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/connector.h>
33 #include <linux/blkpg.h>
34 #include <linux/cpumask.h>
37 #include "drbd_wrappers.h"
38 #include <asm/unaligned.h>
39 #include <linux/drbd_tag_magic.h>
40 #include <linux/drbd_limits.h>
41 #include <linux/compiler.h>
42 #include <linux/kthread.h>
44 static unsigned short *tl_add_blob(unsigned short *, enum drbd_tags
, const void *, int);
45 static unsigned short *tl_add_str(unsigned short *, enum drbd_tags
, const char *);
46 static unsigned short *tl_add_int(unsigned short *, enum drbd_tags
, const void *);
48 /* see get_sb_bdev and bd_claim */
49 static char *drbd_m_holder
= "Hands off! this is DRBD's meta data device.";
51 /* Generate the tag_list to struct functions */
52 #define NL_PACKET(name, number, fields) \
53 static int name ## _from_tags(struct drbd_conf *mdev, \
54 unsigned short *tags, struct name *arg) __attribute__ ((unused)); \
55 static int name ## _from_tags(struct drbd_conf *mdev, \
56 unsigned short *tags, struct name *arg) \
61 while ((tag = get_unaligned(tags++)) != TT_END) { \
62 dlen = get_unaligned(tags++); \
63 switch (tag_number(tag)) { \
66 if (tag & T_MANDATORY) { \
67 dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \
71 tags = (unsigned short *)((char *)tags + dlen); \
75 #define NL_INTEGER(pn, pr, member) \
76 case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \
77 arg->member = get_unaligned((int *)(tags)); \
79 #define NL_INT64(pn, pr, member) \
80 case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \
81 arg->member = get_unaligned((u64 *)(tags)); \
83 #define NL_BIT(pn, pr, member) \
84 case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \
85 arg->member = *(char *)(tags) ? 1 : 0; \
87 #define NL_STRING(pn, pr, member, len) \
88 case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \
90 dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \
91 #member, dlen, (unsigned int)len); \
94 arg->member ## _len = dlen; \
95 memcpy(arg->member, tags, min_t(size_t, dlen, len)); \
97 #include "linux/drbd_nl.h"
99 /* Generate the struct to tag_list functions */
100 #define NL_PACKET(name, number, fields) \
101 static unsigned short* \
102 name ## _to_tags(struct drbd_conf *mdev, \
103 struct name *arg, unsigned short *tags) __attribute__ ((unused)); \
104 static unsigned short* \
105 name ## _to_tags(struct drbd_conf *mdev, \
106 struct name *arg, unsigned short *tags) \
112 #define NL_INTEGER(pn, pr, member) \
113 put_unaligned(pn | pr | TT_INTEGER, tags++); \
114 put_unaligned(sizeof(int), tags++); \
115 put_unaligned(arg->member, (int *)tags); \
116 tags = (unsigned short *)((char *)tags+sizeof(int));
117 #define NL_INT64(pn, pr, member) \
118 put_unaligned(pn | pr | TT_INT64, tags++); \
119 put_unaligned(sizeof(u64), tags++); \
120 put_unaligned(arg->member, (u64 *)tags); \
121 tags = (unsigned short *)((char *)tags+sizeof(u64));
122 #define NL_BIT(pn, pr, member) \
123 put_unaligned(pn | pr | TT_BIT, tags++); \
124 put_unaligned(sizeof(char), tags++); \
125 *(char *)tags = arg->member; \
126 tags = (unsigned short *)((char *)tags+sizeof(char));
127 #define NL_STRING(pn, pr, member, len) \
128 put_unaligned(pn | pr | TT_STRING, tags++); \
129 put_unaligned(arg->member ## _len, tags++); \
130 memcpy(tags, arg->member, arg->member ## _len); \
131 tags = (unsigned short *)((char *)tags + arg->member ## _len);
132 #include "linux/drbd_nl.h"
134 void drbd_bcast_ev_helper(struct drbd_conf
*mdev
, char *helper_name
);
135 void drbd_nl_send_reply(struct cn_msg
*, int);
137 int drbd_khelper(struct drbd_conf
*mdev
, char *cmd
)
139 char *envp
[] = { "HOME=/",
141 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
142 NULL
, /* Will be set to address family */
143 NULL
, /* Will be set to address */
146 char mb
[12], af
[20], ad
[60], *afs
;
147 char *argv
[] = {usermode_helper
, cmd
, mb
, NULL
};
150 snprintf(mb
, 12, "minor-%d", mdev_to_minor(mdev
));
152 if (get_net_conf(mdev
)) {
153 switch (((struct sockaddr
*)mdev
->net_conf
->peer_addr
)->sa_family
) {
156 snprintf(ad
, 60, "DRBD_PEER_ADDRESS=%pI6",
157 &((struct sockaddr_in6
*)mdev
->net_conf
->peer_addr
)->sin6_addr
);
161 snprintf(ad
, 60, "DRBD_PEER_ADDRESS=%pI4",
162 &((struct sockaddr_in
*)mdev
->net_conf
->peer_addr
)->sin_addr
);
166 snprintf(ad
, 60, "DRBD_PEER_ADDRESS=%pI4",
167 &((struct sockaddr_in
*)mdev
->net_conf
->peer_addr
)->sin_addr
);
169 snprintf(af
, 20, "DRBD_PEER_AF=%s", afs
);
175 /* The helper may take some time.
176 * write out any unsynced meta data changes now */
179 dev_info(DEV
, "helper command: %s %s %s\n", usermode_helper
, cmd
, mb
);
181 drbd_bcast_ev_helper(mdev
, cmd
);
182 ret
= call_usermodehelper(usermode_helper
, argv
, envp
, 1);
184 dev_warn(DEV
, "helper command: %s %s %s exit code %u (0x%x)\n",
185 usermode_helper
, cmd
, mb
,
186 (ret
>> 8) & 0xff, ret
);
188 dev_info(DEV
, "helper command: %s %s %s exit code %u (0x%x)\n",
189 usermode_helper
, cmd
, mb
,
190 (ret
>> 8) & 0xff, ret
);
192 if (ret
< 0) /* Ignore any ERRNOs we got. */
198 enum drbd_disk_state
drbd_try_outdate_peer(struct drbd_conf
*mdev
)
202 enum drbd_disk_state nps
;
203 enum drbd_fencing_p fp
;
205 D_ASSERT(mdev
->state
.pdsk
== D_UNKNOWN
);
207 if (get_ldev_if_state(mdev
, D_CONSISTENT
)) {
208 fp
= mdev
->ldev
->dc
.fencing
;
211 dev_warn(DEV
, "Not fencing peer, I'm not even Consistent myself.\n");
212 nps
= mdev
->state
.pdsk
;
216 r
= drbd_khelper(mdev
, "fence-peer");
218 switch ((r
>>8) & 0xff) {
219 case 3: /* peer is inconsistent */
220 ex_to_string
= "peer is inconsistent or worse";
221 nps
= D_INCONSISTENT
;
223 case 4: /* peer got outdated, or was already outdated */
224 ex_to_string
= "peer was fenced";
227 case 5: /* peer was down */
228 if (mdev
->state
.disk
== D_UP_TO_DATE
) {
229 /* we will(have) create(d) a new UUID anyways... */
230 ex_to_string
= "peer is unreachable, assumed to be dead";
233 ex_to_string
= "peer unreachable, doing nothing since disk != UpToDate";
234 nps
= mdev
->state
.pdsk
;
237 case 6: /* Peer is primary, voluntarily outdate myself.
238 * This is useful when an unconnected R_SECONDARY is asked to
239 * become R_PRIMARY, but finds the other peer being active. */
240 ex_to_string
= "peer is active";
241 dev_warn(DEV
, "Peer is primary, outdating myself.\n");
243 _drbd_request_state(mdev
, NS(disk
, D_OUTDATED
), CS_WAIT_COMPLETE
);
246 if (fp
!= FP_STONITH
)
247 dev_err(DEV
, "fence-peer() = 7 && fencing != Stonith !!!\n");
248 ex_to_string
= "peer was stonithed";
252 /* The script is broken ... */
254 dev_err(DEV
, "fence-peer helper broken, returned %d\n", (r
>>8)&0xff);
258 dev_info(DEV
, "fence-peer helper returned %d (%s)\n",
259 (r
>>8) & 0xff, ex_to_string
);
262 if (mdev
->state
.susp_fen
&& nps
>= D_UNKNOWN
) {
263 /* The handler was not successful... unfreeze here, the
264 state engine can not unfreeze... */
265 _drbd_request_state(mdev
, NS(susp_fen
, 0), CS_VERBOSE
);
271 static int _try_outdate_peer_async(void *data
)
273 struct drbd_conf
*mdev
= (struct drbd_conf
*)data
;
274 enum drbd_disk_state nps
;
276 nps
= drbd_try_outdate_peer(mdev
);
277 drbd_request_state(mdev
, NS(pdsk
, nps
));
282 void drbd_try_outdate_peer_async(struct drbd_conf
*mdev
)
284 struct task_struct
*opa
;
286 opa
= kthread_run(_try_outdate_peer_async
, mdev
, "drbd%d_a_helper", mdev_to_minor(mdev
));
288 dev_err(DEV
, "out of mem, failed to invoke fence-peer helper\n");
291 int drbd_set_role(struct drbd_conf
*mdev
, enum drbd_role new_role
, int force
)
293 const int max_tries
= 4;
297 union drbd_state mask
, val
;
298 enum drbd_disk_state nps
;
300 if (new_role
== R_PRIMARY
)
301 request_ping(mdev
); /* Detect a dead peer ASAP */
303 mutex_lock(&mdev
->state_mutex
);
305 mask
.i
= 0; mask
.role
= R_MASK
;
306 val
.i
= 0; val
.role
= new_role
;
308 while (try++ < max_tries
) {
309 r
= _drbd_request_state(mdev
, mask
, val
, CS_WAIT_COMPLETE
);
311 /* in case we first succeeded to outdate,
312 * but now suddenly could establish a connection */
313 if (r
== SS_CW_FAILED_BY_PEER
&& mask
.pdsk
!= 0) {
319 if (r
== SS_NO_UP_TO_DATE_DISK
&& force
&&
320 (mdev
->state
.disk
< D_UP_TO_DATE
&&
321 mdev
->state
.disk
>= D_INCONSISTENT
)) {
323 val
.disk
= D_UP_TO_DATE
;
328 if (r
== SS_NO_UP_TO_DATE_DISK
&&
329 mdev
->state
.disk
== D_CONSISTENT
&& mask
.pdsk
== 0) {
330 D_ASSERT(mdev
->state
.pdsk
== D_UNKNOWN
);
331 nps
= drbd_try_outdate_peer(mdev
);
333 if (nps
== D_OUTDATED
|| nps
== D_INCONSISTENT
) {
334 val
.disk
= D_UP_TO_DATE
;
344 if (r
== SS_NOTHING_TO_DO
)
346 if (r
== SS_PRIMARY_NOP
&& mask
.pdsk
== 0) {
347 nps
= drbd_try_outdate_peer(mdev
);
349 if (force
&& nps
> D_OUTDATED
) {
350 dev_warn(DEV
, "Forced into split brain situation!\n");
359 if (r
== SS_TWO_PRIMARIES
) {
360 /* Maybe the peer is detected as dead very soon...
361 retry at most once more in this case. */
362 __set_current_state(TASK_INTERRUPTIBLE
);
363 schedule_timeout((mdev
->net_conf
->ping_timeo
+1)*HZ
/10);
368 if (r
< SS_SUCCESS
) {
369 r
= _drbd_request_state(mdev
, mask
, val
,
370 CS_VERBOSE
+ CS_WAIT_COMPLETE
);
381 dev_warn(DEV
, "Forced to consider local data as UpToDate!\n");
383 /* Wait until nothing is on the fly :) */
384 wait_event(mdev
->misc_wait
, atomic_read(&mdev
->ap_pending_cnt
) == 0);
386 if (new_role
== R_SECONDARY
) {
387 set_disk_ro(mdev
->vdisk
, TRUE
);
388 if (get_ldev(mdev
)) {
389 mdev
->ldev
->md
.uuid
[UI_CURRENT
] &= ~(u64
)1;
393 if (get_net_conf(mdev
)) {
394 mdev
->net_conf
->want_lose
= 0;
397 set_disk_ro(mdev
->vdisk
, FALSE
);
398 if (get_ldev(mdev
)) {
399 if (((mdev
->state
.conn
< C_CONNECTED
||
400 mdev
->state
.pdsk
<= D_FAILED
)
401 && mdev
->ldev
->md
.uuid
[UI_BITMAP
] == 0) || forced
)
402 drbd_uuid_new_current(mdev
);
404 mdev
->ldev
->md
.uuid
[UI_CURRENT
] |= (u64
)1;
409 if ((new_role
== R_SECONDARY
) && get_ldev(mdev
)) {
410 drbd_al_to_on_disk_bm(mdev
);
414 if (mdev
->state
.conn
>= C_WF_REPORT_PARAMS
) {
415 /* if this was forced, we should consider sync */
417 drbd_send_uuids(mdev
);
418 drbd_send_state(mdev
);
423 kobject_uevent(&disk_to_dev(mdev
->vdisk
)->kobj
, KOBJ_CHANGE
);
425 mutex_unlock(&mdev
->state_mutex
);
429 static struct drbd_conf
*ensure_mdev(int minor
, int create
)
431 struct drbd_conf
*mdev
;
433 if (minor
>= minor_count
)
436 mdev
= minor_to_mdev(minor
);
438 if (!mdev
&& create
) {
439 struct gendisk
*disk
= NULL
;
440 mdev
= drbd_new_device(minor
);
442 spin_lock_irq(&drbd_pp_lock
);
443 if (minor_table
[minor
] == NULL
) {
444 minor_table
[minor
] = mdev
;
447 } /* else: we lost the race */
448 spin_unlock_irq(&drbd_pp_lock
);
450 if (disk
) /* we won the race above */
451 /* in case we ever add a drbd_delete_device(),
452 * don't forget the del_gendisk! */
454 else /* we lost the race above */
455 drbd_free_mdev(mdev
);
457 mdev
= minor_to_mdev(minor
);
463 static int drbd_nl_primary(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
464 struct drbd_nl_cfg_reply
*reply
)
466 struct primary primary_args
;
468 memset(&primary_args
, 0, sizeof(struct primary
));
469 if (!primary_from_tags(mdev
, nlp
->tag_list
, &primary_args
)) {
470 reply
->ret_code
= ERR_MANDATORY_TAG
;
475 drbd_set_role(mdev
, R_PRIMARY
, primary_args
.primary_force
);
480 static int drbd_nl_secondary(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
481 struct drbd_nl_cfg_reply
*reply
)
483 reply
->ret_code
= drbd_set_role(mdev
, R_SECONDARY
, 0);
488 /* initializes the md.*_offset members, so we are able to find
489 * the on disk meta data */
490 static void drbd_md_set_sector_offsets(struct drbd_conf
*mdev
,
491 struct drbd_backing_dev
*bdev
)
493 sector_t md_size_sect
= 0;
494 switch (bdev
->dc
.meta_dev_idx
) {
496 /* v07 style fixed size indexed meta data */
497 bdev
->md
.md_size_sect
= MD_RESERVED_SECT
;
498 bdev
->md
.md_offset
= drbd_md_ss__(mdev
, bdev
);
499 bdev
->md
.al_offset
= MD_AL_OFFSET
;
500 bdev
->md
.bm_offset
= MD_BM_OFFSET
;
502 case DRBD_MD_INDEX_FLEX_EXT
:
503 /* just occupy the full device; unit: sectors */
504 bdev
->md
.md_size_sect
= drbd_get_capacity(bdev
->md_bdev
);
505 bdev
->md
.md_offset
= 0;
506 bdev
->md
.al_offset
= MD_AL_OFFSET
;
507 bdev
->md
.bm_offset
= MD_BM_OFFSET
;
509 case DRBD_MD_INDEX_INTERNAL
:
510 case DRBD_MD_INDEX_FLEX_INT
:
511 bdev
->md
.md_offset
= drbd_md_ss__(mdev
, bdev
);
512 /* al size is still fixed */
513 bdev
->md
.al_offset
= -MD_AL_MAX_SIZE
;
514 /* we need (slightly less than) ~ this much bitmap sectors: */
515 md_size_sect
= drbd_get_capacity(bdev
->backing_bdev
);
516 md_size_sect
= ALIGN(md_size_sect
, BM_SECT_PER_EXT
);
517 md_size_sect
= BM_SECT_TO_EXT(md_size_sect
);
518 md_size_sect
= ALIGN(md_size_sect
, 8);
520 /* plus the "drbd meta data super block",
521 * and the activity log; */
522 md_size_sect
+= MD_BM_OFFSET
;
524 bdev
->md
.md_size_sect
= md_size_sect
;
525 /* bitmap offset is adjusted by 'super' block size */
526 bdev
->md
.bm_offset
= -md_size_sect
+ MD_AL_OFFSET
;
531 char *ppsize(char *buf
, unsigned long long size
)
533 /* Needs 9 bytes at max. */
534 static char units
[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
536 while (size
>= 10000) {
538 size
= (size
>> 10) + !!(size
& (1<<9));
541 sprintf(buf
, "%lu %cB", (long)size
, units
[base
]);
546 /* there is still a theoretical deadlock when called from receiver
547 * on an D_INCONSISTENT R_PRIMARY:
548 * remote READ does inc_ap_bio, receiver would need to receive answer
549 * packet from remote to dec_ap_bio again.
550 * receiver receive_sizes(), comes here,
551 * waits for ap_bio_cnt == 0. -> deadlock.
552 * but this cannot happen, actually, because:
553 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
554 * (not connected, or bad/no disk on peer):
555 * see drbd_fail_request_early, ap_bio_cnt is zero.
556 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
557 * peer may not initiate a resize.
559 void drbd_suspend_io(struct drbd_conf
*mdev
)
561 set_bit(SUSPEND_IO
, &mdev
->flags
);
562 if (is_susp(mdev
->state
))
564 wait_event(mdev
->misc_wait
, !atomic_read(&mdev
->ap_bio_cnt
));
567 void drbd_resume_io(struct drbd_conf
*mdev
)
569 clear_bit(SUSPEND_IO
, &mdev
->flags
);
570 wake_up(&mdev
->misc_wait
);
574 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
575 * @mdev: DRBD device.
577 * Returns 0 on success, negative return values indicate errors.
578 * You should call drbd_md_sync() after calling this function.
580 enum determine_dev_size
drbd_determin_dev_size(struct drbd_conf
*mdev
, enum dds_flags flags
) __must_hold(local
)
582 sector_t prev_first_sect
, prev_size
; /* previous meta location */
587 int md_moved
, la_size_changed
;
588 enum determine_dev_size rv
= unchanged
;
591 * application request passes inc_ap_bio,
592 * but then cannot get an AL-reference.
593 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
596 * Suspend IO right here.
597 * still lock the act_log to not trigger ASSERTs there.
599 drbd_suspend_io(mdev
);
601 /* no wait necessary anymore, actually we could assert that */
602 wait_event(mdev
->al_wait
, lc_try_lock(mdev
->act_log
));
604 prev_first_sect
= drbd_md_first_sector(mdev
->ldev
);
605 prev_size
= mdev
->ldev
->md
.md_size_sect
;
606 la_size
= mdev
->ldev
->md
.la_size_sect
;
608 /* TODO: should only be some assert here, not (re)init... */
609 drbd_md_set_sector_offsets(mdev
, mdev
->ldev
);
611 size
= drbd_new_dev_size(mdev
, mdev
->ldev
, flags
& DDSF_FORCED
);
613 if (drbd_get_capacity(mdev
->this_bdev
) != size
||
614 drbd_bm_capacity(mdev
) != size
) {
616 err
= drbd_bm_resize(mdev
, size
, !(flags
& DDSF_NO_RESYNC
));
618 /* currently there is only one error: ENOMEM! */
619 size
= drbd_bm_capacity(mdev
)>>1;
621 dev_err(DEV
, "OUT OF MEMORY! "
622 "Could not allocate bitmap!\n");
624 dev_err(DEV
, "BM resizing failed. "
625 "Leaving size unchanged at size = %lu KB\n",
626 (unsigned long)size
);
630 /* racy, see comments above. */
631 drbd_set_my_capacity(mdev
, size
);
632 mdev
->ldev
->md
.la_size_sect
= size
;
633 dev_info(DEV
, "size = %s (%llu KB)\n", ppsize(ppb
, size
>>1),
634 (unsigned long long)size
>>1);
636 if (rv
== dev_size_error
)
639 la_size_changed
= (la_size
!= mdev
->ldev
->md
.la_size_sect
);
641 md_moved
= prev_first_sect
!= drbd_md_first_sector(mdev
->ldev
)
642 || prev_size
!= mdev
->ldev
->md
.md_size_sect
;
644 if (la_size_changed
|| md_moved
) {
645 drbd_al_shrink(mdev
); /* All extents inactive. */
646 dev_info(DEV
, "Writing the whole bitmap, %s\n",
647 la_size_changed
&& md_moved
? "size changed and md moved" :
648 la_size_changed
? "size changed" : "md moved");
649 rv
= drbd_bitmap_io(mdev
, &drbd_bm_write
, "size changed"); /* does drbd_resume_io() ! */
650 drbd_md_mark_dirty(mdev
);
658 lc_unlock(mdev
->act_log
);
659 wake_up(&mdev
->al_wait
);
660 drbd_resume_io(mdev
);
666 drbd_new_dev_size(struct drbd_conf
*mdev
, struct drbd_backing_dev
*bdev
, int assume_peer_has_space
)
668 sector_t p_size
= mdev
->p_size
; /* partner's disk size. */
669 sector_t la_size
= bdev
->md
.la_size_sect
; /* last agreed size. */
670 sector_t m_size
; /* my size */
671 sector_t u_size
= bdev
->dc
.disk_size
; /* size requested by user. */
674 m_size
= drbd_get_max_capacity(bdev
);
676 if (mdev
->state
.conn
< C_CONNECTED
&& assume_peer_has_space
) {
677 dev_warn(DEV
, "Resize while not connected was forced by the user!\n");
681 if (p_size
&& m_size
) {
682 size
= min_t(sector_t
, p_size
, m_size
);
686 if (m_size
&& m_size
< size
)
688 if (p_size
&& p_size
< size
)
699 dev_err(DEV
, "Both nodes diskless!\n");
703 dev_err(DEV
, "Requested disk size is too big (%lu > %lu)\n",
704 (unsigned long)u_size
>>1, (unsigned long)size
>>1);
713 * drbd_check_al_size() - Ensures that the AL is of the right size
714 * @mdev: DRBD device.
716 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
717 * failed, and 0 on success. You should call drbd_md_sync() after you called
720 static int drbd_check_al_size(struct drbd_conf
*mdev
)
722 struct lru_cache
*n
, *t
;
723 struct lc_element
*e
;
727 ERR_IF(mdev
->sync_conf
.al_extents
< 7)
728 mdev
->sync_conf
.al_extents
= 127;
731 mdev
->act_log
->nr_elements
== mdev
->sync_conf
.al_extents
)
736 n
= lc_create("act_log", drbd_al_ext_cache
,
737 mdev
->sync_conf
.al_extents
, sizeof(struct lc_element
), 0);
740 dev_err(DEV
, "Cannot allocate act_log lru!\n");
743 spin_lock_irq(&mdev
->al_lock
);
745 for (i
= 0; i
< t
->nr_elements
; i
++) {
746 e
= lc_element_by_index(t
, i
);
748 dev_err(DEV
, "refcnt(%d)==%d\n",
749 e
->lc_number
, e
->refcnt
);
755 spin_unlock_irq(&mdev
->al_lock
);
757 dev_err(DEV
, "Activity log still in use!\n");
764 drbd_md_mark_dirty(mdev
); /* we changed mdev->act_log->nr_elemens */
768 void drbd_setup_queue_param(struct drbd_conf
*mdev
, unsigned int max_seg_s
) __must_hold(local
)
770 struct request_queue
* const q
= mdev
->rq_queue
;
771 struct request_queue
* const b
= mdev
->ldev
->backing_bdev
->bd_disk
->queue
;
772 int max_segments
= mdev
->ldev
->dc
.max_bio_bvecs
;
774 max_seg_s
= min(queue_max_sectors(b
) * queue_logical_block_size(b
), max_seg_s
);
776 blk_queue_max_hw_sectors(q
, max_seg_s
>> 9);
777 blk_queue_max_segments(q
, max_segments
? max_segments
: BLK_MAX_SEGMENTS
);
778 blk_queue_max_segment_size(q
, max_seg_s
);
779 blk_queue_logical_block_size(q
, 512);
780 blk_queue_segment_boundary(q
, PAGE_SIZE
-1);
781 blk_stack_limits(&q
->limits
, &b
->limits
, 0);
783 dev_info(DEV
, "max_segment_size ( = BIO size ) = %u\n", queue_max_segment_size(q
));
785 if (q
->backing_dev_info
.ra_pages
!= b
->backing_dev_info
.ra_pages
) {
786 dev_info(DEV
, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
787 q
->backing_dev_info
.ra_pages
,
788 b
->backing_dev_info
.ra_pages
);
789 q
->backing_dev_info
.ra_pages
= b
->backing_dev_info
.ra_pages
;
793 /* serialize deconfig (worker exiting, doing cleanup)
794 * and reconfig (drbdsetup disk, drbdsetup net)
796 * Wait for a potentially exiting worker, then restart it,
797 * or start a new one. Flush any pending work, there may still be an
798 * after_state_change queued.
800 static void drbd_reconfig_start(struct drbd_conf
*mdev
)
802 wait_event(mdev
->state_wait
, !test_and_set_bit(CONFIG_PENDING
, &mdev
->flags
));
803 wait_event(mdev
->state_wait
, !test_bit(DEVICE_DYING
, &mdev
->flags
));
804 drbd_thread_start(&mdev
->worker
);
805 drbd_flush_workqueue(mdev
);
808 /* if still unconfigured, stops worker again.
809 * if configured now, clears CONFIG_PENDING.
810 * wakes potential waiters */
811 static void drbd_reconfig_done(struct drbd_conf
*mdev
)
813 spin_lock_irq(&mdev
->req_lock
);
814 if (mdev
->state
.disk
== D_DISKLESS
&&
815 mdev
->state
.conn
== C_STANDALONE
&&
816 mdev
->state
.role
== R_SECONDARY
) {
817 set_bit(DEVICE_DYING
, &mdev
->flags
);
818 drbd_thread_stop_nowait(&mdev
->worker
);
820 clear_bit(CONFIG_PENDING
, &mdev
->flags
);
821 spin_unlock_irq(&mdev
->req_lock
);
822 wake_up(&mdev
->state_wait
);
825 /* Make sure IO is suspended before calling this function(). */
826 static void drbd_suspend_al(struct drbd_conf
*mdev
)
830 if (lc_try_lock(mdev
->act_log
)) {
831 drbd_al_shrink(mdev
);
832 lc_unlock(mdev
->act_log
);
834 dev_warn(DEV
, "Failed to lock al in drbd_suspend_al()\n");
838 spin_lock_irq(&mdev
->req_lock
);
839 if (mdev
->state
.conn
< C_CONNECTED
)
840 s
= !test_and_set_bit(AL_SUSPENDED
, &mdev
->flags
);
842 spin_unlock_irq(&mdev
->req_lock
);
845 dev_info(DEV
, "Suspended AL updates\n");
848 /* does always return 0;
849 * interesting return code is in reply->ret_code */
850 static int drbd_nl_disk_conf(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
851 struct drbd_nl_cfg_reply
*reply
)
853 enum drbd_ret_codes retcode
;
854 enum determine_dev_size dd
;
855 sector_t max_possible_sectors
;
856 sector_t min_md_device_sectors
;
857 struct drbd_backing_dev
*nbc
= NULL
; /* new_backing_conf */
858 struct inode
*inode
, *inode2
;
859 struct lru_cache
*resync_lru
= NULL
;
860 union drbd_state ns
, os
;
861 unsigned int max_seg_s
;
863 int cp_discovered
= 0;
864 int logical_block_size
;
866 drbd_reconfig_start(mdev
);
868 /* if you want to reconfigure, please tear down first */
869 if (mdev
->state
.disk
> D_DISKLESS
) {
870 retcode
= ERR_DISK_CONFIGURED
;
874 /* allocation not in the IO path, cqueue thread context */
875 nbc
= kzalloc(sizeof(struct drbd_backing_dev
), GFP_KERNEL
);
881 nbc
->dc
.disk_size
= DRBD_DISK_SIZE_SECT_DEF
;
882 nbc
->dc
.on_io_error
= DRBD_ON_IO_ERROR_DEF
;
883 nbc
->dc
.fencing
= DRBD_FENCING_DEF
;
884 nbc
->dc
.max_bio_bvecs
= DRBD_MAX_BIO_BVECS_DEF
;
886 if (!disk_conf_from_tags(mdev
, nlp
->tag_list
, &nbc
->dc
)) {
887 retcode
= ERR_MANDATORY_TAG
;
891 if (nbc
->dc
.meta_dev_idx
< DRBD_MD_INDEX_FLEX_INT
) {
892 retcode
= ERR_MD_IDX_INVALID
;
896 if (get_net_conf(mdev
)) {
897 int prot
= mdev
->net_conf
->wire_protocol
;
899 if (nbc
->dc
.fencing
== FP_STONITH
&& prot
== DRBD_PROT_A
) {
900 retcode
= ERR_STONITH_AND_PROT_A
;
905 nbc
->lo_file
= filp_open(nbc
->dc
.backing_dev
, O_RDWR
, 0);
906 if (IS_ERR(nbc
->lo_file
)) {
907 dev_err(DEV
, "open(\"%s\") failed with %ld\n", nbc
->dc
.backing_dev
,
908 PTR_ERR(nbc
->lo_file
));
910 retcode
= ERR_OPEN_DISK
;
914 inode
= nbc
->lo_file
->f_dentry
->d_inode
;
916 if (!S_ISBLK(inode
->i_mode
)) {
917 retcode
= ERR_DISK_NOT_BDEV
;
921 nbc
->md_file
= filp_open(nbc
->dc
.meta_dev
, O_RDWR
, 0);
922 if (IS_ERR(nbc
->md_file
)) {
923 dev_err(DEV
, "open(\"%s\") failed with %ld\n", nbc
->dc
.meta_dev
,
924 PTR_ERR(nbc
->md_file
));
926 retcode
= ERR_OPEN_MD_DISK
;
930 inode2
= nbc
->md_file
->f_dentry
->d_inode
;
932 if (!S_ISBLK(inode2
->i_mode
)) {
933 retcode
= ERR_MD_NOT_BDEV
;
937 nbc
->backing_bdev
= inode
->i_bdev
;
938 if (bd_claim(nbc
->backing_bdev
, mdev
)) {
939 printk(KERN_ERR
"drbd: bd_claim(%p,%p); failed [%p;%p;%u]\n",
940 nbc
->backing_bdev
, mdev
,
941 nbc
->backing_bdev
->bd_holder
,
942 nbc
->backing_bdev
->bd_contains
->bd_holder
,
943 nbc
->backing_bdev
->bd_holders
);
944 retcode
= ERR_BDCLAIM_DISK
;
948 resync_lru
= lc_create("resync", drbd_bm_ext_cache
,
949 61, sizeof(struct bm_extent
),
950 offsetof(struct bm_extent
, lce
));
953 goto release_bdev_fail
;
956 /* meta_dev_idx >= 0: external fixed size,
957 * possibly multiple drbd sharing one meta device.
958 * TODO in that case, paranoia check that [md_bdev, meta_dev_idx] is
959 * not yet used by some other drbd minor!
960 * (if you use drbd.conf + drbdadm,
961 * that should check it for you already; but if you don't, or someone
962 * fooled it, we need to double check here) */
963 nbc
->md_bdev
= inode2
->i_bdev
;
964 if (bd_claim(nbc
->md_bdev
, (nbc
->dc
.meta_dev_idx
< 0) ? (void *)mdev
965 : (void *) drbd_m_holder
)) {
966 retcode
= ERR_BDCLAIM_MD_DISK
;
967 goto release_bdev_fail
;
970 if ((nbc
->backing_bdev
== nbc
->md_bdev
) !=
971 (nbc
->dc
.meta_dev_idx
== DRBD_MD_INDEX_INTERNAL
||
972 nbc
->dc
.meta_dev_idx
== DRBD_MD_INDEX_FLEX_INT
)) {
973 retcode
= ERR_MD_IDX_INVALID
;
974 goto release_bdev2_fail
;
977 /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
978 drbd_md_set_sector_offsets(mdev
, nbc
);
980 if (drbd_get_max_capacity(nbc
) < nbc
->dc
.disk_size
) {
981 dev_err(DEV
, "max capacity %llu smaller than disk size %llu\n",
982 (unsigned long long) drbd_get_max_capacity(nbc
),
983 (unsigned long long) nbc
->dc
.disk_size
);
984 retcode
= ERR_DISK_TO_SMALL
;
985 goto release_bdev2_fail
;
988 if (nbc
->dc
.meta_dev_idx
< 0) {
989 max_possible_sectors
= DRBD_MAX_SECTORS_FLEX
;
990 /* at least one MB, otherwise it does not make sense */
991 min_md_device_sectors
= (2<<10);
993 max_possible_sectors
= DRBD_MAX_SECTORS
;
994 min_md_device_sectors
= MD_RESERVED_SECT
* (nbc
->dc
.meta_dev_idx
+ 1);
997 if (drbd_get_capacity(nbc
->md_bdev
) < min_md_device_sectors
) {
998 retcode
= ERR_MD_DISK_TO_SMALL
;
999 dev_warn(DEV
, "refusing attach: md-device too small, "
1000 "at least %llu sectors needed for this meta-disk type\n",
1001 (unsigned long long) min_md_device_sectors
);
1002 goto release_bdev2_fail
;
1005 /* Make sure the new disk is big enough
1006 * (we may currently be R_PRIMARY with no local disk...) */
1007 if (drbd_get_max_capacity(nbc
) <
1008 drbd_get_capacity(mdev
->this_bdev
)) {
1009 retcode
= ERR_DISK_TO_SMALL
;
1010 goto release_bdev2_fail
;
1013 nbc
->known_size
= drbd_get_capacity(nbc
->backing_bdev
);
1015 if (nbc
->known_size
> max_possible_sectors
) {
1016 dev_warn(DEV
, "==> truncating very big lower level device "
1017 "to currently maximum possible %llu sectors <==\n",
1018 (unsigned long long) max_possible_sectors
);
1019 if (nbc
->dc
.meta_dev_idx
>= 0)
1020 dev_warn(DEV
, "==>> using internal or flexible "
1021 "meta data may help <<==\n");
1024 drbd_suspend_io(mdev
);
1025 /* also wait for the last barrier ack. */
1026 wait_event(mdev
->misc_wait
, !atomic_read(&mdev
->ap_pending_cnt
) || is_susp(mdev
->state
));
1027 /* and for any other previously queued work */
1028 drbd_flush_workqueue(mdev
);
1030 retcode
= _drbd_request_state(mdev
, NS(disk
, D_ATTACHING
), CS_VERBOSE
);
1031 drbd_resume_io(mdev
);
1032 if (retcode
< SS_SUCCESS
)
1033 goto release_bdev2_fail
;
1035 if (!get_ldev_if_state(mdev
, D_ATTACHING
))
1036 goto force_diskless
;
1038 drbd_md_set_sector_offsets(mdev
, nbc
);
1040 /* allocate a second IO page if logical_block_size != 512 */
1041 logical_block_size
= bdev_logical_block_size(nbc
->md_bdev
);
1042 if (logical_block_size
== 0)
1043 logical_block_size
= MD_SECTOR_SIZE
;
1045 if (logical_block_size
!= MD_SECTOR_SIZE
) {
1046 if (!mdev
->md_io_tmpp
) {
1047 struct page
*page
= alloc_page(GFP_NOIO
);
1049 goto force_diskless_dec
;
1051 dev_warn(DEV
, "Meta data's bdev logical_block_size = %d != %d\n",
1052 logical_block_size
, MD_SECTOR_SIZE
);
1053 dev_warn(DEV
, "Workaround engaged (has performance impact).\n");
1055 mdev
->md_io_tmpp
= page
;
1059 if (!mdev
->bitmap
) {
1060 if (drbd_bm_init(mdev
)) {
1061 retcode
= ERR_NOMEM
;
1062 goto force_diskless_dec
;
1066 retcode
= drbd_md_read(mdev
, nbc
);
1067 if (retcode
!= NO_ERROR
)
1068 goto force_diskless_dec
;
1070 if (mdev
->state
.conn
< C_CONNECTED
&&
1071 mdev
->state
.role
== R_PRIMARY
&&
1072 (mdev
->ed_uuid
& ~((u64
)1)) != (nbc
->md
.uuid
[UI_CURRENT
] & ~((u64
)1))) {
1073 dev_err(DEV
, "Can only attach to data with current UUID=%016llX\n",
1074 (unsigned long long)mdev
->ed_uuid
);
1075 retcode
= ERR_DATA_NOT_CURRENT
;
1076 goto force_diskless_dec
;
1079 /* Since we are diskless, fix the activity log first... */
1080 if (drbd_check_al_size(mdev
)) {
1081 retcode
= ERR_NOMEM
;
1082 goto force_diskless_dec
;
1085 /* Prevent shrinking of consistent devices ! */
1086 if (drbd_md_test_flag(nbc
, MDF_CONSISTENT
) &&
1087 drbd_new_dev_size(mdev
, nbc
, 0) < nbc
->md
.la_size_sect
) {
1088 dev_warn(DEV
, "refusing to truncate a consistent device\n");
1089 retcode
= ERR_DISK_TO_SMALL
;
1090 goto force_diskless_dec
;
1093 if (!drbd_al_read_log(mdev
, nbc
)) {
1094 retcode
= ERR_IO_MD_DISK
;
1095 goto force_diskless_dec
;
1098 /* Reset the "barriers don't work" bits here, then force meta data to
1099 * be written, to ensure we determine if barriers are supported. */
1100 if (nbc
->dc
.no_md_flush
)
1101 set_bit(MD_NO_BARRIER
, &mdev
->flags
);
1103 clear_bit(MD_NO_BARRIER
, &mdev
->flags
);
1105 /* Point of no return reached.
1106 * Devices and memory are no longer released by error cleanup below.
1107 * now mdev takes over responsibility, and the state engine should
1108 * clean it up somewhere. */
1109 D_ASSERT(mdev
->ldev
== NULL
);
1111 mdev
->resync
= resync_lru
;
1115 mdev
->write_ordering
= WO_bio_barrier
;
1116 drbd_bump_write_ordering(mdev
, WO_bio_barrier
);
1118 if (drbd_md_test_flag(mdev
->ldev
, MDF_CRASHED_PRIMARY
))
1119 set_bit(CRASHED_PRIMARY
, &mdev
->flags
);
1121 clear_bit(CRASHED_PRIMARY
, &mdev
->flags
);
1123 if (drbd_md_test_flag(mdev
->ldev
, MDF_PRIMARY_IND
) &&
1124 !(mdev
->state
.role
== R_PRIMARY
&& mdev
->state
.susp_nod
)) {
1125 set_bit(CRASHED_PRIMARY
, &mdev
->flags
);
1134 max_seg_s
= DRBD_MAX_SEGMENT_SIZE
;
1135 if (mdev
->state
.conn
== C_CONNECTED
) {
1136 /* We are Primary, Connected, and now attach a new local
1137 * backing store. We must not increase the user visible maximum
1138 * bio size on this device to something the peer may not be
1139 * able to handle. */
1140 if (mdev
->agreed_pro_version
< 94)
1141 max_seg_s
= queue_max_segment_size(mdev
->rq_queue
);
1142 else if (mdev
->agreed_pro_version
== 94)
1143 max_seg_s
= DRBD_MAX_SIZE_H80_PACKET
;
1144 /* else: drbd 8.3.9 and later, stay with default */
1147 drbd_setup_queue_param(mdev
, max_seg_s
);
1149 /* If I am currently not R_PRIMARY,
1150 * but meta data primary indicator is set,
1151 * I just now recover from a hard crash,
1152 * and have been R_PRIMARY before that crash.
1154 * Now, if I had no connection before that crash
1155 * (have been degraded R_PRIMARY), chances are that
1156 * I won't find my peer now either.
1158 * In that case, and _only_ in that case,
1159 * we use the degr-wfc-timeout instead of the default,
1160 * so we can automatically recover from a crash of a
1161 * degraded but active "cluster" after a certain timeout.
1163 clear_bit(USE_DEGR_WFC_T
, &mdev
->flags
);
1164 if (mdev
->state
.role
!= R_PRIMARY
&&
1165 drbd_md_test_flag(mdev
->ldev
, MDF_PRIMARY_IND
) &&
1166 !drbd_md_test_flag(mdev
->ldev
, MDF_CONNECTED_IND
))
1167 set_bit(USE_DEGR_WFC_T
, &mdev
->flags
);
1169 dd
= drbd_determin_dev_size(mdev
, 0);
1170 if (dd
== dev_size_error
) {
1171 retcode
= ERR_NOMEM_BITMAP
;
1172 goto force_diskless_dec
;
1173 } else if (dd
== grew
)
1174 set_bit(RESYNC_AFTER_NEG
, &mdev
->flags
);
1176 if (drbd_md_test_flag(mdev
->ldev
, MDF_FULL_SYNC
)) {
1177 dev_info(DEV
, "Assuming that all blocks are out of sync "
1178 "(aka FullSync)\n");
1179 if (drbd_bitmap_io(mdev
, &drbd_bmio_set_n_write
, "set_n_write from attaching")) {
1180 retcode
= ERR_IO_MD_DISK
;
1181 goto force_diskless_dec
;
1184 if (drbd_bitmap_io(mdev
, &drbd_bm_read
, "read from attaching") < 0) {
1185 retcode
= ERR_IO_MD_DISK
;
1186 goto force_diskless_dec
;
1190 if (cp_discovered
) {
1191 drbd_al_apply_to_bm(mdev
);
1192 drbd_al_to_on_disk_bm(mdev
);
1195 if (_drbd_bm_total_weight(mdev
) == drbd_bm_bits(mdev
))
1196 drbd_suspend_al(mdev
); /* IO is still suspended here... */
1198 spin_lock_irq(&mdev
->req_lock
);
1201 /* If MDF_CONSISTENT is not set go into inconsistent state,
1202 otherwise investigate MDF_WasUpToDate...
1203 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1204 otherwise into D_CONSISTENT state.
1206 if (drbd_md_test_flag(mdev
->ldev
, MDF_CONSISTENT
)) {
1207 if (drbd_md_test_flag(mdev
->ldev
, MDF_WAS_UP_TO_DATE
))
1208 ns
.disk
= D_CONSISTENT
;
1210 ns
.disk
= D_OUTDATED
;
1212 ns
.disk
= D_INCONSISTENT
;
1215 if (drbd_md_test_flag(mdev
->ldev
, MDF_PEER_OUT_DATED
))
1216 ns
.pdsk
= D_OUTDATED
;
1218 if ( ns
.disk
== D_CONSISTENT
&&
1219 (ns
.pdsk
== D_OUTDATED
|| mdev
->ldev
->dc
.fencing
== FP_DONT_CARE
))
1220 ns
.disk
= D_UP_TO_DATE
;
1222 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1223 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1224 this point, because drbd_request_state() modifies these
1227 /* In case we are C_CONNECTED postpone any decision on the new disk
1228 state after the negotiation phase. */
1229 if (mdev
->state
.conn
== C_CONNECTED
) {
1230 mdev
->new_state_tmp
.i
= ns
.i
;
1232 ns
.disk
= D_NEGOTIATING
;
1234 /* We expect to receive up-to-date UUIDs soon.
1235 To avoid a race in receive_state, free p_uuid while
1236 holding req_lock. I.e. atomic with the state change */
1237 kfree(mdev
->p_uuid
);
1238 mdev
->p_uuid
= NULL
;
1241 rv
= _drbd_set_state(mdev
, ns
, CS_VERBOSE
, NULL
);
1243 spin_unlock_irq(&mdev
->req_lock
);
1245 if (rv
< SS_SUCCESS
)
1246 goto force_diskless_dec
;
1248 if (mdev
->state
.role
== R_PRIMARY
)
1249 mdev
->ldev
->md
.uuid
[UI_CURRENT
] |= (u64
)1;
1251 mdev
->ldev
->md
.uuid
[UI_CURRENT
] &= ~(u64
)1;
1253 drbd_md_mark_dirty(mdev
);
1256 kobject_uevent(&disk_to_dev(mdev
->vdisk
)->kobj
, KOBJ_CHANGE
);
1258 reply
->ret_code
= retcode
;
1259 drbd_reconfig_done(mdev
);
1265 drbd_force_state(mdev
, NS(disk
, D_DISKLESS
));
1269 bd_release(nbc
->md_bdev
);
1272 bd_release(nbc
->backing_bdev
);
1281 lc_destroy(resync_lru
);
1283 reply
->ret_code
= retcode
;
1284 drbd_reconfig_done(mdev
);
1288 static int drbd_nl_detach(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1289 struct drbd_nl_cfg_reply
*reply
)
1291 reply
->ret_code
= drbd_request_state(mdev
, NS(disk
, D_DISKLESS
));
1295 static int drbd_nl_net_conf(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1296 struct drbd_nl_cfg_reply
*reply
)
1299 enum drbd_ret_codes retcode
;
1300 struct net_conf
*new_conf
= NULL
;
1301 struct crypto_hash
*tfm
= NULL
;
1302 struct crypto_hash
*integrity_w_tfm
= NULL
;
1303 struct crypto_hash
*integrity_r_tfm
= NULL
;
1304 struct hlist_head
*new_tl_hash
= NULL
;
1305 struct hlist_head
*new_ee_hash
= NULL
;
1306 struct drbd_conf
*odev
;
1307 char hmac_name
[CRYPTO_MAX_ALG_NAME
];
1308 void *int_dig_out
= NULL
;
1309 void *int_dig_in
= NULL
;
1310 void *int_dig_vv
= NULL
;
1311 struct sockaddr
*new_my_addr
, *new_peer_addr
, *taken_addr
;
1313 drbd_reconfig_start(mdev
);
1315 if (mdev
->state
.conn
> C_STANDALONE
) {
1316 retcode
= ERR_NET_CONFIGURED
;
1320 /* allocation not in the IO path, cqueue thread context */
1321 new_conf
= kzalloc(sizeof(struct net_conf
), GFP_KERNEL
);
1323 retcode
= ERR_NOMEM
;
1327 new_conf
->timeout
= DRBD_TIMEOUT_DEF
;
1328 new_conf
->try_connect_int
= DRBD_CONNECT_INT_DEF
;
1329 new_conf
->ping_int
= DRBD_PING_INT_DEF
;
1330 new_conf
->max_epoch_size
= DRBD_MAX_EPOCH_SIZE_DEF
;
1331 new_conf
->max_buffers
= DRBD_MAX_BUFFERS_DEF
;
1332 new_conf
->unplug_watermark
= DRBD_UNPLUG_WATERMARK_DEF
;
1333 new_conf
->sndbuf_size
= DRBD_SNDBUF_SIZE_DEF
;
1334 new_conf
->rcvbuf_size
= DRBD_RCVBUF_SIZE_DEF
;
1335 new_conf
->ko_count
= DRBD_KO_COUNT_DEF
;
1336 new_conf
->after_sb_0p
= DRBD_AFTER_SB_0P_DEF
;
1337 new_conf
->after_sb_1p
= DRBD_AFTER_SB_1P_DEF
;
1338 new_conf
->after_sb_2p
= DRBD_AFTER_SB_2P_DEF
;
1339 new_conf
->want_lose
= 0;
1340 new_conf
->two_primaries
= 0;
1341 new_conf
->wire_protocol
= DRBD_PROT_C
;
1342 new_conf
->ping_timeo
= DRBD_PING_TIMEO_DEF
;
1343 new_conf
->rr_conflict
= DRBD_RR_CONFLICT_DEF
;
1345 if (!net_conf_from_tags(mdev
, nlp
->tag_list
, new_conf
)) {
1346 retcode
= ERR_MANDATORY_TAG
;
1350 if (new_conf
->two_primaries
1351 && (new_conf
->wire_protocol
!= DRBD_PROT_C
)) {
1352 retcode
= ERR_NOT_PROTO_C
;
1356 if (get_ldev(mdev
)) {
1357 enum drbd_fencing_p fp
= mdev
->ldev
->dc
.fencing
;
1359 if (new_conf
->wire_protocol
== DRBD_PROT_A
&& fp
== FP_STONITH
) {
1360 retcode
= ERR_STONITH_AND_PROT_A
;
1365 if (mdev
->state
.role
== R_PRIMARY
&& new_conf
->want_lose
) {
1366 retcode
= ERR_DISCARD
;
1372 new_my_addr
= (struct sockaddr
*)&new_conf
->my_addr
;
1373 new_peer_addr
= (struct sockaddr
*)&new_conf
->peer_addr
;
1374 for (i
= 0; i
< minor_count
; i
++) {
1375 odev
= minor_to_mdev(i
);
1376 if (!odev
|| odev
== mdev
)
1378 if (get_net_conf(odev
)) {
1379 taken_addr
= (struct sockaddr
*)&odev
->net_conf
->my_addr
;
1380 if (new_conf
->my_addr_len
== odev
->net_conf
->my_addr_len
&&
1381 !memcmp(new_my_addr
, taken_addr
, new_conf
->my_addr_len
))
1382 retcode
= ERR_LOCAL_ADDR
;
1384 taken_addr
= (struct sockaddr
*)&odev
->net_conf
->peer_addr
;
1385 if (new_conf
->peer_addr_len
== odev
->net_conf
->peer_addr_len
&&
1386 !memcmp(new_peer_addr
, taken_addr
, new_conf
->peer_addr_len
))
1387 retcode
= ERR_PEER_ADDR
;
1390 if (retcode
!= NO_ERROR
)
1395 if (new_conf
->cram_hmac_alg
[0] != 0) {
1396 snprintf(hmac_name
, CRYPTO_MAX_ALG_NAME
, "hmac(%s)",
1397 new_conf
->cram_hmac_alg
);
1398 tfm
= crypto_alloc_hash(hmac_name
, 0, CRYPTO_ALG_ASYNC
);
1401 retcode
= ERR_AUTH_ALG
;
1405 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm
))) {
1406 retcode
= ERR_AUTH_ALG_ND
;
1411 if (new_conf
->integrity_alg
[0]) {
1412 integrity_w_tfm
= crypto_alloc_hash(new_conf
->integrity_alg
, 0, CRYPTO_ALG_ASYNC
);
1413 if (IS_ERR(integrity_w_tfm
)) {
1414 integrity_w_tfm
= NULL
;
1415 retcode
=ERR_INTEGRITY_ALG
;
1419 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm
))) {
1420 retcode
=ERR_INTEGRITY_ALG_ND
;
1424 integrity_r_tfm
= crypto_alloc_hash(new_conf
->integrity_alg
, 0, CRYPTO_ALG_ASYNC
);
1425 if (IS_ERR(integrity_r_tfm
)) {
1426 integrity_r_tfm
= NULL
;
1427 retcode
=ERR_INTEGRITY_ALG
;
1432 ns
= new_conf
->max_epoch_size
/8;
1433 if (mdev
->tl_hash_s
!= ns
) {
1434 new_tl_hash
= kzalloc(ns
*sizeof(void *), GFP_KERNEL
);
1436 retcode
= ERR_NOMEM
;
1441 ns
= new_conf
->max_buffers
/8;
1442 if (new_conf
->two_primaries
&& (mdev
->ee_hash_s
!= ns
)) {
1443 new_ee_hash
= kzalloc(ns
*sizeof(void *), GFP_KERNEL
);
1445 retcode
= ERR_NOMEM
;
1450 ((char *)new_conf
->shared_secret
)[SHARED_SECRET_MAX
-1] = 0;
1452 if (integrity_w_tfm
) {
1453 i
= crypto_hash_digestsize(integrity_w_tfm
);
1454 int_dig_out
= kmalloc(i
, GFP_KERNEL
);
1456 retcode
= ERR_NOMEM
;
1459 int_dig_in
= kmalloc(i
, GFP_KERNEL
);
1461 retcode
= ERR_NOMEM
;
1464 int_dig_vv
= kmalloc(i
, GFP_KERNEL
);
1466 retcode
= ERR_NOMEM
;
1471 if (!mdev
->bitmap
) {
1472 if(drbd_bm_init(mdev
)) {
1473 retcode
= ERR_NOMEM
;
1478 drbd_flush_workqueue(mdev
);
1479 spin_lock_irq(&mdev
->req_lock
);
1480 if (mdev
->net_conf
!= NULL
) {
1481 retcode
= ERR_NET_CONFIGURED
;
1482 spin_unlock_irq(&mdev
->req_lock
);
1485 mdev
->net_conf
= new_conf
;
1491 kfree(mdev
->tl_hash
);
1492 mdev
->tl_hash_s
= mdev
->net_conf
->max_epoch_size
/8;
1493 mdev
->tl_hash
= new_tl_hash
;
1497 kfree(mdev
->ee_hash
);
1498 mdev
->ee_hash_s
= mdev
->net_conf
->max_buffers
/8;
1499 mdev
->ee_hash
= new_ee_hash
;
1502 crypto_free_hash(mdev
->cram_hmac_tfm
);
1503 mdev
->cram_hmac_tfm
= tfm
;
1505 crypto_free_hash(mdev
->integrity_w_tfm
);
1506 mdev
->integrity_w_tfm
= integrity_w_tfm
;
1508 crypto_free_hash(mdev
->integrity_r_tfm
);
1509 mdev
->integrity_r_tfm
= integrity_r_tfm
;
1511 kfree(mdev
->int_dig_out
);
1512 kfree(mdev
->int_dig_in
);
1513 kfree(mdev
->int_dig_vv
);
1514 mdev
->int_dig_out
=int_dig_out
;
1515 mdev
->int_dig_in
=int_dig_in
;
1516 mdev
->int_dig_vv
=int_dig_vv
;
1517 retcode
= _drbd_set_state(_NS(mdev
, conn
, C_UNCONNECTED
), CS_VERBOSE
, NULL
);
1518 spin_unlock_irq(&mdev
->req_lock
);
1520 kobject_uevent(&disk_to_dev(mdev
->vdisk
)->kobj
, KOBJ_CHANGE
);
1521 reply
->ret_code
= retcode
;
1522 drbd_reconfig_done(mdev
);
1529 crypto_free_hash(tfm
);
1530 crypto_free_hash(integrity_w_tfm
);
1531 crypto_free_hash(integrity_r_tfm
);
1536 reply
->ret_code
= retcode
;
1537 drbd_reconfig_done(mdev
);
1541 static int drbd_nl_disconnect(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1542 struct drbd_nl_cfg_reply
*reply
)
1546 retcode
= _drbd_request_state(mdev
, NS(conn
, C_DISCONNECTING
), CS_ORDERED
);
1548 if (retcode
== SS_NOTHING_TO_DO
)
1550 else if (retcode
== SS_ALREADY_STANDALONE
)
1552 else if (retcode
== SS_PRIMARY_NOP
) {
1553 /* Our statche checking code wants to see the peer outdated. */
1554 retcode
= drbd_request_state(mdev
, NS2(conn
, C_DISCONNECTING
,
1556 } else if (retcode
== SS_CW_FAILED_BY_PEER
) {
1557 /* The peer probably wants to see us outdated. */
1558 retcode
= _drbd_request_state(mdev
, NS2(conn
, C_DISCONNECTING
,
1561 if (retcode
== SS_IS_DISKLESS
|| retcode
== SS_LOWER_THAN_OUTDATED
) {
1562 drbd_force_state(mdev
, NS(conn
, C_DISCONNECTING
));
1563 retcode
= SS_SUCCESS
;
1567 if (retcode
< SS_SUCCESS
)
1570 if (wait_event_interruptible(mdev
->state_wait
,
1571 mdev
->state
.conn
!= C_DISCONNECTING
)) {
1572 /* Do not test for mdev->state.conn == C_STANDALONE, since
1573 someone else might connect us in the mean time! */
1582 reply
->ret_code
= retcode
;
1586 void resync_after_online_grow(struct drbd_conf
*mdev
)
1588 int iass
; /* I am sync source */
1590 dev_info(DEV
, "Resync of new storage after online grow\n");
1591 if (mdev
->state
.role
!= mdev
->state
.peer
)
1592 iass
= (mdev
->state
.role
== R_PRIMARY
);
1594 iass
= test_bit(DISCARD_CONCURRENT
, &mdev
->flags
);
1597 drbd_start_resync(mdev
, C_SYNC_SOURCE
);
1599 _drbd_request_state(mdev
, NS(conn
, C_WF_SYNC_UUID
), CS_VERBOSE
+ CS_SERIALIZE
);
1602 static int drbd_nl_resize(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1603 struct drbd_nl_cfg_reply
*reply
)
1606 int retcode
= NO_ERROR
;
1607 enum determine_dev_size dd
;
1608 enum dds_flags ddsf
;
1610 memset(&rs
, 0, sizeof(struct resize
));
1611 if (!resize_from_tags(mdev
, nlp
->tag_list
, &rs
)) {
1612 retcode
= ERR_MANDATORY_TAG
;
1616 if (mdev
->state
.conn
> C_CONNECTED
) {
1617 retcode
= ERR_RESIZE_RESYNC
;
1621 if (mdev
->state
.role
== R_SECONDARY
&&
1622 mdev
->state
.peer
== R_SECONDARY
) {
1623 retcode
= ERR_NO_PRIMARY
;
1627 if (!get_ldev(mdev
)) {
1628 retcode
= ERR_NO_DISK
;
1632 if (rs
.no_resync
&& mdev
->agreed_pro_version
< 93) {
1633 retcode
= ERR_NEED_APV_93
;
1637 if (mdev
->ldev
->known_size
!= drbd_get_capacity(mdev
->ldev
->backing_bdev
))
1638 mdev
->ldev
->known_size
= drbd_get_capacity(mdev
->ldev
->backing_bdev
);
1640 mdev
->ldev
->dc
.disk_size
= (sector_t
)rs
.resize_size
;
1641 ddsf
= (rs
.resize_force
? DDSF_FORCED
: 0) | (rs
.no_resync
? DDSF_NO_RESYNC
: 0);
1642 dd
= drbd_determin_dev_size(mdev
, ddsf
);
1645 if (dd
== dev_size_error
) {
1646 retcode
= ERR_NOMEM_BITMAP
;
1650 if (mdev
->state
.conn
== C_CONNECTED
) {
1652 set_bit(RESIZE_PENDING
, &mdev
->flags
);
1654 drbd_send_uuids(mdev
);
1655 drbd_send_sizes(mdev
, 1, ddsf
);
1659 reply
->ret_code
= retcode
;
1663 static int drbd_nl_syncer_conf(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1664 struct drbd_nl_cfg_reply
*reply
)
1666 int retcode
= NO_ERROR
;
1668 int ovr
; /* online verify running */
1669 int rsr
; /* re-sync running */
1670 struct crypto_hash
*verify_tfm
= NULL
;
1671 struct crypto_hash
*csums_tfm
= NULL
;
1672 struct syncer_conf sc
;
1673 cpumask_var_t new_cpu_mask
;
1674 int *rs_plan_s
= NULL
;
1677 if (!zalloc_cpumask_var(&new_cpu_mask
, GFP_KERNEL
)) {
1678 retcode
= ERR_NOMEM
;
1682 if (nlp
->flags
& DRBD_NL_SET_DEFAULTS
) {
1683 memset(&sc
, 0, sizeof(struct syncer_conf
));
1684 sc
.rate
= DRBD_RATE_DEF
;
1685 sc
.after
= DRBD_AFTER_DEF
;
1686 sc
.al_extents
= DRBD_AL_EXTENTS_DEF
;
1687 sc
.on_no_data
= DRBD_ON_NO_DATA_DEF
;
1688 sc
.c_plan_ahead
= DRBD_C_PLAN_AHEAD_DEF
;
1689 sc
.c_delay_target
= DRBD_C_DELAY_TARGET_DEF
;
1690 sc
.c_fill_target
= DRBD_C_FILL_TARGET_DEF
;
1691 sc
.c_max_rate
= DRBD_C_MAX_RATE_DEF
;
1692 sc
.c_min_rate
= DRBD_C_MIN_RATE_DEF
;
1694 memcpy(&sc
, &mdev
->sync_conf
, sizeof(struct syncer_conf
));
1696 if (!syncer_conf_from_tags(mdev
, nlp
->tag_list
, &sc
)) {
1697 retcode
= ERR_MANDATORY_TAG
;
1701 /* re-sync running */
1702 rsr
= ( mdev
->state
.conn
== C_SYNC_SOURCE
||
1703 mdev
->state
.conn
== C_SYNC_TARGET
||
1704 mdev
->state
.conn
== C_PAUSED_SYNC_S
||
1705 mdev
->state
.conn
== C_PAUSED_SYNC_T
);
1707 if (rsr
&& strcmp(sc
.csums_alg
, mdev
->sync_conf
.csums_alg
)) {
1708 retcode
= ERR_CSUMS_RESYNC_RUNNING
;
1712 if (!rsr
&& sc
.csums_alg
[0]) {
1713 csums_tfm
= crypto_alloc_hash(sc
.csums_alg
, 0, CRYPTO_ALG_ASYNC
);
1714 if (IS_ERR(csums_tfm
)) {
1716 retcode
= ERR_CSUMS_ALG
;
1720 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm
))) {
1721 retcode
= ERR_CSUMS_ALG_ND
;
1726 /* online verify running */
1727 ovr
= (mdev
->state
.conn
== C_VERIFY_S
|| mdev
->state
.conn
== C_VERIFY_T
);
1730 if (strcmp(sc
.verify_alg
, mdev
->sync_conf
.verify_alg
)) {
1731 retcode
= ERR_VERIFY_RUNNING
;
1736 if (!ovr
&& sc
.verify_alg
[0]) {
1737 verify_tfm
= crypto_alloc_hash(sc
.verify_alg
, 0, CRYPTO_ALG_ASYNC
);
1738 if (IS_ERR(verify_tfm
)) {
1740 retcode
= ERR_VERIFY_ALG
;
1744 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm
))) {
1745 retcode
= ERR_VERIFY_ALG_ND
;
1750 /* silently ignore cpu mask on UP kernel */
1751 if (nr_cpu_ids
> 1 && sc
.cpu_mask
[0] != 0) {
1752 err
= __bitmap_parse(sc
.cpu_mask
, 32, 0,
1753 cpumask_bits(new_cpu_mask
), nr_cpu_ids
);
1755 dev_warn(DEV
, "__bitmap_parse() failed with %d\n", err
);
1756 retcode
= ERR_CPU_MASK_PARSE
;
1761 ERR_IF (sc
.rate
< 1) sc
.rate
= 1;
1762 ERR_IF (sc
.al_extents
< 7) sc
.al_extents
= 127; /* arbitrary minimum */
1763 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1764 if (sc
.al_extents
> AL_MAX
) {
1765 dev_err(DEV
, "sc.al_extents > %d\n", AL_MAX
);
1766 sc
.al_extents
= AL_MAX
;
1770 /* to avoid spurious errors when configuring minors before configuring
1771 * the minors they depend on: if necessary, first create the minor we
1774 ensure_mdev(sc
.after
, 1);
1776 /* most sanity checks done, try to assign the new sync-after
1777 * dependency. need to hold the global lock in there,
1778 * to avoid a race in the dependency loop check. */
1779 retcode
= drbd_alter_sa(mdev
, sc
.after
);
1780 if (retcode
!= NO_ERROR
)
1783 fifo_size
= (sc
.c_plan_ahead
* 10 * SLEEP_TIME
) / HZ
;
1784 if (fifo_size
!= mdev
->rs_plan_s
.size
&& fifo_size
> 0) {
1785 rs_plan_s
= kzalloc(sizeof(int) * fifo_size
, GFP_KERNEL
);
1787 dev_err(DEV
, "kmalloc of fifo_buffer failed");
1788 retcode
= ERR_NOMEM
;
1793 /* ok, assign the rest of it as well.
1794 * lock against receive_SyncParam() */
1795 spin_lock(&mdev
->peer_seq_lock
);
1796 mdev
->sync_conf
= sc
;
1799 crypto_free_hash(mdev
->csums_tfm
);
1800 mdev
->csums_tfm
= csums_tfm
;
1805 crypto_free_hash(mdev
->verify_tfm
);
1806 mdev
->verify_tfm
= verify_tfm
;
1810 if (fifo_size
!= mdev
->rs_plan_s
.size
) {
1811 kfree(mdev
->rs_plan_s
.values
);
1812 mdev
->rs_plan_s
.values
= rs_plan_s
;
1813 mdev
->rs_plan_s
.size
= fifo_size
;
1814 mdev
->rs_planed
= 0;
1818 spin_unlock(&mdev
->peer_seq_lock
);
1820 if (get_ldev(mdev
)) {
1821 wait_event(mdev
->al_wait
, lc_try_lock(mdev
->act_log
));
1822 drbd_al_shrink(mdev
);
1823 err
= drbd_check_al_size(mdev
);
1824 lc_unlock(mdev
->act_log
);
1825 wake_up(&mdev
->al_wait
);
1831 retcode
= ERR_NOMEM
;
1836 if (mdev
->state
.conn
>= C_CONNECTED
)
1837 drbd_send_sync_param(mdev
, &sc
);
1839 if (!cpumask_equal(mdev
->cpu_mask
, new_cpu_mask
)) {
1840 cpumask_copy(mdev
->cpu_mask
, new_cpu_mask
);
1841 drbd_calc_cpu_mask(mdev
);
1842 mdev
->receiver
.reset_cpu_mask
= 1;
1843 mdev
->asender
.reset_cpu_mask
= 1;
1844 mdev
->worker
.reset_cpu_mask
= 1;
1847 kobject_uevent(&disk_to_dev(mdev
->vdisk
)->kobj
, KOBJ_CHANGE
);
1850 free_cpumask_var(new_cpu_mask
);
1851 crypto_free_hash(csums_tfm
);
1852 crypto_free_hash(verify_tfm
);
1853 reply
->ret_code
= retcode
;
1857 static int drbd_nl_invalidate(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1858 struct drbd_nl_cfg_reply
*reply
)
1862 retcode
= _drbd_request_state(mdev
, NS(conn
, C_STARTING_SYNC_T
), CS_ORDERED
);
1864 if (retcode
< SS_SUCCESS
&& retcode
!= SS_NEED_CONNECTION
)
1865 retcode
= drbd_request_state(mdev
, NS(conn
, C_STARTING_SYNC_T
));
1867 while (retcode
== SS_NEED_CONNECTION
) {
1868 spin_lock_irq(&mdev
->req_lock
);
1869 if (mdev
->state
.conn
< C_CONNECTED
)
1870 retcode
= _drbd_set_state(_NS(mdev
, disk
, D_INCONSISTENT
), CS_VERBOSE
, NULL
);
1871 spin_unlock_irq(&mdev
->req_lock
);
1873 if (retcode
!= SS_NEED_CONNECTION
)
1876 retcode
= drbd_request_state(mdev
, NS(conn
, C_STARTING_SYNC_T
));
1879 reply
->ret_code
= retcode
;
1883 static int drbd_bmio_set_susp_al(struct drbd_conf
*mdev
)
1887 rv
= drbd_bmio_set_n_write(mdev
);
1888 drbd_suspend_al(mdev
);
1892 static int drbd_nl_invalidate_peer(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1893 struct drbd_nl_cfg_reply
*reply
)
1897 retcode
= _drbd_request_state(mdev
, NS(conn
, C_STARTING_SYNC_S
), CS_ORDERED
);
1899 if (retcode
< SS_SUCCESS
) {
1900 if (retcode
== SS_NEED_CONNECTION
&& mdev
->state
.role
== R_PRIMARY
) {
1901 /* The peer will get a resync upon connect anyways. Just make that
1902 into a full resync. */
1903 retcode
= drbd_request_state(mdev
, NS(pdsk
, D_INCONSISTENT
));
1904 if (retcode
>= SS_SUCCESS
) {
1905 /* open coded drbd_bitmap_io() */
1906 if (drbd_bitmap_io(mdev
, &drbd_bmio_set_susp_al
,
1907 "set_n_write from invalidate_peer"))
1908 retcode
= ERR_IO_MD_DISK
;
1911 retcode
= drbd_request_state(mdev
, NS(conn
, C_STARTING_SYNC_S
));
1914 reply
->ret_code
= retcode
;
1918 static int drbd_nl_pause_sync(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1919 struct drbd_nl_cfg_reply
*reply
)
1921 int retcode
= NO_ERROR
;
1923 if (drbd_request_state(mdev
, NS(user_isp
, 1)) == SS_NOTHING_TO_DO
)
1924 retcode
= ERR_PAUSE_IS_SET
;
1926 reply
->ret_code
= retcode
;
1930 static int drbd_nl_resume_sync(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1931 struct drbd_nl_cfg_reply
*reply
)
1933 int retcode
= NO_ERROR
;
1935 if (drbd_request_state(mdev
, NS(user_isp
, 0)) == SS_NOTHING_TO_DO
)
1936 retcode
= ERR_PAUSE_IS_CLEAR
;
1938 reply
->ret_code
= retcode
;
1942 static int drbd_nl_suspend_io(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1943 struct drbd_nl_cfg_reply
*reply
)
1945 reply
->ret_code
= drbd_request_state(mdev
, NS(susp
, 1));
1950 static int drbd_nl_resume_io(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1951 struct drbd_nl_cfg_reply
*reply
)
1953 if (test_bit(NEW_CUR_UUID
, &mdev
->flags
)) {
1954 drbd_uuid_new_current(mdev
);
1955 clear_bit(NEW_CUR_UUID
, &mdev
->flags
);
1958 drbd_suspend_io(mdev
);
1959 reply
->ret_code
= drbd_request_state(mdev
, NS3(susp
, 0, susp_nod
, 0, susp_fen
, 0));
1960 if (reply
->ret_code
== SS_SUCCESS
) {
1961 if (mdev
->state
.conn
< C_CONNECTED
)
1963 if (mdev
->state
.disk
== D_DISKLESS
|| mdev
->state
.disk
== D_FAILED
)
1964 tl_restart(mdev
, fail_frozen_disk_io
);
1966 drbd_resume_io(mdev
);
1971 static int drbd_nl_outdate(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1972 struct drbd_nl_cfg_reply
*reply
)
1974 reply
->ret_code
= drbd_request_state(mdev
, NS(disk
, D_OUTDATED
));
1978 static int drbd_nl_get_config(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1979 struct drbd_nl_cfg_reply
*reply
)
1983 tl
= reply
->tag_list
;
1985 if (get_ldev(mdev
)) {
1986 tl
= disk_conf_to_tags(mdev
, &mdev
->ldev
->dc
, tl
);
1990 if (get_net_conf(mdev
)) {
1991 tl
= net_conf_to_tags(mdev
, mdev
->net_conf
, tl
);
1994 tl
= syncer_conf_to_tags(mdev
, &mdev
->sync_conf
, tl
);
1996 put_unaligned(TT_END
, tl
++); /* Close the tag list */
1998 return (int)((char *)tl
- (char *)reply
->tag_list
);
2001 static int drbd_nl_get_state(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
2002 struct drbd_nl_cfg_reply
*reply
)
2004 unsigned short *tl
= reply
->tag_list
;
2005 union drbd_state s
= mdev
->state
;
2006 unsigned long rs_left
;
2009 tl
= get_state_to_tags(mdev
, (struct get_state
*)&s
, tl
);
2011 /* no local ref, no bitmap, no syncer progress. */
2012 if (s
.conn
>= C_SYNC_SOURCE
&& s
.conn
<= C_PAUSED_SYNC_T
) {
2013 if (get_ldev(mdev
)) {
2014 drbd_get_syncer_progress(mdev
, &rs_left
, &res
);
2015 tl
= tl_add_int(tl
, T_sync_progress
, &res
);
2019 put_unaligned(TT_END
, tl
++); /* Close the tag list */
2021 return (int)((char *)tl
- (char *)reply
->tag_list
);
2024 static int drbd_nl_get_uuids(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
2025 struct drbd_nl_cfg_reply
*reply
)
2029 tl
= reply
->tag_list
;
2031 if (get_ldev(mdev
)) {
2032 tl
= tl_add_blob(tl
, T_uuids
, mdev
->ldev
->md
.uuid
, UI_SIZE
*sizeof(u64
));
2033 tl
= tl_add_int(tl
, T_uuids_flags
, &mdev
->ldev
->md
.flags
);
2036 put_unaligned(TT_END
, tl
++); /* Close the tag list */
2038 return (int)((char *)tl
- (char *)reply
->tag_list
);
2042 * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
2043 * @mdev: DRBD device.
2044 * @nlp: Netlink/connector packet from drbdsetup
2045 * @reply: Reply packet for drbdsetup
2047 static int drbd_nl_get_timeout_flag(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
2048 struct drbd_nl_cfg_reply
*reply
)
2053 tl
= reply
->tag_list
;
2055 rv
= mdev
->state
.pdsk
== D_OUTDATED
? UT_PEER_OUTDATED
:
2056 test_bit(USE_DEGR_WFC_T
, &mdev
->flags
) ? UT_DEGRADED
: UT_DEFAULT
;
2058 tl
= tl_add_blob(tl
, T_use_degraded
, &rv
, sizeof(rv
));
2059 put_unaligned(TT_END
, tl
++); /* Close the tag list */
2061 return (int)((char *)tl
- (char *)reply
->tag_list
);
2064 static int drbd_nl_start_ov(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
2065 struct drbd_nl_cfg_reply
*reply
)
2067 /* default to resume from last known position, if possible */
2068 struct start_ov args
=
2069 { .start_sector
= mdev
->ov_start_sector
};
2071 if (!start_ov_from_tags(mdev
, nlp
->tag_list
, &args
)) {
2072 reply
->ret_code
= ERR_MANDATORY_TAG
;
2075 /* w_make_ov_request expects position to be aligned */
2076 mdev
->ov_start_sector
= args
.start_sector
& ~BM_SECT_PER_BIT
;
2077 reply
->ret_code
= drbd_request_state(mdev
,NS(conn
,C_VERIFY_S
));
2082 static int drbd_nl_new_c_uuid(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
2083 struct drbd_nl_cfg_reply
*reply
)
2085 int retcode
= NO_ERROR
;
2086 int skip_initial_sync
= 0;
2089 struct new_c_uuid args
;
2091 memset(&args
, 0, sizeof(struct new_c_uuid
));
2092 if (!new_c_uuid_from_tags(mdev
, nlp
->tag_list
, &args
)) {
2093 reply
->ret_code
= ERR_MANDATORY_TAG
;
2097 mutex_lock(&mdev
->state_mutex
); /* Protects us against serialized state changes. */
2099 if (!get_ldev(mdev
)) {
2100 retcode
= ERR_NO_DISK
;
2104 /* this is "skip initial sync", assume to be clean */
2105 if (mdev
->state
.conn
== C_CONNECTED
&& mdev
->agreed_pro_version
>= 90 &&
2106 mdev
->ldev
->md
.uuid
[UI_CURRENT
] == UUID_JUST_CREATED
&& args
.clear_bm
) {
2107 dev_info(DEV
, "Preparing to skip initial sync\n");
2108 skip_initial_sync
= 1;
2109 } else if (mdev
->state
.conn
!= C_STANDALONE
) {
2110 retcode
= ERR_CONNECTED
;
2114 drbd_uuid_set(mdev
, UI_BITMAP
, 0); /* Rotate UI_BITMAP to History 1, etc... */
2115 drbd_uuid_new_current(mdev
); /* New current, previous to UI_BITMAP */
2117 if (args
.clear_bm
) {
2118 err
= drbd_bitmap_io(mdev
, &drbd_bmio_clear_n_write
, "clear_n_write from new_c_uuid");
2120 dev_err(DEV
, "Writing bitmap failed with %d\n",err
);
2121 retcode
= ERR_IO_MD_DISK
;
2123 if (skip_initial_sync
) {
2124 drbd_send_uuids_skip_initial_sync(mdev
);
2125 _drbd_uuid_set(mdev
, UI_BITMAP
, 0);
2126 spin_lock_irq(&mdev
->req_lock
);
2127 _drbd_set_state(_NS2(mdev
, disk
, D_UP_TO_DATE
, pdsk
, D_UP_TO_DATE
),
2129 spin_unlock_irq(&mdev
->req_lock
);
2137 mutex_unlock(&mdev
->state_mutex
);
2139 reply
->ret_code
= retcode
;
2143 struct cn_handler_struct
{
2144 int (*function
)(struct drbd_conf
*,
2145 struct drbd_nl_cfg_req
*,
2146 struct drbd_nl_cfg_reply
*);
2147 int reply_body_size
;
2150 static struct cn_handler_struct cnd_table
[] = {
2151 [ P_primary
] = { &drbd_nl_primary
, 0 },
2152 [ P_secondary
] = { &drbd_nl_secondary
, 0 },
2153 [ P_disk_conf
] = { &drbd_nl_disk_conf
, 0 },
2154 [ P_detach
] = { &drbd_nl_detach
, 0 },
2155 [ P_net_conf
] = { &drbd_nl_net_conf
, 0 },
2156 [ P_disconnect
] = { &drbd_nl_disconnect
, 0 },
2157 [ P_resize
] = { &drbd_nl_resize
, 0 },
2158 [ P_syncer_conf
] = { &drbd_nl_syncer_conf
, 0 },
2159 [ P_invalidate
] = { &drbd_nl_invalidate
, 0 },
2160 [ P_invalidate_peer
] = { &drbd_nl_invalidate_peer
, 0 },
2161 [ P_pause_sync
] = { &drbd_nl_pause_sync
, 0 },
2162 [ P_resume_sync
] = { &drbd_nl_resume_sync
, 0 },
2163 [ P_suspend_io
] = { &drbd_nl_suspend_io
, 0 },
2164 [ P_resume_io
] = { &drbd_nl_resume_io
, 0 },
2165 [ P_outdate
] = { &drbd_nl_outdate
, 0 },
2166 [ P_get_config
] = { &drbd_nl_get_config
,
2167 sizeof(struct syncer_conf_tag_len_struct
) +
2168 sizeof(struct disk_conf_tag_len_struct
) +
2169 sizeof(struct net_conf_tag_len_struct
) },
2170 [ P_get_state
] = { &drbd_nl_get_state
,
2171 sizeof(struct get_state_tag_len_struct
) +
2172 sizeof(struct sync_progress_tag_len_struct
) },
2173 [ P_get_uuids
] = { &drbd_nl_get_uuids
,
2174 sizeof(struct get_uuids_tag_len_struct
) },
2175 [ P_get_timeout_flag
] = { &drbd_nl_get_timeout_flag
,
2176 sizeof(struct get_timeout_flag_tag_len_struct
)},
2177 [ P_start_ov
] = { &drbd_nl_start_ov
, 0 },
2178 [ P_new_c_uuid
] = { &drbd_nl_new_c_uuid
, 0 },
2181 static void drbd_connector_callback(struct cn_msg
*req
, struct netlink_skb_parms
*nsp
)
2183 struct drbd_nl_cfg_req
*nlp
= (struct drbd_nl_cfg_req
*)req
->data
;
2184 struct cn_handler_struct
*cm
;
2185 struct cn_msg
*cn_reply
;
2186 struct drbd_nl_cfg_reply
*reply
;
2187 struct drbd_conf
*mdev
;
2189 int reply_size
= sizeof(struct cn_msg
)
2190 + sizeof(struct drbd_nl_cfg_reply
)
2191 + sizeof(short int);
2193 if (!try_module_get(THIS_MODULE
)) {
2194 printk(KERN_ERR
"drbd: try_module_get() failed!\n");
2198 if (!cap_raised(nsp
->eff_cap
, CAP_SYS_ADMIN
)) {
2203 mdev
= ensure_mdev(nlp
->drbd_minor
,
2204 (nlp
->flags
& DRBD_NL_CREATE_DEVICE
));
2206 retcode
= ERR_MINOR_INVALID
;
2210 if (nlp
->packet_type
>= P_nl_after_last_packet
) {
2211 retcode
= ERR_PACKET_NR
;
2215 cm
= cnd_table
+ nlp
->packet_type
;
2217 /* This may happen if packet number is 0: */
2218 if (cm
->function
== NULL
) {
2219 retcode
= ERR_PACKET_NR
;
2223 reply_size
+= cm
->reply_body_size
;
2225 /* allocation not in the IO path, cqueue thread context */
2226 cn_reply
= kmalloc(reply_size
, GFP_KERNEL
);
2228 retcode
= ERR_NOMEM
;
2231 reply
= (struct drbd_nl_cfg_reply
*) cn_reply
->data
;
2233 reply
->packet_type
=
2234 cm
->reply_body_size
? nlp
->packet_type
: P_nl_after_last_packet
;
2235 reply
->minor
= nlp
->drbd_minor
;
2236 reply
->ret_code
= NO_ERROR
; /* Might by modified by cm->function. */
2237 /* reply->tag_list; might be modified by cm->function. */
2239 rr
= cm
->function(mdev
, nlp
, reply
);
2241 cn_reply
->id
= req
->id
;
2242 cn_reply
->seq
= req
->seq
;
2243 cn_reply
->ack
= req
->ack
+ 1;
2244 cn_reply
->len
= sizeof(struct drbd_nl_cfg_reply
) + rr
;
2245 cn_reply
->flags
= 0;
2247 rr
= cn_netlink_send(cn_reply
, CN_IDX_DRBD
, GFP_KERNEL
);
2248 if (rr
&& rr
!= -ESRCH
)
2249 printk(KERN_INFO
"drbd: cn_netlink_send()=%d\n", rr
);
2252 module_put(THIS_MODULE
);
2255 drbd_nl_send_reply(req
, retcode
);
2256 module_put(THIS_MODULE
);
2259 static atomic_t drbd_nl_seq
= ATOMIC_INIT(2); /* two. */
2261 static unsigned short *
2262 __tl_add_blob(unsigned short *tl
, enum drbd_tags tag
, const void *data
,
2263 unsigned short len
, int nul_terminated
)
2265 unsigned short l
= tag_descriptions
[tag_number(tag
)].max_len
;
2266 len
= (len
< l
) ? len
: l
;
2267 put_unaligned(tag
, tl
++);
2268 put_unaligned(len
, tl
++);
2269 memcpy(tl
, data
, len
);
2270 tl
= (unsigned short*)((char*)tl
+ len
);
2272 *((char*)tl
- 1) = 0;
2276 static unsigned short *
2277 tl_add_blob(unsigned short *tl
, enum drbd_tags tag
, const void *data
, int len
)
2279 return __tl_add_blob(tl
, tag
, data
, len
, 0);
2282 static unsigned short *
2283 tl_add_str(unsigned short *tl
, enum drbd_tags tag
, const char *str
)
2285 return __tl_add_blob(tl
, tag
, str
, strlen(str
)+1, 0);
2288 static unsigned short *
2289 tl_add_int(unsigned short *tl
, enum drbd_tags tag
, const void *val
)
2291 put_unaligned(tag
, tl
++);
2292 switch(tag_type(tag
)) {
2294 put_unaligned(sizeof(int), tl
++);
2295 put_unaligned(*(int *)val
, (int *)tl
);
2296 tl
= (unsigned short*)((char*)tl
+sizeof(int));
2299 put_unaligned(sizeof(u64
), tl
++);
2300 put_unaligned(*(u64
*)val
, (u64
*)tl
);
2301 tl
= (unsigned short*)((char*)tl
+sizeof(u64
));
2304 /* someone did something stupid. */
2310 void drbd_bcast_state(struct drbd_conf
*mdev
, union drbd_state state
)
2312 char buffer
[sizeof(struct cn_msg
)+
2313 sizeof(struct drbd_nl_cfg_reply
)+
2314 sizeof(struct get_state_tag_len_struct
)+
2316 struct cn_msg
*cn_reply
= (struct cn_msg
*) buffer
;
2317 struct drbd_nl_cfg_reply
*reply
=
2318 (struct drbd_nl_cfg_reply
*)cn_reply
->data
;
2319 unsigned short *tl
= reply
->tag_list
;
2321 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2323 tl
= get_state_to_tags(mdev
, (struct get_state
*)&state
, tl
);
2325 put_unaligned(TT_END
, tl
++); /* Close the tag list */
2327 cn_reply
->id
.idx
= CN_IDX_DRBD
;
2328 cn_reply
->id
.val
= CN_VAL_DRBD
;
2330 cn_reply
->seq
= atomic_add_return(1, &drbd_nl_seq
);
2331 cn_reply
->ack
= 0; /* not used here. */
2332 cn_reply
->len
= sizeof(struct drbd_nl_cfg_reply
) +
2333 (int)((char *)tl
- (char *)reply
->tag_list
);
2334 cn_reply
->flags
= 0;
2336 reply
->packet_type
= P_get_state
;
2337 reply
->minor
= mdev_to_minor(mdev
);
2338 reply
->ret_code
= NO_ERROR
;
2340 cn_netlink_send(cn_reply
, CN_IDX_DRBD
, GFP_NOIO
);
2343 void drbd_bcast_ev_helper(struct drbd_conf
*mdev
, char *helper_name
)
2345 char buffer
[sizeof(struct cn_msg
)+
2346 sizeof(struct drbd_nl_cfg_reply
)+
2347 sizeof(struct call_helper_tag_len_struct
)+
2349 struct cn_msg
*cn_reply
= (struct cn_msg
*) buffer
;
2350 struct drbd_nl_cfg_reply
*reply
=
2351 (struct drbd_nl_cfg_reply
*)cn_reply
->data
;
2352 unsigned short *tl
= reply
->tag_list
;
2354 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2356 tl
= tl_add_str(tl
, T_helper
, helper_name
);
2357 put_unaligned(TT_END
, tl
++); /* Close the tag list */
2359 cn_reply
->id
.idx
= CN_IDX_DRBD
;
2360 cn_reply
->id
.val
= CN_VAL_DRBD
;
2362 cn_reply
->seq
= atomic_add_return(1, &drbd_nl_seq
);
2363 cn_reply
->ack
= 0; /* not used here. */
2364 cn_reply
->len
= sizeof(struct drbd_nl_cfg_reply
) +
2365 (int)((char *)tl
- (char *)reply
->tag_list
);
2366 cn_reply
->flags
= 0;
2368 reply
->packet_type
= P_call_helper
;
2369 reply
->minor
= mdev_to_minor(mdev
);
2370 reply
->ret_code
= NO_ERROR
;
2372 cn_netlink_send(cn_reply
, CN_IDX_DRBD
, GFP_NOIO
);
2375 void drbd_bcast_ee(struct drbd_conf
*mdev
,
2376 const char *reason
, const int dgs
,
2377 const char* seen_hash
, const char* calc_hash
,
2378 const struct drbd_epoch_entry
* e
)
2380 struct cn_msg
*cn_reply
;
2381 struct drbd_nl_cfg_reply
*reply
;
2388 if (!reason
|| !reason
[0])
2391 /* apparently we have to memcpy twice, first to prepare the data for the
2392 * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2394 /* receiver thread context, which is not in the writeout path (of this node),
2395 * but may be in the writeout path of the _other_ node.
2396 * GFP_NOIO to avoid potential "distributed deadlock". */
2398 sizeof(struct cn_msg
)+
2399 sizeof(struct drbd_nl_cfg_reply
)+
2400 sizeof(struct dump_ee_tag_len_struct
)+
2405 dev_err(DEV
, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2406 (unsigned long long)e
->sector
, e
->size
);
2410 reply
= (struct drbd_nl_cfg_reply
*)cn_reply
->data
;
2411 tl
= reply
->tag_list
;
2413 tl
= tl_add_str(tl
, T_dump_ee_reason
, reason
);
2414 tl
= tl_add_blob(tl
, T_seen_digest
, seen_hash
, dgs
);
2415 tl
= tl_add_blob(tl
, T_calc_digest
, calc_hash
, dgs
);
2416 tl
= tl_add_int(tl
, T_ee_sector
, &e
->sector
);
2417 tl
= tl_add_int(tl
, T_ee_block_id
, &e
->block_id
);
2419 put_unaligned(T_ee_data
, tl
++);
2420 put_unaligned(e
->size
, tl
++);
2424 page_chain_for_each(page
) {
2425 void *d
= kmap_atomic(page
, KM_USER0
);
2426 unsigned l
= min_t(unsigned, len
, PAGE_SIZE
);
2428 kunmap_atomic(d
, KM_USER0
);
2429 tl
= (unsigned short*)((char*)tl
+ l
);
2432 put_unaligned(TT_END
, tl
++); /* Close the tag list */
2434 cn_reply
->id
.idx
= CN_IDX_DRBD
;
2435 cn_reply
->id
.val
= CN_VAL_DRBD
;
2437 cn_reply
->seq
= atomic_add_return(1,&drbd_nl_seq
);
2438 cn_reply
->ack
= 0; // not used here.
2439 cn_reply
->len
= sizeof(struct drbd_nl_cfg_reply
) +
2440 (int)((char*)tl
- (char*)reply
->tag_list
);
2441 cn_reply
->flags
= 0;
2443 reply
->packet_type
= P_dump_ee
;
2444 reply
->minor
= mdev_to_minor(mdev
);
2445 reply
->ret_code
= NO_ERROR
;
2447 cn_netlink_send(cn_reply
, CN_IDX_DRBD
, GFP_NOIO
);
2451 void drbd_bcast_sync_progress(struct drbd_conf
*mdev
)
2453 char buffer
[sizeof(struct cn_msg
)+
2454 sizeof(struct drbd_nl_cfg_reply
)+
2455 sizeof(struct sync_progress_tag_len_struct
)+
2457 struct cn_msg
*cn_reply
= (struct cn_msg
*) buffer
;
2458 struct drbd_nl_cfg_reply
*reply
=
2459 (struct drbd_nl_cfg_reply
*)cn_reply
->data
;
2460 unsigned short *tl
= reply
->tag_list
;
2461 unsigned long rs_left
;
2464 /* no local ref, no bitmap, no syncer progress, no broadcast. */
2465 if (!get_ldev(mdev
))
2467 drbd_get_syncer_progress(mdev
, &rs_left
, &res
);
2470 tl
= tl_add_int(tl
, T_sync_progress
, &res
);
2471 put_unaligned(TT_END
, tl
++); /* Close the tag list */
2473 cn_reply
->id
.idx
= CN_IDX_DRBD
;
2474 cn_reply
->id
.val
= CN_VAL_DRBD
;
2476 cn_reply
->seq
= atomic_add_return(1, &drbd_nl_seq
);
2477 cn_reply
->ack
= 0; /* not used here. */
2478 cn_reply
->len
= sizeof(struct drbd_nl_cfg_reply
) +
2479 (int)((char *)tl
- (char *)reply
->tag_list
);
2480 cn_reply
->flags
= 0;
2482 reply
->packet_type
= P_sync_progress
;
2483 reply
->minor
= mdev_to_minor(mdev
);
2484 reply
->ret_code
= NO_ERROR
;
2486 cn_netlink_send(cn_reply
, CN_IDX_DRBD
, GFP_NOIO
);
2489 int __init
drbd_nl_init(void)
2491 static struct cb_id cn_id_drbd
;
2494 cn_id_drbd
.val
= CN_VAL_DRBD
;
2496 cn_id_drbd
.idx
= cn_idx
;
2497 err
= cn_add_callback(&cn_id_drbd
, "cn_drbd", &drbd_connector_callback
);
2500 cn_idx
= (cn_idx
+ CN_IDX_STEP
);
2504 printk(KERN_ERR
"drbd: cn_drbd failed to register\n");
2511 void drbd_nl_cleanup(void)
2513 static struct cb_id cn_id_drbd
;
2515 cn_id_drbd
.idx
= cn_idx
;
2516 cn_id_drbd
.val
= CN_VAL_DRBD
;
2518 cn_del_callback(&cn_id_drbd
);
2521 void drbd_nl_send_reply(struct cn_msg
*req
, int ret_code
)
2523 char buffer
[sizeof(struct cn_msg
)+sizeof(struct drbd_nl_cfg_reply
)];
2524 struct cn_msg
*cn_reply
= (struct cn_msg
*) buffer
;
2525 struct drbd_nl_cfg_reply
*reply
=
2526 (struct drbd_nl_cfg_reply
*)cn_reply
->data
;
2529 cn_reply
->id
= req
->id
;
2531 cn_reply
->seq
= req
->seq
;
2532 cn_reply
->ack
= req
->ack
+ 1;
2533 cn_reply
->len
= sizeof(struct drbd_nl_cfg_reply
);
2534 cn_reply
->flags
= 0;
2536 reply
->minor
= ((struct drbd_nl_cfg_req
*)req
->data
)->drbd_minor
;
2537 reply
->ret_code
= ret_code
;
2539 rr
= cn_netlink_send(cn_reply
, CN_IDX_DRBD
, GFP_NOIO
);
2540 if (rr
&& rr
!= -ESRCH
)
2541 printk(KERN_INFO
"drbd: cn_netlink_send()=%d\n", rr
);