4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
6 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
7 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
8 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
10 drbd is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation; either version 2, or (at your option)
15 drbd is distributed in the hope that it will be useful,
16 but WITHOUT ANY WARRANTY; without even the implied warranty of
17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 GNU General Public License for more details.
20 You should have received a copy of the GNU General Public License
21 along with drbd; see the file COPYING. If not, write to
22 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
26 #include <linux/module.h>
27 #include <linux/drbd.h>
30 #include <linux/file.h>
31 #include <linux/slab.h>
32 #include <linux/blkpg.h>
33 #include <linux/cpumask.h>
35 #include "drbd_protocol.h"
37 #include <asm/unaligned.h>
38 #include <linux/drbd_limits.h>
39 #include <linux/kthread.h>
41 #include <net/genetlink.h>
44 // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info);
45 // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info);
47 int drbd_adm_new_minor(struct sk_buff
*skb
, struct genl_info
*info
);
48 int drbd_adm_del_minor(struct sk_buff
*skb
, struct genl_info
*info
);
50 int drbd_adm_new_resource(struct sk_buff
*skb
, struct genl_info
*info
);
51 int drbd_adm_del_resource(struct sk_buff
*skb
, struct genl_info
*info
);
52 int drbd_adm_down(struct sk_buff
*skb
, struct genl_info
*info
);
54 int drbd_adm_set_role(struct sk_buff
*skb
, struct genl_info
*info
);
55 int drbd_adm_attach(struct sk_buff
*skb
, struct genl_info
*info
);
56 int drbd_adm_disk_opts(struct sk_buff
*skb
, struct genl_info
*info
);
57 int drbd_adm_detach(struct sk_buff
*skb
, struct genl_info
*info
);
58 int drbd_adm_connect(struct sk_buff
*skb
, struct genl_info
*info
);
59 int drbd_adm_net_opts(struct sk_buff
*skb
, struct genl_info
*info
);
60 int drbd_adm_resize(struct sk_buff
*skb
, struct genl_info
*info
);
61 int drbd_adm_start_ov(struct sk_buff
*skb
, struct genl_info
*info
);
62 int drbd_adm_new_c_uuid(struct sk_buff
*skb
, struct genl_info
*info
);
63 int drbd_adm_disconnect(struct sk_buff
*skb
, struct genl_info
*info
);
64 int drbd_adm_invalidate(struct sk_buff
*skb
, struct genl_info
*info
);
65 int drbd_adm_invalidate_peer(struct sk_buff
*skb
, struct genl_info
*info
);
66 int drbd_adm_pause_sync(struct sk_buff
*skb
, struct genl_info
*info
);
67 int drbd_adm_resume_sync(struct sk_buff
*skb
, struct genl_info
*info
);
68 int drbd_adm_suspend_io(struct sk_buff
*skb
, struct genl_info
*info
);
69 int drbd_adm_resume_io(struct sk_buff
*skb
, struct genl_info
*info
);
70 int drbd_adm_outdate(struct sk_buff
*skb
, struct genl_info
*info
);
71 int drbd_adm_resource_opts(struct sk_buff
*skb
, struct genl_info
*info
);
72 int drbd_adm_get_status(struct sk_buff
*skb
, struct genl_info
*info
);
73 int drbd_adm_get_timeout_type(struct sk_buff
*skb
, struct genl_info
*info
);
75 int drbd_adm_get_status_all(struct sk_buff
*skb
, struct netlink_callback
*cb
);
77 #include <linux/drbd_genl_api.h>
79 #include <linux/genl_magic_func.h>
81 /* used blkdev_get_by_path, to claim our meta data device(s) */
82 static char *drbd_m_holder
= "Hands off! this is DRBD's meta data device.";
84 static void drbd_adm_send_reply(struct sk_buff
*skb
, struct genl_info
*info
)
86 genlmsg_end(skb
, genlmsg_data(nlmsg_data(nlmsg_hdr(skb
))));
87 if (genlmsg_reply(skb
, info
))
88 printk(KERN_ERR
"drbd: error sending genl reply\n");
91 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
92 * reason it could fail was no space in skb, and there are 4k available. */
93 int drbd_msg_put_info(struct sk_buff
*skb
, const char *info
)
98 if (!info
|| !info
[0])
101 nla
= nla_nest_start(skb
, DRBD_NLA_CFG_REPLY
);
105 err
= nla_put_string(skb
, T_info_text
, info
);
107 nla_nest_cancel(skb
, nla
);
110 nla_nest_end(skb
, nla
);
114 /* This would be a good candidate for a "pre_doit" hook,
115 * and per-family private info->pointers.
116 * But we need to stay compatible with older kernels.
117 * If it returns successfully, adm_ctx members are valid.
119 * At this point, we still rely on the global genl_lock().
120 * If we want to avoid that, and allow "genl_family.parallel_ops", we may need
121 * to add additional synchronization against object destruction/modification.
123 #define DRBD_ADM_NEED_MINOR 1
124 #define DRBD_ADM_NEED_RESOURCE 2
125 #define DRBD_ADM_NEED_CONNECTION 4
126 static int drbd_adm_prepare(struct drbd_config_context
*adm_ctx
,
127 struct sk_buff
*skb
, struct genl_info
*info
, unsigned flags
)
129 struct drbd_genlmsghdr
*d_in
= info
->userhdr
;
130 const u8 cmd
= info
->genlhdr
->cmd
;
133 memset(adm_ctx
, 0, sizeof(*adm_ctx
));
135 /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
136 if (cmd
!= DRBD_ADM_GET_STATUS
&& !capable(CAP_NET_ADMIN
))
139 adm_ctx
->reply_skb
= genlmsg_new(NLMSG_GOODSIZE
, GFP_KERNEL
);
140 if (!adm_ctx
->reply_skb
) {
145 adm_ctx
->reply_dh
= genlmsg_put_reply(adm_ctx
->reply_skb
,
146 info
, &drbd_genl_family
, 0, cmd
);
147 /* put of a few bytes into a fresh skb of >= 4k will always succeed.
149 if (!adm_ctx
->reply_dh
) {
154 adm_ctx
->reply_dh
->minor
= d_in
->minor
;
155 adm_ctx
->reply_dh
->ret_code
= NO_ERROR
;
157 adm_ctx
->volume
= VOLUME_UNSPECIFIED
;
158 if (info
->attrs
[DRBD_NLA_CFG_CONTEXT
]) {
160 /* parse and validate only */
161 err
= drbd_cfg_context_from_attrs(NULL
, info
);
165 /* It was present, and valid,
166 * copy it over to the reply skb. */
167 err
= nla_put_nohdr(adm_ctx
->reply_skb
,
168 info
->attrs
[DRBD_NLA_CFG_CONTEXT
]->nla_len
,
169 info
->attrs
[DRBD_NLA_CFG_CONTEXT
]);
173 /* and assign stuff to the adm_ctx */
174 nla
= nested_attr_tb
[__nla_type(T_ctx_volume
)];
176 adm_ctx
->volume
= nla_get_u32(nla
);
177 nla
= nested_attr_tb
[__nla_type(T_ctx_resource_name
)];
179 adm_ctx
->resource_name
= nla_data(nla
);
180 adm_ctx
->my_addr
= nested_attr_tb
[__nla_type(T_ctx_my_addr
)];
181 adm_ctx
->peer_addr
= nested_attr_tb
[__nla_type(T_ctx_peer_addr
)];
182 if ((adm_ctx
->my_addr
&&
183 nla_len(adm_ctx
->my_addr
) > sizeof(adm_ctx
->connection
->my_addr
)) ||
184 (adm_ctx
->peer_addr
&&
185 nla_len(adm_ctx
->peer_addr
) > sizeof(adm_ctx
->connection
->peer_addr
))) {
191 adm_ctx
->minor
= d_in
->minor
;
192 adm_ctx
->device
= minor_to_device(d_in
->minor
);
194 /* We are protected by the global genl_lock().
195 * But we may explicitly drop it/retake it in drbd_adm_set_role(),
196 * so make sure this object stays around. */
198 kref_get(&adm_ctx
->device
->kref
);
200 if (adm_ctx
->resource_name
) {
201 adm_ctx
->resource
= drbd_find_resource(adm_ctx
->resource_name
);
204 if (!adm_ctx
->device
&& (flags
& DRBD_ADM_NEED_MINOR
)) {
205 drbd_msg_put_info(adm_ctx
->reply_skb
, "unknown minor");
206 return ERR_MINOR_INVALID
;
208 if (!adm_ctx
->resource
&& (flags
& DRBD_ADM_NEED_RESOURCE
)) {
209 drbd_msg_put_info(adm_ctx
->reply_skb
, "unknown resource");
210 if (adm_ctx
->resource_name
)
211 return ERR_RES_NOT_KNOWN
;
212 return ERR_INVALID_REQUEST
;
215 if (flags
& DRBD_ADM_NEED_CONNECTION
) {
216 if (adm_ctx
->resource
) {
217 drbd_msg_put_info(adm_ctx
->reply_skb
, "no resource name expected");
218 return ERR_INVALID_REQUEST
;
220 if (adm_ctx
->device
) {
221 drbd_msg_put_info(adm_ctx
->reply_skb
, "no minor number expected");
222 return ERR_INVALID_REQUEST
;
224 if (adm_ctx
->my_addr
&& adm_ctx
->peer_addr
)
225 adm_ctx
->connection
= conn_get_by_addrs(nla_data(adm_ctx
->my_addr
),
226 nla_len(adm_ctx
->my_addr
),
227 nla_data(adm_ctx
->peer_addr
),
228 nla_len(adm_ctx
->peer_addr
));
229 if (!adm_ctx
->connection
) {
230 drbd_msg_put_info(adm_ctx
->reply_skb
, "unknown connection");
231 return ERR_INVALID_REQUEST
;
235 /* some more paranoia, if the request was over-determined */
236 if (adm_ctx
->device
&& adm_ctx
->resource
&&
237 adm_ctx
->device
->resource
!= adm_ctx
->resource
) {
238 pr_warning("request: minor=%u, resource=%s; but that minor belongs to resource %s\n",
239 adm_ctx
->minor
, adm_ctx
->resource
->name
,
240 adm_ctx
->device
->resource
->name
);
241 drbd_msg_put_info(adm_ctx
->reply_skb
, "minor exists in different resource");
242 return ERR_INVALID_REQUEST
;
244 if (adm_ctx
->device
&&
245 adm_ctx
->volume
!= VOLUME_UNSPECIFIED
&&
246 adm_ctx
->volume
!= adm_ctx
->device
->vnr
) {
247 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
248 adm_ctx
->minor
, adm_ctx
->volume
,
249 adm_ctx
->device
->vnr
,
250 adm_ctx
->device
->resource
->name
);
251 drbd_msg_put_info(adm_ctx
->reply_skb
, "minor exists as different volume");
252 return ERR_INVALID_REQUEST
;
255 /* still, provide adm_ctx->resource always, if possible. */
256 if (!adm_ctx
->resource
) {
257 adm_ctx
->resource
= adm_ctx
->device
? adm_ctx
->device
->resource
258 : adm_ctx
->connection
? adm_ctx
->connection
->resource
: NULL
;
259 if (adm_ctx
->resource
)
260 kref_get(&adm_ctx
->resource
->kref
);
266 nlmsg_free(adm_ctx
->reply_skb
);
267 adm_ctx
->reply_skb
= NULL
;
271 static int drbd_adm_finish(struct drbd_config_context
*adm_ctx
,
272 struct genl_info
*info
, int retcode
)
274 if (adm_ctx
->device
) {
275 kref_put(&adm_ctx
->device
->kref
, drbd_destroy_device
);
276 adm_ctx
->device
= NULL
;
278 if (adm_ctx
->connection
) {
279 kref_put(&adm_ctx
->connection
->kref
, &drbd_destroy_connection
);
280 adm_ctx
->connection
= NULL
;
282 if (adm_ctx
->resource
) {
283 kref_put(&adm_ctx
->resource
->kref
, drbd_destroy_resource
);
284 adm_ctx
->resource
= NULL
;
287 if (!adm_ctx
->reply_skb
)
290 adm_ctx
->reply_dh
->ret_code
= retcode
;
291 drbd_adm_send_reply(adm_ctx
->reply_skb
, info
);
295 static void setup_khelper_env(struct drbd_connection
*connection
, char **envp
)
299 /* FIXME: A future version will not allow this case. */
300 if (connection
->my_addr_len
== 0 || connection
->peer_addr_len
== 0)
303 switch (((struct sockaddr
*)&connection
->peer_addr
)->sa_family
) {
306 snprintf(envp
[4], 60, "DRBD_PEER_ADDRESS=%pI6",
307 &((struct sockaddr_in6
*)&connection
->peer_addr
)->sin6_addr
);
311 snprintf(envp
[4], 60, "DRBD_PEER_ADDRESS=%pI4",
312 &((struct sockaddr_in
*)&connection
->peer_addr
)->sin_addr
);
316 snprintf(envp
[4], 60, "DRBD_PEER_ADDRESS=%pI4",
317 &((struct sockaddr_in
*)&connection
->peer_addr
)->sin_addr
);
319 snprintf(envp
[3], 20, "DRBD_PEER_AF=%s", afs
);
322 int drbd_khelper(struct drbd_device
*device
, char *cmd
)
324 char *envp
[] = { "HOME=/",
326 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
327 (char[20]) { }, /* address family */
328 (char[60]) { }, /* address */
331 char *argv
[] = {usermode_helper
, cmd
, mb
, NULL
};
332 struct drbd_connection
*connection
= first_peer_device(device
)->connection
;
336 if (current
== connection
->worker
.task
)
337 set_bit(CALLBACK_PENDING
, &connection
->flags
);
339 snprintf(mb
, 12, "minor-%d", device_to_minor(device
));
340 setup_khelper_env(connection
, envp
);
342 /* The helper may take some time.
343 * write out any unsynced meta data changes now */
344 drbd_md_sync(device
);
346 drbd_info(device
, "helper command: %s %s %s\n", usermode_helper
, cmd
, mb
);
347 sib
.sib_reason
= SIB_HELPER_PRE
;
348 sib
.helper_name
= cmd
;
349 drbd_bcast_event(device
, &sib
);
350 ret
= call_usermodehelper(usermode_helper
, argv
, envp
, UMH_WAIT_PROC
);
352 drbd_warn(device
, "helper command: %s %s %s exit code %u (0x%x)\n",
353 usermode_helper
, cmd
, mb
,
354 (ret
>> 8) & 0xff, ret
);
356 drbd_info(device
, "helper command: %s %s %s exit code %u (0x%x)\n",
357 usermode_helper
, cmd
, mb
,
358 (ret
>> 8) & 0xff, ret
);
359 sib
.sib_reason
= SIB_HELPER_POST
;
360 sib
.helper_exit_code
= ret
;
361 drbd_bcast_event(device
, &sib
);
363 if (current
== connection
->worker
.task
)
364 clear_bit(CALLBACK_PENDING
, &connection
->flags
);
366 if (ret
< 0) /* Ignore any ERRNOs we got. */
372 static int conn_khelper(struct drbd_connection
*connection
, char *cmd
)
374 char *envp
[] = { "HOME=/",
376 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
377 (char[20]) { }, /* address family */
378 (char[60]) { }, /* address */
380 char *resource_name
= connection
->resource
->name
;
381 char *argv
[] = {usermode_helper
, cmd
, resource_name
, NULL
};
384 setup_khelper_env(connection
, envp
);
385 conn_md_sync(connection
);
387 drbd_info(connection
, "helper command: %s %s %s\n", usermode_helper
, cmd
, resource_name
);
388 /* TODO: conn_bcast_event() ?? */
390 ret
= call_usermodehelper(usermode_helper
, argv
, envp
, UMH_WAIT_PROC
);
392 drbd_warn(connection
, "helper command: %s %s %s exit code %u (0x%x)\n",
393 usermode_helper
, cmd
, resource_name
,
394 (ret
>> 8) & 0xff, ret
);
396 drbd_info(connection
, "helper command: %s %s %s exit code %u (0x%x)\n",
397 usermode_helper
, cmd
, resource_name
,
398 (ret
>> 8) & 0xff, ret
);
399 /* TODO: conn_bcast_event() ?? */
401 if (ret
< 0) /* Ignore any ERRNOs we got. */
407 static enum drbd_fencing_p
highest_fencing_policy(struct drbd_connection
*connection
)
409 enum drbd_fencing_p fp
= FP_NOT_AVAIL
;
410 struct drbd_peer_device
*peer_device
;
414 idr_for_each_entry(&connection
->peer_devices
, peer_device
, vnr
) {
415 struct drbd_device
*device
= peer_device
->device
;
416 if (get_ldev_if_state(device
, D_CONSISTENT
)) {
417 struct disk_conf
*disk_conf
=
418 rcu_dereference(peer_device
->device
->ldev
->disk_conf
);
419 fp
= max_t(enum drbd_fencing_p
, fp
, disk_conf
->fencing
);
425 if (fp
== FP_NOT_AVAIL
) {
426 /* IO Suspending works on the whole resource.
427 Do it only for one device. */
429 peer_device
= idr_get_next(&connection
->peer_devices
, &vnr
);
430 drbd_change_state(peer_device
->device
, CS_VERBOSE
| CS_HARD
, NS(susp_fen
, 0));
436 bool conn_try_outdate_peer(struct drbd_connection
*connection
)
438 unsigned int connect_cnt
;
439 union drbd_state mask
= { };
440 union drbd_state val
= { };
441 enum drbd_fencing_p fp
;
445 spin_lock_irq(&connection
->resource
->req_lock
);
446 if (connection
->cstate
>= C_WF_REPORT_PARAMS
) {
447 drbd_err(connection
, "Expected cstate < C_WF_REPORT_PARAMS\n");
448 spin_unlock_irq(&connection
->resource
->req_lock
);
452 connect_cnt
= connection
->connect_cnt
;
453 spin_unlock_irq(&connection
->resource
->req_lock
);
455 fp
= highest_fencing_policy(connection
);
458 drbd_warn(connection
, "Not fencing peer, I'm not even Consistent myself.\n");
465 r
= conn_khelper(connection
, "fence-peer");
467 switch ((r
>>8) & 0xff) {
468 case 3: /* peer is inconsistent */
469 ex_to_string
= "peer is inconsistent or worse";
471 val
.pdsk
= D_INCONSISTENT
;
473 case 4: /* peer got outdated, or was already outdated */
474 ex_to_string
= "peer was fenced";
476 val
.pdsk
= D_OUTDATED
;
478 case 5: /* peer was down */
479 if (conn_highest_disk(connection
) == D_UP_TO_DATE
) {
480 /* we will(have) create(d) a new UUID anyways... */
481 ex_to_string
= "peer is unreachable, assumed to be dead";
483 val
.pdsk
= D_OUTDATED
;
485 ex_to_string
= "peer unreachable, doing nothing since disk != UpToDate";
488 case 6: /* Peer is primary, voluntarily outdate myself.
489 * This is useful when an unconnected R_SECONDARY is asked to
490 * become R_PRIMARY, but finds the other peer being active. */
491 ex_to_string
= "peer is active";
492 drbd_warn(connection
, "Peer is primary, outdating myself.\n");
494 val
.disk
= D_OUTDATED
;
497 if (fp
!= FP_STONITH
)
498 drbd_err(connection
, "fence-peer() = 7 && fencing != Stonith !!!\n");
499 ex_to_string
= "peer was stonithed";
501 val
.pdsk
= D_OUTDATED
;
504 /* The script is broken ... */
505 drbd_err(connection
, "fence-peer helper broken, returned %d\n", (r
>>8)&0xff);
506 return false; /* Eventually leave IO frozen */
509 drbd_info(connection
, "fence-peer helper returned %d (%s)\n",
510 (r
>>8) & 0xff, ex_to_string
);
515 conn_request_state(connection, mask, val, CS_VERBOSE);
516 here, because we might were able to re-establish the connection in the
518 spin_lock_irq(&connection
->resource
->req_lock
);
519 if (connection
->cstate
< C_WF_REPORT_PARAMS
&& !test_bit(STATE_SENT
, &connection
->flags
)) {
520 if (connection
->connect_cnt
!= connect_cnt
)
521 /* In case the connection was established and droped
522 while the fence-peer handler was running, ignore it */
523 drbd_info(connection
, "Ignoring fence-peer exit code\n");
525 _conn_request_state(connection
, mask
, val
, CS_VERBOSE
);
527 spin_unlock_irq(&connection
->resource
->req_lock
);
529 return conn_highest_pdsk(connection
) <= D_OUTDATED
;
532 static int _try_outdate_peer_async(void *data
)
534 struct drbd_connection
*connection
= (struct drbd_connection
*)data
;
536 conn_try_outdate_peer(connection
);
538 kref_put(&connection
->kref
, drbd_destroy_connection
);
542 void conn_try_outdate_peer_async(struct drbd_connection
*connection
)
544 struct task_struct
*opa
;
546 kref_get(&connection
->kref
);
547 /* We may just have force_sig()'ed this thread
548 * to get it out of some blocking network function.
549 * Clear signals; otherwise kthread_run(), which internally uses
550 * wait_on_completion_killable(), will mistake our pending signal
551 * for a new fatal signal and fail. */
552 flush_signals(current
);
553 opa
= kthread_run(_try_outdate_peer_async
, connection
, "drbd_async_h");
555 drbd_err(connection
, "out of mem, failed to invoke fence-peer helper\n");
556 kref_put(&connection
->kref
, drbd_destroy_connection
);
561 drbd_set_role(struct drbd_device
*device
, enum drbd_role new_role
, int force
)
563 const int max_tries
= 4;
564 enum drbd_state_rv rv
= SS_UNKNOWN_ERROR
;
568 union drbd_state mask
, val
;
570 if (new_role
== R_PRIMARY
) {
571 struct drbd_connection
*connection
;
573 /* Detect dead peers as soon as possible. */
576 for_each_connection(connection
, device
->resource
)
577 request_ping(connection
);
581 mutex_lock(device
->state_mutex
);
583 mask
.i
= 0; mask
.role
= R_MASK
;
584 val
.i
= 0; val
.role
= new_role
;
586 while (try++ < max_tries
) {
587 rv
= _drbd_request_state(device
, mask
, val
, CS_WAIT_COMPLETE
);
589 /* in case we first succeeded to outdate,
590 * but now suddenly could establish a connection */
591 if (rv
== SS_CW_FAILED_BY_PEER
&& mask
.pdsk
!= 0) {
597 if (rv
== SS_NO_UP_TO_DATE_DISK
&& force
&&
598 (device
->state
.disk
< D_UP_TO_DATE
&&
599 device
->state
.disk
>= D_INCONSISTENT
)) {
601 val
.disk
= D_UP_TO_DATE
;
606 if (rv
== SS_NO_UP_TO_DATE_DISK
&&
607 device
->state
.disk
== D_CONSISTENT
&& mask
.pdsk
== 0) {
608 D_ASSERT(device
, device
->state
.pdsk
== D_UNKNOWN
);
610 if (conn_try_outdate_peer(first_peer_device(device
)->connection
)) {
611 val
.disk
= D_UP_TO_DATE
;
617 if (rv
== SS_NOTHING_TO_DO
)
619 if (rv
== SS_PRIMARY_NOP
&& mask
.pdsk
== 0) {
620 if (!conn_try_outdate_peer(first_peer_device(device
)->connection
) && force
) {
621 drbd_warn(device
, "Forced into split brain situation!\n");
623 val
.pdsk
= D_OUTDATED
;
628 if (rv
== SS_TWO_PRIMARIES
) {
629 /* Maybe the peer is detected as dead very soon...
630 retry at most once more in this case. */
633 nc
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
);
634 timeo
= nc
? (nc
->ping_timeo
+ 1) * HZ
/ 10 : 1;
636 schedule_timeout_interruptible(timeo
);
641 if (rv
< SS_SUCCESS
) {
642 rv
= _drbd_request_state(device
, mask
, val
,
643 CS_VERBOSE
+ CS_WAIT_COMPLETE
);
654 drbd_warn(device
, "Forced to consider local data as UpToDate!\n");
656 /* Wait until nothing is on the fly :) */
657 wait_event(device
->misc_wait
, atomic_read(&device
->ap_pending_cnt
) == 0);
659 /* FIXME also wait for all pending P_BARRIER_ACK? */
661 if (new_role
== R_SECONDARY
) {
662 set_disk_ro(device
->vdisk
, true);
663 if (get_ldev(device
)) {
664 device
->ldev
->md
.uuid
[UI_CURRENT
] &= ~(u64
)1;
668 /* Called from drbd_adm_set_role only.
669 * We are still holding the conf_update mutex. */
670 nc
= first_peer_device(device
)->connection
->net_conf
;
672 nc
->discard_my_data
= 0; /* without copy; single bit op is atomic */
674 set_disk_ro(device
->vdisk
, false);
675 if (get_ldev(device
)) {
676 if (((device
->state
.conn
< C_CONNECTED
||
677 device
->state
.pdsk
<= D_FAILED
)
678 && device
->ldev
->md
.uuid
[UI_BITMAP
] == 0) || forced
)
679 drbd_uuid_new_current(device
);
681 device
->ldev
->md
.uuid
[UI_CURRENT
] |= (u64
)1;
686 /* writeout of activity log covered areas of the bitmap
687 * to stable storage done in after state change already */
689 if (device
->state
.conn
>= C_WF_REPORT_PARAMS
) {
690 /* if this was forced, we should consider sync */
692 drbd_send_uuids(first_peer_device(device
));
693 drbd_send_current_state(first_peer_device(device
));
696 drbd_md_sync(device
);
698 kobject_uevent(&disk_to_dev(device
->vdisk
)->kobj
, KOBJ_CHANGE
);
700 mutex_unlock(device
->state_mutex
);
704 static const char *from_attrs_err_to_txt(int err
)
706 return err
== -ENOMSG
? "required attribute missing" :
707 err
== -EOPNOTSUPP
? "unknown mandatory attribute" :
708 err
== -EEXIST
? "can not change invariant setting" :
709 "invalid attribute value";
712 int drbd_adm_set_role(struct sk_buff
*skb
, struct genl_info
*info
)
714 struct drbd_config_context adm_ctx
;
715 struct set_role_parms parms
;
717 enum drbd_ret_code retcode
;
719 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
720 if (!adm_ctx
.reply_skb
)
722 if (retcode
!= NO_ERROR
)
725 memset(&parms
, 0, sizeof(parms
));
726 if (info
->attrs
[DRBD_NLA_SET_ROLE_PARMS
]) {
727 err
= set_role_parms_from_attrs(&parms
, info
);
729 retcode
= ERR_MANDATORY_TAG
;
730 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
735 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
737 if (info
->genlhdr
->cmd
== DRBD_ADM_PRIMARY
)
738 retcode
= drbd_set_role(adm_ctx
.device
, R_PRIMARY
, parms
.assume_uptodate
);
740 retcode
= drbd_set_role(adm_ctx
.device
, R_SECONDARY
, 0);
742 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
745 drbd_adm_finish(&adm_ctx
, info
, retcode
);
749 /* Initializes the md.*_offset members, so we are able to find
750 * the on disk meta data.
752 * We currently have two possible layouts:
754 * |----------- md_size_sect ------------------|
755 * [ 4k superblock ][ activity log ][ Bitmap ]
757 * | bm_offset = al_offset + X |
758 * ==> bitmap sectors = md_size_sect - bm_offset
761 * |----------- md_size_sect ------------------|
762 * [data.....][ Bitmap ][ activity log ][ 4k superblock ]
764 * | bm_offset = al_offset - Y |
765 * ==> bitmap sectors = Y = al_offset - bm_offset
767 * Activity log size used to be fixed 32kB,
768 * but is about to become configurable.
770 static void drbd_md_set_sector_offsets(struct drbd_device
*device
,
771 struct drbd_backing_dev
*bdev
)
773 sector_t md_size_sect
= 0;
774 unsigned int al_size_sect
= bdev
->md
.al_size_4k
* 8;
776 bdev
->md
.md_offset
= drbd_md_ss(bdev
);
778 switch (bdev
->md
.meta_dev_idx
) {
780 /* v07 style fixed size indexed meta data */
781 bdev
->md
.md_size_sect
= MD_128MB_SECT
;
782 bdev
->md
.al_offset
= MD_4kB_SECT
;
783 bdev
->md
.bm_offset
= MD_4kB_SECT
+ al_size_sect
;
785 case DRBD_MD_INDEX_FLEX_EXT
:
786 /* just occupy the full device; unit: sectors */
787 bdev
->md
.md_size_sect
= drbd_get_capacity(bdev
->md_bdev
);
788 bdev
->md
.al_offset
= MD_4kB_SECT
;
789 bdev
->md
.bm_offset
= MD_4kB_SECT
+ al_size_sect
;
791 case DRBD_MD_INDEX_INTERNAL
:
792 case DRBD_MD_INDEX_FLEX_INT
:
793 /* al size is still fixed */
794 bdev
->md
.al_offset
= -al_size_sect
;
795 /* we need (slightly less than) ~ this much bitmap sectors: */
796 md_size_sect
= drbd_get_capacity(bdev
->backing_bdev
);
797 md_size_sect
= ALIGN(md_size_sect
, BM_SECT_PER_EXT
);
798 md_size_sect
= BM_SECT_TO_EXT(md_size_sect
);
799 md_size_sect
= ALIGN(md_size_sect
, 8);
801 /* plus the "drbd meta data super block",
802 * and the activity log; */
803 md_size_sect
+= MD_4kB_SECT
+ al_size_sect
;
805 bdev
->md
.md_size_sect
= md_size_sect
;
806 /* bitmap offset is adjusted by 'super' block size */
807 bdev
->md
.bm_offset
= -md_size_sect
+ MD_4kB_SECT
;
812 /* input size is expected to be in KB */
813 char *ppsize(char *buf
, unsigned long long size
)
815 /* Needs 9 bytes at max including trailing NUL:
816 * -1ULL ==> "16384 EB" */
817 static char units
[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
819 while (size
>= 10000 && base
< sizeof(units
)-1) {
821 size
= (size
>> 10) + !!(size
& (1<<9));
824 sprintf(buf
, "%u %cB", (unsigned)size
, units
[base
]);
829 /* there is still a theoretical deadlock when called from receiver
830 * on an D_INCONSISTENT R_PRIMARY:
831 * remote READ does inc_ap_bio, receiver would need to receive answer
832 * packet from remote to dec_ap_bio again.
833 * receiver receive_sizes(), comes here,
834 * waits for ap_bio_cnt == 0. -> deadlock.
835 * but this cannot happen, actually, because:
836 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
837 * (not connected, or bad/no disk on peer):
838 * see drbd_fail_request_early, ap_bio_cnt is zero.
839 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
840 * peer may not initiate a resize.
842 /* Note these are not to be confused with
843 * drbd_adm_suspend_io/drbd_adm_resume_io,
844 * which are (sub) state changes triggered by admin (drbdsetup),
845 * and can be long lived.
846 * This changes an device->flag, is triggered by drbd internals,
847 * and should be short-lived. */
848 void drbd_suspend_io(struct drbd_device
*device
)
850 set_bit(SUSPEND_IO
, &device
->flags
);
851 if (drbd_suspended(device
))
853 wait_event(device
->misc_wait
, !atomic_read(&device
->ap_bio_cnt
));
856 void drbd_resume_io(struct drbd_device
*device
)
858 clear_bit(SUSPEND_IO
, &device
->flags
);
859 wake_up(&device
->misc_wait
);
863 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
864 * @device: DRBD device.
866 * Returns 0 on success, negative return values indicate errors.
867 * You should call drbd_md_sync() after calling this function.
869 enum determine_dev_size
870 drbd_determine_dev_size(struct drbd_device
*device
, enum dds_flags flags
, struct resize_parms
*rs
) __must_hold(local
)
872 sector_t prev_first_sect
, prev_size
; /* previous meta location */
873 sector_t la_size_sect
, u_size
;
874 struct drbd_md
*md
= &device
->ldev
->md
;
875 u32 prev_al_stripe_size_4k
;
881 int md_moved
, la_size_changed
;
882 enum determine_dev_size rv
= DS_UNCHANGED
;
885 * application request passes inc_ap_bio,
886 * but then cannot get an AL-reference.
887 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
890 * Suspend IO right here.
891 * still lock the act_log to not trigger ASSERTs there.
893 drbd_suspend_io(device
);
894 buffer
= drbd_md_get_buffer(device
); /* Lock meta-data IO */
896 drbd_resume_io(device
);
900 /* no wait necessary anymore, actually we could assert that */
901 wait_event(device
->al_wait
, lc_try_lock(device
->act_log
));
903 prev_first_sect
= drbd_md_first_sector(device
->ldev
);
904 prev_size
= device
->ldev
->md
.md_size_sect
;
905 la_size_sect
= device
->ldev
->md
.la_size_sect
;
908 /* rs is non NULL if we should change the AL layout only */
910 prev_al_stripes
= md
->al_stripes
;
911 prev_al_stripe_size_4k
= md
->al_stripe_size_4k
;
913 md
->al_stripes
= rs
->al_stripes
;
914 md
->al_stripe_size_4k
= rs
->al_stripe_size
/ 4;
915 md
->al_size_4k
= (u64
)rs
->al_stripes
* rs
->al_stripe_size
/ 4;
918 drbd_md_set_sector_offsets(device
, device
->ldev
);
921 u_size
= rcu_dereference(device
->ldev
->disk_conf
)->disk_size
;
923 size
= drbd_new_dev_size(device
, device
->ldev
, u_size
, flags
& DDSF_FORCED
);
925 if (size
< la_size_sect
) {
926 if (rs
&& u_size
== 0) {
927 /* Remove "rs &&" later. This check should always be active, but
928 right now the receiver expects the permissive behavior */
929 drbd_warn(device
, "Implicit shrink not allowed. "
930 "Use --size=%llus for explicit shrink.\n",
931 (unsigned long long)size
);
932 rv
= DS_ERROR_SHRINK
;
935 rv
= DS_ERROR_SPACE_MD
;
936 if (rv
!= DS_UNCHANGED
)
940 if (drbd_get_capacity(device
->this_bdev
) != size
||
941 drbd_bm_capacity(device
) != size
) {
943 err
= drbd_bm_resize(device
, size
, !(flags
& DDSF_NO_RESYNC
));
945 /* currently there is only one error: ENOMEM! */
946 size
= drbd_bm_capacity(device
)>>1;
948 drbd_err(device
, "OUT OF MEMORY! "
949 "Could not allocate bitmap!\n");
951 drbd_err(device
, "BM resizing failed. "
952 "Leaving size unchanged at size = %lu KB\n",
953 (unsigned long)size
);
957 /* racy, see comments above. */
958 drbd_set_my_capacity(device
, size
);
959 device
->ldev
->md
.la_size_sect
= size
;
960 drbd_info(device
, "size = %s (%llu KB)\n", ppsize(ppb
, size
>>1),
961 (unsigned long long)size
>>1);
966 la_size_changed
= (la_size_sect
!= device
->ldev
->md
.la_size_sect
);
968 md_moved
= prev_first_sect
!= drbd_md_first_sector(device
->ldev
)
969 || prev_size
!= device
->ldev
->md
.md_size_sect
;
971 if (la_size_changed
|| md_moved
|| rs
) {
974 drbd_al_shrink(device
); /* All extents inactive. */
976 prev_flags
= md
->flags
;
977 md
->flags
&= ~MDF_PRIMARY_IND
;
978 drbd_md_write(device
, buffer
);
980 drbd_info(device
, "Writing the whole bitmap, %s\n",
981 la_size_changed
&& md_moved
? "size changed and md moved" :
982 la_size_changed
? "size changed" : "md moved");
983 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
984 drbd_bitmap_io(device
, md_moved
? &drbd_bm_write_all
: &drbd_bm_write
,
985 "size changed", BM_LOCKED_MASK
);
986 drbd_initialize_al(device
, buffer
);
988 md
->flags
= prev_flags
;
989 drbd_md_write(device
, buffer
);
992 drbd_info(device
, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n",
993 md
->al_stripes
, md
->al_stripe_size_4k
* 4);
996 if (size
> la_size_sect
)
997 rv
= la_size_sect
? DS_GREW
: DS_GREW_FROM_ZERO
;
998 if (size
< la_size_sect
)
1004 md
->al_stripes
= prev_al_stripes
;
1005 md
->al_stripe_size_4k
= prev_al_stripe_size_4k
;
1006 md
->al_size_4k
= (u64
)prev_al_stripes
* prev_al_stripe_size_4k
;
1008 drbd_md_set_sector_offsets(device
, device
->ldev
);
1011 lc_unlock(device
->act_log
);
1012 wake_up(&device
->al_wait
);
1013 drbd_md_put_buffer(device
);
1014 drbd_resume_io(device
);
1020 drbd_new_dev_size(struct drbd_device
*device
, struct drbd_backing_dev
*bdev
,
1021 sector_t u_size
, int assume_peer_has_space
)
1023 sector_t p_size
= device
->p_size
; /* partner's disk size. */
1024 sector_t la_size_sect
= bdev
->md
.la_size_sect
; /* last agreed size. */
1025 sector_t m_size
; /* my size */
1028 m_size
= drbd_get_max_capacity(bdev
);
1030 if (device
->state
.conn
< C_CONNECTED
&& assume_peer_has_space
) {
1031 drbd_warn(device
, "Resize while not connected was forced by the user!\n");
1035 if (p_size
&& m_size
) {
1036 size
= min_t(sector_t
, p_size
, m_size
);
1039 size
= la_size_sect
;
1040 if (m_size
&& m_size
< size
)
1042 if (p_size
&& p_size
< size
)
1053 drbd_err(device
, "Both nodes diskless!\n");
1057 drbd_err(device
, "Requested disk size is too big (%lu > %lu)\n",
1058 (unsigned long)u_size
>>1, (unsigned long)size
>>1);
1067 * drbd_check_al_size() - Ensures that the AL is of the right size
1068 * @device: DRBD device.
1070 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
1071 * failed, and 0 on success. You should call drbd_md_sync() after you called
1074 static int drbd_check_al_size(struct drbd_device
*device
, struct disk_conf
*dc
)
1076 struct lru_cache
*n
, *t
;
1077 struct lc_element
*e
;
1078 unsigned int in_use
;
1081 if (device
->act_log
&&
1082 device
->act_log
->nr_elements
== dc
->al_extents
)
1086 t
= device
->act_log
;
1087 n
= lc_create("act_log", drbd_al_ext_cache
, AL_UPDATES_PER_TRANSACTION
,
1088 dc
->al_extents
, sizeof(struct lc_element
), 0);
1091 drbd_err(device
, "Cannot allocate act_log lru!\n");
1094 spin_lock_irq(&device
->al_lock
);
1096 for (i
= 0; i
< t
->nr_elements
; i
++) {
1097 e
= lc_element_by_index(t
, i
);
1099 drbd_err(device
, "refcnt(%d)==%d\n",
1100 e
->lc_number
, e
->refcnt
);
1101 in_use
+= e
->refcnt
;
1105 device
->act_log
= n
;
1106 spin_unlock_irq(&device
->al_lock
);
1108 drbd_err(device
, "Activity log still in use!\n");
1115 drbd_md_mark_dirty(device
); /* we changed device->act_log->nr_elemens */
1119 static void drbd_setup_queue_param(struct drbd_device
*device
, unsigned int max_bio_size
)
1121 struct request_queue
* const q
= device
->rq_queue
;
1122 unsigned int max_hw_sectors
= max_bio_size
>> 9;
1123 unsigned int max_segments
= 0;
1124 struct request_queue
*b
= NULL
;
1126 if (get_ldev_if_state(device
, D_ATTACHING
)) {
1127 b
= device
->ldev
->backing_bdev
->bd_disk
->queue
;
1129 max_hw_sectors
= min(queue_max_hw_sectors(b
), max_bio_size
>> 9);
1131 max_segments
= rcu_dereference(device
->ldev
->disk_conf
)->max_bio_bvecs
;
1134 blk_set_stacking_limits(&q
->limits
);
1135 blk_queue_max_write_same_sectors(q
, 0);
1138 blk_queue_logical_block_size(q
, 512);
1139 blk_queue_max_hw_sectors(q
, max_hw_sectors
);
1140 /* This is the workaround for "bio would need to, but cannot, be split" */
1141 blk_queue_max_segments(q
, max_segments
? max_segments
: BLK_MAX_SEGMENTS
);
1142 blk_queue_segment_boundary(q
, PAGE_CACHE_SIZE
-1);
1145 struct drbd_connection
*connection
= first_peer_device(device
)->connection
;
1147 if (blk_queue_discard(b
) &&
1148 (connection
->cstate
< C_CONNECTED
|| connection
->agreed_features
& FF_TRIM
)) {
1149 /* For now, don't allow more than one activity log extent worth of data
1150 * to be discarded in one go. We may need to rework drbd_al_begin_io()
1151 * to allow for even larger discard ranges */
1152 q
->limits
.max_discard_sectors
= DRBD_MAX_DISCARD_SECTORS
;
1154 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
1155 /* REALLY? Is stacking secdiscard "legal"? */
1156 if (blk_queue_secdiscard(b
))
1157 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD
, q
);
1159 q
->limits
.max_discard_sectors
= 0;
1160 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
1161 queue_flag_clear_unlocked(QUEUE_FLAG_SECDISCARD
, q
);
1164 blk_queue_stack_limits(q
, b
);
1166 if (q
->backing_dev_info
.ra_pages
!= b
->backing_dev_info
.ra_pages
) {
1167 drbd_info(device
, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1168 q
->backing_dev_info
.ra_pages
,
1169 b
->backing_dev_info
.ra_pages
);
1170 q
->backing_dev_info
.ra_pages
= b
->backing_dev_info
.ra_pages
;
1176 void drbd_reconsider_max_bio_size(struct drbd_device
*device
)
1178 unsigned int now
, new, local
, peer
;
1180 now
= queue_max_hw_sectors(device
->rq_queue
) << 9;
1181 local
= device
->local_max_bio_size
; /* Eventually last known value, from volatile memory */
1182 peer
= device
->peer_max_bio_size
; /* Eventually last known value, from meta data */
1184 if (get_ldev_if_state(device
, D_ATTACHING
)) {
1185 local
= queue_max_hw_sectors(device
->ldev
->backing_bdev
->bd_disk
->queue
) << 9;
1186 device
->local_max_bio_size
= local
;
1189 local
= min(local
, DRBD_MAX_BIO_SIZE
);
1191 /* We may ignore peer limits if the peer is modern enough.
1192 Because new from 8.3.8 onwards the peer can use multiple
1193 BIOs for a single peer_request */
1194 if (device
->state
.conn
>= C_WF_REPORT_PARAMS
) {
1195 if (first_peer_device(device
)->connection
->agreed_pro_version
< 94)
1196 peer
= min(device
->peer_max_bio_size
, DRBD_MAX_SIZE_H80_PACKET
);
1197 /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1198 else if (first_peer_device(device
)->connection
->agreed_pro_version
== 94)
1199 peer
= DRBD_MAX_SIZE_H80_PACKET
;
1200 else if (first_peer_device(device
)->connection
->agreed_pro_version
< 100)
1201 peer
= DRBD_MAX_BIO_SIZE_P95
; /* drbd 8.3.8 onwards, before 8.4.0 */
1203 peer
= DRBD_MAX_BIO_SIZE
;
1205 /* We may later detach and re-attach on a disconnected Primary.
1206 * Avoid this setting to jump back in that case.
1207 * We want to store what we know the peer DRBD can handle,
1208 * not what the peer IO backend can handle. */
1209 if (peer
> device
->peer_max_bio_size
)
1210 device
->peer_max_bio_size
= peer
;
1212 new = min(local
, peer
);
1214 if (device
->state
.role
== R_PRIMARY
&& new < now
)
1215 drbd_err(device
, "ASSERT FAILED new < now; (%u < %u)\n", new, now
);
1218 drbd_info(device
, "max BIO size = %u\n", new);
1220 drbd_setup_queue_param(device
, new);
1223 /* Starts the worker thread */
1224 static void conn_reconfig_start(struct drbd_connection
*connection
)
1226 drbd_thread_start(&connection
->worker
);
1227 drbd_flush_workqueue(&connection
->sender_work
);
1230 /* if still unconfigured, stops worker again. */
1231 static void conn_reconfig_done(struct drbd_connection
*connection
)
1234 spin_lock_irq(&connection
->resource
->req_lock
);
1235 stop_threads
= conn_all_vols_unconf(connection
) &&
1236 connection
->cstate
== C_STANDALONE
;
1237 spin_unlock_irq(&connection
->resource
->req_lock
);
1239 /* asender is implicitly stopped by receiver
1240 * in conn_disconnect() */
1241 drbd_thread_stop(&connection
->receiver
);
1242 drbd_thread_stop(&connection
->worker
);
1246 /* Make sure IO is suspended before calling this function(). */
1247 static void drbd_suspend_al(struct drbd_device
*device
)
1251 if (!lc_try_lock(device
->act_log
)) {
1252 drbd_warn(device
, "Failed to lock al in drbd_suspend_al()\n");
1256 drbd_al_shrink(device
);
1257 spin_lock_irq(&device
->resource
->req_lock
);
1258 if (device
->state
.conn
< C_CONNECTED
)
1259 s
= !test_and_set_bit(AL_SUSPENDED
, &device
->flags
);
1260 spin_unlock_irq(&device
->resource
->req_lock
);
1261 lc_unlock(device
->act_log
);
1264 drbd_info(device
, "Suspended AL updates\n");
1268 static bool should_set_defaults(struct genl_info
*info
)
1270 unsigned flags
= ((struct drbd_genlmsghdr
*)info
->userhdr
)->flags
;
1271 return 0 != (flags
& DRBD_GENL_F_SET_DEFAULTS
);
1274 static unsigned int drbd_al_extents_max(struct drbd_backing_dev
*bdev
)
1276 /* This is limited by 16 bit "slot" numbers,
1277 * and by available on-disk context storage.
1279 * Also (u16)~0 is special (denotes a "free" extent).
1281 * One transaction occupies one 4kB on-disk block,
1282 * we have n such blocks in the on disk ring buffer,
1283 * the "current" transaction may fail (n-1),
1284 * and there is 919 slot numbers context information per transaction.
1286 * 72 transaction blocks amounts to more than 2**16 context slots,
1287 * so cap there first.
1289 const unsigned int max_al_nr
= DRBD_AL_EXTENTS_MAX
;
1290 const unsigned int sufficient_on_disk
=
1291 (max_al_nr
+ AL_CONTEXT_PER_TRANSACTION
-1)
1292 /AL_CONTEXT_PER_TRANSACTION
;
1294 unsigned int al_size_4k
= bdev
->md
.al_size_4k
;
1296 if (al_size_4k
> sufficient_on_disk
)
1299 return (al_size_4k
- 1) * AL_CONTEXT_PER_TRANSACTION
;
1302 int drbd_adm_disk_opts(struct sk_buff
*skb
, struct genl_info
*info
)
1304 struct drbd_config_context adm_ctx
;
1305 enum drbd_ret_code retcode
;
1306 struct drbd_device
*device
;
1307 struct disk_conf
*new_disk_conf
, *old_disk_conf
;
1308 struct fifo_buffer
*old_plan
= NULL
, *new_plan
= NULL
;
1311 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
1312 if (!adm_ctx
.reply_skb
)
1314 if (retcode
!= NO_ERROR
)
1317 device
= adm_ctx
.device
;
1318 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
1320 /* we also need a disk
1321 * to change the options on */
1322 if (!get_ldev(device
)) {
1323 retcode
= ERR_NO_DISK
;
1327 new_disk_conf
= kmalloc(sizeof(struct disk_conf
), GFP_KERNEL
);
1328 if (!new_disk_conf
) {
1329 retcode
= ERR_NOMEM
;
1333 mutex_lock(&device
->resource
->conf_update
);
1334 old_disk_conf
= device
->ldev
->disk_conf
;
1335 *new_disk_conf
= *old_disk_conf
;
1336 if (should_set_defaults(info
))
1337 set_disk_conf_defaults(new_disk_conf
);
1339 err
= disk_conf_from_attrs_for_change(new_disk_conf
, info
);
1340 if (err
&& err
!= -ENOMSG
) {
1341 retcode
= ERR_MANDATORY_TAG
;
1342 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
1346 if (!expect(new_disk_conf
->resync_rate
>= 1))
1347 new_disk_conf
->resync_rate
= 1;
1349 if (new_disk_conf
->al_extents
< DRBD_AL_EXTENTS_MIN
)
1350 new_disk_conf
->al_extents
= DRBD_AL_EXTENTS_MIN
;
1351 if (new_disk_conf
->al_extents
> drbd_al_extents_max(device
->ldev
))
1352 new_disk_conf
->al_extents
= drbd_al_extents_max(device
->ldev
);
1354 if (new_disk_conf
->c_plan_ahead
> DRBD_C_PLAN_AHEAD_MAX
)
1355 new_disk_conf
->c_plan_ahead
= DRBD_C_PLAN_AHEAD_MAX
;
1357 fifo_size
= (new_disk_conf
->c_plan_ahead
* 10 * SLEEP_TIME
) / HZ
;
1358 if (fifo_size
!= device
->rs_plan_s
->size
) {
1359 new_plan
= fifo_alloc(fifo_size
);
1361 drbd_err(device
, "kmalloc of fifo_buffer failed");
1362 retcode
= ERR_NOMEM
;
1367 drbd_suspend_io(device
);
1368 wait_event(device
->al_wait
, lc_try_lock(device
->act_log
));
1369 drbd_al_shrink(device
);
1370 err
= drbd_check_al_size(device
, new_disk_conf
);
1371 lc_unlock(device
->act_log
);
1372 wake_up(&device
->al_wait
);
1373 drbd_resume_io(device
);
1376 retcode
= ERR_NOMEM
;
1380 write_lock_irq(&global_state_lock
);
1381 retcode
= drbd_resync_after_valid(device
, new_disk_conf
->resync_after
);
1382 if (retcode
== NO_ERROR
) {
1383 rcu_assign_pointer(device
->ldev
->disk_conf
, new_disk_conf
);
1384 drbd_resync_after_changed(device
);
1386 write_unlock_irq(&global_state_lock
);
1388 if (retcode
!= NO_ERROR
)
1392 old_plan
= device
->rs_plan_s
;
1393 rcu_assign_pointer(device
->rs_plan_s
, new_plan
);
1396 mutex_unlock(&device
->resource
->conf_update
);
1398 if (new_disk_conf
->al_updates
)
1399 device
->ldev
->md
.flags
&= ~MDF_AL_DISABLED
;
1401 device
->ldev
->md
.flags
|= MDF_AL_DISABLED
;
1403 if (new_disk_conf
->md_flushes
)
1404 clear_bit(MD_NO_FUA
, &device
->flags
);
1406 set_bit(MD_NO_FUA
, &device
->flags
);
1408 drbd_bump_write_ordering(first_peer_device(device
)->connection
, WO_bdev_flush
);
1410 drbd_md_sync(device
);
1412 if (device
->state
.conn
>= C_CONNECTED
) {
1413 struct drbd_peer_device
*peer_device
;
1415 for_each_peer_device(peer_device
, device
)
1416 drbd_send_sync_param(peer_device
);
1420 kfree(old_disk_conf
);
1422 mod_timer(&device
->request_timer
, jiffies
+ HZ
);
1426 mutex_unlock(&device
->resource
->conf_update
);
1428 kfree(new_disk_conf
);
1433 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
1435 drbd_adm_finish(&adm_ctx
, info
, retcode
);
1439 int drbd_adm_attach(struct sk_buff
*skb
, struct genl_info
*info
)
1441 struct drbd_config_context adm_ctx
;
1442 struct drbd_device
*device
;
1444 enum drbd_ret_code retcode
;
1445 enum determine_dev_size dd
;
1446 sector_t max_possible_sectors
;
1447 sector_t min_md_device_sectors
;
1448 struct drbd_backing_dev
*nbc
= NULL
; /* new_backing_conf */
1449 struct disk_conf
*new_disk_conf
= NULL
;
1450 struct block_device
*bdev
;
1451 struct lru_cache
*resync_lru
= NULL
;
1452 struct fifo_buffer
*new_plan
= NULL
;
1453 union drbd_state ns
, os
;
1454 enum drbd_state_rv rv
;
1455 struct net_conf
*nc
;
1457 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
1458 if (!adm_ctx
.reply_skb
)
1460 if (retcode
!= NO_ERROR
)
1463 device
= adm_ctx
.device
;
1464 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
1465 conn_reconfig_start(first_peer_device(device
)->connection
);
1467 /* if you want to reconfigure, please tear down first */
1468 if (device
->state
.disk
> D_DISKLESS
) {
1469 retcode
= ERR_DISK_CONFIGURED
;
1472 /* It may just now have detached because of IO error. Make sure
1473 * drbd_ldev_destroy is done already, we may end up here very fast,
1474 * e.g. if someone calls attach from the on-io-error handler,
1475 * to realize a "hot spare" feature (not that I'd recommend that) */
1476 wait_event(device
->misc_wait
, !atomic_read(&device
->local_cnt
));
1478 /* make sure there is no leftover from previous force-detach attempts */
1479 clear_bit(FORCE_DETACH
, &device
->flags
);
1480 clear_bit(WAS_IO_ERROR
, &device
->flags
);
1481 clear_bit(WAS_READ_ERROR
, &device
->flags
);
1483 /* and no leftover from previously aborted resync or verify, either */
1484 device
->rs_total
= 0;
1485 device
->rs_failed
= 0;
1486 atomic_set(&device
->rs_pending_cnt
, 0);
1488 /* allocation not in the IO path, drbdsetup context */
1489 nbc
= kzalloc(sizeof(struct drbd_backing_dev
), GFP_KERNEL
);
1491 retcode
= ERR_NOMEM
;
1494 spin_lock_init(&nbc
->md
.uuid_lock
);
1496 new_disk_conf
= kzalloc(sizeof(struct disk_conf
), GFP_KERNEL
);
1497 if (!new_disk_conf
) {
1498 retcode
= ERR_NOMEM
;
1501 nbc
->disk_conf
= new_disk_conf
;
1503 set_disk_conf_defaults(new_disk_conf
);
1504 err
= disk_conf_from_attrs(new_disk_conf
, info
);
1506 retcode
= ERR_MANDATORY_TAG
;
1507 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
1511 if (new_disk_conf
->c_plan_ahead
> DRBD_C_PLAN_AHEAD_MAX
)
1512 new_disk_conf
->c_plan_ahead
= DRBD_C_PLAN_AHEAD_MAX
;
1514 new_plan
= fifo_alloc((new_disk_conf
->c_plan_ahead
* 10 * SLEEP_TIME
) / HZ
);
1516 retcode
= ERR_NOMEM
;
1520 if (new_disk_conf
->meta_dev_idx
< DRBD_MD_INDEX_FLEX_INT
) {
1521 retcode
= ERR_MD_IDX_INVALID
;
1525 write_lock_irq(&global_state_lock
);
1526 retcode
= drbd_resync_after_valid(device
, new_disk_conf
->resync_after
);
1527 write_unlock_irq(&global_state_lock
);
1528 if (retcode
!= NO_ERROR
)
1532 nc
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
);
1534 if (new_disk_conf
->fencing
== FP_STONITH
&& nc
->wire_protocol
== DRBD_PROT_A
) {
1536 retcode
= ERR_STONITH_AND_PROT_A
;
1542 bdev
= blkdev_get_by_path(new_disk_conf
->backing_dev
,
1543 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, device
);
1545 drbd_err(device
, "open(\"%s\") failed with %ld\n", new_disk_conf
->backing_dev
,
1547 retcode
= ERR_OPEN_DISK
;
1550 nbc
->backing_bdev
= bdev
;
1553 * meta_dev_idx >= 0: external fixed size, possibly multiple
1554 * drbd sharing one meta device. TODO in that case, paranoia
1555 * check that [md_bdev, meta_dev_idx] is not yet used by some
1556 * other drbd minor! (if you use drbd.conf + drbdadm, that
1557 * should check it for you already; but if you don't, or
1558 * someone fooled it, we need to double check here)
1560 bdev
= blkdev_get_by_path(new_disk_conf
->meta_dev
,
1561 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
,
1562 (new_disk_conf
->meta_dev_idx
< 0) ?
1563 (void *)device
: (void *)drbd_m_holder
);
1565 drbd_err(device
, "open(\"%s\") failed with %ld\n", new_disk_conf
->meta_dev
,
1567 retcode
= ERR_OPEN_MD_DISK
;
1570 nbc
->md_bdev
= bdev
;
1572 if ((nbc
->backing_bdev
== nbc
->md_bdev
) !=
1573 (new_disk_conf
->meta_dev_idx
== DRBD_MD_INDEX_INTERNAL
||
1574 new_disk_conf
->meta_dev_idx
== DRBD_MD_INDEX_FLEX_INT
)) {
1575 retcode
= ERR_MD_IDX_INVALID
;
1579 resync_lru
= lc_create("resync", drbd_bm_ext_cache
,
1580 1, 61, sizeof(struct bm_extent
),
1581 offsetof(struct bm_extent
, lce
));
1583 retcode
= ERR_NOMEM
;
1587 /* Read our meta data super block early.
1588 * This also sets other on-disk offsets. */
1589 retcode
= drbd_md_read(device
, nbc
);
1590 if (retcode
!= NO_ERROR
)
1593 if (new_disk_conf
->al_extents
< DRBD_AL_EXTENTS_MIN
)
1594 new_disk_conf
->al_extents
= DRBD_AL_EXTENTS_MIN
;
1595 if (new_disk_conf
->al_extents
> drbd_al_extents_max(nbc
))
1596 new_disk_conf
->al_extents
= drbd_al_extents_max(nbc
);
1598 if (drbd_get_max_capacity(nbc
) < new_disk_conf
->disk_size
) {
1599 drbd_err(device
, "max capacity %llu smaller than disk size %llu\n",
1600 (unsigned long long) drbd_get_max_capacity(nbc
),
1601 (unsigned long long) new_disk_conf
->disk_size
);
1602 retcode
= ERR_DISK_TOO_SMALL
;
1606 if (new_disk_conf
->meta_dev_idx
< 0) {
1607 max_possible_sectors
= DRBD_MAX_SECTORS_FLEX
;
1608 /* at least one MB, otherwise it does not make sense */
1609 min_md_device_sectors
= (2<<10);
1611 max_possible_sectors
= DRBD_MAX_SECTORS
;
1612 min_md_device_sectors
= MD_128MB_SECT
* (new_disk_conf
->meta_dev_idx
+ 1);
1615 if (drbd_get_capacity(nbc
->md_bdev
) < min_md_device_sectors
) {
1616 retcode
= ERR_MD_DISK_TOO_SMALL
;
1617 drbd_warn(device
, "refusing attach: md-device too small, "
1618 "at least %llu sectors needed for this meta-disk type\n",
1619 (unsigned long long) min_md_device_sectors
);
1623 /* Make sure the new disk is big enough
1624 * (we may currently be R_PRIMARY with no local disk...) */
1625 if (drbd_get_max_capacity(nbc
) <
1626 drbd_get_capacity(device
->this_bdev
)) {
1627 retcode
= ERR_DISK_TOO_SMALL
;
1631 nbc
->known_size
= drbd_get_capacity(nbc
->backing_bdev
);
1633 if (nbc
->known_size
> max_possible_sectors
) {
1634 drbd_warn(device
, "==> truncating very big lower level device "
1635 "to currently maximum possible %llu sectors <==\n",
1636 (unsigned long long) max_possible_sectors
);
1637 if (new_disk_conf
->meta_dev_idx
>= 0)
1638 drbd_warn(device
, "==>> using internal or flexible "
1639 "meta data may help <<==\n");
1642 drbd_suspend_io(device
);
1643 /* also wait for the last barrier ack. */
1644 /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1645 * We need a way to either ignore barrier acks for barriers sent before a device
1646 * was attached, or a way to wait for all pending barrier acks to come in.
1647 * As barriers are counted per resource,
1648 * we'd need to suspend io on all devices of a resource.
1650 wait_event(device
->misc_wait
, !atomic_read(&device
->ap_pending_cnt
) || drbd_suspended(device
));
1651 /* and for any other previously queued work */
1652 drbd_flush_workqueue(&first_peer_device(device
)->connection
->sender_work
);
1654 rv
= _drbd_request_state(device
, NS(disk
, D_ATTACHING
), CS_VERBOSE
);
1655 retcode
= rv
; /* FIXME: Type mismatch. */
1656 drbd_resume_io(device
);
1657 if (rv
< SS_SUCCESS
)
1660 if (!get_ldev_if_state(device
, D_ATTACHING
))
1661 goto force_diskless
;
1663 if (!device
->bitmap
) {
1664 if (drbd_bm_init(device
)) {
1665 retcode
= ERR_NOMEM
;
1666 goto force_diskless_dec
;
1670 if (device
->state
.conn
< C_CONNECTED
&&
1671 device
->state
.role
== R_PRIMARY
&& device
->ed_uuid
&&
1672 (device
->ed_uuid
& ~((u64
)1)) != (nbc
->md
.uuid
[UI_CURRENT
] & ~((u64
)1))) {
1673 drbd_err(device
, "Can only attach to data with current UUID=%016llX\n",
1674 (unsigned long long)device
->ed_uuid
);
1675 retcode
= ERR_DATA_NOT_CURRENT
;
1676 goto force_diskless_dec
;
1679 /* Since we are diskless, fix the activity log first... */
1680 if (drbd_check_al_size(device
, new_disk_conf
)) {
1681 retcode
= ERR_NOMEM
;
1682 goto force_diskless_dec
;
1685 /* Prevent shrinking of consistent devices ! */
1686 if (drbd_md_test_flag(nbc
, MDF_CONSISTENT
) &&
1687 drbd_new_dev_size(device
, nbc
, nbc
->disk_conf
->disk_size
, 0) < nbc
->md
.la_size_sect
) {
1688 drbd_warn(device
, "refusing to truncate a consistent device\n");
1689 retcode
= ERR_DISK_TOO_SMALL
;
1690 goto force_diskless_dec
;
1693 /* Reset the "barriers don't work" bits here, then force meta data to
1694 * be written, to ensure we determine if barriers are supported. */
1695 if (new_disk_conf
->md_flushes
)
1696 clear_bit(MD_NO_FUA
, &device
->flags
);
1698 set_bit(MD_NO_FUA
, &device
->flags
);
1700 /* Point of no return reached.
1701 * Devices and memory are no longer released by error cleanup below.
1702 * now device takes over responsibility, and the state engine should
1703 * clean it up somewhere. */
1704 D_ASSERT(device
, device
->ldev
== NULL
);
1706 device
->resync
= resync_lru
;
1707 device
->rs_plan_s
= new_plan
;
1710 new_disk_conf
= NULL
;
1713 drbd_bump_write_ordering(first_peer_device(device
)->connection
, WO_bdev_flush
);
1715 if (drbd_md_test_flag(device
->ldev
, MDF_CRASHED_PRIMARY
))
1716 set_bit(CRASHED_PRIMARY
, &device
->flags
);
1718 clear_bit(CRASHED_PRIMARY
, &device
->flags
);
1720 if (drbd_md_test_flag(device
->ldev
, MDF_PRIMARY_IND
) &&
1721 !(device
->state
.role
== R_PRIMARY
&& device
->resource
->susp_nod
))
1722 set_bit(CRASHED_PRIMARY
, &device
->flags
);
1724 device
->send_cnt
= 0;
1725 device
->recv_cnt
= 0;
1726 device
->read_cnt
= 0;
1727 device
->writ_cnt
= 0;
1729 drbd_reconsider_max_bio_size(device
);
1731 /* If I am currently not R_PRIMARY,
1732 * but meta data primary indicator is set,
1733 * I just now recover from a hard crash,
1734 * and have been R_PRIMARY before that crash.
1736 * Now, if I had no connection before that crash
1737 * (have been degraded R_PRIMARY), chances are that
1738 * I won't find my peer now either.
1740 * In that case, and _only_ in that case,
1741 * we use the degr-wfc-timeout instead of the default,
1742 * so we can automatically recover from a crash of a
1743 * degraded but active "cluster" after a certain timeout.
1745 clear_bit(USE_DEGR_WFC_T
, &device
->flags
);
1746 if (device
->state
.role
!= R_PRIMARY
&&
1747 drbd_md_test_flag(device
->ldev
, MDF_PRIMARY_IND
) &&
1748 !drbd_md_test_flag(device
->ldev
, MDF_CONNECTED_IND
))
1749 set_bit(USE_DEGR_WFC_T
, &device
->flags
);
1751 dd
= drbd_determine_dev_size(device
, 0, NULL
);
1752 if (dd
<= DS_ERROR
) {
1753 retcode
= ERR_NOMEM_BITMAP
;
1754 goto force_diskless_dec
;
1755 } else if (dd
== DS_GREW
)
1756 set_bit(RESYNC_AFTER_NEG
, &device
->flags
);
1758 if (drbd_md_test_flag(device
->ldev
, MDF_FULL_SYNC
) ||
1759 (test_bit(CRASHED_PRIMARY
, &device
->flags
) &&
1760 drbd_md_test_flag(device
->ldev
, MDF_AL_DISABLED
))) {
1761 drbd_info(device
, "Assuming that all blocks are out of sync "
1762 "(aka FullSync)\n");
1763 if (drbd_bitmap_io(device
, &drbd_bmio_set_n_write
,
1764 "set_n_write from attaching", BM_LOCKED_MASK
)) {
1765 retcode
= ERR_IO_MD_DISK
;
1766 goto force_diskless_dec
;
1769 if (drbd_bitmap_io(device
, &drbd_bm_read
,
1770 "read from attaching", BM_LOCKED_MASK
)) {
1771 retcode
= ERR_IO_MD_DISK
;
1772 goto force_diskless_dec
;
1776 if (_drbd_bm_total_weight(device
) == drbd_bm_bits(device
))
1777 drbd_suspend_al(device
); /* IO is still suspended here... */
1779 spin_lock_irq(&device
->resource
->req_lock
);
1780 os
= drbd_read_state(device
);
1782 /* If MDF_CONSISTENT is not set go into inconsistent state,
1783 otherwise investigate MDF_WasUpToDate...
1784 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1785 otherwise into D_CONSISTENT state.
1787 if (drbd_md_test_flag(device
->ldev
, MDF_CONSISTENT
)) {
1788 if (drbd_md_test_flag(device
->ldev
, MDF_WAS_UP_TO_DATE
))
1789 ns
.disk
= D_CONSISTENT
;
1791 ns
.disk
= D_OUTDATED
;
1793 ns
.disk
= D_INCONSISTENT
;
1796 if (drbd_md_test_flag(device
->ldev
, MDF_PEER_OUT_DATED
))
1797 ns
.pdsk
= D_OUTDATED
;
1800 if (ns
.disk
== D_CONSISTENT
&&
1801 (ns
.pdsk
== D_OUTDATED
|| rcu_dereference(device
->ldev
->disk_conf
)->fencing
== FP_DONT_CARE
))
1802 ns
.disk
= D_UP_TO_DATE
;
1804 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1805 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1806 this point, because drbd_request_state() modifies these
1809 if (rcu_dereference(device
->ldev
->disk_conf
)->al_updates
)
1810 device
->ldev
->md
.flags
&= ~MDF_AL_DISABLED
;
1812 device
->ldev
->md
.flags
|= MDF_AL_DISABLED
;
1816 /* In case we are C_CONNECTED postpone any decision on the new disk
1817 state after the negotiation phase. */
1818 if (device
->state
.conn
== C_CONNECTED
) {
1819 device
->new_state_tmp
.i
= ns
.i
;
1821 ns
.disk
= D_NEGOTIATING
;
1823 /* We expect to receive up-to-date UUIDs soon.
1824 To avoid a race in receive_state, free p_uuid while
1825 holding req_lock. I.e. atomic with the state change */
1826 kfree(device
->p_uuid
);
1827 device
->p_uuid
= NULL
;
1830 rv
= _drbd_set_state(device
, ns
, CS_VERBOSE
, NULL
);
1831 spin_unlock_irq(&device
->resource
->req_lock
);
1833 if (rv
< SS_SUCCESS
)
1834 goto force_diskless_dec
;
1836 mod_timer(&device
->request_timer
, jiffies
+ HZ
);
1838 if (device
->state
.role
== R_PRIMARY
)
1839 device
->ldev
->md
.uuid
[UI_CURRENT
] |= (u64
)1;
1841 device
->ldev
->md
.uuid
[UI_CURRENT
] &= ~(u64
)1;
1843 drbd_md_mark_dirty(device
);
1844 drbd_md_sync(device
);
1846 kobject_uevent(&disk_to_dev(device
->vdisk
)->kobj
, KOBJ_CHANGE
);
1848 conn_reconfig_done(first_peer_device(device
)->connection
);
1849 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
1850 drbd_adm_finish(&adm_ctx
, info
, retcode
);
1856 drbd_force_state(device
, NS(disk
, D_DISKLESS
));
1857 drbd_md_sync(device
);
1859 conn_reconfig_done(first_peer_device(device
)->connection
);
1861 if (nbc
->backing_bdev
)
1862 blkdev_put(nbc
->backing_bdev
,
1863 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
1865 blkdev_put(nbc
->md_bdev
,
1866 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
1869 kfree(new_disk_conf
);
1870 lc_destroy(resync_lru
);
1872 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
1874 drbd_adm_finish(&adm_ctx
, info
, retcode
);
1878 static int adm_detach(struct drbd_device
*device
, int force
)
1880 enum drbd_state_rv retcode
;
1884 set_bit(FORCE_DETACH
, &device
->flags
);
1885 drbd_force_state(device
, NS(disk
, D_FAILED
));
1886 retcode
= SS_SUCCESS
;
1890 drbd_suspend_io(device
); /* so no-one is stuck in drbd_al_begin_io */
1891 drbd_md_get_buffer(device
); /* make sure there is no in-flight meta-data IO */
1892 retcode
= drbd_request_state(device
, NS(disk
, D_FAILED
));
1893 drbd_md_put_buffer(device
);
1894 /* D_FAILED will transition to DISKLESS. */
1895 ret
= wait_event_interruptible(device
->misc_wait
,
1896 device
->state
.disk
!= D_FAILED
);
1897 drbd_resume_io(device
);
1898 if ((int)retcode
== (int)SS_IS_DISKLESS
)
1899 retcode
= SS_NOTHING_TO_DO
;
1906 /* Detaching the disk is a process in multiple stages. First we need to lock
1907 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1908 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1909 * internal references as well.
1910 * Only then we have finally detached. */
1911 int drbd_adm_detach(struct sk_buff
*skb
, struct genl_info
*info
)
1913 struct drbd_config_context adm_ctx
;
1914 enum drbd_ret_code retcode
;
1915 struct detach_parms parms
= { };
1918 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
1919 if (!adm_ctx
.reply_skb
)
1921 if (retcode
!= NO_ERROR
)
1924 if (info
->attrs
[DRBD_NLA_DETACH_PARMS
]) {
1925 err
= detach_parms_from_attrs(&parms
, info
);
1927 retcode
= ERR_MANDATORY_TAG
;
1928 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
1933 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
1934 retcode
= adm_detach(adm_ctx
.device
, parms
.force_detach
);
1935 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
1937 drbd_adm_finish(&adm_ctx
, info
, retcode
);
1941 static bool conn_resync_running(struct drbd_connection
*connection
)
1943 struct drbd_peer_device
*peer_device
;
1948 idr_for_each_entry(&connection
->peer_devices
, peer_device
, vnr
) {
1949 struct drbd_device
*device
= peer_device
->device
;
1950 if (device
->state
.conn
== C_SYNC_SOURCE
||
1951 device
->state
.conn
== C_SYNC_TARGET
||
1952 device
->state
.conn
== C_PAUSED_SYNC_S
||
1953 device
->state
.conn
== C_PAUSED_SYNC_T
) {
1963 static bool conn_ov_running(struct drbd_connection
*connection
)
1965 struct drbd_peer_device
*peer_device
;
1970 idr_for_each_entry(&connection
->peer_devices
, peer_device
, vnr
) {
1971 struct drbd_device
*device
= peer_device
->device
;
1972 if (device
->state
.conn
== C_VERIFY_S
||
1973 device
->state
.conn
== C_VERIFY_T
) {
1983 static enum drbd_ret_code
1984 _check_net_options(struct drbd_connection
*connection
, struct net_conf
*old_net_conf
, struct net_conf
*new_net_conf
)
1986 struct drbd_peer_device
*peer_device
;
1989 if (old_net_conf
&& connection
->cstate
== C_WF_REPORT_PARAMS
&& connection
->agreed_pro_version
< 100) {
1990 if (new_net_conf
->wire_protocol
!= old_net_conf
->wire_protocol
)
1991 return ERR_NEED_APV_100
;
1993 if (new_net_conf
->two_primaries
!= old_net_conf
->two_primaries
)
1994 return ERR_NEED_APV_100
;
1996 if (strcmp(new_net_conf
->integrity_alg
, old_net_conf
->integrity_alg
))
1997 return ERR_NEED_APV_100
;
2000 if (!new_net_conf
->two_primaries
&&
2001 conn_highest_role(connection
) == R_PRIMARY
&&
2002 conn_highest_peer(connection
) == R_PRIMARY
)
2003 return ERR_NEED_ALLOW_TWO_PRI
;
2005 if (new_net_conf
->two_primaries
&&
2006 (new_net_conf
->wire_protocol
!= DRBD_PROT_C
))
2007 return ERR_NOT_PROTO_C
;
2009 idr_for_each_entry(&connection
->peer_devices
, peer_device
, i
) {
2010 struct drbd_device
*device
= peer_device
->device
;
2011 if (get_ldev(device
)) {
2012 enum drbd_fencing_p fp
= rcu_dereference(device
->ldev
->disk_conf
)->fencing
;
2014 if (new_net_conf
->wire_protocol
== DRBD_PROT_A
&& fp
== FP_STONITH
)
2015 return ERR_STONITH_AND_PROT_A
;
2017 if (device
->state
.role
== R_PRIMARY
&& new_net_conf
->discard_my_data
)
2018 return ERR_DISCARD_IMPOSSIBLE
;
2021 if (new_net_conf
->on_congestion
!= OC_BLOCK
&& new_net_conf
->wire_protocol
!= DRBD_PROT_A
)
2022 return ERR_CONG_NOT_PROTO_A
;
2027 static enum drbd_ret_code
2028 check_net_options(struct drbd_connection
*connection
, struct net_conf
*new_net_conf
)
2030 static enum drbd_ret_code rv
;
2031 struct drbd_peer_device
*peer_device
;
2035 rv
= _check_net_options(connection
, rcu_dereference(connection
->net_conf
), new_net_conf
);
2038 /* connection->volumes protected by genl_lock() here */
2039 idr_for_each_entry(&connection
->peer_devices
, peer_device
, i
) {
2040 struct drbd_device
*device
= peer_device
->device
;
2041 if (!device
->bitmap
) {
2042 if (drbd_bm_init(device
))
2051 struct crypto_hash
*verify_tfm
;
2052 struct crypto_hash
*csums_tfm
;
2053 struct crypto_hash
*cram_hmac_tfm
;
2054 struct crypto_hash
*integrity_tfm
;
2058 alloc_hash(struct crypto_hash
**tfm
, char *tfm_name
, int err_alg
)
2063 *tfm
= crypto_alloc_hash(tfm_name
, 0, CRYPTO_ALG_ASYNC
);
2072 static enum drbd_ret_code
2073 alloc_crypto(struct crypto
*crypto
, struct net_conf
*new_net_conf
)
2075 char hmac_name
[CRYPTO_MAX_ALG_NAME
];
2076 enum drbd_ret_code rv
;
2078 rv
= alloc_hash(&crypto
->csums_tfm
, new_net_conf
->csums_alg
,
2082 rv
= alloc_hash(&crypto
->verify_tfm
, new_net_conf
->verify_alg
,
2086 rv
= alloc_hash(&crypto
->integrity_tfm
, new_net_conf
->integrity_alg
,
2090 if (new_net_conf
->cram_hmac_alg
[0] != 0) {
2091 snprintf(hmac_name
, CRYPTO_MAX_ALG_NAME
, "hmac(%s)",
2092 new_net_conf
->cram_hmac_alg
);
2094 rv
= alloc_hash(&crypto
->cram_hmac_tfm
, hmac_name
,
2101 static void free_crypto(struct crypto
*crypto
)
2103 crypto_free_hash(crypto
->cram_hmac_tfm
);
2104 crypto_free_hash(crypto
->integrity_tfm
);
2105 crypto_free_hash(crypto
->csums_tfm
);
2106 crypto_free_hash(crypto
->verify_tfm
);
2109 int drbd_adm_net_opts(struct sk_buff
*skb
, struct genl_info
*info
)
2111 struct drbd_config_context adm_ctx
;
2112 enum drbd_ret_code retcode
;
2113 struct drbd_connection
*connection
;
2114 struct net_conf
*old_net_conf
, *new_net_conf
= NULL
;
2116 int ovr
; /* online verify running */
2117 int rsr
; /* re-sync running */
2118 struct crypto crypto
= { };
2120 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_CONNECTION
);
2121 if (!adm_ctx
.reply_skb
)
2123 if (retcode
!= NO_ERROR
)
2126 connection
= adm_ctx
.connection
;
2127 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2129 new_net_conf
= kzalloc(sizeof(struct net_conf
), GFP_KERNEL
);
2130 if (!new_net_conf
) {
2131 retcode
= ERR_NOMEM
;
2135 conn_reconfig_start(connection
);
2137 mutex_lock(&connection
->data
.mutex
);
2138 mutex_lock(&connection
->resource
->conf_update
);
2139 old_net_conf
= connection
->net_conf
;
2141 if (!old_net_conf
) {
2142 drbd_msg_put_info(adm_ctx
.reply_skb
, "net conf missing, try connect");
2143 retcode
= ERR_INVALID_REQUEST
;
2147 *new_net_conf
= *old_net_conf
;
2148 if (should_set_defaults(info
))
2149 set_net_conf_defaults(new_net_conf
);
2151 err
= net_conf_from_attrs_for_change(new_net_conf
, info
);
2152 if (err
&& err
!= -ENOMSG
) {
2153 retcode
= ERR_MANDATORY_TAG
;
2154 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2158 retcode
= check_net_options(connection
, new_net_conf
);
2159 if (retcode
!= NO_ERROR
)
2162 /* re-sync running */
2163 rsr
= conn_resync_running(connection
);
2164 if (rsr
&& strcmp(new_net_conf
->csums_alg
, old_net_conf
->csums_alg
)) {
2165 retcode
= ERR_CSUMS_RESYNC_RUNNING
;
2169 /* online verify running */
2170 ovr
= conn_ov_running(connection
);
2171 if (ovr
&& strcmp(new_net_conf
->verify_alg
, old_net_conf
->verify_alg
)) {
2172 retcode
= ERR_VERIFY_RUNNING
;
2176 retcode
= alloc_crypto(&crypto
, new_net_conf
);
2177 if (retcode
!= NO_ERROR
)
2180 rcu_assign_pointer(connection
->net_conf
, new_net_conf
);
2183 crypto_free_hash(connection
->csums_tfm
);
2184 connection
->csums_tfm
= crypto
.csums_tfm
;
2185 crypto
.csums_tfm
= NULL
;
2188 crypto_free_hash(connection
->verify_tfm
);
2189 connection
->verify_tfm
= crypto
.verify_tfm
;
2190 crypto
.verify_tfm
= NULL
;
2193 crypto_free_hash(connection
->integrity_tfm
);
2194 connection
->integrity_tfm
= crypto
.integrity_tfm
;
2195 if (connection
->cstate
>= C_WF_REPORT_PARAMS
&& connection
->agreed_pro_version
>= 100)
2196 /* Do this without trying to take connection->data.mutex again. */
2197 __drbd_send_protocol(connection
, P_PROTOCOL_UPDATE
);
2199 crypto_free_hash(connection
->cram_hmac_tfm
);
2200 connection
->cram_hmac_tfm
= crypto
.cram_hmac_tfm
;
2202 mutex_unlock(&connection
->resource
->conf_update
);
2203 mutex_unlock(&connection
->data
.mutex
);
2205 kfree(old_net_conf
);
2207 if (connection
->cstate
>= C_WF_REPORT_PARAMS
) {
2208 struct drbd_peer_device
*peer_device
;
2211 idr_for_each_entry(&connection
->peer_devices
, peer_device
, vnr
)
2212 drbd_send_sync_param(peer_device
);
2218 mutex_unlock(&connection
->resource
->conf_update
);
2219 mutex_unlock(&connection
->data
.mutex
);
2220 free_crypto(&crypto
);
2221 kfree(new_net_conf
);
2223 conn_reconfig_done(connection
);
2225 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2227 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2231 int drbd_adm_connect(struct sk_buff
*skb
, struct genl_info
*info
)
2233 struct drbd_config_context adm_ctx
;
2234 struct drbd_peer_device
*peer_device
;
2235 struct net_conf
*old_net_conf
, *new_net_conf
= NULL
;
2236 struct crypto crypto
= { };
2237 struct drbd_resource
*resource
;
2238 struct drbd_connection
*connection
;
2239 enum drbd_ret_code retcode
;
2243 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
2245 if (!adm_ctx
.reply_skb
)
2247 if (retcode
!= NO_ERROR
)
2249 if (!(adm_ctx
.my_addr
&& adm_ctx
.peer_addr
)) {
2250 drbd_msg_put_info(adm_ctx
.reply_skb
, "connection endpoint(s) missing");
2251 retcode
= ERR_INVALID_REQUEST
;
2255 /* No need for _rcu here. All reconfiguration is
2256 * strictly serialized on genl_lock(). We are protected against
2257 * concurrent reconfiguration/addition/deletion */
2258 for_each_resource(resource
, &drbd_resources
) {
2259 for_each_connection(connection
, resource
) {
2260 if (nla_len(adm_ctx
.my_addr
) == connection
->my_addr_len
&&
2261 !memcmp(nla_data(adm_ctx
.my_addr
), &connection
->my_addr
,
2262 connection
->my_addr_len
)) {
2263 retcode
= ERR_LOCAL_ADDR
;
2267 if (nla_len(adm_ctx
.peer_addr
) == connection
->peer_addr_len
&&
2268 !memcmp(nla_data(adm_ctx
.peer_addr
), &connection
->peer_addr
,
2269 connection
->peer_addr_len
)) {
2270 retcode
= ERR_PEER_ADDR
;
2276 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2277 connection
= first_connection(adm_ctx
.resource
);
2278 conn_reconfig_start(connection
);
2280 if (connection
->cstate
> C_STANDALONE
) {
2281 retcode
= ERR_NET_CONFIGURED
;
2285 /* allocation not in the IO path, drbdsetup / netlink process context */
2286 new_net_conf
= kzalloc(sizeof(*new_net_conf
), GFP_KERNEL
);
2287 if (!new_net_conf
) {
2288 retcode
= ERR_NOMEM
;
2292 set_net_conf_defaults(new_net_conf
);
2294 err
= net_conf_from_attrs(new_net_conf
, info
);
2295 if (err
&& err
!= -ENOMSG
) {
2296 retcode
= ERR_MANDATORY_TAG
;
2297 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2301 retcode
= check_net_options(connection
, new_net_conf
);
2302 if (retcode
!= NO_ERROR
)
2305 retcode
= alloc_crypto(&crypto
, new_net_conf
);
2306 if (retcode
!= NO_ERROR
)
2309 ((char *)new_net_conf
->shared_secret
)[SHARED_SECRET_MAX
-1] = 0;
2311 drbd_flush_workqueue(&connection
->sender_work
);
2313 mutex_lock(&adm_ctx
.resource
->conf_update
);
2314 old_net_conf
= connection
->net_conf
;
2316 retcode
= ERR_NET_CONFIGURED
;
2317 mutex_unlock(&adm_ctx
.resource
->conf_update
);
2320 rcu_assign_pointer(connection
->net_conf
, new_net_conf
);
2322 conn_free_crypto(connection
);
2323 connection
->cram_hmac_tfm
= crypto
.cram_hmac_tfm
;
2324 connection
->integrity_tfm
= crypto
.integrity_tfm
;
2325 connection
->csums_tfm
= crypto
.csums_tfm
;
2326 connection
->verify_tfm
= crypto
.verify_tfm
;
2328 connection
->my_addr_len
= nla_len(adm_ctx
.my_addr
);
2329 memcpy(&connection
->my_addr
, nla_data(adm_ctx
.my_addr
), connection
->my_addr_len
);
2330 connection
->peer_addr_len
= nla_len(adm_ctx
.peer_addr
);
2331 memcpy(&connection
->peer_addr
, nla_data(adm_ctx
.peer_addr
), connection
->peer_addr_len
);
2333 mutex_unlock(&adm_ctx
.resource
->conf_update
);
2336 idr_for_each_entry(&connection
->peer_devices
, peer_device
, i
) {
2337 struct drbd_device
*device
= peer_device
->device
;
2338 device
->send_cnt
= 0;
2339 device
->recv_cnt
= 0;
2343 retcode
= conn_request_state(connection
, NS(conn
, C_UNCONNECTED
), CS_VERBOSE
);
2345 conn_reconfig_done(connection
);
2346 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2347 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2351 free_crypto(&crypto
);
2352 kfree(new_net_conf
);
2354 conn_reconfig_done(connection
);
2355 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2357 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2361 static enum drbd_state_rv
conn_try_disconnect(struct drbd_connection
*connection
, bool force
)
2363 enum drbd_state_rv rv
;
2365 rv
= conn_request_state(connection
, NS(conn
, C_DISCONNECTING
),
2366 force
? CS_HARD
: 0);
2369 case SS_NOTHING_TO_DO
:
2371 case SS_ALREADY_STANDALONE
:
2373 case SS_PRIMARY_NOP
:
2374 /* Our state checking code wants to see the peer outdated. */
2375 rv
= conn_request_state(connection
, NS2(conn
, C_DISCONNECTING
, pdsk
, D_OUTDATED
), 0);
2377 if (rv
== SS_OUTDATE_WO_CONN
) /* lost connection before graceful disconnect succeeded */
2378 rv
= conn_request_state(connection
, NS(conn
, C_DISCONNECTING
), CS_VERBOSE
);
2381 case SS_CW_FAILED_BY_PEER
:
2382 /* The peer probably wants to see us outdated. */
2383 rv
= conn_request_state(connection
, NS2(conn
, C_DISCONNECTING
,
2384 disk
, D_OUTDATED
), 0);
2385 if (rv
== SS_IS_DISKLESS
|| rv
== SS_LOWER_THAN_OUTDATED
) {
2386 rv
= conn_request_state(connection
, NS(conn
, C_DISCONNECTING
),
2391 /* no special handling necessary */
2394 if (rv
>= SS_SUCCESS
) {
2395 enum drbd_state_rv rv2
;
2396 /* No one else can reconfigure the network while I am here.
2397 * The state handling only uses drbd_thread_stop_nowait(),
2398 * we want to really wait here until the receiver is no more.
2400 drbd_thread_stop(&connection
->receiver
);
2402 /* Race breaker. This additional state change request may be
2403 * necessary, if this was a forced disconnect during a receiver
2404 * restart. We may have "killed" the receiver thread just
2405 * after drbd_receiver() returned. Typically, we should be
2406 * C_STANDALONE already, now, and this becomes a no-op.
2408 rv2
= conn_request_state(connection
, NS(conn
, C_STANDALONE
),
2409 CS_VERBOSE
| CS_HARD
);
2410 if (rv2
< SS_SUCCESS
)
2411 drbd_err(connection
,
2412 "unexpected rv2=%d in conn_try_disconnect()\n",
2418 int drbd_adm_disconnect(struct sk_buff
*skb
, struct genl_info
*info
)
2420 struct drbd_config_context adm_ctx
;
2421 struct disconnect_parms parms
;
2422 struct drbd_connection
*connection
;
2423 enum drbd_state_rv rv
;
2424 enum drbd_ret_code retcode
;
2427 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_CONNECTION
);
2428 if (!adm_ctx
.reply_skb
)
2430 if (retcode
!= NO_ERROR
)
2433 connection
= adm_ctx
.connection
;
2434 memset(&parms
, 0, sizeof(parms
));
2435 if (info
->attrs
[DRBD_NLA_DISCONNECT_PARMS
]) {
2436 err
= disconnect_parms_from_attrs(&parms
, info
);
2438 retcode
= ERR_MANDATORY_TAG
;
2439 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2444 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2445 rv
= conn_try_disconnect(connection
, parms
.force_disconnect
);
2446 if (rv
< SS_SUCCESS
)
2447 retcode
= rv
; /* FIXME: Type mismatch. */
2450 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2452 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2456 void resync_after_online_grow(struct drbd_device
*device
)
2458 int iass
; /* I am sync source */
2460 drbd_info(device
, "Resync of new storage after online grow\n");
2461 if (device
->state
.role
!= device
->state
.peer
)
2462 iass
= (device
->state
.role
== R_PRIMARY
);
2464 iass
= test_bit(RESOLVE_CONFLICTS
, &first_peer_device(device
)->connection
->flags
);
2467 drbd_start_resync(device
, C_SYNC_SOURCE
);
2469 _drbd_request_state(device
, NS(conn
, C_WF_SYNC_UUID
), CS_VERBOSE
+ CS_SERIALIZE
);
2472 int drbd_adm_resize(struct sk_buff
*skb
, struct genl_info
*info
)
2474 struct drbd_config_context adm_ctx
;
2475 struct disk_conf
*old_disk_conf
, *new_disk_conf
= NULL
;
2476 struct resize_parms rs
;
2477 struct drbd_device
*device
;
2478 enum drbd_ret_code retcode
;
2479 enum determine_dev_size dd
;
2480 bool change_al_layout
= false;
2481 enum dds_flags ddsf
;
2485 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2486 if (!adm_ctx
.reply_skb
)
2488 if (retcode
!= NO_ERROR
)
2491 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2492 device
= adm_ctx
.device
;
2493 if (!get_ldev(device
)) {
2494 retcode
= ERR_NO_DISK
;
2498 memset(&rs
, 0, sizeof(struct resize_parms
));
2499 rs
.al_stripes
= device
->ldev
->md
.al_stripes
;
2500 rs
.al_stripe_size
= device
->ldev
->md
.al_stripe_size_4k
* 4;
2501 if (info
->attrs
[DRBD_NLA_RESIZE_PARMS
]) {
2502 err
= resize_parms_from_attrs(&rs
, info
);
2504 retcode
= ERR_MANDATORY_TAG
;
2505 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2510 if (device
->state
.conn
> C_CONNECTED
) {
2511 retcode
= ERR_RESIZE_RESYNC
;
2515 if (device
->state
.role
== R_SECONDARY
&&
2516 device
->state
.peer
== R_SECONDARY
) {
2517 retcode
= ERR_NO_PRIMARY
;
2521 if (rs
.no_resync
&& first_peer_device(device
)->connection
->agreed_pro_version
< 93) {
2522 retcode
= ERR_NEED_APV_93
;
2527 u_size
= rcu_dereference(device
->ldev
->disk_conf
)->disk_size
;
2529 if (u_size
!= (sector_t
)rs
.resize_size
) {
2530 new_disk_conf
= kmalloc(sizeof(struct disk_conf
), GFP_KERNEL
);
2531 if (!new_disk_conf
) {
2532 retcode
= ERR_NOMEM
;
2537 if (device
->ldev
->md
.al_stripes
!= rs
.al_stripes
||
2538 device
->ldev
->md
.al_stripe_size_4k
!= rs
.al_stripe_size
/ 4) {
2539 u32 al_size_k
= rs
.al_stripes
* rs
.al_stripe_size
;
2541 if (al_size_k
> (16 * 1024 * 1024)) {
2542 retcode
= ERR_MD_LAYOUT_TOO_BIG
;
2546 if (al_size_k
< MD_32kB_SECT
/2) {
2547 retcode
= ERR_MD_LAYOUT_TOO_SMALL
;
2551 if (device
->state
.conn
!= C_CONNECTED
&& !rs
.resize_force
) {
2552 retcode
= ERR_MD_LAYOUT_CONNECTED
;
2556 change_al_layout
= true;
2559 if (device
->ldev
->known_size
!= drbd_get_capacity(device
->ldev
->backing_bdev
))
2560 device
->ldev
->known_size
= drbd_get_capacity(device
->ldev
->backing_bdev
);
2562 if (new_disk_conf
) {
2563 mutex_lock(&device
->resource
->conf_update
);
2564 old_disk_conf
= device
->ldev
->disk_conf
;
2565 *new_disk_conf
= *old_disk_conf
;
2566 new_disk_conf
->disk_size
= (sector_t
)rs
.resize_size
;
2567 rcu_assign_pointer(device
->ldev
->disk_conf
, new_disk_conf
);
2568 mutex_unlock(&device
->resource
->conf_update
);
2570 kfree(old_disk_conf
);
2573 ddsf
= (rs
.resize_force
? DDSF_FORCED
: 0) | (rs
.no_resync
? DDSF_NO_RESYNC
: 0);
2574 dd
= drbd_determine_dev_size(device
, ddsf
, change_al_layout
? &rs
: NULL
);
2575 drbd_md_sync(device
);
2577 if (dd
== DS_ERROR
) {
2578 retcode
= ERR_NOMEM_BITMAP
;
2580 } else if (dd
== DS_ERROR_SPACE_MD
) {
2581 retcode
= ERR_MD_LAYOUT_NO_FIT
;
2583 } else if (dd
== DS_ERROR_SHRINK
) {
2584 retcode
= ERR_IMPLICIT_SHRINK
;
2588 if (device
->state
.conn
== C_CONNECTED
) {
2590 set_bit(RESIZE_PENDING
, &device
->flags
);
2592 drbd_send_uuids(first_peer_device(device
));
2593 drbd_send_sizes(first_peer_device(device
), 1, ddsf
);
2597 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2599 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2607 int drbd_adm_resource_opts(struct sk_buff
*skb
, struct genl_info
*info
)
2609 struct drbd_config_context adm_ctx
;
2610 enum drbd_ret_code retcode
;
2611 struct res_opts res_opts
;
2614 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
2615 if (!adm_ctx
.reply_skb
)
2617 if (retcode
!= NO_ERROR
)
2620 res_opts
= adm_ctx
.resource
->res_opts
;
2621 if (should_set_defaults(info
))
2622 set_res_opts_defaults(&res_opts
);
2624 err
= res_opts_from_attrs(&res_opts
, info
);
2625 if (err
&& err
!= -ENOMSG
) {
2626 retcode
= ERR_MANDATORY_TAG
;
2627 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
2631 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2632 err
= set_resource_options(adm_ctx
.resource
, &res_opts
);
2634 retcode
= ERR_INVALID_REQUEST
;
2636 retcode
= ERR_NOMEM
;
2638 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2641 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2645 int drbd_adm_invalidate(struct sk_buff
*skb
, struct genl_info
*info
)
2647 struct drbd_config_context adm_ctx
;
2648 struct drbd_device
*device
;
2649 int retcode
; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2651 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2652 if (!adm_ctx
.reply_skb
)
2654 if (retcode
!= NO_ERROR
)
2657 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2658 device
= adm_ctx
.device
;
2660 /* If there is still bitmap IO pending, probably because of a previous
2661 * resync just being finished, wait for it before requesting a new resync.
2662 * Also wait for it's after_state_ch(). */
2663 drbd_suspend_io(device
);
2664 wait_event(device
->misc_wait
, !test_bit(BITMAP_IO
, &device
->flags
));
2665 drbd_flush_workqueue(&first_peer_device(device
)->connection
->sender_work
);
2667 /* If we happen to be C_STANDALONE R_SECONDARY, just change to
2668 * D_INCONSISTENT, and set all bits in the bitmap. Otherwise,
2669 * try to start a resync handshake as sync target for full sync.
2671 if (device
->state
.conn
== C_STANDALONE
&& device
->state
.role
== R_SECONDARY
) {
2672 retcode
= drbd_request_state(device
, NS(disk
, D_INCONSISTENT
));
2673 if (retcode
>= SS_SUCCESS
) {
2674 if (drbd_bitmap_io(device
, &drbd_bmio_set_n_write
,
2675 "set_n_write from invalidate", BM_LOCKED_MASK
))
2676 retcode
= ERR_IO_MD_DISK
;
2679 retcode
= drbd_request_state(device
, NS(conn
, C_STARTING_SYNC_T
));
2680 drbd_resume_io(device
);
2681 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2683 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2687 static int drbd_adm_simple_request_state(struct sk_buff
*skb
, struct genl_info
*info
,
2688 union drbd_state mask
, union drbd_state val
)
2690 struct drbd_config_context adm_ctx
;
2691 enum drbd_ret_code retcode
;
2693 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2694 if (!adm_ctx
.reply_skb
)
2696 if (retcode
!= NO_ERROR
)
2699 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2700 retcode
= drbd_request_state(adm_ctx
.device
, mask
, val
);
2701 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2703 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2707 static int drbd_bmio_set_susp_al(struct drbd_device
*device
)
2711 rv
= drbd_bmio_set_n_write(device
);
2712 drbd_suspend_al(device
);
2716 int drbd_adm_invalidate_peer(struct sk_buff
*skb
, struct genl_info
*info
)
2718 struct drbd_config_context adm_ctx
;
2719 int retcode
; /* drbd_ret_code, drbd_state_rv */
2720 struct drbd_device
*device
;
2722 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2723 if (!adm_ctx
.reply_skb
)
2725 if (retcode
!= NO_ERROR
)
2728 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2729 device
= adm_ctx
.device
;
2731 /* If there is still bitmap IO pending, probably because of a previous
2732 * resync just being finished, wait for it before requesting a new resync.
2733 * Also wait for it's after_state_ch(). */
2734 drbd_suspend_io(device
);
2735 wait_event(device
->misc_wait
, !test_bit(BITMAP_IO
, &device
->flags
));
2736 drbd_flush_workqueue(&first_peer_device(device
)->connection
->sender_work
);
2738 /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
2739 * in the bitmap. Otherwise, try to start a resync handshake
2740 * as sync source for full sync.
2742 if (device
->state
.conn
== C_STANDALONE
&& device
->state
.role
== R_PRIMARY
) {
2743 /* The peer will get a resync upon connect anyways. Just make that
2744 into a full resync. */
2745 retcode
= drbd_request_state(device
, NS(pdsk
, D_INCONSISTENT
));
2746 if (retcode
>= SS_SUCCESS
) {
2747 if (drbd_bitmap_io(device
, &drbd_bmio_set_susp_al
,
2748 "set_n_write from invalidate_peer",
2749 BM_LOCKED_SET_ALLOWED
))
2750 retcode
= ERR_IO_MD_DISK
;
2753 retcode
= drbd_request_state(device
, NS(conn
, C_STARTING_SYNC_S
));
2754 drbd_resume_io(device
);
2755 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2757 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2761 int drbd_adm_pause_sync(struct sk_buff
*skb
, struct genl_info
*info
)
2763 struct drbd_config_context adm_ctx
;
2764 enum drbd_ret_code retcode
;
2766 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2767 if (!adm_ctx
.reply_skb
)
2769 if (retcode
!= NO_ERROR
)
2772 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2773 if (drbd_request_state(adm_ctx
.device
, NS(user_isp
, 1)) == SS_NOTHING_TO_DO
)
2774 retcode
= ERR_PAUSE_IS_SET
;
2775 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2777 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2781 int drbd_adm_resume_sync(struct sk_buff
*skb
, struct genl_info
*info
)
2783 struct drbd_config_context adm_ctx
;
2784 union drbd_dev_state s
;
2785 enum drbd_ret_code retcode
;
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 if (drbd_request_state(adm_ctx
.device
, NS(user_isp
, 0)) == SS_NOTHING_TO_DO
) {
2795 s
= adm_ctx
.device
->state
;
2796 if (s
.conn
== C_PAUSED_SYNC_S
|| s
.conn
== C_PAUSED_SYNC_T
) {
2797 retcode
= s
.aftr_isp
? ERR_PIC_AFTER_DEP
:
2798 s
.peer_isp
? ERR_PIC_PEER_DEP
: ERR_PAUSE_IS_CLEAR
;
2800 retcode
= ERR_PAUSE_IS_CLEAR
;
2803 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2805 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2809 int drbd_adm_suspend_io(struct sk_buff
*skb
, struct genl_info
*info
)
2811 return drbd_adm_simple_request_state(skb
, info
, NS(susp
, 1));
2814 int drbd_adm_resume_io(struct sk_buff
*skb
, struct genl_info
*info
)
2816 struct drbd_config_context adm_ctx
;
2817 struct drbd_device
*device
;
2818 int retcode
; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2820 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
2821 if (!adm_ctx
.reply_skb
)
2823 if (retcode
!= NO_ERROR
)
2826 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
2827 device
= adm_ctx
.device
;
2828 if (test_bit(NEW_CUR_UUID
, &device
->flags
)) {
2829 drbd_uuid_new_current(device
);
2830 clear_bit(NEW_CUR_UUID
, &device
->flags
);
2832 drbd_suspend_io(device
);
2833 retcode
= drbd_request_state(device
, NS3(susp
, 0, susp_nod
, 0, susp_fen
, 0));
2834 if (retcode
== SS_SUCCESS
) {
2835 if (device
->state
.conn
< C_CONNECTED
)
2836 tl_clear(first_peer_device(device
)->connection
);
2837 if (device
->state
.disk
== D_DISKLESS
|| device
->state
.disk
== D_FAILED
)
2838 tl_restart(first_peer_device(device
)->connection
, FAIL_FROZEN_DISK_IO
);
2840 drbd_resume_io(device
);
2841 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
2843 drbd_adm_finish(&adm_ctx
, info
, retcode
);
2847 int drbd_adm_outdate(struct sk_buff
*skb
, struct genl_info
*info
)
2849 return drbd_adm_simple_request_state(skb
, info
, NS(disk
, D_OUTDATED
));
2852 static int nla_put_drbd_cfg_context(struct sk_buff
*skb
,
2853 struct drbd_resource
*resource
,
2854 struct drbd_connection
*connection
,
2855 struct drbd_device
*device
)
2858 nla
= nla_nest_start(skb
, DRBD_NLA_CFG_CONTEXT
);
2860 goto nla_put_failure
;
2862 nla_put_u32(skb
, T_ctx_volume
, device
->vnr
))
2863 goto nla_put_failure
;
2864 if (nla_put_string(skb
, T_ctx_resource_name
, resource
->name
))
2865 goto nla_put_failure
;
2867 if (connection
->my_addr_len
&&
2868 nla_put(skb
, T_ctx_my_addr
, connection
->my_addr_len
, &connection
->my_addr
))
2869 goto nla_put_failure
;
2870 if (connection
->peer_addr_len
&&
2871 nla_put(skb
, T_ctx_peer_addr
, connection
->peer_addr_len
, &connection
->peer_addr
))
2872 goto nla_put_failure
;
2874 nla_nest_end(skb
, nla
);
2879 nla_nest_cancel(skb
, nla
);
2884 * Return the connection of @resource if @resource has exactly one connection.
2886 static struct drbd_connection
*the_only_connection(struct drbd_resource
*resource
)
2888 struct list_head
*connections
= &resource
->connections
;
2890 if (list_empty(connections
) || connections
->next
->next
!= connections
)
2892 return list_first_entry(&resource
->connections
, struct drbd_connection
, connections
);
2895 int nla_put_status_info(struct sk_buff
*skb
, struct drbd_device
*device
,
2896 const struct sib_info
*sib
)
2898 struct drbd_resource
*resource
= device
->resource
;
2899 struct state_info
*si
= NULL
; /* for sizeof(si->member); */
2903 int exclude_sensitive
;
2905 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2906 * to. So we better exclude_sensitive information.
2908 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2909 * in the context of the requesting user process. Exclude sensitive
2910 * information, unless current has superuser.
2912 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2913 * relies on the current implementation of netlink_dump(), which
2914 * executes the dump callback successively from netlink_recvmsg(),
2915 * always in the context of the receiving process */
2916 exclude_sensitive
= sib
|| !capable(CAP_SYS_ADMIN
);
2918 got_ldev
= get_ldev(device
);
2920 /* We need to add connection name and volume number information still.
2921 * Minor number is in drbd_genlmsghdr. */
2922 if (nla_put_drbd_cfg_context(skb
, resource
, the_only_connection(resource
), device
))
2923 goto nla_put_failure
;
2925 if (res_opts_to_skb(skb
, &device
->resource
->res_opts
, exclude_sensitive
))
2926 goto nla_put_failure
;
2930 struct disk_conf
*disk_conf
;
2932 disk_conf
= rcu_dereference(device
->ldev
->disk_conf
);
2933 err
= disk_conf_to_skb(skb
, disk_conf
, exclude_sensitive
);
2936 struct net_conf
*nc
;
2938 nc
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
);
2940 err
= net_conf_to_skb(skb
, nc
, exclude_sensitive
);
2944 goto nla_put_failure
;
2946 nla
= nla_nest_start(skb
, DRBD_NLA_STATE_INFO
);
2948 goto nla_put_failure
;
2949 if (nla_put_u32(skb
, T_sib_reason
, sib
? sib
->sib_reason
: SIB_GET_STATUS_REPLY
) ||
2950 nla_put_u32(skb
, T_current_state
, device
->state
.i
) ||
2951 nla_put_u64(skb
, T_ed_uuid
, device
->ed_uuid
) ||
2952 nla_put_u64(skb
, T_capacity
, drbd_get_capacity(device
->this_bdev
)) ||
2953 nla_put_u64(skb
, T_send_cnt
, device
->send_cnt
) ||
2954 nla_put_u64(skb
, T_recv_cnt
, device
->recv_cnt
) ||
2955 nla_put_u64(skb
, T_read_cnt
, device
->read_cnt
) ||
2956 nla_put_u64(skb
, T_writ_cnt
, device
->writ_cnt
) ||
2957 nla_put_u64(skb
, T_al_writ_cnt
, device
->al_writ_cnt
) ||
2958 nla_put_u64(skb
, T_bm_writ_cnt
, device
->bm_writ_cnt
) ||
2959 nla_put_u32(skb
, T_ap_bio_cnt
, atomic_read(&device
->ap_bio_cnt
)) ||
2960 nla_put_u32(skb
, T_ap_pending_cnt
, atomic_read(&device
->ap_pending_cnt
)) ||
2961 nla_put_u32(skb
, T_rs_pending_cnt
, atomic_read(&device
->rs_pending_cnt
)))
2962 goto nla_put_failure
;
2967 spin_lock_irq(&device
->ldev
->md
.uuid_lock
);
2968 err
= nla_put(skb
, T_uuids
, sizeof(si
->uuids
), device
->ldev
->md
.uuid
);
2969 spin_unlock_irq(&device
->ldev
->md
.uuid_lock
);
2972 goto nla_put_failure
;
2974 if (nla_put_u32(skb
, T_disk_flags
, device
->ldev
->md
.flags
) ||
2975 nla_put_u64(skb
, T_bits_total
, drbd_bm_bits(device
)) ||
2976 nla_put_u64(skb
, T_bits_oos
, drbd_bm_total_weight(device
)))
2977 goto nla_put_failure
;
2978 if (C_SYNC_SOURCE
<= device
->state
.conn
&&
2979 C_PAUSED_SYNC_T
>= device
->state
.conn
) {
2980 if (nla_put_u64(skb
, T_bits_rs_total
, device
->rs_total
) ||
2981 nla_put_u64(skb
, T_bits_rs_failed
, device
->rs_failed
))
2982 goto nla_put_failure
;
2987 switch(sib
->sib_reason
) {
2988 case SIB_SYNC_PROGRESS
:
2989 case SIB_GET_STATUS_REPLY
:
2991 case SIB_STATE_CHANGE
:
2992 if (nla_put_u32(skb
, T_prev_state
, sib
->os
.i
) ||
2993 nla_put_u32(skb
, T_new_state
, sib
->ns
.i
))
2994 goto nla_put_failure
;
2996 case SIB_HELPER_POST
:
2997 if (nla_put_u32(skb
, T_helper_exit_code
,
2998 sib
->helper_exit_code
))
2999 goto nla_put_failure
;
3001 case SIB_HELPER_PRE
:
3002 if (nla_put_string(skb
, T_helper
, sib
->helper_name
))
3003 goto nla_put_failure
;
3007 nla_nest_end(skb
, nla
);
3017 int drbd_adm_get_status(struct sk_buff
*skb
, struct genl_info
*info
)
3019 struct drbd_config_context adm_ctx
;
3020 enum drbd_ret_code retcode
;
3023 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3024 if (!adm_ctx
.reply_skb
)
3026 if (retcode
!= NO_ERROR
)
3029 err
= nla_put_status_info(adm_ctx
.reply_skb
, adm_ctx
.device
, NULL
);
3031 nlmsg_free(adm_ctx
.reply_skb
);
3035 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3039 static int get_one_status(struct sk_buff
*skb
, struct netlink_callback
*cb
)
3041 struct drbd_device
*device
;
3042 struct drbd_genlmsghdr
*dh
;
3043 struct drbd_resource
*pos
= (struct drbd_resource
*)cb
->args
[0];
3044 struct drbd_resource
*resource
= NULL
;
3045 struct drbd_resource
*tmp
;
3046 unsigned volume
= cb
->args
[1];
3048 /* Open coded, deferred, iteration:
3049 * for_each_resource_safe(resource, tmp, &drbd_resources) {
3050 * connection = "first connection of resource or undefined";
3051 * idr_for_each_entry(&resource->devices, device, i) {
3055 * where resource is cb->args[0];
3056 * and i is cb->args[1];
3058 * cb->args[2] indicates if we shall loop over all resources,
3059 * or just dump all volumes of a single resource.
3061 * This may miss entries inserted after this dump started,
3062 * or entries deleted before they are reached.
3064 * We need to make sure the device won't disappear while
3065 * we are looking at it, and revalidate our iterators
3066 * on each iteration.
3069 /* synchronize with conn_create()/drbd_destroy_connection() */
3071 /* revalidate iterator position */
3072 for_each_resource_rcu(tmp
, &drbd_resources
) {
3074 /* first iteration */
3086 device
= idr_get_next(&resource
->devices
, &volume
);
3088 /* No more volumes to dump on this resource.
3089 * Advance resource iterator. */
3090 pos
= list_entry_rcu(resource
->resources
.next
,
3091 struct drbd_resource
, resources
);
3092 /* Did we dump any volume of this resource yet? */
3094 /* If we reached the end of the list,
3095 * or only a single resource dump was requested,
3097 if (&pos
->resources
== &drbd_resources
|| cb
->args
[2])
3105 dh
= genlmsg_put(skb
, NETLINK_CB(cb
->skb
).portid
,
3106 cb
->nlh
->nlmsg_seq
, &drbd_genl_family
,
3107 NLM_F_MULTI
, DRBD_ADM_GET_STATUS
);
3112 /* This is a connection without a single volume.
3113 * Suprisingly enough, it may have a network
3115 struct drbd_connection
*connection
;
3118 dh
->ret_code
= NO_ERROR
;
3119 connection
= the_only_connection(resource
);
3120 if (nla_put_drbd_cfg_context(skb
, resource
, connection
, NULL
))
3123 struct net_conf
*nc
;
3125 nc
= rcu_dereference(connection
->net_conf
);
3126 if (nc
&& net_conf_to_skb(skb
, nc
, 1) != 0)
3132 D_ASSERT(device
, device
->vnr
== volume
);
3133 D_ASSERT(device
, device
->resource
== resource
);
3135 dh
->minor
= device_to_minor(device
);
3136 dh
->ret_code
= NO_ERROR
;
3138 if (nla_put_status_info(skb
, device
, NULL
)) {
3140 genlmsg_cancel(skb
, dh
);
3144 genlmsg_end(skb
, dh
);
3149 /* where to start the next iteration */
3150 cb
->args
[0] = (long)pos
;
3151 cb
->args
[1] = (pos
== resource
) ? volume
+ 1 : 0;
3153 /* No more resources/volumes/minors found results in an empty skb.
3154 * Which will terminate the dump. */
3159 * Request status of all resources, or of all volumes within a single resource.
3161 * This is a dump, as the answer may not fit in a single reply skb otherwise.
3162 * Which means we cannot use the family->attrbuf or other such members, because
3163 * dump is NOT protected by the genl_lock(). During dump, we only have access
3164 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
3166 * Once things are setup properly, we call into get_one_status().
3168 int drbd_adm_get_status_all(struct sk_buff
*skb
, struct netlink_callback
*cb
)
3170 const unsigned hdrlen
= GENL_HDRLEN
+ GENL_MAGIC_FAMILY_HDRSZ
;
3172 const char *resource_name
;
3173 struct drbd_resource
*resource
;
3176 /* Is this a followup call? */
3178 /* ... of a single resource dump,
3179 * and the resource iterator has been advanced already? */
3180 if (cb
->args
[2] && cb
->args
[2] != cb
->args
[0])
3181 return 0; /* DONE. */
3185 /* First call (from netlink_dump_start). We need to figure out
3186 * which resource(s) the user wants us to dump. */
3187 nla
= nla_find(nlmsg_attrdata(cb
->nlh
, hdrlen
),
3188 nlmsg_attrlen(cb
->nlh
, hdrlen
),
3189 DRBD_NLA_CFG_CONTEXT
);
3191 /* No explicit context given. Dump all. */
3194 maxtype
= ARRAY_SIZE(drbd_cfg_context_nl_policy
) - 1;
3195 nla
= drbd_nla_find_nested(maxtype
, nla
, __nla_type(T_ctx_resource_name
));
3197 return PTR_ERR(nla
);
3198 /* context given, but no name present? */
3201 resource_name
= nla_data(nla
);
3202 if (!*resource_name
)
3204 resource
= drbd_find_resource(resource_name
);
3208 kref_put(&resource
->kref
, drbd_destroy_resource
); /* get_one_status() revalidates the resource */
3210 /* prime iterators, and set "filter" mode mark:
3211 * only dump this connection. */
3212 cb
->args
[0] = (long)resource
;
3213 /* cb->args[1] = 0; passed in this way. */
3214 cb
->args
[2] = (long)resource
;
3217 return get_one_status(skb
, cb
);
3220 int drbd_adm_get_timeout_type(struct sk_buff
*skb
, struct genl_info
*info
)
3222 struct drbd_config_context adm_ctx
;
3223 enum drbd_ret_code retcode
;
3224 struct timeout_parms tp
;
3227 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3228 if (!adm_ctx
.reply_skb
)
3230 if (retcode
!= NO_ERROR
)
3234 adm_ctx
.device
->state
.pdsk
== D_OUTDATED
? UT_PEER_OUTDATED
:
3235 test_bit(USE_DEGR_WFC_T
, &adm_ctx
.device
->flags
) ? UT_DEGRADED
:
3238 err
= timeout_parms_to_priv_skb(adm_ctx
.reply_skb
, &tp
);
3240 nlmsg_free(adm_ctx
.reply_skb
);
3244 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3248 int drbd_adm_start_ov(struct sk_buff
*skb
, struct genl_info
*info
)
3250 struct drbd_config_context adm_ctx
;
3251 struct drbd_device
*device
;
3252 enum drbd_ret_code retcode
;
3253 struct start_ov_parms parms
;
3255 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3256 if (!adm_ctx
.reply_skb
)
3258 if (retcode
!= NO_ERROR
)
3261 device
= adm_ctx
.device
;
3263 /* resume from last known position, if possible */
3264 parms
.ov_start_sector
= device
->ov_start_sector
;
3265 parms
.ov_stop_sector
= ULLONG_MAX
;
3266 if (info
->attrs
[DRBD_NLA_START_OV_PARMS
]) {
3267 int err
= start_ov_parms_from_attrs(&parms
, info
);
3269 retcode
= ERR_MANDATORY_TAG
;
3270 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
3274 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
3276 /* w_make_ov_request expects position to be aligned */
3277 device
->ov_start_sector
= parms
.ov_start_sector
& ~(BM_SECT_PER_BIT
-1);
3278 device
->ov_stop_sector
= parms
.ov_stop_sector
;
3280 /* If there is still bitmap IO pending, e.g. previous resync or verify
3281 * just being finished, wait for it before requesting a new resync. */
3282 drbd_suspend_io(device
);
3283 wait_event(device
->misc_wait
, !test_bit(BITMAP_IO
, &device
->flags
));
3284 retcode
= drbd_request_state(device
, NS(conn
, C_VERIFY_S
));
3285 drbd_resume_io(device
);
3287 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
3289 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3294 int drbd_adm_new_c_uuid(struct sk_buff
*skb
, struct genl_info
*info
)
3296 struct drbd_config_context adm_ctx
;
3297 struct drbd_device
*device
;
3298 enum drbd_ret_code retcode
;
3299 int skip_initial_sync
= 0;
3301 struct new_c_uuid_parms args
;
3303 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3304 if (!adm_ctx
.reply_skb
)
3306 if (retcode
!= NO_ERROR
)
3309 device
= adm_ctx
.device
;
3310 memset(&args
, 0, sizeof(args
));
3311 if (info
->attrs
[DRBD_NLA_NEW_C_UUID_PARMS
]) {
3312 err
= new_c_uuid_parms_from_attrs(&args
, info
);
3314 retcode
= ERR_MANDATORY_TAG
;
3315 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
3320 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
3321 mutex_lock(device
->state_mutex
); /* Protects us against serialized state changes. */
3323 if (!get_ldev(device
)) {
3324 retcode
= ERR_NO_DISK
;
3328 /* this is "skip initial sync", assume to be clean */
3329 if (device
->state
.conn
== C_CONNECTED
&&
3330 first_peer_device(device
)->connection
->agreed_pro_version
>= 90 &&
3331 device
->ldev
->md
.uuid
[UI_CURRENT
] == UUID_JUST_CREATED
&& args
.clear_bm
) {
3332 drbd_info(device
, "Preparing to skip initial sync\n");
3333 skip_initial_sync
= 1;
3334 } else if (device
->state
.conn
!= C_STANDALONE
) {
3335 retcode
= ERR_CONNECTED
;
3339 drbd_uuid_set(device
, UI_BITMAP
, 0); /* Rotate UI_BITMAP to History 1, etc... */
3340 drbd_uuid_new_current(device
); /* New current, previous to UI_BITMAP */
3342 if (args
.clear_bm
) {
3343 err
= drbd_bitmap_io(device
, &drbd_bmio_clear_n_write
,
3344 "clear_n_write from new_c_uuid", BM_LOCKED_MASK
);
3346 drbd_err(device
, "Writing bitmap failed with %d\n", err
);
3347 retcode
= ERR_IO_MD_DISK
;
3349 if (skip_initial_sync
) {
3350 drbd_send_uuids_skip_initial_sync(first_peer_device(device
));
3351 _drbd_uuid_set(device
, UI_BITMAP
, 0);
3352 drbd_print_uuids(device
, "cleared bitmap UUID");
3353 spin_lock_irq(&device
->resource
->req_lock
);
3354 _drbd_set_state(_NS2(device
, disk
, D_UP_TO_DATE
, pdsk
, D_UP_TO_DATE
),
3356 spin_unlock_irq(&device
->resource
->req_lock
);
3360 drbd_md_sync(device
);
3364 mutex_unlock(device
->state_mutex
);
3365 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
3367 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3371 static enum drbd_ret_code
3372 drbd_check_resource_name(struct drbd_config_context
*adm_ctx
)
3374 const char *name
= adm_ctx
->resource_name
;
3375 if (!name
|| !name
[0]) {
3376 drbd_msg_put_info(adm_ctx
->reply_skb
, "resource name missing");
3377 return ERR_MANDATORY_TAG
;
3379 /* if we want to use these in sysfs/configfs/debugfs some day,
3380 * we must not allow slashes */
3381 if (strchr(name
, '/')) {
3382 drbd_msg_put_info(adm_ctx
->reply_skb
, "invalid resource name");
3383 return ERR_INVALID_REQUEST
;
3388 int drbd_adm_new_resource(struct sk_buff
*skb
, struct genl_info
*info
)
3390 struct drbd_config_context adm_ctx
;
3391 enum drbd_ret_code retcode
;
3392 struct res_opts res_opts
;
3395 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, 0);
3396 if (!adm_ctx
.reply_skb
)
3398 if (retcode
!= NO_ERROR
)
3401 set_res_opts_defaults(&res_opts
);
3402 err
= res_opts_from_attrs(&res_opts
, info
);
3403 if (err
&& err
!= -ENOMSG
) {
3404 retcode
= ERR_MANDATORY_TAG
;
3405 drbd_msg_put_info(adm_ctx
.reply_skb
, from_attrs_err_to_txt(err
));
3409 retcode
= drbd_check_resource_name(&adm_ctx
);
3410 if (retcode
!= NO_ERROR
)
3413 if (adm_ctx
.resource
) {
3414 if (info
->nlhdr
->nlmsg_flags
& NLM_F_EXCL
) {
3415 retcode
= ERR_INVALID_REQUEST
;
3416 drbd_msg_put_info(adm_ctx
.reply_skb
, "resource exists");
3418 /* else: still NO_ERROR */
3422 /* not yet safe for genl_family.parallel_ops */
3423 if (!conn_create(adm_ctx
.resource_name
, &res_opts
))
3424 retcode
= ERR_NOMEM
;
3426 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3430 int drbd_adm_new_minor(struct sk_buff
*skb
, struct genl_info
*info
)
3432 struct drbd_config_context adm_ctx
;
3433 struct drbd_genlmsghdr
*dh
= info
->userhdr
;
3434 enum drbd_ret_code retcode
;
3436 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
3437 if (!adm_ctx
.reply_skb
)
3439 if (retcode
!= NO_ERROR
)
3442 if (dh
->minor
> MINORMASK
) {
3443 drbd_msg_put_info(adm_ctx
.reply_skb
, "requested minor out of range");
3444 retcode
= ERR_INVALID_REQUEST
;
3447 if (adm_ctx
.volume
> DRBD_VOLUME_MAX
) {
3448 drbd_msg_put_info(adm_ctx
.reply_skb
, "requested volume id out of range");
3449 retcode
= ERR_INVALID_REQUEST
;
3453 /* drbd_adm_prepare made sure already
3454 * that first_peer_device(device)->connection and device->vnr match the request. */
3455 if (adm_ctx
.device
) {
3456 if (info
->nlhdr
->nlmsg_flags
& NLM_F_EXCL
)
3457 retcode
= ERR_MINOR_EXISTS
;
3458 /* else: still NO_ERROR */
3462 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
3463 retcode
= drbd_create_device(&adm_ctx
, dh
->minor
);
3464 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
3466 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3470 static enum drbd_ret_code
adm_del_minor(struct drbd_device
*device
)
3472 if (device
->state
.disk
== D_DISKLESS
&&
3473 /* no need to be device->state.conn == C_STANDALONE &&
3474 * we may want to delete a minor from a live replication group.
3476 device
->state
.role
== R_SECONDARY
) {
3477 _drbd_request_state(device
, NS(conn
, C_WF_REPORT_PARAMS
),
3478 CS_VERBOSE
+ CS_WAIT_COMPLETE
);
3479 drbd_delete_device(device
);
3482 return ERR_MINOR_CONFIGURED
;
3485 int drbd_adm_del_minor(struct sk_buff
*skb
, struct genl_info
*info
)
3487 struct drbd_config_context adm_ctx
;
3488 enum drbd_ret_code retcode
;
3490 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_MINOR
);
3491 if (!adm_ctx
.reply_skb
)
3493 if (retcode
!= NO_ERROR
)
3496 mutex_lock(&adm_ctx
.resource
->adm_mutex
);
3497 retcode
= adm_del_minor(adm_ctx
.device
);
3498 mutex_unlock(&adm_ctx
.resource
->adm_mutex
);
3500 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3504 int drbd_adm_down(struct sk_buff
*skb
, struct genl_info
*info
)
3506 struct drbd_config_context adm_ctx
;
3507 struct drbd_resource
*resource
;
3508 struct drbd_connection
*connection
;
3509 struct drbd_device
*device
;
3510 int retcode
; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3513 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
3514 if (!adm_ctx
.reply_skb
)
3516 if (retcode
!= NO_ERROR
)
3519 resource
= adm_ctx
.resource
;
3520 mutex_lock(&resource
->adm_mutex
);
3522 for_each_connection(connection
, resource
) {
3523 struct drbd_peer_device
*peer_device
;
3525 idr_for_each_entry(&connection
->peer_devices
, peer_device
, i
) {
3526 retcode
= drbd_set_role(peer_device
->device
, R_SECONDARY
, 0);
3527 if (retcode
< SS_SUCCESS
) {
3528 drbd_msg_put_info(adm_ctx
.reply_skb
, "failed to demote");
3533 retcode
= conn_try_disconnect(connection
, 0);
3534 if (retcode
< SS_SUCCESS
) {
3535 drbd_msg_put_info(adm_ctx
.reply_skb
, "failed to disconnect");
3541 idr_for_each_entry(&resource
->devices
, device
, i
) {
3542 retcode
= adm_detach(device
, 0);
3543 if (retcode
< SS_SUCCESS
|| retcode
> NO_ERROR
) {
3544 drbd_msg_put_info(adm_ctx
.reply_skb
, "failed to detach");
3549 /* If we reach this, all volumes (of this connection) are Secondary,
3550 * Disconnected, Diskless, aka Unconfigured. Make sure all threads have
3551 * actually stopped, state handling only does drbd_thread_stop_nowait(). */
3552 for_each_connection(connection
, resource
)
3553 drbd_thread_stop(&connection
->worker
);
3555 /* Now, nothing can fail anymore */
3557 /* delete volumes */
3558 idr_for_each_entry(&resource
->devices
, device
, i
) {
3559 retcode
= adm_del_minor(device
);
3560 if (retcode
!= NO_ERROR
) {
3561 /* "can not happen" */
3562 drbd_msg_put_info(adm_ctx
.reply_skb
, "failed to delete volume");
3567 list_del_rcu(&resource
->resources
);
3569 drbd_free_resource(resource
);
3572 mutex_unlock(&resource
->adm_mutex
);
3574 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3578 int drbd_adm_del_resource(struct sk_buff
*skb
, struct genl_info
*info
)
3580 struct drbd_config_context adm_ctx
;
3581 struct drbd_resource
*resource
;
3582 struct drbd_connection
*connection
;
3583 enum drbd_ret_code retcode
;
3585 retcode
= drbd_adm_prepare(&adm_ctx
, skb
, info
, DRBD_ADM_NEED_RESOURCE
);
3586 if (!adm_ctx
.reply_skb
)
3588 if (retcode
!= NO_ERROR
)
3591 resource
= adm_ctx
.resource
;
3592 mutex_lock(&resource
->adm_mutex
);
3593 for_each_connection(connection
, resource
) {
3594 if (connection
->cstate
> C_STANDALONE
) {
3595 retcode
= ERR_NET_CONFIGURED
;
3599 if (!idr_is_empty(&resource
->devices
)) {
3600 retcode
= ERR_RES_IN_USE
;
3604 list_del_rcu(&resource
->resources
);
3605 for_each_connection(connection
, resource
)
3606 drbd_thread_stop(&connection
->worker
);
3608 drbd_free_resource(resource
);
3611 mutex_unlock(&resource
->adm_mutex
);
3613 drbd_adm_finish(&adm_ctx
, info
, retcode
);
3617 void drbd_bcast_event(struct drbd_device
*device
, const struct sib_info
*sib
)
3619 static atomic_t drbd_genl_seq
= ATOMIC_INIT(2); /* two. */
3620 struct sk_buff
*msg
;
3621 struct drbd_genlmsghdr
*d_out
;
3625 if (sib
->sib_reason
== SIB_SYNC_PROGRESS
) {
3626 if (time_after(jiffies
, device
->rs_last_bcast
+ HZ
))
3627 device
->rs_last_bcast
= jiffies
;
3632 seq
= atomic_inc_return(&drbd_genl_seq
);
3633 msg
= genlmsg_new(NLMSG_GOODSIZE
, GFP_NOIO
);
3638 d_out
= genlmsg_put(msg
, 0, seq
, &drbd_genl_family
, 0, DRBD_EVENT
);
3639 if (!d_out
) /* cannot happen, but anyways. */
3640 goto nla_put_failure
;
3641 d_out
->minor
= device_to_minor(device
);
3642 d_out
->ret_code
= NO_ERROR
;
3644 if (nla_put_status_info(msg
, device
, sib
))
3645 goto nla_put_failure
;
3646 genlmsg_end(msg
, d_out
);
3647 err
= drbd_genl_multicast_events(msg
, 0);
3648 /* msg has been consumed or freed in netlink_broadcast() */
3649 if (err
&& err
!= -ESRCH
)
3657 drbd_err(device
, "Error %d while broadcasting event. "
3658 "Event seq:%u sib_reason:%u\n",
3659 err
, seq
, sib
->sib_reason
);