]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - drivers/block/drbd/drbd_nl.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
drbd: fix race on meta-data update
[mirror_ubuntu-artful-kernel.git] / drivers / block / drbd / drbd_nl.c
CommitLineData
b411b363
PR		 1/*
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
b411b363
PR		 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"
265be2d0 36#include "drbd_req.h"
b411b363
PR		 37#include "drbd_wrappers.h"
38#include <asm/unaligned.h>
39#include <linux/drbd_tag_magic.h>
40#include <linux/drbd_limits.h>
87f7be4c
PR		 41#include <linux/compiler.h>
42#include <linux/kthread.h>
b411b363
PR		 43
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
b411b363
PR		 211 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
87f7be4c
PR		 258static 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}
b411b363
PR		 277
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 &&
d10a33c6
PR		 307 (mdev->state.disk < D_UP_TO_DATE &&
308 mdev->state.disk >= D_INCONSISTENT)) {
b411b363
PR		 309 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 =
1f552430 429 drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force);
b411b363
PR		 430
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);
265be2d0
PR		 516 if (mdev->state.susp)
517 return;
b411b363
PR		 518 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 */
d845030f 534enum determine_dev_size drbd_determin_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
b411b363
PR		 535{
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
d845030f 565 size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
b411b363
PR		 566
567 if (drbd_get_capacity(mdev->this_bdev) != size ||
568 drbd_bm_capacity(mdev) != size) {
569 int err;
02d9a94b 570 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
b411b363
PR		 571 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
a393db6f 620drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
b411b363
PR		 621{
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
a393db6f
PR		 630 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
b411b363
PR		 635 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
b411b363
PR		 728 max_seg_s = min(queue_max_sectors(b) * queue_logical_block_size(b), max_seg_s);
729
086fa5ff 730 blk_queue_max_hw_sectors(q, max_seg_s >> 9);
8a78362c 731 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
b411b363
PR		 732 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 *
c518d04f
LE		 753 * Wait for a potentially exiting worker, then restart it,
754 * or start a new one. Flush any pending work, there may still be an
755 * after_state_change queued.
b411b363
PR		 756 */
757static void drbd_reconfig_start(struct drbd_conf *mdev)
758{
6c6c7951 759 wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
b411b363
PR		 760 wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
761 drbd_thread_start(&mdev->worker);
c518d04f 762 drbd_flush_workqueue(mdev);
b411b363
PR		 763}
764
765/* if still unconfigured, stops worker again.
766 * if configured now, clears CONFIG_PENDING.
767 * wakes potential waiters */
768static void drbd_reconfig_done(struct drbd_conf *mdev)
769{
770 spin_lock_irq(&mdev->req_lock);
771 if (mdev->state.disk == D_DISKLESS &&
772 mdev->state.conn == C_STANDALONE &&
773 mdev->state.role == R_SECONDARY) {
774 set_bit(DEVICE_DYING, &mdev->flags);
775 drbd_thread_stop_nowait(&mdev->worker);
776 } else
777 clear_bit(CONFIG_PENDING, &mdev->flags);
778 spin_unlock_irq(&mdev->req_lock);
779 wake_up(&mdev->state_wait);
780}
781
0778286a
PR		 782/* Make sure IO is suspended before calling this function(). */
783static void drbd_suspend_al(struct drbd_conf *mdev)
784{
785 int s = 0;
786
787 if (lc_try_lock(mdev->act_log)) {
788 drbd_al_shrink(mdev);
789 lc_unlock(mdev->act_log);
790 } else {
791 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
792 return;
793 }
794
795 spin_lock_irq(&mdev->req_lock);
796 if (mdev->state.conn < C_CONNECTED)
797 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
798
799 spin_unlock_irq(&mdev->req_lock);
800
801 if (s)
802 dev_info(DEV, "Suspended AL updates\n");
803}
804
b411b363
PR		 805/* does always return 0;
806 * interesting return code is in reply->ret_code */
807static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
808 struct drbd_nl_cfg_reply *reply)
809{
810 enum drbd_ret_codes retcode;
811 enum determine_dev_size dd;
812 sector_t max_possible_sectors;
813 sector_t min_md_device_sectors;
814 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
815 struct inode *inode, *inode2;
816 struct lru_cache *resync_lru = NULL;
817 union drbd_state ns, os;
818 int rv;
819 int cp_discovered = 0;
820 int logical_block_size;
821
822 drbd_reconfig_start(mdev);
823
824 /* if you want to reconfigure, please tear down first */
825 if (mdev->state.disk > D_DISKLESS) {
826 retcode = ERR_DISK_CONFIGURED;
827 goto fail;
828 }
829
830 /* allocation not in the IO path, cqueue thread context */
831 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
832 if (!nbc) {
833 retcode = ERR_NOMEM;
834 goto fail;
835 }
836
837 nbc->dc.disk_size = DRBD_DISK_SIZE_SECT_DEF;
838 nbc->dc.on_io_error = DRBD_ON_IO_ERROR_DEF;
839 nbc->dc.fencing = DRBD_FENCING_DEF;
840 nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
841
842 if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
843 retcode = ERR_MANDATORY_TAG;
844 goto fail;
845 }
846
847 if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
848 retcode = ERR_MD_IDX_INVALID;
849 goto fail;
850 }
851
47ff2d0a
PR		 852 if (get_net_conf(mdev)) {
853 int prot = mdev->net_conf->wire_protocol;
854 put_net_conf(mdev);
855 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
856 retcode = ERR_STONITH_AND_PROT_A;
857 goto fail;
858 }
859 }
860
b411b363
PR		 861 nbc->lo_file = filp_open(nbc->dc.backing_dev, O_RDWR, 0);
862 if (IS_ERR(nbc->lo_file)) {
863 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
864 PTR_ERR(nbc->lo_file));
865 nbc->lo_file = NULL;
866 retcode = ERR_OPEN_DISK;
867 goto fail;
868 }
869
870 inode = nbc->lo_file->f_dentry->d_inode;
871
872 if (!S_ISBLK(inode->i_mode)) {
873 retcode = ERR_DISK_NOT_BDEV;
874 goto fail;
875 }
876
877 nbc->md_file = filp_open(nbc->dc.meta_dev, O_RDWR, 0);
878 if (IS_ERR(nbc->md_file)) {
879 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
880 PTR_ERR(nbc->md_file));
881 nbc->md_file = NULL;
882 retcode = ERR_OPEN_MD_DISK;
883 goto fail;
884 }
885
886 inode2 = nbc->md_file->f_dentry->d_inode;
887
888 if (!S_ISBLK(inode2->i_mode)) {
889 retcode = ERR_MD_NOT_BDEV;
890 goto fail;
891 }
892
893 nbc->backing_bdev = inode->i_bdev;
894 if (bd_claim(nbc->backing_bdev, mdev)) {
895 printk(KERN_ERR "drbd: bd_claim(%p,%p); failed [%p;%p;%u]\n",
896 nbc->backing_bdev, mdev,
897 nbc->backing_bdev->bd_holder,
898 nbc->backing_bdev->bd_contains->bd_holder,
899 nbc->backing_bdev->bd_holders);
900 retcode = ERR_BDCLAIM_DISK;
901 goto fail;
902 }
903
904 resync_lru = lc_create("resync", drbd_bm_ext_cache,
905 61, sizeof(struct bm_extent),
906 offsetof(struct bm_extent, lce));
907 if (!resync_lru) {
908 retcode = ERR_NOMEM;
909 goto release_bdev_fail;
910 }
911
912 /* meta_dev_idx >= 0: external fixed size,
913 * possibly multiple drbd sharing one meta device.
914 * TODO in that case, paranoia check that [md_bdev, meta_dev_idx] is
915 * not yet used by some other drbd minor!
916 * (if you use drbd.conf + drbdadm,
917 * that should check it for you already; but if you don't, or someone
918 * fooled it, we need to double check here) */
919 nbc->md_bdev = inode2->i_bdev;
920 if (bd_claim(nbc->md_bdev, (nbc->dc.meta_dev_idx < 0) ? (void *)mdev
921 : (void *) drbd_m_holder)) {
922 retcode = ERR_BDCLAIM_MD_DISK;
923 goto release_bdev_fail;
924 }
925
926 if ((nbc->backing_bdev == nbc->md_bdev) !=
927 (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
928 nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
929 retcode = ERR_MD_IDX_INVALID;
930 goto release_bdev2_fail;
931 }
932
933 /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
934 drbd_md_set_sector_offsets(mdev, nbc);
935
936 if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
937 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
938 (unsigned long long) drbd_get_max_capacity(nbc),
939 (unsigned long long) nbc->dc.disk_size);
940 retcode = ERR_DISK_TO_SMALL;
941 goto release_bdev2_fail;
942 }
943
944 if (nbc->dc.meta_dev_idx < 0) {
945 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
946 /* at least one MB, otherwise it does not make sense */
947 min_md_device_sectors = (2<<10);
948 } else {
949 max_possible_sectors = DRBD_MAX_SECTORS;
950 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
951 }
952
b411b363
PR		 953 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
954 retcode = ERR_MD_DISK_TO_SMALL;
955 dev_warn(DEV, "refusing attach: md-device too small, "
956 "at least %llu sectors needed for this meta-disk type\n",
957 (unsigned long long) min_md_device_sectors);
958 goto release_bdev2_fail;
959 }
960
961 /* Make sure the new disk is big enough
962 * (we may currently be R_PRIMARY with no local disk...) */
963 if (drbd_get_max_capacity(nbc) <
964 drbd_get_capacity(mdev->this_bdev)) {
965 retcode = ERR_DISK_TO_SMALL;
966 goto release_bdev2_fail;
967 }
968
969 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
970
1352994b
LE		 971 if (nbc->known_size > max_possible_sectors) {
972 dev_warn(DEV, "==> truncating very big lower level device "
973 "to currently maximum possible %llu sectors <==\n",
974 (unsigned long long) max_possible_sectors);
975 if (nbc->dc.meta_dev_idx >= 0)
976 dev_warn(DEV, "==>> using internal or flexible "
977 "meta data may help <<==\n");
978 }
979
b411b363
PR		 980 drbd_suspend_io(mdev);
981 /* also wait for the last barrier ack. */
8f488156 982 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || mdev->state.susp);
b411b363
PR		 983 /* and for any other previously queued work */
984 drbd_flush_workqueue(mdev);
985
986 retcode = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
987 drbd_resume_io(mdev);
988 if (retcode < SS_SUCCESS)
989 goto release_bdev2_fail;
990
991 if (!get_ldev_if_state(mdev, D_ATTACHING))
992 goto force_diskless;
993
994 drbd_md_set_sector_offsets(mdev, nbc);
995
4aa83b7b
LE		 996 /* allocate a second IO page if logical_block_size != 512 */
997 logical_block_size = bdev_logical_block_size(nbc->md_bdev);
998 if (logical_block_size == 0)
999 logical_block_size = MD_SECTOR_SIZE;
1000
1001 if (logical_block_size != MD_SECTOR_SIZE) {
1002 if (!mdev->md_io_tmpp) {
1003 struct page *page = alloc_page(GFP_NOIO);
1004 if (!page)
1005 goto force_diskless_dec;
1006
1007 dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
1008 logical_block_size, MD_SECTOR_SIZE);
1009 dev_warn(DEV, "Workaround engaged (has performance impact).\n");
1010
1011 mdev->md_io_tmpp = page;
1012 }
1013 }
1014
b411b363
PR		 1015 if (!mdev->bitmap) {
1016 if (drbd_bm_init(mdev)) {
1017 retcode = ERR_NOMEM;
1018 goto force_diskless_dec;
1019 }
1020 }
1021
1022 retcode = drbd_md_read(mdev, nbc);
1023 if (retcode != NO_ERROR)
1024 goto force_diskless_dec;
1025
1026 if (mdev->state.conn < C_CONNECTED &&
1027 mdev->state.role == R_PRIMARY &&
1028 (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1029 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1030 (unsigned long long)mdev->ed_uuid);
1031 retcode = ERR_DATA_NOT_CURRENT;
1032 goto force_diskless_dec;
1033 }
1034
1035 /* Since we are diskless, fix the activity log first... */
1036 if (drbd_check_al_size(mdev)) {
1037 retcode = ERR_NOMEM;
1038 goto force_diskless_dec;
1039 }
1040
1041 /* Prevent shrinking of consistent devices ! */
1042 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
a393db6f 1043 drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
b411b363
PR		 1044 dev_warn(DEV, "refusing to truncate a consistent device\n");
1045 retcode = ERR_DISK_TO_SMALL;
1046 goto force_diskless_dec;
1047 }
1048
1049 if (!drbd_al_read_log(mdev, nbc)) {
1050 retcode = ERR_IO_MD_DISK;
1051 goto force_diskless_dec;
1052 }
1053
b411b363
PR		 1054 /* Reset the "barriers don't work" bits here, then force meta data to
1055 * be written, to ensure we determine if barriers are supported. */
1056 if (nbc->dc.no_md_flush)
1057 set_bit(MD_NO_BARRIER, &mdev->flags);
1058 else
1059 clear_bit(MD_NO_BARRIER, &mdev->flags);
1060
1061 /* Point of no return reached.
1062 * Devices and memory are no longer released by error cleanup below.
1063 * now mdev takes over responsibility, and the state engine should
1064 * clean it up somewhere. */
1065 D_ASSERT(mdev->ldev == NULL);
1066 mdev->ldev = nbc;
1067 mdev->resync = resync_lru;
1068 nbc = NULL;
1069 resync_lru = NULL;
1070
1071 mdev->write_ordering = WO_bio_barrier;
1072 drbd_bump_write_ordering(mdev, WO_bio_barrier);
1073
1074 if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1075 set_bit(CRASHED_PRIMARY, &mdev->flags);
1076 else
1077 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1078
894c6a94
PR		 1079 if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1080 !(mdev->state.role == R_PRIMARY && mdev->state.susp &&
1081 mdev->sync_conf.on_no_data == OND_SUSPEND_IO)) {
b411b363
PR		 1082 set_bit(CRASHED_PRIMARY, &mdev->flags);
1083 cp_discovered = 1;
1084 }
1085
1086 mdev->send_cnt = 0;
1087 mdev->recv_cnt = 0;
1088 mdev->read_cnt = 0;
1089 mdev->writ_cnt = 0;
1090
d5373389
PR		 1091 drbd_setup_queue_param(mdev, mdev->state.conn == C_CONNECTED &&
1092 mdev->agreed_pro_version < 95 ?
1093 DRBD_MAX_SIZE_H80_PACKET : DRBD_MAX_SEGMENT_SIZE);
b411b363
PR		 1094
1095 /* If I am currently not R_PRIMARY,
1096 * but meta data primary indicator is set,
1097 * I just now recover from a hard crash,
1098 * and have been R_PRIMARY before that crash.
1099 *
1100 * Now, if I had no connection before that crash
1101 * (have been degraded R_PRIMARY), chances are that
1102 * I won't find my peer now either.
1103 *
1104 * In that case, and _only_ in that case,
1105 * we use the degr-wfc-timeout instead of the default,
1106 * so we can automatically recover from a crash of a
1107 * degraded but active "cluster" after a certain timeout.
1108 */
1109 clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1110 if (mdev->state.role != R_PRIMARY &&
1111 drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1112 !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1113 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1114
a393db6f 1115 dd = drbd_determin_dev_size(mdev, 0);
b411b363
PR		 1116 if (dd == dev_size_error) {
1117 retcode = ERR_NOMEM_BITMAP;
1118 goto force_diskless_dec;
1119 } else if (dd == grew)
1120 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1121
1122 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1123 dev_info(DEV, "Assuming that all blocks are out of sync "
1124 "(aka FullSync)\n");
1125 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from attaching")) {
1126 retcode = ERR_IO_MD_DISK;
1127 goto force_diskless_dec;
1128 }
1129 } else {
1130 if (drbd_bitmap_io(mdev, &drbd_bm_read, "read from attaching") < 0) {
1131 retcode = ERR_IO_MD_DISK;
1132 goto force_diskless_dec;
1133 }
1134 }
1135
1136 if (cp_discovered) {
1137 drbd_al_apply_to_bm(mdev);
1138 drbd_al_to_on_disk_bm(mdev);
1139 }
1140
0778286a
PR		 1141 if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1142 drbd_suspend_al(mdev); /* IO is still suspended here... */
1143
b411b363
PR		 1144 spin_lock_irq(&mdev->req_lock);
1145 os = mdev->state;
1146 ns.i = os.i;
1147 /* If MDF_CONSISTENT is not set go into inconsistent state,
1148 otherwise investigate MDF_WasUpToDate...
1149 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1150 otherwise into D_CONSISTENT state.
1151 */
1152 if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1153 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1154 ns.disk = D_CONSISTENT;
1155 else
1156 ns.disk = D_OUTDATED;
1157 } else {
1158 ns.disk = D_INCONSISTENT;
1159 }
1160
1161 if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1162 ns.pdsk = D_OUTDATED;
1163
1164 if ( ns.disk == D_CONSISTENT &&
1165 (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1166 ns.disk = D_UP_TO_DATE;
1167
1168 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1169 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1170 this point, because drbd_request_state() modifies these
1171 flags. */
1172
1173 /* In case we are C_CONNECTED postpone any decision on the new disk
1174 state after the negotiation phase. */
1175 if (mdev->state.conn == C_CONNECTED) {
1176 mdev->new_state_tmp.i = ns.i;
1177 ns.i = os.i;
1178 ns.disk = D_NEGOTIATING;
dc66c74d
PR		 1179
1180 /* We expect to receive up-to-date UUIDs soon.
1181 To avoid a race in receive_state, free p_uuid while
1182 holding req_lock. I.e. atomic with the state change */
1183 kfree(mdev->p_uuid);
1184 mdev->p_uuid = NULL;
b411b363
PR		 1185 }
1186
1187 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1188 ns = mdev->state;
1189 spin_unlock_irq(&mdev->req_lock);
1190
1191 if (rv < SS_SUCCESS)
1192 goto force_diskless_dec;
1193
1194 if (mdev->state.role == R_PRIMARY)
1195 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
1196 else
1197 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1198
1199 drbd_md_mark_dirty(mdev);
1200 drbd_md_sync(mdev);
1201
1202 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1203 put_ldev(mdev);
1204 reply->ret_code = retcode;
1205 drbd_reconfig_done(mdev);
1206 return 0;
1207
1208 force_diskless_dec:
1209 put_ldev(mdev);
1210 force_diskless:
1211 drbd_force_state(mdev, NS(disk, D_DISKLESS));
1212 drbd_md_sync(mdev);
1213 release_bdev2_fail:
1214 if (nbc)
1215 bd_release(nbc->md_bdev);
1216 release_bdev_fail:
1217 if (nbc)
1218 bd_release(nbc->backing_bdev);
1219 fail:
1220 if (nbc) {
1221 if (nbc->lo_file)
1222 fput(nbc->lo_file);
1223 if (nbc->md_file)
1224 fput(nbc->md_file);
1225 kfree(nbc);
1226 }
1227 lc_destroy(resync_lru);
1228
1229 reply->ret_code = retcode;
1230 drbd_reconfig_done(mdev);
1231 return 0;
1232}
1233
1234static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1235 struct drbd_nl_cfg_reply *reply)
1236{
1237 reply->ret_code = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1238 return 0;
1239}
1240
1241static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1242 struct drbd_nl_cfg_reply *reply)
1243{
1244 int i, ns;
1245 enum drbd_ret_codes retcode;
1246 struct net_conf *new_conf = NULL;
1247 struct crypto_hash *tfm = NULL;
1248 struct crypto_hash *integrity_w_tfm = NULL;
1249 struct crypto_hash *integrity_r_tfm = NULL;
1250 struct hlist_head *new_tl_hash = NULL;
1251 struct hlist_head *new_ee_hash = NULL;
1252 struct drbd_conf *odev;
1253 char hmac_name[CRYPTO_MAX_ALG_NAME];
1254 void *int_dig_out = NULL;
1255 void *int_dig_in = NULL;
1256 void *int_dig_vv = NULL;
1257 struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1258
1259 drbd_reconfig_start(mdev);
1260
1261 if (mdev->state.conn > C_STANDALONE) {
1262 retcode = ERR_NET_CONFIGURED;
1263 goto fail;
1264 }
1265
1266 /* allocation not in the IO path, cqueue thread context */
2db4e42e 1267 new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
b411b363
PR		 1268 if (!new_conf) {
1269 retcode = ERR_NOMEM;
1270 goto fail;
1271 }
1272
b411b363
PR		 1273 new_conf->timeout = DRBD_TIMEOUT_DEF;
1274 new_conf->try_connect_int = DRBD_CONNECT_INT_DEF;
1275 new_conf->ping_int = DRBD_PING_INT_DEF;
1276 new_conf->max_epoch_size = DRBD_MAX_EPOCH_SIZE_DEF;
1277 new_conf->max_buffers = DRBD_MAX_BUFFERS_DEF;
1278 new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1279 new_conf->sndbuf_size = DRBD_SNDBUF_SIZE_DEF;
1280 new_conf->rcvbuf_size = DRBD_RCVBUF_SIZE_DEF;
1281 new_conf->ko_count = DRBD_KO_COUNT_DEF;
1282 new_conf->after_sb_0p = DRBD_AFTER_SB_0P_DEF;
1283 new_conf->after_sb_1p = DRBD_AFTER_SB_1P_DEF;
1284 new_conf->after_sb_2p = DRBD_AFTER_SB_2P_DEF;
1285 new_conf->want_lose = 0;
1286 new_conf->two_primaries = 0;
1287 new_conf->wire_protocol = DRBD_PROT_C;
1288 new_conf->ping_timeo = DRBD_PING_TIMEO_DEF;
1289 new_conf->rr_conflict = DRBD_RR_CONFLICT_DEF;
1290
1291 if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1292 retcode = ERR_MANDATORY_TAG;
1293 goto fail;
1294 }
1295
1296 if (new_conf->two_primaries
1297 && (new_conf->wire_protocol != DRBD_PROT_C)) {
1298 retcode = ERR_NOT_PROTO_C;
1299 goto fail;
47ff2d0a
PR		 1300 }
1301
1302 if (get_ldev(mdev)) {
1303 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1304 put_ldev(mdev);
1305 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1306 retcode = ERR_STONITH_AND_PROT_A;
1307 goto fail;
1308 }
1309 }
b411b363
PR		 1310
1311 if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1312 retcode = ERR_DISCARD;
1313 goto fail;
1314 }
1315
1316 retcode = NO_ERROR;
1317
1318 new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1319 new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1320 for (i = 0; i < minor_count; i++) {
1321 odev = minor_to_mdev(i);
1322 if (!odev || odev == mdev)
1323 continue;
1324 if (get_net_conf(odev)) {
1325 taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1326 if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1327 !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1328 retcode = ERR_LOCAL_ADDR;
1329
1330 taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1331 if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1332 !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1333 retcode = ERR_PEER_ADDR;
1334
1335 put_net_conf(odev);
1336 if (retcode != NO_ERROR)
1337 goto fail;
1338 }
1339 }
1340
1341 if (new_conf->cram_hmac_alg[0] != 0) {
1342 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1343 new_conf->cram_hmac_alg);
1344 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1345 if (IS_ERR(tfm)) {
1346 tfm = NULL;
1347 retcode = ERR_AUTH_ALG;
1348 goto fail;
1349 }
1350
0798219f 1351 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
b411b363
PR		 1352 retcode = ERR_AUTH_ALG_ND;
1353 goto fail;
1354 }
1355 }
1356
1357 if (new_conf->integrity_alg[0]) {
1358 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1359 if (IS_ERR(integrity_w_tfm)) {
1360 integrity_w_tfm = NULL;
1361 retcode=ERR_INTEGRITY_ALG;
1362 goto fail;
1363 }
1364
1365 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1366 retcode=ERR_INTEGRITY_ALG_ND;
1367 goto fail;
1368 }
1369
1370 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1371 if (IS_ERR(integrity_r_tfm)) {
1372 integrity_r_tfm = NULL;
1373 retcode=ERR_INTEGRITY_ALG;
1374 goto fail;
1375 }
1376 }
1377
1378 ns = new_conf->max_epoch_size/8;
1379 if (mdev->tl_hash_s != ns) {
1380 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1381 if (!new_tl_hash) {
1382 retcode = ERR_NOMEM;
1383 goto fail;
1384 }
1385 }
1386
1387 ns = new_conf->max_buffers/8;
1388 if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1389 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1390 if (!new_ee_hash) {
1391 retcode = ERR_NOMEM;
1392 goto fail;
1393 }
1394 }
1395
1396 ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1397
1398 if (integrity_w_tfm) {
1399 i = crypto_hash_digestsize(integrity_w_tfm);
1400 int_dig_out = kmalloc(i, GFP_KERNEL);
1401 if (!int_dig_out) {
1402 retcode = ERR_NOMEM;
1403 goto fail;
1404 }
1405 int_dig_in = kmalloc(i, GFP_KERNEL);
1406 if (!int_dig_in) {
1407 retcode = ERR_NOMEM;
1408 goto fail;
1409 }
1410 int_dig_vv = kmalloc(i, GFP_KERNEL);
1411 if (!int_dig_vv) {
1412 retcode = ERR_NOMEM;
1413 goto fail;
1414 }
1415 }
1416
1417 if (!mdev->bitmap) {
1418 if(drbd_bm_init(mdev)) {
1419 retcode = ERR_NOMEM;
1420 goto fail;
1421 }
1422 }
1423
f70b3511 1424 drbd_flush_workqueue(mdev);
b411b363
PR		 1425 spin_lock_irq(&mdev->req_lock);
1426 if (mdev->net_conf != NULL) {
1427 retcode = ERR_NET_CONFIGURED;
1428 spin_unlock_irq(&mdev->req_lock);
1429 goto fail;
1430 }
1431 mdev->net_conf = new_conf;
1432
1433 mdev->send_cnt = 0;
1434 mdev->recv_cnt = 0;
1435
1436 if (new_tl_hash) {
1437 kfree(mdev->tl_hash);
1438 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1439 mdev->tl_hash = new_tl_hash;
1440 }
1441
1442 if (new_ee_hash) {
1443 kfree(mdev->ee_hash);
1444 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1445 mdev->ee_hash = new_ee_hash;
1446 }
1447
1448 crypto_free_hash(mdev->cram_hmac_tfm);
1449 mdev->cram_hmac_tfm = tfm;
1450
1451 crypto_free_hash(mdev->integrity_w_tfm);
1452 mdev->integrity_w_tfm = integrity_w_tfm;
1453
1454 crypto_free_hash(mdev->integrity_r_tfm);
1455 mdev->integrity_r_tfm = integrity_r_tfm;
1456
1457 kfree(mdev->int_dig_out);
1458 kfree(mdev->int_dig_in);
1459 kfree(mdev->int_dig_vv);
1460 mdev->int_dig_out=int_dig_out;
1461 mdev->int_dig_in=int_dig_in;
1462 mdev->int_dig_vv=int_dig_vv;
f70b3511 1463 retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL);
b411b363
PR		 1464 spin_unlock_irq(&mdev->req_lock);
1465
b411b363
PR		 1466 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1467 reply->ret_code = retcode;
1468 drbd_reconfig_done(mdev);
1469 return 0;
1470
1471fail:
1472 kfree(int_dig_out);
1473 kfree(int_dig_in);
1474 kfree(int_dig_vv);
1475 crypto_free_hash(tfm);
1476 crypto_free_hash(integrity_w_tfm);
1477 crypto_free_hash(integrity_r_tfm);
1478 kfree(new_tl_hash);
1479 kfree(new_ee_hash);
1480 kfree(new_conf);
1481
1482 reply->ret_code = retcode;
1483 drbd_reconfig_done(mdev);
1484 return 0;
1485}
1486
1487static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1488 struct drbd_nl_cfg_reply *reply)
1489{
1490 int retcode;
1491
1492 retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1493
1494 if (retcode == SS_NOTHING_TO_DO)
1495 goto done;
1496 else if (retcode == SS_ALREADY_STANDALONE)
1497 goto done;
1498 else if (retcode == SS_PRIMARY_NOP) {
1499 /* Our statche checking code wants to see the peer outdated. */
1500 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1501 pdsk, D_OUTDATED));
1502 } else if (retcode == SS_CW_FAILED_BY_PEER) {
1503 /* The peer probably wants to see us outdated. */
1504 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1505 disk, D_OUTDATED),
1506 CS_ORDERED);
1507 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1508 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1509 retcode = SS_SUCCESS;
1510 }
1511 }
1512
1513 if (retcode < SS_SUCCESS)
1514 goto fail;
1515
1516 if (wait_event_interruptible(mdev->state_wait,
1517 mdev->state.conn != C_DISCONNECTING)) {
1518 /* Do not test for mdev->state.conn == C_STANDALONE, since
1519 someone else might connect us in the mean time! */
1520 retcode = ERR_INTR;
1521 goto fail;
1522 }
1523
1524 done:
1525 retcode = NO_ERROR;
1526 fail:
1527 drbd_md_sync(mdev);
1528 reply->ret_code = retcode;
1529 return 0;
1530}
1531
1532void resync_after_online_grow(struct drbd_conf *mdev)
1533{
1534 int iass; /* I am sync source */
1535
1536 dev_info(DEV, "Resync of new storage after online grow\n");
1537 if (mdev->state.role != mdev->state.peer)
1538 iass = (mdev->state.role == R_PRIMARY);
1539 else
1540 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1541
1542 if (iass)
1543 drbd_start_resync(mdev, C_SYNC_SOURCE);
1544 else
1545 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1546}
1547
1548static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1549 struct drbd_nl_cfg_reply *reply)
1550{
1551 struct resize rs;
1552 int retcode = NO_ERROR;
b411b363 1553 enum determine_dev_size dd;
6495d2c6 1554 enum dds_flags ddsf;
b411b363
PR		 1555
1556 memset(&rs, 0, sizeof(struct resize));
1557 if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1558 retcode = ERR_MANDATORY_TAG;
1559 goto fail;
1560 }
1561
1562 if (mdev->state.conn > C_CONNECTED) {
1563 retcode = ERR_RESIZE_RESYNC;
1564 goto fail;
1565 }
1566
1567 if (mdev->state.role == R_SECONDARY &&
1568 mdev->state.peer == R_SECONDARY) {
1569 retcode = ERR_NO_PRIMARY;
1570 goto fail;
1571 }
1572
1573 if (!get_ldev(mdev)) {
1574 retcode = ERR_NO_DISK;
1575 goto fail;
1576 }
1577
6495d2c6
PR		 1578 if (rs.no_resync && mdev->agreed_pro_version < 93) {
1579 retcode = ERR_NEED_APV_93;
1580 goto fail;
1581 }
1582
087c2492 1583 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
b411b363 1584 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
b411b363
PR		 1585
1586 mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
6495d2c6
PR		 1587 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1588 dd = drbd_determin_dev_size(mdev, ddsf);
b411b363
PR		 1589 drbd_md_sync(mdev);
1590 put_ldev(mdev);
1591 if (dd == dev_size_error) {
1592 retcode = ERR_NOMEM_BITMAP;
1593 goto fail;
1594 }
1595
087c2492 1596 if (mdev->state.conn == C_CONNECTED) {
b411b363
PR		 1597 if (dd == grew)
1598 set_bit(RESIZE_PENDING, &mdev->flags);
1599
1600 drbd_send_uuids(mdev);
6495d2c6 1601 drbd_send_sizes(mdev, 1, ddsf);
b411b363
PR		 1602 }
1603
1604 fail:
1605 reply->ret_code = retcode;
1606 return 0;
1607}
1608
1609static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1610 struct drbd_nl_cfg_reply *reply)
1611{
1612 int retcode = NO_ERROR;
1613 int err;
1614 int ovr; /* online verify running */
1615 int rsr; /* re-sync running */
1616 struct crypto_hash *verify_tfm = NULL;
1617 struct crypto_hash *csums_tfm = NULL;
1618 struct syncer_conf sc;
1619 cpumask_var_t new_cpu_mask;
778f271d
PR		 1620 int *rs_plan_s = NULL;
1621 int fifo_size;
b411b363
PR		 1622
1623 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1624 retcode = ERR_NOMEM;
1625 goto fail;
1626 }
1627
1628 if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1629 memset(&sc, 0, sizeof(struct syncer_conf));
1630 sc.rate = DRBD_RATE_DEF;
1631 sc.after = DRBD_AFTER_DEF;
1632 sc.al_extents = DRBD_AL_EXTENTS_DEF;
265be2d0 1633 sc.on_no_data = DRBD_ON_NO_DATA_DEF;
9a31d716
PR		 1634 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1635 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1636 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
0f0601f4
LE		 1637 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1638 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
b411b363
PR		 1639 } else
1640 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1641
1642 if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1643 retcode = ERR_MANDATORY_TAG;
1644 goto fail;
1645 }
1646
1647 /* re-sync running */
1648 rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1649 mdev->state.conn == C_SYNC_TARGET ||
1650 mdev->state.conn == C_PAUSED_SYNC_S ||
1651 mdev->state.conn == C_PAUSED_SYNC_T );
1652
1653 if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1654 retcode = ERR_CSUMS_RESYNC_RUNNING;
1655 goto fail;
1656 }
1657
1658 if (!rsr && sc.csums_alg[0]) {
1659 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1660 if (IS_ERR(csums_tfm)) {
1661 csums_tfm = NULL;
1662 retcode = ERR_CSUMS_ALG;
1663 goto fail;
1664 }
1665
1666 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1667 retcode = ERR_CSUMS_ALG_ND;
1668 goto fail;
1669 }
1670 }
1671
1672 /* online verify running */
1673 ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1674
1675 if (ovr) {
1676 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1677 retcode = ERR_VERIFY_RUNNING;
1678 goto fail;
1679 }
1680 }
1681
1682 if (!ovr && sc.verify_alg[0]) {
1683 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1684 if (IS_ERR(verify_tfm)) {
1685 verify_tfm = NULL;
1686 retcode = ERR_VERIFY_ALG;
1687 goto fail;
1688 }
1689
1690 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1691 retcode = ERR_VERIFY_ALG_ND;
1692 goto fail;
1693 }
1694 }
1695
1696 /* silently ignore cpu mask on UP kernel */
1697 if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1698 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1699 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1700 if (err) {
1701 dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1702 retcode = ERR_CPU_MASK_PARSE;
1703 goto fail;
1704 }
1705 }
1706
1707 ERR_IF (sc.rate < 1) sc.rate = 1;
1708 ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1709#define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1710 if (sc.al_extents > AL_MAX) {
1711 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1712 sc.al_extents = AL_MAX;
1713 }
1714#undef AL_MAX
1715
1716 /* most sanity checks done, try to assign the new sync-after
1717 * dependency. need to hold the global lock in there,
1718 * to avoid a race in the dependency loop check. */
1719 retcode = drbd_alter_sa(mdev, sc.after);
1720 if (retcode != NO_ERROR)
1721 goto fail;
1722
778f271d
PR		 1723 fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1724 if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1725 rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1726 if (!rs_plan_s) {
1727 dev_err(DEV, "kmalloc of fifo_buffer failed");
1728 retcode = ERR_NOMEM;
1729 goto fail;
1730 }
1731 }
1732
b411b363
PR		 1733 /* ok, assign the rest of it as well.
1734 * lock against receive_SyncParam() */
1735 spin_lock(&mdev->peer_seq_lock);
1736 mdev->sync_conf = sc;
1737
1738 if (!rsr) {
1739 crypto_free_hash(mdev->csums_tfm);
1740 mdev->csums_tfm = csums_tfm;
1741 csums_tfm = NULL;
1742 }
1743
1744 if (!ovr) {
1745 crypto_free_hash(mdev->verify_tfm);
1746 mdev->verify_tfm = verify_tfm;
1747 verify_tfm = NULL;
1748 }
778f271d
PR		 1749
1750 if (fifo_size != mdev->rs_plan_s.size) {
1751 kfree(mdev->rs_plan_s.values);
1752 mdev->rs_plan_s.values = rs_plan_s;
1753 mdev->rs_plan_s.size = fifo_size;
1754 mdev->rs_planed = 0;
1755 rs_plan_s = NULL;
1756 }
1757
b411b363
PR		 1758 spin_unlock(&mdev->peer_seq_lock);
1759
1760 if (get_ldev(mdev)) {
1761 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1762 drbd_al_shrink(mdev);
1763 err = drbd_check_al_size(mdev);
1764 lc_unlock(mdev->act_log);
1765 wake_up(&mdev->al_wait);
1766
1767 put_ldev(mdev);
1768 drbd_md_sync(mdev);
1769
1770 if (err) {
1771 retcode = ERR_NOMEM;
1772 goto fail;
1773 }
1774 }
1775
1776 if (mdev->state.conn >= C_CONNECTED)
1777 drbd_send_sync_param(mdev, &sc);
1778
1779 if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1780 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1781 drbd_calc_cpu_mask(mdev);
1782 mdev->receiver.reset_cpu_mask = 1;
1783 mdev->asender.reset_cpu_mask = 1;
1784 mdev->worker.reset_cpu_mask = 1;
1785 }
1786
1787 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1788fail:
778f271d 1789 kfree(rs_plan_s);
b411b363
PR		 1790 free_cpumask_var(new_cpu_mask);
1791 crypto_free_hash(csums_tfm);
1792 crypto_free_hash(verify_tfm);
1793 reply->ret_code = retcode;
1794 return 0;
1795}
1796
1797static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1798 struct drbd_nl_cfg_reply *reply)
1799{
1800 int retcode;
1801
1802 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1803
1804 if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1805 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1806
1807 while (retcode == SS_NEED_CONNECTION) {
1808 spin_lock_irq(&mdev->req_lock);
1809 if (mdev->state.conn < C_CONNECTED)
1810 retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1811 spin_unlock_irq(&mdev->req_lock);
1812
1813 if (retcode != SS_NEED_CONNECTION)
1814 break;
1815
1816 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1817 }
1818
1819 reply->ret_code = retcode;
1820 return 0;
1821}
1822
0778286a
PR		 1823static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
1824{
1825 int rv;
1826
1827 rv = drbd_bmio_set_n_write(mdev);
1828 drbd_suspend_al(mdev);
1829 return rv;
1830}
1831
b411b363
PR		 1832static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1833 struct drbd_nl_cfg_reply *reply)
1834{
0778286a 1835 int retcode;
b411b363 1836
0778286a
PR		 1837 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
1838
1839 if (retcode < SS_SUCCESS) {
1840 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
1841 /* The peer will get a resync upon connect anyways. Just make that
1842 into a full resync. */
1843 retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
1844 if (retcode >= SS_SUCCESS) {
1845 /* open coded drbd_bitmap_io() */
1846 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
1847 "set_n_write from invalidate_peer"))
1848 retcode = ERR_IO_MD_DISK;
1849 }
1850 } else
1851 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
1852 }
b411b363 1853
0778286a 1854 reply->ret_code = retcode;
b411b363
PR		 1855 return 0;
1856}
1857
1858static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1859 struct drbd_nl_cfg_reply *reply)
1860{
1861 int retcode = NO_ERROR;
1862
1863 if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
1864 retcode = ERR_PAUSE_IS_SET;
1865
1866 reply->ret_code = retcode;
1867 return 0;
1868}
1869
1870static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1871 struct drbd_nl_cfg_reply *reply)
1872{
1873 int retcode = NO_ERROR;
1874
1875 if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO)
1876 retcode = ERR_PAUSE_IS_CLEAR;
1877
1878 reply->ret_code = retcode;
1879 return 0;
1880}
1881
1882static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1883 struct drbd_nl_cfg_reply *reply)
1884{
1885 reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
1886
1887 return 0;
1888}
1889
1890static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1891 struct drbd_nl_cfg_reply *reply)
1892{
43a5182c
PR		 1893 if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
1894 drbd_uuid_new_current(mdev);
1895 clear_bit(NEW_CUR_UUID, &mdev->flags);
1896 drbd_md_sync(mdev);
1897 }
265be2d0 1898 drbd_suspend_io(mdev);
b411b363 1899 reply->ret_code = drbd_request_state(mdev, NS(susp, 0));
265be2d0
PR		 1900 if (reply->ret_code == SS_SUCCESS) {
1901 if (mdev->state.conn < C_CONNECTED)
1902 tl_clear(mdev);
1903 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
1904 tl_restart(mdev, fail_frozen_disk_io);
1905 }
1906 drbd_resume_io(mdev);
1907
b411b363
PR		 1908 return 0;
1909}
1910
1911static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1912 struct drbd_nl_cfg_reply *reply)
1913{
1914 reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
1915 return 0;
1916}
1917
1918static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1919 struct drbd_nl_cfg_reply *reply)
1920{
1921 unsigned short *tl;
1922
1923 tl = reply->tag_list;
1924
1925 if (get_ldev(mdev)) {
1926 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
1927 put_ldev(mdev);
1928 }
1929
1930 if (get_net_conf(mdev)) {
1931 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
1932 put_net_conf(mdev);
1933 }
1934 tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
1935
1936 put_unaligned(TT_END, tl++); /* Close the tag list */
1937
1938 return (int)((char *)tl - (char *)reply->tag_list);
1939}
1940
1941static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1942 struct drbd_nl_cfg_reply *reply)
1943{
1944 unsigned short *tl = reply->tag_list;
1945 union drbd_state s = mdev->state;
1946 unsigned long rs_left;
1947 unsigned int res;
1948
1949 tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
1950
1951 /* no local ref, no bitmap, no syncer progress. */
1952 if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
1953 if (get_ldev(mdev)) {
1954 drbd_get_syncer_progress(mdev, &rs_left, &res);
1955 tl = tl_add_int(tl, T_sync_progress, &res);
1956 put_ldev(mdev);
1957 }
1958 }
1959 put_unaligned(TT_END, tl++); /* Close the tag list */
1960
1961 return (int)((char *)tl - (char *)reply->tag_list);
1962}
1963
1964static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1965 struct drbd_nl_cfg_reply *reply)
1966{
1967 unsigned short *tl;
1968
1969 tl = reply->tag_list;
1970
1971 if (get_ldev(mdev)) {
1972 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
1973 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
1974 put_ldev(mdev);
1975 }
1976 put_unaligned(TT_END, tl++); /* Close the tag list */
1977
1978 return (int)((char *)tl - (char *)reply->tag_list);
1979}
1980
1981/**
1982 * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
1983 * @mdev: DRBD device.
1984 * @nlp: Netlink/connector packet from drbdsetup
1985 * @reply: Reply packet for drbdsetup
1986 */
1987static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1988 struct drbd_nl_cfg_reply *reply)
1989{
1990 unsigned short *tl;
1991 char rv;
1992
1993 tl = reply->tag_list;
1994
1995 rv = mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
1996 test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
1997
1998 tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
1999 put_unaligned(TT_END, tl++); /* Close the tag list */
2000
2001 return (int)((char *)tl - (char *)reply->tag_list);
2002}
2003
2004static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2005 struct drbd_nl_cfg_reply *reply)
2006{
2007 /* default to resume from last known position, if possible */
2008 struct start_ov args =
2009 { .start_sector = mdev->ov_start_sector };
2010
2011 if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2012 reply->ret_code = ERR_MANDATORY_TAG;
2013 return 0;
2014 }
2015 /* w_make_ov_request expects position to be aligned */
2016 mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
2017 reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2018 return 0;
2019}
2020
2021
2022static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2023 struct drbd_nl_cfg_reply *reply)
2024{
2025 int retcode = NO_ERROR;
2026 int skip_initial_sync = 0;
2027 int err;
2028
2029 struct new_c_uuid args;
2030
2031 memset(&args, 0, sizeof(struct new_c_uuid));
2032 if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2033 reply->ret_code = ERR_MANDATORY_TAG;
2034 return 0;
2035 }
2036
2037 mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
2038
2039 if (!get_ldev(mdev)) {
2040 retcode = ERR_NO_DISK;
2041 goto out;
2042 }
2043
2044 /* this is "skip initial sync", assume to be clean */
2045 if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
2046 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2047 dev_info(DEV, "Preparing to skip initial sync\n");
2048 skip_initial_sync = 1;
2049 } else if (mdev->state.conn != C_STANDALONE) {
2050 retcode = ERR_CONNECTED;
2051 goto out_dec;
2052 }
2053
2054 drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2055 drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2056
2057 if (args.clear_bm) {
2058 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, "clear_n_write from new_c_uuid");
2059 if (err) {
2060 dev_err(DEV, "Writing bitmap failed with %d\n",err);
2061 retcode = ERR_IO_MD_DISK;
2062 }
2063 if (skip_initial_sync) {
2064 drbd_send_uuids_skip_initial_sync(mdev);
2065 _drbd_uuid_set(mdev, UI_BITMAP, 0);
2066 spin_lock_irq(&mdev->req_lock);
2067 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2068 CS_VERBOSE, NULL);
2069 spin_unlock_irq(&mdev->req_lock);
2070 }
2071 }
2072
2073 drbd_md_sync(mdev);
2074out_dec:
2075 put_ldev(mdev);
2076out:
2077 mutex_unlock(&mdev->state_mutex);
2078
2079 reply->ret_code = retcode;
2080 return 0;
2081}
2082
2083static struct drbd_conf *ensure_mdev(struct drbd_nl_cfg_req *nlp)
2084{
2085 struct drbd_conf *mdev;
2086
2087 if (nlp->drbd_minor >= minor_count)
2088 return NULL;
2089
2090 mdev = minor_to_mdev(nlp->drbd_minor);
2091
2092 if (!mdev && (nlp->flags & DRBD_NL_CREATE_DEVICE)) {
2093 struct gendisk *disk = NULL;
2094 mdev = drbd_new_device(nlp->drbd_minor);
2095
2096 spin_lock_irq(&drbd_pp_lock);
2097 if (minor_table[nlp->drbd_minor] == NULL) {
2098 minor_table[nlp->drbd_minor] = mdev;
2099 disk = mdev->vdisk;
2100 mdev = NULL;
2101 } /* else: we lost the race */
2102 spin_unlock_irq(&drbd_pp_lock);
2103
2104 if (disk) /* we won the race above */
2105 /* in case we ever add a drbd_delete_device(),
2106 * don't forget the del_gendisk! */
2107 add_disk(disk);
2108 else /* we lost the race above */
2109 drbd_free_mdev(mdev);
2110
2111 mdev = minor_to_mdev(nlp->drbd_minor);
2112 }
2113
2114 return mdev;
2115}
2116
2117struct cn_handler_struct {
2118 int (*function)(struct drbd_conf *,
2119 struct drbd_nl_cfg_req *,
2120 struct drbd_nl_cfg_reply *);
2121 int reply_body_size;
2122};
2123
2124static struct cn_handler_struct cnd_table[] = {
2125 [ P_primary ] = { &drbd_nl_primary, 0 },
2126 [ P_secondary ] = { &drbd_nl_secondary, 0 },
2127 [ P_disk_conf ] = { &drbd_nl_disk_conf, 0 },
2128 [ P_detach ] = { &drbd_nl_detach, 0 },
2129 [ P_net_conf ] = { &drbd_nl_net_conf, 0 },
2130 [ P_disconnect ] = { &drbd_nl_disconnect, 0 },
2131 [ P_resize ] = { &drbd_nl_resize, 0 },
2132 [ P_syncer_conf ] = { &drbd_nl_syncer_conf, 0 },
2133 [ P_invalidate ] = { &drbd_nl_invalidate, 0 },
2134 [ P_invalidate_peer ] = { &drbd_nl_invalidate_peer, 0 },
2135 [ P_pause_sync ] = { &drbd_nl_pause_sync, 0 },
2136 [ P_resume_sync ] = { &drbd_nl_resume_sync, 0 },
2137 [ P_suspend_io ] = { &drbd_nl_suspend_io, 0 },
2138 [ P_resume_io ] = { &drbd_nl_resume_io, 0 },
2139 [ P_outdate ] = { &drbd_nl_outdate, 0 },
2140 [ P_get_config ] = { &drbd_nl_get_config,
2141 sizeof(struct syncer_conf_tag_len_struct) +
2142 sizeof(struct disk_conf_tag_len_struct) +
2143 sizeof(struct net_conf_tag_len_struct) },
2144 [ P_get_state ] = { &drbd_nl_get_state,
2145 sizeof(struct get_state_tag_len_struct) +
2146 sizeof(struct sync_progress_tag_len_struct) },
2147 [ P_get_uuids ] = { &drbd_nl_get_uuids,
2148 sizeof(struct get_uuids_tag_len_struct) },
2149 [ P_get_timeout_flag ] = { &drbd_nl_get_timeout_flag,
2150 sizeof(struct get_timeout_flag_tag_len_struct)},
2151 [ P_start_ov ] = { &drbd_nl_start_ov, 0 },
2152 [ P_new_c_uuid ] = { &drbd_nl_new_c_uuid, 0 },
2153};
2154
9f5180e5 2155static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
b411b363
PR		 2156{
2157 struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2158 struct cn_handler_struct *cm;
2159 struct cn_msg *cn_reply;
2160 struct drbd_nl_cfg_reply *reply;
2161 struct drbd_conf *mdev;
2162 int retcode, rr;
2163 int reply_size = sizeof(struct cn_msg)
2164 + sizeof(struct drbd_nl_cfg_reply)
2165 + sizeof(short int);
2166
2167 if (!try_module_get(THIS_MODULE)) {
2168 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2169 return;
2170 }
2171
9f5180e5
PR		 2172 if (!cap_raised(nsp->eff_cap, CAP_SYS_ADMIN)) {
2173 retcode = ERR_PERM;
2174 goto fail;
2175 }
2176
b411b363
PR		 2177 mdev = ensure_mdev(nlp);
2178 if (!mdev) {
2179 retcode = ERR_MINOR_INVALID;
2180 goto fail;
2181 }
2182
b411b363
PR		 2183 if (nlp->packet_type >= P_nl_after_last_packet) {
2184 retcode = ERR_PACKET_NR;
2185 goto fail;
2186 }
2187
2188 cm = cnd_table + nlp->packet_type;
2189
2190 /* This may happen if packet number is 0: */
2191 if (cm->function == NULL) {
2192 retcode = ERR_PACKET_NR;
2193 goto fail;
2194 }
2195
2196 reply_size += cm->reply_body_size;
2197
2198 /* allocation not in the IO path, cqueue thread context */
2199 cn_reply = kmalloc(reply_size, GFP_KERNEL);
2200 if (!cn_reply) {
2201 retcode = ERR_NOMEM;
2202 goto fail;
2203 }
2204 reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2205
2206 reply->packet_type =
2207 cm->reply_body_size ? nlp->packet_type : P_nl_after_last_packet;
2208 reply->minor = nlp->drbd_minor;
2209 reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2210 /* reply->tag_list; might be modified by cm->function. */
2211
2212 rr = cm->function(mdev, nlp, reply);
2213
2214 cn_reply->id = req->id;
2215 cn_reply->seq = req->seq;
2216 cn_reply->ack = req->ack + 1;
2217 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2218 cn_reply->flags = 0;
2219
b411b363
PR		 2220 rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2221 if (rr && rr != -ESRCH)
2222 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2223
2224 kfree(cn_reply);
2225 module_put(THIS_MODULE);
2226 return;
2227 fail:
2228 drbd_nl_send_reply(req, retcode);
2229 module_put(THIS_MODULE);
2230}
2231
2232static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2233
2234static unsigned short *
2235__tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2236 unsigned short len, int nul_terminated)
2237{
2238 unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2239 len = (len < l) ? len : l;
2240 put_unaligned(tag, tl++);
2241 put_unaligned(len, tl++);
2242 memcpy(tl, data, len);
2243 tl = (unsigned short*)((char*)tl + len);
2244 if (nul_terminated)
2245 *((char*)tl - 1) = 0;
2246 return tl;
2247}
2248
2249static unsigned short *
2250tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2251{
2252 return __tl_add_blob(tl, tag, data, len, 0);
2253}
2254
2255static unsigned short *
2256tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2257{
2258 return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2259}
2260
2261static unsigned short *
2262tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2263{
2264 put_unaligned(tag, tl++);
2265 switch(tag_type(tag)) {
2266 case TT_INTEGER:
2267 put_unaligned(sizeof(int), tl++);
2268 put_unaligned(*(int *)val, (int *)tl);
2269 tl = (unsigned short*)((char*)tl+sizeof(int));
2270 break;
2271 case TT_INT64:
2272 put_unaligned(sizeof(u64), tl++);
2273 put_unaligned(*(u64 *)val, (u64 *)tl);
2274 tl = (unsigned short*)((char*)tl+sizeof(u64));
2275 break;
2276 default:
2277 /* someone did something stupid. */
2278 ;
2279 }
2280 return tl;
2281}
2282
2283void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2284{
2285 char buffer[sizeof(struct cn_msg)+
2286 sizeof(struct drbd_nl_cfg_reply)+
2287 sizeof(struct get_state_tag_len_struct)+
2288 sizeof(short int)];
2289 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2290 struct drbd_nl_cfg_reply *reply =
2291 (struct drbd_nl_cfg_reply *)cn_reply->data;
2292 unsigned short *tl = reply->tag_list;
2293
2294 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2295
2296 tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2297
2298 put_unaligned(TT_END, tl++); /* Close the tag list */
2299
2300 cn_reply->id.idx = CN_IDX_DRBD;
2301 cn_reply->id.val = CN_VAL_DRBD;
2302
2303 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2304 cn_reply->ack = 0; /* not used here. */
2305 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2306 (int)((char *)tl - (char *)reply->tag_list);
2307 cn_reply->flags = 0;
2308
2309 reply->packet_type = P_get_state;
2310 reply->minor = mdev_to_minor(mdev);
2311 reply->ret_code = NO_ERROR;
2312
b411b363
PR		 2313 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2314}
2315
2316void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2317{
2318 char buffer[sizeof(struct cn_msg)+
2319 sizeof(struct drbd_nl_cfg_reply)+
2320 sizeof(struct call_helper_tag_len_struct)+
2321 sizeof(short int)];
2322 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2323 struct drbd_nl_cfg_reply *reply =
2324 (struct drbd_nl_cfg_reply *)cn_reply->data;
2325 unsigned short *tl = reply->tag_list;
2326
2327 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2328
2329 tl = tl_add_str(tl, T_helper, helper_name);
2330 put_unaligned(TT_END, tl++); /* Close the tag list */
2331
2332 cn_reply->id.idx = CN_IDX_DRBD;
2333 cn_reply->id.val = CN_VAL_DRBD;
2334
2335 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2336 cn_reply->ack = 0; /* not used here. */
2337 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2338 (int)((char *)tl - (char *)reply->tag_list);
2339 cn_reply->flags = 0;
2340
2341 reply->packet_type = P_call_helper;
2342 reply->minor = mdev_to_minor(mdev);
2343 reply->ret_code = NO_ERROR;
2344
b411b363
PR		 2345 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2346}
2347
2348void drbd_bcast_ee(struct drbd_conf *mdev,
2349 const char *reason, const int dgs,
2350 const char* seen_hash, const char* calc_hash,
2351 const struct drbd_epoch_entry* e)
2352{
2353 struct cn_msg *cn_reply;
2354 struct drbd_nl_cfg_reply *reply;
b411b363 2355 unsigned short *tl;
45bb912b
LE		 2356 struct page *page;
2357 unsigned len;
b411b363
PR		 2358
2359 if (!e)
2360 return;
2361 if (!reason || !reason[0])
2362 return;
2363
2364 /* apparently we have to memcpy twice, first to prepare the data for the
2365 * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2366 * netlink skb. */
2367 /* receiver thread context, which is not in the writeout path (of this node),
2368 * but may be in the writeout path of the _other_ node.
2369 * GFP_NOIO to avoid potential "distributed deadlock". */
2370 cn_reply = kmalloc(
2371 sizeof(struct cn_msg)+
2372 sizeof(struct drbd_nl_cfg_reply)+
2373 sizeof(struct dump_ee_tag_len_struct)+
2374 sizeof(short int),
2375 GFP_NOIO);
2376
2377 if (!cn_reply) {
2378 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2379 (unsigned long long)e->sector, e->size);
2380 return;
2381 }
2382
2383 reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2384 tl = reply->tag_list;
2385
2386 tl = tl_add_str(tl, T_dump_ee_reason, reason);
2387 tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2388 tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2389 tl = tl_add_int(tl, T_ee_sector, &e->sector);
2390 tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2391
2392 put_unaligned(T_ee_data, tl++);
2393 put_unaligned(e->size, tl++);
2394
45bb912b
LE		 2395 len = e->size;
2396 page = e->pages;
2397 page_chain_for_each(page) {
2398 void *d = kmap_atomic(page, KM_USER0);
2399 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2400 memcpy(tl, d, l);
2401 kunmap_atomic(d, KM_USER0);
2402 tl = (unsigned short*)((char*)tl + l);
2403 len -= l;
b411b363
PR		 2404 }
2405 put_unaligned(TT_END, tl++); /* Close the tag list */
2406
2407 cn_reply->id.idx = CN_IDX_DRBD;
2408 cn_reply->id.val = CN_VAL_DRBD;
2409
2410 cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2411 cn_reply->ack = 0; // not used here.
2412 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2413 (int)((char*)tl - (char*)reply->tag_list);
2414 cn_reply->flags = 0;
2415
2416 reply->packet_type = P_dump_ee;
2417 reply->minor = mdev_to_minor(mdev);
2418 reply->ret_code = NO_ERROR;
2419
b411b363
PR		 2420 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2421 kfree(cn_reply);
2422}
2423
2424void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2425{
2426 char buffer[sizeof(struct cn_msg)+
2427 sizeof(struct drbd_nl_cfg_reply)+
2428 sizeof(struct sync_progress_tag_len_struct)+
2429 sizeof(short int)];
2430 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2431 struct drbd_nl_cfg_reply *reply =
2432 (struct drbd_nl_cfg_reply *)cn_reply->data;
2433 unsigned short *tl = reply->tag_list;
2434 unsigned long rs_left;
2435 unsigned int res;
2436
2437 /* no local ref, no bitmap, no syncer progress, no broadcast. */
2438 if (!get_ldev(mdev))
2439 return;
2440 drbd_get_syncer_progress(mdev, &rs_left, &res);
2441 put_ldev(mdev);
2442
2443 tl = tl_add_int(tl, T_sync_progress, &res);
2444 put_unaligned(TT_END, tl++); /* Close the tag list */
2445
2446 cn_reply->id.idx = CN_IDX_DRBD;
2447 cn_reply->id.val = CN_VAL_DRBD;
2448
2449 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2450 cn_reply->ack = 0; /* not used here. */
2451 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2452 (int)((char *)tl - (char *)reply->tag_list);
2453 cn_reply->flags = 0;
2454
2455 reply->packet_type = P_sync_progress;
2456 reply->minor = mdev_to_minor(mdev);
2457 reply->ret_code = NO_ERROR;
2458
b411b363
PR		 2459 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2460}
2461
2462int __init drbd_nl_init(void)
2463{
2464 static struct cb_id cn_id_drbd;
2465 int err, try=10;
2466
2467 cn_id_drbd.val = CN_VAL_DRBD;
2468 do {
2469 cn_id_drbd.idx = cn_idx;
2470 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2471 if (!err)
2472 break;
2473 cn_idx = (cn_idx + CN_IDX_STEP);
2474 } while (try--);
2475
2476 if (err) {
2477 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2478 return err;
2479 }
2480
2481 return 0;
2482}
2483
2484void drbd_nl_cleanup(void)
2485{
2486 static struct cb_id cn_id_drbd;
2487
2488 cn_id_drbd.idx = cn_idx;
2489 cn_id_drbd.val = CN_VAL_DRBD;
2490
2491 cn_del_callback(&cn_id_drbd);
2492}
2493
2494void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2495{
2496 char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2497 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2498 struct drbd_nl_cfg_reply *reply =
2499 (struct drbd_nl_cfg_reply *)cn_reply->data;
2500 int rr;
2501
2502 cn_reply->id = req->id;
2503
2504 cn_reply->seq = req->seq;
2505 cn_reply->ack = req->ack + 1;
2506 cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2507 cn_reply->flags = 0;
2508
2509 reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2510 reply->ret_code = ret_code;
2511
b411b363
PR		 2512 rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2513 if (rr && rr != -ESRCH)
2514 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2515}
2516
Ubuntu Artful kernel mirror