blob: 6742652c8abc7f8b65fe0b211cd4d11bc62961e9 [file] [log] [blame]
Philipp Reisnerb411b362009-09-25 16:07:19 -07001/*
2 drbd_nl.c
3
4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
5
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>.
9
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.
14
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.
19
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.
23
24 */
25
Philipp Reisnerb411b362009-09-25 16:07:19 -070026#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"
Philipp Reisner265be2d2010-05-31 10:14:17 +020036#include "drbd_req.h"
Philipp Reisnerb411b362009-09-25 16:07:19 -070037#include "drbd_wrappers.h"
38#include <asm/unaligned.h>
39#include <linux/drbd_tag_magic.h>
40#include <linux/drbd_limits.h>
Philipp Reisner87f7be42010-06-11 13:56:33 +020041#include <linux/compiler.h>
42#include <linux/kthread.h>
Philipp Reisnerb411b362009-09-25 16:07:19 -070043
44static unsigned short *tl_add_blob(unsigned short *, enum drbd_tags, const void *, int);
45static unsigned short *tl_add_str(unsigned short *, enum drbd_tags, const char *);
46static unsigned short *tl_add_int(unsigned short *, enum drbd_tags, const void *);
47
48/* see get_sb_bdev and bd_claim */
49static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
50
51/* Generate the tag_list to struct functions */
52#define NL_PACKET(name, number, fields) \
53static int name ## _from_tags(struct drbd_conf *mdev, \
54 unsigned short *tags, struct name *arg) __attribute__ ((unused)); \
55static int name ## _from_tags(struct drbd_conf *mdev, \
56 unsigned short *tags, struct name *arg) \
57{ \
58 int tag; \
59 int dlen; \
60 \
61 while ((tag = get_unaligned(tags++)) != TT_END) { \
62 dlen = get_unaligned(tags++); \
63 switch (tag_number(tag)) { \
64 fields \
65 default: \
66 if (tag & T_MANDATORY) { \
67 dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \
68 return 0; \
69 } \
70 } \
71 tags = (unsigned short *)((char *)tags + dlen); \
72 } \
73 return 1; \
74}
75#define NL_INTEGER(pn, pr, member) \
76 case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \
77 arg->member = get_unaligned((int *)(tags)); \
78 break;
79#define NL_INT64(pn, pr, member) \
80 case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \
81 arg->member = get_unaligned((u64 *)(tags)); \
82 break;
83#define NL_BIT(pn, pr, member) \
84 case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \
85 arg->member = *(char *)(tags) ? 1 : 0; \
86 break;
87#define NL_STRING(pn, pr, member, len) \
88 case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \
89 if (dlen > len) { \
90 dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \
91 #member, dlen, (unsigned int)len); \
92 return 0; \
93 } \
94 arg->member ## _len = dlen; \
95 memcpy(arg->member, tags, min_t(size_t, dlen, len)); \
96 break;
97#include "linux/drbd_nl.h"
98
99/* Generate the struct to tag_list functions */
100#define NL_PACKET(name, number, fields) \
101static unsigned short* \
102name ## _to_tags(struct drbd_conf *mdev, \
103 struct name *arg, unsigned short *tags) __attribute__ ((unused)); \
104static unsigned short* \
105name ## _to_tags(struct drbd_conf *mdev, \
106 struct name *arg, unsigned short *tags) \
107{ \
108 fields \
109 return tags; \
110}
111
112#define NL_INTEGER(pn, pr, member) \
113 put_unaligned(pn | pr | TT_INTEGER, tags++); \
114 put_unaligned(sizeof(int), tags++); \
115 put_unaligned(arg->member, (int *)tags); \
116 tags = (unsigned short *)((char *)tags+sizeof(int));
117#define NL_INT64(pn, pr, member) \
118 put_unaligned(pn | pr | TT_INT64, tags++); \
119 put_unaligned(sizeof(u64), tags++); \
120 put_unaligned(arg->member, (u64 *)tags); \
121 tags = (unsigned short *)((char *)tags+sizeof(u64));
122#define NL_BIT(pn, pr, member) \
123 put_unaligned(pn | pr | TT_BIT, tags++); \
124 put_unaligned(sizeof(char), tags++); \
125 *(char *)tags = arg->member; \
126 tags = (unsigned short *)((char *)tags+sizeof(char));
127#define NL_STRING(pn, pr, member, len) \
128 put_unaligned(pn | pr | TT_STRING, tags++); \
129 put_unaligned(arg->member ## _len, tags++); \
130 memcpy(tags, arg->member, arg->member ## _len); \
131 tags = (unsigned short *)((char *)tags + arg->member ## _len);
132#include "linux/drbd_nl.h"
133
134void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name);
135void drbd_nl_send_reply(struct cn_msg *, int);
136
137int drbd_khelper(struct drbd_conf *mdev, char *cmd)
138{
139 char *envp[] = { "HOME=/",
140 "TERM=linux",
141 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
142 NULL, /* Will be set to address family */
143 NULL, /* Will be set to address */
144 NULL };
145
146 char mb[12], af[20], ad[60], *afs;
147 char *argv[] = {usermode_helper, cmd, mb, NULL };
148 int ret;
149
150 snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
151
152 if (get_net_conf(mdev)) {
153 switch (((struct sockaddr *)mdev->net_conf->peer_addr)->sa_family) {
154 case AF_INET6:
155 afs = "ipv6";
156 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6",
157 &((struct sockaddr_in6 *)mdev->net_conf->peer_addr)->sin6_addr);
158 break;
159 case AF_INET:
160 afs = "ipv4";
161 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
162 &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
163 break;
164 default:
165 afs = "ssocks";
166 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
167 &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
168 }
169 snprintf(af, 20, "DRBD_PEER_AF=%s", afs);
170 envp[3]=af;
171 envp[4]=ad;
172 put_net_conf(mdev);
173 }
174
175 dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
176
177 drbd_bcast_ev_helper(mdev, cmd);
178 ret = call_usermodehelper(usermode_helper, argv, envp, 1);
179 if (ret)
180 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
181 usermode_helper, cmd, mb,
182 (ret >> 8) & 0xff, ret);
183 else
184 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
185 usermode_helper, cmd, mb,
186 (ret >> 8) & 0xff, ret);
187
188 if (ret < 0) /* Ignore any ERRNOs we got. */
189 ret = 0;
190
191 return ret;
192}
193
194enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
195{
196 char *ex_to_string;
197 int r;
198 enum drbd_disk_state nps;
199 enum drbd_fencing_p fp;
200
201 D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
202
203 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
204 fp = mdev->ldev->dc.fencing;
205 put_ldev(mdev);
206 } else {
207 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n");
208 return mdev->state.pdsk;
209 }
210
Philipp Reisnerb411b362009-09-25 16:07:19 -0700211 r = drbd_khelper(mdev, "fence-peer");
212
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;
230 }
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;
251 }
252
253 dev_info(DEV, "fence-peer helper returned %d (%s)\n",
254 (r>>8) & 0xff, ex_to_string);
255 return nps;
256}
257
Philipp Reisner87f7be42010-06-11 13:56:33 +0200258static int _try_outdate_peer_async(void *data)
259{
260 struct drbd_conf *mdev = (struct drbd_conf *)data;
261 enum drbd_disk_state nps;
262
263 nps = drbd_try_outdate_peer(mdev);
264 drbd_request_state(mdev, NS(pdsk, nps));
265
266 return 0;
267}
268
269void drbd_try_outdate_peer_async(struct drbd_conf *mdev)
270{
271 struct task_struct *opa;
272
273 opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev));
274 if (IS_ERR(opa))
275 dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n");
276}
Philipp Reisnerb411b362009-09-25 16:07:19 -0700277
278int drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
279{
280 const int max_tries = 4;
281 int r = 0;
282 int try = 0;
283 int forced = 0;
284 union drbd_state mask, val;
285 enum drbd_disk_state nps;
286
287 if (new_role == R_PRIMARY)
288 request_ping(mdev); /* Detect a dead peer ASAP */
289
290 mutex_lock(&mdev->state_mutex);
291
292 mask.i = 0; mask.role = R_MASK;
293 val.i = 0; val.role = new_role;
294
295 while (try++ < max_tries) {
296 r = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
297
298 /* in case we first succeeded to outdate,
299 * but now suddenly could establish a connection */
300 if (r == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
301 val.pdsk = 0;
302 mask.pdsk = 0;
303 continue;
304 }
305
306 if (r == SS_NO_UP_TO_DATE_DISK && force &&
Philipp Reisnerd10a33c2010-03-04 15:11:39 +0100307 (mdev->state.disk < D_UP_TO_DATE &&
308 mdev->state.disk >= D_INCONSISTENT)) {
Philipp Reisnerb411b362009-09-25 16:07:19 -0700309 mask.disk = D_MASK;
310 val.disk = D_UP_TO_DATE;
311 forced = 1;
312 continue;
313 }
314
315 if (r == SS_NO_UP_TO_DATE_DISK &&
316 mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
317 D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
318 nps = drbd_try_outdate_peer(mdev);
319
320 if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
321 val.disk = D_UP_TO_DATE;
322 mask.disk = D_MASK;
323 }
324
325 val.pdsk = nps;
326 mask.pdsk = D_MASK;
327
328 continue;
329 }
330
331 if (r == SS_NOTHING_TO_DO)
332 goto fail;
333 if (r == SS_PRIMARY_NOP && mask.pdsk == 0) {
334 nps = drbd_try_outdate_peer(mdev);
335
336 if (force && nps > D_OUTDATED) {
337 dev_warn(DEV, "Forced into split brain situation!\n");
338 nps = D_OUTDATED;
339 }
340
341 mask.pdsk = D_MASK;
342 val.pdsk = nps;
343
344 continue;
345 }
346 if (r == SS_TWO_PRIMARIES) {
347 /* Maybe the peer is detected as dead very soon...
348 retry at most once more in this case. */
349 __set_current_state(TASK_INTERRUPTIBLE);
350 schedule_timeout((mdev->net_conf->ping_timeo+1)*HZ/10);
351 if (try < max_tries)
352 try = max_tries - 1;
353 continue;
354 }
355 if (r < SS_SUCCESS) {
356 r = _drbd_request_state(mdev, mask, val,
357 CS_VERBOSE + CS_WAIT_COMPLETE);
358 if (r < SS_SUCCESS)
359 goto fail;
360 }
361 break;
362 }
363
364 if (r < SS_SUCCESS)
365 goto fail;
366
367 if (forced)
368 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
369
370 /* Wait until nothing is on the fly :) */
371 wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
372
373 if (new_role == R_SECONDARY) {
374 set_disk_ro(mdev->vdisk, TRUE);
375 if (get_ldev(mdev)) {
376 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
377 put_ldev(mdev);
378 }
379 } else {
380 if (get_net_conf(mdev)) {
381 mdev->net_conf->want_lose = 0;
382 put_net_conf(mdev);
383 }
384 set_disk_ro(mdev->vdisk, FALSE);
385 if (get_ldev(mdev)) {
386 if (((mdev->state.conn < C_CONNECTED ||
387 mdev->state.pdsk <= D_FAILED)
388 && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
389 drbd_uuid_new_current(mdev);
390
391 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
392 put_ldev(mdev);
393 }
394 }
395
396 if ((new_role == R_SECONDARY) && get_ldev(mdev)) {
397 drbd_al_to_on_disk_bm(mdev);
398 put_ldev(mdev);
399 }
400
401 if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
402 /* if this was forced, we should consider sync */
403 if (forced)
404 drbd_send_uuids(mdev);
405 drbd_send_state(mdev);
406 }
407
408 drbd_md_sync(mdev);
409
410 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
411 fail:
412 mutex_unlock(&mdev->state_mutex);
413 return r;
414}
415
416
417static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
418 struct drbd_nl_cfg_reply *reply)
419{
420 struct primary primary_args;
421
422 memset(&primary_args, 0, sizeof(struct primary));
423 if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) {
424 reply->ret_code = ERR_MANDATORY_TAG;
425 return 0;
426 }
427
428 reply->ret_code =
Philipp Reisner1f552432010-03-04 15:51:01 +0100429 drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700430
431 return 0;
432}
433
434static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
435 struct drbd_nl_cfg_reply *reply)
436{
437 reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0);
438
439 return 0;
440}
441
442/* initializes the md.*_offset members, so we are able to find
443 * the on disk meta data */
444static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
445 struct drbd_backing_dev *bdev)
446{
447 sector_t md_size_sect = 0;
448 switch (bdev->dc.meta_dev_idx) {
449 default:
450 /* v07 style fixed size indexed meta data */
451 bdev->md.md_size_sect = MD_RESERVED_SECT;
452 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
453 bdev->md.al_offset = MD_AL_OFFSET;
454 bdev->md.bm_offset = MD_BM_OFFSET;
455 break;
456 case DRBD_MD_INDEX_FLEX_EXT:
457 /* just occupy the full device; unit: sectors */
458 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
459 bdev->md.md_offset = 0;
460 bdev->md.al_offset = MD_AL_OFFSET;
461 bdev->md.bm_offset = MD_BM_OFFSET;
462 break;
463 case DRBD_MD_INDEX_INTERNAL:
464 case DRBD_MD_INDEX_FLEX_INT:
465 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
466 /* al size is still fixed */
467 bdev->md.al_offset = -MD_AL_MAX_SIZE;
468 /* we need (slightly less than) ~ this much bitmap sectors: */
469 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
470 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
471 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
472 md_size_sect = ALIGN(md_size_sect, 8);
473
474 /* plus the "drbd meta data super block",
475 * and the activity log; */
476 md_size_sect += MD_BM_OFFSET;
477
478 bdev->md.md_size_sect = md_size_sect;
479 /* bitmap offset is adjusted by 'super' block size */
480 bdev->md.bm_offset = -md_size_sect + MD_AL_OFFSET;
481 break;
482 }
483}
484
485char *ppsize(char *buf, unsigned long long size)
486{
487 /* Needs 9 bytes at max. */
488 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
489 int base = 0;
490 while (size >= 10000) {
491 /* shift + round */
492 size = (size >> 10) + !!(size & (1<<9));
493 base++;
494 }
495 sprintf(buf, "%lu %cB", (long)size, units[base]);
496
497 return buf;
498}
499
500/* there is still a theoretical deadlock when called from receiver
501 * on an D_INCONSISTENT R_PRIMARY:
502 * remote READ does inc_ap_bio, receiver would need to receive answer
503 * packet from remote to dec_ap_bio again.
504 * receiver receive_sizes(), comes here,
505 * waits for ap_bio_cnt == 0. -> deadlock.
506 * but this cannot happen, actually, because:
507 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
508 * (not connected, or bad/no disk on peer):
509 * see drbd_fail_request_early, ap_bio_cnt is zero.
510 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
511 * peer may not initiate a resize.
512 */
513void drbd_suspend_io(struct drbd_conf *mdev)
514{
515 set_bit(SUSPEND_IO, &mdev->flags);
Philipp Reisner265be2d2010-05-31 10:14:17 +0200516 if (mdev->state.susp)
517 return;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700518 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
519}
520
521void drbd_resume_io(struct drbd_conf *mdev)
522{
523 clear_bit(SUSPEND_IO, &mdev->flags);
524 wake_up(&mdev->misc_wait);
525}
526
527/**
528 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
529 * @mdev: DRBD device.
530 *
531 * Returns 0 on success, negative return values indicate errors.
532 * You should call drbd_md_sync() after calling this function.
533 */
Philipp Reisnerd8450302010-03-24 15:51:26 +0100534enum determine_dev_size drbd_determin_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700535{
536 sector_t prev_first_sect, prev_size; /* previous meta location */
537 sector_t la_size;
538 sector_t size;
539 char ppb[10];
540
541 int md_moved, la_size_changed;
542 enum determine_dev_size rv = unchanged;
543
544 /* race:
545 * application request passes inc_ap_bio,
546 * but then cannot get an AL-reference.
547 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
548 *
549 * to avoid that:
550 * Suspend IO right here.
551 * still lock the act_log to not trigger ASSERTs there.
552 */
553 drbd_suspend_io(mdev);
554
555 /* no wait necessary anymore, actually we could assert that */
556 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
557
558 prev_first_sect = drbd_md_first_sector(mdev->ldev);
559 prev_size = mdev->ldev->md.md_size_sect;
560 la_size = mdev->ldev->md.la_size_sect;
561
562 /* TODO: should only be some assert here, not (re)init... */
563 drbd_md_set_sector_offsets(mdev, mdev->ldev);
564
Philipp Reisnerd8450302010-03-24 15:51:26 +0100565 size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700566
567 if (drbd_get_capacity(mdev->this_bdev) != size ||
568 drbd_bm_capacity(mdev) != size) {
569 int err;
Philipp Reisner02d9a942010-03-24 16:23:03 +0100570 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
Philipp Reisnerb411b362009-09-25 16:07:19 -0700571 if (unlikely(err)) {
572 /* currently there is only one error: ENOMEM! */
573 size = drbd_bm_capacity(mdev)>>1;
574 if (size == 0) {
575 dev_err(DEV, "OUT OF MEMORY! "
576 "Could not allocate bitmap!\n");
577 } else {
578 dev_err(DEV, "BM resizing failed. "
579 "Leaving size unchanged at size = %lu KB\n",
580 (unsigned long)size);
581 }
582 rv = dev_size_error;
583 }
584 /* racy, see comments above. */
585 drbd_set_my_capacity(mdev, size);
586 mdev->ldev->md.la_size_sect = size;
587 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
588 (unsigned long long)size>>1);
589 }
590 if (rv == dev_size_error)
591 goto out;
592
593 la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
594
595 md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
596 || prev_size != mdev->ldev->md.md_size_sect;
597
598 if (la_size_changed || md_moved) {
599 drbd_al_shrink(mdev); /* All extents inactive. */
600 dev_info(DEV, "Writing the whole bitmap, %s\n",
601 la_size_changed && md_moved ? "size changed and md moved" :
602 la_size_changed ? "size changed" : "md moved");
603 rv = drbd_bitmap_io(mdev, &drbd_bm_write, "size changed"); /* does drbd_resume_io() ! */
604 drbd_md_mark_dirty(mdev);
605 }
606
607 if (size > la_size)
608 rv = grew;
609 if (size < la_size)
610 rv = shrunk;
611out:
612 lc_unlock(mdev->act_log);
613 wake_up(&mdev->al_wait);
614 drbd_resume_io(mdev);
615
616 return rv;
617}
618
619sector_t
Philipp Reisnera393db62009-12-22 13:35:52 +0100620drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700621{
622 sector_t p_size = mdev->p_size; /* partner's disk size. */
623 sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
624 sector_t m_size; /* my size */
625 sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
626 sector_t size = 0;
627
628 m_size = drbd_get_max_capacity(bdev);
629
Philipp Reisnera393db62009-12-22 13:35:52 +0100630 if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
631 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
632 p_size = m_size;
633 }
634
Philipp Reisnerb411b362009-09-25 16:07:19 -0700635 if (p_size && m_size) {
636 size = min_t(sector_t, p_size, m_size);
637 } else {
638 if (la_size) {
639 size = la_size;
640 if (m_size && m_size < size)
641 size = m_size;
642 if (p_size && p_size < size)
643 size = p_size;
644 } else {
645 if (m_size)
646 size = m_size;
647 if (p_size)
648 size = p_size;
649 }
650 }
651
652 if (size == 0)
653 dev_err(DEV, "Both nodes diskless!\n");
654
655 if (u_size) {
656 if (u_size > size)
657 dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
658 (unsigned long)u_size>>1, (unsigned long)size>>1);
659 else
660 size = u_size;
661 }
662
663 return size;
664}
665
666/**
667 * drbd_check_al_size() - Ensures that the AL is of the right size
668 * @mdev: DRBD device.
669 *
670 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
671 * failed, and 0 on success. You should call drbd_md_sync() after you called
672 * this function.
673 */
674static int drbd_check_al_size(struct drbd_conf *mdev)
675{
676 struct lru_cache *n, *t;
677 struct lc_element *e;
678 unsigned int in_use;
679 int i;
680
681 ERR_IF(mdev->sync_conf.al_extents < 7)
682 mdev->sync_conf.al_extents = 127;
683
684 if (mdev->act_log &&
685 mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
686 return 0;
687
688 in_use = 0;
689 t = mdev->act_log;
690 n = lc_create("act_log", drbd_al_ext_cache,
691 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
692
693 if (n == NULL) {
694 dev_err(DEV, "Cannot allocate act_log lru!\n");
695 return -ENOMEM;
696 }
697 spin_lock_irq(&mdev->al_lock);
698 if (t) {
699 for (i = 0; i < t->nr_elements; i++) {
700 e = lc_element_by_index(t, i);
701 if (e->refcnt)
702 dev_err(DEV, "refcnt(%d)==%d\n",
703 e->lc_number, e->refcnt);
704 in_use += e->refcnt;
705 }
706 }
707 if (!in_use)
708 mdev->act_log = n;
709 spin_unlock_irq(&mdev->al_lock);
710 if (in_use) {
711 dev_err(DEV, "Activity log still in use!\n");
712 lc_destroy(n);
713 return -EBUSY;
714 } else {
715 if (t)
716 lc_destroy(t);
717 }
718 drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
719 return 0;
720}
721
722void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_seg_s) __must_hold(local)
723{
724 struct request_queue * const q = mdev->rq_queue;
725 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
726 int max_segments = mdev->ldev->dc.max_bio_bvecs;
727
Philipp Reisnerb411b362009-09-25 16:07:19 -0700728 max_seg_s = min(queue_max_sectors(b) * queue_logical_block_size(b), max_seg_s);
729
Martin K. Petersen086fa5f2010-02-26 00:20:38 -0500730 blk_queue_max_hw_sectors(q, max_seg_s >> 9);
Martin K. Petersen8a783622010-02-26 00:20:39 -0500731 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700732 blk_queue_max_segment_size(q, max_seg_s);
733 blk_queue_logical_block_size(q, 512);
734 blk_queue_segment_boundary(q, PAGE_SIZE-1);
735 blk_stack_limits(&q->limits, &b->limits, 0);
736
737 if (b->merge_bvec_fn)
738 dev_warn(DEV, "Backing device's merge_bvec_fn() = %p\n",
739 b->merge_bvec_fn);
740 dev_info(DEV, "max_segment_size ( = BIO size ) = %u\n", queue_max_segment_size(q));
741
742 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
743 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
744 q->backing_dev_info.ra_pages,
745 b->backing_dev_info.ra_pages);
746 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
747 }
748}
749
750/* serialize deconfig (worker exiting, doing cleanup)
751 * and reconfig (drbdsetup disk, drbdsetup net)
752 *
753 * wait for a potentially exiting worker, then restart it,
754 * or start a new one.
755 */
756static void drbd_reconfig_start(struct drbd_conf *mdev)
757{
Lars Ellenberg6c6c7952009-11-16 15:48:54 +0100758 wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
Philipp Reisnerb411b362009-09-25 16:07:19 -0700759 wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
760 drbd_thread_start(&mdev->worker);
761}
762
763/* if still unconfigured, stops worker again.
764 * if configured now, clears CONFIG_PENDING.
765 * wakes potential waiters */
766static void drbd_reconfig_done(struct drbd_conf *mdev)
767{
768 spin_lock_irq(&mdev->req_lock);
769 if (mdev->state.disk == D_DISKLESS &&
770 mdev->state.conn == C_STANDALONE &&
771 mdev->state.role == R_SECONDARY) {
772 set_bit(DEVICE_DYING, &mdev->flags);
773 drbd_thread_stop_nowait(&mdev->worker);
774 } else
775 clear_bit(CONFIG_PENDING, &mdev->flags);
776 spin_unlock_irq(&mdev->req_lock);
777 wake_up(&mdev->state_wait);
778}
779
Philipp Reisner07782862010-08-31 12:00:50 +0200780/* Make sure IO is suspended before calling this function(). */
781static void drbd_suspend_al(struct drbd_conf *mdev)
782{
783 int s = 0;
784
785 if (lc_try_lock(mdev->act_log)) {
786 drbd_al_shrink(mdev);
787 lc_unlock(mdev->act_log);
788 } else {
789 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
790 return;
791 }
792
793 spin_lock_irq(&mdev->req_lock);
794 if (mdev->state.conn < C_CONNECTED)
795 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
796
797 spin_unlock_irq(&mdev->req_lock);
798
799 if (s)
800 dev_info(DEV, "Suspended AL updates\n");
801}
802
Philipp Reisnerb411b362009-09-25 16:07:19 -0700803/* does always return 0;
804 * interesting return code is in reply->ret_code */
805static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
806 struct drbd_nl_cfg_reply *reply)
807{
808 enum drbd_ret_codes retcode;
809 enum determine_dev_size dd;
810 sector_t max_possible_sectors;
811 sector_t min_md_device_sectors;
812 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
813 struct inode *inode, *inode2;
814 struct lru_cache *resync_lru = NULL;
815 union drbd_state ns, os;
816 int rv;
817 int cp_discovered = 0;
818 int logical_block_size;
819
820 drbd_reconfig_start(mdev);
821
822 /* if you want to reconfigure, please tear down first */
823 if (mdev->state.disk > D_DISKLESS) {
824 retcode = ERR_DISK_CONFIGURED;
825 goto fail;
826 }
827
828 /* allocation not in the IO path, cqueue thread context */
829 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
830 if (!nbc) {
831 retcode = ERR_NOMEM;
832 goto fail;
833 }
834
835 nbc->dc.disk_size = DRBD_DISK_SIZE_SECT_DEF;
836 nbc->dc.on_io_error = DRBD_ON_IO_ERROR_DEF;
837 nbc->dc.fencing = DRBD_FENCING_DEF;
838 nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
839
840 if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
841 retcode = ERR_MANDATORY_TAG;
842 goto fail;
843 }
844
845 if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
846 retcode = ERR_MD_IDX_INVALID;
847 goto fail;
848 }
849
Philipp Reisner47ff2d02010-06-18 13:56:57 +0200850 if (get_net_conf(mdev)) {
851 int prot = mdev->net_conf->wire_protocol;
852 put_net_conf(mdev);
853 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
854 retcode = ERR_STONITH_AND_PROT_A;
855 goto fail;
856 }
857 }
858
Philipp Reisnerb411b362009-09-25 16:07:19 -0700859 nbc->lo_file = filp_open(nbc->dc.backing_dev, O_RDWR, 0);
860 if (IS_ERR(nbc->lo_file)) {
861 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
862 PTR_ERR(nbc->lo_file));
863 nbc->lo_file = NULL;
864 retcode = ERR_OPEN_DISK;
865 goto fail;
866 }
867
868 inode = nbc->lo_file->f_dentry->d_inode;
869
870 if (!S_ISBLK(inode->i_mode)) {
871 retcode = ERR_DISK_NOT_BDEV;
872 goto fail;
873 }
874
875 nbc->md_file = filp_open(nbc->dc.meta_dev, O_RDWR, 0);
876 if (IS_ERR(nbc->md_file)) {
877 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
878 PTR_ERR(nbc->md_file));
879 nbc->md_file = NULL;
880 retcode = ERR_OPEN_MD_DISK;
881 goto fail;
882 }
883
884 inode2 = nbc->md_file->f_dentry->d_inode;
885
886 if (!S_ISBLK(inode2->i_mode)) {
887 retcode = ERR_MD_NOT_BDEV;
888 goto fail;
889 }
890
891 nbc->backing_bdev = inode->i_bdev;
892 if (bd_claim(nbc->backing_bdev, mdev)) {
893 printk(KERN_ERR "drbd: bd_claim(%p,%p); failed [%p;%p;%u]\n",
894 nbc->backing_bdev, mdev,
895 nbc->backing_bdev->bd_holder,
896 nbc->backing_bdev->bd_contains->bd_holder,
897 nbc->backing_bdev->bd_holders);
898 retcode = ERR_BDCLAIM_DISK;
899 goto fail;
900 }
901
902 resync_lru = lc_create("resync", drbd_bm_ext_cache,
903 61, sizeof(struct bm_extent),
904 offsetof(struct bm_extent, lce));
905 if (!resync_lru) {
906 retcode = ERR_NOMEM;
907 goto release_bdev_fail;
908 }
909
910 /* meta_dev_idx >= 0: external fixed size,
911 * possibly multiple drbd sharing one meta device.
912 * TODO in that case, paranoia check that [md_bdev, meta_dev_idx] is
913 * not yet used by some other drbd minor!
914 * (if you use drbd.conf + drbdadm,
915 * that should check it for you already; but if you don't, or someone
916 * fooled it, we need to double check here) */
917 nbc->md_bdev = inode2->i_bdev;
918 if (bd_claim(nbc->md_bdev, (nbc->dc.meta_dev_idx < 0) ? (void *)mdev
919 : (void *) drbd_m_holder)) {
920 retcode = ERR_BDCLAIM_MD_DISK;
921 goto release_bdev_fail;
922 }
923
924 if ((nbc->backing_bdev == nbc->md_bdev) !=
925 (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
926 nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
927 retcode = ERR_MD_IDX_INVALID;
928 goto release_bdev2_fail;
929 }
930
931 /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
932 drbd_md_set_sector_offsets(mdev, nbc);
933
934 if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
935 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
936 (unsigned long long) drbd_get_max_capacity(nbc),
937 (unsigned long long) nbc->dc.disk_size);
938 retcode = ERR_DISK_TO_SMALL;
939 goto release_bdev2_fail;
940 }
941
942 if (nbc->dc.meta_dev_idx < 0) {
943 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
944 /* at least one MB, otherwise it does not make sense */
945 min_md_device_sectors = (2<<10);
946 } else {
947 max_possible_sectors = DRBD_MAX_SECTORS;
948 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
949 }
950
Philipp Reisnerb411b362009-09-25 16:07:19 -0700951 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
952 retcode = ERR_MD_DISK_TO_SMALL;
953 dev_warn(DEV, "refusing attach: md-device too small, "
954 "at least %llu sectors needed for this meta-disk type\n",
955 (unsigned long long) min_md_device_sectors);
956 goto release_bdev2_fail;
957 }
958
959 /* Make sure the new disk is big enough
960 * (we may currently be R_PRIMARY with no local disk...) */
961 if (drbd_get_max_capacity(nbc) <
962 drbd_get_capacity(mdev->this_bdev)) {
963 retcode = ERR_DISK_TO_SMALL;
964 goto release_bdev2_fail;
965 }
966
967 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
968
Lars Ellenberg13529942009-10-12 19:07:49 +0200969 if (nbc->known_size > max_possible_sectors) {
970 dev_warn(DEV, "==> truncating very big lower level device "
971 "to currently maximum possible %llu sectors <==\n",
972 (unsigned long long) max_possible_sectors);
973 if (nbc->dc.meta_dev_idx >= 0)
974 dev_warn(DEV, "==>> using internal or flexible "
975 "meta data may help <<==\n");
976 }
977
Philipp Reisnerb411b362009-09-25 16:07:19 -0700978 drbd_suspend_io(mdev);
979 /* also wait for the last barrier ack. */
Philipp Reisner8f488152010-06-24 12:05:53 +0200980 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || mdev->state.susp);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700981 /* and for any other previously queued work */
982 drbd_flush_workqueue(mdev);
983
984 retcode = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
985 drbd_resume_io(mdev);
986 if (retcode < SS_SUCCESS)
987 goto release_bdev2_fail;
988
989 if (!get_ldev_if_state(mdev, D_ATTACHING))
990 goto force_diskless;
991
992 drbd_md_set_sector_offsets(mdev, nbc);
993
Lars Ellenberg4aa83b72010-02-26 16:53:24 +0100994 /* allocate a second IO page if logical_block_size != 512 */
995 logical_block_size = bdev_logical_block_size(nbc->md_bdev);
996 if (logical_block_size == 0)
997 logical_block_size = MD_SECTOR_SIZE;
998
999 if (logical_block_size != MD_SECTOR_SIZE) {
1000 if (!mdev->md_io_tmpp) {
1001 struct page *page = alloc_page(GFP_NOIO);
1002 if (!page)
1003 goto force_diskless_dec;
1004
1005 dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
1006 logical_block_size, MD_SECTOR_SIZE);
1007 dev_warn(DEV, "Workaround engaged (has performance impact).\n");
1008
1009 mdev->md_io_tmpp = page;
1010 }
1011 }
1012
Philipp Reisnerb411b362009-09-25 16:07:19 -07001013 if (!mdev->bitmap) {
1014 if (drbd_bm_init(mdev)) {
1015 retcode = ERR_NOMEM;
1016 goto force_diskless_dec;
1017 }
1018 }
1019
1020 retcode = drbd_md_read(mdev, nbc);
1021 if (retcode != NO_ERROR)
1022 goto force_diskless_dec;
1023
1024 if (mdev->state.conn < C_CONNECTED &&
1025 mdev->state.role == R_PRIMARY &&
1026 (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1027 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1028 (unsigned long long)mdev->ed_uuid);
1029 retcode = ERR_DATA_NOT_CURRENT;
1030 goto force_diskless_dec;
1031 }
1032
1033 /* Since we are diskless, fix the activity log first... */
1034 if (drbd_check_al_size(mdev)) {
1035 retcode = ERR_NOMEM;
1036 goto force_diskless_dec;
1037 }
1038
1039 /* Prevent shrinking of consistent devices ! */
1040 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
Philipp Reisnera393db62009-12-22 13:35:52 +01001041 drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001042 dev_warn(DEV, "refusing to truncate a consistent device\n");
1043 retcode = ERR_DISK_TO_SMALL;
1044 goto force_diskless_dec;
1045 }
1046
1047 if (!drbd_al_read_log(mdev, nbc)) {
1048 retcode = ERR_IO_MD_DISK;
1049 goto force_diskless_dec;
1050 }
1051
Philipp Reisnerb411b362009-09-25 16:07:19 -07001052 /* Reset the "barriers don't work" bits here, then force meta data to
1053 * be written, to ensure we determine if barriers are supported. */
1054 if (nbc->dc.no_md_flush)
1055 set_bit(MD_NO_BARRIER, &mdev->flags);
1056 else
1057 clear_bit(MD_NO_BARRIER, &mdev->flags);
1058
1059 /* Point of no return reached.
1060 * Devices and memory are no longer released by error cleanup below.
1061 * now mdev takes over responsibility, and the state engine should
1062 * clean it up somewhere. */
1063 D_ASSERT(mdev->ldev == NULL);
1064 mdev->ldev = nbc;
1065 mdev->resync = resync_lru;
1066 nbc = NULL;
1067 resync_lru = NULL;
1068
1069 mdev->write_ordering = WO_bio_barrier;
1070 drbd_bump_write_ordering(mdev, WO_bio_barrier);
1071
1072 if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1073 set_bit(CRASHED_PRIMARY, &mdev->flags);
1074 else
1075 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1076
Philipp Reisner894c6a92010-06-18 16:03:20 +02001077 if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1078 !(mdev->state.role == R_PRIMARY && mdev->state.susp &&
1079 mdev->sync_conf.on_no_data == OND_SUSPEND_IO)) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001080 set_bit(CRASHED_PRIMARY, &mdev->flags);
1081 cp_discovered = 1;
1082 }
1083
1084 mdev->send_cnt = 0;
1085 mdev->recv_cnt = 0;
1086 mdev->read_cnt = 0;
1087 mdev->writ_cnt = 0;
1088
Philipp Reisnerd5373382010-08-23 15:18:33 +02001089 drbd_setup_queue_param(mdev, mdev->state.conn == C_CONNECTED &&
1090 mdev->agreed_pro_version < 95 ?
1091 DRBD_MAX_SIZE_H80_PACKET : DRBD_MAX_SEGMENT_SIZE);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001092
1093 /* If I am currently not R_PRIMARY,
1094 * but meta data primary indicator is set,
1095 * I just now recover from a hard crash,
1096 * and have been R_PRIMARY before that crash.
1097 *
1098 * Now, if I had no connection before that crash
1099 * (have been degraded R_PRIMARY), chances are that
1100 * I won't find my peer now either.
1101 *
1102 * In that case, and _only_ in that case,
1103 * we use the degr-wfc-timeout instead of the default,
1104 * so we can automatically recover from a crash of a
1105 * degraded but active "cluster" after a certain timeout.
1106 */
1107 clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1108 if (mdev->state.role != R_PRIMARY &&
1109 drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1110 !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1111 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1112
Philipp Reisnera393db62009-12-22 13:35:52 +01001113 dd = drbd_determin_dev_size(mdev, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001114 if (dd == dev_size_error) {
1115 retcode = ERR_NOMEM_BITMAP;
1116 goto force_diskless_dec;
1117 } else if (dd == grew)
1118 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1119
1120 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1121 dev_info(DEV, "Assuming that all blocks are out of sync "
1122 "(aka FullSync)\n");
1123 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from attaching")) {
1124 retcode = ERR_IO_MD_DISK;
1125 goto force_diskless_dec;
1126 }
1127 } else {
1128 if (drbd_bitmap_io(mdev, &drbd_bm_read, "read from attaching") < 0) {
1129 retcode = ERR_IO_MD_DISK;
1130 goto force_diskless_dec;
1131 }
1132 }
1133
1134 if (cp_discovered) {
1135 drbd_al_apply_to_bm(mdev);
1136 drbd_al_to_on_disk_bm(mdev);
1137 }
1138
Philipp Reisner07782862010-08-31 12:00:50 +02001139 if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1140 drbd_suspend_al(mdev); /* IO is still suspended here... */
1141
Philipp Reisnerb411b362009-09-25 16:07:19 -07001142 spin_lock_irq(&mdev->req_lock);
1143 os = mdev->state;
1144 ns.i = os.i;
1145 /* If MDF_CONSISTENT is not set go into inconsistent state,
1146 otherwise investigate MDF_WasUpToDate...
1147 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1148 otherwise into D_CONSISTENT state.
1149 */
1150 if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1151 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1152 ns.disk = D_CONSISTENT;
1153 else
1154 ns.disk = D_OUTDATED;
1155 } else {
1156 ns.disk = D_INCONSISTENT;
1157 }
1158
1159 if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1160 ns.pdsk = D_OUTDATED;
1161
1162 if ( ns.disk == D_CONSISTENT &&
1163 (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1164 ns.disk = D_UP_TO_DATE;
1165
1166 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1167 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1168 this point, because drbd_request_state() modifies these
1169 flags. */
1170
1171 /* In case we are C_CONNECTED postpone any decision on the new disk
1172 state after the negotiation phase. */
1173 if (mdev->state.conn == C_CONNECTED) {
1174 mdev->new_state_tmp.i = ns.i;
1175 ns.i = os.i;
1176 ns.disk = D_NEGOTIATING;
Philipp Reisnerdc66c742010-06-02 14:31:29 +02001177
1178 /* We expect to receive up-to-date UUIDs soon.
1179 To avoid a race in receive_state, free p_uuid while
1180 holding req_lock. I.e. atomic with the state change */
1181 kfree(mdev->p_uuid);
1182 mdev->p_uuid = NULL;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001183 }
1184
1185 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1186 ns = mdev->state;
1187 spin_unlock_irq(&mdev->req_lock);
1188
1189 if (rv < SS_SUCCESS)
1190 goto force_diskless_dec;
1191
1192 if (mdev->state.role == R_PRIMARY)
1193 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
1194 else
1195 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1196
1197 drbd_md_mark_dirty(mdev);
1198 drbd_md_sync(mdev);
1199
1200 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1201 put_ldev(mdev);
1202 reply->ret_code = retcode;
1203 drbd_reconfig_done(mdev);
1204 return 0;
1205
1206 force_diskless_dec:
1207 put_ldev(mdev);
1208 force_diskless:
1209 drbd_force_state(mdev, NS(disk, D_DISKLESS));
1210 drbd_md_sync(mdev);
1211 release_bdev2_fail:
1212 if (nbc)
1213 bd_release(nbc->md_bdev);
1214 release_bdev_fail:
1215 if (nbc)
1216 bd_release(nbc->backing_bdev);
1217 fail:
1218 if (nbc) {
1219 if (nbc->lo_file)
1220 fput(nbc->lo_file);
1221 if (nbc->md_file)
1222 fput(nbc->md_file);
1223 kfree(nbc);
1224 }
1225 lc_destroy(resync_lru);
1226
1227 reply->ret_code = retcode;
1228 drbd_reconfig_done(mdev);
1229 return 0;
1230}
1231
1232static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1233 struct drbd_nl_cfg_reply *reply)
1234{
1235 reply->ret_code = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1236 return 0;
1237}
1238
1239static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1240 struct drbd_nl_cfg_reply *reply)
1241{
1242 int i, ns;
1243 enum drbd_ret_codes retcode;
1244 struct net_conf *new_conf = NULL;
1245 struct crypto_hash *tfm = NULL;
1246 struct crypto_hash *integrity_w_tfm = NULL;
1247 struct crypto_hash *integrity_r_tfm = NULL;
1248 struct hlist_head *new_tl_hash = NULL;
1249 struct hlist_head *new_ee_hash = NULL;
1250 struct drbd_conf *odev;
1251 char hmac_name[CRYPTO_MAX_ALG_NAME];
1252 void *int_dig_out = NULL;
1253 void *int_dig_in = NULL;
1254 void *int_dig_vv = NULL;
1255 struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1256
1257 drbd_reconfig_start(mdev);
1258
1259 if (mdev->state.conn > C_STANDALONE) {
1260 retcode = ERR_NET_CONFIGURED;
1261 goto fail;
1262 }
1263
1264 /* allocation not in the IO path, cqueue thread context */
Julia Lawall2db4e422010-05-13 22:02:21 +02001265 new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001266 if (!new_conf) {
1267 retcode = ERR_NOMEM;
1268 goto fail;
1269 }
1270
Philipp Reisnerb411b362009-09-25 16:07:19 -07001271 new_conf->timeout = DRBD_TIMEOUT_DEF;
1272 new_conf->try_connect_int = DRBD_CONNECT_INT_DEF;
1273 new_conf->ping_int = DRBD_PING_INT_DEF;
1274 new_conf->max_epoch_size = DRBD_MAX_EPOCH_SIZE_DEF;
1275 new_conf->max_buffers = DRBD_MAX_BUFFERS_DEF;
1276 new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1277 new_conf->sndbuf_size = DRBD_SNDBUF_SIZE_DEF;
1278 new_conf->rcvbuf_size = DRBD_RCVBUF_SIZE_DEF;
1279 new_conf->ko_count = DRBD_KO_COUNT_DEF;
1280 new_conf->after_sb_0p = DRBD_AFTER_SB_0P_DEF;
1281 new_conf->after_sb_1p = DRBD_AFTER_SB_1P_DEF;
1282 new_conf->after_sb_2p = DRBD_AFTER_SB_2P_DEF;
1283 new_conf->want_lose = 0;
1284 new_conf->two_primaries = 0;
1285 new_conf->wire_protocol = DRBD_PROT_C;
1286 new_conf->ping_timeo = DRBD_PING_TIMEO_DEF;
1287 new_conf->rr_conflict = DRBD_RR_CONFLICT_DEF;
1288
1289 if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1290 retcode = ERR_MANDATORY_TAG;
1291 goto fail;
1292 }
1293
1294 if (new_conf->two_primaries
1295 && (new_conf->wire_protocol != DRBD_PROT_C)) {
1296 retcode = ERR_NOT_PROTO_C;
1297 goto fail;
Philipp Reisner47ff2d02010-06-18 13:56:57 +02001298 }
1299
1300 if (get_ldev(mdev)) {
1301 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1302 put_ldev(mdev);
1303 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1304 retcode = ERR_STONITH_AND_PROT_A;
1305 goto fail;
1306 }
1307 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07001308
1309 if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1310 retcode = ERR_DISCARD;
1311 goto fail;
1312 }
1313
1314 retcode = NO_ERROR;
1315
1316 new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1317 new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1318 for (i = 0; i < minor_count; i++) {
1319 odev = minor_to_mdev(i);
1320 if (!odev || odev == mdev)
1321 continue;
1322 if (get_net_conf(odev)) {
1323 taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1324 if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1325 !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1326 retcode = ERR_LOCAL_ADDR;
1327
1328 taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1329 if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1330 !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1331 retcode = ERR_PEER_ADDR;
1332
1333 put_net_conf(odev);
1334 if (retcode != NO_ERROR)
1335 goto fail;
1336 }
1337 }
1338
1339 if (new_conf->cram_hmac_alg[0] != 0) {
1340 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1341 new_conf->cram_hmac_alg);
1342 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1343 if (IS_ERR(tfm)) {
1344 tfm = NULL;
1345 retcode = ERR_AUTH_ALG;
1346 goto fail;
1347 }
1348
Philipp Reisner07982192009-12-28 16:58:38 +01001349 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001350 retcode = ERR_AUTH_ALG_ND;
1351 goto fail;
1352 }
1353 }
1354
1355 if (new_conf->integrity_alg[0]) {
1356 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1357 if (IS_ERR(integrity_w_tfm)) {
1358 integrity_w_tfm = NULL;
1359 retcode=ERR_INTEGRITY_ALG;
1360 goto fail;
1361 }
1362
1363 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1364 retcode=ERR_INTEGRITY_ALG_ND;
1365 goto fail;
1366 }
1367
1368 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1369 if (IS_ERR(integrity_r_tfm)) {
1370 integrity_r_tfm = NULL;
1371 retcode=ERR_INTEGRITY_ALG;
1372 goto fail;
1373 }
1374 }
1375
1376 ns = new_conf->max_epoch_size/8;
1377 if (mdev->tl_hash_s != ns) {
1378 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1379 if (!new_tl_hash) {
1380 retcode = ERR_NOMEM;
1381 goto fail;
1382 }
1383 }
1384
1385 ns = new_conf->max_buffers/8;
1386 if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1387 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1388 if (!new_ee_hash) {
1389 retcode = ERR_NOMEM;
1390 goto fail;
1391 }
1392 }
1393
1394 ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1395
1396 if (integrity_w_tfm) {
1397 i = crypto_hash_digestsize(integrity_w_tfm);
1398 int_dig_out = kmalloc(i, GFP_KERNEL);
1399 if (!int_dig_out) {
1400 retcode = ERR_NOMEM;
1401 goto fail;
1402 }
1403 int_dig_in = kmalloc(i, GFP_KERNEL);
1404 if (!int_dig_in) {
1405 retcode = ERR_NOMEM;
1406 goto fail;
1407 }
1408 int_dig_vv = kmalloc(i, GFP_KERNEL);
1409 if (!int_dig_vv) {
1410 retcode = ERR_NOMEM;
1411 goto fail;
1412 }
1413 }
1414
1415 if (!mdev->bitmap) {
1416 if(drbd_bm_init(mdev)) {
1417 retcode = ERR_NOMEM;
1418 goto fail;
1419 }
1420 }
1421
Philipp Reisnerf70b35112010-06-24 14:34:40 +02001422 drbd_flush_workqueue(mdev);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001423 spin_lock_irq(&mdev->req_lock);
1424 if (mdev->net_conf != NULL) {
1425 retcode = ERR_NET_CONFIGURED;
1426 spin_unlock_irq(&mdev->req_lock);
1427 goto fail;
1428 }
1429 mdev->net_conf = new_conf;
1430
1431 mdev->send_cnt = 0;
1432 mdev->recv_cnt = 0;
1433
1434 if (new_tl_hash) {
1435 kfree(mdev->tl_hash);
1436 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1437 mdev->tl_hash = new_tl_hash;
1438 }
1439
1440 if (new_ee_hash) {
1441 kfree(mdev->ee_hash);
1442 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1443 mdev->ee_hash = new_ee_hash;
1444 }
1445
1446 crypto_free_hash(mdev->cram_hmac_tfm);
1447 mdev->cram_hmac_tfm = tfm;
1448
1449 crypto_free_hash(mdev->integrity_w_tfm);
1450 mdev->integrity_w_tfm = integrity_w_tfm;
1451
1452 crypto_free_hash(mdev->integrity_r_tfm);
1453 mdev->integrity_r_tfm = integrity_r_tfm;
1454
1455 kfree(mdev->int_dig_out);
1456 kfree(mdev->int_dig_in);
1457 kfree(mdev->int_dig_vv);
1458 mdev->int_dig_out=int_dig_out;
1459 mdev->int_dig_in=int_dig_in;
1460 mdev->int_dig_vv=int_dig_vv;
Philipp Reisnerf70b35112010-06-24 14:34:40 +02001461 retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001462 spin_unlock_irq(&mdev->req_lock);
1463
Philipp Reisnerb411b362009-09-25 16:07:19 -07001464 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1465 reply->ret_code = retcode;
1466 drbd_reconfig_done(mdev);
1467 return 0;
1468
1469fail:
1470 kfree(int_dig_out);
1471 kfree(int_dig_in);
1472 kfree(int_dig_vv);
1473 crypto_free_hash(tfm);
1474 crypto_free_hash(integrity_w_tfm);
1475 crypto_free_hash(integrity_r_tfm);
1476 kfree(new_tl_hash);
1477 kfree(new_ee_hash);
1478 kfree(new_conf);
1479
1480 reply->ret_code = retcode;
1481 drbd_reconfig_done(mdev);
1482 return 0;
1483}
1484
1485static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1486 struct drbd_nl_cfg_reply *reply)
1487{
1488 int retcode;
1489
1490 retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1491
1492 if (retcode == SS_NOTHING_TO_DO)
1493 goto done;
1494 else if (retcode == SS_ALREADY_STANDALONE)
1495 goto done;
1496 else if (retcode == SS_PRIMARY_NOP) {
1497 /* Our statche checking code wants to see the peer outdated. */
1498 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1499 pdsk, D_OUTDATED));
1500 } else if (retcode == SS_CW_FAILED_BY_PEER) {
1501 /* The peer probably wants to see us outdated. */
1502 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1503 disk, D_OUTDATED),
1504 CS_ORDERED);
1505 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1506 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1507 retcode = SS_SUCCESS;
1508 }
1509 }
1510
1511 if (retcode < SS_SUCCESS)
1512 goto fail;
1513
1514 if (wait_event_interruptible(mdev->state_wait,
1515 mdev->state.conn != C_DISCONNECTING)) {
1516 /* Do not test for mdev->state.conn == C_STANDALONE, since
1517 someone else might connect us in the mean time! */
1518 retcode = ERR_INTR;
1519 goto fail;
1520 }
1521
1522 done:
1523 retcode = NO_ERROR;
1524 fail:
1525 drbd_md_sync(mdev);
1526 reply->ret_code = retcode;
1527 return 0;
1528}
1529
1530void resync_after_online_grow(struct drbd_conf *mdev)
1531{
1532 int iass; /* I am sync source */
1533
1534 dev_info(DEV, "Resync of new storage after online grow\n");
1535 if (mdev->state.role != mdev->state.peer)
1536 iass = (mdev->state.role == R_PRIMARY);
1537 else
1538 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1539
1540 if (iass)
1541 drbd_start_resync(mdev, C_SYNC_SOURCE);
1542 else
1543 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1544}
1545
1546static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1547 struct drbd_nl_cfg_reply *reply)
1548{
1549 struct resize rs;
1550 int retcode = NO_ERROR;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001551 enum determine_dev_size dd;
Philipp Reisner6495d2c2010-03-24 16:07:04 +01001552 enum dds_flags ddsf;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001553
1554 memset(&rs, 0, sizeof(struct resize));
1555 if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1556 retcode = ERR_MANDATORY_TAG;
1557 goto fail;
1558 }
1559
1560 if (mdev->state.conn > C_CONNECTED) {
1561 retcode = ERR_RESIZE_RESYNC;
1562 goto fail;
1563 }
1564
1565 if (mdev->state.role == R_SECONDARY &&
1566 mdev->state.peer == R_SECONDARY) {
1567 retcode = ERR_NO_PRIMARY;
1568 goto fail;
1569 }
1570
1571 if (!get_ldev(mdev)) {
1572 retcode = ERR_NO_DISK;
1573 goto fail;
1574 }
1575
Philipp Reisner6495d2c2010-03-24 16:07:04 +01001576 if (rs.no_resync && mdev->agreed_pro_version < 93) {
1577 retcode = ERR_NEED_APV_93;
1578 goto fail;
1579 }
1580
Philipp Reisner087c2492010-03-26 13:49:56 +01001581 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
Philipp Reisnerb411b362009-09-25 16:07:19 -07001582 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001583
1584 mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
Philipp Reisner6495d2c2010-03-24 16:07:04 +01001585 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1586 dd = drbd_determin_dev_size(mdev, ddsf);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001587 drbd_md_sync(mdev);
1588 put_ldev(mdev);
1589 if (dd == dev_size_error) {
1590 retcode = ERR_NOMEM_BITMAP;
1591 goto fail;
1592 }
1593
Philipp Reisner087c2492010-03-26 13:49:56 +01001594 if (mdev->state.conn == C_CONNECTED) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001595 if (dd == grew)
1596 set_bit(RESIZE_PENDING, &mdev->flags);
1597
1598 drbd_send_uuids(mdev);
Philipp Reisner6495d2c2010-03-24 16:07:04 +01001599 drbd_send_sizes(mdev, 1, ddsf);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001600 }
1601
1602 fail:
1603 reply->ret_code = retcode;
1604 return 0;
1605}
1606
1607static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1608 struct drbd_nl_cfg_reply *reply)
1609{
1610 int retcode = NO_ERROR;
1611 int err;
1612 int ovr; /* online verify running */
1613 int rsr; /* re-sync running */
1614 struct crypto_hash *verify_tfm = NULL;
1615 struct crypto_hash *csums_tfm = NULL;
1616 struct syncer_conf sc;
1617 cpumask_var_t new_cpu_mask;
Philipp Reisner778f2712010-07-06 11:14:00 +02001618 int *rs_plan_s = NULL;
1619 int fifo_size;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001620
1621 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1622 retcode = ERR_NOMEM;
1623 goto fail;
1624 }
1625
1626 if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1627 memset(&sc, 0, sizeof(struct syncer_conf));
1628 sc.rate = DRBD_RATE_DEF;
1629 sc.after = DRBD_AFTER_DEF;
1630 sc.al_extents = DRBD_AL_EXTENTS_DEF;
Philipp Reisner265be2d2010-05-31 10:14:17 +02001631 sc.on_no_data = DRBD_ON_NO_DATA_DEF;
Philipp Reisner9a31d712010-07-05 13:42:03 +02001632 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1633 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1634 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
Lars Ellenberg0f0601f2010-08-11 23:40:24 +02001635 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1636 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001637 } else
1638 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1639
1640 if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1641 retcode = ERR_MANDATORY_TAG;
1642 goto fail;
1643 }
1644
1645 /* re-sync running */
1646 rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1647 mdev->state.conn == C_SYNC_TARGET ||
1648 mdev->state.conn == C_PAUSED_SYNC_S ||
1649 mdev->state.conn == C_PAUSED_SYNC_T );
1650
1651 if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1652 retcode = ERR_CSUMS_RESYNC_RUNNING;
1653 goto fail;
1654 }
1655
1656 if (!rsr && sc.csums_alg[0]) {
1657 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1658 if (IS_ERR(csums_tfm)) {
1659 csums_tfm = NULL;
1660 retcode = ERR_CSUMS_ALG;
1661 goto fail;
1662 }
1663
1664 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1665 retcode = ERR_CSUMS_ALG_ND;
1666 goto fail;
1667 }
1668 }
1669
1670 /* online verify running */
1671 ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1672
1673 if (ovr) {
1674 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1675 retcode = ERR_VERIFY_RUNNING;
1676 goto fail;
1677 }
1678 }
1679
1680 if (!ovr && sc.verify_alg[0]) {
1681 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1682 if (IS_ERR(verify_tfm)) {
1683 verify_tfm = NULL;
1684 retcode = ERR_VERIFY_ALG;
1685 goto fail;
1686 }
1687
1688 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1689 retcode = ERR_VERIFY_ALG_ND;
1690 goto fail;
1691 }
1692 }
1693
1694 /* silently ignore cpu mask on UP kernel */
1695 if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1696 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1697 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1698 if (err) {
1699 dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1700 retcode = ERR_CPU_MASK_PARSE;
1701 goto fail;
1702 }
1703 }
1704
1705 ERR_IF (sc.rate < 1) sc.rate = 1;
1706 ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1707#define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1708 if (sc.al_extents > AL_MAX) {
1709 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1710 sc.al_extents = AL_MAX;
1711 }
1712#undef AL_MAX
1713
1714 /* most sanity checks done, try to assign the new sync-after
1715 * dependency. need to hold the global lock in there,
1716 * to avoid a race in the dependency loop check. */
1717 retcode = drbd_alter_sa(mdev, sc.after);
1718 if (retcode != NO_ERROR)
1719 goto fail;
1720
Philipp Reisner778f2712010-07-06 11:14:00 +02001721 fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1722 if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1723 rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1724 if (!rs_plan_s) {
1725 dev_err(DEV, "kmalloc of fifo_buffer failed");
1726 retcode = ERR_NOMEM;
1727 goto fail;
1728 }
1729 }
1730
Philipp Reisnerb411b362009-09-25 16:07:19 -07001731 /* ok, assign the rest of it as well.
1732 * lock against receive_SyncParam() */
1733 spin_lock(&mdev->peer_seq_lock);
1734 mdev->sync_conf = sc;
1735
1736 if (!rsr) {
1737 crypto_free_hash(mdev->csums_tfm);
1738 mdev->csums_tfm = csums_tfm;
1739 csums_tfm = NULL;
1740 }
1741
1742 if (!ovr) {
1743 crypto_free_hash(mdev->verify_tfm);
1744 mdev->verify_tfm = verify_tfm;
1745 verify_tfm = NULL;
1746 }
Philipp Reisner778f2712010-07-06 11:14:00 +02001747
1748 if (fifo_size != mdev->rs_plan_s.size) {
1749 kfree(mdev->rs_plan_s.values);
1750 mdev->rs_plan_s.values = rs_plan_s;
1751 mdev->rs_plan_s.size = fifo_size;
1752 mdev->rs_planed = 0;
1753 rs_plan_s = NULL;
1754 }
1755
Philipp Reisnerb411b362009-09-25 16:07:19 -07001756 spin_unlock(&mdev->peer_seq_lock);
1757
1758 if (get_ldev(mdev)) {
1759 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1760 drbd_al_shrink(mdev);
1761 err = drbd_check_al_size(mdev);
1762 lc_unlock(mdev->act_log);
1763 wake_up(&mdev->al_wait);
1764
1765 put_ldev(mdev);
1766 drbd_md_sync(mdev);
1767
1768 if (err) {
1769 retcode = ERR_NOMEM;
1770 goto fail;
1771 }
1772 }
1773
1774 if (mdev->state.conn >= C_CONNECTED)
1775 drbd_send_sync_param(mdev, &sc);
1776
1777 if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1778 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1779 drbd_calc_cpu_mask(mdev);
1780 mdev->receiver.reset_cpu_mask = 1;
1781 mdev->asender.reset_cpu_mask = 1;
1782 mdev->worker.reset_cpu_mask = 1;
1783 }
1784
1785 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1786fail:
Philipp Reisner778f2712010-07-06 11:14:00 +02001787 kfree(rs_plan_s);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001788 free_cpumask_var(new_cpu_mask);
1789 crypto_free_hash(csums_tfm);
1790 crypto_free_hash(verify_tfm);
1791 reply->ret_code = retcode;
1792 return 0;
1793}
1794
1795static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1796 struct drbd_nl_cfg_reply *reply)
1797{
1798 int retcode;
1799
1800 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1801
1802 if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1803 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1804
1805 while (retcode == SS_NEED_CONNECTION) {
1806 spin_lock_irq(&mdev->req_lock);
1807 if (mdev->state.conn < C_CONNECTED)
1808 retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1809 spin_unlock_irq(&mdev->req_lock);
1810
1811 if (retcode != SS_NEED_CONNECTION)
1812 break;
1813
1814 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1815 }
1816
1817 reply->ret_code = retcode;
1818 return 0;
1819}
1820
Philipp Reisner07782862010-08-31 12:00:50 +02001821static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
1822{
1823 int rv;
1824
1825 rv = drbd_bmio_set_n_write(mdev);
1826 drbd_suspend_al(mdev);
1827 return rv;
1828}
1829
Philipp Reisnerb411b362009-09-25 16:07:19 -07001830static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1831 struct drbd_nl_cfg_reply *reply)
1832{
Philipp Reisner07782862010-08-31 12:00:50 +02001833 int retcode;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001834
Philipp Reisner07782862010-08-31 12:00:50 +02001835 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001836
Philipp Reisner07782862010-08-31 12:00:50 +02001837 if (retcode < SS_SUCCESS) {
1838 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
1839 /* The peer will get a resync upon connect anyways. Just make that
1840 into a full resync. */
1841 retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
1842 if (retcode >= SS_SUCCESS) {
1843 /* open coded drbd_bitmap_io() */
1844 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
1845 "set_n_write from invalidate_peer"))
1846 retcode = ERR_IO_MD_DISK;
1847 }
1848 } else
1849 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
1850 }
1851
1852 reply->ret_code = retcode;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001853 return 0;
1854}
1855
1856static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1857 struct drbd_nl_cfg_reply *reply)
1858{
1859 int retcode = NO_ERROR;
1860
1861 if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
1862 retcode = ERR_PAUSE_IS_SET;
1863
1864 reply->ret_code = retcode;
1865 return 0;
1866}
1867
1868static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1869 struct drbd_nl_cfg_reply *reply)
1870{
1871 int retcode = NO_ERROR;
1872
1873 if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO)
1874 retcode = ERR_PAUSE_IS_CLEAR;
1875
1876 reply->ret_code = retcode;
1877 return 0;
1878}
1879
1880static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1881 struct drbd_nl_cfg_reply *reply)
1882{
1883 reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
1884
1885 return 0;
1886}
1887
1888static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1889 struct drbd_nl_cfg_reply *reply)
1890{
Philipp Reisner43a51822010-06-11 11:26:34 +02001891 if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
1892 drbd_uuid_new_current(mdev);
1893 clear_bit(NEW_CUR_UUID, &mdev->flags);
1894 drbd_md_sync(mdev);
1895 }
Philipp Reisner265be2d2010-05-31 10:14:17 +02001896 drbd_suspend_io(mdev);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001897 reply->ret_code = drbd_request_state(mdev, NS(susp, 0));
Philipp Reisner265be2d2010-05-31 10:14:17 +02001898 if (reply->ret_code == SS_SUCCESS) {
1899 if (mdev->state.conn < C_CONNECTED)
1900 tl_clear(mdev);
1901 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
1902 tl_restart(mdev, fail_frozen_disk_io);
1903 }
1904 drbd_resume_io(mdev);
1905
Philipp Reisnerb411b362009-09-25 16:07:19 -07001906 return 0;
1907}
1908
1909static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1910 struct drbd_nl_cfg_reply *reply)
1911{
1912 reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
1913 return 0;
1914}
1915
1916static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1917 struct drbd_nl_cfg_reply *reply)
1918{
1919 unsigned short *tl;
1920
1921 tl = reply->tag_list;
1922
1923 if (get_ldev(mdev)) {
1924 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
1925 put_ldev(mdev);
1926 }
1927
1928 if (get_net_conf(mdev)) {
1929 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
1930 put_net_conf(mdev);
1931 }
1932 tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
1933
1934 put_unaligned(TT_END, tl++); /* Close the tag list */
1935
1936 return (int)((char *)tl - (char *)reply->tag_list);
1937}
1938
1939static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1940 struct drbd_nl_cfg_reply *reply)
1941{
1942 unsigned short *tl = reply->tag_list;
1943 union drbd_state s = mdev->state;
1944 unsigned long rs_left;
1945 unsigned int res;
1946
1947 tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
1948
1949 /* no local ref, no bitmap, no syncer progress. */
1950 if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
1951 if (get_ldev(mdev)) {
1952 drbd_get_syncer_progress(mdev, &rs_left, &res);
1953 tl = tl_add_int(tl, T_sync_progress, &res);
1954 put_ldev(mdev);
1955 }
1956 }
1957 put_unaligned(TT_END, tl++); /* Close the tag list */
1958
1959 return (int)((char *)tl - (char *)reply->tag_list);
1960}
1961
1962static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1963 struct drbd_nl_cfg_reply *reply)
1964{
1965 unsigned short *tl;
1966
1967 tl = reply->tag_list;
1968
1969 if (get_ldev(mdev)) {
1970 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
1971 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
1972 put_ldev(mdev);
1973 }
1974 put_unaligned(TT_END, tl++); /* Close the tag list */
1975
1976 return (int)((char *)tl - (char *)reply->tag_list);
1977}
1978
1979/**
1980 * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
1981 * @mdev: DRBD device.
1982 * @nlp: Netlink/connector packet from drbdsetup
1983 * @reply: Reply packet for drbdsetup
1984 */
1985static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1986 struct drbd_nl_cfg_reply *reply)
1987{
1988 unsigned short *tl;
1989 char rv;
1990
1991 tl = reply->tag_list;
1992
1993 rv = mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
1994 test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
1995
1996 tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
1997 put_unaligned(TT_END, tl++); /* Close the tag list */
1998
1999 return (int)((char *)tl - (char *)reply->tag_list);
2000}
2001
2002static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2003 struct drbd_nl_cfg_reply *reply)
2004{
2005 /* default to resume from last known position, if possible */
2006 struct start_ov args =
2007 { .start_sector = mdev->ov_start_sector };
2008
2009 if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2010 reply->ret_code = ERR_MANDATORY_TAG;
2011 return 0;
2012 }
2013 /* w_make_ov_request expects position to be aligned */
2014 mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
2015 reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2016 return 0;
2017}
2018
2019
2020static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2021 struct drbd_nl_cfg_reply *reply)
2022{
2023 int retcode = NO_ERROR;
2024 int skip_initial_sync = 0;
2025 int err;
2026
2027 struct new_c_uuid args;
2028
2029 memset(&args, 0, sizeof(struct new_c_uuid));
2030 if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2031 reply->ret_code = ERR_MANDATORY_TAG;
2032 return 0;
2033 }
2034
2035 mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
2036
2037 if (!get_ldev(mdev)) {
2038 retcode = ERR_NO_DISK;
2039 goto out;
2040 }
2041
2042 /* this is "skip initial sync", assume to be clean */
2043 if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
2044 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2045 dev_info(DEV, "Preparing to skip initial sync\n");
2046 skip_initial_sync = 1;
2047 } else if (mdev->state.conn != C_STANDALONE) {
2048 retcode = ERR_CONNECTED;
2049 goto out_dec;
2050 }
2051
2052 drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2053 drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2054
2055 if (args.clear_bm) {
2056 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, "clear_n_write from new_c_uuid");
2057 if (err) {
2058 dev_err(DEV, "Writing bitmap failed with %d\n",err);
2059 retcode = ERR_IO_MD_DISK;
2060 }
2061 if (skip_initial_sync) {
2062 drbd_send_uuids_skip_initial_sync(mdev);
2063 _drbd_uuid_set(mdev, UI_BITMAP, 0);
2064 spin_lock_irq(&mdev->req_lock);
2065 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2066 CS_VERBOSE, NULL);
2067 spin_unlock_irq(&mdev->req_lock);
2068 }
2069 }
2070
2071 drbd_md_sync(mdev);
2072out_dec:
2073 put_ldev(mdev);
2074out:
2075 mutex_unlock(&mdev->state_mutex);
2076
2077 reply->ret_code = retcode;
2078 return 0;
2079}
2080
2081static struct drbd_conf *ensure_mdev(struct drbd_nl_cfg_req *nlp)
2082{
2083 struct drbd_conf *mdev;
2084
2085 if (nlp->drbd_minor >= minor_count)
2086 return NULL;
2087
2088 mdev = minor_to_mdev(nlp->drbd_minor);
2089
2090 if (!mdev && (nlp->flags & DRBD_NL_CREATE_DEVICE)) {
2091 struct gendisk *disk = NULL;
2092 mdev = drbd_new_device(nlp->drbd_minor);
2093
2094 spin_lock_irq(&drbd_pp_lock);
2095 if (minor_table[nlp->drbd_minor] == NULL) {
2096 minor_table[nlp->drbd_minor] = mdev;
2097 disk = mdev->vdisk;
2098 mdev = NULL;
2099 } /* else: we lost the race */
2100 spin_unlock_irq(&drbd_pp_lock);
2101
2102 if (disk) /* we won the race above */
2103 /* in case we ever add a drbd_delete_device(),
2104 * don't forget the del_gendisk! */
2105 add_disk(disk);
2106 else /* we lost the race above */
2107 drbd_free_mdev(mdev);
2108
2109 mdev = minor_to_mdev(nlp->drbd_minor);
2110 }
2111
2112 return mdev;
2113}
2114
2115struct cn_handler_struct {
2116 int (*function)(struct drbd_conf *,
2117 struct drbd_nl_cfg_req *,
2118 struct drbd_nl_cfg_reply *);
2119 int reply_body_size;
2120};
2121
2122static struct cn_handler_struct cnd_table[] = {
2123 [ P_primary ] = { &drbd_nl_primary, 0 },
2124 [ P_secondary ] = { &drbd_nl_secondary, 0 },
2125 [ P_disk_conf ] = { &drbd_nl_disk_conf, 0 },
2126 [ P_detach ] = { &drbd_nl_detach, 0 },
2127 [ P_net_conf ] = { &drbd_nl_net_conf, 0 },
2128 [ P_disconnect ] = { &drbd_nl_disconnect, 0 },
2129 [ P_resize ] = { &drbd_nl_resize, 0 },
2130 [ P_syncer_conf ] = { &drbd_nl_syncer_conf, 0 },
2131 [ P_invalidate ] = { &drbd_nl_invalidate, 0 },
2132 [ P_invalidate_peer ] = { &drbd_nl_invalidate_peer, 0 },
2133 [ P_pause_sync ] = { &drbd_nl_pause_sync, 0 },
2134 [ P_resume_sync ] = { &drbd_nl_resume_sync, 0 },
2135 [ P_suspend_io ] = { &drbd_nl_suspend_io, 0 },
2136 [ P_resume_io ] = { &drbd_nl_resume_io, 0 },
2137 [ P_outdate ] = { &drbd_nl_outdate, 0 },
2138 [ P_get_config ] = { &drbd_nl_get_config,
2139 sizeof(struct syncer_conf_tag_len_struct) +
2140 sizeof(struct disk_conf_tag_len_struct) +
2141 sizeof(struct net_conf_tag_len_struct) },
2142 [ P_get_state ] = { &drbd_nl_get_state,
2143 sizeof(struct get_state_tag_len_struct) +
2144 sizeof(struct sync_progress_tag_len_struct) },
2145 [ P_get_uuids ] = { &drbd_nl_get_uuids,
2146 sizeof(struct get_uuids_tag_len_struct) },
2147 [ P_get_timeout_flag ] = { &drbd_nl_get_timeout_flag,
2148 sizeof(struct get_timeout_flag_tag_len_struct)},
2149 [ P_start_ov ] = { &drbd_nl_start_ov, 0 },
2150 [ P_new_c_uuid ] = { &drbd_nl_new_c_uuid, 0 },
2151};
2152
Philipp Reisner9f5180e2009-10-06 09:30:14 +02002153static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002154{
2155 struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2156 struct cn_handler_struct *cm;
2157 struct cn_msg *cn_reply;
2158 struct drbd_nl_cfg_reply *reply;
2159 struct drbd_conf *mdev;
2160 int retcode, rr;
2161 int reply_size = sizeof(struct cn_msg)
2162 + sizeof(struct drbd_nl_cfg_reply)
2163 + sizeof(short int);
2164
2165 if (!try_module_get(THIS_MODULE)) {
2166 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2167 return;
2168 }
2169
Philipp Reisner9f5180e2009-10-06 09:30:14 +02002170 if (!cap_raised(nsp->eff_cap, CAP_SYS_ADMIN)) {
2171 retcode = ERR_PERM;
2172 goto fail;
2173 }
2174
Philipp Reisnerb411b362009-09-25 16:07:19 -07002175 mdev = ensure_mdev(nlp);
2176 if (!mdev) {
2177 retcode = ERR_MINOR_INVALID;
2178 goto fail;
2179 }
2180
Philipp Reisnerb411b362009-09-25 16:07:19 -07002181 if (nlp->packet_type >= P_nl_after_last_packet) {
2182 retcode = ERR_PACKET_NR;
2183 goto fail;
2184 }
2185
2186 cm = cnd_table + nlp->packet_type;
2187
2188 /* This may happen if packet number is 0: */
2189 if (cm->function == NULL) {
2190 retcode = ERR_PACKET_NR;
2191 goto fail;
2192 }
2193
2194 reply_size += cm->reply_body_size;
2195
2196 /* allocation not in the IO path, cqueue thread context */
2197 cn_reply = kmalloc(reply_size, GFP_KERNEL);
2198 if (!cn_reply) {
2199 retcode = ERR_NOMEM;
2200 goto fail;
2201 }
2202 reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2203
2204 reply->packet_type =
2205 cm->reply_body_size ? nlp->packet_type : P_nl_after_last_packet;
2206 reply->minor = nlp->drbd_minor;
2207 reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2208 /* reply->tag_list; might be modified by cm->function. */
2209
2210 rr = cm->function(mdev, nlp, reply);
2211
2212 cn_reply->id = req->id;
2213 cn_reply->seq = req->seq;
2214 cn_reply->ack = req->ack + 1;
2215 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2216 cn_reply->flags = 0;
2217
Philipp Reisnerb411b362009-09-25 16:07:19 -07002218 rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2219 if (rr && rr != -ESRCH)
2220 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2221
2222 kfree(cn_reply);
2223 module_put(THIS_MODULE);
2224 return;
2225 fail:
2226 drbd_nl_send_reply(req, retcode);
2227 module_put(THIS_MODULE);
2228}
2229
2230static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2231
2232static unsigned short *
2233__tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2234 unsigned short len, int nul_terminated)
2235{
2236 unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2237 len = (len < l) ? len : l;
2238 put_unaligned(tag, tl++);
2239 put_unaligned(len, tl++);
2240 memcpy(tl, data, len);
2241 tl = (unsigned short*)((char*)tl + len);
2242 if (nul_terminated)
2243 *((char*)tl - 1) = 0;
2244 return tl;
2245}
2246
2247static unsigned short *
2248tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2249{
2250 return __tl_add_blob(tl, tag, data, len, 0);
2251}
2252
2253static unsigned short *
2254tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2255{
2256 return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2257}
2258
2259static unsigned short *
2260tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2261{
2262 put_unaligned(tag, tl++);
2263 switch(tag_type(tag)) {
2264 case TT_INTEGER:
2265 put_unaligned(sizeof(int), tl++);
2266 put_unaligned(*(int *)val, (int *)tl);
2267 tl = (unsigned short*)((char*)tl+sizeof(int));
2268 break;
2269 case TT_INT64:
2270 put_unaligned(sizeof(u64), tl++);
2271 put_unaligned(*(u64 *)val, (u64 *)tl);
2272 tl = (unsigned short*)((char*)tl+sizeof(u64));
2273 break;
2274 default:
2275 /* someone did something stupid. */
2276 ;
2277 }
2278 return tl;
2279}
2280
2281void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2282{
2283 char buffer[sizeof(struct cn_msg)+
2284 sizeof(struct drbd_nl_cfg_reply)+
2285 sizeof(struct get_state_tag_len_struct)+
2286 sizeof(short int)];
2287 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2288 struct drbd_nl_cfg_reply *reply =
2289 (struct drbd_nl_cfg_reply *)cn_reply->data;
2290 unsigned short *tl = reply->tag_list;
2291
2292 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2293
2294 tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2295
2296 put_unaligned(TT_END, tl++); /* Close the tag list */
2297
2298 cn_reply->id.idx = CN_IDX_DRBD;
2299 cn_reply->id.val = CN_VAL_DRBD;
2300
2301 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2302 cn_reply->ack = 0; /* not used here. */
2303 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2304 (int)((char *)tl - (char *)reply->tag_list);
2305 cn_reply->flags = 0;
2306
2307 reply->packet_type = P_get_state;
2308 reply->minor = mdev_to_minor(mdev);
2309 reply->ret_code = NO_ERROR;
2310
Philipp Reisnerb411b362009-09-25 16:07:19 -07002311 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2312}
2313
2314void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2315{
2316 char buffer[sizeof(struct cn_msg)+
2317 sizeof(struct drbd_nl_cfg_reply)+
2318 sizeof(struct call_helper_tag_len_struct)+
2319 sizeof(short int)];
2320 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2321 struct drbd_nl_cfg_reply *reply =
2322 (struct drbd_nl_cfg_reply *)cn_reply->data;
2323 unsigned short *tl = reply->tag_list;
2324
2325 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2326
2327 tl = tl_add_str(tl, T_helper, helper_name);
2328 put_unaligned(TT_END, tl++); /* Close the tag list */
2329
2330 cn_reply->id.idx = CN_IDX_DRBD;
2331 cn_reply->id.val = CN_VAL_DRBD;
2332
2333 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2334 cn_reply->ack = 0; /* not used here. */
2335 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2336 (int)((char *)tl - (char *)reply->tag_list);
2337 cn_reply->flags = 0;
2338
2339 reply->packet_type = P_call_helper;
2340 reply->minor = mdev_to_minor(mdev);
2341 reply->ret_code = NO_ERROR;
2342
Philipp Reisnerb411b362009-09-25 16:07:19 -07002343 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2344}
2345
2346void drbd_bcast_ee(struct drbd_conf *mdev,
2347 const char *reason, const int dgs,
2348 const char* seen_hash, const char* calc_hash,
2349 const struct drbd_epoch_entry* e)
2350{
2351 struct cn_msg *cn_reply;
2352 struct drbd_nl_cfg_reply *reply;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002353 unsigned short *tl;
Lars Ellenberg45bb9122010-05-14 17:10:48 +02002354 struct page *page;
2355 unsigned len;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002356
2357 if (!e)
2358 return;
2359 if (!reason || !reason[0])
2360 return;
2361
2362 /* apparently we have to memcpy twice, first to prepare the data for the
2363 * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2364 * netlink skb. */
2365 /* receiver thread context, which is not in the writeout path (of this node),
2366 * but may be in the writeout path of the _other_ node.
2367 * GFP_NOIO to avoid potential "distributed deadlock". */
2368 cn_reply = kmalloc(
2369 sizeof(struct cn_msg)+
2370 sizeof(struct drbd_nl_cfg_reply)+
2371 sizeof(struct dump_ee_tag_len_struct)+
2372 sizeof(short int),
2373 GFP_NOIO);
2374
2375 if (!cn_reply) {
2376 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2377 (unsigned long long)e->sector, e->size);
2378 return;
2379 }
2380
2381 reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2382 tl = reply->tag_list;
2383
2384 tl = tl_add_str(tl, T_dump_ee_reason, reason);
2385 tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2386 tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2387 tl = tl_add_int(tl, T_ee_sector, &e->sector);
2388 tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2389
2390 put_unaligned(T_ee_data, tl++);
2391 put_unaligned(e->size, tl++);
2392
Lars Ellenberg45bb9122010-05-14 17:10:48 +02002393 len = e->size;
2394 page = e->pages;
2395 page_chain_for_each(page) {
2396 void *d = kmap_atomic(page, KM_USER0);
2397 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2398 memcpy(tl, d, l);
2399 kunmap_atomic(d, KM_USER0);
2400 tl = (unsigned short*)((char*)tl + l);
2401 len -= l;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002402 }
2403 put_unaligned(TT_END, tl++); /* Close the tag list */
2404
2405 cn_reply->id.idx = CN_IDX_DRBD;
2406 cn_reply->id.val = CN_VAL_DRBD;
2407
2408 cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2409 cn_reply->ack = 0; // not used here.
2410 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2411 (int)((char*)tl - (char*)reply->tag_list);
2412 cn_reply->flags = 0;
2413
2414 reply->packet_type = P_dump_ee;
2415 reply->minor = mdev_to_minor(mdev);
2416 reply->ret_code = NO_ERROR;
2417
Philipp Reisnerb411b362009-09-25 16:07:19 -07002418 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2419 kfree(cn_reply);
2420}
2421
2422void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2423{
2424 char buffer[sizeof(struct cn_msg)+
2425 sizeof(struct drbd_nl_cfg_reply)+
2426 sizeof(struct sync_progress_tag_len_struct)+
2427 sizeof(short int)];
2428 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2429 struct drbd_nl_cfg_reply *reply =
2430 (struct drbd_nl_cfg_reply *)cn_reply->data;
2431 unsigned short *tl = reply->tag_list;
2432 unsigned long rs_left;
2433 unsigned int res;
2434
2435 /* no local ref, no bitmap, no syncer progress, no broadcast. */
2436 if (!get_ldev(mdev))
2437 return;
2438 drbd_get_syncer_progress(mdev, &rs_left, &res);
2439 put_ldev(mdev);
2440
2441 tl = tl_add_int(tl, T_sync_progress, &res);
2442 put_unaligned(TT_END, tl++); /* Close the tag list */
2443
2444 cn_reply->id.idx = CN_IDX_DRBD;
2445 cn_reply->id.val = CN_VAL_DRBD;
2446
2447 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2448 cn_reply->ack = 0; /* not used here. */
2449 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2450 (int)((char *)tl - (char *)reply->tag_list);
2451 cn_reply->flags = 0;
2452
2453 reply->packet_type = P_sync_progress;
2454 reply->minor = mdev_to_minor(mdev);
2455 reply->ret_code = NO_ERROR;
2456
Philipp Reisnerb411b362009-09-25 16:07:19 -07002457 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2458}
2459
2460int __init drbd_nl_init(void)
2461{
2462 static struct cb_id cn_id_drbd;
2463 int err, try=10;
2464
2465 cn_id_drbd.val = CN_VAL_DRBD;
2466 do {
2467 cn_id_drbd.idx = cn_idx;
2468 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2469 if (!err)
2470 break;
2471 cn_idx = (cn_idx + CN_IDX_STEP);
2472 } while (try--);
2473
2474 if (err) {
2475 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2476 return err;
2477 }
2478
2479 return 0;
2480}
2481
2482void drbd_nl_cleanup(void)
2483{
2484 static struct cb_id cn_id_drbd;
2485
2486 cn_id_drbd.idx = cn_idx;
2487 cn_id_drbd.val = CN_VAL_DRBD;
2488
2489 cn_del_callback(&cn_id_drbd);
2490}
2491
2492void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2493{
2494 char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2495 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2496 struct drbd_nl_cfg_reply *reply =
2497 (struct drbd_nl_cfg_reply *)cn_reply->data;
2498 int rr;
2499
2500 cn_reply->id = req->id;
2501
2502 cn_reply->seq = req->seq;
2503 cn_reply->ack = req->ack + 1;
2504 cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2505 cn_reply->flags = 0;
2506
2507 reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2508 reply->ret_code = ret_code;
2509
Philipp Reisnerb411b362009-09-25 16:07:19 -07002510 rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2511 if (rr && rr != -ESRCH)
2512 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2513}
2514