Linux 2.6.34-rc3
[pohmelfs.git] / drivers / block / drbd / drbd_nl.c
blob4df3b40b1057d225fe310601c77021a6f37f0d7c
1 /*
2 drbd_nl.c
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)
13 any later version.
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>
28 #include <linux/in.h>
29 #include <linux/fs.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>
35 #include "drbd_int.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) \
54 { \
55 int tag; \
56 int dlen; \
58 while ((tag = get_unaligned(tags++)) != TT_END) { \
59 dlen = get_unaligned(tags++); \
60 switch (tag_number(tag)) { \
61 fields \
62 default: \
63 if (tag & T_MANDATORY) { \
64 dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \
65 return 0; \
66 } \
67 } \
68 tags = (unsigned short *)((char *)tags + dlen); \
69 } \
70 return 1; \
72 #define NL_INTEGER(pn, pr, member) \
73 case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \
74 arg->member = get_unaligned((int *)(tags)); \
75 break;
76 #define NL_INT64(pn, pr, member) \
77 case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \
78 arg->member = get_unaligned((u64 *)(tags)); \
79 break;
80 #define NL_BIT(pn, pr, member) \
81 case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \
82 arg->member = *(char *)(tags) ? 1 : 0; \
83 break;
84 #define NL_STRING(pn, pr, member, len) \
85 case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \
86 if (dlen > len) { \
87 dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \
88 #member, dlen, (unsigned int)len); \
89 return 0; \
90 } \
91 arg->member ## _len = dlen; \
92 memcpy(arg->member, tags, min_t(size_t, dlen, len)); \
93 break;
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) \
105 fields \
106 return 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=/",
137 "TERM=linux",
138 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
139 NULL, /* Will be set to address family */
140 NULL, /* Will be set to address */
141 NULL };
143 char mb[12], af[20], ad[60], *afs;
144 char *argv[] = {usermode_helper, cmd, mb, NULL };
145 int ret;
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) {
151 case AF_INET6:
152 afs = "ipv6";
153 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6",
154 &((struct sockaddr_in6 *)mdev->net_conf->peer_addr)->sin6_addr);
155 break;
156 case AF_INET:
157 afs = "ipv4";
158 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
159 &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
160 break;
161 default:
162 afs = "ssocks";
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);
167 envp[3]=af;
168 envp[4]=ad;
169 put_net_conf(mdev);
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);
176 if (ret)
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);
180 else
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. */
186 ret = 0;
188 return ret;
191 enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
193 char *ex_to_string;
194 int r;
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;
202 put_ldev(mdev);
203 } else {
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;
217 break;
218 case 4: /* peer got outdated, or was already outdated */
219 ex_to_string = "peer was fenced";
220 nps = D_OUTDATED;
221 break;
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";
226 nps = D_OUTDATED;
227 } else {
228 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
229 nps = mdev->state.pdsk;
231 break;
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");
237 nps = D_UNKNOWN;
238 _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE);
239 break;
240 case 7:
241 if (fp != FP_STONITH)
242 dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n");
243 ex_to_string = "peer was stonithed";
244 nps = D_OUTDATED;
245 break;
246 default:
247 /* The script is broken ... */
248 nps = D_UNKNOWN;
249 dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
250 return nps;
253 dev_info(DEV, "fence-peer helper returned %d (%s)\n",
254 (r>>8) & 0xff, ex_to_string);
255 return nps;
259 int drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
261 const int max_tries = 4;
262 int r = 0;
263 int try = 0;
264 int forced = 0;
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) {
282 val.pdsk = 0;
283 mask.pdsk = 0;
284 continue;
287 if (r == SS_NO_UP_TO_DATE_DISK && force &&
288 (mdev->state.disk == D_INCONSISTENT ||
289 mdev->state.disk == D_OUTDATED)) {
290 mask.disk = D_MASK;
291 val.disk = D_UP_TO_DATE;
292 forced = 1;
293 continue;
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;
303 mask.disk = D_MASK;
306 val.pdsk = nps;
307 mask.pdsk = D_MASK;
309 continue;
312 if (r == SS_NOTHING_TO_DO)
313 goto fail;
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");
319 nps = D_OUTDATED;
322 mask.pdsk = D_MASK;
323 val.pdsk = nps;
325 continue;
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);
332 if (try < max_tries)
333 try = max_tries - 1;
334 continue;
336 if (r < SS_SUCCESS) {
337 r = _drbd_request_state(mdev, mask, val,
338 CS_VERBOSE + CS_WAIT_COMPLETE);
339 if (r < SS_SUCCESS)
340 goto fail;
342 break;
345 if (r < SS_SUCCESS)
346 goto fail;
348 if (forced)
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;
358 put_ldev(mdev);
360 } else {
361 if (get_net_conf(mdev)) {
362 mdev->net_conf->want_lose = 0;
363 put_net_conf(mdev);
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;
373 put_ldev(mdev);
377 if ((new_role == R_SECONDARY) && get_ldev(mdev)) {
378 drbd_al_to_on_disk_bm(mdev);
379 put_ldev(mdev);
382 if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
383 /* if this was forced, we should consider sync */
384 if (forced)
385 drbd_send_uuids(mdev);
386 drbd_send_state(mdev);
389 drbd_md_sync(mdev);
391 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
392 fail:
393 mutex_unlock(&mdev->state_mutex);
394 return r;
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;
406 return 0;
409 reply->ret_code =
410 drbd_set_role(mdev, R_PRIMARY, primary_args.overwrite_peer);
412 return 0;
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);
420 return 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) {
430 default:
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;
436 break;
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;
443 break;
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;
462 break;
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' };
470 int base = 0;
471 while (size >= 10000) {
472 /* shift + round */
473 size = (size >> 10) + !!(size & (1<<9));
474 base++;
476 sprintf(buf, "%lu %cB", (long)size, units[base]);
478 return buf;
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, int force) __must_hold(local)
515 sector_t prev_first_sect, prev_size; /* previous meta location */
516 sector_t la_size;
517 sector_t size;
518 char ppb[10];
520 int md_moved, la_size_changed;
521 enum determine_dev_size rv = unchanged;
523 /* race:
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.
528 * to avoid that:
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, force);
546 if (drbd_get_capacity(mdev->this_bdev) != size ||
547 drbd_bm_capacity(mdev) != size) {
548 int err;
549 err = drbd_bm_resize(mdev, size);
550 if (unlikely(err)) {
551 /* currently there is only one error: ENOMEM! */
552 size = drbd_bm_capacity(mdev)>>1;
553 if (size == 0) {
554 dev_err(DEV, "OUT OF MEMORY! "
555 "Could not allocate bitmap!\n");
556 } else {
557 dev_err(DEV, "BM resizing failed. "
558 "Leaving size unchanged at size = %lu KB\n",
559 (unsigned long)size);
561 rv = dev_size_error;
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)
570 goto out;
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);
586 if (size > la_size)
587 rv = grew;
588 if (size < la_size)
589 rv = shrunk;
590 out:
591 lc_unlock(mdev->act_log);
592 wake_up(&mdev->al_wait);
593 drbd_resume_io(mdev);
595 return rv;
598 sector_t
599 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
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. */
605 sector_t size = 0;
607 m_size = drbd_get_max_capacity(bdev);
609 if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
610 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
611 p_size = m_size;
614 if (p_size && m_size) {
615 size = min_t(sector_t, p_size, m_size);
616 } else {
617 if (la_size) {
618 size = la_size;
619 if (m_size && m_size < size)
620 size = m_size;
621 if (p_size && p_size < size)
622 size = p_size;
623 } else {
624 if (m_size)
625 size = m_size;
626 if (p_size)
627 size = p_size;
631 if (size == 0)
632 dev_err(DEV, "Both nodes diskless!\n");
634 if (u_size) {
635 if (u_size > size)
636 dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
637 (unsigned long)u_size>>1, (unsigned long)size>>1);
638 else
639 size = u_size;
642 return size;
646 * drbd_check_al_size() - Ensures that the AL is of the right size
647 * @mdev: DRBD device.
649 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
650 * failed, and 0 on success. You should call drbd_md_sync() after you called
651 * this function.
653 static int drbd_check_al_size(struct drbd_conf *mdev)
655 struct lru_cache *n, *t;
656 struct lc_element *e;
657 unsigned int in_use;
658 int i;
660 ERR_IF(mdev->sync_conf.al_extents < 7)
661 mdev->sync_conf.al_extents = 127;
663 if (mdev->act_log &&
664 mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
665 return 0;
667 in_use = 0;
668 t = mdev->act_log;
669 n = lc_create("act_log", drbd_al_ext_cache,
670 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
672 if (n == NULL) {
673 dev_err(DEV, "Cannot allocate act_log lru!\n");
674 return -ENOMEM;
676 spin_lock_irq(&mdev->al_lock);
677 if (t) {
678 for (i = 0; i < t->nr_elements; i++) {
679 e = lc_element_by_index(t, i);
680 if (e->refcnt)
681 dev_err(DEV, "refcnt(%d)==%d\n",
682 e->lc_number, e->refcnt);
683 in_use += e->refcnt;
686 if (!in_use)
687 mdev->act_log = n;
688 spin_unlock_irq(&mdev->al_lock);
689 if (in_use) {
690 dev_err(DEV, "Activity log still in use!\n");
691 lc_destroy(n);
692 return -EBUSY;
693 } else {
694 if (t)
695 lc_destroy(t);
697 drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
698 return 0;
701 void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_seg_s) __must_hold(local)
703 struct request_queue * const q = mdev->rq_queue;
704 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
705 int max_segments = mdev->ldev->dc.max_bio_bvecs;
707 if (b->merge_bvec_fn && !mdev->ldev->dc.use_bmbv)
708 max_seg_s = PAGE_SIZE;
710 max_seg_s = min(queue_max_sectors(b) * queue_logical_block_size(b), max_seg_s);
712 blk_queue_max_hw_sectors(q, max_seg_s >> 9);
713 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
714 blk_queue_max_segment_size(q, max_seg_s);
715 blk_queue_logical_block_size(q, 512);
716 blk_queue_segment_boundary(q, PAGE_SIZE-1);
717 blk_stack_limits(&q->limits, &b->limits, 0);
719 if (b->merge_bvec_fn)
720 dev_warn(DEV, "Backing device's merge_bvec_fn() = %p\n",
721 b->merge_bvec_fn);
722 dev_info(DEV, "max_segment_size ( = BIO size ) = %u\n", queue_max_segment_size(q));
724 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
725 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
726 q->backing_dev_info.ra_pages,
727 b->backing_dev_info.ra_pages);
728 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
732 /* serialize deconfig (worker exiting, doing cleanup)
733 * and reconfig (drbdsetup disk, drbdsetup net)
735 * wait for a potentially exiting worker, then restart it,
736 * or start a new one.
738 static void drbd_reconfig_start(struct drbd_conf *mdev)
740 wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
741 wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
742 drbd_thread_start(&mdev->worker);
745 /* if still unconfigured, stops worker again.
746 * if configured now, clears CONFIG_PENDING.
747 * wakes potential waiters */
748 static void drbd_reconfig_done(struct drbd_conf *mdev)
750 spin_lock_irq(&mdev->req_lock);
751 if (mdev->state.disk == D_DISKLESS &&
752 mdev->state.conn == C_STANDALONE &&
753 mdev->state.role == R_SECONDARY) {
754 set_bit(DEVICE_DYING, &mdev->flags);
755 drbd_thread_stop_nowait(&mdev->worker);
756 } else
757 clear_bit(CONFIG_PENDING, &mdev->flags);
758 spin_unlock_irq(&mdev->req_lock);
759 wake_up(&mdev->state_wait);
762 /* does always return 0;
763 * interesting return code is in reply->ret_code */
764 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
765 struct drbd_nl_cfg_reply *reply)
767 enum drbd_ret_codes retcode;
768 enum determine_dev_size dd;
769 sector_t max_possible_sectors;
770 sector_t min_md_device_sectors;
771 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
772 struct inode *inode, *inode2;
773 struct lru_cache *resync_lru = NULL;
774 union drbd_state ns, os;
775 int rv;
776 int cp_discovered = 0;
777 int logical_block_size;
779 drbd_reconfig_start(mdev);
781 /* if you want to reconfigure, please tear down first */
782 if (mdev->state.disk > D_DISKLESS) {
783 retcode = ERR_DISK_CONFIGURED;
784 goto fail;
787 /* allocation not in the IO path, cqueue thread context */
788 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
789 if (!nbc) {
790 retcode = ERR_NOMEM;
791 goto fail;
794 nbc->dc.disk_size = DRBD_DISK_SIZE_SECT_DEF;
795 nbc->dc.on_io_error = DRBD_ON_IO_ERROR_DEF;
796 nbc->dc.fencing = DRBD_FENCING_DEF;
797 nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
799 if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
800 retcode = ERR_MANDATORY_TAG;
801 goto fail;
804 if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
805 retcode = ERR_MD_IDX_INVALID;
806 goto fail;
809 nbc->lo_file = filp_open(nbc->dc.backing_dev, O_RDWR, 0);
810 if (IS_ERR(nbc->lo_file)) {
811 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
812 PTR_ERR(nbc->lo_file));
813 nbc->lo_file = NULL;
814 retcode = ERR_OPEN_DISK;
815 goto fail;
818 inode = nbc->lo_file->f_dentry->d_inode;
820 if (!S_ISBLK(inode->i_mode)) {
821 retcode = ERR_DISK_NOT_BDEV;
822 goto fail;
825 nbc->md_file = filp_open(nbc->dc.meta_dev, O_RDWR, 0);
826 if (IS_ERR(nbc->md_file)) {
827 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
828 PTR_ERR(nbc->md_file));
829 nbc->md_file = NULL;
830 retcode = ERR_OPEN_MD_DISK;
831 goto fail;
834 inode2 = nbc->md_file->f_dentry->d_inode;
836 if (!S_ISBLK(inode2->i_mode)) {
837 retcode = ERR_MD_NOT_BDEV;
838 goto fail;
841 nbc->backing_bdev = inode->i_bdev;
842 if (bd_claim(nbc->backing_bdev, mdev)) {
843 printk(KERN_ERR "drbd: bd_claim(%p,%p); failed [%p;%p;%u]\n",
844 nbc->backing_bdev, mdev,
845 nbc->backing_bdev->bd_holder,
846 nbc->backing_bdev->bd_contains->bd_holder,
847 nbc->backing_bdev->bd_holders);
848 retcode = ERR_BDCLAIM_DISK;
849 goto fail;
852 resync_lru = lc_create("resync", drbd_bm_ext_cache,
853 61, sizeof(struct bm_extent),
854 offsetof(struct bm_extent, lce));
855 if (!resync_lru) {
856 retcode = ERR_NOMEM;
857 goto release_bdev_fail;
860 /* meta_dev_idx >= 0: external fixed size,
861 * possibly multiple drbd sharing one meta device.
862 * TODO in that case, paranoia check that [md_bdev, meta_dev_idx] is
863 * not yet used by some other drbd minor!
864 * (if you use drbd.conf + drbdadm,
865 * that should check it for you already; but if you don't, or someone
866 * fooled it, we need to double check here) */
867 nbc->md_bdev = inode2->i_bdev;
868 if (bd_claim(nbc->md_bdev, (nbc->dc.meta_dev_idx < 0) ? (void *)mdev
869 : (void *) drbd_m_holder)) {
870 retcode = ERR_BDCLAIM_MD_DISK;
871 goto release_bdev_fail;
874 if ((nbc->backing_bdev == nbc->md_bdev) !=
875 (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
876 nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
877 retcode = ERR_MD_IDX_INVALID;
878 goto release_bdev2_fail;
881 /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
882 drbd_md_set_sector_offsets(mdev, nbc);
884 if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
885 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
886 (unsigned long long) drbd_get_max_capacity(nbc),
887 (unsigned long long) nbc->dc.disk_size);
888 retcode = ERR_DISK_TO_SMALL;
889 goto release_bdev2_fail;
892 if (nbc->dc.meta_dev_idx < 0) {
893 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
894 /* at least one MB, otherwise it does not make sense */
895 min_md_device_sectors = (2<<10);
896 } else {
897 max_possible_sectors = DRBD_MAX_SECTORS;
898 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
901 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
902 retcode = ERR_MD_DISK_TO_SMALL;
903 dev_warn(DEV, "refusing attach: md-device too small, "
904 "at least %llu sectors needed for this meta-disk type\n",
905 (unsigned long long) min_md_device_sectors);
906 goto release_bdev2_fail;
909 /* Make sure the new disk is big enough
910 * (we may currently be R_PRIMARY with no local disk...) */
911 if (drbd_get_max_capacity(nbc) <
912 drbd_get_capacity(mdev->this_bdev)) {
913 retcode = ERR_DISK_TO_SMALL;
914 goto release_bdev2_fail;
917 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
919 if (nbc->known_size > max_possible_sectors) {
920 dev_warn(DEV, "==> truncating very big lower level device "
921 "to currently maximum possible %llu sectors <==\n",
922 (unsigned long long) max_possible_sectors);
923 if (nbc->dc.meta_dev_idx >= 0)
924 dev_warn(DEV, "==>> using internal or flexible "
925 "meta data may help <<==\n");
928 drbd_suspend_io(mdev);
929 /* also wait for the last barrier ack. */
930 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt));
931 /* and for any other previously queued work */
932 drbd_flush_workqueue(mdev);
934 retcode = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
935 drbd_resume_io(mdev);
936 if (retcode < SS_SUCCESS)
937 goto release_bdev2_fail;
939 if (!get_ldev_if_state(mdev, D_ATTACHING))
940 goto force_diskless;
942 drbd_md_set_sector_offsets(mdev, nbc);
944 if (!mdev->bitmap) {
945 if (drbd_bm_init(mdev)) {
946 retcode = ERR_NOMEM;
947 goto force_diskless_dec;
951 retcode = drbd_md_read(mdev, nbc);
952 if (retcode != NO_ERROR)
953 goto force_diskless_dec;
955 if (mdev->state.conn < C_CONNECTED &&
956 mdev->state.role == R_PRIMARY &&
957 (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
958 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
959 (unsigned long long)mdev->ed_uuid);
960 retcode = ERR_DATA_NOT_CURRENT;
961 goto force_diskless_dec;
964 /* Since we are diskless, fix the activity log first... */
965 if (drbd_check_al_size(mdev)) {
966 retcode = ERR_NOMEM;
967 goto force_diskless_dec;
970 /* Prevent shrinking of consistent devices ! */
971 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
972 drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
973 dev_warn(DEV, "refusing to truncate a consistent device\n");
974 retcode = ERR_DISK_TO_SMALL;
975 goto force_diskless_dec;
978 if (!drbd_al_read_log(mdev, nbc)) {
979 retcode = ERR_IO_MD_DISK;
980 goto force_diskless_dec;
983 /* allocate a second IO page if logical_block_size != 512 */
984 logical_block_size = bdev_logical_block_size(nbc->md_bdev);
985 if (logical_block_size == 0)
986 logical_block_size = MD_SECTOR_SIZE;
988 if (logical_block_size != MD_SECTOR_SIZE) {
989 if (!mdev->md_io_tmpp) {
990 struct page *page = alloc_page(GFP_NOIO);
991 if (!page)
992 goto force_diskless_dec;
994 dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
995 logical_block_size, MD_SECTOR_SIZE);
996 dev_warn(DEV, "Workaround engaged (has performance impact).\n");
998 mdev->md_io_tmpp = page;
1002 /* Reset the "barriers don't work" bits here, then force meta data to
1003 * be written, to ensure we determine if barriers are supported. */
1004 if (nbc->dc.no_md_flush)
1005 set_bit(MD_NO_BARRIER, &mdev->flags);
1006 else
1007 clear_bit(MD_NO_BARRIER, &mdev->flags);
1009 /* Point of no return reached.
1010 * Devices and memory are no longer released by error cleanup below.
1011 * now mdev takes over responsibility, and the state engine should
1012 * clean it up somewhere. */
1013 D_ASSERT(mdev->ldev == NULL);
1014 mdev->ldev = nbc;
1015 mdev->resync = resync_lru;
1016 nbc = NULL;
1017 resync_lru = NULL;
1019 mdev->write_ordering = WO_bio_barrier;
1020 drbd_bump_write_ordering(mdev, WO_bio_barrier);
1022 if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1023 set_bit(CRASHED_PRIMARY, &mdev->flags);
1024 else
1025 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1027 if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND)) {
1028 set_bit(CRASHED_PRIMARY, &mdev->flags);
1029 cp_discovered = 1;
1032 mdev->send_cnt = 0;
1033 mdev->recv_cnt = 0;
1034 mdev->read_cnt = 0;
1035 mdev->writ_cnt = 0;
1037 drbd_setup_queue_param(mdev, DRBD_MAX_SEGMENT_SIZE);
1039 /* If I am currently not R_PRIMARY,
1040 * but meta data primary indicator is set,
1041 * I just now recover from a hard crash,
1042 * and have been R_PRIMARY before that crash.
1044 * Now, if I had no connection before that crash
1045 * (have been degraded R_PRIMARY), chances are that
1046 * I won't find my peer now either.
1048 * In that case, and _only_ in that case,
1049 * we use the degr-wfc-timeout instead of the default,
1050 * so we can automatically recover from a crash of a
1051 * degraded but active "cluster" after a certain timeout.
1053 clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1054 if (mdev->state.role != R_PRIMARY &&
1055 drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1056 !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1057 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1059 dd = drbd_determin_dev_size(mdev, 0);
1060 if (dd == dev_size_error) {
1061 retcode = ERR_NOMEM_BITMAP;
1062 goto force_diskless_dec;
1063 } else if (dd == grew)
1064 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1066 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1067 dev_info(DEV, "Assuming that all blocks are out of sync "
1068 "(aka FullSync)\n");
1069 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from attaching")) {
1070 retcode = ERR_IO_MD_DISK;
1071 goto force_diskless_dec;
1073 } else {
1074 if (drbd_bitmap_io(mdev, &drbd_bm_read, "read from attaching") < 0) {
1075 retcode = ERR_IO_MD_DISK;
1076 goto force_diskless_dec;
1080 if (cp_discovered) {
1081 drbd_al_apply_to_bm(mdev);
1082 drbd_al_to_on_disk_bm(mdev);
1085 spin_lock_irq(&mdev->req_lock);
1086 os = mdev->state;
1087 ns.i = os.i;
1088 /* If MDF_CONSISTENT is not set go into inconsistent state,
1089 otherwise investigate MDF_WasUpToDate...
1090 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1091 otherwise into D_CONSISTENT state.
1093 if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1094 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1095 ns.disk = D_CONSISTENT;
1096 else
1097 ns.disk = D_OUTDATED;
1098 } else {
1099 ns.disk = D_INCONSISTENT;
1102 if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1103 ns.pdsk = D_OUTDATED;
1105 if ( ns.disk == D_CONSISTENT &&
1106 (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1107 ns.disk = D_UP_TO_DATE;
1109 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1110 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1111 this point, because drbd_request_state() modifies these
1112 flags. */
1114 /* In case we are C_CONNECTED postpone any decision on the new disk
1115 state after the negotiation phase. */
1116 if (mdev->state.conn == C_CONNECTED) {
1117 mdev->new_state_tmp.i = ns.i;
1118 ns.i = os.i;
1119 ns.disk = D_NEGOTIATING;
1122 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1123 ns = mdev->state;
1124 spin_unlock_irq(&mdev->req_lock);
1126 if (rv < SS_SUCCESS)
1127 goto force_diskless_dec;
1129 if (mdev->state.role == R_PRIMARY)
1130 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
1131 else
1132 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1134 drbd_md_mark_dirty(mdev);
1135 drbd_md_sync(mdev);
1137 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1138 put_ldev(mdev);
1139 reply->ret_code = retcode;
1140 drbd_reconfig_done(mdev);
1141 return 0;
1143 force_diskless_dec:
1144 put_ldev(mdev);
1145 force_diskless:
1146 drbd_force_state(mdev, NS(disk, D_DISKLESS));
1147 drbd_md_sync(mdev);
1148 release_bdev2_fail:
1149 if (nbc)
1150 bd_release(nbc->md_bdev);
1151 release_bdev_fail:
1152 if (nbc)
1153 bd_release(nbc->backing_bdev);
1154 fail:
1155 if (nbc) {
1156 if (nbc->lo_file)
1157 fput(nbc->lo_file);
1158 if (nbc->md_file)
1159 fput(nbc->md_file);
1160 kfree(nbc);
1162 lc_destroy(resync_lru);
1164 reply->ret_code = retcode;
1165 drbd_reconfig_done(mdev);
1166 return 0;
1169 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1170 struct drbd_nl_cfg_reply *reply)
1172 reply->ret_code = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1173 return 0;
1176 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1177 struct drbd_nl_cfg_reply *reply)
1179 int i, ns;
1180 enum drbd_ret_codes retcode;
1181 struct net_conf *new_conf = NULL;
1182 struct crypto_hash *tfm = NULL;
1183 struct crypto_hash *integrity_w_tfm = NULL;
1184 struct crypto_hash *integrity_r_tfm = NULL;
1185 struct hlist_head *new_tl_hash = NULL;
1186 struct hlist_head *new_ee_hash = NULL;
1187 struct drbd_conf *odev;
1188 char hmac_name[CRYPTO_MAX_ALG_NAME];
1189 void *int_dig_out = NULL;
1190 void *int_dig_in = NULL;
1191 void *int_dig_vv = NULL;
1192 struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1194 drbd_reconfig_start(mdev);
1196 if (mdev->state.conn > C_STANDALONE) {
1197 retcode = ERR_NET_CONFIGURED;
1198 goto fail;
1201 /* allocation not in the IO path, cqueue thread context */
1202 new_conf = kmalloc(sizeof(struct net_conf), GFP_KERNEL);
1203 if (!new_conf) {
1204 retcode = ERR_NOMEM;
1205 goto fail;
1208 memset(new_conf, 0, sizeof(struct net_conf));
1209 new_conf->timeout = DRBD_TIMEOUT_DEF;
1210 new_conf->try_connect_int = DRBD_CONNECT_INT_DEF;
1211 new_conf->ping_int = DRBD_PING_INT_DEF;
1212 new_conf->max_epoch_size = DRBD_MAX_EPOCH_SIZE_DEF;
1213 new_conf->max_buffers = DRBD_MAX_BUFFERS_DEF;
1214 new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1215 new_conf->sndbuf_size = DRBD_SNDBUF_SIZE_DEF;
1216 new_conf->rcvbuf_size = DRBD_RCVBUF_SIZE_DEF;
1217 new_conf->ko_count = DRBD_KO_COUNT_DEF;
1218 new_conf->after_sb_0p = DRBD_AFTER_SB_0P_DEF;
1219 new_conf->after_sb_1p = DRBD_AFTER_SB_1P_DEF;
1220 new_conf->after_sb_2p = DRBD_AFTER_SB_2P_DEF;
1221 new_conf->want_lose = 0;
1222 new_conf->two_primaries = 0;
1223 new_conf->wire_protocol = DRBD_PROT_C;
1224 new_conf->ping_timeo = DRBD_PING_TIMEO_DEF;
1225 new_conf->rr_conflict = DRBD_RR_CONFLICT_DEF;
1227 if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1228 retcode = ERR_MANDATORY_TAG;
1229 goto fail;
1232 if (new_conf->two_primaries
1233 && (new_conf->wire_protocol != DRBD_PROT_C)) {
1234 retcode = ERR_NOT_PROTO_C;
1235 goto fail;
1238 if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1239 retcode = ERR_DISCARD;
1240 goto fail;
1243 retcode = NO_ERROR;
1245 new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1246 new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1247 for (i = 0; i < minor_count; i++) {
1248 odev = minor_to_mdev(i);
1249 if (!odev || odev == mdev)
1250 continue;
1251 if (get_net_conf(odev)) {
1252 taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1253 if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1254 !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1255 retcode = ERR_LOCAL_ADDR;
1257 taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1258 if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1259 !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1260 retcode = ERR_PEER_ADDR;
1262 put_net_conf(odev);
1263 if (retcode != NO_ERROR)
1264 goto fail;
1268 if (new_conf->cram_hmac_alg[0] != 0) {
1269 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1270 new_conf->cram_hmac_alg);
1271 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1272 if (IS_ERR(tfm)) {
1273 tfm = NULL;
1274 retcode = ERR_AUTH_ALG;
1275 goto fail;
1278 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1279 retcode = ERR_AUTH_ALG_ND;
1280 goto fail;
1284 if (new_conf->integrity_alg[0]) {
1285 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1286 if (IS_ERR(integrity_w_tfm)) {
1287 integrity_w_tfm = NULL;
1288 retcode=ERR_INTEGRITY_ALG;
1289 goto fail;
1292 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1293 retcode=ERR_INTEGRITY_ALG_ND;
1294 goto fail;
1297 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1298 if (IS_ERR(integrity_r_tfm)) {
1299 integrity_r_tfm = NULL;
1300 retcode=ERR_INTEGRITY_ALG;
1301 goto fail;
1305 ns = new_conf->max_epoch_size/8;
1306 if (mdev->tl_hash_s != ns) {
1307 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1308 if (!new_tl_hash) {
1309 retcode = ERR_NOMEM;
1310 goto fail;
1314 ns = new_conf->max_buffers/8;
1315 if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1316 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1317 if (!new_ee_hash) {
1318 retcode = ERR_NOMEM;
1319 goto fail;
1323 ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1325 if (integrity_w_tfm) {
1326 i = crypto_hash_digestsize(integrity_w_tfm);
1327 int_dig_out = kmalloc(i, GFP_KERNEL);
1328 if (!int_dig_out) {
1329 retcode = ERR_NOMEM;
1330 goto fail;
1332 int_dig_in = kmalloc(i, GFP_KERNEL);
1333 if (!int_dig_in) {
1334 retcode = ERR_NOMEM;
1335 goto fail;
1337 int_dig_vv = kmalloc(i, GFP_KERNEL);
1338 if (!int_dig_vv) {
1339 retcode = ERR_NOMEM;
1340 goto fail;
1344 if (!mdev->bitmap) {
1345 if(drbd_bm_init(mdev)) {
1346 retcode = ERR_NOMEM;
1347 goto fail;
1351 spin_lock_irq(&mdev->req_lock);
1352 if (mdev->net_conf != NULL) {
1353 retcode = ERR_NET_CONFIGURED;
1354 spin_unlock_irq(&mdev->req_lock);
1355 goto fail;
1357 mdev->net_conf = new_conf;
1359 mdev->send_cnt = 0;
1360 mdev->recv_cnt = 0;
1362 if (new_tl_hash) {
1363 kfree(mdev->tl_hash);
1364 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1365 mdev->tl_hash = new_tl_hash;
1368 if (new_ee_hash) {
1369 kfree(mdev->ee_hash);
1370 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1371 mdev->ee_hash = new_ee_hash;
1374 crypto_free_hash(mdev->cram_hmac_tfm);
1375 mdev->cram_hmac_tfm = tfm;
1377 crypto_free_hash(mdev->integrity_w_tfm);
1378 mdev->integrity_w_tfm = integrity_w_tfm;
1380 crypto_free_hash(mdev->integrity_r_tfm);
1381 mdev->integrity_r_tfm = integrity_r_tfm;
1383 kfree(mdev->int_dig_out);
1384 kfree(mdev->int_dig_in);
1385 kfree(mdev->int_dig_vv);
1386 mdev->int_dig_out=int_dig_out;
1387 mdev->int_dig_in=int_dig_in;
1388 mdev->int_dig_vv=int_dig_vv;
1389 spin_unlock_irq(&mdev->req_lock);
1391 retcode = _drbd_request_state(mdev, NS(conn, C_UNCONNECTED), CS_VERBOSE);
1393 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1394 reply->ret_code = retcode;
1395 drbd_reconfig_done(mdev);
1396 return 0;
1398 fail:
1399 kfree(int_dig_out);
1400 kfree(int_dig_in);
1401 kfree(int_dig_vv);
1402 crypto_free_hash(tfm);
1403 crypto_free_hash(integrity_w_tfm);
1404 crypto_free_hash(integrity_r_tfm);
1405 kfree(new_tl_hash);
1406 kfree(new_ee_hash);
1407 kfree(new_conf);
1409 reply->ret_code = retcode;
1410 drbd_reconfig_done(mdev);
1411 return 0;
1414 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1415 struct drbd_nl_cfg_reply *reply)
1417 int retcode;
1419 retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1421 if (retcode == SS_NOTHING_TO_DO)
1422 goto done;
1423 else if (retcode == SS_ALREADY_STANDALONE)
1424 goto done;
1425 else if (retcode == SS_PRIMARY_NOP) {
1426 /* Our statche checking code wants to see the peer outdated. */
1427 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1428 pdsk, D_OUTDATED));
1429 } else if (retcode == SS_CW_FAILED_BY_PEER) {
1430 /* The peer probably wants to see us outdated. */
1431 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1432 disk, D_OUTDATED),
1433 CS_ORDERED);
1434 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1435 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1436 retcode = SS_SUCCESS;
1440 if (retcode < SS_SUCCESS)
1441 goto fail;
1443 if (wait_event_interruptible(mdev->state_wait,
1444 mdev->state.conn != C_DISCONNECTING)) {
1445 /* Do not test for mdev->state.conn == C_STANDALONE, since
1446 someone else might connect us in the mean time! */
1447 retcode = ERR_INTR;
1448 goto fail;
1451 done:
1452 retcode = NO_ERROR;
1453 fail:
1454 drbd_md_sync(mdev);
1455 reply->ret_code = retcode;
1456 return 0;
1459 void resync_after_online_grow(struct drbd_conf *mdev)
1461 int iass; /* I am sync source */
1463 dev_info(DEV, "Resync of new storage after online grow\n");
1464 if (mdev->state.role != mdev->state.peer)
1465 iass = (mdev->state.role == R_PRIMARY);
1466 else
1467 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1469 if (iass)
1470 drbd_start_resync(mdev, C_SYNC_SOURCE);
1471 else
1472 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1475 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1476 struct drbd_nl_cfg_reply *reply)
1478 struct resize rs;
1479 int retcode = NO_ERROR;
1480 int ldsc = 0; /* local disk size changed */
1481 enum determine_dev_size dd;
1483 memset(&rs, 0, sizeof(struct resize));
1484 if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1485 retcode = ERR_MANDATORY_TAG;
1486 goto fail;
1489 if (mdev->state.conn > C_CONNECTED) {
1490 retcode = ERR_RESIZE_RESYNC;
1491 goto fail;
1494 if (mdev->state.role == R_SECONDARY &&
1495 mdev->state.peer == R_SECONDARY) {
1496 retcode = ERR_NO_PRIMARY;
1497 goto fail;
1500 if (!get_ldev(mdev)) {
1501 retcode = ERR_NO_DISK;
1502 goto fail;
1505 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) {
1506 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1507 ldsc = 1;
1510 mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1511 dd = drbd_determin_dev_size(mdev, rs.resize_force);
1512 drbd_md_sync(mdev);
1513 put_ldev(mdev);
1514 if (dd == dev_size_error) {
1515 retcode = ERR_NOMEM_BITMAP;
1516 goto fail;
1519 if (mdev->state.conn == C_CONNECTED && (dd != unchanged || ldsc)) {
1520 if (dd == grew)
1521 set_bit(RESIZE_PENDING, &mdev->flags);
1523 drbd_send_uuids(mdev);
1524 drbd_send_sizes(mdev, 1);
1527 fail:
1528 reply->ret_code = retcode;
1529 return 0;
1532 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1533 struct drbd_nl_cfg_reply *reply)
1535 int retcode = NO_ERROR;
1536 int err;
1537 int ovr; /* online verify running */
1538 int rsr; /* re-sync running */
1539 struct crypto_hash *verify_tfm = NULL;
1540 struct crypto_hash *csums_tfm = NULL;
1541 struct syncer_conf sc;
1542 cpumask_var_t new_cpu_mask;
1544 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1545 retcode = ERR_NOMEM;
1546 goto fail;
1549 if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1550 memset(&sc, 0, sizeof(struct syncer_conf));
1551 sc.rate = DRBD_RATE_DEF;
1552 sc.after = DRBD_AFTER_DEF;
1553 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1554 } else
1555 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1557 if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1558 retcode = ERR_MANDATORY_TAG;
1559 goto fail;
1562 /* re-sync running */
1563 rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1564 mdev->state.conn == C_SYNC_TARGET ||
1565 mdev->state.conn == C_PAUSED_SYNC_S ||
1566 mdev->state.conn == C_PAUSED_SYNC_T );
1568 if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1569 retcode = ERR_CSUMS_RESYNC_RUNNING;
1570 goto fail;
1573 if (!rsr && sc.csums_alg[0]) {
1574 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1575 if (IS_ERR(csums_tfm)) {
1576 csums_tfm = NULL;
1577 retcode = ERR_CSUMS_ALG;
1578 goto fail;
1581 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1582 retcode = ERR_CSUMS_ALG_ND;
1583 goto fail;
1587 /* online verify running */
1588 ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1590 if (ovr) {
1591 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1592 retcode = ERR_VERIFY_RUNNING;
1593 goto fail;
1597 if (!ovr && sc.verify_alg[0]) {
1598 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1599 if (IS_ERR(verify_tfm)) {
1600 verify_tfm = NULL;
1601 retcode = ERR_VERIFY_ALG;
1602 goto fail;
1605 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1606 retcode = ERR_VERIFY_ALG_ND;
1607 goto fail;
1611 /* silently ignore cpu mask on UP kernel */
1612 if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1613 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1614 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1615 if (err) {
1616 dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1617 retcode = ERR_CPU_MASK_PARSE;
1618 goto fail;
1622 ERR_IF (sc.rate < 1) sc.rate = 1;
1623 ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1624 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1625 if (sc.al_extents > AL_MAX) {
1626 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1627 sc.al_extents = AL_MAX;
1629 #undef AL_MAX
1631 /* most sanity checks done, try to assign the new sync-after
1632 * dependency. need to hold the global lock in there,
1633 * to avoid a race in the dependency loop check. */
1634 retcode = drbd_alter_sa(mdev, sc.after);
1635 if (retcode != NO_ERROR)
1636 goto fail;
1638 /* ok, assign the rest of it as well.
1639 * lock against receive_SyncParam() */
1640 spin_lock(&mdev->peer_seq_lock);
1641 mdev->sync_conf = sc;
1643 if (!rsr) {
1644 crypto_free_hash(mdev->csums_tfm);
1645 mdev->csums_tfm = csums_tfm;
1646 csums_tfm = NULL;
1649 if (!ovr) {
1650 crypto_free_hash(mdev->verify_tfm);
1651 mdev->verify_tfm = verify_tfm;
1652 verify_tfm = NULL;
1654 spin_unlock(&mdev->peer_seq_lock);
1656 if (get_ldev(mdev)) {
1657 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1658 drbd_al_shrink(mdev);
1659 err = drbd_check_al_size(mdev);
1660 lc_unlock(mdev->act_log);
1661 wake_up(&mdev->al_wait);
1663 put_ldev(mdev);
1664 drbd_md_sync(mdev);
1666 if (err) {
1667 retcode = ERR_NOMEM;
1668 goto fail;
1672 if (mdev->state.conn >= C_CONNECTED)
1673 drbd_send_sync_param(mdev, &sc);
1675 if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1676 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1677 drbd_calc_cpu_mask(mdev);
1678 mdev->receiver.reset_cpu_mask = 1;
1679 mdev->asender.reset_cpu_mask = 1;
1680 mdev->worker.reset_cpu_mask = 1;
1683 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1684 fail:
1685 free_cpumask_var(new_cpu_mask);
1686 crypto_free_hash(csums_tfm);
1687 crypto_free_hash(verify_tfm);
1688 reply->ret_code = retcode;
1689 return 0;
1692 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1693 struct drbd_nl_cfg_reply *reply)
1695 int retcode;
1697 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1699 if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1700 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1702 while (retcode == SS_NEED_CONNECTION) {
1703 spin_lock_irq(&mdev->req_lock);
1704 if (mdev->state.conn < C_CONNECTED)
1705 retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1706 spin_unlock_irq(&mdev->req_lock);
1708 if (retcode != SS_NEED_CONNECTION)
1709 break;
1711 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1714 reply->ret_code = retcode;
1715 return 0;
1718 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1719 struct drbd_nl_cfg_reply *reply)
1722 reply->ret_code = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
1724 return 0;
1727 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1728 struct drbd_nl_cfg_reply *reply)
1730 int retcode = NO_ERROR;
1732 if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
1733 retcode = ERR_PAUSE_IS_SET;
1735 reply->ret_code = retcode;
1736 return 0;
1739 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1740 struct drbd_nl_cfg_reply *reply)
1742 int retcode = NO_ERROR;
1744 if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO)
1745 retcode = ERR_PAUSE_IS_CLEAR;
1747 reply->ret_code = retcode;
1748 return 0;
1751 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1752 struct drbd_nl_cfg_reply *reply)
1754 reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
1756 return 0;
1759 static int drbd_nl_resume_io(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(susp, 0));
1763 return 0;
1766 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1767 struct drbd_nl_cfg_reply *reply)
1769 reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
1770 return 0;
1773 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1774 struct drbd_nl_cfg_reply *reply)
1776 unsigned short *tl;
1778 tl = reply->tag_list;
1780 if (get_ldev(mdev)) {
1781 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
1782 put_ldev(mdev);
1785 if (get_net_conf(mdev)) {
1786 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
1787 put_net_conf(mdev);
1789 tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
1791 put_unaligned(TT_END, tl++); /* Close the tag list */
1793 return (int)((char *)tl - (char *)reply->tag_list);
1796 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1797 struct drbd_nl_cfg_reply *reply)
1799 unsigned short *tl = reply->tag_list;
1800 union drbd_state s = mdev->state;
1801 unsigned long rs_left;
1802 unsigned int res;
1804 tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
1806 /* no local ref, no bitmap, no syncer progress. */
1807 if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
1808 if (get_ldev(mdev)) {
1809 drbd_get_syncer_progress(mdev, &rs_left, &res);
1810 tl = tl_add_int(tl, T_sync_progress, &res);
1811 put_ldev(mdev);
1814 put_unaligned(TT_END, tl++); /* Close the tag list */
1816 return (int)((char *)tl - (char *)reply->tag_list);
1819 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1820 struct drbd_nl_cfg_reply *reply)
1822 unsigned short *tl;
1824 tl = reply->tag_list;
1826 if (get_ldev(mdev)) {
1827 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
1828 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
1829 put_ldev(mdev);
1831 put_unaligned(TT_END, tl++); /* Close the tag list */
1833 return (int)((char *)tl - (char *)reply->tag_list);
1837 * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
1838 * @mdev: DRBD device.
1839 * @nlp: Netlink/connector packet from drbdsetup
1840 * @reply: Reply packet for drbdsetup
1842 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1843 struct drbd_nl_cfg_reply *reply)
1845 unsigned short *tl;
1846 char rv;
1848 tl = reply->tag_list;
1850 rv = mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
1851 test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
1853 tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
1854 put_unaligned(TT_END, tl++); /* Close the tag list */
1856 return (int)((char *)tl - (char *)reply->tag_list);
1859 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1860 struct drbd_nl_cfg_reply *reply)
1862 /* default to resume from last known position, if possible */
1863 struct start_ov args =
1864 { .start_sector = mdev->ov_start_sector };
1866 if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
1867 reply->ret_code = ERR_MANDATORY_TAG;
1868 return 0;
1870 /* w_make_ov_request expects position to be aligned */
1871 mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
1872 reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
1873 return 0;
1877 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1878 struct drbd_nl_cfg_reply *reply)
1880 int retcode = NO_ERROR;
1881 int skip_initial_sync = 0;
1882 int err;
1884 struct new_c_uuid args;
1886 memset(&args, 0, sizeof(struct new_c_uuid));
1887 if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
1888 reply->ret_code = ERR_MANDATORY_TAG;
1889 return 0;
1892 mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
1894 if (!get_ldev(mdev)) {
1895 retcode = ERR_NO_DISK;
1896 goto out;
1899 /* this is "skip initial sync", assume to be clean */
1900 if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
1901 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
1902 dev_info(DEV, "Preparing to skip initial sync\n");
1903 skip_initial_sync = 1;
1904 } else if (mdev->state.conn != C_STANDALONE) {
1905 retcode = ERR_CONNECTED;
1906 goto out_dec;
1909 drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
1910 drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
1912 if (args.clear_bm) {
1913 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, "clear_n_write from new_c_uuid");
1914 if (err) {
1915 dev_err(DEV, "Writing bitmap failed with %d\n",err);
1916 retcode = ERR_IO_MD_DISK;
1918 if (skip_initial_sync) {
1919 drbd_send_uuids_skip_initial_sync(mdev);
1920 _drbd_uuid_set(mdev, UI_BITMAP, 0);
1921 spin_lock_irq(&mdev->req_lock);
1922 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
1923 CS_VERBOSE, NULL);
1924 spin_unlock_irq(&mdev->req_lock);
1928 drbd_md_sync(mdev);
1929 out_dec:
1930 put_ldev(mdev);
1931 out:
1932 mutex_unlock(&mdev->state_mutex);
1934 reply->ret_code = retcode;
1935 return 0;
1938 static struct drbd_conf *ensure_mdev(struct drbd_nl_cfg_req *nlp)
1940 struct drbd_conf *mdev;
1942 if (nlp->drbd_minor >= minor_count)
1943 return NULL;
1945 mdev = minor_to_mdev(nlp->drbd_minor);
1947 if (!mdev && (nlp->flags & DRBD_NL_CREATE_DEVICE)) {
1948 struct gendisk *disk = NULL;
1949 mdev = drbd_new_device(nlp->drbd_minor);
1951 spin_lock_irq(&drbd_pp_lock);
1952 if (minor_table[nlp->drbd_minor] == NULL) {
1953 minor_table[nlp->drbd_minor] = mdev;
1954 disk = mdev->vdisk;
1955 mdev = NULL;
1956 } /* else: we lost the race */
1957 spin_unlock_irq(&drbd_pp_lock);
1959 if (disk) /* we won the race above */
1960 /* in case we ever add a drbd_delete_device(),
1961 * don't forget the del_gendisk! */
1962 add_disk(disk);
1963 else /* we lost the race above */
1964 drbd_free_mdev(mdev);
1966 mdev = minor_to_mdev(nlp->drbd_minor);
1969 return mdev;
1972 struct cn_handler_struct {
1973 int (*function)(struct drbd_conf *,
1974 struct drbd_nl_cfg_req *,
1975 struct drbd_nl_cfg_reply *);
1976 int reply_body_size;
1979 static struct cn_handler_struct cnd_table[] = {
1980 [ P_primary ] = { &drbd_nl_primary, 0 },
1981 [ P_secondary ] = { &drbd_nl_secondary, 0 },
1982 [ P_disk_conf ] = { &drbd_nl_disk_conf, 0 },
1983 [ P_detach ] = { &drbd_nl_detach, 0 },
1984 [ P_net_conf ] = { &drbd_nl_net_conf, 0 },
1985 [ P_disconnect ] = { &drbd_nl_disconnect, 0 },
1986 [ P_resize ] = { &drbd_nl_resize, 0 },
1987 [ P_syncer_conf ] = { &drbd_nl_syncer_conf, 0 },
1988 [ P_invalidate ] = { &drbd_nl_invalidate, 0 },
1989 [ P_invalidate_peer ] = { &drbd_nl_invalidate_peer, 0 },
1990 [ P_pause_sync ] = { &drbd_nl_pause_sync, 0 },
1991 [ P_resume_sync ] = { &drbd_nl_resume_sync, 0 },
1992 [ P_suspend_io ] = { &drbd_nl_suspend_io, 0 },
1993 [ P_resume_io ] = { &drbd_nl_resume_io, 0 },
1994 [ P_outdate ] = { &drbd_nl_outdate, 0 },
1995 [ P_get_config ] = { &drbd_nl_get_config,
1996 sizeof(struct syncer_conf_tag_len_struct) +
1997 sizeof(struct disk_conf_tag_len_struct) +
1998 sizeof(struct net_conf_tag_len_struct) },
1999 [ P_get_state ] = { &drbd_nl_get_state,
2000 sizeof(struct get_state_tag_len_struct) +
2001 sizeof(struct sync_progress_tag_len_struct) },
2002 [ P_get_uuids ] = { &drbd_nl_get_uuids,
2003 sizeof(struct get_uuids_tag_len_struct) },
2004 [ P_get_timeout_flag ] = { &drbd_nl_get_timeout_flag,
2005 sizeof(struct get_timeout_flag_tag_len_struct)},
2006 [ P_start_ov ] = { &drbd_nl_start_ov, 0 },
2007 [ P_new_c_uuid ] = { &drbd_nl_new_c_uuid, 0 },
2010 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2012 struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2013 struct cn_handler_struct *cm;
2014 struct cn_msg *cn_reply;
2015 struct drbd_nl_cfg_reply *reply;
2016 struct drbd_conf *mdev;
2017 int retcode, rr;
2018 int reply_size = sizeof(struct cn_msg)
2019 + sizeof(struct drbd_nl_cfg_reply)
2020 + sizeof(short int);
2022 if (!try_module_get(THIS_MODULE)) {
2023 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2024 return;
2027 if (!cap_raised(nsp->eff_cap, CAP_SYS_ADMIN)) {
2028 retcode = ERR_PERM;
2029 goto fail;
2032 mdev = ensure_mdev(nlp);
2033 if (!mdev) {
2034 retcode = ERR_MINOR_INVALID;
2035 goto fail;
2038 if (nlp->packet_type >= P_nl_after_last_packet) {
2039 retcode = ERR_PACKET_NR;
2040 goto fail;
2043 cm = cnd_table + nlp->packet_type;
2045 /* This may happen if packet number is 0: */
2046 if (cm->function == NULL) {
2047 retcode = ERR_PACKET_NR;
2048 goto fail;
2051 reply_size += cm->reply_body_size;
2053 /* allocation not in the IO path, cqueue thread context */
2054 cn_reply = kmalloc(reply_size, GFP_KERNEL);
2055 if (!cn_reply) {
2056 retcode = ERR_NOMEM;
2057 goto fail;
2059 reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2061 reply->packet_type =
2062 cm->reply_body_size ? nlp->packet_type : P_nl_after_last_packet;
2063 reply->minor = nlp->drbd_minor;
2064 reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2065 /* reply->tag_list; might be modified by cm->function. */
2067 rr = cm->function(mdev, nlp, reply);
2069 cn_reply->id = req->id;
2070 cn_reply->seq = req->seq;
2071 cn_reply->ack = req->ack + 1;
2072 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2073 cn_reply->flags = 0;
2075 rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2076 if (rr && rr != -ESRCH)
2077 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2079 kfree(cn_reply);
2080 module_put(THIS_MODULE);
2081 return;
2082 fail:
2083 drbd_nl_send_reply(req, retcode);
2084 module_put(THIS_MODULE);
2087 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2089 static unsigned short *
2090 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2091 unsigned short len, int nul_terminated)
2093 unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2094 len = (len < l) ? len : l;
2095 put_unaligned(tag, tl++);
2096 put_unaligned(len, tl++);
2097 memcpy(tl, data, len);
2098 tl = (unsigned short*)((char*)tl + len);
2099 if (nul_terminated)
2100 *((char*)tl - 1) = 0;
2101 return tl;
2104 static unsigned short *
2105 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2107 return __tl_add_blob(tl, tag, data, len, 0);
2110 static unsigned short *
2111 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2113 return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2116 static unsigned short *
2117 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2119 put_unaligned(tag, tl++);
2120 switch(tag_type(tag)) {
2121 case TT_INTEGER:
2122 put_unaligned(sizeof(int), tl++);
2123 put_unaligned(*(int *)val, (int *)tl);
2124 tl = (unsigned short*)((char*)tl+sizeof(int));
2125 break;
2126 case TT_INT64:
2127 put_unaligned(sizeof(u64), tl++);
2128 put_unaligned(*(u64 *)val, (u64 *)tl);
2129 tl = (unsigned short*)((char*)tl+sizeof(u64));
2130 break;
2131 default:
2132 /* someone did something stupid. */
2135 return tl;
2138 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2140 char buffer[sizeof(struct cn_msg)+
2141 sizeof(struct drbd_nl_cfg_reply)+
2142 sizeof(struct get_state_tag_len_struct)+
2143 sizeof(short int)];
2144 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2145 struct drbd_nl_cfg_reply *reply =
2146 (struct drbd_nl_cfg_reply *)cn_reply->data;
2147 unsigned short *tl = reply->tag_list;
2149 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2151 tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2153 put_unaligned(TT_END, tl++); /* Close the tag list */
2155 cn_reply->id.idx = CN_IDX_DRBD;
2156 cn_reply->id.val = CN_VAL_DRBD;
2158 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2159 cn_reply->ack = 0; /* not used here. */
2160 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2161 (int)((char *)tl - (char *)reply->tag_list);
2162 cn_reply->flags = 0;
2164 reply->packet_type = P_get_state;
2165 reply->minor = mdev_to_minor(mdev);
2166 reply->ret_code = NO_ERROR;
2168 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2171 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2173 char buffer[sizeof(struct cn_msg)+
2174 sizeof(struct drbd_nl_cfg_reply)+
2175 sizeof(struct call_helper_tag_len_struct)+
2176 sizeof(short int)];
2177 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2178 struct drbd_nl_cfg_reply *reply =
2179 (struct drbd_nl_cfg_reply *)cn_reply->data;
2180 unsigned short *tl = reply->tag_list;
2182 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2184 tl = tl_add_str(tl, T_helper, helper_name);
2185 put_unaligned(TT_END, tl++); /* Close the tag list */
2187 cn_reply->id.idx = CN_IDX_DRBD;
2188 cn_reply->id.val = CN_VAL_DRBD;
2190 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2191 cn_reply->ack = 0; /* not used here. */
2192 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2193 (int)((char *)tl - (char *)reply->tag_list);
2194 cn_reply->flags = 0;
2196 reply->packet_type = P_call_helper;
2197 reply->minor = mdev_to_minor(mdev);
2198 reply->ret_code = NO_ERROR;
2200 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2203 void drbd_bcast_ee(struct drbd_conf *mdev,
2204 const char *reason, const int dgs,
2205 const char* seen_hash, const char* calc_hash,
2206 const struct drbd_epoch_entry* e)
2208 struct cn_msg *cn_reply;
2209 struct drbd_nl_cfg_reply *reply;
2210 struct bio_vec *bvec;
2211 unsigned short *tl;
2212 int i;
2214 if (!e)
2215 return;
2216 if (!reason || !reason[0])
2217 return;
2219 /* apparently we have to memcpy twice, first to prepare the data for the
2220 * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2221 * netlink skb. */
2222 /* receiver thread context, which is not in the writeout path (of this node),
2223 * but may be in the writeout path of the _other_ node.
2224 * GFP_NOIO to avoid potential "distributed deadlock". */
2225 cn_reply = kmalloc(
2226 sizeof(struct cn_msg)+
2227 sizeof(struct drbd_nl_cfg_reply)+
2228 sizeof(struct dump_ee_tag_len_struct)+
2229 sizeof(short int),
2230 GFP_NOIO);
2232 if (!cn_reply) {
2233 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2234 (unsigned long long)e->sector, e->size);
2235 return;
2238 reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2239 tl = reply->tag_list;
2241 tl = tl_add_str(tl, T_dump_ee_reason, reason);
2242 tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2243 tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2244 tl = tl_add_int(tl, T_ee_sector, &e->sector);
2245 tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2247 put_unaligned(T_ee_data, tl++);
2248 put_unaligned(e->size, tl++);
2250 __bio_for_each_segment(bvec, e->private_bio, i, 0) {
2251 void *d = kmap(bvec->bv_page);
2252 memcpy(tl, d + bvec->bv_offset, bvec->bv_len);
2253 kunmap(bvec->bv_page);
2254 tl=(unsigned short*)((char*)tl + bvec->bv_len);
2256 put_unaligned(TT_END, tl++); /* Close the tag list */
2258 cn_reply->id.idx = CN_IDX_DRBD;
2259 cn_reply->id.val = CN_VAL_DRBD;
2261 cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2262 cn_reply->ack = 0; // not used here.
2263 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2264 (int)((char*)tl - (char*)reply->tag_list);
2265 cn_reply->flags = 0;
2267 reply->packet_type = P_dump_ee;
2268 reply->minor = mdev_to_minor(mdev);
2269 reply->ret_code = NO_ERROR;
2271 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2272 kfree(cn_reply);
2275 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2277 char buffer[sizeof(struct cn_msg)+
2278 sizeof(struct drbd_nl_cfg_reply)+
2279 sizeof(struct sync_progress_tag_len_struct)+
2280 sizeof(short int)];
2281 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2282 struct drbd_nl_cfg_reply *reply =
2283 (struct drbd_nl_cfg_reply *)cn_reply->data;
2284 unsigned short *tl = reply->tag_list;
2285 unsigned long rs_left;
2286 unsigned int res;
2288 /* no local ref, no bitmap, no syncer progress, no broadcast. */
2289 if (!get_ldev(mdev))
2290 return;
2291 drbd_get_syncer_progress(mdev, &rs_left, &res);
2292 put_ldev(mdev);
2294 tl = tl_add_int(tl, T_sync_progress, &res);
2295 put_unaligned(TT_END, tl++); /* Close the tag list */
2297 cn_reply->id.idx = CN_IDX_DRBD;
2298 cn_reply->id.val = CN_VAL_DRBD;
2300 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2301 cn_reply->ack = 0; /* not used here. */
2302 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2303 (int)((char *)tl - (char *)reply->tag_list);
2304 cn_reply->flags = 0;
2306 reply->packet_type = P_sync_progress;
2307 reply->minor = mdev_to_minor(mdev);
2308 reply->ret_code = NO_ERROR;
2310 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2313 int __init drbd_nl_init(void)
2315 static struct cb_id cn_id_drbd;
2316 int err, try=10;
2318 cn_id_drbd.val = CN_VAL_DRBD;
2319 do {
2320 cn_id_drbd.idx = cn_idx;
2321 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2322 if (!err)
2323 break;
2324 cn_idx = (cn_idx + CN_IDX_STEP);
2325 } while (try--);
2327 if (err) {
2328 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2329 return err;
2332 return 0;
2335 void drbd_nl_cleanup(void)
2337 static struct cb_id cn_id_drbd;
2339 cn_id_drbd.idx = cn_idx;
2340 cn_id_drbd.val = CN_VAL_DRBD;
2342 cn_del_callback(&cn_id_drbd);
2345 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2347 char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2348 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2349 struct drbd_nl_cfg_reply *reply =
2350 (struct drbd_nl_cfg_reply *)cn_reply->data;
2351 int rr;
2353 cn_reply->id = req->id;
2355 cn_reply->seq = req->seq;
2356 cn_reply->ack = req->ack + 1;
2357 cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2358 cn_reply->flags = 0;
2360 reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2361 reply->ret_code = ret_code;
2363 rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2364 if (rr && rr != -ESRCH)
2365 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);