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>
36 #include "drbd_wrappers.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_add_minor(struct sk_buff
*skb
, struct genl_info
*info
);
48 int drbd_adm_delete_minor(struct sk_buff
*skb
, struct genl_info
*info
);
50 int drbd_adm_create_connection(struct sk_buff
*skb
, struct genl_info
*info
);
51 int drbd_adm_delete_connection(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>
78 #include <linux/genl_magic_func.h>
80 /* used blkdev_get_by_path, to claim our meta data device(s) */
81 static char *drbd_m_holder
= "Hands off! this is DRBD's meta data device.";
83 /* Configuration is strictly serialized, because generic netlink message
84 * processing is strictly serialized by the genl_lock().
85 * Which means we can use one static global drbd_config_context struct.
87 static struct drbd_config_context
{
88 /* assigned from drbd_genlmsghdr */
90 /* assigned from request attributes, if present */
92 #define VOLUME_UNSPECIFIED (-1U)
93 /* pointer into the request skb,
94 * limited lifetime! */
98 struct sk_buff
*reply_skb
;
99 /* pointer into reply buffer */
100 struct drbd_genlmsghdr
*reply_dh
;
101 /* resolved from attributes, if possible */
102 struct drbd_conf
*mdev
;
103 struct drbd_tconn
*tconn
;
106 static void drbd_adm_send_reply(struct sk_buff
*skb
, struct genl_info
*info
)
108 genlmsg_end(skb
, genlmsg_data(nlmsg_data(nlmsg_hdr(skb
))));
109 if (genlmsg_reply(skb
, info
))
110 printk(KERN_ERR
"drbd: error sending genl reply\n");
113 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
114 * reason it could fail was no space in skb, and there are 4k available. */
115 int drbd_msg_put_info(const char *info
)
117 struct sk_buff
*skb
= adm_ctx
.reply_skb
;
121 if (!info
|| !info
[0])
124 nla
= nla_nest_start(skb
, DRBD_NLA_CFG_REPLY
);
128 err
= nla_put_string(skb
, T_info_text
, info
);
130 nla_nest_cancel(skb
, nla
);
133 nla_nest_end(skb
, nla
);
137 /* This would be a good candidate for a "pre_doit" hook,
138 * and per-family private info->pointers.
139 * But we need to stay compatible with older kernels.
140 * If it returns successfully, adm_ctx members are valid.
142 #define DRBD_ADM_NEED_MINOR 1
143 #define DRBD_ADM_NEED_CONN 2
144 static int drbd_adm_prepare(struct sk_buff
*skb
, struct genl_info
*info
,
147 struct drbd_genlmsghdr
*d_in
= info
->userhdr
;
148 const u8 cmd
= info
->genlhdr
->cmd
;
151 memset(&adm_ctx
, 0, sizeof(adm_ctx
));
153 /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
154 if (cmd
!= DRBD_ADM_GET_STATUS
155 && security_netlink_recv(skb
, CAP_SYS_ADMIN
))
158 adm_ctx
.reply_skb
= genlmsg_new(NLMSG_GOODSIZE
, GFP_KERNEL
);
159 if (!adm_ctx
.reply_skb
)
162 adm_ctx
.reply_dh
= genlmsg_put_reply(adm_ctx
.reply_skb
,
163 info
, &drbd_genl_family
, 0, cmd
);
164 /* put of a few bytes into a fresh skb of >= 4k will always succeed.
166 if (!adm_ctx
.reply_dh
)
169 adm_ctx
.reply_dh
->minor
= d_in
->minor
;
170 adm_ctx
.reply_dh
->ret_code
= NO_ERROR
;
172 if (info
->attrs
[DRBD_NLA_CFG_CONTEXT
]) {
174 /* parse and validate only */
175 err
= drbd_cfg_context_from_attrs(NULL
, info
);
179 /* It was present, and valid,
180 * copy it over to the reply skb. */
181 err
= nla_put_nohdr(adm_ctx
.reply_skb
,
182 info
->attrs
[DRBD_NLA_CFG_CONTEXT
]->nla_len
,
183 info
->attrs
[DRBD_NLA_CFG_CONTEXT
]);
187 /* and assign stuff to the global adm_ctx */
188 nla
= nested_attr_tb
[__nla_type(T_ctx_volume
)];
189 adm_ctx
.volume
= nla
? nla_get_u32(nla
) : VOLUME_UNSPECIFIED
;
190 nla
= nested_attr_tb
[__nla_type(T_ctx_conn_name
)];
192 adm_ctx
.conn_name
= nla_data(nla
);
194 adm_ctx
.volume
= VOLUME_UNSPECIFIED
;
196 adm_ctx
.minor
= d_in
->minor
;
197 adm_ctx
.mdev
= minor_to_mdev(d_in
->minor
);
198 adm_ctx
.tconn
= conn_by_name(adm_ctx
.conn_name
);
200 if (!adm_ctx
.mdev
&& (flags
& DRBD_ADM_NEED_MINOR
)) {
201 drbd_msg_put_info("unknown minor");
202 return ERR_MINOR_INVALID
;
204 if (!adm_ctx
.tconn
&& (flags
& DRBD_ADM_NEED_CONN
)) {
205 drbd_msg_put_info("unknown connection");
206 return ERR_INVALID_REQUEST
;
209 /* some more paranoia, if the request was over-determined */
210 if (adm_ctx
.mdev
&& adm_ctx
.tconn
&&
211 adm_ctx
.mdev
->tconn
!= adm_ctx
.tconn
) {
212 pr_warning("request: minor=%u, conn=%s; but that minor belongs to connection %s\n",
213 adm_ctx
.minor
, adm_ctx
.conn_name
, adm_ctx
.mdev
->tconn
->name
);
214 drbd_msg_put_info("minor exists in different connection");
215 return ERR_INVALID_REQUEST
;
218 adm_ctx
.volume
!= VOLUME_UNSPECIFIED
&&
219 adm_ctx
.volume
!= adm_ctx
.mdev
->vnr
) {
220 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
221 adm_ctx
.minor
, adm_ctx
.volume
,
222 adm_ctx
.mdev
->vnr
, adm_ctx
.mdev
->tconn
->name
);
223 drbd_msg_put_info("minor exists as different volume");
224 return ERR_INVALID_REQUEST
;
226 if (adm_ctx
.mdev
&& !adm_ctx
.tconn
)
227 adm_ctx
.tconn
= adm_ctx
.mdev
->tconn
;
231 nlmsg_free(adm_ctx
.reply_skb
);
232 adm_ctx
.reply_skb
= NULL
;
236 static int drbd_adm_finish(struct genl_info
*info
, int retcode
)
239 const char *conn_name
= NULL
;
241 if (!adm_ctx
.reply_skb
)
244 adm_ctx
.reply_dh
->ret_code
= retcode
;
246 nla
= info
->attrs
[DRBD_NLA_CFG_CONTEXT
];
248 nla
= nla_find_nested(nla
, __nla_type(T_ctx_conn_name
));
250 conn_name
= nla_data(nla
);
253 drbd_adm_send_reply(adm_ctx
.reply_skb
, info
);
257 static void setup_khelper_env(struct drbd_tconn
*tconn
, char **envp
)
261 if (get_net_conf(tconn
)) {
262 switch (((struct sockaddr
*)tconn
->net_conf
->peer_addr
)->sa_family
) {
265 snprintf(envp
[4], 60, "DRBD_PEER_ADDRESS=%pI6",
266 &((struct sockaddr_in6
*)tconn
->net_conf
->peer_addr
)->sin6_addr
);
270 snprintf(envp
[4], 60, "DRBD_PEER_ADDRESS=%pI4",
271 &((struct sockaddr_in
*)tconn
->net_conf
->peer_addr
)->sin_addr
);
275 snprintf(envp
[4], 60, "DRBD_PEER_ADDRESS=%pI4",
276 &((struct sockaddr_in
*)tconn
->net_conf
->peer_addr
)->sin_addr
);
278 snprintf(envp
[3], 20, "DRBD_PEER_AF=%s", afs
);
283 int drbd_khelper(struct drbd_conf
*mdev
, char *cmd
)
285 char *envp
[] = { "HOME=/",
287 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
288 (char[20]) { }, /* address family */
289 (char[60]) { }, /* address */
292 char *argv
[] = {usermode_helper
, cmd
, mb
, NULL
};
296 snprintf(mb
, 12, "minor-%d", mdev_to_minor(mdev
));
297 setup_khelper_env(mdev
->tconn
, envp
);
299 /* The helper may take some time.
300 * write out any unsynced meta data changes now */
303 dev_info(DEV
, "helper command: %s %s %s\n", usermode_helper
, cmd
, mb
);
304 sib
.sib_reason
= SIB_HELPER_PRE
;
305 sib
.helper_name
= cmd
;
306 drbd_bcast_event(mdev
, &sib
);
307 ret
= call_usermodehelper(usermode_helper
, argv
, envp
, 1);
309 dev_warn(DEV
, "helper command: %s %s %s exit code %u (0x%x)\n",
310 usermode_helper
, cmd
, mb
,
311 (ret
>> 8) & 0xff, ret
);
313 dev_info(DEV
, "helper command: %s %s %s exit code %u (0x%x)\n",
314 usermode_helper
, cmd
, mb
,
315 (ret
>> 8) & 0xff, ret
);
316 sib
.sib_reason
= SIB_HELPER_POST
;
317 sib
.helper_exit_code
= ret
;
318 drbd_bcast_event(mdev
, &sib
);
320 if (ret
< 0) /* Ignore any ERRNOs we got. */
326 static void conn_md_sync(struct drbd_tconn
*tconn
)
328 struct drbd_conf
*mdev
;
331 idr_for_each_entry(&tconn
->volumes
, mdev
, vnr
)
335 int conn_khelper(struct drbd_tconn
*tconn
, char *cmd
)
337 char *envp
[] = { "HOME=/",
339 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
340 (char[20]) { }, /* address family */
341 (char[60]) { }, /* address */
343 char *argv
[] = {usermode_helper
, cmd
, tconn
->name
, NULL
};
346 setup_khelper_env(tconn
, envp
);
349 conn_info(tconn
, "helper command: %s %s %s\n", usermode_helper
, cmd
, tconn
->name
);
350 /* TODO: conn_bcast_event() ?? */
352 ret
= call_usermodehelper(usermode_helper
, argv
, envp
, 1);
354 conn_warn(tconn
, "helper command: %s %s %s exit code %u (0x%x)\n",
355 usermode_helper
, cmd
, tconn
->name
,
356 (ret
>> 8) & 0xff, ret
);
358 conn_info(tconn
, "helper command: %s %s %s exit code %u (0x%x)\n",
359 usermode_helper
, cmd
, tconn
->name
,
360 (ret
>> 8) & 0xff, ret
);
361 /* TODO: conn_bcast_event() ?? */
363 if (ret
< 0) /* Ignore any ERRNOs we got. */
369 static enum drbd_fencing_p
highest_fencing_policy(struct drbd_tconn
*tconn
)
371 enum drbd_fencing_p fp
= FP_NOT_AVAIL
;
372 struct drbd_conf
*mdev
;
375 idr_for_each_entry(&tconn
->volumes
, mdev
, vnr
) {
376 if (get_ldev_if_state(mdev
, D_CONSISTENT
)) {
377 fp
= max_t(enum drbd_fencing_p
, fp
, mdev
->ldev
->dc
.fencing
);
385 bool conn_try_outdate_peer(struct drbd_tconn
*tconn
)
387 union drbd_state mask
= { };
388 union drbd_state val
= { };
389 enum drbd_fencing_p fp
;
393 if (tconn
->cstate
>= C_WF_REPORT_PARAMS
) {
394 conn_err(tconn
, "Expected cstate < C_WF_REPORT_PARAMS\n");
398 fp
= highest_fencing_policy(tconn
);
401 conn_warn(tconn
, "Not fencing peer, I'm not even Consistent myself.\n");
408 r
= conn_khelper(tconn
, "fence-peer");
410 switch ((r
>>8) & 0xff) {
411 case 3: /* peer is inconsistent */
412 ex_to_string
= "peer is inconsistent or worse";
414 val
.pdsk
= D_INCONSISTENT
;
416 case 4: /* peer got outdated, or was already outdated */
417 ex_to_string
= "peer was fenced";
419 val
.pdsk
= D_OUTDATED
;
421 case 5: /* peer was down */
422 if (conn_highest_disk(tconn
) == D_UP_TO_DATE
) {
423 /* we will(have) create(d) a new UUID anyways... */
424 ex_to_string
= "peer is unreachable, assumed to be dead";
426 val
.pdsk
= D_OUTDATED
;
428 ex_to_string
= "peer unreachable, doing nothing since disk != UpToDate";
431 case 6: /* Peer is primary, voluntarily outdate myself.
432 * This is useful when an unconnected R_SECONDARY is asked to
433 * become R_PRIMARY, but finds the other peer being active. */
434 ex_to_string
= "peer is active";
435 conn_warn(tconn
, "Peer is primary, outdating myself.\n");
437 val
.disk
= D_OUTDATED
;
440 if (fp
!= FP_STONITH
)
441 conn_err(tconn
, "fence-peer() = 7 && fencing != Stonith !!!\n");
442 ex_to_string
= "peer was stonithed";
444 val
.pdsk
= D_OUTDATED
;
447 /* The script is broken ... */
448 conn_err(tconn
, "fence-peer helper broken, returned %d\n", (r
>>8)&0xff);
449 return false; /* Eventually leave IO frozen */
452 conn_info(tconn
, "fence-peer helper returned %d (%s)\n",
453 (r
>>8) & 0xff, ex_to_string
);
458 conn_request_state(tconn, mask, val, CS_VERBOSE);
459 here, because we might were able to re-establish the connection in the
461 spin_lock_irq(&tconn
->req_lock
);
462 if (tconn
->cstate
< C_WF_REPORT_PARAMS
)
463 _conn_request_state(tconn
, mask
, val
, CS_VERBOSE
);
464 spin_unlock_irq(&tconn
->req_lock
);
466 return conn_highest_pdsk(tconn
) <= D_OUTDATED
;
469 static int _try_outdate_peer_async(void *data
)
471 struct drbd_tconn
*tconn
= (struct drbd_tconn
*)data
;
473 conn_try_outdate_peer(tconn
);
478 void conn_try_outdate_peer_async(struct drbd_tconn
*tconn
)
480 struct task_struct
*opa
;
482 opa
= kthread_run(_try_outdate_peer_async
, tconn
, "drbd_async_h");
484 conn_err(tconn
, "out of mem, failed to invoke fence-peer helper\n");
488 drbd_set_role(struct drbd_conf
*mdev
, enum drbd_role new_role
, int force
)
490 const int max_tries
= 4;
491 enum drbd_state_rv rv
= SS_UNKNOWN_ERROR
;
494 union drbd_state mask
, val
;
496 if (new_role
== R_PRIMARY
)
497 request_ping(mdev
->tconn
); /* Detect a dead peer ASAP */
499 mutex_lock(mdev
->state_mutex
);
501 mask
.i
= 0; mask
.role
= R_MASK
;
502 val
.i
= 0; val
.role
= new_role
;
504 while (try++ < max_tries
) {
505 rv
= _drbd_request_state(mdev
, mask
, val
, CS_WAIT_COMPLETE
);
507 /* in case we first succeeded to outdate,
508 * but now suddenly could establish a connection */
509 if (rv
== SS_CW_FAILED_BY_PEER
&& mask
.pdsk
!= 0) {
515 if (rv
== SS_NO_UP_TO_DATE_DISK
&& force
&&
516 (mdev
->state
.disk
< D_UP_TO_DATE
&&
517 mdev
->state
.disk
>= D_INCONSISTENT
)) {
519 val
.disk
= D_UP_TO_DATE
;
524 if (rv
== SS_NO_UP_TO_DATE_DISK
&&
525 mdev
->state
.disk
== D_CONSISTENT
&& mask
.pdsk
== 0) {
526 D_ASSERT(mdev
->state
.pdsk
== D_UNKNOWN
);
528 if (conn_try_outdate_peer(mdev
->tconn
)) {
529 val
.disk
= D_UP_TO_DATE
;
535 if (rv
== SS_NOTHING_TO_DO
)
537 if (rv
== SS_PRIMARY_NOP
&& mask
.pdsk
== 0) {
538 if (!conn_try_outdate_peer(mdev
->tconn
) && force
) {
539 dev_warn(DEV
, "Forced into split brain situation!\n");
541 val
.pdsk
= D_OUTDATED
;
546 if (rv
== SS_TWO_PRIMARIES
) {
547 /* Maybe the peer is detected as dead very soon...
548 retry at most once more in this case. */
549 schedule_timeout_interruptible((mdev
->tconn
->net_conf
->ping_timeo
+1)*HZ
/10);
554 if (rv
< SS_SUCCESS
) {
555 rv
= _drbd_request_state(mdev
, mask
, val
,
556 CS_VERBOSE
+ CS_WAIT_COMPLETE
);
567 dev_warn(DEV
, "Forced to consider local data as UpToDate!\n");
569 /* Wait until nothing is on the fly :) */
570 wait_event(mdev
->misc_wait
, atomic_read(&mdev
->ap_pending_cnt
) == 0);
572 if (new_role
== R_SECONDARY
) {
573 set_disk_ro(mdev
->vdisk
, true);
574 if (get_ldev(mdev
)) {
575 mdev
->ldev
->md
.uuid
[UI_CURRENT
] &= ~(u64
)1;
579 if (get_net_conf(mdev
->tconn
)) {
580 mdev
->tconn
->net_conf
->want_lose
= 0;
581 put_net_conf(mdev
->tconn
);
583 set_disk_ro(mdev
->vdisk
, false);
584 if (get_ldev(mdev
)) {
585 if (((mdev
->state
.conn
< C_CONNECTED
||
586 mdev
->state
.pdsk
<= D_FAILED
)
587 && mdev
->ldev
->md
.uuid
[UI_BITMAP
] == 0) || forced
)
588 drbd_uuid_new_current(mdev
);
590 mdev
->ldev
->md
.uuid
[UI_CURRENT
] |= (u64
)1;
595 /* writeout of activity log covered areas of the bitmap
596 * to stable storage done in after state change already */
598 if (mdev
->state
.conn
>= C_WF_REPORT_PARAMS
) {
599 /* if this was forced, we should consider sync */
601 drbd_send_uuids(mdev
);
602 drbd_send_state(mdev
);
607 kobject_uevent(&disk_to_dev(mdev
->vdisk
)->kobj
, KOBJ_CHANGE
);
609 mutex_unlock(mdev
->state_mutex
);
613 static const char *from_attrs_err_to_txt(int err
)
615 return err
== -ENOMSG
? "required attribute missing" :
616 err
== -EOPNOTSUPP
? "unknown mandatory attribute" :
617 err
== -EEXIST
? "can not change invariant setting" :
618 "invalid attribute value";
621 int drbd_adm_set_role(struct sk_buff
*skb
, struct genl_info
*info
)
623 struct set_role_parms parms
;
625 enum drbd_ret_code retcode
;
627 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
628 if (!adm_ctx
.reply_skb
)
630 if (retcode
!= NO_ERROR
)
633 memset(&parms
, 0, sizeof(parms
));
634 if (info
->attrs
[DRBD_NLA_SET_ROLE_PARMS
]) {
635 err
= set_role_parms_from_attrs(&parms
, info
);
637 retcode
= ERR_MANDATORY_TAG
;
638 drbd_msg_put_info(from_attrs_err_to_txt(err
));
643 if (info
->genlhdr
->cmd
== DRBD_ADM_PRIMARY
)
644 retcode
= drbd_set_role(adm_ctx
.mdev
, R_PRIMARY
, parms
.assume_uptodate
);
646 retcode
= drbd_set_role(adm_ctx
.mdev
, R_SECONDARY
, 0);
648 drbd_adm_finish(info
, retcode
);
652 /* initializes the md.*_offset members, so we are able to find
653 * the on disk meta data */
654 static void drbd_md_set_sector_offsets(struct drbd_conf
*mdev
,
655 struct drbd_backing_dev
*bdev
)
657 sector_t md_size_sect
= 0;
658 switch (bdev
->dc
.meta_dev_idx
) {
660 /* v07 style fixed size indexed meta data */
661 bdev
->md
.md_size_sect
= MD_RESERVED_SECT
;
662 bdev
->md
.md_offset
= drbd_md_ss__(mdev
, bdev
);
663 bdev
->md
.al_offset
= MD_AL_OFFSET
;
664 bdev
->md
.bm_offset
= MD_BM_OFFSET
;
666 case DRBD_MD_INDEX_FLEX_EXT
:
667 /* just occupy the full device; unit: sectors */
668 bdev
->md
.md_size_sect
= drbd_get_capacity(bdev
->md_bdev
);
669 bdev
->md
.md_offset
= 0;
670 bdev
->md
.al_offset
= MD_AL_OFFSET
;
671 bdev
->md
.bm_offset
= MD_BM_OFFSET
;
673 case DRBD_MD_INDEX_INTERNAL
:
674 case DRBD_MD_INDEX_FLEX_INT
:
675 bdev
->md
.md_offset
= drbd_md_ss__(mdev
, bdev
);
676 /* al size is still fixed */
677 bdev
->md
.al_offset
= -MD_AL_SECTORS
;
678 /* we need (slightly less than) ~ this much bitmap sectors: */
679 md_size_sect
= drbd_get_capacity(bdev
->backing_bdev
);
680 md_size_sect
= ALIGN(md_size_sect
, BM_SECT_PER_EXT
);
681 md_size_sect
= BM_SECT_TO_EXT(md_size_sect
);
682 md_size_sect
= ALIGN(md_size_sect
, 8);
684 /* plus the "drbd meta data super block",
685 * and the activity log; */
686 md_size_sect
+= MD_BM_OFFSET
;
688 bdev
->md
.md_size_sect
= md_size_sect
;
689 /* bitmap offset is adjusted by 'super' block size */
690 bdev
->md
.bm_offset
= -md_size_sect
+ MD_AL_OFFSET
;
695 /* input size is expected to be in KB */
696 char *ppsize(char *buf
, unsigned long long size
)
698 /* Needs 9 bytes at max including trailing NUL:
699 * -1ULL ==> "16384 EB" */
700 static char units
[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
702 while (size
>= 10000 && base
< sizeof(units
)-1) {
704 size
= (size
>> 10) + !!(size
& (1<<9));
707 sprintf(buf
, "%u %cB", (unsigned)size
, units
[base
]);
712 /* there is still a theoretical deadlock when called from receiver
713 * on an D_INCONSISTENT R_PRIMARY:
714 * remote READ does inc_ap_bio, receiver would need to receive answer
715 * packet from remote to dec_ap_bio again.
716 * receiver receive_sizes(), comes here,
717 * waits for ap_bio_cnt == 0. -> deadlock.
718 * but this cannot happen, actually, because:
719 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
720 * (not connected, or bad/no disk on peer):
721 * see drbd_fail_request_early, ap_bio_cnt is zero.
722 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
723 * peer may not initiate a resize.
725 /* Note these are not to be confused with
726 * drbd_adm_suspend_io/drbd_adm_resume_io,
727 * which are (sub) state changes triggered by admin (drbdsetup),
728 * and can be long lived.
729 * This changes an mdev->flag, is triggered by drbd internals,
730 * and should be short-lived. */
731 void drbd_suspend_io(struct drbd_conf
*mdev
)
733 set_bit(SUSPEND_IO
, &mdev
->flags
);
734 if (drbd_suspended(mdev
))
736 wait_event(mdev
->misc_wait
, !atomic_read(&mdev
->ap_bio_cnt
));
739 void drbd_resume_io(struct drbd_conf
*mdev
)
741 clear_bit(SUSPEND_IO
, &mdev
->flags
);
742 wake_up(&mdev
->misc_wait
);
746 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
747 * @mdev: DRBD device.
749 * Returns 0 on success, negative return values indicate errors.
750 * You should call drbd_md_sync() after calling this function.
752 enum determine_dev_size
drbd_determine_dev_size(struct drbd_conf
*mdev
, enum dds_flags flags
) __must_hold(local
)
754 sector_t prev_first_sect
, prev_size
; /* previous meta location */
759 int md_moved
, la_size_changed
;
760 enum determine_dev_size rv
= unchanged
;
763 * application request passes inc_ap_bio,
764 * but then cannot get an AL-reference.
765 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
768 * Suspend IO right here.
769 * still lock the act_log to not trigger ASSERTs there.
771 drbd_suspend_io(mdev
);
773 /* no wait necessary anymore, actually we could assert that */
774 wait_event(mdev
->al_wait
, lc_try_lock(mdev
->act_log
));
776 prev_first_sect
= drbd_md_first_sector(mdev
->ldev
);
777 prev_size
= mdev
->ldev
->md
.md_size_sect
;
778 la_size
= mdev
->ldev
->md
.la_size_sect
;
780 /* TODO: should only be some assert here, not (re)init... */
781 drbd_md_set_sector_offsets(mdev
, mdev
->ldev
);
783 size
= drbd_new_dev_size(mdev
, mdev
->ldev
, flags
& DDSF_FORCED
);
785 if (drbd_get_capacity(mdev
->this_bdev
) != size
||
786 drbd_bm_capacity(mdev
) != size
) {
788 err
= drbd_bm_resize(mdev
, size
, !(flags
& DDSF_NO_RESYNC
));
790 /* currently there is only one error: ENOMEM! */
791 size
= drbd_bm_capacity(mdev
)>>1;
793 dev_err(DEV
, "OUT OF MEMORY! "
794 "Could not allocate bitmap!\n");
796 dev_err(DEV
, "BM resizing failed. "
797 "Leaving size unchanged at size = %lu KB\n",
798 (unsigned long)size
);
802 /* racy, see comments above. */
803 drbd_set_my_capacity(mdev
, size
);
804 mdev
->ldev
->md
.la_size_sect
= size
;
805 dev_info(DEV
, "size = %s (%llu KB)\n", ppsize(ppb
, size
>>1),
806 (unsigned long long)size
>>1);
808 if (rv
== dev_size_error
)
811 la_size_changed
= (la_size
!= mdev
->ldev
->md
.la_size_sect
);
813 md_moved
= prev_first_sect
!= drbd_md_first_sector(mdev
->ldev
)
814 || prev_size
!= mdev
->ldev
->md
.md_size_sect
;
816 if (la_size_changed
|| md_moved
) {
819 drbd_al_shrink(mdev
); /* All extents inactive. */
820 dev_info(DEV
, "Writing the whole bitmap, %s\n",
821 la_size_changed
&& md_moved
? "size changed and md moved" :
822 la_size_changed
? "size changed" : "md moved");
823 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
824 err
= drbd_bitmap_io(mdev
, &drbd_bm_write
,
825 "size changed", BM_LOCKED_MASK
);
830 drbd_md_mark_dirty(mdev
);
838 lc_unlock(mdev
->act_log
);
839 wake_up(&mdev
->al_wait
);
840 drbd_resume_io(mdev
);
846 drbd_new_dev_size(struct drbd_conf
*mdev
, struct drbd_backing_dev
*bdev
, int assume_peer_has_space
)
848 sector_t p_size
= mdev
->p_size
; /* partner's disk size. */
849 sector_t la_size
= bdev
->md
.la_size_sect
; /* last agreed size. */
850 sector_t m_size
; /* my size */
851 sector_t u_size
= bdev
->dc
.disk_size
; /* size requested by user. */
854 m_size
= drbd_get_max_capacity(bdev
);
856 if (mdev
->state
.conn
< C_CONNECTED
&& assume_peer_has_space
) {
857 dev_warn(DEV
, "Resize while not connected was forced by the user!\n");
861 if (p_size
&& m_size
) {
862 size
= min_t(sector_t
, p_size
, m_size
);
866 if (m_size
&& m_size
< size
)
868 if (p_size
&& p_size
< size
)
879 dev_err(DEV
, "Both nodes diskless!\n");
883 dev_err(DEV
, "Requested disk size is too big (%lu > %lu)\n",
884 (unsigned long)u_size
>>1, (unsigned long)size
>>1);
893 * drbd_check_al_size() - Ensures that the AL is of the right size
894 * @mdev: DRBD device.
896 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
897 * failed, and 0 on success. You should call drbd_md_sync() after you called
900 static int drbd_check_al_size(struct drbd_conf
*mdev
, struct disk_conf
*dc
)
902 struct lru_cache
*n
, *t
;
903 struct lc_element
*e
;
907 if (!expect(dc
->al_extents
>= DRBD_AL_EXTENTS_MIN
))
908 dc
->al_extents
= DRBD_AL_EXTENTS_MIN
;
911 mdev
->act_log
->nr_elements
== dc
->al_extents
)
916 n
= lc_create("act_log", drbd_al_ext_cache
, AL_UPDATES_PER_TRANSACTION
,
917 dc
->al_extents
, sizeof(struct lc_element
), 0);
920 dev_err(DEV
, "Cannot allocate act_log lru!\n");
923 spin_lock_irq(&mdev
->al_lock
);
925 for (i
= 0; i
< t
->nr_elements
; i
++) {
926 e
= lc_element_by_index(t
, i
);
928 dev_err(DEV
, "refcnt(%d)==%d\n",
929 e
->lc_number
, e
->refcnt
);
935 spin_unlock_irq(&mdev
->al_lock
);
937 dev_err(DEV
, "Activity log still in use!\n");
944 drbd_md_mark_dirty(mdev
); /* we changed mdev->act_log->nr_elemens */
948 static void drbd_setup_queue_param(struct drbd_conf
*mdev
, unsigned int max_bio_size
)
950 struct request_queue
* const q
= mdev
->rq_queue
;
951 int max_hw_sectors
= max_bio_size
>> 9;
952 int max_segments
= 0;
954 if (get_ldev_if_state(mdev
, D_ATTACHING
)) {
955 struct request_queue
* const b
= mdev
->ldev
->backing_bdev
->bd_disk
->queue
;
957 max_hw_sectors
= min(queue_max_hw_sectors(b
), max_bio_size
>> 9);
958 max_segments
= mdev
->ldev
->dc
.max_bio_bvecs
;
962 blk_queue_logical_block_size(q
, 512);
963 blk_queue_max_hw_sectors(q
, max_hw_sectors
);
964 /* This is the workaround for "bio would need to, but cannot, be split" */
965 blk_queue_max_segments(q
, max_segments
? max_segments
: BLK_MAX_SEGMENTS
);
966 blk_queue_segment_boundary(q
, PAGE_CACHE_SIZE
-1);
968 if (get_ldev_if_state(mdev
, D_ATTACHING
)) {
969 struct request_queue
* const b
= mdev
->ldev
->backing_bdev
->bd_disk
->queue
;
971 blk_queue_stack_limits(q
, b
);
973 if (q
->backing_dev_info
.ra_pages
!= b
->backing_dev_info
.ra_pages
) {
974 dev_info(DEV
, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
975 q
->backing_dev_info
.ra_pages
,
976 b
->backing_dev_info
.ra_pages
);
977 q
->backing_dev_info
.ra_pages
= b
->backing_dev_info
.ra_pages
;
983 void drbd_reconsider_max_bio_size(struct drbd_conf
*mdev
)
985 int now
, new, local
, peer
;
987 now
= queue_max_hw_sectors(mdev
->rq_queue
) << 9;
988 local
= mdev
->local_max_bio_size
; /* Eventually last known value, from volatile memory */
989 peer
= mdev
->peer_max_bio_size
; /* Eventually last known value, from meta data */
991 if (get_ldev_if_state(mdev
, D_ATTACHING
)) {
992 local
= queue_max_hw_sectors(mdev
->ldev
->backing_bdev
->bd_disk
->queue
) << 9;
993 mdev
->local_max_bio_size
= local
;
997 /* We may ignore peer limits if the peer is modern enough.
998 Because new from 8.3.8 onwards the peer can use multiple
999 BIOs for a single peer_request */
1000 if (mdev
->state
.conn
>= C_CONNECTED
) {
1001 if (mdev
->tconn
->agreed_pro_version
< 94)
1002 peer
= mdev
->peer_max_bio_size
;
1003 else if (mdev
->tconn
->agreed_pro_version
== 94)
1004 peer
= DRBD_MAX_SIZE_H80_PACKET
;
1005 else /* drbd 8.3.8 onwards */
1006 peer
= DRBD_MAX_BIO_SIZE
;
1009 new = min_t(int, local
, peer
);
1011 if (mdev
->state
.role
== R_PRIMARY
&& new < now
)
1012 dev_err(DEV
, "ASSERT FAILED new < now; (%d < %d)\n", new, now
);
1015 dev_info(DEV
, "max BIO size = %u\n", new);
1017 drbd_setup_queue_param(mdev
, new);
1020 /* serialize deconfig (worker exiting, doing cleanup)
1021 * and reconfig (drbdsetup disk, drbdsetup net)
1023 * Wait for a potentially exiting worker, then restart it,
1024 * or start a new one. Flush any pending work, there may still be an
1025 * after_state_change queued.
1027 static void conn_reconfig_start(struct drbd_tconn
*tconn
)
1029 wait_event(tconn
->ping_wait
, !test_and_set_bit(CONFIG_PENDING
, &tconn
->flags
));
1030 wait_event(tconn
->ping_wait
, !test_bit(OBJECT_DYING
, &tconn
->flags
));
1031 drbd_thread_start(&tconn
->worker
);
1032 conn_flush_workqueue(tconn
);
1035 /* if still unconfigured, stops worker again.
1036 * if configured now, clears CONFIG_PENDING.
1037 * wakes potential waiters */
1038 static void conn_reconfig_done(struct drbd_tconn
*tconn
)
1040 spin_lock_irq(&tconn
->req_lock
);
1041 if (conn_all_vols_unconf(tconn
)) {
1042 set_bit(OBJECT_DYING
, &tconn
->flags
);
1043 drbd_thread_stop_nowait(&tconn
->worker
);
1045 clear_bit(CONFIG_PENDING
, &tconn
->flags
);
1046 spin_unlock_irq(&tconn
->req_lock
);
1047 wake_up(&tconn
->ping_wait
);
1050 /* Make sure IO is suspended before calling this function(). */
1051 static void drbd_suspend_al(struct drbd_conf
*mdev
)
1055 if (!lc_try_lock(mdev
->act_log
)) {
1056 dev_warn(DEV
, "Failed to lock al in drbd_suspend_al()\n");
1060 drbd_al_shrink(mdev
);
1061 spin_lock_irq(&mdev
->tconn
->req_lock
);
1062 if (mdev
->state
.conn
< C_CONNECTED
)
1063 s
= !test_and_set_bit(AL_SUSPENDED
, &mdev
->flags
);
1064 spin_unlock_irq(&mdev
->tconn
->req_lock
);
1065 lc_unlock(mdev
->act_log
);
1068 dev_info(DEV
, "Suspended AL updates\n");
1071 int drbd_adm_disk_opts(struct sk_buff
*skb
, struct genl_info
*info
)
1073 enum drbd_ret_code retcode
;
1074 struct drbd_conf
*mdev
;
1075 struct disk_conf
*ndc
; /* new disk conf */
1077 int *rs_plan_s
= NULL
;
1079 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
1080 if (!adm_ctx
.reply_skb
)
1082 if (retcode
!= NO_ERROR
)
1085 mdev
= adm_ctx
.mdev
;
1087 /* we also need a disk
1088 * to change the options on */
1089 if (!get_ldev(mdev
)) {
1090 retcode
= ERR_NO_DISK
;
1094 /* FIXME freeze IO, cluster wide.
1096 * We should make sure no-one uses
1097 * some half-updated struct when we
1098 * assign it later. */
1100 ndc
= kmalloc(sizeof(*ndc
), GFP_KERNEL
);
1102 retcode
= ERR_NOMEM
;
1106 memcpy(ndc
, &mdev
->ldev
->dc
, sizeof(*ndc
));
1107 err
= disk_conf_from_attrs_for_change(ndc
, info
);
1109 retcode
= ERR_MANDATORY_TAG
;
1110 drbd_msg_put_info(from_attrs_err_to_txt(err
));
1113 if (!expect(ndc
->resync_rate
>= 1))
1114 ndc
->resync_rate
= 1;
1116 /* clip to allowed range */
1117 if (!expect(ndc
->al_extents
>= DRBD_AL_EXTENTS_MIN
))
1118 ndc
->al_extents
= DRBD_AL_EXTENTS_MIN
;
1119 if (!expect(ndc
->al_extents
<= DRBD_AL_EXTENTS_MAX
))
1120 ndc
->al_extents
= DRBD_AL_EXTENTS_MAX
;
1122 /* most sanity checks done, try to assign the new sync-after
1123 * dependency. need to hold the global lock in there,
1124 * to avoid a race in the dependency loop check. */
1125 retcode
= drbd_alter_sa(mdev
, ndc
->resync_after
);
1126 if (retcode
!= NO_ERROR
)
1129 fifo_size
= (ndc
->c_plan_ahead
* 10 * SLEEP_TIME
) / HZ
;
1130 if (fifo_size
!= mdev
->rs_plan_s
.size
&& fifo_size
> 0) {
1131 rs_plan_s
= kzalloc(sizeof(int) * fifo_size
, GFP_KERNEL
);
1133 dev_err(DEV
, "kmalloc of fifo_buffer failed");
1134 retcode
= ERR_NOMEM
;
1139 if (fifo_size
!= mdev
->rs_plan_s
.size
) {
1140 kfree(mdev
->rs_plan_s
.values
);
1141 mdev
->rs_plan_s
.values
= rs_plan_s
;
1142 mdev
->rs_plan_s
.size
= fifo_size
;
1143 mdev
->rs_planed
= 0;
1147 wait_event(mdev
->al_wait
, lc_try_lock(mdev
->act_log
));
1148 drbd_al_shrink(mdev
);
1149 err
= drbd_check_al_size(mdev
, ndc
);
1150 lc_unlock(mdev
->act_log
);
1151 wake_up(&mdev
->al_wait
);
1154 retcode
= ERR_NOMEM
;
1159 * To avoid someone looking at a half-updated struct, we probably
1160 * should have a rw-semaphor on net_conf and disk_conf.
1162 mdev
->ldev
->dc
= *ndc
;
1167 if (mdev
->state
.conn
>= C_CONNECTED
)
1168 drbd_send_sync_param(mdev
);
1175 drbd_adm_finish(info
, retcode
);
1179 int drbd_adm_attach(struct sk_buff
*skb
, struct genl_info
*info
)
1181 struct drbd_conf
*mdev
;
1183 enum drbd_ret_code retcode
;
1184 enum determine_dev_size dd
;
1185 sector_t max_possible_sectors
;
1186 sector_t min_md_device_sectors
;
1187 struct drbd_backing_dev
*nbc
= NULL
; /* new_backing_conf */
1188 struct block_device
*bdev
;
1189 struct lru_cache
*resync_lru
= NULL
;
1190 union drbd_state ns
, os
;
1191 enum drbd_state_rv rv
;
1192 int cp_discovered
= 0;
1194 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
1195 if (!adm_ctx
.reply_skb
)
1197 if (retcode
!= NO_ERROR
)
1200 mdev
= adm_ctx
.mdev
;
1201 conn_reconfig_start(mdev
->tconn
);
1203 /* if you want to reconfigure, please tear down first */
1204 if (mdev
->state
.disk
> D_DISKLESS
) {
1205 retcode
= ERR_DISK_CONFIGURED
;
1208 /* It may just now have detached because of IO error. Make sure
1209 * drbd_ldev_destroy is done already, we may end up here very fast,
1210 * e.g. if someone calls attach from the on-io-error handler,
1211 * to realize a "hot spare" feature (not that I'd recommend that) */
1212 wait_event(mdev
->misc_wait
, !atomic_read(&mdev
->local_cnt
));
1214 /* allocation not in the IO path, drbdsetup context */
1215 nbc
= kzalloc(sizeof(struct drbd_backing_dev
), GFP_KERNEL
);
1217 retcode
= ERR_NOMEM
;
1221 nbc
->dc
= (struct disk_conf
) {
1222 {}, 0, /* backing_dev */
1223 {}, 0, /* meta_dev */
1224 0, /* meta_dev_idx */
1225 DRBD_DISK_SIZE_SECT_DEF
, /* disk_size */
1226 DRBD_MAX_BIO_BVECS_DEF
, /* max_bio_bvecs */
1227 DRBD_ON_IO_ERROR_DEF
, /* on_io_error */
1228 DRBD_FENCING_DEF
, /* fencing */
1229 DRBD_RATE_DEF
, /* resync_rate */
1230 DRBD_AFTER_DEF
, /* resync_after */
1231 DRBD_AL_EXTENTS_DEF
, /* al_extents */
1232 DRBD_C_PLAN_AHEAD_DEF
, /* c_plan_ahead */
1233 DRBD_C_DELAY_TARGET_DEF
, /* c_delay_target */
1234 DRBD_C_FILL_TARGET_DEF
, /* c_fill_target */
1235 DRBD_C_MAX_RATE_DEF
, /* c_max_rate */
1236 DRBD_C_MIN_RATE_DEF
, /* c_min_rate */
1237 0, /* no_disk_barrier */
1238 0, /* no_disk_flush */
1239 0, /* no_disk_drain */
1240 0, /* no_md_flush */
1243 err
= disk_conf_from_attrs(&nbc
->dc
, info
);
1245 retcode
= ERR_MANDATORY_TAG
;
1246 drbd_msg_put_info(from_attrs_err_to_txt(err
));
1250 if ((int)nbc
->dc
.meta_dev_idx
< DRBD_MD_INDEX_FLEX_INT
) {
1251 retcode
= ERR_MD_IDX_INVALID
;
1255 if (get_net_conf(mdev
->tconn
)) {
1256 int prot
= mdev
->tconn
->net_conf
->wire_protocol
;
1257 put_net_conf(mdev
->tconn
);
1258 if (nbc
->dc
.fencing
== FP_STONITH
&& prot
== DRBD_PROT_A
) {
1259 retcode
= ERR_STONITH_AND_PROT_A
;
1264 bdev
= blkdev_get_by_path(nbc
->dc
.backing_dev
,
1265 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, mdev
);
1267 dev_err(DEV
, "open(\"%s\") failed with %ld\n", nbc
->dc
.backing_dev
,
1269 retcode
= ERR_OPEN_DISK
;
1272 nbc
->backing_bdev
= bdev
;
1275 * meta_dev_idx >= 0: external fixed size, possibly multiple
1276 * drbd sharing one meta device. TODO in that case, paranoia
1277 * check that [md_bdev, meta_dev_idx] is not yet used by some
1278 * other drbd minor! (if you use drbd.conf + drbdadm, that
1279 * should check it for you already; but if you don't, or
1280 * someone fooled it, we need to double check here)
1282 bdev
= blkdev_get_by_path(nbc
->dc
.meta_dev
,
1283 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
,
1284 ((int)nbc
->dc
.meta_dev_idx
< 0) ?
1285 (void *)mdev
: (void *)drbd_m_holder
);
1287 dev_err(DEV
, "open(\"%s\") failed with %ld\n", nbc
->dc
.meta_dev
,
1289 retcode
= ERR_OPEN_MD_DISK
;
1292 nbc
->md_bdev
= bdev
;
1294 if ((nbc
->backing_bdev
== nbc
->md_bdev
) !=
1295 (nbc
->dc
.meta_dev_idx
== DRBD_MD_INDEX_INTERNAL
||
1296 nbc
->dc
.meta_dev_idx
== DRBD_MD_INDEX_FLEX_INT
)) {
1297 retcode
= ERR_MD_IDX_INVALID
;
1301 resync_lru
= lc_create("resync", drbd_bm_ext_cache
,
1302 1, 61, sizeof(struct bm_extent
),
1303 offsetof(struct bm_extent
, lce
));
1305 retcode
= ERR_NOMEM
;
1309 /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1310 drbd_md_set_sector_offsets(mdev
, nbc
);
1312 if (drbd_get_max_capacity(nbc
) < nbc
->dc
.disk_size
) {
1313 dev_err(DEV
, "max capacity %llu smaller than disk size %llu\n",
1314 (unsigned long long) drbd_get_max_capacity(nbc
),
1315 (unsigned long long) nbc
->dc
.disk_size
);
1316 retcode
= ERR_DISK_TO_SMALL
;
1320 if ((int)nbc
->dc
.meta_dev_idx
< 0) {
1321 max_possible_sectors
= DRBD_MAX_SECTORS_FLEX
;
1322 /* at least one MB, otherwise it does not make sense */
1323 min_md_device_sectors
= (2<<10);
1325 max_possible_sectors
= DRBD_MAX_SECTORS
;
1326 min_md_device_sectors
= MD_RESERVED_SECT
* (nbc
->dc
.meta_dev_idx
+ 1);
1329 if (drbd_get_capacity(nbc
->md_bdev
) < min_md_device_sectors
) {
1330 retcode
= ERR_MD_DISK_TO_SMALL
;
1331 dev_warn(DEV
, "refusing attach: md-device too small, "
1332 "at least %llu sectors needed for this meta-disk type\n",
1333 (unsigned long long) min_md_device_sectors
);
1337 /* Make sure the new disk is big enough
1338 * (we may currently be R_PRIMARY with no local disk...) */
1339 if (drbd_get_max_capacity(nbc
) <
1340 drbd_get_capacity(mdev
->this_bdev
)) {
1341 retcode
= ERR_DISK_TO_SMALL
;
1345 nbc
->known_size
= drbd_get_capacity(nbc
->backing_bdev
);
1347 if (nbc
->known_size
> max_possible_sectors
) {
1348 dev_warn(DEV
, "==> truncating very big lower level device "
1349 "to currently maximum possible %llu sectors <==\n",
1350 (unsigned long long) max_possible_sectors
);
1351 if ((int)nbc
->dc
.meta_dev_idx
>= 0)
1352 dev_warn(DEV
, "==>> using internal or flexible "
1353 "meta data may help <<==\n");
1356 drbd_suspend_io(mdev
);
1357 /* also wait for the last barrier ack. */
1358 wait_event(mdev
->misc_wait
, !atomic_read(&mdev
->ap_pending_cnt
) || drbd_suspended(mdev
));
1359 /* and for any other previously queued work */
1360 drbd_flush_workqueue(mdev
);
1362 rv
= _drbd_request_state(mdev
, NS(disk
, D_ATTACHING
), CS_VERBOSE
);
1363 retcode
= rv
; /* FIXME: Type mismatch. */
1364 drbd_resume_io(mdev
);
1365 if (rv
< SS_SUCCESS
)
1368 if (!get_ldev_if_state(mdev
, D_ATTACHING
))
1369 goto force_diskless
;
1371 drbd_md_set_sector_offsets(mdev
, nbc
);
1373 if (!mdev
->bitmap
) {
1374 if (drbd_bm_init(mdev
)) {
1375 retcode
= ERR_NOMEM
;
1376 goto force_diskless_dec
;
1380 retcode
= drbd_md_read(mdev
, nbc
);
1381 if (retcode
!= NO_ERROR
)
1382 goto force_diskless_dec
;
1384 if (mdev
->state
.conn
< C_CONNECTED
&&
1385 mdev
->state
.role
== R_PRIMARY
&&
1386 (mdev
->ed_uuid
& ~((u64
)1)) != (nbc
->md
.uuid
[UI_CURRENT
] & ~((u64
)1))) {
1387 dev_err(DEV
, "Can only attach to data with current UUID=%016llX\n",
1388 (unsigned long long)mdev
->ed_uuid
);
1389 retcode
= ERR_DATA_NOT_CURRENT
;
1390 goto force_diskless_dec
;
1393 /* Since we are diskless, fix the activity log first... */
1394 if (drbd_check_al_size(mdev
, &nbc
->dc
)) {
1395 retcode
= ERR_NOMEM
;
1396 goto force_diskless_dec
;
1399 /* Prevent shrinking of consistent devices ! */
1400 if (drbd_md_test_flag(nbc
, MDF_CONSISTENT
) &&
1401 drbd_new_dev_size(mdev
, nbc
, 0) < nbc
->md
.la_size_sect
) {
1402 dev_warn(DEV
, "refusing to truncate a consistent device\n");
1403 retcode
= ERR_DISK_TO_SMALL
;
1404 goto force_diskless_dec
;
1407 if (!drbd_al_read_log(mdev
, nbc
)) {
1408 retcode
= ERR_IO_MD_DISK
;
1409 goto force_diskless_dec
;
1412 /* Reset the "barriers don't work" bits here, then force meta data to
1413 * be written, to ensure we determine if barriers are supported. */
1414 if (nbc
->dc
.no_md_flush
)
1415 set_bit(MD_NO_FUA
, &mdev
->flags
);
1417 clear_bit(MD_NO_FUA
, &mdev
->flags
);
1419 /* Point of no return reached.
1420 * Devices and memory are no longer released by error cleanup below.
1421 * now mdev takes over responsibility, and the state engine should
1422 * clean it up somewhere. */
1423 D_ASSERT(mdev
->ldev
== NULL
);
1425 mdev
->resync
= resync_lru
;
1429 mdev
->write_ordering
= WO_bdev_flush
;
1430 drbd_bump_write_ordering(mdev
, WO_bdev_flush
);
1432 if (drbd_md_test_flag(mdev
->ldev
, MDF_CRASHED_PRIMARY
))
1433 set_bit(CRASHED_PRIMARY
, &mdev
->flags
);
1435 clear_bit(CRASHED_PRIMARY
, &mdev
->flags
);
1437 if (drbd_md_test_flag(mdev
->ldev
, MDF_PRIMARY_IND
) &&
1438 !(mdev
->state
.role
== R_PRIMARY
&& mdev
->tconn
->susp_nod
)) {
1439 set_bit(CRASHED_PRIMARY
, &mdev
->flags
);
1448 drbd_reconsider_max_bio_size(mdev
);
1450 /* If I am currently not R_PRIMARY,
1451 * but meta data primary indicator is set,
1452 * I just now recover from a hard crash,
1453 * and have been R_PRIMARY before that crash.
1455 * Now, if I had no connection before that crash
1456 * (have been degraded R_PRIMARY), chances are that
1457 * I won't find my peer now either.
1459 * In that case, and _only_ in that case,
1460 * we use the degr-wfc-timeout instead of the default,
1461 * so we can automatically recover from a crash of a
1462 * degraded but active "cluster" after a certain timeout.
1464 clear_bit(USE_DEGR_WFC_T
, &mdev
->flags
);
1465 if (mdev
->state
.role
!= R_PRIMARY
&&
1466 drbd_md_test_flag(mdev
->ldev
, MDF_PRIMARY_IND
) &&
1467 !drbd_md_test_flag(mdev
->ldev
, MDF_CONNECTED_IND
))
1468 set_bit(USE_DEGR_WFC_T
, &mdev
->flags
);
1470 dd
= drbd_determine_dev_size(mdev
, 0);
1471 if (dd
== dev_size_error
) {
1472 retcode
= ERR_NOMEM_BITMAP
;
1473 goto force_diskless_dec
;
1474 } else if (dd
== grew
)
1475 set_bit(RESYNC_AFTER_NEG
, &mdev
->flags
);
1477 if (drbd_md_test_flag(mdev
->ldev
, MDF_FULL_SYNC
)) {
1478 dev_info(DEV
, "Assuming that all blocks are out of sync "
1479 "(aka FullSync)\n");
1480 if (drbd_bitmap_io(mdev
, &drbd_bmio_set_n_write
,
1481 "set_n_write from attaching", BM_LOCKED_MASK
)) {
1482 retcode
= ERR_IO_MD_DISK
;
1483 goto force_diskless_dec
;
1486 if (drbd_bitmap_io(mdev
, &drbd_bm_read
,
1487 "read from attaching", BM_LOCKED_MASK
)) {
1488 retcode
= ERR_IO_MD_DISK
;
1489 goto force_diskless_dec
;
1493 if (cp_discovered
) {
1494 drbd_al_apply_to_bm(mdev
);
1495 if (drbd_bitmap_io(mdev
, &drbd_bm_write
,
1496 "crashed primary apply AL", BM_LOCKED_MASK
)) {
1497 retcode
= ERR_IO_MD_DISK
;
1498 goto force_diskless_dec
;
1502 if (_drbd_bm_total_weight(mdev
) == drbd_bm_bits(mdev
))
1503 drbd_suspend_al(mdev
); /* IO is still suspended here... */
1505 spin_lock_irq(&mdev
->tconn
->req_lock
);
1506 os
= drbd_read_state(mdev
);
1508 /* If MDF_CONSISTENT is not set go into inconsistent state,
1509 otherwise investigate MDF_WasUpToDate...
1510 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1511 otherwise into D_CONSISTENT state.
1513 if (drbd_md_test_flag(mdev
->ldev
, MDF_CONSISTENT
)) {
1514 if (drbd_md_test_flag(mdev
->ldev
, MDF_WAS_UP_TO_DATE
))
1515 ns
.disk
= D_CONSISTENT
;
1517 ns
.disk
= D_OUTDATED
;
1519 ns
.disk
= D_INCONSISTENT
;
1522 if (drbd_md_test_flag(mdev
->ldev
, MDF_PEER_OUT_DATED
))
1523 ns
.pdsk
= D_OUTDATED
;
1525 if ( ns
.disk
== D_CONSISTENT
&&
1526 (ns
.pdsk
== D_OUTDATED
|| mdev
->ldev
->dc
.fencing
== FP_DONT_CARE
))
1527 ns
.disk
= D_UP_TO_DATE
;
1529 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1530 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1531 this point, because drbd_request_state() modifies these
1534 /* In case we are C_CONNECTED postpone any decision on the new disk
1535 state after the negotiation phase. */
1536 if (mdev
->state
.conn
== C_CONNECTED
) {
1537 mdev
->new_state_tmp
.i
= ns
.i
;
1539 ns
.disk
= D_NEGOTIATING
;
1541 /* We expect to receive up-to-date UUIDs soon.
1542 To avoid a race in receive_state, free p_uuid while
1543 holding req_lock. I.e. atomic with the state change */
1544 kfree(mdev
->p_uuid
);
1545 mdev
->p_uuid
= NULL
;
1548 rv
= _drbd_set_state(mdev
, ns
, CS_VERBOSE
, NULL
);
1549 spin_unlock_irq(&mdev
->tconn
->req_lock
);
1551 if (rv
< SS_SUCCESS
)
1552 goto force_diskless_dec
;
1554 if (mdev
->state
.role
== R_PRIMARY
)
1555 mdev
->ldev
->md
.uuid
[UI_CURRENT
] |= (u64
)1;
1557 mdev
->ldev
->md
.uuid
[UI_CURRENT
] &= ~(u64
)1;
1559 drbd_md_mark_dirty(mdev
);
1562 kobject_uevent(&disk_to_dev(mdev
->vdisk
)->kobj
, KOBJ_CHANGE
);
1564 conn_reconfig_done(mdev
->tconn
);
1565 drbd_adm_finish(info
, retcode
);
1571 drbd_force_state(mdev
, NS(disk
, D_FAILED
));
1574 conn_reconfig_done(mdev
->tconn
);
1576 if (nbc
->backing_bdev
)
1577 blkdev_put(nbc
->backing_bdev
,
1578 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
1580 blkdev_put(nbc
->md_bdev
,
1581 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
1584 lc_destroy(resync_lru
);
1587 drbd_adm_finish(info
, retcode
);
1591 static int adm_detach(struct drbd_conf
*mdev
)
1593 enum drbd_state_rv retcode
;
1594 drbd_suspend_io(mdev
); /* so no-one is stuck in drbd_al_begin_io */
1595 retcode
= drbd_request_state(mdev
, NS(disk
, D_DISKLESS
));
1596 wait_event(mdev
->misc_wait
,
1597 mdev
->state
.disk
!= D_DISKLESS
||
1598 !atomic_read(&mdev
->local_cnt
));
1599 drbd_resume_io(mdev
);
1603 /* Detaching the disk is a process in multiple stages. First we need to lock
1604 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1605 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1606 * internal references as well.
1607 * Only then we have finally detached. */
1608 int drbd_adm_detach(struct sk_buff
*skb
, struct genl_info
*info
)
1610 enum drbd_ret_code retcode
;
1612 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
1613 if (!adm_ctx
.reply_skb
)
1615 if (retcode
!= NO_ERROR
)
1618 retcode
= adm_detach(adm_ctx
.mdev
);
1620 drbd_adm_finish(info
, retcode
);
1624 static bool conn_resync_running(struct drbd_tconn
*tconn
)
1626 struct drbd_conf
*mdev
;
1629 idr_for_each_entry(&tconn
->volumes
, mdev
, vnr
) {
1630 if (mdev
->state
.conn
== C_SYNC_SOURCE
||
1631 mdev
->state
.conn
== C_SYNC_TARGET
||
1632 mdev
->state
.conn
== C_PAUSED_SYNC_S
||
1633 mdev
->state
.conn
== C_PAUSED_SYNC_T
)
1639 static bool conn_ov_running(struct drbd_tconn
*tconn
)
1641 struct drbd_conf
*mdev
;
1644 idr_for_each_entry(&tconn
->volumes
, mdev
, vnr
) {
1645 if (mdev
->state
.conn
== C_VERIFY_S
||
1646 mdev
->state
.conn
== C_VERIFY_T
)
1652 int drbd_adm_net_opts(struct sk_buff
*skb
, struct genl_info
*info
)
1654 enum drbd_ret_code retcode
;
1655 struct drbd_tconn
*tconn
;
1656 struct net_conf
*new_conf
= NULL
;
1658 int ovr
; /* online verify running */
1659 int rsr
; /* re-sync running */
1660 struct crypto_hash
*verify_tfm
= NULL
;
1661 struct crypto_hash
*csums_tfm
= NULL
;
1664 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_CONN
);
1665 if (!adm_ctx
.reply_skb
)
1667 if (retcode
!= NO_ERROR
)
1670 tconn
= adm_ctx
.tconn
;
1672 new_conf
= kzalloc(sizeof(struct net_conf
), GFP_KERNEL
);
1674 retcode
= ERR_NOMEM
;
1678 /* we also need a net config
1679 * to change the options on */
1680 if (!get_net_conf(tconn
)) {
1681 drbd_msg_put_info("net conf missing, try connect");
1682 retcode
= ERR_INVALID_REQUEST
;
1686 conn_reconfig_start(tconn
);
1688 memcpy(new_conf
, tconn
->net_conf
, sizeof(*new_conf
));
1689 err
= net_conf_from_attrs_for_change(new_conf
, info
);
1691 retcode
= ERR_MANDATORY_TAG
;
1692 drbd_msg_put_info(from_attrs_err_to_txt(err
));
1696 /* re-sync running */
1697 rsr
= conn_resync_running(tconn
);
1698 if (rsr
&& strcmp(new_conf
->csums_alg
, tconn
->net_conf
->csums_alg
)) {
1699 retcode
= ERR_CSUMS_RESYNC_RUNNING
;
1703 if (!rsr
&& new_conf
->csums_alg
[0]) {
1704 csums_tfm
= crypto_alloc_hash(new_conf
->csums_alg
, 0, CRYPTO_ALG_ASYNC
);
1705 if (IS_ERR(csums_tfm
)) {
1707 retcode
= ERR_CSUMS_ALG
;
1711 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm
))) {
1712 retcode
= ERR_CSUMS_ALG_ND
;
1717 /* online verify running */
1718 ovr
= conn_ov_running(tconn
);
1720 if (strcmp(new_conf
->verify_alg
, tconn
->net_conf
->verify_alg
)) {
1721 retcode
= ERR_VERIFY_RUNNING
;
1726 if (!ovr
&& new_conf
->verify_alg
[0]) {
1727 verify_tfm
= crypto_alloc_hash(new_conf
->verify_alg
, 0, CRYPTO_ALG_ASYNC
);
1728 if (IS_ERR(verify_tfm
)) {
1730 retcode
= ERR_VERIFY_ALG
;
1734 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm
))) {
1735 retcode
= ERR_VERIFY_ALG_ND
;
1741 /* For now, use struct assignment, not pointer assignment.
1742 * We don't have any means to determine who might still
1743 * keep a local alias into the struct,
1744 * so we cannot just free it and hope for the best :(
1746 * To avoid someone looking at a half-updated struct, we probably
1747 * should have a rw-semaphor on net_conf and disk_conf.
1749 *tconn
->net_conf
= *new_conf
;
1752 crypto_free_hash(tconn
->csums_tfm
);
1753 tconn
->csums_tfm
= csums_tfm
;
1757 crypto_free_hash(tconn
->verify_tfm
);
1758 tconn
->verify_tfm
= verify_tfm
;
1762 if (tconn
->cstate
>= C_WF_REPORT_PARAMS
)
1763 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn
)));
1766 crypto_free_hash(csums_tfm
);
1767 crypto_free_hash(verify_tfm
);
1769 put_net_conf(tconn
);
1770 conn_reconfig_done(tconn
);
1772 drbd_adm_finish(info
, retcode
);
1776 int drbd_adm_connect(struct sk_buff
*skb
, struct genl_info
*info
)
1778 char hmac_name
[CRYPTO_MAX_ALG_NAME
];
1779 struct drbd_conf
*mdev
;
1780 struct net_conf
*new_conf
= NULL
;
1781 struct crypto_hash
*tfm
= NULL
;
1782 struct crypto_hash
*integrity_w_tfm
= NULL
;
1783 struct crypto_hash
*integrity_r_tfm
= NULL
;
1784 void *int_dig_out
= NULL
;
1785 void *int_dig_in
= NULL
;
1786 void *int_dig_vv
= NULL
;
1787 struct drbd_tconn
*oconn
;
1788 struct drbd_tconn
*tconn
;
1789 struct sockaddr
*new_my_addr
, *new_peer_addr
, *taken_addr
;
1790 enum drbd_ret_code retcode
;
1794 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_CONN
);
1795 if (!adm_ctx
.reply_skb
)
1797 if (retcode
!= NO_ERROR
)
1800 tconn
= adm_ctx
.tconn
;
1801 conn_reconfig_start(tconn
);
1803 if (tconn
->cstate
> C_STANDALONE
) {
1804 retcode
= ERR_NET_CONFIGURED
;
1808 /* allocation not in the IO path, cqueue thread context */
1809 new_conf
= kmalloc(sizeof(struct net_conf
), GFP_KERNEL
);
1811 retcode
= ERR_NOMEM
;
1815 *new_conf
= (struct net_conf
) {
1816 {}, 0, /* my_addr */
1817 {}, 0, /* peer_addr */
1818 {}, 0, /* shared_secret */
1819 {}, 0, /* cram_hmac_alg */
1820 {}, 0, /* integrity_alg */
1821 {}, 0, /* verify_alg */
1822 {}, 0, /* csums_alg */
1823 DRBD_PROTOCOL_DEF
, /* wire_protocol */
1824 DRBD_CONNECT_INT_DEF
, /* try_connect_int */
1825 DRBD_TIMEOUT_DEF
, /* timeout */
1826 DRBD_PING_INT_DEF
, /* ping_int */
1827 DRBD_PING_TIMEO_DEF
, /* ping_timeo */
1828 DRBD_SNDBUF_SIZE_DEF
, /* sndbuf_size */
1829 DRBD_RCVBUF_SIZE_DEF
, /* rcvbuf_size */
1830 DRBD_KO_COUNT_DEF
, /* ko_count */
1831 DRBD_MAX_BUFFERS_DEF
, /* max_buffers */
1832 DRBD_MAX_EPOCH_SIZE_DEF
, /* max_epoch_size */
1833 DRBD_UNPLUG_WATERMARK_DEF
, /* unplug_watermark */
1834 DRBD_AFTER_SB_0P_DEF
, /* after_sb_0p */
1835 DRBD_AFTER_SB_1P_DEF
, /* after_sb_1p */
1836 DRBD_AFTER_SB_2P_DEF
, /* after_sb_2p */
1837 DRBD_RR_CONFLICT_DEF
, /* rr_conflict */
1838 DRBD_ON_CONGESTION_DEF
, /* on_congestion */
1839 DRBD_CONG_FILL_DEF
, /* cong_fill */
1840 DRBD_CONG_EXTENTS_DEF
, /* cong_extents */
1841 0, /* two_primaries */
1844 0, /* always_asbp */
1849 err
= net_conf_from_attrs(new_conf
, info
);
1851 retcode
= ERR_MANDATORY_TAG
;
1852 drbd_msg_put_info(from_attrs_err_to_txt(err
));
1856 if (new_conf
->two_primaries
1857 && (new_conf
->wire_protocol
!= DRBD_PROT_C
)) {
1858 retcode
= ERR_NOT_PROTO_C
;
1862 idr_for_each_entry(&tconn
->volumes
, mdev
, i
) {
1863 if (get_ldev(mdev
)) {
1864 enum drbd_fencing_p fp
= mdev
->ldev
->dc
.fencing
;
1866 if (new_conf
->wire_protocol
== DRBD_PROT_A
&& fp
== FP_STONITH
) {
1867 retcode
= ERR_STONITH_AND_PROT_A
;
1871 if (mdev
->state
.role
== R_PRIMARY
&& new_conf
->want_lose
) {
1872 retcode
= ERR_DISCARD
;
1875 if (!mdev
->bitmap
) {
1876 if(drbd_bm_init(mdev
)) {
1877 retcode
= ERR_NOMEM
;
1883 if (new_conf
->on_congestion
!= OC_BLOCK
&& new_conf
->wire_protocol
!= DRBD_PROT_A
) {
1884 retcode
= ERR_CONG_NOT_PROTO_A
;
1890 new_my_addr
= (struct sockaddr
*)&new_conf
->my_addr
;
1891 new_peer_addr
= (struct sockaddr
*)&new_conf
->peer_addr
;
1893 /* No need to take drbd_cfg_mutex here. All reconfiguration is
1894 * strictly serialized on genl_lock(). We are protected against
1895 * concurrent reconfiguration/addition/deletion */
1896 list_for_each_entry(oconn
, &drbd_tconns
, all_tconn
) {
1899 if (get_net_conf(oconn
)) {
1900 taken_addr
= (struct sockaddr
*)&oconn
->net_conf
->my_addr
;
1901 if (new_conf
->my_addr_len
== oconn
->net_conf
->my_addr_len
&&
1902 !memcmp(new_my_addr
, taken_addr
, new_conf
->my_addr_len
))
1903 retcode
= ERR_LOCAL_ADDR
;
1905 taken_addr
= (struct sockaddr
*)&oconn
->net_conf
->peer_addr
;
1906 if (new_conf
->peer_addr_len
== oconn
->net_conf
->peer_addr_len
&&
1907 !memcmp(new_peer_addr
, taken_addr
, new_conf
->peer_addr_len
))
1908 retcode
= ERR_PEER_ADDR
;
1910 put_net_conf(oconn
);
1911 if (retcode
!= NO_ERROR
)
1916 if (new_conf
->cram_hmac_alg
[0] != 0) {
1917 snprintf(hmac_name
, CRYPTO_MAX_ALG_NAME
, "hmac(%s)",
1918 new_conf
->cram_hmac_alg
);
1919 tfm
= crypto_alloc_hash(hmac_name
, 0, CRYPTO_ALG_ASYNC
);
1922 retcode
= ERR_AUTH_ALG
;
1926 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm
))) {
1927 retcode
= ERR_AUTH_ALG_ND
;
1932 if (new_conf
->integrity_alg
[0]) {
1933 integrity_w_tfm
= crypto_alloc_hash(new_conf
->integrity_alg
, 0, CRYPTO_ALG_ASYNC
);
1934 if (IS_ERR(integrity_w_tfm
)) {
1935 integrity_w_tfm
= NULL
;
1936 retcode
=ERR_INTEGRITY_ALG
;
1940 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm
))) {
1941 retcode
=ERR_INTEGRITY_ALG_ND
;
1945 integrity_r_tfm
= crypto_alloc_hash(new_conf
->integrity_alg
, 0, CRYPTO_ALG_ASYNC
);
1946 if (IS_ERR(integrity_r_tfm
)) {
1947 integrity_r_tfm
= NULL
;
1948 retcode
=ERR_INTEGRITY_ALG
;
1953 ((char *)new_conf
->shared_secret
)[SHARED_SECRET_MAX
-1] = 0;
1955 /* allocation not in the IO path, cqueue thread context */
1956 if (integrity_w_tfm
) {
1957 i
= crypto_hash_digestsize(integrity_w_tfm
);
1958 int_dig_out
= kmalloc(i
, GFP_KERNEL
);
1960 retcode
= ERR_NOMEM
;
1963 int_dig_in
= kmalloc(i
, GFP_KERNEL
);
1965 retcode
= ERR_NOMEM
;
1968 int_dig_vv
= kmalloc(i
, GFP_KERNEL
);
1970 retcode
= ERR_NOMEM
;
1975 conn_flush_workqueue(tconn
);
1976 spin_lock_irq(&tconn
->req_lock
);
1977 if (tconn
->net_conf
!= NULL
) {
1978 retcode
= ERR_NET_CONFIGURED
;
1979 spin_unlock_irq(&tconn
->req_lock
);
1982 tconn
->net_conf
= new_conf
;
1984 crypto_free_hash(tconn
->cram_hmac_tfm
);
1985 tconn
->cram_hmac_tfm
= tfm
;
1987 crypto_free_hash(tconn
->integrity_w_tfm
);
1988 tconn
->integrity_w_tfm
= integrity_w_tfm
;
1990 crypto_free_hash(tconn
->integrity_r_tfm
);
1991 tconn
->integrity_r_tfm
= integrity_r_tfm
;
1993 kfree(tconn
->int_dig_out
);
1994 kfree(tconn
->int_dig_in
);
1995 kfree(tconn
->int_dig_vv
);
1996 tconn
->int_dig_out
=int_dig_out
;
1997 tconn
->int_dig_in
=int_dig_in
;
1998 tconn
->int_dig_vv
=int_dig_vv
;
1999 retcode
= _conn_request_state(tconn
, NS(conn
, C_UNCONNECTED
), CS_VERBOSE
);
2000 spin_unlock_irq(&tconn
->req_lock
);
2002 idr_for_each_entry(&tconn
->volumes
, mdev
, i
) {
2005 kobject_uevent(&disk_to_dev(mdev
->vdisk
)->kobj
, KOBJ_CHANGE
);
2007 conn_reconfig_done(tconn
);
2008 drbd_adm_finish(info
, retcode
);
2015 crypto_free_hash(tfm
);
2016 crypto_free_hash(integrity_w_tfm
);
2017 crypto_free_hash(integrity_r_tfm
);
2020 conn_reconfig_done(tconn
);
2022 drbd_adm_finish(info
, retcode
);
2026 static enum drbd_state_rv
conn_try_disconnect(struct drbd_tconn
*tconn
, bool force
)
2028 enum drbd_state_rv rv
;
2030 spin_lock_irq(&tconn
->req_lock
);
2031 if (tconn
->cstate
>= C_WF_CONNECTION
)
2032 _conn_request_state(tconn
, NS(conn
, C_DISCONNECTING
), CS_HARD
);
2033 spin_unlock_irq(&tconn
->req_lock
);
2037 rv
= conn_request_state(tconn
, NS(conn
, C_DISCONNECTING
), 0);
2040 case SS_NOTHING_TO_DO
:
2041 case SS_ALREADY_STANDALONE
:
2043 case SS_PRIMARY_NOP
:
2044 /* Our state checking code wants to see the peer outdated. */
2045 rv
= conn_request_state(tconn
, NS2(conn
, C_DISCONNECTING
,
2046 pdsk
, D_OUTDATED
), CS_VERBOSE
);
2048 case SS_CW_FAILED_BY_PEER
:
2049 /* The peer probably wants to see us outdated. */
2050 rv
= conn_request_state(tconn
, NS2(conn
, C_DISCONNECTING
,
2051 disk
, D_OUTDATED
), 0);
2052 if (rv
== SS_IS_DISKLESS
|| rv
== SS_LOWER_THAN_OUTDATED
) {
2053 conn_request_state(tconn
, NS(conn
, C_DISCONNECTING
), CS_HARD
);
2058 /* no special handling necessary */
2064 int drbd_adm_disconnect(struct sk_buff
*skb
, struct genl_info
*info
)
2066 struct disconnect_parms parms
;
2067 struct drbd_tconn
*tconn
;
2068 enum drbd_state_rv rv
;
2069 enum drbd_ret_code retcode
;
2072 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_CONN
);
2073 if (!adm_ctx
.reply_skb
)
2075 if (retcode
!= NO_ERROR
)
2078 tconn
= adm_ctx
.tconn
;
2079 memset(&parms
, 0, sizeof(parms
));
2080 if (info
->attrs
[DRBD_NLA_DISCONNECT_PARMS
]) {
2081 err
= disconnect_parms_from_attrs(&parms
, info
);
2083 retcode
= ERR_MANDATORY_TAG
;
2084 drbd_msg_put_info(from_attrs_err_to_txt(err
));
2089 rv
= conn_try_disconnect(tconn
, parms
.force_disconnect
);
2090 if (rv
< SS_SUCCESS
)
2093 if (wait_event_interruptible(tconn
->ping_wait
,
2094 tconn
->cstate
!= C_DISCONNECTING
)) {
2095 /* Do not test for mdev->state.conn == C_STANDALONE, since
2096 someone else might connect us in the mean time! */
2103 drbd_adm_finish(info
, retcode
);
2107 void resync_after_online_grow(struct drbd_conf
*mdev
)
2109 int iass
; /* I am sync source */
2111 dev_info(DEV
, "Resync of new storage after online grow\n");
2112 if (mdev
->state
.role
!= mdev
->state
.peer
)
2113 iass
= (mdev
->state
.role
== R_PRIMARY
);
2115 iass
= test_bit(DISCARD_CONCURRENT
, &mdev
->tconn
->flags
);
2118 drbd_start_resync(mdev
, C_SYNC_SOURCE
);
2120 _drbd_request_state(mdev
, NS(conn
, C_WF_SYNC_UUID
), CS_VERBOSE
+ CS_SERIALIZE
);
2123 int drbd_adm_resize(struct sk_buff
*skb
, struct genl_info
*info
)
2125 struct resize_parms rs
;
2126 struct drbd_conf
*mdev
;
2127 enum drbd_ret_code retcode
;
2128 enum determine_dev_size dd
;
2129 enum dds_flags ddsf
;
2132 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
2133 if (!adm_ctx
.reply_skb
)
2135 if (retcode
!= NO_ERROR
)
2138 memset(&rs
, 0, sizeof(struct resize_parms
));
2139 if (info
->attrs
[DRBD_NLA_RESIZE_PARMS
]) {
2140 err
= resize_parms_from_attrs(&rs
, info
);
2142 retcode
= ERR_MANDATORY_TAG
;
2143 drbd_msg_put_info(from_attrs_err_to_txt(err
));
2148 mdev
= adm_ctx
.mdev
;
2149 if (mdev
->state
.conn
> C_CONNECTED
) {
2150 retcode
= ERR_RESIZE_RESYNC
;
2154 if (mdev
->state
.role
== R_SECONDARY
&&
2155 mdev
->state
.peer
== R_SECONDARY
) {
2156 retcode
= ERR_NO_PRIMARY
;
2160 if (!get_ldev(mdev
)) {
2161 retcode
= ERR_NO_DISK
;
2165 if (rs
.no_resync
&& mdev
->tconn
->agreed_pro_version
< 93) {
2166 retcode
= ERR_NEED_APV_93
;
2170 if (mdev
->ldev
->known_size
!= drbd_get_capacity(mdev
->ldev
->backing_bdev
))
2171 mdev
->ldev
->known_size
= drbd_get_capacity(mdev
->ldev
->backing_bdev
);
2173 mdev
->ldev
->dc
.disk_size
= (sector_t
)rs
.resize_size
;
2174 ddsf
= (rs
.resize_force
? DDSF_FORCED
: 0) | (rs
.no_resync
? DDSF_NO_RESYNC
: 0);
2175 dd
= drbd_determine_dev_size(mdev
, ddsf
);
2178 if (dd
== dev_size_error
) {
2179 retcode
= ERR_NOMEM_BITMAP
;
2183 if (mdev
->state
.conn
== C_CONNECTED
) {
2185 set_bit(RESIZE_PENDING
, &mdev
->flags
);
2187 drbd_send_uuids(mdev
);
2188 drbd_send_sizes(mdev
, 1, ddsf
);
2192 drbd_adm_finish(info
, retcode
);
2196 int drbd_adm_resource_opts(struct sk_buff
*skb
, struct genl_info
*info
)
2198 enum drbd_ret_code retcode
;
2199 cpumask_var_t new_cpu_mask
;
2200 struct drbd_tconn
*tconn
;
2201 int *rs_plan_s
= NULL
;
2205 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_CONN
);
2206 if (!adm_ctx
.reply_skb
)
2208 if (retcode
!= NO_ERROR
)
2210 tconn
= adm_ctx
.tconn
;
2212 if (!zalloc_cpumask_var(&new_cpu_mask
, GFP_KERNEL
)) {
2213 retcode
= ERR_NOMEM
;
2214 drbd_msg_put_info("unable to allocate cpumask");
2218 if (((struct drbd_genlmsghdr
*)info
->userhdr
)->flags
2219 & DRBD_GENL_F_SET_DEFAULTS
) {
2220 memset(&sc
, 0, sizeof(struct res_opts
));
2221 sc
.on_no_data
= DRBD_ON_NO_DATA_DEF
;
2223 sc
= tconn
->res_opts
;
2225 err
= res_opts_from_attrs(&sc
, info
);
2227 retcode
= ERR_MANDATORY_TAG
;
2228 drbd_msg_put_info(from_attrs_err_to_txt(err
));
2232 /* silently ignore cpu mask on UP kernel */
2233 if (nr_cpu_ids
> 1 && sc
.cpu_mask
[0] != 0) {
2234 err
= __bitmap_parse(sc
.cpu_mask
, 32, 0,
2235 cpumask_bits(new_cpu_mask
), nr_cpu_ids
);
2237 conn_warn(tconn
, "__bitmap_parse() failed with %d\n", err
);
2238 retcode
= ERR_CPU_MASK_PARSE
;
2244 tconn
->res_opts
= sc
;
2246 if (!cpumask_equal(tconn
->cpu_mask
, new_cpu_mask
)) {
2247 cpumask_copy(tconn
->cpu_mask
, new_cpu_mask
);
2248 drbd_calc_cpu_mask(tconn
);
2249 tconn
->receiver
.reset_cpu_mask
= 1;
2250 tconn
->asender
.reset_cpu_mask
= 1;
2251 tconn
->worker
.reset_cpu_mask
= 1;
2256 free_cpumask_var(new_cpu_mask
);
2258 drbd_adm_finish(info
, retcode
);
2262 int drbd_adm_invalidate(struct sk_buff
*skb
, struct genl_info
*info
)
2264 struct drbd_conf
*mdev
;
2265 int retcode
; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2267 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
2268 if (!adm_ctx
.reply_skb
)
2270 if (retcode
!= NO_ERROR
)
2273 mdev
= adm_ctx
.mdev
;
2275 /* If there is still bitmap IO pending, probably because of a previous
2276 * resync just being finished, wait for it before requesting a new resync. */
2277 wait_event(mdev
->misc_wait
, !test_bit(BITMAP_IO
, &mdev
->flags
));
2279 retcode
= _drbd_request_state(mdev
, NS(conn
, C_STARTING_SYNC_T
), CS_ORDERED
);
2281 if (retcode
< SS_SUCCESS
&& retcode
!= SS_NEED_CONNECTION
)
2282 retcode
= drbd_request_state(mdev
, NS(conn
, C_STARTING_SYNC_T
));
2284 while (retcode
== SS_NEED_CONNECTION
) {
2285 spin_lock_irq(&mdev
->tconn
->req_lock
);
2286 if (mdev
->state
.conn
< C_CONNECTED
)
2287 retcode
= _drbd_set_state(_NS(mdev
, disk
, D_INCONSISTENT
), CS_VERBOSE
, NULL
);
2288 spin_unlock_irq(&mdev
->tconn
->req_lock
);
2290 if (retcode
!= SS_NEED_CONNECTION
)
2293 retcode
= drbd_request_state(mdev
, NS(conn
, C_STARTING_SYNC_T
));
2297 drbd_adm_finish(info
, retcode
);
2301 static int drbd_bmio_set_susp_al(struct drbd_conf
*mdev
)
2305 rv
= drbd_bmio_set_n_write(mdev
);
2306 drbd_suspend_al(mdev
);
2310 static int drbd_adm_simple_request_state(struct sk_buff
*skb
, struct genl_info
*info
,
2311 union drbd_state mask
, union drbd_state val
)
2313 enum drbd_ret_code retcode
;
2315 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
2316 if (!adm_ctx
.reply_skb
)
2318 if (retcode
!= NO_ERROR
)
2321 retcode
= drbd_request_state(adm_ctx
.mdev
, mask
, val
);
2323 drbd_adm_finish(info
, retcode
);
2327 int drbd_adm_invalidate_peer(struct sk_buff
*skb
, struct genl_info
*info
)
2329 return drbd_adm_simple_request_state(skb
, info
, NS(conn
, C_STARTING_SYNC_S
));
2332 int drbd_adm_pause_sync(struct sk_buff
*skb
, struct genl_info
*info
)
2334 enum drbd_ret_code retcode
;
2336 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
2337 if (!adm_ctx
.reply_skb
)
2339 if (retcode
!= NO_ERROR
)
2342 if (drbd_request_state(adm_ctx
.mdev
, NS(user_isp
, 1)) == SS_NOTHING_TO_DO
)
2343 retcode
= ERR_PAUSE_IS_SET
;
2345 drbd_adm_finish(info
, retcode
);
2349 int drbd_adm_resume_sync(struct sk_buff
*skb
, struct genl_info
*info
)
2351 union drbd_dev_state s
;
2352 enum drbd_ret_code retcode
;
2354 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
2355 if (!adm_ctx
.reply_skb
)
2357 if (retcode
!= NO_ERROR
)
2360 if (drbd_request_state(adm_ctx
.mdev
, NS(user_isp
, 0)) == SS_NOTHING_TO_DO
) {
2361 s
= adm_ctx
.mdev
->state
;
2362 if (s
.conn
== C_PAUSED_SYNC_S
|| s
.conn
== C_PAUSED_SYNC_T
) {
2363 retcode
= s
.aftr_isp
? ERR_PIC_AFTER_DEP
:
2364 s
.peer_isp
? ERR_PIC_PEER_DEP
: ERR_PAUSE_IS_CLEAR
;
2366 retcode
= ERR_PAUSE_IS_CLEAR
;
2371 drbd_adm_finish(info
, retcode
);
2375 int drbd_adm_suspend_io(struct sk_buff
*skb
, struct genl_info
*info
)
2377 return drbd_adm_simple_request_state(skb
, info
, NS(susp
, 1));
2380 int drbd_adm_resume_io(struct sk_buff
*skb
, struct genl_info
*info
)
2382 struct drbd_conf
*mdev
;
2383 int retcode
; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2385 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
2386 if (!adm_ctx
.reply_skb
)
2388 if (retcode
!= NO_ERROR
)
2391 mdev
= adm_ctx
.mdev
;
2392 if (test_bit(NEW_CUR_UUID
, &mdev
->flags
)) {
2393 drbd_uuid_new_current(mdev
);
2394 clear_bit(NEW_CUR_UUID
, &mdev
->flags
);
2396 drbd_suspend_io(mdev
);
2397 retcode
= drbd_request_state(mdev
, NS3(susp
, 0, susp_nod
, 0, susp_fen
, 0));
2398 if (retcode
== SS_SUCCESS
) {
2399 if (mdev
->state
.conn
< C_CONNECTED
)
2400 tl_clear(mdev
->tconn
);
2401 if (mdev
->state
.disk
== D_DISKLESS
|| mdev
->state
.disk
== D_FAILED
)
2402 tl_restart(mdev
->tconn
, FAIL_FROZEN_DISK_IO
);
2404 drbd_resume_io(mdev
);
2407 drbd_adm_finish(info
, retcode
);
2411 int drbd_adm_outdate(struct sk_buff
*skb
, struct genl_info
*info
)
2413 return drbd_adm_simple_request_state(skb
, info
, NS(disk
, D_OUTDATED
));
2416 int nla_put_drbd_cfg_context(struct sk_buff
*skb
, const char *conn_name
, unsigned vnr
)
2419 nla
= nla_nest_start(skb
, DRBD_NLA_CFG_CONTEXT
);
2421 goto nla_put_failure
;
2422 if (vnr
!= VOLUME_UNSPECIFIED
)
2423 NLA_PUT_U32(skb
, T_ctx_volume
, vnr
);
2424 NLA_PUT_STRING(skb
, T_ctx_conn_name
, conn_name
);
2425 nla_nest_end(skb
, nla
);
2430 nla_nest_cancel(skb
, nla
);
2434 int nla_put_status_info(struct sk_buff
*skb
, struct drbd_conf
*mdev
,
2435 const struct sib_info
*sib
)
2437 struct state_info
*si
= NULL
; /* for sizeof(si->member); */
2442 int exclude_sensitive
;
2444 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2445 * to. So we better exclude_sensitive information.
2447 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2448 * in the context of the requesting user process. Exclude sensitive
2449 * information, unless current has superuser.
2451 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2452 * relies on the current implementation of netlink_dump(), which
2453 * executes the dump callback successively from netlink_recvmsg(),
2454 * always in the context of the receiving process */
2455 exclude_sensitive
= sib
|| !capable(CAP_SYS_ADMIN
);
2457 got_ldev
= get_ldev(mdev
);
2458 got_net
= get_net_conf(mdev
->tconn
);
2460 /* We need to add connection name and volume number information still.
2461 * Minor number is in drbd_genlmsghdr. */
2462 if (nla_put_drbd_cfg_context(skb
, mdev
->tconn
->name
, mdev
->vnr
))
2463 goto nla_put_failure
;
2465 if (res_opts_to_skb(skb
, &mdev
->tconn
->res_opts
, exclude_sensitive
))
2466 goto nla_put_failure
;
2469 if (disk_conf_to_skb(skb
, &mdev
->ldev
->dc
, exclude_sensitive
))
2470 goto nla_put_failure
;
2472 if (net_conf_to_skb(skb
, mdev
->tconn
->net_conf
, exclude_sensitive
))
2473 goto nla_put_failure
;
2475 nla
= nla_nest_start(skb
, DRBD_NLA_STATE_INFO
);
2477 goto nla_put_failure
;
2478 NLA_PUT_U32(skb
, T_sib_reason
, sib
? sib
->sib_reason
: SIB_GET_STATUS_REPLY
);
2479 NLA_PUT_U32(skb
, T_current_state
, mdev
->state
.i
);
2480 NLA_PUT_U64(skb
, T_ed_uuid
, mdev
->ed_uuid
);
2481 NLA_PUT_U64(skb
, T_capacity
, drbd_get_capacity(mdev
->this_bdev
));
2484 NLA_PUT_U32(skb
, T_disk_flags
, mdev
->ldev
->md
.flags
);
2485 NLA_PUT(skb
, T_uuids
, sizeof(si
->uuids
), mdev
->ldev
->md
.uuid
);
2486 NLA_PUT_U64(skb
, T_bits_total
, drbd_bm_bits(mdev
));
2487 NLA_PUT_U64(skb
, T_bits_oos
, drbd_bm_total_weight(mdev
));
2488 if (C_SYNC_SOURCE
<= mdev
->state
.conn
&&
2489 C_PAUSED_SYNC_T
>= mdev
->state
.conn
) {
2490 NLA_PUT_U64(skb
, T_bits_rs_total
, mdev
->rs_total
);
2491 NLA_PUT_U64(skb
, T_bits_rs_failed
, mdev
->rs_failed
);
2496 switch(sib
->sib_reason
) {
2497 case SIB_SYNC_PROGRESS
:
2498 case SIB_GET_STATUS_REPLY
:
2500 case SIB_STATE_CHANGE
:
2501 NLA_PUT_U32(skb
, T_prev_state
, sib
->os
.i
);
2502 NLA_PUT_U32(skb
, T_new_state
, sib
->ns
.i
);
2504 case SIB_HELPER_POST
:
2506 T_helper_exit_code
, sib
->helper_exit_code
);
2508 case SIB_HELPER_PRE
:
2509 NLA_PUT_STRING(skb
, T_helper
, sib
->helper_name
);
2513 nla_nest_end(skb
, nla
);
2521 put_net_conf(mdev
->tconn
);
2525 int drbd_adm_get_status(struct sk_buff
*skb
, struct genl_info
*info
)
2527 enum drbd_ret_code retcode
;
2530 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
2531 if (!adm_ctx
.reply_skb
)
2533 if (retcode
!= NO_ERROR
)
2536 err
= nla_put_status_info(adm_ctx
.reply_skb
, adm_ctx
.mdev
, NULL
);
2538 nlmsg_free(adm_ctx
.reply_skb
);
2542 drbd_adm_finish(info
, retcode
);
2546 int drbd_adm_get_status_all(struct sk_buff
*skb
, struct netlink_callback
*cb
)
2548 struct drbd_conf
*mdev
;
2549 struct drbd_genlmsghdr
*dh
;
2550 struct drbd_tconn
*pos
= (struct drbd_tconn
*)cb
->args
[0];
2551 struct drbd_tconn
*tconn
= NULL
;
2552 struct drbd_tconn
*tmp
;
2553 unsigned volume
= cb
->args
[1];
2555 /* Open coded, deferred, iteration:
2556 * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2557 * idr_for_each_entry(&tconn->volumes, mdev, i) {
2561 * where tconn is cb->args[0];
2562 * and i is cb->args[1];
2564 * This may miss entries inserted after this dump started,
2565 * or entries deleted before they are reached.
2567 * We need to make sure the mdev won't disappear while
2568 * we are looking at it, and revalidate our iterators
2569 * on each iteration.
2572 /* synchronize with drbd_new_tconn/drbd_free_tconn */
2573 mutex_lock(&drbd_cfg_mutex
);
2574 /* synchronize with drbd_delete_device */
2577 /* revalidate iterator position */
2578 list_for_each_entry(tmp
, &drbd_tconns
, all_tconn
) {
2580 /* first iteration */
2591 mdev
= idr_get_next(&tconn
->volumes
, &volume
);
2593 /* No more volumes to dump on this tconn.
2594 * Advance tconn iterator. */
2595 pos
= list_entry(tconn
->all_tconn
.next
,
2596 struct drbd_tconn
, all_tconn
);
2597 /* But, did we dump any volume on this tconn yet? */
2605 dh
= genlmsg_put(skb
, NETLINK_CB(cb
->skb
).pid
,
2606 cb
->nlh
->nlmsg_seq
, &drbd_genl_family
,
2607 NLM_F_MULTI
, DRBD_ADM_GET_STATUS
);
2612 /* this is a tconn without a single volume */
2614 dh
->ret_code
= NO_ERROR
;
2615 if (nla_put_drbd_cfg_context(skb
, tconn
->name
, VOLUME_UNSPECIFIED
))
2616 genlmsg_cancel(skb
, dh
);
2618 genlmsg_end(skb
, dh
);
2622 D_ASSERT(mdev
->vnr
== volume
);
2623 D_ASSERT(mdev
->tconn
== tconn
);
2625 dh
->minor
= mdev_to_minor(mdev
);
2626 dh
->ret_code
= NO_ERROR
;
2628 if (nla_put_status_info(skb
, mdev
, NULL
)) {
2629 genlmsg_cancel(skb
, dh
);
2632 genlmsg_end(skb
, dh
);
2637 mutex_unlock(&drbd_cfg_mutex
);
2638 /* where to start the next iteration */
2639 cb
->args
[0] = (long)pos
;
2640 cb
->args
[1] = (pos
== tconn
) ? volume
+ 1 : 0;
2642 /* No more tconns/volumes/minors found results in an empty skb.
2643 * Which will terminate the dump. */
2647 int drbd_adm_get_timeout_type(struct sk_buff
*skb
, struct genl_info
*info
)
2649 enum drbd_ret_code retcode
;
2650 struct timeout_parms tp
;
2653 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
2654 if (!adm_ctx
.reply_skb
)
2656 if (retcode
!= NO_ERROR
)
2660 adm_ctx
.mdev
->state
.pdsk
== D_OUTDATED
? UT_PEER_OUTDATED
:
2661 test_bit(USE_DEGR_WFC_T
, &adm_ctx
.mdev
->flags
) ? UT_DEGRADED
:
2664 err
= timeout_parms_to_priv_skb(adm_ctx
.reply_skb
, &tp
);
2666 nlmsg_free(adm_ctx
.reply_skb
);
2670 drbd_adm_finish(info
, retcode
);
2674 int drbd_adm_start_ov(struct sk_buff
*skb
, struct genl_info
*info
)
2676 struct drbd_conf
*mdev
;
2677 enum drbd_ret_code retcode
;
2679 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
2680 if (!adm_ctx
.reply_skb
)
2682 if (retcode
!= NO_ERROR
)
2685 mdev
= adm_ctx
.mdev
;
2686 if (info
->attrs
[DRBD_NLA_START_OV_PARMS
]) {
2687 /* resume from last known position, if possible */
2688 struct start_ov_parms parms
=
2689 { .ov_start_sector
= mdev
->ov_start_sector
};
2690 int err
= start_ov_parms_from_attrs(&parms
, info
);
2692 retcode
= ERR_MANDATORY_TAG
;
2693 drbd_msg_put_info(from_attrs_err_to_txt(err
));
2696 /* w_make_ov_request expects position to be aligned */
2697 mdev
->ov_start_sector
= parms
.ov_start_sector
& ~BM_SECT_PER_BIT
;
2699 /* If there is still bitmap IO pending, e.g. previous resync or verify
2700 * just being finished, wait for it before requesting a new resync. */
2701 wait_event(mdev
->misc_wait
, !test_bit(BITMAP_IO
, &mdev
->flags
));
2702 retcode
= drbd_request_state(mdev
,NS(conn
,C_VERIFY_S
));
2704 drbd_adm_finish(info
, retcode
);
2709 int drbd_adm_new_c_uuid(struct sk_buff
*skb
, struct genl_info
*info
)
2711 struct drbd_conf
*mdev
;
2712 enum drbd_ret_code retcode
;
2713 int skip_initial_sync
= 0;
2715 struct new_c_uuid_parms args
;
2717 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
2718 if (!adm_ctx
.reply_skb
)
2720 if (retcode
!= NO_ERROR
)
2723 mdev
= adm_ctx
.mdev
;
2724 memset(&args
, 0, sizeof(args
));
2725 if (info
->attrs
[DRBD_NLA_NEW_C_UUID_PARMS
]) {
2726 err
= new_c_uuid_parms_from_attrs(&args
, info
);
2728 retcode
= ERR_MANDATORY_TAG
;
2729 drbd_msg_put_info(from_attrs_err_to_txt(err
));
2734 mutex_lock(mdev
->state_mutex
); /* Protects us against serialized state changes. */
2736 if (!get_ldev(mdev
)) {
2737 retcode
= ERR_NO_DISK
;
2741 /* this is "skip initial sync", assume to be clean */
2742 if (mdev
->state
.conn
== C_CONNECTED
&& mdev
->tconn
->agreed_pro_version
>= 90 &&
2743 mdev
->ldev
->md
.uuid
[UI_CURRENT
] == UUID_JUST_CREATED
&& args
.clear_bm
) {
2744 dev_info(DEV
, "Preparing to skip initial sync\n");
2745 skip_initial_sync
= 1;
2746 } else if (mdev
->state
.conn
!= C_STANDALONE
) {
2747 retcode
= ERR_CONNECTED
;
2751 drbd_uuid_set(mdev
, UI_BITMAP
, 0); /* Rotate UI_BITMAP to History 1, etc... */
2752 drbd_uuid_new_current(mdev
); /* New current, previous to UI_BITMAP */
2754 if (args
.clear_bm
) {
2755 err
= drbd_bitmap_io(mdev
, &drbd_bmio_clear_n_write
,
2756 "clear_n_write from new_c_uuid", BM_LOCKED_MASK
);
2758 dev_err(DEV
, "Writing bitmap failed with %d\n",err
);
2759 retcode
= ERR_IO_MD_DISK
;
2761 if (skip_initial_sync
) {
2762 drbd_send_uuids_skip_initial_sync(mdev
);
2763 _drbd_uuid_set(mdev
, UI_BITMAP
, 0);
2764 drbd_print_uuids(mdev
, "cleared bitmap UUID");
2765 spin_lock_irq(&mdev
->tconn
->req_lock
);
2766 _drbd_set_state(_NS2(mdev
, disk
, D_UP_TO_DATE
, pdsk
, D_UP_TO_DATE
),
2768 spin_unlock_irq(&mdev
->tconn
->req_lock
);
2776 mutex_unlock(mdev
->state_mutex
);
2778 drbd_adm_finish(info
, retcode
);
2782 static enum drbd_ret_code
2783 drbd_check_conn_name(const char *name
)
2785 if (!name
|| !name
[0]) {
2786 drbd_msg_put_info("connection name missing");
2787 return ERR_MANDATORY_TAG
;
2789 /* if we want to use these in sysfs/configfs/debugfs some day,
2790 * we must not allow slashes */
2791 if (strchr(name
, '/')) {
2792 drbd_msg_put_info("invalid connection name");
2793 return ERR_INVALID_REQUEST
;
2798 int drbd_adm_create_connection(struct sk_buff
*skb
, struct genl_info
*info
)
2800 enum drbd_ret_code retcode
;
2802 retcode
= drbd_adm_prepare(skb
, info
, 0);
2803 if (!adm_ctx
.reply_skb
)
2805 if (retcode
!= NO_ERROR
)
2808 retcode
= drbd_check_conn_name(adm_ctx
.conn_name
);
2809 if (retcode
!= NO_ERROR
)
2812 if (adm_ctx
.tconn
) {
2813 if (info
->nlhdr
->nlmsg_flags
& NLM_F_EXCL
) {
2814 retcode
= ERR_INVALID_REQUEST
;
2815 drbd_msg_put_info("connection exists");
2817 /* else: still NO_ERROR */
2821 if (!drbd_new_tconn(adm_ctx
.conn_name
))
2822 retcode
= ERR_NOMEM
;
2824 drbd_adm_finish(info
, retcode
);
2828 int drbd_adm_add_minor(struct sk_buff
*skb
, struct genl_info
*info
)
2830 struct drbd_genlmsghdr
*dh
= info
->userhdr
;
2831 enum drbd_ret_code retcode
;
2833 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_CONN
);
2834 if (!adm_ctx
.reply_skb
)
2836 if (retcode
!= NO_ERROR
)
2839 /* FIXME drop minor_count parameter, limit to MINORMASK */
2840 if (dh
->minor
>= minor_count
) {
2841 drbd_msg_put_info("requested minor out of range");
2842 retcode
= ERR_INVALID_REQUEST
;
2845 /* FIXME we need a define here */
2846 if (adm_ctx
.volume
>= 256) {
2847 drbd_msg_put_info("requested volume id out of range");
2848 retcode
= ERR_INVALID_REQUEST
;
2852 /* drbd_adm_prepare made sure already
2853 * that mdev->tconn and mdev->vnr match the request. */
2855 if (info
->nlhdr
->nlmsg_flags
& NLM_F_EXCL
)
2856 retcode
= ERR_MINOR_EXISTS
;
2857 /* else: still NO_ERROR */
2861 retcode
= conn_new_minor(adm_ctx
.tconn
, dh
->minor
, adm_ctx
.volume
);
2863 drbd_adm_finish(info
, retcode
);
2867 static enum drbd_ret_code
adm_delete_minor(struct drbd_conf
*mdev
)
2869 if (mdev
->state
.disk
== D_DISKLESS
&&
2870 /* no need to be mdev->state.conn == C_STANDALONE &&
2871 * we may want to delete a minor from a live replication group.
2873 mdev
->state
.role
== R_SECONDARY
) {
2874 drbd_delete_device(mdev_to_minor(mdev
));
2877 return ERR_MINOR_CONFIGURED
;
2880 int drbd_adm_delete_minor(struct sk_buff
*skb
, struct genl_info
*info
)
2882 enum drbd_ret_code retcode
;
2884 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_MINOR
);
2885 if (!adm_ctx
.reply_skb
)
2887 if (retcode
!= NO_ERROR
)
2890 mutex_lock(&drbd_cfg_mutex
);
2891 retcode
= adm_delete_minor(adm_ctx
.mdev
);
2892 mutex_unlock(&drbd_cfg_mutex
);
2893 /* if this was the last volume of this connection,
2894 * this will terminate all threads */
2895 if (retcode
== NO_ERROR
)
2896 conn_reconfig_done(adm_ctx
.tconn
);
2898 drbd_adm_finish(info
, retcode
);
2902 int drbd_adm_down(struct sk_buff
*skb
, struct genl_info
*info
)
2904 enum drbd_ret_code retcode
;
2905 enum drbd_state_rv rv
;
2906 struct drbd_conf
*mdev
;
2909 retcode
= drbd_adm_prepare(skb
, info
, 0);
2910 if (!adm_ctx
.reply_skb
)
2912 if (retcode
!= NO_ERROR
)
2915 if (!adm_ctx
.tconn
) {
2916 retcode
= ERR_CONN_NOT_KNOWN
;
2920 mutex_lock(&drbd_cfg_mutex
);
2922 idr_for_each_entry(&adm_ctx
.tconn
->volumes
, mdev
, i
) {
2923 retcode
= drbd_set_role(mdev
, R_SECONDARY
, 0);
2924 if (retcode
< SS_SUCCESS
) {
2925 drbd_msg_put_info("failed to demote");
2931 rv
= conn_try_disconnect(adm_ctx
.tconn
, 0);
2932 if (rv
< SS_SUCCESS
) {
2933 retcode
= rv
; /* enum type mismatch! */
2934 drbd_msg_put_info("failed to disconnect");
2939 idr_for_each_entry(&adm_ctx
.tconn
->volumes
, mdev
, i
) {
2940 rv
= adm_detach(mdev
);
2941 if (rv
< SS_SUCCESS
) {
2942 retcode
= rv
; /* enum type mismatch! */
2943 drbd_msg_put_info("failed to detach");
2948 /* delete volumes */
2949 idr_for_each_entry(&adm_ctx
.tconn
->volumes
, mdev
, i
) {
2950 retcode
= adm_delete_minor(mdev
);
2951 if (retcode
!= NO_ERROR
) {
2952 /* "can not happen" */
2953 drbd_msg_put_info("failed to delete volume");
2958 /* stop all threads */
2959 conn_reconfig_done(adm_ctx
.tconn
);
2961 /* delete connection */
2962 if (conn_lowest_minor(adm_ctx
.tconn
) < 0) {
2963 drbd_free_tconn(adm_ctx
.tconn
);
2966 /* "can not happen" */
2967 retcode
= ERR_CONN_IN_USE
;
2968 drbd_msg_put_info("failed to delete connection");
2972 mutex_unlock(&drbd_cfg_mutex
);
2974 drbd_adm_finish(info
, retcode
);
2978 int drbd_adm_delete_connection(struct sk_buff
*skb
, struct genl_info
*info
)
2980 enum drbd_ret_code retcode
;
2982 retcode
= drbd_adm_prepare(skb
, info
, DRBD_ADM_NEED_CONN
);
2983 if (!adm_ctx
.reply_skb
)
2985 if (retcode
!= NO_ERROR
)
2988 mutex_lock(&drbd_cfg_mutex
);
2989 if (conn_lowest_minor(adm_ctx
.tconn
) < 0) {
2990 drbd_free_tconn(adm_ctx
.tconn
);
2993 retcode
= ERR_CONN_IN_USE
;
2995 mutex_unlock(&drbd_cfg_mutex
);
2998 drbd_adm_finish(info
, retcode
);
3002 void drbd_bcast_event(struct drbd_conf
*mdev
, const struct sib_info
*sib
)
3004 static atomic_t drbd_genl_seq
= ATOMIC_INIT(2); /* two. */
3005 struct sk_buff
*msg
;
3006 struct drbd_genlmsghdr
*d_out
;
3010 seq
= atomic_inc_return(&drbd_genl_seq
);
3011 msg
= genlmsg_new(NLMSG_GOODSIZE
, GFP_NOIO
);
3016 d_out
= genlmsg_put(msg
, 0, seq
, &drbd_genl_family
, 0, DRBD_EVENT
);
3017 if (!d_out
) /* cannot happen, but anyways. */
3018 goto nla_put_failure
;
3019 d_out
->minor
= mdev_to_minor(mdev
);
3020 d_out
->ret_code
= 0;
3022 if (nla_put_status_info(msg
, mdev
, sib
))
3023 goto nla_put_failure
;
3024 genlmsg_end(msg
, d_out
);
3025 err
= drbd_genl_multicast_events(msg
, 0);
3026 /* msg has been consumed or freed in netlink_broadcast() */
3027 if (err
&& err
!= -ESRCH
)
3035 dev_err(DEV
, "Error %d while broadcasting event. "
3036 "Event seq:%u sib_reason:%u\n",
3037 err
, seq
, sib
->sib_reason
);