Merge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block
[sfrench/cifs-2.6.git] / drivers / block / drbd / drbd_nl.c
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
26 #include <linux/module.h>
27 #include <linux/drbd.h>
28 #include <linux/in.h>
29 #include <linux/fs.h>
30 #include <linux/file.h>
31 #include <linux/slab.h>
32 #include <linux/connector.h>
33 #include <linux/blkpg.h>
34 #include <linux/cpumask.h>
35 #include "drbd_int.h"
36 #include "drbd_wrappers.h"
37 #include <asm/unaligned.h>
38 #include <linux/drbd_tag_magic.h>
39 #include <linux/drbd_limits.h>
40
41 static unsigned short *tl_add_blob(unsigned short *, enum drbd_tags, const void *, int);
42 static unsigned short *tl_add_str(unsigned short *, enum drbd_tags, const char *);
43 static unsigned short *tl_add_int(unsigned short *, enum drbd_tags, const void *);
44
45 /* see get_sb_bdev and bd_claim */
46 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
47
48 /* Generate the tag_list to struct functions */
49 #define NL_PACKET(name, number, fields) \
50 static int name ## _from_tags(struct drbd_conf *mdev, \
51         unsigned short *tags, struct name *arg) __attribute__ ((unused)); \
52 static int name ## _from_tags(struct drbd_conf *mdev, \
53         unsigned short *tags, struct name *arg) \
54 { \
55         int tag; \
56         int dlen; \
57         \
58         while ((tag = get_unaligned(tags++)) != TT_END) {       \
59                 dlen = get_unaligned(tags++);                   \
60                 switch (tag_number(tag)) { \
61                 fields \
62                 default: \
63                         if (tag & T_MANDATORY) { \
64                                 dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \
65                                 return 0; \
66                         } \
67                 } \
68                 tags = (unsigned short *)((char *)tags + dlen); \
69         } \
70         return 1; \
71 }
72 #define NL_INTEGER(pn, pr, member) \
73         case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \
74                 arg->member = get_unaligned((int *)(tags));     \
75                 break;
76 #define NL_INT64(pn, pr, member) \
77         case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \
78                 arg->member = get_unaligned((u64 *)(tags));     \
79                 break;
80 #define NL_BIT(pn, pr, member) \
81         case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \
82                 arg->member = *(char *)(tags) ? 1 : 0; \
83                 break;
84 #define NL_STRING(pn, pr, member, len) \
85         case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \
86                 if (dlen > len) { \
87                         dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \
88                                 #member, dlen, (unsigned int)len); \
89                         return 0; \
90                 } \
91                  arg->member ## _len = dlen; \
92                  memcpy(arg->member, tags, min_t(size_t, dlen, len)); \
93                  break;
94 #include "linux/drbd_nl.h"
95
96 /* Generate the struct to tag_list functions */
97 #define NL_PACKET(name, number, fields) \
98 static unsigned short* \
99 name ## _to_tags(struct drbd_conf *mdev, \
100         struct name *arg, unsigned short *tags) __attribute__ ((unused)); \
101 static unsigned short* \
102 name ## _to_tags(struct drbd_conf *mdev, \
103         struct name *arg, unsigned short *tags) \
104 { \
105         fields \
106         return tags; \
107 }
108
109 #define NL_INTEGER(pn, pr, member) \
110         put_unaligned(pn | pr | TT_INTEGER, tags++);    \
111         put_unaligned(sizeof(int), tags++);             \
112         put_unaligned(arg->member, (int *)tags);        \
113         tags = (unsigned short *)((char *)tags+sizeof(int));
114 #define NL_INT64(pn, pr, member) \
115         put_unaligned(pn | pr | TT_INT64, tags++);      \
116         put_unaligned(sizeof(u64), tags++);             \
117         put_unaligned(arg->member, (u64 *)tags);        \
118         tags = (unsigned short *)((char *)tags+sizeof(u64));
119 #define NL_BIT(pn, pr, member) \
120         put_unaligned(pn | pr | TT_BIT, tags++);        \
121         put_unaligned(sizeof(char), tags++);            \
122         *(char *)tags = arg->member; \
123         tags = (unsigned short *)((char *)tags+sizeof(char));
124 #define NL_STRING(pn, pr, member, len) \
125         put_unaligned(pn | pr | TT_STRING, tags++);     \
126         put_unaligned(arg->member ## _len, tags++);     \
127         memcpy(tags, arg->member, arg->member ## _len); \
128         tags = (unsigned short *)((char *)tags + arg->member ## _len);
129 #include "linux/drbd_nl.h"
130
131 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name);
132 void drbd_nl_send_reply(struct cn_msg *, int);
133
134 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
135 {
136         char *envp[] = { "HOME=/",
137                         "TERM=linux",
138                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
139                         NULL, /* Will be set to address family */
140                         NULL, /* Will be set to address */
141                         NULL };
142
143         char mb[12], af[20], ad[60], *afs;
144         char *argv[] = {usermode_helper, cmd, mb, NULL };
145         int ret;
146
147         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
148
149         if (get_net_conf(mdev)) {
150                 switch (((struct sockaddr *)mdev->net_conf->peer_addr)->sa_family) {
151                 case AF_INET6:
152                         afs = "ipv6";
153                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6",
154                                  &((struct sockaddr_in6 *)mdev->net_conf->peer_addr)->sin6_addr);
155                         break;
156                 case AF_INET:
157                         afs = "ipv4";
158                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
159                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
160                         break;
161                 default:
162                         afs = "ssocks";
163                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
164                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
165                 }
166                 snprintf(af, 20, "DRBD_PEER_AF=%s", afs);
167                 envp[3]=af;
168                 envp[4]=ad;
169                 put_net_conf(mdev);
170         }
171
172         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
173
174         drbd_bcast_ev_helper(mdev, cmd);
175         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
176         if (ret)
177                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
178                                 usermode_helper, cmd, mb,
179                                 (ret >> 8) & 0xff, ret);
180         else
181                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
182                                 usermode_helper, cmd, mb,
183                                 (ret >> 8) & 0xff, ret);
184
185         if (ret < 0) /* Ignore any ERRNOs we got. */
186                 ret = 0;
187
188         return ret;
189 }
190
191 enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
192 {
193         char *ex_to_string;
194         int r;
195         enum drbd_disk_state nps;
196         enum drbd_fencing_p fp;
197
198         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
199
200         if (get_ldev_if_state(mdev, D_CONSISTENT)) {
201                 fp = mdev->ldev->dc.fencing;
202                 put_ldev(mdev);
203         } else {
204                 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n");
205                 return mdev->state.pdsk;
206         }
207
208         if (fp == FP_STONITH)
209                 _drbd_request_state(mdev, NS(susp, 1), CS_WAIT_COMPLETE);
210
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
258
259 int drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
260 {
261         const int max_tries = 4;
262         int r = 0;
263         int try = 0;
264         int forced = 0;
265         union drbd_state mask, val;
266         enum drbd_disk_state nps;
267
268         if (new_role == R_PRIMARY)
269                 request_ping(mdev); /* Detect a dead peer ASAP */
270
271         mutex_lock(&mdev->state_mutex);
272
273         mask.i = 0; mask.role = R_MASK;
274         val.i  = 0; val.role  = new_role;
275
276         while (try++ < max_tries) {
277                 r = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
278
279                 /* in case we first succeeded to outdate,
280                  * but now suddenly could establish a connection */
281                 if (r == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
282                         val.pdsk = 0;
283                         mask.pdsk = 0;
284                         continue;
285                 }
286
287                 if (r == SS_NO_UP_TO_DATE_DISK && force &&
288                     (mdev->state.disk < D_UP_TO_DATE &&
289                      mdev->state.disk >= D_INCONSISTENT)) {
290                         mask.disk = D_MASK;
291                         val.disk  = D_UP_TO_DATE;
292                         forced = 1;
293                         continue;
294                 }
295
296                 if (r == SS_NO_UP_TO_DATE_DISK &&
297                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
298                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
299                         nps = drbd_try_outdate_peer(mdev);
300
301                         if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
302                                 val.disk = D_UP_TO_DATE;
303                                 mask.disk = D_MASK;
304                         }
305
306                         val.pdsk = nps;
307                         mask.pdsk = D_MASK;
308
309                         continue;
310                 }
311
312                 if (r == SS_NOTHING_TO_DO)
313                         goto fail;
314                 if (r == SS_PRIMARY_NOP && mask.pdsk == 0) {
315                         nps = drbd_try_outdate_peer(mdev);
316
317                         if (force && nps > D_OUTDATED) {
318                                 dev_warn(DEV, "Forced into split brain situation!\n");
319                                 nps = D_OUTDATED;
320                         }
321
322                         mask.pdsk = D_MASK;
323                         val.pdsk  = nps;
324
325                         continue;
326                 }
327                 if (r == SS_TWO_PRIMARIES) {
328                         /* Maybe the peer is detected as dead very soon...
329                            retry at most once more in this case. */
330                         __set_current_state(TASK_INTERRUPTIBLE);
331                         schedule_timeout((mdev->net_conf->ping_timeo+1)*HZ/10);
332                         if (try < max_tries)
333                                 try = max_tries - 1;
334                         continue;
335                 }
336                 if (r < SS_SUCCESS) {
337                         r = _drbd_request_state(mdev, mask, val,
338                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
339                         if (r < SS_SUCCESS)
340                                 goto fail;
341                 }
342                 break;
343         }
344
345         if (r < SS_SUCCESS)
346                 goto fail;
347
348         if (forced)
349                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
350
351         /* Wait until nothing is on the fly :) */
352         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
353
354         if (new_role == R_SECONDARY) {
355                 set_disk_ro(mdev->vdisk, TRUE);
356                 if (get_ldev(mdev)) {
357                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
358                         put_ldev(mdev);
359                 }
360         } else {
361                 if (get_net_conf(mdev)) {
362                         mdev->net_conf->want_lose = 0;
363                         put_net_conf(mdev);
364                 }
365                 set_disk_ro(mdev->vdisk, FALSE);
366                 if (get_ldev(mdev)) {
367                         if (((mdev->state.conn < C_CONNECTED ||
368                                mdev->state.pdsk <= D_FAILED)
369                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
370                                 drbd_uuid_new_current(mdev);
371
372                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
373                         put_ldev(mdev);
374                 }
375         }
376
377         if ((new_role == R_SECONDARY) && get_ldev(mdev)) {
378                 drbd_al_to_on_disk_bm(mdev);
379                 put_ldev(mdev);
380         }
381
382         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
383                 /* if this was forced, we should consider sync */
384                 if (forced)
385                         drbd_send_uuids(mdev);
386                 drbd_send_state(mdev);
387         }
388
389         drbd_md_sync(mdev);
390
391         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
392  fail:
393         mutex_unlock(&mdev->state_mutex);
394         return r;
395 }
396
397
398 static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
399                            struct drbd_nl_cfg_reply *reply)
400 {
401         struct primary primary_args;
402
403         memset(&primary_args, 0, sizeof(struct primary));
404         if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) {
405                 reply->ret_code = ERR_MANDATORY_TAG;
406                 return 0;
407         }
408
409         reply->ret_code =
410                 drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force);
411
412         return 0;
413 }
414
415 static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
416                              struct drbd_nl_cfg_reply *reply)
417 {
418         reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0);
419
420         return 0;
421 }
422
423 /* initializes the md.*_offset members, so we are able to find
424  * the on disk meta data */
425 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
426                                        struct drbd_backing_dev *bdev)
427 {
428         sector_t md_size_sect = 0;
429         switch (bdev->dc.meta_dev_idx) {
430         default:
431                 /* v07 style fixed size indexed meta data */
432                 bdev->md.md_size_sect = MD_RESERVED_SECT;
433                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
434                 bdev->md.al_offset = MD_AL_OFFSET;
435                 bdev->md.bm_offset = MD_BM_OFFSET;
436                 break;
437         case DRBD_MD_INDEX_FLEX_EXT:
438                 /* just occupy the full device; unit: sectors */
439                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
440                 bdev->md.md_offset = 0;
441                 bdev->md.al_offset = MD_AL_OFFSET;
442                 bdev->md.bm_offset = MD_BM_OFFSET;
443                 break;
444         case DRBD_MD_INDEX_INTERNAL:
445         case DRBD_MD_INDEX_FLEX_INT:
446                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
447                 /* al size is still fixed */
448                 bdev->md.al_offset = -MD_AL_MAX_SIZE;
449                 /* we need (slightly less than) ~ this much bitmap sectors: */
450                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
451                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
452                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
453                 md_size_sect = ALIGN(md_size_sect, 8);
454
455                 /* plus the "drbd meta data super block",
456                  * and the activity log; */
457                 md_size_sect += MD_BM_OFFSET;
458
459                 bdev->md.md_size_sect = md_size_sect;
460                 /* bitmap offset is adjusted by 'super' block size */
461                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
462                 break;
463         }
464 }
465
466 char *ppsize(char *buf, unsigned long long size)
467 {
468         /* Needs 9 bytes at max. */
469         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
470         int base = 0;
471         while (size >= 10000) {
472                 /* shift + round */
473                 size = (size >> 10) + !!(size & (1<<9));
474                 base++;
475         }
476         sprintf(buf, "%lu %cB", (long)size, units[base]);
477
478         return buf;
479 }
480
481 /* there is still a theoretical deadlock when called from receiver
482  * on an D_INCONSISTENT R_PRIMARY:
483  *  remote READ does inc_ap_bio, receiver would need to receive answer
484  *  packet from remote to dec_ap_bio again.
485  *  receiver receive_sizes(), comes here,
486  *  waits for ap_bio_cnt == 0. -> deadlock.
487  * but this cannot happen, actually, because:
488  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
489  *  (not connected, or bad/no disk on peer):
490  *  see drbd_fail_request_early, ap_bio_cnt is zero.
491  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
492  *  peer may not initiate a resize.
493  */
494 void drbd_suspend_io(struct drbd_conf *mdev)
495 {
496         set_bit(SUSPEND_IO, &mdev->flags);
497         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
498 }
499
500 void drbd_resume_io(struct drbd_conf *mdev)
501 {
502         clear_bit(SUSPEND_IO, &mdev->flags);
503         wake_up(&mdev->misc_wait);
504 }
505
506 /**
507  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
508  * @mdev:       DRBD device.
509  *
510  * Returns 0 on success, negative return values indicate errors.
511  * You should call drbd_md_sync() after calling this function.
512  */
513 enum determine_dev_size drbd_determin_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
514 {
515         sector_t prev_first_sect, prev_size; /* previous meta location */
516         sector_t la_size;
517         sector_t size;
518         char ppb[10];
519
520         int md_moved, la_size_changed;
521         enum determine_dev_size rv = unchanged;
522
523         /* race:
524          * application request passes inc_ap_bio,
525          * but then cannot get an AL-reference.
526          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
527          *
528          * to avoid that:
529          * Suspend IO right here.
530          * still lock the act_log to not trigger ASSERTs there.
531          */
532         drbd_suspend_io(mdev);
533
534         /* no wait necessary anymore, actually we could assert that */
535         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
536
537         prev_first_sect = drbd_md_first_sector(mdev->ldev);
538         prev_size = mdev->ldev->md.md_size_sect;
539         la_size = mdev->ldev->md.la_size_sect;
540
541         /* TODO: should only be some assert here, not (re)init... */
542         drbd_md_set_sector_offsets(mdev, mdev->ldev);
543
544         size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
545
546         if (drbd_get_capacity(mdev->this_bdev) != size ||
547             drbd_bm_capacity(mdev) != size) {
548                 int err;
549                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
550                 if (unlikely(err)) {
551                         /* currently there is only one error: ENOMEM! */
552                         size = drbd_bm_capacity(mdev)>>1;
553                         if (size == 0) {
554                                 dev_err(DEV, "OUT OF MEMORY! "
555                                     "Could not allocate bitmap!\n");
556                         } else {
557                                 dev_err(DEV, "BM resizing failed. "
558                                     "Leaving size unchanged at size = %lu KB\n",
559                                     (unsigned long)size);
560                         }
561                         rv = dev_size_error;
562                 }
563                 /* racy, see comments above. */
564                 drbd_set_my_capacity(mdev, size);
565                 mdev->ldev->md.la_size_sect = size;
566                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
567                      (unsigned long long)size>>1);
568         }
569         if (rv == dev_size_error)
570                 goto out;
571
572         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
573
574         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
575                 || prev_size       != mdev->ldev->md.md_size_sect;
576
577         if (la_size_changed || md_moved) {
578                 drbd_al_shrink(mdev); /* All extents inactive. */
579                 dev_info(DEV, "Writing the whole bitmap, %s\n",
580                          la_size_changed && md_moved ? "size changed and md moved" :
581                          la_size_changed ? "size changed" : "md moved");
582                 rv = drbd_bitmap_io(mdev, &drbd_bm_write, "size changed"); /* does drbd_resume_io() ! */
583                 drbd_md_mark_dirty(mdev);
584         }
585
586         if (size > la_size)
587                 rv = grew;
588         if (size < la_size)
589                 rv = shrunk;
590 out:
591         lc_unlock(mdev->act_log);
592         wake_up(&mdev->al_wait);
593         drbd_resume_io(mdev);
594
595         return rv;
596 }
597
598 sector_t
599 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
600 {
601         sector_t p_size = mdev->p_size;   /* partner's disk size. */
602         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
603         sector_t m_size; /* my size */
604         sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
605         sector_t size = 0;
606
607         m_size = drbd_get_max_capacity(bdev);
608
609         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
610                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
611                 p_size = m_size;
612         }
613
614         if (p_size && m_size) {
615                 size = min_t(sector_t, p_size, m_size);
616         } else {
617                 if (la_size) {
618                         size = la_size;
619                         if (m_size && m_size < size)
620                                 size = m_size;
621                         if (p_size && p_size < size)
622                                 size = p_size;
623                 } else {
624                         if (m_size)
625                                 size = m_size;
626                         if (p_size)
627                                 size = p_size;
628                 }
629         }
630
631         if (size == 0)
632                 dev_err(DEV, "Both nodes diskless!\n");
633
634         if (u_size) {
635                 if (u_size > size)
636                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
637                             (unsigned long)u_size>>1, (unsigned long)size>>1);
638                 else
639                         size = u_size;
640         }
641
642         return size;
643 }
644
645 /**
646  * drbd_check_al_size() - Ensures that the AL is of the right size
647  * @mdev:       DRBD device.
648  *
649  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
650  * failed, and 0 on success. You should call drbd_md_sync() after you called
651  * this function.
652  */
653 static int drbd_check_al_size(struct drbd_conf *mdev)
654 {
655         struct lru_cache *n, *t;
656         struct lc_element *e;
657         unsigned int in_use;
658         int i;
659
660         ERR_IF(mdev->sync_conf.al_extents < 7)
661                 mdev->sync_conf.al_extents = 127;
662
663         if (mdev->act_log &&
664             mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
665                 return 0;
666
667         in_use = 0;
668         t = mdev->act_log;
669         n = lc_create("act_log", drbd_al_ext_cache,
670                 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
671
672         if (n == NULL) {
673                 dev_err(DEV, "Cannot allocate act_log lru!\n");
674                 return -ENOMEM;
675         }
676         spin_lock_irq(&mdev->al_lock);
677         if (t) {
678                 for (i = 0; i < t->nr_elements; i++) {
679                         e = lc_element_by_index(t, i);
680                         if (e->refcnt)
681                                 dev_err(DEV, "refcnt(%d)==%d\n",
682                                     e->lc_number, e->refcnt);
683                         in_use += e->refcnt;
684                 }
685         }
686         if (!in_use)
687                 mdev->act_log = n;
688         spin_unlock_irq(&mdev->al_lock);
689         if (in_use) {
690                 dev_err(DEV, "Activity log still in use!\n");
691                 lc_destroy(n);
692                 return -EBUSY;
693         } else {
694                 if (t)
695                         lc_destroy(t);
696         }
697         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
698         return 0;
699 }
700
701 void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_seg_s) __must_hold(local)
702 {
703         struct request_queue * const q = mdev->rq_queue;
704         struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
705         int max_segments = mdev->ldev->dc.max_bio_bvecs;
706
707         max_seg_s = min(queue_max_sectors(b) * queue_logical_block_size(b), max_seg_s);
708
709         blk_queue_max_hw_sectors(q, max_seg_s >> 9);
710         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
711         blk_queue_max_segment_size(q, max_seg_s);
712         blk_queue_logical_block_size(q, 512);
713         blk_queue_segment_boundary(q, PAGE_SIZE-1);
714         blk_stack_limits(&q->limits, &b->limits, 0);
715
716         if (b->merge_bvec_fn)
717                 dev_warn(DEV, "Backing device's merge_bvec_fn() = %p\n",
718                      b->merge_bvec_fn);
719         dev_info(DEV, "max_segment_size ( = BIO size ) = %u\n", queue_max_segment_size(q));
720
721         if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
722                 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
723                      q->backing_dev_info.ra_pages,
724                      b->backing_dev_info.ra_pages);
725                 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
726         }
727 }
728
729 /* serialize deconfig (worker exiting, doing cleanup)
730  * and reconfig (drbdsetup disk, drbdsetup net)
731  *
732  * wait for a potentially exiting worker, then restart it,
733  * or start a new one.
734  */
735 static void drbd_reconfig_start(struct drbd_conf *mdev)
736 {
737         wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
738         wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
739         drbd_thread_start(&mdev->worker);
740 }
741
742 /* if still unconfigured, stops worker again.
743  * if configured now, clears CONFIG_PENDING.
744  * wakes potential waiters */
745 static void drbd_reconfig_done(struct drbd_conf *mdev)
746 {
747         spin_lock_irq(&mdev->req_lock);
748         if (mdev->state.disk == D_DISKLESS &&
749             mdev->state.conn == C_STANDALONE &&
750             mdev->state.role == R_SECONDARY) {
751                 set_bit(DEVICE_DYING, &mdev->flags);
752                 drbd_thread_stop_nowait(&mdev->worker);
753         } else
754                 clear_bit(CONFIG_PENDING, &mdev->flags);
755         spin_unlock_irq(&mdev->req_lock);
756         wake_up(&mdev->state_wait);
757 }
758
759 /* does always return 0;
760  * interesting return code is in reply->ret_code */
761 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
762                              struct drbd_nl_cfg_reply *reply)
763 {
764         enum drbd_ret_codes retcode;
765         enum determine_dev_size dd;
766         sector_t max_possible_sectors;
767         sector_t min_md_device_sectors;
768         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
769         struct inode *inode, *inode2;
770         struct lru_cache *resync_lru = NULL;
771         union drbd_state ns, os;
772         int rv;
773         int cp_discovered = 0;
774         int logical_block_size;
775
776         drbd_reconfig_start(mdev);
777
778         /* if you want to reconfigure, please tear down first */
779         if (mdev->state.disk > D_DISKLESS) {
780                 retcode = ERR_DISK_CONFIGURED;
781                 goto fail;
782         }
783
784         /* allocation not in the IO path, cqueue thread context */
785         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
786         if (!nbc) {
787                 retcode = ERR_NOMEM;
788                 goto fail;
789         }
790
791         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
792         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
793         nbc->dc.fencing       = DRBD_FENCING_DEF;
794         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
795
796         if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
797                 retcode = ERR_MANDATORY_TAG;
798                 goto fail;
799         }
800
801         if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
802                 retcode = ERR_MD_IDX_INVALID;
803                 goto fail;
804         }
805
806         nbc->lo_file = filp_open(nbc->dc.backing_dev, O_RDWR, 0);
807         if (IS_ERR(nbc->lo_file)) {
808                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
809                     PTR_ERR(nbc->lo_file));
810                 nbc->lo_file = NULL;
811                 retcode = ERR_OPEN_DISK;
812                 goto fail;
813         }
814
815         inode = nbc->lo_file->f_dentry->d_inode;
816
817         if (!S_ISBLK(inode->i_mode)) {
818                 retcode = ERR_DISK_NOT_BDEV;
819                 goto fail;
820         }
821
822         nbc->md_file = filp_open(nbc->dc.meta_dev, O_RDWR, 0);
823         if (IS_ERR(nbc->md_file)) {
824                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
825                     PTR_ERR(nbc->md_file));
826                 nbc->md_file = NULL;
827                 retcode = ERR_OPEN_MD_DISK;
828                 goto fail;
829         }
830
831         inode2 = nbc->md_file->f_dentry->d_inode;
832
833         if (!S_ISBLK(inode2->i_mode)) {
834                 retcode = ERR_MD_NOT_BDEV;
835                 goto fail;
836         }
837
838         nbc->backing_bdev = inode->i_bdev;
839         if (bd_claim(nbc->backing_bdev, mdev)) {
840                 printk(KERN_ERR "drbd: bd_claim(%p,%p); failed [%p;%p;%u]\n",
841                        nbc->backing_bdev, mdev,
842                        nbc->backing_bdev->bd_holder,
843                        nbc->backing_bdev->bd_contains->bd_holder,
844                        nbc->backing_bdev->bd_holders);
845                 retcode = ERR_BDCLAIM_DISK;
846                 goto fail;
847         }
848
849         resync_lru = lc_create("resync", drbd_bm_ext_cache,
850                         61, sizeof(struct bm_extent),
851                         offsetof(struct bm_extent, lce));
852         if (!resync_lru) {
853                 retcode = ERR_NOMEM;
854                 goto release_bdev_fail;
855         }
856
857         /* meta_dev_idx >= 0: external fixed size,
858          * possibly multiple drbd sharing one meta device.
859          * TODO in that case, paranoia check that [md_bdev, meta_dev_idx] is
860          * not yet used by some other drbd minor!
861          * (if you use drbd.conf + drbdadm,
862          * that should check it for you already; but if you don't, or someone
863          * fooled it, we need to double check here) */
864         nbc->md_bdev = inode2->i_bdev;
865         if (bd_claim(nbc->md_bdev, (nbc->dc.meta_dev_idx < 0) ? (void *)mdev
866                                 : (void *) drbd_m_holder)) {
867                 retcode = ERR_BDCLAIM_MD_DISK;
868                 goto release_bdev_fail;
869         }
870
871         if ((nbc->backing_bdev == nbc->md_bdev) !=
872             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
873              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
874                 retcode = ERR_MD_IDX_INVALID;
875                 goto release_bdev2_fail;
876         }
877
878         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
879         drbd_md_set_sector_offsets(mdev, nbc);
880
881         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
882                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
883                         (unsigned long long) drbd_get_max_capacity(nbc),
884                         (unsigned long long) nbc->dc.disk_size);
885                 retcode = ERR_DISK_TO_SMALL;
886                 goto release_bdev2_fail;
887         }
888
889         if (nbc->dc.meta_dev_idx < 0) {
890                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
891                 /* at least one MB, otherwise it does not make sense */
892                 min_md_device_sectors = (2<<10);
893         } else {
894                 max_possible_sectors = DRBD_MAX_SECTORS;
895                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
896         }
897
898         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
899                 retcode = ERR_MD_DISK_TO_SMALL;
900                 dev_warn(DEV, "refusing attach: md-device too small, "
901                      "at least %llu sectors needed for this meta-disk type\n",
902                      (unsigned long long) min_md_device_sectors);
903                 goto release_bdev2_fail;
904         }
905
906         /* Make sure the new disk is big enough
907          * (we may currently be R_PRIMARY with no local disk...) */
908         if (drbd_get_max_capacity(nbc) <
909             drbd_get_capacity(mdev->this_bdev)) {
910                 retcode = ERR_DISK_TO_SMALL;
911                 goto release_bdev2_fail;
912         }
913
914         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
915
916         if (nbc->known_size > max_possible_sectors) {
917                 dev_warn(DEV, "==> truncating very big lower level device "
918                         "to currently maximum possible %llu sectors <==\n",
919                         (unsigned long long) max_possible_sectors);
920                 if (nbc->dc.meta_dev_idx >= 0)
921                         dev_warn(DEV, "==>> using internal or flexible "
922                                       "meta data may help <<==\n");
923         }
924
925         drbd_suspend_io(mdev);
926         /* also wait for the last barrier ack. */
927         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt));
928         /* and for any other previously queued work */
929         drbd_flush_workqueue(mdev);
930
931         retcode = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
932         drbd_resume_io(mdev);
933         if (retcode < SS_SUCCESS)
934                 goto release_bdev2_fail;
935
936         if (!get_ldev_if_state(mdev, D_ATTACHING))
937                 goto force_diskless;
938
939         drbd_md_set_sector_offsets(mdev, nbc);
940
941         /* allocate a second IO page if logical_block_size != 512 */
942         logical_block_size = bdev_logical_block_size(nbc->md_bdev);
943         if (logical_block_size == 0)
944                 logical_block_size = MD_SECTOR_SIZE;
945
946         if (logical_block_size != MD_SECTOR_SIZE) {
947                 if (!mdev->md_io_tmpp) {
948                         struct page *page = alloc_page(GFP_NOIO);
949                         if (!page)
950                                 goto force_diskless_dec;
951
952                         dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
953                              logical_block_size, MD_SECTOR_SIZE);
954                         dev_warn(DEV, "Workaround engaged (has performance impact).\n");
955
956                         mdev->md_io_tmpp = page;
957                 }
958         }
959
960         if (!mdev->bitmap) {
961                 if (drbd_bm_init(mdev)) {
962                         retcode = ERR_NOMEM;
963                         goto force_diskless_dec;
964                 }
965         }
966
967         retcode = drbd_md_read(mdev, nbc);
968         if (retcode != NO_ERROR)
969                 goto force_diskless_dec;
970
971         if (mdev->state.conn < C_CONNECTED &&
972             mdev->state.role == R_PRIMARY &&
973             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
974                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
975                     (unsigned long long)mdev->ed_uuid);
976                 retcode = ERR_DATA_NOT_CURRENT;
977                 goto force_diskless_dec;
978         }
979
980         /* Since we are diskless, fix the activity log first... */
981         if (drbd_check_al_size(mdev)) {
982                 retcode = ERR_NOMEM;
983                 goto force_diskless_dec;
984         }
985
986         /* Prevent shrinking of consistent devices ! */
987         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
988             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
989                 dev_warn(DEV, "refusing to truncate a consistent device\n");
990                 retcode = ERR_DISK_TO_SMALL;
991                 goto force_diskless_dec;
992         }
993
994         if (!drbd_al_read_log(mdev, nbc)) {
995                 retcode = ERR_IO_MD_DISK;
996                 goto force_diskless_dec;
997         }
998
999         /* Reset the "barriers don't work" bits here, then force meta data to
1000          * be written, to ensure we determine if barriers are supported. */
1001         if (nbc->dc.no_md_flush)
1002                 set_bit(MD_NO_BARRIER, &mdev->flags);
1003         else
1004                 clear_bit(MD_NO_BARRIER, &mdev->flags);
1005
1006         /* Point of no return reached.
1007          * Devices and memory are no longer released by error cleanup below.
1008          * now mdev takes over responsibility, and the state engine should
1009          * clean it up somewhere.  */
1010         D_ASSERT(mdev->ldev == NULL);
1011         mdev->ldev = nbc;
1012         mdev->resync = resync_lru;
1013         nbc = NULL;
1014         resync_lru = NULL;
1015
1016         mdev->write_ordering = WO_bio_barrier;
1017         drbd_bump_write_ordering(mdev, WO_bio_barrier);
1018
1019         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1020                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1021         else
1022                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1023
1024         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND)) {
1025                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1026                 cp_discovered = 1;
1027         }
1028
1029         mdev->send_cnt = 0;
1030         mdev->recv_cnt = 0;
1031         mdev->read_cnt = 0;
1032         mdev->writ_cnt = 0;
1033
1034         drbd_setup_queue_param(mdev, DRBD_MAX_SEGMENT_SIZE);
1035
1036         /* If I am currently not R_PRIMARY,
1037          * but meta data primary indicator is set,
1038          * I just now recover from a hard crash,
1039          * and have been R_PRIMARY before that crash.
1040          *
1041          * Now, if I had no connection before that crash
1042          * (have been degraded R_PRIMARY), chances are that
1043          * I won't find my peer now either.
1044          *
1045          * In that case, and _only_ in that case,
1046          * we use the degr-wfc-timeout instead of the default,
1047          * so we can automatically recover from a crash of a
1048          * degraded but active "cluster" after a certain timeout.
1049          */
1050         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1051         if (mdev->state.role != R_PRIMARY &&
1052              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1053             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1054                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1055
1056         dd = drbd_determin_dev_size(mdev, 0);
1057         if (dd == dev_size_error) {
1058                 retcode = ERR_NOMEM_BITMAP;
1059                 goto force_diskless_dec;
1060         } else if (dd == grew)
1061                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1062
1063         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1064                 dev_info(DEV, "Assuming that all blocks are out of sync "
1065                      "(aka FullSync)\n");
1066                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from attaching")) {
1067                         retcode = ERR_IO_MD_DISK;
1068                         goto force_diskless_dec;
1069                 }
1070         } else {
1071                 if (drbd_bitmap_io(mdev, &drbd_bm_read, "read from attaching") < 0) {
1072                         retcode = ERR_IO_MD_DISK;
1073                         goto force_diskless_dec;
1074                 }
1075         }
1076
1077         if (cp_discovered) {
1078                 drbd_al_apply_to_bm(mdev);
1079                 drbd_al_to_on_disk_bm(mdev);
1080         }
1081
1082         spin_lock_irq(&mdev->req_lock);
1083         os = mdev->state;
1084         ns.i = os.i;
1085         /* If MDF_CONSISTENT is not set go into inconsistent state,
1086            otherwise investigate MDF_WasUpToDate...
1087            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1088            otherwise into D_CONSISTENT state.
1089         */
1090         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1091                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1092                         ns.disk = D_CONSISTENT;
1093                 else
1094                         ns.disk = D_OUTDATED;
1095         } else {
1096                 ns.disk = D_INCONSISTENT;
1097         }
1098
1099         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1100                 ns.pdsk = D_OUTDATED;
1101
1102         if ( ns.disk == D_CONSISTENT &&
1103             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1104                 ns.disk = D_UP_TO_DATE;
1105
1106         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1107            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1108            this point, because drbd_request_state() modifies these
1109            flags. */
1110
1111         /* In case we are C_CONNECTED postpone any decision on the new disk
1112            state after the negotiation phase. */
1113         if (mdev->state.conn == C_CONNECTED) {
1114                 mdev->new_state_tmp.i = ns.i;
1115                 ns.i = os.i;
1116                 ns.disk = D_NEGOTIATING;
1117
1118                 /* We expect to receive up-to-date UUIDs soon.
1119                    To avoid a race in receive_state, free p_uuid while
1120                    holding req_lock. I.e. atomic with the state change */
1121                 kfree(mdev->p_uuid);
1122                 mdev->p_uuid = NULL;
1123         }
1124
1125         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1126         ns = mdev->state;
1127         spin_unlock_irq(&mdev->req_lock);
1128
1129         if (rv < SS_SUCCESS)
1130                 goto force_diskless_dec;
1131
1132         if (mdev->state.role == R_PRIMARY)
1133                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1134         else
1135                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1136
1137         drbd_md_mark_dirty(mdev);
1138         drbd_md_sync(mdev);
1139
1140         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1141         put_ldev(mdev);
1142         reply->ret_code = retcode;
1143         drbd_reconfig_done(mdev);
1144         return 0;
1145
1146  force_diskless_dec:
1147         put_ldev(mdev);
1148  force_diskless:
1149         drbd_force_state(mdev, NS(disk, D_DISKLESS));
1150         drbd_md_sync(mdev);
1151  release_bdev2_fail:
1152         if (nbc)
1153                 bd_release(nbc->md_bdev);
1154  release_bdev_fail:
1155         if (nbc)
1156                 bd_release(nbc->backing_bdev);
1157  fail:
1158         if (nbc) {
1159                 if (nbc->lo_file)
1160                         fput(nbc->lo_file);
1161                 if (nbc->md_file)
1162                         fput(nbc->md_file);
1163                 kfree(nbc);
1164         }
1165         lc_destroy(resync_lru);
1166
1167         reply->ret_code = retcode;
1168         drbd_reconfig_done(mdev);
1169         return 0;
1170 }
1171
1172 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1173                           struct drbd_nl_cfg_reply *reply)
1174 {
1175         reply->ret_code = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1176         return 0;
1177 }
1178
1179 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1180                             struct drbd_nl_cfg_reply *reply)
1181 {
1182         int i, ns;
1183         enum drbd_ret_codes retcode;
1184         struct net_conf *new_conf = NULL;
1185         struct crypto_hash *tfm = NULL;
1186         struct crypto_hash *integrity_w_tfm = NULL;
1187         struct crypto_hash *integrity_r_tfm = NULL;
1188         struct hlist_head *new_tl_hash = NULL;
1189         struct hlist_head *new_ee_hash = NULL;
1190         struct drbd_conf *odev;
1191         char hmac_name[CRYPTO_MAX_ALG_NAME];
1192         void *int_dig_out = NULL;
1193         void *int_dig_in = NULL;
1194         void *int_dig_vv = NULL;
1195         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1196
1197         drbd_reconfig_start(mdev);
1198
1199         if (mdev->state.conn > C_STANDALONE) {
1200                 retcode = ERR_NET_CONFIGURED;
1201                 goto fail;
1202         }
1203
1204         /* allocation not in the IO path, cqueue thread context */
1205         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1206         if (!new_conf) {
1207                 retcode = ERR_NOMEM;
1208                 goto fail;
1209         }
1210
1211         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1212         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1213         new_conf->ping_int         = DRBD_PING_INT_DEF;
1214         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1215         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1216         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1217         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1218         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1219         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1220         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1221         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1222         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1223         new_conf->want_lose        = 0;
1224         new_conf->two_primaries    = 0;
1225         new_conf->wire_protocol    = DRBD_PROT_C;
1226         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1227         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1228
1229         if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1230                 retcode = ERR_MANDATORY_TAG;
1231                 goto fail;
1232         }
1233
1234         if (new_conf->two_primaries
1235             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1236                 retcode = ERR_NOT_PROTO_C;
1237                 goto fail;
1238         };
1239
1240         if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1241                 retcode = ERR_DISCARD;
1242                 goto fail;
1243         }
1244
1245         retcode = NO_ERROR;
1246
1247         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1248         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1249         for (i = 0; i < minor_count; i++) {
1250                 odev = minor_to_mdev(i);
1251                 if (!odev || odev == mdev)
1252                         continue;
1253                 if (get_net_conf(odev)) {
1254                         taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1255                         if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1256                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1257                                 retcode = ERR_LOCAL_ADDR;
1258
1259                         taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1260                         if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1261                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1262                                 retcode = ERR_PEER_ADDR;
1263
1264                         put_net_conf(odev);
1265                         if (retcode != NO_ERROR)
1266                                 goto fail;
1267                 }
1268         }
1269
1270         if (new_conf->cram_hmac_alg[0] != 0) {
1271                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1272                         new_conf->cram_hmac_alg);
1273                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1274                 if (IS_ERR(tfm)) {
1275                         tfm = NULL;
1276                         retcode = ERR_AUTH_ALG;
1277                         goto fail;
1278                 }
1279
1280                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1281                         retcode = ERR_AUTH_ALG_ND;
1282                         goto fail;
1283                 }
1284         }
1285
1286         if (new_conf->integrity_alg[0]) {
1287                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1288                 if (IS_ERR(integrity_w_tfm)) {
1289                         integrity_w_tfm = NULL;
1290                         retcode=ERR_INTEGRITY_ALG;
1291                         goto fail;
1292                 }
1293
1294                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1295                         retcode=ERR_INTEGRITY_ALG_ND;
1296                         goto fail;
1297                 }
1298
1299                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1300                 if (IS_ERR(integrity_r_tfm)) {
1301                         integrity_r_tfm = NULL;
1302                         retcode=ERR_INTEGRITY_ALG;
1303                         goto fail;
1304                 }
1305         }
1306
1307         ns = new_conf->max_epoch_size/8;
1308         if (mdev->tl_hash_s != ns) {
1309                 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1310                 if (!new_tl_hash) {
1311                         retcode = ERR_NOMEM;
1312                         goto fail;
1313                 }
1314         }
1315
1316         ns = new_conf->max_buffers/8;
1317         if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1318                 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1319                 if (!new_ee_hash) {
1320                         retcode = ERR_NOMEM;
1321                         goto fail;
1322                 }
1323         }
1324
1325         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1326
1327         if (integrity_w_tfm) {
1328                 i = crypto_hash_digestsize(integrity_w_tfm);
1329                 int_dig_out = kmalloc(i, GFP_KERNEL);
1330                 if (!int_dig_out) {
1331                         retcode = ERR_NOMEM;
1332                         goto fail;
1333                 }
1334                 int_dig_in = kmalloc(i, GFP_KERNEL);
1335                 if (!int_dig_in) {
1336                         retcode = ERR_NOMEM;
1337                         goto fail;
1338                 }
1339                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1340                 if (!int_dig_vv) {
1341                         retcode = ERR_NOMEM;
1342                         goto fail;
1343                 }
1344         }
1345
1346         if (!mdev->bitmap) {
1347                 if(drbd_bm_init(mdev)) {
1348                         retcode = ERR_NOMEM;
1349                         goto fail;
1350                 }
1351         }
1352
1353         spin_lock_irq(&mdev->req_lock);
1354         if (mdev->net_conf != NULL) {
1355                 retcode = ERR_NET_CONFIGURED;
1356                 spin_unlock_irq(&mdev->req_lock);
1357                 goto fail;
1358         }
1359         mdev->net_conf = new_conf;
1360
1361         mdev->send_cnt = 0;
1362         mdev->recv_cnt = 0;
1363
1364         if (new_tl_hash) {
1365                 kfree(mdev->tl_hash);
1366                 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1367                 mdev->tl_hash = new_tl_hash;
1368         }
1369
1370         if (new_ee_hash) {
1371                 kfree(mdev->ee_hash);
1372                 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1373                 mdev->ee_hash = new_ee_hash;
1374         }
1375
1376         crypto_free_hash(mdev->cram_hmac_tfm);
1377         mdev->cram_hmac_tfm = tfm;
1378
1379         crypto_free_hash(mdev->integrity_w_tfm);
1380         mdev->integrity_w_tfm = integrity_w_tfm;
1381
1382         crypto_free_hash(mdev->integrity_r_tfm);
1383         mdev->integrity_r_tfm = integrity_r_tfm;
1384
1385         kfree(mdev->int_dig_out);
1386         kfree(mdev->int_dig_in);
1387         kfree(mdev->int_dig_vv);
1388         mdev->int_dig_out=int_dig_out;
1389         mdev->int_dig_in=int_dig_in;
1390         mdev->int_dig_vv=int_dig_vv;
1391         spin_unlock_irq(&mdev->req_lock);
1392
1393         retcode = _drbd_request_state(mdev, NS(conn, C_UNCONNECTED), CS_VERBOSE);
1394
1395         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1396         reply->ret_code = retcode;
1397         drbd_reconfig_done(mdev);
1398         return 0;
1399
1400 fail:
1401         kfree(int_dig_out);
1402         kfree(int_dig_in);
1403         kfree(int_dig_vv);
1404         crypto_free_hash(tfm);
1405         crypto_free_hash(integrity_w_tfm);
1406         crypto_free_hash(integrity_r_tfm);
1407         kfree(new_tl_hash);
1408         kfree(new_ee_hash);
1409         kfree(new_conf);
1410
1411         reply->ret_code = retcode;
1412         drbd_reconfig_done(mdev);
1413         return 0;
1414 }
1415
1416 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1417                               struct drbd_nl_cfg_reply *reply)
1418 {
1419         int retcode;
1420
1421         retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1422
1423         if (retcode == SS_NOTHING_TO_DO)
1424                 goto done;
1425         else if (retcode == SS_ALREADY_STANDALONE)
1426                 goto done;
1427         else if (retcode == SS_PRIMARY_NOP) {
1428                 /* Our statche checking code wants to see the peer outdated. */
1429                 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1430                                                       pdsk, D_OUTDATED));
1431         } else if (retcode == SS_CW_FAILED_BY_PEER) {
1432                 /* The peer probably wants to see us outdated. */
1433                 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1434                                                         disk, D_OUTDATED),
1435                                               CS_ORDERED);
1436                 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1437                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1438                         retcode = SS_SUCCESS;
1439                 }
1440         }
1441
1442         if (retcode < SS_SUCCESS)
1443                 goto fail;
1444
1445         if (wait_event_interruptible(mdev->state_wait,
1446                                      mdev->state.conn != C_DISCONNECTING)) {
1447                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1448                    someone else might connect us in the mean time! */
1449                 retcode = ERR_INTR;
1450                 goto fail;
1451         }
1452
1453  done:
1454         retcode = NO_ERROR;
1455  fail:
1456         drbd_md_sync(mdev);
1457         reply->ret_code = retcode;
1458         return 0;
1459 }
1460
1461 void resync_after_online_grow(struct drbd_conf *mdev)
1462 {
1463         int iass; /* I am sync source */
1464
1465         dev_info(DEV, "Resync of new storage after online grow\n");
1466         if (mdev->state.role != mdev->state.peer)
1467                 iass = (mdev->state.role == R_PRIMARY);
1468         else
1469                 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1470
1471         if (iass)
1472                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1473         else
1474                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1475 }
1476
1477 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1478                           struct drbd_nl_cfg_reply *reply)
1479 {
1480         struct resize rs;
1481         int retcode = NO_ERROR;
1482         enum determine_dev_size dd;
1483         enum dds_flags ddsf;
1484
1485         memset(&rs, 0, sizeof(struct resize));
1486         if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1487                 retcode = ERR_MANDATORY_TAG;
1488                 goto fail;
1489         }
1490
1491         if (mdev->state.conn > C_CONNECTED) {
1492                 retcode = ERR_RESIZE_RESYNC;
1493                 goto fail;
1494         }
1495
1496         if (mdev->state.role == R_SECONDARY &&
1497             mdev->state.peer == R_SECONDARY) {
1498                 retcode = ERR_NO_PRIMARY;
1499                 goto fail;
1500         }
1501
1502         if (!get_ldev(mdev)) {
1503                 retcode = ERR_NO_DISK;
1504                 goto fail;
1505         }
1506
1507         if (rs.no_resync && mdev->agreed_pro_version < 93) {
1508                 retcode = ERR_NEED_APV_93;
1509                 goto fail;
1510         }
1511
1512         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1513                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1514
1515         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1516         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1517         dd = drbd_determin_dev_size(mdev, ddsf);
1518         drbd_md_sync(mdev);
1519         put_ldev(mdev);
1520         if (dd == dev_size_error) {
1521                 retcode = ERR_NOMEM_BITMAP;
1522                 goto fail;
1523         }
1524
1525         if (mdev->state.conn == C_CONNECTED) {
1526                 if (dd == grew)
1527                         set_bit(RESIZE_PENDING, &mdev->flags);
1528
1529                 drbd_send_uuids(mdev);
1530                 drbd_send_sizes(mdev, 1, ddsf);
1531         }
1532
1533  fail:
1534         reply->ret_code = retcode;
1535         return 0;
1536 }
1537
1538 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1539                                struct drbd_nl_cfg_reply *reply)
1540 {
1541         int retcode = NO_ERROR;
1542         int err;
1543         int ovr; /* online verify running */
1544         int rsr; /* re-sync running */
1545         struct crypto_hash *verify_tfm = NULL;
1546         struct crypto_hash *csums_tfm = NULL;
1547         struct syncer_conf sc;
1548         cpumask_var_t new_cpu_mask;
1549
1550         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1551                 retcode = ERR_NOMEM;
1552                 goto fail;
1553         }
1554
1555         if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1556                 memset(&sc, 0, sizeof(struct syncer_conf));
1557                 sc.rate       = DRBD_RATE_DEF;
1558                 sc.after      = DRBD_AFTER_DEF;
1559                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1560                 sc.dp_volume  = DRBD_DP_VOLUME_DEF;
1561                 sc.dp_interval = DRBD_DP_INTERVAL_DEF;
1562                 sc.throttle_th = DRBD_RS_THROTTLE_TH_DEF;
1563                 sc.hold_off_th = DRBD_RS_HOLD_OFF_TH_DEF;
1564         } else
1565                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1566
1567         if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1568                 retcode = ERR_MANDATORY_TAG;
1569                 goto fail;
1570         }
1571
1572         /* re-sync running */
1573         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1574                 mdev->state.conn == C_SYNC_TARGET ||
1575                 mdev->state.conn == C_PAUSED_SYNC_S ||
1576                 mdev->state.conn == C_PAUSED_SYNC_T );
1577
1578         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1579                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1580                 goto fail;
1581         }
1582
1583         if (!rsr && sc.csums_alg[0]) {
1584                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1585                 if (IS_ERR(csums_tfm)) {
1586                         csums_tfm = NULL;
1587                         retcode = ERR_CSUMS_ALG;
1588                         goto fail;
1589                 }
1590
1591                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1592                         retcode = ERR_CSUMS_ALG_ND;
1593                         goto fail;
1594                 }
1595         }
1596
1597         /* online verify running */
1598         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1599
1600         if (ovr) {
1601                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1602                         retcode = ERR_VERIFY_RUNNING;
1603                         goto fail;
1604                 }
1605         }
1606
1607         if (!ovr && sc.verify_alg[0]) {
1608                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1609                 if (IS_ERR(verify_tfm)) {
1610                         verify_tfm = NULL;
1611                         retcode = ERR_VERIFY_ALG;
1612                         goto fail;
1613                 }
1614
1615                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1616                         retcode = ERR_VERIFY_ALG_ND;
1617                         goto fail;
1618                 }
1619         }
1620
1621         /* silently ignore cpu mask on UP kernel */
1622         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1623                 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1624                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1625                 if (err) {
1626                         dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1627                         retcode = ERR_CPU_MASK_PARSE;
1628                         goto fail;
1629                 }
1630         }
1631
1632         ERR_IF (sc.rate < 1) sc.rate = 1;
1633         ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1634 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1635         if (sc.al_extents > AL_MAX) {
1636                 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1637                 sc.al_extents = AL_MAX;
1638         }
1639 #undef AL_MAX
1640
1641         /* most sanity checks done, try to assign the new sync-after
1642          * dependency.  need to hold the global lock in there,
1643          * to avoid a race in the dependency loop check. */
1644         retcode = drbd_alter_sa(mdev, sc.after);
1645         if (retcode != NO_ERROR)
1646                 goto fail;
1647
1648         /* ok, assign the rest of it as well.
1649          * lock against receive_SyncParam() */
1650         spin_lock(&mdev->peer_seq_lock);
1651         mdev->sync_conf = sc;
1652
1653         if (!rsr) {
1654                 crypto_free_hash(mdev->csums_tfm);
1655                 mdev->csums_tfm = csums_tfm;
1656                 csums_tfm = NULL;
1657         }
1658
1659         if (!ovr) {
1660                 crypto_free_hash(mdev->verify_tfm);
1661                 mdev->verify_tfm = verify_tfm;
1662                 verify_tfm = NULL;
1663         }
1664         spin_unlock(&mdev->peer_seq_lock);
1665
1666         if (get_ldev(mdev)) {
1667                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1668                 drbd_al_shrink(mdev);
1669                 err = drbd_check_al_size(mdev);
1670                 lc_unlock(mdev->act_log);
1671                 wake_up(&mdev->al_wait);
1672
1673                 put_ldev(mdev);
1674                 drbd_md_sync(mdev);
1675
1676                 if (err) {
1677                         retcode = ERR_NOMEM;
1678                         goto fail;
1679                 }
1680         }
1681
1682         if (mdev->state.conn >= C_CONNECTED)
1683                 drbd_send_sync_param(mdev, &sc);
1684
1685         if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1686                 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1687                 drbd_calc_cpu_mask(mdev);
1688                 mdev->receiver.reset_cpu_mask = 1;
1689                 mdev->asender.reset_cpu_mask = 1;
1690                 mdev->worker.reset_cpu_mask = 1;
1691         }
1692
1693         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1694 fail:
1695         free_cpumask_var(new_cpu_mask);
1696         crypto_free_hash(csums_tfm);
1697         crypto_free_hash(verify_tfm);
1698         reply->ret_code = retcode;
1699         return 0;
1700 }
1701
1702 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1703                               struct drbd_nl_cfg_reply *reply)
1704 {
1705         int retcode;
1706
1707         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1708
1709         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1710                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1711
1712         while (retcode == SS_NEED_CONNECTION) {
1713                 spin_lock_irq(&mdev->req_lock);
1714                 if (mdev->state.conn < C_CONNECTED)
1715                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1716                 spin_unlock_irq(&mdev->req_lock);
1717
1718                 if (retcode != SS_NEED_CONNECTION)
1719                         break;
1720
1721                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1722         }
1723
1724         reply->ret_code = retcode;
1725         return 0;
1726 }
1727
1728 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1729                                    struct drbd_nl_cfg_reply *reply)
1730 {
1731
1732         reply->ret_code = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
1733
1734         return 0;
1735 }
1736
1737 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1738                               struct drbd_nl_cfg_reply *reply)
1739 {
1740         int retcode = NO_ERROR;
1741
1742         if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
1743                 retcode = ERR_PAUSE_IS_SET;
1744
1745         reply->ret_code = retcode;
1746         return 0;
1747 }
1748
1749 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1750                                struct drbd_nl_cfg_reply *reply)
1751 {
1752         int retcode = NO_ERROR;
1753
1754         if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO)
1755                 retcode = ERR_PAUSE_IS_CLEAR;
1756
1757         reply->ret_code = retcode;
1758         return 0;
1759 }
1760
1761 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1762                               struct drbd_nl_cfg_reply *reply)
1763 {
1764         reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
1765
1766         return 0;
1767 }
1768
1769 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1770                              struct drbd_nl_cfg_reply *reply)
1771 {
1772         reply->ret_code = drbd_request_state(mdev, NS(susp, 0));
1773         return 0;
1774 }
1775
1776 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1777                            struct drbd_nl_cfg_reply *reply)
1778 {
1779         reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
1780         return 0;
1781 }
1782
1783 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1784                            struct drbd_nl_cfg_reply *reply)
1785 {
1786         unsigned short *tl;
1787
1788         tl = reply->tag_list;
1789
1790         if (get_ldev(mdev)) {
1791                 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
1792                 put_ldev(mdev);
1793         }
1794
1795         if (get_net_conf(mdev)) {
1796                 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
1797                 put_net_conf(mdev);
1798         }
1799         tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
1800
1801         put_unaligned(TT_END, tl++); /* Close the tag list */
1802
1803         return (int)((char *)tl - (char *)reply->tag_list);
1804 }
1805
1806 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1807                              struct drbd_nl_cfg_reply *reply)
1808 {
1809         unsigned short *tl = reply->tag_list;
1810         union drbd_state s = mdev->state;
1811         unsigned long rs_left;
1812         unsigned int res;
1813
1814         tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
1815
1816         /* no local ref, no bitmap, no syncer progress. */
1817         if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
1818                 if (get_ldev(mdev)) {
1819                         drbd_get_syncer_progress(mdev, &rs_left, &res);
1820                         tl = tl_add_int(tl, T_sync_progress, &res);
1821                         put_ldev(mdev);
1822                 }
1823         }
1824         put_unaligned(TT_END, tl++); /* Close the tag list */
1825
1826         return (int)((char *)tl - (char *)reply->tag_list);
1827 }
1828
1829 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1830                              struct drbd_nl_cfg_reply *reply)
1831 {
1832         unsigned short *tl;
1833
1834         tl = reply->tag_list;
1835
1836         if (get_ldev(mdev)) {
1837                 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
1838                 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
1839                 put_ldev(mdev);
1840         }
1841         put_unaligned(TT_END, tl++); /* Close the tag list */
1842
1843         return (int)((char *)tl - (char *)reply->tag_list);
1844 }
1845
1846 /**
1847  * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
1848  * @mdev:       DRBD device.
1849  * @nlp:        Netlink/connector packet from drbdsetup
1850  * @reply:      Reply packet for drbdsetup
1851  */
1852 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1853                                     struct drbd_nl_cfg_reply *reply)
1854 {
1855         unsigned short *tl;
1856         char rv;
1857
1858         tl = reply->tag_list;
1859
1860         rv = mdev->state.pdsk == D_OUTDATED        ? UT_PEER_OUTDATED :
1861           test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
1862
1863         tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
1864         put_unaligned(TT_END, tl++); /* Close the tag list */
1865
1866         return (int)((char *)tl - (char *)reply->tag_list);
1867 }
1868
1869 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1870                                     struct drbd_nl_cfg_reply *reply)
1871 {
1872         /* default to resume from last known position, if possible */
1873         struct start_ov args =
1874                 { .start_sector = mdev->ov_start_sector };
1875
1876         if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
1877                 reply->ret_code = ERR_MANDATORY_TAG;
1878                 return 0;
1879         }
1880         /* w_make_ov_request expects position to be aligned */
1881         mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
1882         reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
1883         return 0;
1884 }
1885
1886
1887 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1888                               struct drbd_nl_cfg_reply *reply)
1889 {
1890         int retcode = NO_ERROR;
1891         int skip_initial_sync = 0;
1892         int err;
1893
1894         struct new_c_uuid args;
1895
1896         memset(&args, 0, sizeof(struct new_c_uuid));
1897         if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
1898                 reply->ret_code = ERR_MANDATORY_TAG;
1899                 return 0;
1900         }
1901
1902         mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
1903
1904         if (!get_ldev(mdev)) {
1905                 retcode = ERR_NO_DISK;
1906                 goto out;
1907         }
1908
1909         /* this is "skip initial sync", assume to be clean */
1910         if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
1911             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
1912                 dev_info(DEV, "Preparing to skip initial sync\n");
1913                 skip_initial_sync = 1;
1914         } else if (mdev->state.conn != C_STANDALONE) {
1915                 retcode = ERR_CONNECTED;
1916                 goto out_dec;
1917         }
1918
1919         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
1920         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
1921
1922         if (args.clear_bm) {
1923                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, "clear_n_write from new_c_uuid");
1924                 if (err) {
1925                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
1926                         retcode = ERR_IO_MD_DISK;
1927                 }
1928                 if (skip_initial_sync) {
1929                         drbd_send_uuids_skip_initial_sync(mdev);
1930                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
1931                         spin_lock_irq(&mdev->req_lock);
1932                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
1933                                         CS_VERBOSE, NULL);
1934                         spin_unlock_irq(&mdev->req_lock);
1935                 }
1936         }
1937
1938         drbd_md_sync(mdev);
1939 out_dec:
1940         put_ldev(mdev);
1941 out:
1942         mutex_unlock(&mdev->state_mutex);
1943
1944         reply->ret_code = retcode;
1945         return 0;
1946 }
1947
1948 static struct drbd_conf *ensure_mdev(struct drbd_nl_cfg_req *nlp)
1949 {
1950         struct drbd_conf *mdev;
1951
1952         if (nlp->drbd_minor >= minor_count)
1953                 return NULL;
1954
1955         mdev = minor_to_mdev(nlp->drbd_minor);
1956
1957         if (!mdev && (nlp->flags & DRBD_NL_CREATE_DEVICE)) {
1958                 struct gendisk *disk = NULL;
1959                 mdev = drbd_new_device(nlp->drbd_minor);
1960
1961                 spin_lock_irq(&drbd_pp_lock);
1962                 if (minor_table[nlp->drbd_minor] == NULL) {
1963                         minor_table[nlp->drbd_minor] = mdev;
1964                         disk = mdev->vdisk;
1965                         mdev = NULL;
1966                 } /* else: we lost the race */
1967                 spin_unlock_irq(&drbd_pp_lock);
1968
1969                 if (disk) /* we won the race above */
1970                         /* in case we ever add a drbd_delete_device(),
1971                          * don't forget the del_gendisk! */
1972                         add_disk(disk);
1973                 else /* we lost the race above */
1974                         drbd_free_mdev(mdev);
1975
1976                 mdev = minor_to_mdev(nlp->drbd_minor);
1977         }
1978
1979         return mdev;
1980 }
1981
1982 struct cn_handler_struct {
1983         int (*function)(struct drbd_conf *,
1984                          struct drbd_nl_cfg_req *,
1985                          struct drbd_nl_cfg_reply *);
1986         int reply_body_size;
1987 };
1988
1989 static struct cn_handler_struct cnd_table[] = {
1990         [ P_primary ]           = { &drbd_nl_primary,           0 },
1991         [ P_secondary ]         = { &drbd_nl_secondary,         0 },
1992         [ P_disk_conf ]         = { &drbd_nl_disk_conf,         0 },
1993         [ P_detach ]            = { &drbd_nl_detach,            0 },
1994         [ P_net_conf ]          = { &drbd_nl_net_conf,          0 },
1995         [ P_disconnect ]        = { &drbd_nl_disconnect,        0 },
1996         [ P_resize ]            = { &drbd_nl_resize,            0 },
1997         [ P_syncer_conf ]       = { &drbd_nl_syncer_conf,       0 },
1998         [ P_invalidate ]        = { &drbd_nl_invalidate,        0 },
1999         [ P_invalidate_peer ]   = { &drbd_nl_invalidate_peer,   0 },
2000         [ P_pause_sync ]        = { &drbd_nl_pause_sync,        0 },
2001         [ P_resume_sync ]       = { &drbd_nl_resume_sync,       0 },
2002         [ P_suspend_io ]        = { &drbd_nl_suspend_io,        0 },
2003         [ P_resume_io ]         = { &drbd_nl_resume_io,         0 },
2004         [ P_outdate ]           = { &drbd_nl_outdate,           0 },
2005         [ P_get_config ]        = { &drbd_nl_get_config,
2006                                     sizeof(struct syncer_conf_tag_len_struct) +
2007                                     sizeof(struct disk_conf_tag_len_struct) +
2008                                     sizeof(struct net_conf_tag_len_struct) },
2009         [ P_get_state ]         = { &drbd_nl_get_state,
2010                                     sizeof(struct get_state_tag_len_struct) +
2011                                     sizeof(struct sync_progress_tag_len_struct) },
2012         [ P_get_uuids ]         = { &drbd_nl_get_uuids,
2013                                     sizeof(struct get_uuids_tag_len_struct) },
2014         [ P_get_timeout_flag ]  = { &drbd_nl_get_timeout_flag,
2015                                     sizeof(struct get_timeout_flag_tag_len_struct)},
2016         [ P_start_ov ]          = { &drbd_nl_start_ov,          0 },
2017         [ P_new_c_uuid ]        = { &drbd_nl_new_c_uuid,        0 },
2018 };
2019
2020 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2021 {
2022         struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2023         struct cn_handler_struct *cm;
2024         struct cn_msg *cn_reply;
2025         struct drbd_nl_cfg_reply *reply;
2026         struct drbd_conf *mdev;
2027         int retcode, rr;
2028         int reply_size = sizeof(struct cn_msg)
2029                 + sizeof(struct drbd_nl_cfg_reply)
2030                 + sizeof(short int);
2031
2032         if (!try_module_get(THIS_MODULE)) {
2033                 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2034                 return;
2035         }
2036
2037         if (!cap_raised(nsp->eff_cap, CAP_SYS_ADMIN)) {
2038                 retcode = ERR_PERM;
2039                 goto fail;
2040         }
2041
2042         mdev = ensure_mdev(nlp);
2043         if (!mdev) {
2044                 retcode = ERR_MINOR_INVALID;
2045                 goto fail;
2046         }
2047
2048         if (nlp->packet_type >= P_nl_after_last_packet) {
2049                 retcode = ERR_PACKET_NR;
2050                 goto fail;
2051         }
2052
2053         cm = cnd_table + nlp->packet_type;
2054
2055         /* This may happen if packet number is 0: */
2056         if (cm->function == NULL) {
2057                 retcode = ERR_PACKET_NR;
2058                 goto fail;
2059         }
2060
2061         reply_size += cm->reply_body_size;
2062
2063         /* allocation not in the IO path, cqueue thread context */
2064         cn_reply = kmalloc(reply_size, GFP_KERNEL);
2065         if (!cn_reply) {
2066                 retcode = ERR_NOMEM;
2067                 goto fail;
2068         }
2069         reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2070
2071         reply->packet_type =
2072                 cm->reply_body_size ? nlp->packet_type : P_nl_after_last_packet;
2073         reply->minor = nlp->drbd_minor;
2074         reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2075         /* reply->tag_list; might be modified by cm->function. */
2076
2077         rr = cm->function(mdev, nlp, reply);
2078
2079         cn_reply->id = req->id;
2080         cn_reply->seq = req->seq;
2081         cn_reply->ack = req->ack  + 1;
2082         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2083         cn_reply->flags = 0;
2084
2085         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2086         if (rr && rr != -ESRCH)
2087                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2088
2089         kfree(cn_reply);
2090         module_put(THIS_MODULE);
2091         return;
2092  fail:
2093         drbd_nl_send_reply(req, retcode);
2094         module_put(THIS_MODULE);
2095 }
2096
2097 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2098
2099 static unsigned short *
2100 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2101         unsigned short len, int nul_terminated)
2102 {
2103         unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2104         len = (len < l) ? len :  l;
2105         put_unaligned(tag, tl++);
2106         put_unaligned(len, tl++);
2107         memcpy(tl, data, len);
2108         tl = (unsigned short*)((char*)tl + len);
2109         if (nul_terminated)
2110                 *((char*)tl - 1) = 0;
2111         return tl;
2112 }
2113
2114 static unsigned short *
2115 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2116 {
2117         return __tl_add_blob(tl, tag, data, len, 0);
2118 }
2119
2120 static unsigned short *
2121 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2122 {
2123         return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2124 }
2125
2126 static unsigned short *
2127 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2128 {
2129         put_unaligned(tag, tl++);
2130         switch(tag_type(tag)) {
2131         case TT_INTEGER:
2132                 put_unaligned(sizeof(int), tl++);
2133                 put_unaligned(*(int *)val, (int *)tl);
2134                 tl = (unsigned short*)((char*)tl+sizeof(int));
2135                 break;
2136         case TT_INT64:
2137                 put_unaligned(sizeof(u64), tl++);
2138                 put_unaligned(*(u64 *)val, (u64 *)tl);
2139                 tl = (unsigned short*)((char*)tl+sizeof(u64));
2140                 break;
2141         default:
2142                 /* someone did something stupid. */
2143                 ;
2144         }
2145         return tl;
2146 }
2147
2148 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2149 {
2150         char buffer[sizeof(struct cn_msg)+
2151                     sizeof(struct drbd_nl_cfg_reply)+
2152                     sizeof(struct get_state_tag_len_struct)+
2153                     sizeof(short int)];
2154         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2155         struct drbd_nl_cfg_reply *reply =
2156                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2157         unsigned short *tl = reply->tag_list;
2158
2159         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2160
2161         tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2162
2163         put_unaligned(TT_END, tl++); /* Close the tag list */
2164
2165         cn_reply->id.idx = CN_IDX_DRBD;
2166         cn_reply->id.val = CN_VAL_DRBD;
2167
2168         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2169         cn_reply->ack = 0; /* not used here. */
2170         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2171                 (int)((char *)tl - (char *)reply->tag_list);
2172         cn_reply->flags = 0;
2173
2174         reply->packet_type = P_get_state;
2175         reply->minor = mdev_to_minor(mdev);
2176         reply->ret_code = NO_ERROR;
2177
2178         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2179 }
2180
2181 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2182 {
2183         char buffer[sizeof(struct cn_msg)+
2184                     sizeof(struct drbd_nl_cfg_reply)+
2185                     sizeof(struct call_helper_tag_len_struct)+
2186                     sizeof(short int)];
2187         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2188         struct drbd_nl_cfg_reply *reply =
2189                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2190         unsigned short *tl = reply->tag_list;
2191
2192         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2193
2194         tl = tl_add_str(tl, T_helper, helper_name);
2195         put_unaligned(TT_END, tl++); /* Close the tag list */
2196
2197         cn_reply->id.idx = CN_IDX_DRBD;
2198         cn_reply->id.val = CN_VAL_DRBD;
2199
2200         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2201         cn_reply->ack = 0; /* not used here. */
2202         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2203                 (int)((char *)tl - (char *)reply->tag_list);
2204         cn_reply->flags = 0;
2205
2206         reply->packet_type = P_call_helper;
2207         reply->minor = mdev_to_minor(mdev);
2208         reply->ret_code = NO_ERROR;
2209
2210         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2211 }
2212
2213 void drbd_bcast_ee(struct drbd_conf *mdev,
2214                 const char *reason, const int dgs,
2215                 const char* seen_hash, const char* calc_hash,
2216                 const struct drbd_epoch_entry* e)
2217 {
2218         struct cn_msg *cn_reply;
2219         struct drbd_nl_cfg_reply *reply;
2220         unsigned short *tl;
2221         struct page *page;
2222         unsigned len;
2223
2224         if (!e)
2225                 return;
2226         if (!reason || !reason[0])
2227                 return;
2228
2229         /* apparently we have to memcpy twice, first to prepare the data for the
2230          * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2231          * netlink skb. */
2232         /* receiver thread context, which is not in the writeout path (of this node),
2233          * but may be in the writeout path of the _other_ node.
2234          * GFP_NOIO to avoid potential "distributed deadlock". */
2235         cn_reply = kmalloc(
2236                 sizeof(struct cn_msg)+
2237                 sizeof(struct drbd_nl_cfg_reply)+
2238                 sizeof(struct dump_ee_tag_len_struct)+
2239                 sizeof(short int),
2240                 GFP_NOIO);
2241
2242         if (!cn_reply) {
2243                 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2244                                 (unsigned long long)e->sector, e->size);
2245                 return;
2246         }
2247
2248         reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2249         tl = reply->tag_list;
2250
2251         tl = tl_add_str(tl, T_dump_ee_reason, reason);
2252         tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2253         tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2254         tl = tl_add_int(tl, T_ee_sector, &e->sector);
2255         tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2256
2257         put_unaligned(T_ee_data, tl++);
2258         put_unaligned(e->size, tl++);
2259
2260         len = e->size;
2261         page = e->pages;
2262         page_chain_for_each(page) {
2263                 void *d = kmap_atomic(page, KM_USER0);
2264                 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2265                 memcpy(tl, d, l);
2266                 kunmap_atomic(d, KM_USER0);
2267                 tl = (unsigned short*)((char*)tl + l);
2268                 len -= l;
2269         }
2270         put_unaligned(TT_END, tl++); /* Close the tag list */
2271
2272         cn_reply->id.idx = CN_IDX_DRBD;
2273         cn_reply->id.val = CN_VAL_DRBD;
2274
2275         cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2276         cn_reply->ack = 0; // not used here.
2277         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2278                 (int)((char*)tl - (char*)reply->tag_list);
2279         cn_reply->flags = 0;
2280
2281         reply->packet_type = P_dump_ee;
2282         reply->minor = mdev_to_minor(mdev);
2283         reply->ret_code = NO_ERROR;
2284
2285         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2286         kfree(cn_reply);
2287 }
2288
2289 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2290 {
2291         char buffer[sizeof(struct cn_msg)+
2292                     sizeof(struct drbd_nl_cfg_reply)+
2293                     sizeof(struct sync_progress_tag_len_struct)+
2294                     sizeof(short int)];
2295         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2296         struct drbd_nl_cfg_reply *reply =
2297                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2298         unsigned short *tl = reply->tag_list;
2299         unsigned long rs_left;
2300         unsigned int res;
2301
2302         /* no local ref, no bitmap, no syncer progress, no broadcast. */
2303         if (!get_ldev(mdev))
2304                 return;
2305         drbd_get_syncer_progress(mdev, &rs_left, &res);
2306         put_ldev(mdev);
2307
2308         tl = tl_add_int(tl, T_sync_progress, &res);
2309         put_unaligned(TT_END, tl++); /* Close the tag list */
2310
2311         cn_reply->id.idx = CN_IDX_DRBD;
2312         cn_reply->id.val = CN_VAL_DRBD;
2313
2314         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2315         cn_reply->ack = 0; /* not used here. */
2316         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2317                 (int)((char *)tl - (char *)reply->tag_list);
2318         cn_reply->flags = 0;
2319
2320         reply->packet_type = P_sync_progress;
2321         reply->minor = mdev_to_minor(mdev);
2322         reply->ret_code = NO_ERROR;
2323
2324         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2325 }
2326
2327 int __init drbd_nl_init(void)
2328 {
2329         static struct cb_id cn_id_drbd;
2330         int err, try=10;
2331
2332         cn_id_drbd.val = CN_VAL_DRBD;
2333         do {
2334                 cn_id_drbd.idx = cn_idx;
2335                 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2336                 if (!err)
2337                         break;
2338                 cn_idx = (cn_idx + CN_IDX_STEP);
2339         } while (try--);
2340
2341         if (err) {
2342                 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2343                 return err;
2344         }
2345
2346         return 0;
2347 }
2348
2349 void drbd_nl_cleanup(void)
2350 {
2351         static struct cb_id cn_id_drbd;
2352
2353         cn_id_drbd.idx = cn_idx;
2354         cn_id_drbd.val = CN_VAL_DRBD;
2355
2356         cn_del_callback(&cn_id_drbd);
2357 }
2358
2359 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2360 {
2361         char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2362         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2363         struct drbd_nl_cfg_reply *reply =
2364                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2365         int rr;
2366
2367         cn_reply->id = req->id;
2368
2369         cn_reply->seq = req->seq;
2370         cn_reply->ack = req->ack  + 1;
2371         cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2372         cn_reply->flags = 0;
2373
2374         reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2375         reply->ret_code = ret_code;
2376
2377         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2378         if (rr && rr != -ESRCH)
2379                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2380 }
2381