iterator has been advanced already? */
3147 if (cb
->args
[2] && cb
->args
[2] != cb
->args
[0])
3148 return 0; /* DONE. */
3152 /* First call (from netlink_dump_start). We need to figure out
3153 * which resource(s) the user wants us to dump. */
3154 nla
= nla_find(nlmsg_attrdata(cb
->nlh
, hdrlen
),
3155 nlmsg_attrlen(cb
->nlh
, hdrlen
),
3156 DRBD_NLA_CFG_CONTEXT
);
3158 /* No explicit context given. Dump all. */
3161 maxtype
= ARRAY_SIZE(drbd_cfg_context_nl_policy
) - 1;
3162 nla
= drbd_nla_find_nested(maxtype
, nla
, __nla_type(T_ctx_resource_name
));
3164 return PTR_ERR(nla
);
3165 /* context given, but no name present? */
3168 resource_name
= nla_data(nla
);
3169 if (!*resource_name
)
3171 resource
= drbd_find_resource(resource_name
);
3175 kref_put(&resource
->kref
, drbd_destroy_resource
); /* get_one_status() revalidates the resource */
3177 /* prime iterators, and set "filter" mode mark:
3178 * only dump this connection. */
3179 cb
->args
[0] = (long)resource
;
3180 /* cb->args[1] = 0; passed in this way. */
3181 cb
->args
[2] = (long)resource
;
3184 return get_one_status(skb
, cb
);
3187 int drbd_adm_get_timeout_type(struct sk_buff
*skb
, struct genl_info
*info
)
3189 struct drbd_config_context adm_ctx
;
3190 enum drbd_ret_code retcode
;
3191 struct timeout_parms tp
;
3194 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3195 if (!adm_ctx
.reply_skb
)
3197 if (retcode
!= NO_ERROR
)
3201 adm_ctx
.device
->state
.pdsk
== D_OUTDATED
? UT_PEER_OUTDATED
:
3202 test_bit(USE_DEGR_WFC_T
, &adm_ctx
.device
->flags
) ? UT_DEGRADED
:
3205 err
= timeout_parms_to_priv_skb(adm_ctx
.reply_skb
, &tp
);
3207 nlmsg_free(adm_ctx
.reply_skb
);
3211 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3215 int drbd_adm_start_ov(struct sk_buff
*skb
, struct genl_info
*info
)
3217 struct drbd_config_context adm_ctx
;
3218 struct drbd_device
*device
;
3219 enum drbd_ret_code retcode
;
3220 struct start_ov_parms parms
;
3222 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3223 if (!adm_ctx
.reply_skb
)
3225 if (retcode
!= NO_ERROR
)
3228 device
= adm_ctx
.device
;
3230 /* resume from last known position, if possible */
3231 parms
.ov_start_sector
= device
->ov_start_sector
;
3232 parms
.ov_stop_sector
= ULLONG_MAX
;
3233 if (info
->attrs
[DRBD_NLA_START_OV_PARMS
]) {
3234 int err
= start_ov_parms_from_attrs(&parms
, info
);
3236 retcode
= ERR_MANDATORY_TAG
;
3237 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
3241 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
3243 /* w_make_ov_request expects position to be aligned */
3244 device
->ov_start_sector
= parms
.ov_start_sector
& ~(BM_SECT_PER_BIT
-1);
3245 device
->ov_stop_sector
= parms
.ov_stop_sector
;
3247 /* If there is still bitmap IO pending, e.g. previous resync or verify
3248 * just being finished, wait for it before requesting a new resync. */
3249 drbd_suspend_io(device
);
3250 wait_event(device
->misc_wait
, !test_bit(BITMAP_IO
, &device
->flags
));
3251 retcode
= drbd_request_state(device
, NS(conn
, C_VERIFY_S
));
3252 drbd_resume_io(device
);
3254 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
3256 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3261 int drbd_adm_new_c_uuid(struct sk_buff
*skb
, struct genl_info
*info
)
3263 struct drbd_config_context adm_ctx
;
3264 struct drbd_device
*device
;
3265 enum drbd_ret_code retcode
;
3266 int skip_initial_sync
= 0;
3268 struct new_c_uuid_parms args
;
3270 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3271 if (!adm_ctx
.reply_skb
)
3273 if (retcode
!= NO_ERROR
)
3276 device
= adm_ctx
.device
;
3277 memset(&args
, 0, sizeof(args
));
3278 if (info
->attrs
[DRBD_NLA_NEW_C_UUID_PARMS
]) {
3279 err
= new_c_uuid_parms_from_attrs(&args
, info
);
3281 retcode
= ERR_MANDATORY_TAG
;
3282 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
3287 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
3288 mutex_lock(device
->state_mutex
); /* Protects us against serialized state changes. */
3290 if (!get_ldev(device
)) {
3291 retcode
= ERR_NO_DISK
;
3295 /* this is "skip initial sync", assume to be clean */
3296 if (device
->state
.conn
== C_CONNECTED
&&
3297 first_peer_device(device
)->connection
->agreed_pro_version
>= 90 &&
3298 device
->ldev
->md
.uuid
[UI_CURRENT
] == UUID_JUST_CREATED
&& args
.clear_bm
) {
3299 drbd_info(device
, "Preparing to skip initial sync\n");
3300 skip_initial_sync
= 1;
3301 } else if (device
->state
.conn
!= C_STANDALONE
) {
3302 retcode
= ERR_CONNECTED
;
3306 drbd_uuid_set(device
, UI_BITMAP
, 0); /* Rotate UI_BITMAP to History 1, etc... */
3307 drbd_uuid_new_current(device
); /* New current, previous to UI_BITMAP */
3309 if (args
.clear_bm
) {
3310 err
= drbd_bitmap_io(device
, &drbd_bmio_clear_n_write
,
3311 "clear_n_write from new_c_uuid", BM_LOCKED_MASK
);
3313 drbd_err(device
, "Writing bitmap failed with %d\n", err
);
3314 retcode
= ERR_IO_MD_DISK
;
3316 if (skip_initial_sync
) {
3317 drbd_send_uuids_skip_initial_sync(first_peer_device(device
));
3318 _drbd_uuid_set(device
, UI_BITMAP
, 0);
3319 drbd_print_uuids(device
, "cleared bitmap UUID");
3320 spin_lock_irq(&device
->resource
->req_lock
);
3321 _drbd_set_state(_NS2(device
, disk
, D_UP_TO_DATE
, pdsk
, D_UP_TO_DATE
),
3323 spin_unlock_irq(&device
->resource
->req_lock
);
3327 drbd_md_sync(device
);
3331 mutex_unlock(device
->state_mutex
);
3332 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
3334 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3338 static enum drbd_ret_code
3339 drbd_check_resource_name(struct drbd_config_context
*adm_ctx
)
3341 const char *name
= adm_ctx
->resource_name
;
3342 if (!name
|| !name
[0]) {
3343 drbd_msg_put_info(adm_ctx
->reply_skb
, "resource name missing");
3344 return ERR_MANDATORY_TAG
;
3346 /* if we want to use these in sysfs/configfs/debugfs some day,
3347 * we must not allow slashes */
3348 if (strchr(name
, '/')) {
3349 drbd_msg_put_info(adm_ctx
->reply_skb
, "invalid resource name");
3350 return ERR_INVALID_REQUEST
;
3355 int drbd_adm_new_resource(struct sk_buff
*skb
, struct genl_info
*info
)
3357 struct drbd_config_context adm_ctx
;
3358 enum drbd_ret_code retcode
;
3359 struct res_opts res_opts
;
3362 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, 0);
3363 if (!adm_ctx
.reply_skb
)
3365 if (retcode
!= NO_ERROR
)
3368 set_res_opts_defaults(&res_opts
);
3369 err
= res_opts_from_attrs(&res_opts
, info
);
3370 if (err
&& err
!= -ENOMSG
) {
3371 retcode
= ERR_MANDATORY_TAG
;
3372 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
3376 retcode
= drbd_check_resource_name(&adm_ctx
);
3377 if (retcode
!= NO_ERROR
)
3380 if (adm_ctx
.resource
) {
3381 if (info
->nlhdr
->nlmsg_flags
& NLM_F_EXCL
) {
3382 retcode
= ERR_INVALID_REQUEST
;
3383 drbd_msg_put_info(adm_ctx
.reply_skb
, "resource exists");
3385 /* else: still NO_ERROR */
3389 /* not yet safe for genl_family.parallel_ops */
3390 if (!conn_create(adm_ctx
.resource_name
, &res_opts
))
3391 retcode
= ERR_NOMEM
;
3393 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3397 int drbd_adm_new_minor(struct sk_buff
*skb
, struct genl_info
*info
)
3399 struct drbd_config_context adm_ctx
;
3400 struct drbd_genlmsghdr
*dh
= info
->userhdr
;
3401 enum drbd_ret_code retcode
;
3403 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
3404 if (!adm_ctx
.reply_skb
)
3406 if (retcode
!= NO_ERROR
)
3409 if (dh
->minor
> MINORMASK
) {
3410 drbd_msg_put_info(adm_ctx
.reply_skb
, "requested minor out of range");
3411 retcode
= ERR_INVALID_REQUEST
;
3414 if (adm_ctx
.volume
> DRBD_VOLUME_MAX
) {
3415 drbd_msg_put_info(adm_ctx
.reply_skb
, "requested volume id out of range");
3416 retcode
= ERR_INVALID_REQUEST
;
3420 /* drbd_adm_prepare made sure already
3421 * that first_peer_device(device)->connection and device->vnr match the request. */
3422 if (adm_ctx
.device
) {
3423 if (info
->nlhdr
->nlmsg_flags
& NLM_F_EXCL
)
3424 retcode
= ERR_MINOR_EXISTS
;
3425 /* else: still NO_ERROR */
3429 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
3430 retcode
= drbd_create_device(&adm_ctx
, dh
->minor
);
3431 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
3433 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3437 static enum drbd_ret_code
adm_del_minor(struct drbd_device
*device
)
3439 if (device
->state
.disk
== D_DISKLESS
&&
3440 /* no need to be device->state.conn == C_STANDALONE &&
3441 * we may want to delete a minor from a live replication group.
3443 device
->state
.role
== R_SECONDARY
) {
3444 _drbd_request_state(device
, NS(conn
, C_WF_REPORT_PARAMS
),
3445 CS_VERBOSE
+ CS_WAIT_COMPLETE
);
3446 drbd_delete_device(device
);
3449 return ERR_MINOR_CONFIGURED
;
3452 int drbd_adm_del_minor(struct sk_buff
*skb
, struct genl_info
*info
)
3454 struct drbd_config_context adm_ctx
;
3455 enum drbd_ret_code retcode
;
3457 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3458 if (!adm_ctx
.reply_skb
)
3460 if (retcode
!= NO_ERROR
)
3463 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
3464 retcode
= adm_del_minor(adm_ctx
.device
);
3465 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
3467 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3471 int drbd_adm_down(struct sk_buff
*skb
, struct genl_info
*info
)
3473 struct drbd_config_context adm_ctx
;
3474 struct drbd_resource
*resource
;
3475 struct drbd_connection
*connection
;
3476 struct drbd_device
*device
;
3477 int retcode
; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3480 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
3481 if (!adm_ctx
.reply_skb
)
3483 if (retcode
!= NO_ERROR
)
3486 resource
= adm_ctx
.resource
;
3487 mutex_lock(&resource
->adm_mutex
);
3489 for_each_connection(connection
, resource
) {
3490 struct drbd_peer_device
*peer_device
;
3492 idr_for_each_entry(&connection
->peer_devices
, peer_device
, i
) {
3493 retcode
= drbd_set_role(peer_device
->device
, R_SECONDARY
, 0);
3494 if (retcode
< SS_SUCCESS
) {
3495 drbd_msg_put_info(adm_ctx
.reply_skb
, "failed to demote");
3500 retcode
= conn_try_disconnect(connection
, 0);
3501 if (retcode
< SS_SUCCESS
) {
3502 drbd_msg_put_info(adm_ctx
.reply_skb
, "failed to disconnect");
3508 idr_for_each_entry(&resource
->devices
, device
, i
) {
3509 retcode
= adm_detach(device
, 0);
3510 if (retcode
< SS_SUCCESS
|| retcode
> NO_ERROR
) {
3511 drbd_msg_put_info(adm_ctx
.reply_skb
, "failed to detach");
3516 /* If we reach this, all volumes (of this connection) are Secondary,
3517 * Disconnected, Diskless, aka Unconfigured. Make sure all threads have
3518 * actually stopped, state handling only does drbd_thread_stop_nowait(). */
3519 for_each_connection(connection
, resource
)
3520 drbd_thread_stop(&connection
->worker
);
3522 /* Now, nothing can fail anymore */
3524 /* delete volumes */
3525 idr_for_each_entry(&resource
->devices
, device
, i
) {
3526 retcode
= adm_del_minor(device
);
3527 if (retcode
!= NO_ERROR
) {
3528 /* "can not happen" */
3529 drbd_msg_put_info(adm_ctx
.reply_skb
, "failed to delete volume");
3534 list_del_rcu(&resource
->resources
);
3536 drbd_free_resource(resource
);
3539 mutex_unlock(&resource
->adm_mutex
);
3541 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3545 int drbd_adm_del_resource(struct sk_buff
*skb
, struct genl_info
*info
)
3547 struct drbd_config_context adm_ctx
;
3548 struct drbd_resource
*resource
;
3549 struct drbd_connection
*connection
;
3550 enum drbd_ret_code retcode
;
3552 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
3553 if (!adm_ctx
.reply_skb
)
3555 if (retcode
!= NO_ERROR
)
3558 resource
= adm_ctx
.resource
;
3559 mutex_lock(&resource
->adm_mutex
);
3560 for_each_connection(connection
, resource
) {
3561 if (connection
->cstate
> C_STANDALONE
) {
3562 retcode
= ERR_NET_CONFIGURED
;
3566 if (!idr_is_empty(&resource
->devices
)) {
3567 retcode
= ERR_RES_IN_USE
;
3571 list_del_rcu(&resource
->resources
);
3572 for_each_connection(connection
, resource
)
3573 drbd_thread_stop(&connection
->worker
);
3575 drbd_free_resource(resource
);
3578 mutex_unlock(&resource
->adm_mutex
);
3580 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3584 void drbd_bcast_event(struct drbd_device
*device
, const struct sib_info
*sib
)
3586 static atomic_t drbd_genl_seq
= ATOMIC_INIT(2); /* two. */
3587 struct sk_buff
*msg
;
3588 struct drbd_genlmsghdr
*d_out
;
3592 if (sib
->sib_reason
== SIB_SYNC_PROGRESS
) {
3593 if (time_after(jiffies
, device
->rs_last_bcast
+ HZ
))
3594 device
->rs_last_bcast
= jiffies
;
3599 seq
= atomic_inc_return(&drbd_genl_seq
);
3600 msg
= genlmsg_new(NLMSG_GOODSIZE
, GFP_NOIO
);
3605 d_out
= genlmsg_put(msg
, 0, seq
, &drbd_genl_family
, 0, DRBD_EVENT
);
3606 if (!d_out
) /* cannot happen, but anyways. */
3607 goto nla_put_failure
;
3608 d_out
->minor
= device_to_minor(device
);
3609 d_out
->ret_code
= NO_ERROR
;
3611 if (nla_put_status_info(msg
, device
, sib
))
3612 goto nla_put_failure
;
3613 genlmsg_end(msg
, d_out
);
3614 err
= drbd_genl_multicast_events(msg
, 0);
3615 /* msg has been consumed or freed in netlink_broadcast() */
3616 if (err
&& err
!= -ESRCH
)
3624 drbd_err(device
, "Error %d while broadcasting event. "
3625 "Event seq:%u sib_reason:%u\n",
3626 err
, seq
, sib
->sib_reason
);