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>
36 #include "drbd_wrappers.h"
37 #include <asm/unaligned.h>
38 #include <linux/drbd_tag_magic.h>
39 #include <linux/drbd_limits.h>
41 static unsigned short *tl_add_blob(unsigned short *, enum drbd_tags
, const void *, int);
42 static unsigned short *tl_add_str(unsigned short *, enum drbd_tags
, const char *);
43 static unsigned short *tl_add_int(unsigned short *, enum drbd_tags
, const void *);
45 /* see get_sb_bdev and bd_claim */
46 static char *drbd_m_holder
= "Hands off! this is DRBD's meta data device.";
48 /* Generate the tag_list to struct functions */
49 #define NL_PACKET(name, number, fields) \
50 static int name ## _from_tags(struct drbd_conf *mdev, \
51 unsigned short *tags, struct name *arg) __attribute__ ((unused)); \
52 static int name ## _from_tags(struct drbd_conf *mdev, \
53 unsigned short *tags, struct name *arg) \
58 while ((tag = get_unaligned(tags++)) != TT_END) { \
59 dlen = get_unaligned(tags++); \
60 switch (tag_number(tag)) { \
63 if (tag & T_MANDATORY) { \
64 dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \
68 tags = (unsigned short *)((char *)tags + dlen); \
72 #define NL_INTEGER(pn, pr, member) \
73 case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \
74 arg->member = get_unaligned((int *)(tags)); \
76 #define NL_INT64(pn, pr, member) \
77 case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \
78 arg->member = get_unaligned((u64 *)(tags)); \
80 #define NL_BIT(pn, pr, member) \
81 case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \
82 arg->member = *(char *)(tags) ? 1 : 0; \
84 #define NL_STRING(pn, pr, member, len) \
85 case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \
87 dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \
88 #member, dlen, (unsigned int)len); \
91 arg->member ## _len = dlen; \
92 memcpy(arg->member, tags, min_t(size_t, dlen, len)); \
94 #include "linux/drbd_nl.h"
96 /* Generate the struct to tag_list functions */
97 #define NL_PACKET(name, number, fields) \
98 static unsigned short* \
99 name ## _to_tags(struct drbd_conf *mdev, \
100 struct name *arg, unsigned short *tags) __attribute__ ((unused)); \
101 static unsigned short* \
102 name ## _to_tags(struct drbd_conf *mdev, \
103 struct name *arg, unsigned short *tags) \
109 #define NL_INTEGER(pn, pr, member) \
110 put_unaligned(pn | pr | TT_INTEGER, tags++); \
111 put_unaligned(sizeof(int), tags++); \
112 put_unaligned(arg->member, (int *)tags); \
113 tags = (unsigned short *)((char *)tags+sizeof(int));
114 #define NL_INT64(pn, pr, member) \
115 put_unaligned(pn | pr | TT_INT64, tags++); \
116 put_unaligned(sizeof(u64), tags++); \
117 put_unaligned(arg->member, (u64 *)tags); \
118 tags = (unsigned short *)((char *)tags+sizeof(u64));
119 #define NL_BIT(pn, pr, member) \
120 put_unaligned(pn | pr | TT_BIT, tags++); \
121 put_unaligned(sizeof(char), tags++); \
122 *(char *)tags = arg->member; \
123 tags = (unsigned short *)((char *)tags+sizeof(char));
124 #define NL_STRING(pn, pr, member, len) \
125 put_unaligned(pn | pr | TT_STRING, tags++); \
126 put_unaligned(arg->member ## _len, tags++); \
127 memcpy(tags, arg->member, arg->member ## _len); \
128 tags = (unsigned short *)((char *)tags + arg->member ## _len);
129 #include "linux/drbd_nl.h"
131 void drbd_bcast_ev_helper(struct drbd_conf
*mdev
, char *helper_name
);
132 void drbd_nl_send_reply(struct cn_msg
*, int);
134 int drbd_khelper(struct drbd_conf
*mdev
, char *cmd
)
136 char *envp
[] = { "HOME=/",
138 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
139 NULL
, /* Will be set to address family */
140 NULL
, /* Will be set to address */
143 char mb
[12], af
[20], ad
[60], *afs
;
144 char *argv
[] = {usermode_helper
, cmd
, mb
, NULL
};
147 snprintf(mb
, 12, "minor-%d", mdev_to_minor(mdev
));
149 if (get_net_conf(mdev
)) {
150 switch (((struct sockaddr
*)mdev
->net_conf
->peer_addr
)->sa_family
) {
153 snprintf(ad
, 60, "DRBD_PEER_ADDRESS=%pI6",
154 &((struct sockaddr_in6
*)mdev
->net_conf
->peer_addr
)->sin6_addr
);
158 snprintf(ad
, 60, "DRBD_PEER_ADDRESS=%pI4",
159 &((struct sockaddr_in
*)mdev
->net_conf
->peer_addr
)->sin_addr
);
163 snprintf(ad
, 60, "DRBD_PEER_ADDRESS=%pI4",
164 &((struct sockaddr_in
*)mdev
->net_conf
->peer_addr
)->sin_addr
);
166 snprintf(af
, 20, "DRBD_PEER_AF=%s", afs
);
172 dev_info(DEV
, "helper command: %s %s %s\n", usermode_helper
, cmd
, mb
);
174 drbd_bcast_ev_helper(mdev
, cmd
);
175 ret
= call_usermodehelper(usermode_helper
, argv
, envp
, 1);
177 dev_warn(DEV
, "helper command: %s %s %s exit code %u (0x%x)\n",
178 usermode_helper
, cmd
, mb
,
179 (ret
>> 8) & 0xff, ret
);
181 dev_info(DEV
, "helper command: %s %s %s exit code %u (0x%x)\n",
182 usermode_helper
, cmd
, mb
,
183 (ret
>> 8) & 0xff, ret
);
185 if (ret
< 0) /* Ignore any ERRNOs we got. */
191 enum drbd_disk_state
drbd_try_outdate_peer(struct drbd_conf
*mdev
)
195 enum drbd_disk_state nps
;
196 enum drbd_fencing_p fp
;
198 D_ASSERT(mdev
->state
.pdsk
== D_UNKNOWN
);
200 if (get_ldev_if_state(mdev
, D_CONSISTENT
)) {
201 fp
= mdev
->ldev
->dc
.fencing
;
204 dev_warn(DEV
, "Not fencing peer, I'm not even Consistent myself.\n");
205 return mdev
->state
.pdsk
;
208 if (fp
== FP_STONITH
)
209 _drbd_request_state(mdev
, NS(susp
, 1), CS_WAIT_COMPLETE
);
211 r
= drbd_khelper(mdev
, "fence-peer");
213 switch ((r
>>8) & 0xff) {
214 case 3: /* peer is inconsistent */
215 ex_to_string
= "peer is inconsistent or worse";
216 nps
= D_INCONSISTENT
;
218 case 4: /* peer got outdated, or was already outdated */
219 ex_to_string
= "peer was fenced";
222 case 5: /* peer was down */
223 if (mdev
->state
.disk
== D_UP_TO_DATE
) {
224 /* we will(have) create(d) a new UUID anyways... */
225 ex_to_string
= "peer is unreachable, assumed to be dead";
228 ex_to_string
= "peer unreachable, doing nothing since disk != UpToDate";
229 nps
= mdev
->state
.pdsk
;
232 case 6: /* Peer is primary, voluntarily outdate myself.
233 * This is useful when an unconnected R_SECONDARY is asked to
234 * become R_PRIMARY, but finds the other peer being active. */
235 ex_to_string
= "peer is active";
236 dev_warn(DEV
, "Peer is primary, outdating myself.\n");
238 _drbd_request_state(mdev
, NS(disk
, D_OUTDATED
), CS_WAIT_COMPLETE
);
241 if (fp
!= FP_STONITH
)
242 dev_err(DEV
, "fence-peer() = 7 && fencing != Stonith !!!\n");
243 ex_to_string
= "peer was stonithed";
247 /* The script is broken ... */
249 dev_err(DEV
, "fence-peer helper broken, returned %d\n", (r
>>8)&0xff);
253 dev_info(DEV
, "fence-peer helper returned %d (%s)\n",
254 (r
>>8) & 0xff, ex_to_string
);
259 int drbd_set_role(struct drbd_conf
*mdev
, enum drbd_role new_role
, int force
)
261 const int max_tries
= 4;
265 union drbd_state mask
, val
;
266 enum drbd_disk_state nps
;
268 if (new_role
== R_PRIMARY
)
269 request_ping(mdev
); /* Detect a dead peer ASAP */
271 mutex_lock(&mdev
->state_mutex
);
273 mask
.i
= 0; mask
.role
= R_MASK
;
274 val
.i
= 0; val
.role
= new_role
;
276 while (try++ < max_tries
) {
277 r
= _drbd_request_state(mdev
, mask
, val
, CS_WAIT_COMPLETE
);
279 /* in case we first succeeded to outdate,
280 * but now suddenly could establish a connection */
281 if (r
== SS_CW_FAILED_BY_PEER
&& mask
.pdsk
!= 0) {
287 if (r
== SS_NO_UP_TO_DATE_DISK
&& force
&&
288 (mdev
->state
.disk
== D_INCONSISTENT
||
289 mdev
->state
.disk
== D_OUTDATED
)) {
291 val
.disk
= D_UP_TO_DATE
;
296 if (r
== SS_NO_UP_TO_DATE_DISK
&&
297 mdev
->state
.disk
== D_CONSISTENT
&& mask
.pdsk
== 0) {
298 D_ASSERT(mdev
->state
.pdsk
== D_UNKNOWN
);
299 nps
= drbd_try_outdate_peer(mdev
);
301 if (nps
== D_OUTDATED
|| nps
== D_INCONSISTENT
) {
302 val
.disk
= D_UP_TO_DATE
;
312 if (r
== SS_NOTHING_TO_DO
)
314 if (r
== SS_PRIMARY_NOP
&& mask
.pdsk
== 0) {
315 nps
= drbd_try_outdate_peer(mdev
);
317 if (force
&& nps
> D_OUTDATED
) {
318 dev_warn(DEV
, "Forced into split brain situation!\n");
327 if (r
== SS_TWO_PRIMARIES
) {
328 /* Maybe the peer is detected as dead very soon...
329 retry at most once more in this case. */
330 __set_current_state(TASK_INTERRUPTIBLE
);
331 schedule_timeout((mdev
->net_conf
->ping_timeo
+1)*HZ
/10);
336 if (r
< SS_SUCCESS
) {
337 r
= _drbd_request_state(mdev
, mask
, val
,
338 CS_VERBOSE
+ CS_WAIT_COMPLETE
);
349 dev_warn(DEV
, "Forced to consider local data as UpToDate!\n");
351 /* Wait until nothing is on the fly :) */
352 wait_event(mdev
->misc_wait
, atomic_read(&mdev
->ap_pending_cnt
) == 0);
354 if (new_role
== R_SECONDARY
) {
355 set_disk_ro(mdev
->vdisk
, TRUE
);
356 if (get_ldev(mdev
)) {
357 mdev
->ldev
->md
.uuid
[UI_CURRENT
] &= ~(u64
)1;
361 if (get_net_conf(mdev
)) {
362 mdev
->net_conf
->want_lose
= 0;
365 set_disk_ro(mdev
->vdisk
, FALSE
);
366 if (get_ldev(mdev
)) {
367 if (((mdev
->state
.conn
< C_CONNECTED
||
368 mdev
->state
.pdsk
<= D_FAILED
)
369 && mdev
->ldev
->md
.uuid
[UI_BITMAP
] == 0) || forced
)
370 drbd_uuid_new_current(mdev
);
372 mdev
->ldev
->md
.uuid
[UI_CURRENT
] |= (u64
)1;
377 if ((new_role
== R_SECONDARY
) && get_ldev(mdev
)) {
378 drbd_al_to_on_disk_bm(mdev
);
382 if (mdev
->state
.conn
>= C_WF_REPORT_PARAMS
) {
383 /* if this was forced, we should consider sync */
385 drbd_send_uuids(mdev
);
386 drbd_send_state(mdev
);
391 kobject_uevent(&disk_to_dev(mdev
->vdisk
)->kobj
, KOBJ_CHANGE
);
393 mutex_unlock(&mdev
->state_mutex
);
398 static int drbd_nl_primary(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
399 struct drbd_nl_cfg_reply
*reply
)
401 struct primary primary_args
;
403 memset(&primary_args
, 0, sizeof(struct primary
));
404 if (!primary_from_tags(mdev
, nlp
->tag_list
, &primary_args
)) {
405 reply
->ret_code
= ERR_MANDATORY_TAG
;
410 drbd_set_role(mdev
, R_PRIMARY
, primary_args
.overwrite_peer
);
415 static int drbd_nl_secondary(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
416 struct drbd_nl_cfg_reply
*reply
)
418 reply
->ret_code
= drbd_set_role(mdev
, R_SECONDARY
, 0);
423 /* initializes the md.*_offset members, so we are able to find
424 * the on disk meta data */
425 static void drbd_md_set_sector_offsets(struct drbd_conf
*mdev
,
426 struct drbd_backing_dev
*bdev
)
428 sector_t md_size_sect
= 0;
429 switch (bdev
->dc
.meta_dev_idx
) {
431 /* v07 style fixed size indexed meta data */
432 bdev
->md
.md_size_sect
= MD_RESERVED_SECT
;
433 bdev
->md
.md_offset
= drbd_md_ss__(mdev
, bdev
);
434 bdev
->md
.al_offset
= MD_AL_OFFSET
;
435 bdev
->md
.bm_offset
= MD_BM_OFFSET
;
437 case DRBD_MD_INDEX_FLEX_EXT
:
438 /* just occupy the full device; unit: sectors */
439 bdev
->md
.md_size_sect
= drbd_get_capacity(bdev
->md_bdev
);
440 bdev
->md
.md_offset
= 0;
441 bdev
->md
.al_offset
= MD_AL_OFFSET
;
442 bdev
->md
.bm_offset
= MD_BM_OFFSET
;
444 case DRBD_MD_INDEX_INTERNAL
:
445 case DRBD_MD_INDEX_FLEX_INT
:
446 bdev
->md
.md_offset
= drbd_md_ss__(mdev
, bdev
);
447 /* al size is still fixed */
448 bdev
->md
.al_offset
= -MD_AL_MAX_SIZE
;
449 /* we need (slightly less than) ~ this much bitmap sectors: */
450 md_size_sect
= drbd_get_capacity(bdev
->backing_bdev
);
451 md_size_sect
= ALIGN(md_size_sect
, BM_SECT_PER_EXT
);
452 md_size_sect
= BM_SECT_TO_EXT(md_size_sect
);
453 md_size_sect
= ALIGN(md_size_sect
, 8);
455 /* plus the "drbd meta data super block",
456 * and the activity log; */
457 md_size_sect
+= MD_BM_OFFSET
;
459 bdev
->md
.md_size_sect
= md_size_sect
;
460 /* bitmap offset is adjusted by 'super' block size */
461 bdev
->md
.bm_offset
= -md_size_sect
+ MD_AL_OFFSET
;
466 char *ppsize(char *buf
, unsigned long long size
)
468 /* Needs 9 bytes at max. */
469 static char units
[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
471 while (size
>= 10000) {
473 size
= (size
>> 10) + !!(size
& (1<<9));
476 sprintf(buf
, "%lu %cB", (long)size
, units
[base
]);
481 /* there is still a theoretical deadlock when called from receiver
482 * on an D_INCONSISTENT R_PRIMARY:
483 * remote READ does inc_ap_bio, receiver would need to receive answer
484 * packet from remote to dec_ap_bio again.
485 * receiver receive_sizes(), comes here,
486 * waits for ap_bio_cnt == 0. -> deadlock.
487 * but this cannot happen, actually, because:
488 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
489 * (not connected, or bad/no disk on peer):
490 * see drbd_fail_request_early, ap_bio_cnt is zero.
491 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
492 * peer may not initiate a resize.
494 void drbd_suspend_io(struct drbd_conf
*mdev
)
496 set_bit(SUSPEND_IO
, &mdev
->flags
);
497 wait_event(mdev
->misc_wait
, !atomic_read(&mdev
->ap_bio_cnt
));
500 void drbd_resume_io(struct drbd_conf
*mdev
)
502 clear_bit(SUSPEND_IO
, &mdev
->flags
);
503 wake_up(&mdev
->misc_wait
);
507 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
508 * @mdev: DRBD device.
510 * Returns 0 on success, negative return values indicate errors.
511 * You should call drbd_md_sync() after calling this function.
513 enum determine_dev_size
drbd_determin_dev_size(struct drbd_conf
*mdev
) __must_hold(local
)
515 sector_t prev_first_sect
, prev_size
; /* previous meta location */
520 int md_moved
, la_size_changed
;
521 enum determine_dev_size rv
= unchanged
;
524 * application request passes inc_ap_bio,
525 * but then cannot get an AL-reference.
526 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
529 * Suspend IO right here.
530 * still lock the act_log to not trigger ASSERTs there.
532 drbd_suspend_io(mdev
);
534 /* no wait necessary anymore, actually we could assert that */
535 wait_event(mdev
->al_wait
, lc_try_lock(mdev
->act_log
));
537 prev_first_sect
= drbd_md_first_sector(mdev
->ldev
);
538 prev_size
= mdev
->ldev
->md
.md_size_sect
;
539 la_size
= mdev
->ldev
->md
.la_size_sect
;
541 /* TODO: should only be some assert here, not (re)init... */
542 drbd_md_set_sector_offsets(mdev
, mdev
->ldev
);
544 size
= drbd_new_dev_size(mdev
, mdev
->ldev
);
546 if (drbd_get_capacity(mdev
->this_bdev
) != size
||
547 drbd_bm_capacity(mdev
) != size
) {
549 err
= drbd_bm_resize(mdev
, size
);
551 /* currently there is only one error: ENOMEM! */
552 size
= drbd_bm_capacity(mdev
)>>1;
554 dev_err(DEV
, "OUT OF MEMORY! "
555 "Could not allocate bitmap!\n");
557 dev_err(DEV
, "BM resizing failed. "
558 "Leaving size unchanged at size = %lu KB\n",
559 (unsigned long)size
);
563 /* racy, see comments above. */
564 drbd_set_my_capacity(mdev
, size
);
565 mdev
->ldev
->md
.la_size_sect
= size
;
566 dev_info(DEV
, "size = %s (%llu KB)\n", ppsize(ppb
, size
>>1),
567 (unsigned long long)size
>>1);
569 if (rv
== dev_size_error
)
572 la_size_changed
= (la_size
!= mdev
->ldev
->md
.la_size_sect
);
574 md_moved
= prev_first_sect
!= drbd_md_first_sector(mdev
->ldev
)
575 || prev_size
!= mdev
->ldev
->md
.md_size_sect
;
577 if (la_size_changed
|| md_moved
) {
578 drbd_al_shrink(mdev
); /* All extents inactive. */
579 dev_info(DEV
, "Writing the whole bitmap, %s\n",
580 la_size_changed
&& md_moved
? "size changed and md moved" :
581 la_size_changed
? "size changed" : "md moved");
582 rv
= drbd_bitmap_io(mdev
, &drbd_bm_write
, "size changed"); /* does drbd_resume_io() ! */
583 drbd_md_mark_dirty(mdev
);
591 lc_unlock(mdev
->act_log
);
592 wake_up(&mdev
->al_wait
);
593 drbd_resume_io(mdev
);
599 drbd_new_dev_size(struct drbd_conf
*mdev
, struct drbd_backing_dev
*bdev
)
601 sector_t p_size
= mdev
->p_size
; /* partner's disk size. */
602 sector_t la_size
= bdev
->md
.la_size_sect
; /* last agreed size. */
603 sector_t m_size
; /* my size */
604 sector_t u_size
= bdev
->dc
.disk_size
; /* size requested by user. */
607 m_size
= drbd_get_max_capacity(bdev
);
609 if (p_size
&& m_size
) {
610 size
= min_t(sector_t
, p_size
, m_size
);
614 if (m_size
&& m_size
< size
)
616 if (p_size
&& p_size
< size
)
627 dev_err(DEV
, "Both nodes diskless!\n");
631 dev_err(DEV
, "Requested disk size is too big (%lu > %lu)\n",
632 (unsigned long)u_size
>>1, (unsigned long)size
>>1);
641 * drbd_check_al_size() - Ensures that the AL is of the right size
642 * @mdev: DRBD device.
644 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
645 * failed, and 0 on success. You should call drbd_md_sync() after you called
648 static int drbd_check_al_size(struct drbd_conf
*mdev
)
650 struct lru_cache
*n
, *t
;
651 struct lc_element
*e
;
655 ERR_IF(mdev
->sync_conf
.al_extents
< 7)
656 mdev
->sync_conf
.al_extents
= 127;
659 mdev
->act_log
->nr_elements
== mdev
->sync_conf
.al_extents
)
664 n
= lc_create("act_log", drbd_al_ext_cache
,
665 mdev
->sync_conf
.al_extents
, sizeof(struct lc_element
), 0);
668 dev_err(DEV
, "Cannot allocate act_log lru!\n");
671 spin_lock_irq(&mdev
->al_lock
);
673 for (i
= 0; i
< t
->nr_elements
; i
++) {
674 e
= lc_element_by_index(t
, i
);
676 dev_err(DEV
, "refcnt(%d)==%d\n",
677 e
->lc_number
, e
->refcnt
);
683 spin_unlock_irq(&mdev
->al_lock
);
685 dev_err(DEV
, "Activity log still in use!\n");
692 drbd_md_mark_dirty(mdev
); /* we changed mdev->act_log->nr_elemens */
696 void drbd_setup_queue_param(struct drbd_conf
*mdev
, unsigned int max_seg_s
) __must_hold(local
)
698 struct request_queue
* const q
= mdev
->rq_queue
;
699 struct request_queue
* const b
= mdev
->ldev
->backing_bdev
->bd_disk
->queue
;
700 int max_segments
= mdev
->ldev
->dc
.max_bio_bvecs
;
702 if (b
->merge_bvec_fn
&& !mdev
->ldev
->dc
.use_bmbv
)
703 max_seg_s
= PAGE_SIZE
;
705 max_seg_s
= min(queue_max_sectors(b
) * queue_logical_block_size(b
), max_seg_s
);
707 blk_queue_max_sectors(q
, max_seg_s
>> 9);
708 blk_queue_max_phys_segments(q
, max_segments
? max_segments
: MAX_PHYS_SEGMENTS
);
709 blk_queue_max_hw_segments(q
, max_segments
? max_segments
: MAX_HW_SEGMENTS
);
710 blk_queue_max_segment_size(q
, max_seg_s
);
711 blk_queue_logical_block_size(q
, 512);
712 blk_queue_segment_boundary(q
, PAGE_SIZE
-1);
713 blk_stack_limits(&q
->limits
, &b
->limits
, 0);
715 if (b
->merge_bvec_fn
)
716 dev_warn(DEV
, "Backing device's merge_bvec_fn() = %p\n",
718 dev_info(DEV
, "max_segment_size ( = BIO size ) = %u\n", queue_max_segment_size(q
));
720 if (q
->backing_dev_info
.ra_pages
!= b
->backing_dev_info
.ra_pages
) {
721 dev_info(DEV
, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
722 q
->backing_dev_info
.ra_pages
,
723 b
->backing_dev_info
.ra_pages
);
724 q
->backing_dev_info
.ra_pages
= b
->backing_dev_info
.ra_pages
;
728 /* serialize deconfig (worker exiting, doing cleanup)
729 * and reconfig (drbdsetup disk, drbdsetup net)
731 * wait for a potentially exiting worker, then restart it,
732 * or start a new one.
734 static void drbd_reconfig_start(struct drbd_conf
*mdev
)
736 wait_event(mdev
->state_wait
, test_and_set_bit(CONFIG_PENDING
, &mdev
->flags
));
737 wait_event(mdev
->state_wait
, !test_bit(DEVICE_DYING
, &mdev
->flags
));
738 drbd_thread_start(&mdev
->worker
);
741 /* if still unconfigured, stops worker again.
742 * if configured now, clears CONFIG_PENDING.
743 * wakes potential waiters */
744 static void drbd_reconfig_done(struct drbd_conf
*mdev
)
746 spin_lock_irq(&mdev
->req_lock
);
747 if (mdev
->state
.disk
== D_DISKLESS
&&
748 mdev
->state
.conn
== C_STANDALONE
&&
749 mdev
->state
.role
== R_SECONDARY
) {
750 set_bit(DEVICE_DYING
, &mdev
->flags
);
751 drbd_thread_stop_nowait(&mdev
->worker
);
753 clear_bit(CONFIG_PENDING
, &mdev
->flags
);
754 spin_unlock_irq(&mdev
->req_lock
);
755 wake_up(&mdev
->state_wait
);
758 /* does always return 0;
759 * interesting return code is in reply->ret_code */
760 static int drbd_nl_disk_conf(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
761 struct drbd_nl_cfg_reply
*reply
)
763 enum drbd_ret_codes retcode
;
764 enum determine_dev_size dd
;
765 sector_t max_possible_sectors
;
766 sector_t min_md_device_sectors
;
767 struct drbd_backing_dev
*nbc
= NULL
; /* new_backing_conf */
768 struct inode
*inode
, *inode2
;
769 struct lru_cache
*resync_lru
= NULL
;
770 union drbd_state ns
, os
;
772 int cp_discovered
= 0;
773 int logical_block_size
;
775 drbd_reconfig_start(mdev
);
777 /* if you want to reconfigure, please tear down first */
778 if (mdev
->state
.disk
> D_DISKLESS
) {
779 retcode
= ERR_DISK_CONFIGURED
;
783 /* allocation not in the IO path, cqueue thread context */
784 nbc
= kzalloc(sizeof(struct drbd_backing_dev
), GFP_KERNEL
);
790 nbc
->dc
.disk_size
= DRBD_DISK_SIZE_SECT_DEF
;
791 nbc
->dc
.on_io_error
= DRBD_ON_IO_ERROR_DEF
;
792 nbc
->dc
.fencing
= DRBD_FENCING_DEF
;
793 nbc
->dc
.max_bio_bvecs
= DRBD_MAX_BIO_BVECS_DEF
;
795 if (!disk_conf_from_tags(mdev
, nlp
->tag_list
, &nbc
->dc
)) {
796 retcode
= ERR_MANDATORY_TAG
;
800 if (nbc
->dc
.meta_dev_idx
< DRBD_MD_INDEX_FLEX_INT
) {
801 retcode
= ERR_MD_IDX_INVALID
;
805 nbc
->lo_file
= filp_open(nbc
->dc
.backing_dev
, O_RDWR
, 0);
806 if (IS_ERR(nbc
->lo_file
)) {
807 dev_err(DEV
, "open(\"%s\") failed with %ld\n", nbc
->dc
.backing_dev
,
808 PTR_ERR(nbc
->lo_file
));
810 retcode
= ERR_OPEN_DISK
;
814 inode
= nbc
->lo_file
->f_dentry
->d_inode
;
816 if (!S_ISBLK(inode
->i_mode
)) {
817 retcode
= ERR_DISK_NOT_BDEV
;
821 nbc
->md_file
= filp_open(nbc
->dc
.meta_dev
, O_RDWR
, 0);
822 if (IS_ERR(nbc
->md_file
)) {
823 dev_err(DEV
, "open(\"%s\") failed with %ld\n", nbc
->dc
.meta_dev
,
824 PTR_ERR(nbc
->md_file
));
826 retcode
= ERR_OPEN_MD_DISK
;
830 inode2
= nbc
->md_file
->f_dentry
->d_inode
;
832 if (!S_ISBLK(inode2
->i_mode
)) {
833 retcode
= ERR_MD_NOT_BDEV
;
837 nbc
->backing_bdev
= inode
->i_bdev
;
838 if (bd_claim(nbc
->backing_bdev
, mdev
)) {
839 printk(KERN_ERR
"drbd: bd_claim(%p,%p); failed [%p;%p;%u]\n",
840 nbc
->backing_bdev
, mdev
,
841 nbc
->backing_bdev
->bd_holder
,
842 nbc
->backing_bdev
->bd_contains
->bd_holder
,
843 nbc
->backing_bdev
->bd_holders
);
844 retcode
= ERR_BDCLAIM_DISK
;
848 resync_lru
= lc_create("resync", drbd_bm_ext_cache
,
849 61, sizeof(struct bm_extent
),
850 offsetof(struct bm_extent
, lce
));
853 goto release_bdev_fail
;
856 /* meta_dev_idx >= 0: external fixed size,
857 * possibly multiple drbd sharing one meta device.
858 * TODO in that case, paranoia check that [md_bdev, meta_dev_idx] is
859 * not yet used by some other drbd minor!
860 * (if you use drbd.conf + drbdadm,
861 * that should check it for you already; but if you don't, or someone
862 * fooled it, we need to double check here) */
863 nbc
->md_bdev
= inode2
->i_bdev
;
864 if (bd_claim(nbc
->md_bdev
, (nbc
->dc
.meta_dev_idx
< 0) ? (void *)mdev
865 : (void *) drbd_m_holder
)) {
866 retcode
= ERR_BDCLAIM_MD_DISK
;
867 goto release_bdev_fail
;
870 if ((nbc
->backing_bdev
== nbc
->md_bdev
) !=
871 (nbc
->dc
.meta_dev_idx
== DRBD_MD_INDEX_INTERNAL
||
872 nbc
->dc
.meta_dev_idx
== DRBD_MD_INDEX_FLEX_INT
)) {
873 retcode
= ERR_MD_IDX_INVALID
;
874 goto release_bdev2_fail
;
877 /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
878 drbd_md_set_sector_offsets(mdev
, nbc
);
880 if (drbd_get_max_capacity(nbc
) < nbc
->dc
.disk_size
) {
881 dev_err(DEV
, "max capacity %llu smaller than disk size %llu\n",
882 (unsigned long long) drbd_get_max_capacity(nbc
),
883 (unsigned long long) nbc
->dc
.disk_size
);
884 retcode
= ERR_DISK_TO_SMALL
;
885 goto release_bdev2_fail
;
888 if (nbc
->dc
.meta_dev_idx
< 0) {
889 max_possible_sectors
= DRBD_MAX_SECTORS_FLEX
;
890 /* at least one MB, otherwise it does not make sense */
891 min_md_device_sectors
= (2<<10);
893 max_possible_sectors
= DRBD_MAX_SECTORS
;
894 min_md_device_sectors
= MD_RESERVED_SECT
* (nbc
->dc
.meta_dev_idx
+ 1);
897 if (drbd_get_capacity(nbc
->md_bdev
) > max_possible_sectors
)
898 dev_warn(DEV
, "truncating very big lower level device "
899 "to currently maximum possible %llu sectors\n",
900 (unsigned long long) max_possible_sectors
);
902 if (drbd_get_capacity(nbc
->md_bdev
) < min_md_device_sectors
) {
903 retcode
= ERR_MD_DISK_TO_SMALL
;
904 dev_warn(DEV
, "refusing attach: md-device too small, "
905 "at least %llu sectors needed for this meta-disk type\n",
906 (unsigned long long) min_md_device_sectors
);
907 goto release_bdev2_fail
;
910 /* Make sure the new disk is big enough
911 * (we may currently be R_PRIMARY with no local disk...) */
912 if (drbd_get_max_capacity(nbc
) <
913 drbd_get_capacity(mdev
->this_bdev
)) {
914 retcode
= ERR_DISK_TO_SMALL
;
915 goto release_bdev2_fail
;
918 nbc
->known_size
= drbd_get_capacity(nbc
->backing_bdev
);
920 drbd_suspend_io(mdev
);
921 /* also wait for the last barrier ack. */
922 wait_event(mdev
->misc_wait
, !atomic_read(&mdev
->ap_pending_cnt
));
923 /* and for any other previously queued work */
924 drbd_flush_workqueue(mdev
);
926 retcode
= _drbd_request_state(mdev
, NS(disk
, D_ATTACHING
), CS_VERBOSE
);
927 drbd_resume_io(mdev
);
928 if (retcode
< SS_SUCCESS
)
929 goto release_bdev2_fail
;
931 if (!get_ldev_if_state(mdev
, D_ATTACHING
))
934 drbd_md_set_sector_offsets(mdev
, nbc
);
937 if (drbd_bm_init(mdev
)) {
939 goto force_diskless_dec
;
943 retcode
= drbd_md_read(mdev
, nbc
);
944 if (retcode
!= NO_ERROR
)
945 goto force_diskless_dec
;
947 if (mdev
->state
.conn
< C_CONNECTED
&&
948 mdev
->state
.role
== R_PRIMARY
&&
949 (mdev
->ed_uuid
& ~((u64
)1)) != (nbc
->md
.uuid
[UI_CURRENT
] & ~((u64
)1))) {
950 dev_err(DEV
, "Can only attach to data with current UUID=%016llX\n",
951 (unsigned long long)mdev
->ed_uuid
);
952 retcode
= ERR_DATA_NOT_CURRENT
;
953 goto force_diskless_dec
;
956 /* Since we are diskless, fix the activity log first... */
957 if (drbd_check_al_size(mdev
)) {
959 goto force_diskless_dec
;
962 /* Prevent shrinking of consistent devices ! */
963 if (drbd_md_test_flag(nbc
, MDF_CONSISTENT
) &&
964 drbd_new_dev_size(mdev
, nbc
) < nbc
->md
.la_size_sect
) {
965 dev_warn(DEV
, "refusing to truncate a consistent device\n");
966 retcode
= ERR_DISK_TO_SMALL
;
967 goto force_diskless_dec
;
970 if (!drbd_al_read_log(mdev
, nbc
)) {
971 retcode
= ERR_IO_MD_DISK
;
972 goto force_diskless_dec
;
975 /* allocate a second IO page if logical_block_size != 512 */
976 logical_block_size
= bdev_logical_block_size(nbc
->md_bdev
);
977 if (logical_block_size
== 0)
978 logical_block_size
= MD_SECTOR_SIZE
;
980 if (logical_block_size
!= MD_SECTOR_SIZE
) {
981 if (!mdev
->md_io_tmpp
) {
982 struct page
*page
= alloc_page(GFP_NOIO
);
984 goto force_diskless_dec
;
986 dev_warn(DEV
, "Meta data's bdev logical_block_size = %d != %d\n",
987 logical_block_size
, MD_SECTOR_SIZE
);
988 dev_warn(DEV
, "Workaround engaged (has performance impact).\n");
990 mdev
->md_io_tmpp
= page
;
994 /* Reset the "barriers don't work" bits here, then force meta data to
995 * be written, to ensure we determine if barriers are supported. */
996 if (nbc
->dc
.no_md_flush
)
997 set_bit(MD_NO_BARRIER
, &mdev
->flags
);
999 clear_bit(MD_NO_BARRIER
, &mdev
->flags
);
1001 /* Point of no return reached.
1002 * Devices and memory are no longer released by error cleanup below.
1003 * now mdev takes over responsibility, and the state engine should
1004 * clean it up somewhere. */
1005 D_ASSERT(mdev
->ldev
== NULL
);
1007 mdev
->resync
= resync_lru
;
1011 mdev
->write_ordering
= WO_bio_barrier
;
1012 drbd_bump_write_ordering(mdev
, WO_bio_barrier
);
1014 if (drbd_md_test_flag(mdev
->ldev
, MDF_CRASHED_PRIMARY
))
1015 set_bit(CRASHED_PRIMARY
, &mdev
->flags
);
1017 clear_bit(CRASHED_PRIMARY
, &mdev
->flags
);
1019 if (drbd_md_test_flag(mdev
->ldev
, MDF_PRIMARY_IND
)) {
1020 set_bit(CRASHED_PRIMARY
, &mdev
->flags
);
1029 drbd_setup_queue_param(mdev
, DRBD_MAX_SEGMENT_SIZE
);
1031 /* If I am currently not R_PRIMARY,
1032 * but meta data primary indicator is set,
1033 * I just now recover from a hard crash,
1034 * and have been R_PRIMARY before that crash.
1036 * Now, if I had no connection before that crash
1037 * (have been degraded R_PRIMARY), chances are that
1038 * I won't find my peer now either.
1040 * In that case, and _only_ in that case,
1041 * we use the degr-wfc-timeout instead of the default,
1042 * so we can automatically recover from a crash of a
1043 * degraded but active "cluster" after a certain timeout.
1045 clear_bit(USE_DEGR_WFC_T
, &mdev
->flags
);
1046 if (mdev
->state
.role
!= R_PRIMARY
&&
1047 drbd_md_test_flag(mdev
->ldev
, MDF_PRIMARY_IND
) &&
1048 !drbd_md_test_flag(mdev
->ldev
, MDF_CONNECTED_IND
))
1049 set_bit(USE_DEGR_WFC_T
, &mdev
->flags
);
1051 dd
= drbd_determin_dev_size(mdev
);
1052 if (dd
== dev_size_error
) {
1053 retcode
= ERR_NOMEM_BITMAP
;
1054 goto force_diskless_dec
;
1055 } else if (dd
== grew
)
1056 set_bit(RESYNC_AFTER_NEG
, &mdev
->flags
);
1058 if (drbd_md_test_flag(mdev
->ldev
, MDF_FULL_SYNC
)) {
1059 dev_info(DEV
, "Assuming that all blocks are out of sync "
1060 "(aka FullSync)\n");
1061 if (drbd_bitmap_io(mdev
, &drbd_bmio_set_n_write
, "set_n_write from attaching")) {
1062 retcode
= ERR_IO_MD_DISK
;
1063 goto force_diskless_dec
;
1066 if (drbd_bitmap_io(mdev
, &drbd_bm_read
, "read from attaching") < 0) {
1067 retcode
= ERR_IO_MD_DISK
;
1068 goto force_diskless_dec
;
1072 if (cp_discovered
) {
1073 drbd_al_apply_to_bm(mdev
);
1074 drbd_al_to_on_disk_bm(mdev
);
1077 spin_lock_irq(&mdev
->req_lock
);
1080 /* If MDF_CONSISTENT is not set go into inconsistent state,
1081 otherwise investigate MDF_WasUpToDate...
1082 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1083 otherwise into D_CONSISTENT state.
1085 if (drbd_md_test_flag(mdev
->ldev
, MDF_CONSISTENT
)) {
1086 if (drbd_md_test_flag(mdev
->ldev
, MDF_WAS_UP_TO_DATE
))
1087 ns
.disk
= D_CONSISTENT
;
1089 ns
.disk
= D_OUTDATED
;
1091 ns
.disk
= D_INCONSISTENT
;
1094 if (drbd_md_test_flag(mdev
->ldev
, MDF_PEER_OUT_DATED
))
1095 ns
.pdsk
= D_OUTDATED
;
1097 if ( ns
.disk
== D_CONSISTENT
&&
1098 (ns
.pdsk
== D_OUTDATED
|| mdev
->ldev
->dc
.fencing
== FP_DONT_CARE
))
1099 ns
.disk
= D_UP_TO_DATE
;
1101 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1102 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1103 this point, because drbd_request_state() modifies these
1106 /* In case we are C_CONNECTED postpone any decision on the new disk
1107 state after the negotiation phase. */
1108 if (mdev
->state
.conn
== C_CONNECTED
) {
1109 mdev
->new_state_tmp
.i
= ns
.i
;
1111 ns
.disk
= D_NEGOTIATING
;
1114 rv
= _drbd_set_state(mdev
, ns
, CS_VERBOSE
, NULL
);
1116 spin_unlock_irq(&mdev
->req_lock
);
1118 if (rv
< SS_SUCCESS
)
1119 goto force_diskless_dec
;
1121 if (mdev
->state
.role
== R_PRIMARY
)
1122 mdev
->ldev
->md
.uuid
[UI_CURRENT
] |= (u64
)1;
1124 mdev
->ldev
->md
.uuid
[UI_CURRENT
] &= ~(u64
)1;
1126 drbd_md_mark_dirty(mdev
);
1129 kobject_uevent(&disk_to_dev(mdev
->vdisk
)->kobj
, KOBJ_CHANGE
);
1131 reply
->ret_code
= retcode
;
1132 drbd_reconfig_done(mdev
);
1138 drbd_force_state(mdev
, NS(disk
, D_DISKLESS
));
1142 bd_release(nbc
->md_bdev
);
1145 bd_release(nbc
->backing_bdev
);
1154 lc_destroy(resync_lru
);
1156 reply
->ret_code
= retcode
;
1157 drbd_reconfig_done(mdev
);
1161 static int drbd_nl_detach(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1162 struct drbd_nl_cfg_reply
*reply
)
1164 reply
->ret_code
= drbd_request_state(mdev
, NS(disk
, D_DISKLESS
));
1168 static int drbd_nl_net_conf(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1169 struct drbd_nl_cfg_reply
*reply
)
1172 enum drbd_ret_codes retcode
;
1173 struct net_conf
*new_conf
= NULL
;
1174 struct crypto_hash
*tfm
= NULL
;
1175 struct crypto_hash
*integrity_w_tfm
= NULL
;
1176 struct crypto_hash
*integrity_r_tfm
= NULL
;
1177 struct hlist_head
*new_tl_hash
= NULL
;
1178 struct hlist_head
*new_ee_hash
= NULL
;
1179 struct drbd_conf
*odev
;
1180 char hmac_name
[CRYPTO_MAX_ALG_NAME
];
1181 void *int_dig_out
= NULL
;
1182 void *int_dig_in
= NULL
;
1183 void *int_dig_vv
= NULL
;
1184 struct sockaddr
*new_my_addr
, *new_peer_addr
, *taken_addr
;
1186 drbd_reconfig_start(mdev
);
1188 if (mdev
->state
.conn
> C_STANDALONE
) {
1189 retcode
= ERR_NET_CONFIGURED
;
1193 /* allocation not in the IO path, cqueue thread context */
1194 new_conf
= kmalloc(sizeof(struct net_conf
), GFP_KERNEL
);
1196 retcode
= ERR_NOMEM
;
1200 memset(new_conf
, 0, sizeof(struct net_conf
));
1201 new_conf
->timeout
= DRBD_TIMEOUT_DEF
;
1202 new_conf
->try_connect_int
= DRBD_CONNECT_INT_DEF
;
1203 new_conf
->ping_int
= DRBD_PING_INT_DEF
;
1204 new_conf
->max_epoch_size
= DRBD_MAX_EPOCH_SIZE_DEF
;
1205 new_conf
->max_buffers
= DRBD_MAX_BUFFERS_DEF
;
1206 new_conf
->unplug_watermark
= DRBD_UNPLUG_WATERMARK_DEF
;
1207 new_conf
->sndbuf_size
= DRBD_SNDBUF_SIZE_DEF
;
1208 new_conf
->rcvbuf_size
= DRBD_RCVBUF_SIZE_DEF
;
1209 new_conf
->ko_count
= DRBD_KO_COUNT_DEF
;
1210 new_conf
->after_sb_0p
= DRBD_AFTER_SB_0P_DEF
;
1211 new_conf
->after_sb_1p
= DRBD_AFTER_SB_1P_DEF
;
1212 new_conf
->after_sb_2p
= DRBD_AFTER_SB_2P_DEF
;
1213 new_conf
->want_lose
= 0;
1214 new_conf
->two_primaries
= 0;
1215 new_conf
->wire_protocol
= DRBD_PROT_C
;
1216 new_conf
->ping_timeo
= DRBD_PING_TIMEO_DEF
;
1217 new_conf
->rr_conflict
= DRBD_RR_CONFLICT_DEF
;
1219 if (!net_conf_from_tags(mdev
, nlp
->tag_list
, new_conf
)) {
1220 retcode
= ERR_MANDATORY_TAG
;
1224 if (new_conf
->two_primaries
1225 && (new_conf
->wire_protocol
!= DRBD_PROT_C
)) {
1226 retcode
= ERR_NOT_PROTO_C
;
1230 if (mdev
->state
.role
== R_PRIMARY
&& new_conf
->want_lose
) {
1231 retcode
= ERR_DISCARD
;
1237 new_my_addr
= (struct sockaddr
*)&new_conf
->my_addr
;
1238 new_peer_addr
= (struct sockaddr
*)&new_conf
->peer_addr
;
1239 for (i
= 0; i
< minor_count
; i
++) {
1240 odev
= minor_to_mdev(i
);
1241 if (!odev
|| odev
== mdev
)
1243 if (get_net_conf(odev
)) {
1244 taken_addr
= (struct sockaddr
*)&odev
->net_conf
->my_addr
;
1245 if (new_conf
->my_addr_len
== odev
->net_conf
->my_addr_len
&&
1246 !memcmp(new_my_addr
, taken_addr
, new_conf
->my_addr_len
))
1247 retcode
= ERR_LOCAL_ADDR
;
1249 taken_addr
= (struct sockaddr
*)&odev
->net_conf
->peer_addr
;
1250 if (new_conf
->peer_addr_len
== odev
->net_conf
->peer_addr_len
&&
1251 !memcmp(new_peer_addr
, taken_addr
, new_conf
->peer_addr_len
))
1252 retcode
= ERR_PEER_ADDR
;
1255 if (retcode
!= NO_ERROR
)
1260 if (new_conf
->cram_hmac_alg
[0] != 0) {
1261 snprintf(hmac_name
, CRYPTO_MAX_ALG_NAME
, "hmac(%s)",
1262 new_conf
->cram_hmac_alg
);
1263 tfm
= crypto_alloc_hash(hmac_name
, 0, CRYPTO_ALG_ASYNC
);
1266 retcode
= ERR_AUTH_ALG
;
1270 if (crypto_tfm_alg_type(crypto_hash_tfm(tfm
))
1271 != CRYPTO_ALG_TYPE_HASH
) {
1272 retcode
= ERR_AUTH_ALG_ND
;
1277 if (new_conf
->integrity_alg
[0]) {
1278 integrity_w_tfm
= crypto_alloc_hash(new_conf
->integrity_alg
, 0, CRYPTO_ALG_ASYNC
);
1279 if (IS_ERR(integrity_w_tfm
)) {
1280 integrity_w_tfm
= NULL
;
1281 retcode
=ERR_INTEGRITY_ALG
;
1285 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm
))) {
1286 retcode
=ERR_INTEGRITY_ALG_ND
;
1290 integrity_r_tfm
= crypto_alloc_hash(new_conf
->integrity_alg
, 0, CRYPTO_ALG_ASYNC
);
1291 if (IS_ERR(integrity_r_tfm
)) {
1292 integrity_r_tfm
= NULL
;
1293 retcode
=ERR_INTEGRITY_ALG
;
1298 ns
= new_conf
->max_epoch_size
/8;
1299 if (mdev
->tl_hash_s
!= ns
) {
1300 new_tl_hash
= kzalloc(ns
*sizeof(void *), GFP_KERNEL
);
1302 retcode
= ERR_NOMEM
;
1307 ns
= new_conf
->max_buffers
/8;
1308 if (new_conf
->two_primaries
&& (mdev
->ee_hash_s
!= ns
)) {
1309 new_ee_hash
= kzalloc(ns
*sizeof(void *), GFP_KERNEL
);
1311 retcode
= ERR_NOMEM
;
1316 ((char *)new_conf
->shared_secret
)[SHARED_SECRET_MAX
-1] = 0;
1318 if (integrity_w_tfm
) {
1319 i
= crypto_hash_digestsize(integrity_w_tfm
);
1320 int_dig_out
= kmalloc(i
, GFP_KERNEL
);
1322 retcode
= ERR_NOMEM
;
1325 int_dig_in
= kmalloc(i
, GFP_KERNEL
);
1327 retcode
= ERR_NOMEM
;
1330 int_dig_vv
= kmalloc(i
, GFP_KERNEL
);
1332 retcode
= ERR_NOMEM
;
1337 if (!mdev
->bitmap
) {
1338 if(drbd_bm_init(mdev
)) {
1339 retcode
= ERR_NOMEM
;
1344 spin_lock_irq(&mdev
->req_lock
);
1345 if (mdev
->net_conf
!= NULL
) {
1346 retcode
= ERR_NET_CONFIGURED
;
1347 spin_unlock_irq(&mdev
->req_lock
);
1350 mdev
->net_conf
= new_conf
;
1356 kfree(mdev
->tl_hash
);
1357 mdev
->tl_hash_s
= mdev
->net_conf
->max_epoch_size
/8;
1358 mdev
->tl_hash
= new_tl_hash
;
1362 kfree(mdev
->ee_hash
);
1363 mdev
->ee_hash_s
= mdev
->net_conf
->max_buffers
/8;
1364 mdev
->ee_hash
= new_ee_hash
;
1367 crypto_free_hash(mdev
->cram_hmac_tfm
);
1368 mdev
->cram_hmac_tfm
= tfm
;
1370 crypto_free_hash(mdev
->integrity_w_tfm
);
1371 mdev
->integrity_w_tfm
= integrity_w_tfm
;
1373 crypto_free_hash(mdev
->integrity_r_tfm
);
1374 mdev
->integrity_r_tfm
= integrity_r_tfm
;
1376 kfree(mdev
->int_dig_out
);
1377 kfree(mdev
->int_dig_in
);
1378 kfree(mdev
->int_dig_vv
);
1379 mdev
->int_dig_out
=int_dig_out
;
1380 mdev
->int_dig_in
=int_dig_in
;
1381 mdev
->int_dig_vv
=int_dig_vv
;
1382 spin_unlock_irq(&mdev
->req_lock
);
1384 retcode
= _drbd_request_state(mdev
, NS(conn
, C_UNCONNECTED
), CS_VERBOSE
);
1386 kobject_uevent(&disk_to_dev(mdev
->vdisk
)->kobj
, KOBJ_CHANGE
);
1387 reply
->ret_code
= retcode
;
1388 drbd_reconfig_done(mdev
);
1395 crypto_free_hash(tfm
);
1396 crypto_free_hash(integrity_w_tfm
);
1397 crypto_free_hash(integrity_r_tfm
);
1402 reply
->ret_code
= retcode
;
1403 drbd_reconfig_done(mdev
);
1407 static int drbd_nl_disconnect(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1408 struct drbd_nl_cfg_reply
*reply
)
1412 retcode
= _drbd_request_state(mdev
, NS(conn
, C_DISCONNECTING
), CS_ORDERED
);
1414 if (retcode
== SS_NOTHING_TO_DO
)
1416 else if (retcode
== SS_ALREADY_STANDALONE
)
1418 else if (retcode
== SS_PRIMARY_NOP
) {
1419 /* Our statche checking code wants to see the peer outdated. */
1420 retcode
= drbd_request_state(mdev
, NS2(conn
, C_DISCONNECTING
,
1422 } else if (retcode
== SS_CW_FAILED_BY_PEER
) {
1423 /* The peer probably wants to see us outdated. */
1424 retcode
= _drbd_request_state(mdev
, NS2(conn
, C_DISCONNECTING
,
1427 if (retcode
== SS_IS_DISKLESS
|| retcode
== SS_LOWER_THAN_OUTDATED
) {
1428 drbd_force_state(mdev
, NS(conn
, C_DISCONNECTING
));
1429 retcode
= SS_SUCCESS
;
1433 if (retcode
< SS_SUCCESS
)
1436 if (wait_event_interruptible(mdev
->state_wait
,
1437 mdev
->state
.conn
!= C_DISCONNECTING
)) {
1438 /* Do not test for mdev->state.conn == C_STANDALONE, since
1439 someone else might connect us in the mean time! */
1448 reply
->ret_code
= retcode
;
1452 void resync_after_online_grow(struct drbd_conf
*mdev
)
1454 int iass
; /* I am sync source */
1456 dev_info(DEV
, "Resync of new storage after online grow\n");
1457 if (mdev
->state
.role
!= mdev
->state
.peer
)
1458 iass
= (mdev
->state
.role
== R_PRIMARY
);
1460 iass
= test_bit(DISCARD_CONCURRENT
, &mdev
->flags
);
1463 drbd_start_resync(mdev
, C_SYNC_SOURCE
);
1465 _drbd_request_state(mdev
, NS(conn
, C_WF_SYNC_UUID
), CS_VERBOSE
+ CS_SERIALIZE
);
1468 static int drbd_nl_resize(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1469 struct drbd_nl_cfg_reply
*reply
)
1472 int retcode
= NO_ERROR
;
1473 int ldsc
= 0; /* local disk size changed */
1474 enum determine_dev_size dd
;
1476 memset(&rs
, 0, sizeof(struct resize
));
1477 if (!resize_from_tags(mdev
, nlp
->tag_list
, &rs
)) {
1478 retcode
= ERR_MANDATORY_TAG
;
1482 if (mdev
->state
.conn
> C_CONNECTED
) {
1483 retcode
= ERR_RESIZE_RESYNC
;
1487 if (mdev
->state
.role
== R_SECONDARY
&&
1488 mdev
->state
.peer
== R_SECONDARY
) {
1489 retcode
= ERR_NO_PRIMARY
;
1493 if (!get_ldev(mdev
)) {
1494 retcode
= ERR_NO_DISK
;
1498 if (mdev
->ldev
->known_size
!= drbd_get_capacity(mdev
->ldev
->backing_bdev
)) {
1499 mdev
->ldev
->known_size
= drbd_get_capacity(mdev
->ldev
->backing_bdev
);
1503 mdev
->ldev
->dc
.disk_size
= (sector_t
)rs
.resize_size
;
1504 dd
= drbd_determin_dev_size(mdev
);
1507 if (dd
== dev_size_error
) {
1508 retcode
= ERR_NOMEM_BITMAP
;
1512 if (mdev
->state
.conn
== C_CONNECTED
&& (dd
!= unchanged
|| ldsc
)) {
1514 set_bit(RESIZE_PENDING
, &mdev
->flags
);
1516 drbd_send_uuids(mdev
);
1517 drbd_send_sizes(mdev
, 1);
1521 reply
->ret_code
= retcode
;
1525 static int drbd_nl_syncer_conf(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1526 struct drbd_nl_cfg_reply
*reply
)
1528 int retcode
= NO_ERROR
;
1530 int ovr
; /* online verify running */
1531 int rsr
; /* re-sync running */
1532 struct crypto_hash
*verify_tfm
= NULL
;
1533 struct crypto_hash
*csums_tfm
= NULL
;
1534 struct syncer_conf sc
;
1535 cpumask_var_t new_cpu_mask
;
1537 if (!zalloc_cpumask_var(&new_cpu_mask
, GFP_KERNEL
)) {
1538 retcode
= ERR_NOMEM
;
1542 if (nlp
->flags
& DRBD_NL_SET_DEFAULTS
) {
1543 memset(&sc
, 0, sizeof(struct syncer_conf
));
1544 sc
.rate
= DRBD_RATE_DEF
;
1545 sc
.after
= DRBD_AFTER_DEF
;
1546 sc
.al_extents
= DRBD_AL_EXTENTS_DEF
;
1548 memcpy(&sc
, &mdev
->sync_conf
, sizeof(struct syncer_conf
));
1550 if (!syncer_conf_from_tags(mdev
, nlp
->tag_list
, &sc
)) {
1551 retcode
= ERR_MANDATORY_TAG
;
1555 /* re-sync running */
1556 rsr
= ( mdev
->state
.conn
== C_SYNC_SOURCE
||
1557 mdev
->state
.conn
== C_SYNC_TARGET
||
1558 mdev
->state
.conn
== C_PAUSED_SYNC_S
||
1559 mdev
->state
.conn
== C_PAUSED_SYNC_T
);
1561 if (rsr
&& strcmp(sc
.csums_alg
, mdev
->sync_conf
.csums_alg
)) {
1562 retcode
= ERR_CSUMS_RESYNC_RUNNING
;
1566 if (!rsr
&& sc
.csums_alg
[0]) {
1567 csums_tfm
= crypto_alloc_hash(sc
.csums_alg
, 0, CRYPTO_ALG_ASYNC
);
1568 if (IS_ERR(csums_tfm
)) {
1570 retcode
= ERR_CSUMS_ALG
;
1574 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm
))) {
1575 retcode
= ERR_CSUMS_ALG_ND
;
1580 /* online verify running */
1581 ovr
= (mdev
->state
.conn
== C_VERIFY_S
|| mdev
->state
.conn
== C_VERIFY_T
);
1584 if (strcmp(sc
.verify_alg
, mdev
->sync_conf
.verify_alg
)) {
1585 retcode
= ERR_VERIFY_RUNNING
;
1590 if (!ovr
&& sc
.verify_alg
[0]) {
1591 verify_tfm
= crypto_alloc_hash(sc
.verify_alg
, 0, CRYPTO_ALG_ASYNC
);
1592 if (IS_ERR(verify_tfm
)) {
1594 retcode
= ERR_VERIFY_ALG
;
1598 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm
))) {
1599 retcode
= ERR_VERIFY_ALG_ND
;
1604 /* silently ignore cpu mask on UP kernel */
1605 if (nr_cpu_ids
> 1 && sc
.cpu_mask
[0] != 0) {
1606 err
= __bitmap_parse(sc
.cpu_mask
, 32, 0,
1607 cpumask_bits(new_cpu_mask
), nr_cpu_ids
);
1609 dev_warn(DEV
, "__bitmap_parse() failed with %d\n", err
);
1610 retcode
= ERR_CPU_MASK_PARSE
;
1615 ERR_IF (sc
.rate
< 1) sc
.rate
= 1;
1616 ERR_IF (sc
.al_extents
< 7) sc
.al_extents
= 127; /* arbitrary minimum */
1617 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1618 if (sc
.al_extents
> AL_MAX
) {
1619 dev_err(DEV
, "sc.al_extents > %d\n", AL_MAX
);
1620 sc
.al_extents
= AL_MAX
;
1624 /* most sanity checks done, try to assign the new sync-after
1625 * dependency. need to hold the global lock in there,
1626 * to avoid a race in the dependency loop check. */
1627 retcode
= drbd_alter_sa(mdev
, sc
.after
);
1628 if (retcode
!= NO_ERROR
)
1631 /* ok, assign the rest of it as well.
1632 * lock against receive_SyncParam() */
1633 spin_lock(&mdev
->peer_seq_lock
);
1634 mdev
->sync_conf
= sc
;
1637 crypto_free_hash(mdev
->csums_tfm
);
1638 mdev
->csums_tfm
= csums_tfm
;
1643 crypto_free_hash(mdev
->verify_tfm
);
1644 mdev
->verify_tfm
= verify_tfm
;
1647 spin_unlock(&mdev
->peer_seq_lock
);
1649 if (get_ldev(mdev
)) {
1650 wait_event(mdev
->al_wait
, lc_try_lock(mdev
->act_log
));
1651 drbd_al_shrink(mdev
);
1652 err
= drbd_check_al_size(mdev
);
1653 lc_unlock(mdev
->act_log
);
1654 wake_up(&mdev
->al_wait
);
1660 retcode
= ERR_NOMEM
;
1665 if (mdev
->state
.conn
>= C_CONNECTED
)
1666 drbd_send_sync_param(mdev
, &sc
);
1668 if (!cpumask_equal(mdev
->cpu_mask
, new_cpu_mask
)) {
1669 cpumask_copy(mdev
->cpu_mask
, new_cpu_mask
);
1670 drbd_calc_cpu_mask(mdev
);
1671 mdev
->receiver
.reset_cpu_mask
= 1;
1672 mdev
->asender
.reset_cpu_mask
= 1;
1673 mdev
->worker
.reset_cpu_mask
= 1;
1676 kobject_uevent(&disk_to_dev(mdev
->vdisk
)->kobj
, KOBJ_CHANGE
);
1678 free_cpumask_var(new_cpu_mask
);
1679 crypto_free_hash(csums_tfm
);
1680 crypto_free_hash(verify_tfm
);
1681 reply
->ret_code
= retcode
;
1685 static int drbd_nl_invalidate(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1686 struct drbd_nl_cfg_reply
*reply
)
1690 retcode
= _drbd_request_state(mdev
, NS(conn
, C_STARTING_SYNC_T
), CS_ORDERED
);
1692 if (retcode
< SS_SUCCESS
&& retcode
!= SS_NEED_CONNECTION
)
1693 retcode
= drbd_request_state(mdev
, NS(conn
, C_STARTING_SYNC_T
));
1695 while (retcode
== SS_NEED_CONNECTION
) {
1696 spin_lock_irq(&mdev
->req_lock
);
1697 if (mdev
->state
.conn
< C_CONNECTED
)
1698 retcode
= _drbd_set_state(_NS(mdev
, disk
, D_INCONSISTENT
), CS_VERBOSE
, NULL
);
1699 spin_unlock_irq(&mdev
->req_lock
);
1701 if (retcode
!= SS_NEED_CONNECTION
)
1704 retcode
= drbd_request_state(mdev
, NS(conn
, C_STARTING_SYNC_T
));
1707 reply
->ret_code
= retcode
;
1711 static int drbd_nl_invalidate_peer(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1712 struct drbd_nl_cfg_reply
*reply
)
1715 reply
->ret_code
= drbd_request_state(mdev
, NS(conn
, C_STARTING_SYNC_S
));
1720 static int drbd_nl_pause_sync(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1721 struct drbd_nl_cfg_reply
*reply
)
1723 int retcode
= NO_ERROR
;
1725 if (drbd_request_state(mdev
, NS(user_isp
, 1)) == SS_NOTHING_TO_DO
)
1726 retcode
= ERR_PAUSE_IS_SET
;
1728 reply
->ret_code
= retcode
;
1732 static int drbd_nl_resume_sync(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1733 struct drbd_nl_cfg_reply
*reply
)
1735 int retcode
= NO_ERROR
;
1737 if (drbd_request_state(mdev
, NS(user_isp
, 0)) == SS_NOTHING_TO_DO
)
1738 retcode
= ERR_PAUSE_IS_CLEAR
;
1740 reply
->ret_code
= retcode
;
1744 static int drbd_nl_suspend_io(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1745 struct drbd_nl_cfg_reply
*reply
)
1747 reply
->ret_code
= drbd_request_state(mdev
, NS(susp
, 1));
1752 static int drbd_nl_resume_io(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1753 struct drbd_nl_cfg_reply
*reply
)
1755 reply
->ret_code
= drbd_request_state(mdev
, NS(susp
, 0));
1759 static int drbd_nl_outdate(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1760 struct drbd_nl_cfg_reply
*reply
)
1762 reply
->ret_code
= drbd_request_state(mdev
, NS(disk
, D_OUTDATED
));
1766 static int drbd_nl_get_config(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1767 struct drbd_nl_cfg_reply
*reply
)
1771 tl
= reply
->tag_list
;
1773 if (get_ldev(mdev
)) {
1774 tl
= disk_conf_to_tags(mdev
, &mdev
->ldev
->dc
, tl
);
1778 if (get_net_conf(mdev
)) {
1779 tl
= net_conf_to_tags(mdev
, mdev
->net_conf
, tl
);
1782 tl
= syncer_conf_to_tags(mdev
, &mdev
->sync_conf
, tl
);
1784 put_unaligned(TT_END
, tl
++); /* Close the tag list */
1786 return (int)((char *)tl
- (char *)reply
->tag_list
);
1789 static int drbd_nl_get_state(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1790 struct drbd_nl_cfg_reply
*reply
)
1792 unsigned short *tl
= reply
->tag_list
;
1793 union drbd_state s
= mdev
->state
;
1794 unsigned long rs_left
;
1797 tl
= get_state_to_tags(mdev
, (struct get_state
*)&s
, tl
);
1799 /* no local ref, no bitmap, no syncer progress. */
1800 if (s
.conn
>= C_SYNC_SOURCE
&& s
.conn
<= C_PAUSED_SYNC_T
) {
1801 if (get_ldev(mdev
)) {
1802 drbd_get_syncer_progress(mdev
, &rs_left
, &res
);
1803 tl
= tl_add_int(tl
, T_sync_progress
, &res
);
1807 put_unaligned(TT_END
, tl
++); /* Close the tag list */
1809 return (int)((char *)tl
- (char *)reply
->tag_list
);
1812 static int drbd_nl_get_uuids(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1813 struct drbd_nl_cfg_reply
*reply
)
1817 tl
= reply
->tag_list
;
1819 if (get_ldev(mdev
)) {
1820 tl
= tl_add_blob(tl
, T_uuids
, mdev
->ldev
->md
.uuid
, UI_SIZE
*sizeof(u64
));
1821 tl
= tl_add_int(tl
, T_uuids_flags
, &mdev
->ldev
->md
.flags
);
1824 put_unaligned(TT_END
, tl
++); /* Close the tag list */
1826 return (int)((char *)tl
- (char *)reply
->tag_list
);
1830 * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
1831 * @mdev: DRBD device.
1832 * @nlp: Netlink/connector packet from drbdsetup
1833 * @reply: Reply packet for drbdsetup
1835 static int drbd_nl_get_timeout_flag(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1836 struct drbd_nl_cfg_reply
*reply
)
1841 tl
= reply
->tag_list
;
1843 rv
= mdev
->state
.pdsk
== D_OUTDATED
? UT_PEER_OUTDATED
:
1844 test_bit(USE_DEGR_WFC_T
, &mdev
->flags
) ? UT_DEGRADED
: UT_DEFAULT
;
1846 tl
= tl_add_blob(tl
, T_use_degraded
, &rv
, sizeof(rv
));
1847 put_unaligned(TT_END
, tl
++); /* Close the tag list */
1849 return (int)((char *)tl
- (char *)reply
->tag_list
);
1852 static int drbd_nl_start_ov(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1853 struct drbd_nl_cfg_reply
*reply
)
1855 /* default to resume from last known position, if possible */
1856 struct start_ov args
=
1857 { .start_sector
= mdev
->ov_start_sector
};
1859 if (!start_ov_from_tags(mdev
, nlp
->tag_list
, &args
)) {
1860 reply
->ret_code
= ERR_MANDATORY_TAG
;
1863 /* w_make_ov_request expects position to be aligned */
1864 mdev
->ov_start_sector
= args
.start_sector
& ~BM_SECT_PER_BIT
;
1865 reply
->ret_code
= drbd_request_state(mdev
,NS(conn
,C_VERIFY_S
));
1870 static int drbd_nl_new_c_uuid(struct drbd_conf
*mdev
, struct drbd_nl_cfg_req
*nlp
,
1871 struct drbd_nl_cfg_reply
*reply
)
1873 int retcode
= NO_ERROR
;
1874 int skip_initial_sync
= 0;
1877 struct new_c_uuid args
;
1879 memset(&args
, 0, sizeof(struct new_c_uuid
));
1880 if (!new_c_uuid_from_tags(mdev
, nlp
->tag_list
, &args
)) {
1881 reply
->ret_code
= ERR_MANDATORY_TAG
;
1885 mutex_lock(&mdev
->state_mutex
); /* Protects us against serialized state changes. */
1887 if (!get_ldev(mdev
)) {
1888 retcode
= ERR_NO_DISK
;
1892 /* this is "skip initial sync", assume to be clean */
1893 if (mdev
->state
.conn
== C_CONNECTED
&& mdev
->agreed_pro_version
>= 90 &&
1894 mdev
->ldev
->md
.uuid
[UI_CURRENT
] == UUID_JUST_CREATED
&& args
.clear_bm
) {
1895 dev_info(DEV
, "Preparing to skip initial sync\n");
1896 skip_initial_sync
= 1;
1897 } else if (mdev
->state
.conn
!= C_STANDALONE
) {
1898 retcode
= ERR_CONNECTED
;
1902 drbd_uuid_set(mdev
, UI_BITMAP
, 0); /* Rotate UI_BITMAP to History 1, etc... */
1903 drbd_uuid_new_current(mdev
); /* New current, previous to UI_BITMAP */
1905 if (args
.clear_bm
) {
1906 err
= drbd_bitmap_io(mdev
, &drbd_bmio_clear_n_write
, "clear_n_write from new_c_uuid");
1908 dev_err(DEV
, "Writing bitmap failed with %d\n",err
);
1909 retcode
= ERR_IO_MD_DISK
;
1911 if (skip_initial_sync
) {
1912 drbd_send_uuids_skip_initial_sync(mdev
);
1913 _drbd_uuid_set(mdev
, UI_BITMAP
, 0);
1914 spin_lock_irq(&mdev
->req_lock
);
1915 _drbd_set_state(_NS2(mdev
, disk
, D_UP_TO_DATE
, pdsk
, D_UP_TO_DATE
),
1917 spin_unlock_irq(&mdev
->req_lock
);
1925 mutex_unlock(&mdev
->state_mutex
);
1927 reply
->ret_code
= retcode
;
1931 static struct drbd_conf
*ensure_mdev(struct drbd_nl_cfg_req
*nlp
)
1933 struct drbd_conf
*mdev
;
1935 if (nlp
->drbd_minor
>= minor_count
)
1938 mdev
= minor_to_mdev(nlp
->drbd_minor
);
1940 if (!mdev
&& (nlp
->flags
& DRBD_NL_CREATE_DEVICE
)) {
1941 struct gendisk
*disk
= NULL
;
1942 mdev
= drbd_new_device(nlp
->drbd_minor
);
1944 spin_lock_irq(&drbd_pp_lock
);
1945 if (minor_table
[nlp
->drbd_minor
] == NULL
) {
1946 minor_table
[nlp
->drbd_minor
] = mdev
;
1949 } /* else: we lost the race */
1950 spin_unlock_irq(&drbd_pp_lock
);
1952 if (disk
) /* we won the race above */
1953 /* in case we ever add a drbd_delete_device(),
1954 * don't forget the del_gendisk! */
1956 else /* we lost the race above */
1957 drbd_free_mdev(mdev
);
1959 mdev
= minor_to_mdev(nlp
->drbd_minor
);
1965 struct cn_handler_struct
{
1966 int (*function
)(struct drbd_conf
*,
1967 struct drbd_nl_cfg_req
*,
1968 struct drbd_nl_cfg_reply
*);
1969 int reply_body_size
;
1972 static struct cn_handler_struct cnd_table
[] = {
1973 [ P_primary
] = { &drbd_nl_primary
, 0 },
1974 [ P_secondary
] = { &drbd_nl_secondary
, 0 },
1975 [ P_disk_conf
] = { &drbd_nl_disk_conf
, 0 },
1976 [ P_detach
] = { &drbd_nl_detach
, 0 },
1977 [ P_net_conf
] = { &drbd_nl_net_conf
, 0 },
1978 [ P_disconnect
] = { &drbd_nl_disconnect
, 0 },
1979 [ P_resize
] = { &drbd_nl_resize
, 0 },
1980 [ P_syncer_conf
] = { &drbd_nl_syncer_conf
, 0 },
1981 [ P_invalidate
] = { &drbd_nl_invalidate
, 0 },
1982 [ P_invalidate_peer
] = { &drbd_nl_invalidate_peer
, 0 },
1983 [ P_pause_sync
] = { &drbd_nl_pause_sync
, 0 },
1984 [ P_resume_sync
] = { &drbd_nl_resume_sync
, 0 },
1985 [ P_suspend_io
] = { &drbd_nl_suspend_io
, 0 },
1986 [ P_resume_io
] = { &drbd_nl_resume_io
, 0 },
1987 [ P_outdate
] = { &drbd_nl_outdate
, 0 },
1988 [ P_get_config
] = { &drbd_nl_get_config
,
1989 sizeof(struct syncer_conf_tag_len_struct
) +
1990 sizeof(struct disk_conf_tag_len_struct
) +
1991 sizeof(struct net_conf_tag_len_struct
) },
1992 [ P_get_state
] = { &drbd_nl_get_state
,
1993 sizeof(struct get_state_tag_len_struct
) +
1994 sizeof(struct sync_progress_tag_len_struct
) },
1995 [ P_get_uuids
] = { &drbd_nl_get_uuids
,
1996 sizeof(struct get_uuids_tag_len_struct
) },
1997 [ P_get_timeout_flag
] = { &drbd_nl_get_timeout_flag
,
1998 sizeof(struct get_timeout_flag_tag_len_struct
)},
1999 [ P_start_ov
] = { &drbd_nl_start_ov
, 0 },
2000 [ P_new_c_uuid
] = { &drbd_nl_new_c_uuid
, 0 },
2003 static void drbd_connector_callback(struct cn_msg
*req
)
2005 struct drbd_nl_cfg_req
*nlp
= (struct drbd_nl_cfg_req
*)req
->data
;
2006 struct cn_handler_struct
*cm
;
2007 struct cn_msg
*cn_reply
;
2008 struct drbd_nl_cfg_reply
*reply
;
2009 struct drbd_conf
*mdev
;
2011 int reply_size
= sizeof(struct cn_msg
)
2012 + sizeof(struct drbd_nl_cfg_reply
)
2013 + sizeof(short int);
2015 if (!try_module_get(THIS_MODULE
)) {
2016 printk(KERN_ERR
"drbd: try_module_get() failed!\n");
2020 mdev
= ensure_mdev(nlp
);
2022 retcode
= ERR_MINOR_INVALID
;
2026 if (nlp
->packet_type
>= P_nl_after_last_packet
) {
2027 retcode
= ERR_PACKET_NR
;
2031 cm
= cnd_table
+ nlp
->packet_type
;
2033 /* This may happen if packet number is 0: */
2034 if (cm
->function
== NULL
) {
2035 retcode
= ERR_PACKET_NR
;
2039 reply_size
+= cm
->reply_body_size
;
2041 /* allocation not in the IO path, cqueue thread context */
2042 cn_reply
= kmalloc(reply_size
, GFP_KERNEL
);
2044 retcode
= ERR_NOMEM
;
2047 reply
= (struct drbd_nl_cfg_reply
*) cn_reply
->data
;
2049 reply
->packet_type
=
2050 cm
->reply_body_size
? nlp
->packet_type
: P_nl_after_last_packet
;
2051 reply
->minor
= nlp
->drbd_minor
;
2052 reply
->ret_code
= NO_ERROR
; /* Might by modified by cm->function. */
2053 /* reply->tag_list; might be modified by cm->function. */
2055 rr
= cm
->function(mdev
, nlp
, reply
);
2057 cn_reply
->id
= req
->id
;
2058 cn_reply
->seq
= req
->seq
;
2059 cn_reply
->ack
= req
->ack
+ 1;
2060 cn_reply
->len
= sizeof(struct drbd_nl_cfg_reply
) + rr
;
2061 cn_reply
->flags
= 0;
2063 rr
= cn_netlink_send(cn_reply
, CN_IDX_DRBD
, GFP_KERNEL
);
2064 if (rr
&& rr
!= -ESRCH
)
2065 printk(KERN_INFO
"drbd: cn_netlink_send()=%d\n", rr
);
2068 module_put(THIS_MODULE
);
2071 drbd_nl_send_reply(req
, retcode
);
2072 module_put(THIS_MODULE
);
2075 static atomic_t drbd_nl_seq
= ATOMIC_INIT(2); /* two. */
2077 static unsigned short *
2078 __tl_add_blob(unsigned short *tl
, enum drbd_tags tag
, const void *data
,
2079 unsigned short len
, int nul_terminated
)
2081 unsigned short l
= tag_descriptions
[tag_number(tag
)].max_len
;
2082 len
= (len
< l
) ? len
: l
;
2083 put_unaligned(tag
, tl
++);
2084 put_unaligned(len
, tl
++);
2085 memcpy(tl
, data
, len
);
2086 tl
= (unsigned short*)((char*)tl
+ len
);
2088 *((char*)tl
- 1) = 0;
2092 static unsigned short *
2093 tl_add_blob(unsigned short *tl
, enum drbd_tags tag
, const void *data
, int len
)
2095 return __tl_add_blob(tl
, tag
, data
, len
, 0);
2098 static unsigned short *
2099 tl_add_str(unsigned short *tl
, enum drbd_tags tag
, const char *str
)
2101 return __tl_add_blob(tl
, tag
, str
, strlen(str
)+1, 0);
2104 static unsigned short *
2105 tl_add_int(unsigned short *tl
, enum drbd_tags tag
, const void *val
)
2107 put_unaligned(tag
, tl
++);
2108 switch(tag_type(tag
)) {
2110 put_unaligned(sizeof(int), tl
++);
2111 put_unaligned(*(int *)val
, (int *)tl
);
2112 tl
= (unsigned short*)((char*)tl
+sizeof(int));
2115 put_unaligned(sizeof(u64
), tl
++);
2116 put_unaligned(*(u64
*)val
, (u64
*)tl
);
2117 tl
= (unsigned short*)((char*)tl
+sizeof(u64
));
2120 /* someone did something stupid. */
2126 void drbd_bcast_state(struct drbd_conf
*mdev
, union drbd_state state
)
2128 char buffer
[sizeof(struct cn_msg
)+
2129 sizeof(struct drbd_nl_cfg_reply
)+
2130 sizeof(struct get_state_tag_len_struct
)+
2132 struct cn_msg
*cn_reply
= (struct cn_msg
*) buffer
;
2133 struct drbd_nl_cfg_reply
*reply
=
2134 (struct drbd_nl_cfg_reply
*)cn_reply
->data
;
2135 unsigned short *tl
= reply
->tag_list
;
2137 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2139 tl
= get_state_to_tags(mdev
, (struct get_state
*)&state
, tl
);
2141 put_unaligned(TT_END
, tl
++); /* Close the tag list */
2143 cn_reply
->id
.idx
= CN_IDX_DRBD
;
2144 cn_reply
->id
.val
= CN_VAL_DRBD
;
2146 cn_reply
->seq
= atomic_add_return(1, &drbd_nl_seq
);
2147 cn_reply
->ack
= 0; /* not used here. */
2148 cn_reply
->len
= sizeof(struct drbd_nl_cfg_reply
) +
2149 (int)((char *)tl
- (char *)reply
->tag_list
);
2150 cn_reply
->flags
= 0;
2152 reply
->packet_type
= P_get_state
;
2153 reply
->minor
= mdev_to_minor(mdev
);
2154 reply
->ret_code
= NO_ERROR
;
2156 cn_netlink_send(cn_reply
, CN_IDX_DRBD
, GFP_NOIO
);
2159 void drbd_bcast_ev_helper(struct drbd_conf
*mdev
, char *helper_name
)
2161 char buffer
[sizeof(struct cn_msg
)+
2162 sizeof(struct drbd_nl_cfg_reply
)+
2163 sizeof(struct call_helper_tag_len_struct
)+
2165 struct cn_msg
*cn_reply
= (struct cn_msg
*) buffer
;
2166 struct drbd_nl_cfg_reply
*reply
=
2167 (struct drbd_nl_cfg_reply
*)cn_reply
->data
;
2168 unsigned short *tl
= reply
->tag_list
;
2170 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2172 tl
= tl_add_str(tl
, T_helper
, helper_name
);
2173 put_unaligned(TT_END
, tl
++); /* Close the tag list */
2175 cn_reply
->id
.idx
= CN_IDX_DRBD
;
2176 cn_reply
->id
.val
= CN_VAL_DRBD
;
2178 cn_reply
->seq
= atomic_add_return(1, &drbd_nl_seq
);
2179 cn_reply
->ack
= 0; /* not used here. */
2180 cn_reply
->len
= sizeof(struct drbd_nl_cfg_reply
) +
2181 (int)((char *)tl
- (char *)reply
->tag_list
);
2182 cn_reply
->flags
= 0;
2184 reply
->packet_type
= P_call_helper
;
2185 reply
->minor
= mdev_to_minor(mdev
);
2186 reply
->ret_code
= NO_ERROR
;
2188 cn_netlink_send(cn_reply
, CN_IDX_DRBD
, GFP_NOIO
);
2191 void drbd_bcast_ee(struct drbd_conf
*mdev
,
2192 const char *reason
, const int dgs
,
2193 const char* seen_hash
, const char* calc_hash
,
2194 const struct drbd_epoch_entry
* e
)
2196 struct cn_msg
*cn_reply
;
2197 struct drbd_nl_cfg_reply
*reply
;
2198 struct bio_vec
*bvec
;
2204 if (!reason
|| !reason
[0])
2207 /* apparently we have to memcpy twice, first to prepare the data for the
2208 * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2210 /* receiver thread context, which is not in the writeout path (of this node),
2211 * but may be in the writeout path of the _other_ node.
2212 * GFP_NOIO to avoid potential "distributed deadlock". */
2214 sizeof(struct cn_msg
)+
2215 sizeof(struct drbd_nl_cfg_reply
)+
2216 sizeof(struct dump_ee_tag_len_struct
)+
2221 dev_err(DEV
, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2222 (unsigned long long)e
->sector
, e
->size
);
2226 reply
= (struct drbd_nl_cfg_reply
*)cn_reply
->data
;
2227 tl
= reply
->tag_list
;
2229 tl
= tl_add_str(tl
, T_dump_ee_reason
, reason
);
2230 tl
= tl_add_blob(tl
, T_seen_digest
, seen_hash
, dgs
);
2231 tl
= tl_add_blob(tl
, T_calc_digest
, calc_hash
, dgs
);
2232 tl
= tl_add_int(tl
, T_ee_sector
, &e
->sector
);
2233 tl
= tl_add_int(tl
, T_ee_block_id
, &e
->block_id
);
2235 put_unaligned(T_ee_data
, tl
++);
2236 put_unaligned(e
->size
, tl
++);
2238 __bio_for_each_segment(bvec
, e
->private_bio
, i
, 0) {
2239 void *d
= kmap(bvec
->bv_page
);
2240 memcpy(tl
, d
+ bvec
->bv_offset
, bvec
->bv_len
);
2241 kunmap(bvec
->bv_page
);
2242 tl
=(unsigned short*)((char*)tl
+ bvec
->bv_len
);
2244 put_unaligned(TT_END
, tl
++); /* Close the tag list */
2246 cn_reply
->id
.idx
= CN_IDX_DRBD
;
2247 cn_reply
->id
.val
= CN_VAL_DRBD
;
2249 cn_reply
->seq
= atomic_add_return(1,&drbd_nl_seq
);
2250 cn_reply
->ack
= 0; // not used here.
2251 cn_reply
->len
= sizeof(struct drbd_nl_cfg_reply
) +
2252 (int)((char*)tl
- (char*)reply
->tag_list
);
2253 cn_reply
->flags
= 0;
2255 reply
->packet_type
= P_dump_ee
;
2256 reply
->minor
= mdev_to_minor(mdev
);
2257 reply
->ret_code
= NO_ERROR
;
2259 cn_netlink_send(cn_reply
, CN_IDX_DRBD
, GFP_NOIO
);
2263 void drbd_bcast_sync_progress(struct drbd_conf
*mdev
)
2265 char buffer
[sizeof(struct cn_msg
)+
2266 sizeof(struct drbd_nl_cfg_reply
)+
2267 sizeof(struct sync_progress_tag_len_struct
)+
2269 struct cn_msg
*cn_reply
= (struct cn_msg
*) buffer
;
2270 struct drbd_nl_cfg_reply
*reply
=
2271 (struct drbd_nl_cfg_reply
*)cn_reply
->data
;
2272 unsigned short *tl
= reply
->tag_list
;
2273 unsigned long rs_left
;
2276 /* no local ref, no bitmap, no syncer progress, no broadcast. */
2277 if (!get_ldev(mdev
))
2279 drbd_get_syncer_progress(mdev
, &rs_left
, &res
);
2282 tl
= tl_add_int(tl
, T_sync_progress
, &res
);
2283 put_unaligned(TT_END
, tl
++); /* Close the tag list */
2285 cn_reply
->id
.idx
= CN_IDX_DRBD
;
2286 cn_reply
->id
.val
= CN_VAL_DRBD
;
2288 cn_reply
->seq
= atomic_add_return(1, &drbd_nl_seq
);
2289 cn_reply
->ack
= 0; /* not used here. */
2290 cn_reply
->len
= sizeof(struct drbd_nl_cfg_reply
) +
2291 (int)((char *)tl
- (char *)reply
->tag_list
);
2292 cn_reply
->flags
= 0;
2294 reply
->packet_type
= P_sync_progress
;
2295 reply
->minor
= mdev_to_minor(mdev
);
2296 reply
->ret_code
= NO_ERROR
;
2298 cn_netlink_send(cn_reply
, CN_IDX_DRBD
, GFP_NOIO
);
2301 int __init
drbd_nl_init(void)
2303 static struct cb_id cn_id_drbd
;
2306 cn_id_drbd
.val
= CN_VAL_DRBD
;
2308 cn_id_drbd
.idx
= cn_idx
;
2309 err
= cn_add_callback(&cn_id_drbd
, "cn_drbd", &drbd_connector_callback
);
2312 cn_idx
= (cn_idx
+ CN_IDX_STEP
);
2316 printk(KERN_ERR
"drbd: cn_drbd failed to register\n");
2323 void drbd_nl_cleanup(void)
2325 static struct cb_id cn_id_drbd
;
2327 cn_id_drbd
.idx
= cn_idx
;
2328 cn_id_drbd
.val
= CN_VAL_DRBD
;
2330 cn_del_callback(&cn_id_drbd
);
2333 void drbd_nl_send_reply(struct cn_msg
*req
, int ret_code
)
2335 char buffer
[sizeof(struct cn_msg
)+sizeof(struct drbd_nl_cfg_reply
)];
2336 struct cn_msg
*cn_reply
= (struct cn_msg
*) buffer
;
2337 struct drbd_nl_cfg_reply
*reply
=
2338 (struct drbd_nl_cfg_reply
*)cn_reply
->data
;
2341 cn_reply
->id
= req
->id
;
2343 cn_reply
->seq
= req
->seq
;
2344 cn_reply
->ack
= req
->ack
+ 1;
2345 cn_reply
->len
= sizeof(struct drbd_nl_cfg_reply
);
2346 cn_reply
->flags
= 0;
2348 reply
->minor
= ((struct drbd_nl_cfg_req
*)req
->data
)->drbd_minor
;
2349 reply
->ret_code
= ret_code
;
2351 rr
= cn_netlink_send(cn_reply
, CN_IDX_DRBD
, GFP_NOIO
);
2352 if (rr
&& rr
!= -ESRCH
)
2353 printk(KERN_INFO
"drbd: cn_netlink_send()=%d\n", rr
);