drivers/block/drbd/drbd_nl.c: use bitmap_parse instead of __bitmap_parse
[linux-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_req.h"
37 #include "drbd_wrappers.h"
38 #include <asm/unaligned.h>
39 #include <linux/drbd_tag_magic.h>
40 #include <linux/drbd_limits.h>
41 #include <linux/compiler.h>
42 #include <linux/kthread.h>
43
44 static unsigned short *tl_add_blob(unsigned short *, enum drbd_tags, const void *, int);
45 static unsigned short *tl_add_str(unsigned short *, enum drbd_tags, const char *);
46 static unsigned short *tl_add_int(unsigned short *, enum drbd_tags, const void *);
47
48 /* see get_sb_bdev and bd_claim */
49 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
50
51 /* Generate the tag_list to struct functions */
52 #define NL_PACKET(name, number, fields) \
53 static int name ## _from_tags(struct drbd_conf *mdev, \
54         unsigned short *tags, struct name *arg) __attribute__ ((unused)); \
55 static int name ## _from_tags(struct drbd_conf *mdev, \
56         unsigned short *tags, struct name *arg) \
57 { \
58         int tag; \
59         int dlen; \
60         \
61         while ((tag = get_unaligned(tags++)) != TT_END) {       \
62                 dlen = get_unaligned(tags++);                   \
63                 switch (tag_number(tag)) { \
64                 fields \
65                 default: \
66                         if (tag & T_MANDATORY) { \
67                                 dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \
68                                 return 0; \
69                         } \
70                 } \
71                 tags = (unsigned short *)((char *)tags + dlen); \
72         } \
73         return 1; \
74 }
75 #define NL_INTEGER(pn, pr, member) \
76         case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \
77                 arg->member = get_unaligned((int *)(tags));     \
78                 break;
79 #define NL_INT64(pn, pr, member) \
80         case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \
81                 arg->member = get_unaligned((u64 *)(tags));     \
82                 break;
83 #define NL_BIT(pn, pr, member) \
84         case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \
85                 arg->member = *(char *)(tags) ? 1 : 0; \
86                 break;
87 #define NL_STRING(pn, pr, member, len) \
88         case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \
89                 if (dlen > len) { \
90                         dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \
91                                 #member, dlen, (unsigned int)len); \
92                         return 0; \
93                 } \
94                  arg->member ## _len = dlen; \
95                  memcpy(arg->member, tags, min_t(size_t, dlen, len)); \
96                  break;
97 #include "linux/drbd_nl.h"
98
99 /* Generate the struct to tag_list functions */
100 #define NL_PACKET(name, number, fields) \
101 static unsigned short* \
102 name ## _to_tags(struct drbd_conf *mdev, \
103         struct name *arg, unsigned short *tags) __attribute__ ((unused)); \
104 static unsigned short* \
105 name ## _to_tags(struct drbd_conf *mdev, \
106         struct name *arg, unsigned short *tags) \
107 { \
108         fields \
109         return tags; \
110 }
111
112 #define NL_INTEGER(pn, pr, member) \
113         put_unaligned(pn | pr | TT_INTEGER, tags++);    \
114         put_unaligned(sizeof(int), tags++);             \
115         put_unaligned(arg->member, (int *)tags);        \
116         tags = (unsigned short *)((char *)tags+sizeof(int));
117 #define NL_INT64(pn, pr, member) \
118         put_unaligned(pn | pr | TT_INT64, tags++);      \
119         put_unaligned(sizeof(u64), tags++);             \
120         put_unaligned(arg->member, (u64 *)tags);        \
121         tags = (unsigned short *)((char *)tags+sizeof(u64));
122 #define NL_BIT(pn, pr, member) \
123         put_unaligned(pn | pr | TT_BIT, tags++);        \
124         put_unaligned(sizeof(char), tags++);            \
125         *(char *)tags = arg->member; \
126         tags = (unsigned short *)((char *)tags+sizeof(char));
127 #define NL_STRING(pn, pr, member, len) \
128         put_unaligned(pn | pr | TT_STRING, tags++);     \
129         put_unaligned(arg->member ## _len, tags++);     \
130         memcpy(tags, arg->member, arg->member ## _len); \
131         tags = (unsigned short *)((char *)tags + arg->member ## _len);
132 #include "linux/drbd_nl.h"
133
134 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name);
135 void drbd_nl_send_reply(struct cn_msg *, int);
136
137 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
138 {
139         char *envp[] = { "HOME=/",
140                         "TERM=linux",
141                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
142                         NULL, /* Will be set to address family */
143                         NULL, /* Will be set to address */
144                         NULL };
145
146         char mb[12], af[20], ad[60], *afs;
147         char *argv[] = {usermode_helper, cmd, mb, NULL };
148         int ret;
149
150         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
151
152         if (get_net_conf(mdev)) {
153                 switch (((struct sockaddr *)mdev->net_conf->peer_addr)->sa_family) {
154                 case AF_INET6:
155                         afs = "ipv6";
156                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6",
157                                  &((struct sockaddr_in6 *)mdev->net_conf->peer_addr)->sin6_addr);
158                         break;
159                 case AF_INET:
160                         afs = "ipv4";
161                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
162                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
163                         break;
164                 default:
165                         afs = "ssocks";
166                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
167                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
168                 }
169                 snprintf(af, 20, "DRBD_PEER_AF=%s", afs);
170                 envp[3]=af;
171                 envp[4]=ad;
172                 put_net_conf(mdev);
173         }
174
175         /* The helper may take some time.
176          * write out any unsynced meta data changes now */
177         drbd_md_sync(mdev);
178
179         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
180
181         drbd_bcast_ev_helper(mdev, cmd);
182         ret = call_usermodehelper(usermode_helper, argv, envp, 1);
183         if (ret)
184                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
185                                 usermode_helper, cmd, mb,
186                                 (ret >> 8) & 0xff, ret);
187         else
188                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
189                                 usermode_helper, cmd, mb,
190                                 (ret >> 8) & 0xff, ret);
191
192         if (ret < 0) /* Ignore any ERRNOs we got. */
193                 ret = 0;
194
195         return ret;
196 }
197
198 enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
199 {
200         char *ex_to_string;
201         int r;
202         enum drbd_disk_state nps;
203         enum drbd_fencing_p fp;
204
205         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
206
207         if (get_ldev_if_state(mdev, D_CONSISTENT)) {
208                 fp = mdev->ldev->dc.fencing;
209                 put_ldev(mdev);
210         } else {
211                 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n");
212                 nps = mdev->state.pdsk;
213                 goto out;
214         }
215
216         r = drbd_khelper(mdev, "fence-peer");
217
218         switch ((r>>8) & 0xff) {
219         case 3: /* peer is inconsistent */
220                 ex_to_string = "peer is inconsistent or worse";
221                 nps = D_INCONSISTENT;
222                 break;
223         case 4: /* peer got outdated, or was already outdated */
224                 ex_to_string = "peer was fenced";
225                 nps = D_OUTDATED;
226                 break;
227         case 5: /* peer was down */
228                 if (mdev->state.disk == D_UP_TO_DATE) {
229                         /* we will(have) create(d) a new UUID anyways... */
230                         ex_to_string = "peer is unreachable, assumed to be dead";
231                         nps = D_OUTDATED;
232                 } else {
233                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
234                         nps = mdev->state.pdsk;
235                 }
236                 break;
237         case 6: /* Peer is primary, voluntarily outdate myself.
238                  * This is useful when an unconnected R_SECONDARY is asked to
239                  * become R_PRIMARY, but finds the other peer being active. */
240                 ex_to_string = "peer is active";
241                 dev_warn(DEV, "Peer is primary, outdating myself.\n");
242                 nps = D_UNKNOWN;
243                 _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE);
244                 break;
245         case 7:
246                 if (fp != FP_STONITH)
247                         dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n");
248                 ex_to_string = "peer was stonithed";
249                 nps = D_OUTDATED;
250                 break;
251         default:
252                 /* The script is broken ... */
253                 nps = D_UNKNOWN;
254                 dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
255                 return nps;
256         }
257
258         dev_info(DEV, "fence-peer helper returned %d (%s)\n",
259                         (r>>8) & 0xff, ex_to_string);
260
261 out:
262         if (mdev->state.susp_fen && nps >= D_UNKNOWN) {
263                 /* The handler was not successful... unfreeze here, the
264                    state engine can not unfreeze... */
265                 _drbd_request_state(mdev, NS(susp_fen, 0), CS_VERBOSE);
266         }
267
268         return nps;
269 }
270
271 static int _try_outdate_peer_async(void *data)
272 {
273         struct drbd_conf *mdev = (struct drbd_conf *)data;
274         enum drbd_disk_state nps;
275         union drbd_state ns;
276
277         nps = drbd_try_outdate_peer(mdev);
278
279         /* Not using
280            drbd_request_state(mdev, NS(pdsk, nps));
281            here, because we might were able to re-establish the connection
282            in the meantime. This can only partially be solved in the state's
283            engine is_valid_state() and is_valid_state_transition()
284            functions.
285
286            nps can be D_INCONSISTENT, D_OUTDATED or D_UNKNOWN.
287            pdsk == D_INCONSISTENT while conn >= C_CONNECTED is valid,
288            therefore we have to have the pre state change check here.
289         */
290         spin_lock_irq(&mdev->req_lock);
291         ns = mdev->state;
292         if (ns.conn < C_WF_REPORT_PARAMS) {
293                 ns.pdsk = nps;
294                 _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
295         }
296         spin_unlock_irq(&mdev->req_lock);
297
298         return 0;
299 }
300
301 void drbd_try_outdate_peer_async(struct drbd_conf *mdev)
302 {
303         struct task_struct *opa;
304
305         opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev));
306         if (IS_ERR(opa))
307                 dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n");
308 }
309
310 enum drbd_state_rv
311 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
312 {
313         const int max_tries = 4;
314         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
315         int try = 0;
316         int forced = 0;
317         union drbd_state mask, val;
318         enum drbd_disk_state nps;
319
320         if (new_role == R_PRIMARY)
321                 request_ping(mdev); /* Detect a dead peer ASAP */
322
323         mutex_lock(&mdev->state_mutex);
324
325         mask.i = 0; mask.role = R_MASK;
326         val.i  = 0; val.role  = new_role;
327
328         while (try++ < max_tries) {
329                 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
330
331                 /* in case we first succeeded to outdate,
332                  * but now suddenly could establish a connection */
333                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
334                         val.pdsk = 0;
335                         mask.pdsk = 0;
336                         continue;
337                 }
338
339                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
340                     (mdev->state.disk < D_UP_TO_DATE &&
341                      mdev->state.disk >= D_INCONSISTENT)) {
342                         mask.disk = D_MASK;
343                         val.disk  = D_UP_TO_DATE;
344                         forced = 1;
345                         continue;
346                 }
347
348                 if (rv == SS_NO_UP_TO_DATE_DISK &&
349                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
350                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
351                         nps = drbd_try_outdate_peer(mdev);
352
353                         if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
354                                 val.disk = D_UP_TO_DATE;
355                                 mask.disk = D_MASK;
356                         }
357
358                         val.pdsk = nps;
359                         mask.pdsk = D_MASK;
360
361                         continue;
362                 }
363
364                 if (rv == SS_NOTHING_TO_DO)
365                         goto fail;
366                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
367                         nps = drbd_try_outdate_peer(mdev);
368
369                         if (force && nps > D_OUTDATED) {
370                                 dev_warn(DEV, "Forced into split brain situation!\n");
371                                 nps = D_OUTDATED;
372                         }
373
374                         mask.pdsk = D_MASK;
375                         val.pdsk  = nps;
376
377                         continue;
378                 }
379                 if (rv == SS_TWO_PRIMARIES) {
380                         /* Maybe the peer is detected as dead very soon...
381                            retry at most once more in this case. */
382                         schedule_timeout_interruptible((mdev->net_conf->ping_timeo+1)*HZ/10);
383                         if (try < max_tries)
384                                 try = max_tries - 1;
385                         continue;
386                 }
387                 if (rv < SS_SUCCESS) {
388                         rv = _drbd_request_state(mdev, mask, val,
389                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
390                         if (rv < SS_SUCCESS)
391                                 goto fail;
392                 }
393                 break;
394         }
395
396         if (rv < SS_SUCCESS)
397                 goto fail;
398
399         if (forced)
400                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
401
402         /* Wait until nothing is on the fly :) */
403         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
404
405         if (new_role == R_SECONDARY) {
406                 set_disk_ro(mdev->vdisk, true);
407                 if (get_ldev(mdev)) {
408                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
409                         put_ldev(mdev);
410                 }
411         } else {
412                 if (get_net_conf(mdev)) {
413                         mdev->net_conf->want_lose = 0;
414                         put_net_conf(mdev);
415                 }
416                 set_disk_ro(mdev->vdisk, false);
417                 if (get_ldev(mdev)) {
418                         if (((mdev->state.conn < C_CONNECTED ||
419                                mdev->state.pdsk <= D_FAILED)
420                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
421                                 drbd_uuid_new_current(mdev);
422
423                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
424                         put_ldev(mdev);
425                 }
426         }
427
428         /* writeout of activity log covered areas of the bitmap
429          * to stable storage done in after state change already */
430
431         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
432                 /* if this was forced, we should consider sync */
433                 if (forced)
434                         drbd_send_uuids(mdev);
435                 drbd_send_state(mdev);
436         }
437
438         drbd_md_sync(mdev);
439
440         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
441  fail:
442         mutex_unlock(&mdev->state_mutex);
443         return rv;
444 }
445
446 static struct drbd_conf *ensure_mdev(int minor, int create)
447 {
448         struct drbd_conf *mdev;
449
450         if (minor >= minor_count)
451                 return NULL;
452
453         mdev = minor_to_mdev(minor);
454
455         if (!mdev && create) {
456                 struct gendisk *disk = NULL;
457                 mdev = drbd_new_device(minor);
458
459                 spin_lock_irq(&drbd_pp_lock);
460                 if (minor_table[minor] == NULL) {
461                         minor_table[minor] = mdev;
462                         disk = mdev->vdisk;
463                         mdev = NULL;
464                 } /* else: we lost the race */
465                 spin_unlock_irq(&drbd_pp_lock);
466
467                 if (disk) /* we won the race above */
468                         /* in case we ever add a drbd_delete_device(),
469                          * don't forget the del_gendisk! */
470                         add_disk(disk);
471                 else /* we lost the race above */
472                         drbd_free_mdev(mdev);
473
474                 mdev = minor_to_mdev(minor);
475         }
476
477         return mdev;
478 }
479
480 static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
481                            struct drbd_nl_cfg_reply *reply)
482 {
483         struct primary primary_args;
484
485         memset(&primary_args, 0, sizeof(struct primary));
486         if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) {
487                 reply->ret_code = ERR_MANDATORY_TAG;
488                 return 0;
489         }
490
491         reply->ret_code =
492                 drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force);
493
494         return 0;
495 }
496
497 static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
498                              struct drbd_nl_cfg_reply *reply)
499 {
500         reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0);
501
502         return 0;
503 }
504
505 /* initializes the md.*_offset members, so we are able to find
506  * the on disk meta data */
507 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
508                                        struct drbd_backing_dev *bdev)
509 {
510         sector_t md_size_sect = 0;
511         switch (bdev->dc.meta_dev_idx) {
512         default:
513                 /* v07 style fixed size indexed meta data */
514                 bdev->md.md_size_sect = MD_RESERVED_SECT;
515                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
516                 bdev->md.al_offset = MD_AL_OFFSET;
517                 bdev->md.bm_offset = MD_BM_OFFSET;
518                 break;
519         case DRBD_MD_INDEX_FLEX_EXT:
520                 /* just occupy the full device; unit: sectors */
521                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
522                 bdev->md.md_offset = 0;
523                 bdev->md.al_offset = MD_AL_OFFSET;
524                 bdev->md.bm_offset = MD_BM_OFFSET;
525                 break;
526         case DRBD_MD_INDEX_INTERNAL:
527         case DRBD_MD_INDEX_FLEX_INT:
528                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
529                 /* al size is still fixed */
530                 bdev->md.al_offset = -MD_AL_MAX_SIZE;
531                 /* we need (slightly less than) ~ this much bitmap sectors: */
532                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
533                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
534                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
535                 md_size_sect = ALIGN(md_size_sect, 8);
536
537                 /* plus the "drbd meta data super block",
538                  * and the activity log; */
539                 md_size_sect += MD_BM_OFFSET;
540
541                 bdev->md.md_size_sect = md_size_sect;
542                 /* bitmap offset is adjusted by 'super' block size */
543                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
544                 break;
545         }
546 }
547
548 /* input size is expected to be in KB */
549 char *ppsize(char *buf, unsigned long long size)
550 {
551         /* Needs 9 bytes at max including trailing NUL:
552          * -1ULL ==> "16384 EB" */
553         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
554         int base = 0;
555         while (size >= 10000 && base < sizeof(units)-1) {
556                 /* shift + round */
557                 size = (size >> 10) + !!(size & (1<<9));
558                 base++;
559         }
560         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
561
562         return buf;
563 }
564
565 /* there is still a theoretical deadlock when called from receiver
566  * on an D_INCONSISTENT R_PRIMARY:
567  *  remote READ does inc_ap_bio, receiver would need to receive answer
568  *  packet from remote to dec_ap_bio again.
569  *  receiver receive_sizes(), comes here,
570  *  waits for ap_bio_cnt == 0. -> deadlock.
571  * but this cannot happen, actually, because:
572  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
573  *  (not connected, or bad/no disk on peer):
574  *  see drbd_fail_request_early, ap_bio_cnt is zero.
575  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
576  *  peer may not initiate a resize.
577  */
578 void drbd_suspend_io(struct drbd_conf *mdev)
579 {
580         set_bit(SUSPEND_IO, &mdev->flags);
581         if (is_susp(mdev->state))
582                 return;
583         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
584 }
585
586 void drbd_resume_io(struct drbd_conf *mdev)
587 {
588         clear_bit(SUSPEND_IO, &mdev->flags);
589         wake_up(&mdev->misc_wait);
590 }
591
592 /**
593  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
594  * @mdev:       DRBD device.
595  *
596  * Returns 0 on success, negative return values indicate errors.
597  * You should call drbd_md_sync() after calling this function.
598  */
599 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
600 {
601         sector_t prev_first_sect, prev_size; /* previous meta location */
602         sector_t la_size;
603         sector_t size;
604         char ppb[10];
605
606         int md_moved, la_size_changed;
607         enum determine_dev_size rv = unchanged;
608
609         /* race:
610          * application request passes inc_ap_bio,
611          * but then cannot get an AL-reference.
612          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
613          *
614          * to avoid that:
615          * Suspend IO right here.
616          * still lock the act_log to not trigger ASSERTs there.
617          */
618         drbd_suspend_io(mdev);
619
620         /* no wait necessary anymore, actually we could assert that */
621         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
622
623         prev_first_sect = drbd_md_first_sector(mdev->ldev);
624         prev_size = mdev->ldev->md.md_size_sect;
625         la_size = mdev->ldev->md.la_size_sect;
626
627         /* TODO: should only be some assert here, not (re)init... */
628         drbd_md_set_sector_offsets(mdev, mdev->ldev);
629
630         size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
631
632         if (drbd_get_capacity(mdev->this_bdev) != size ||
633             drbd_bm_capacity(mdev) != size) {
634                 int err;
635                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
636                 if (unlikely(err)) {
637                         /* currently there is only one error: ENOMEM! */
638                         size = drbd_bm_capacity(mdev)>>1;
639                         if (size == 0) {
640                                 dev_err(DEV, "OUT OF MEMORY! "
641                                     "Could not allocate bitmap!\n");
642                         } else {
643                                 dev_err(DEV, "BM resizing failed. "
644                                     "Leaving size unchanged at size = %lu KB\n",
645                                     (unsigned long)size);
646                         }
647                         rv = dev_size_error;
648                 }
649                 /* racy, see comments above. */
650                 drbd_set_my_capacity(mdev, size);
651                 mdev->ldev->md.la_size_sect = size;
652                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
653                      (unsigned long long)size>>1);
654         }
655         if (rv == dev_size_error)
656                 goto out;
657
658         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
659
660         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
661                 || prev_size       != mdev->ldev->md.md_size_sect;
662
663         if (la_size_changed || md_moved) {
664                 int err;
665
666                 drbd_al_shrink(mdev); /* All extents inactive. */
667                 dev_info(DEV, "Writing the whole bitmap, %s\n",
668                          la_size_changed && md_moved ? "size changed and md moved" :
669                          la_size_changed ? "size changed" : "md moved");
670                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
671                 err = drbd_bitmap_io(mdev, &drbd_bm_write,
672                                 "size changed", BM_LOCKED_MASK);
673                 if (err) {
674                         rv = dev_size_error;
675                         goto out;
676                 }
677                 drbd_md_mark_dirty(mdev);
678         }
679
680         if (size > la_size)
681                 rv = grew;
682         if (size < la_size)
683                 rv = shrunk;
684 out:
685         lc_unlock(mdev->act_log);
686         wake_up(&mdev->al_wait);
687         drbd_resume_io(mdev);
688
689         return rv;
690 }
691
692 sector_t
693 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
694 {
695         sector_t p_size = mdev->p_size;   /* partner's disk size. */
696         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
697         sector_t m_size; /* my size */
698         sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
699         sector_t size = 0;
700
701         m_size = drbd_get_max_capacity(bdev);
702
703         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
704                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
705                 p_size = m_size;
706         }
707
708         if (p_size && m_size) {
709                 size = min_t(sector_t, p_size, m_size);
710         } else {
711                 if (la_size) {
712                         size = la_size;
713                         if (m_size && m_size < size)
714                                 size = m_size;
715                         if (p_size && p_size < size)
716                                 size = p_size;
717                 } else {
718                         if (m_size)
719                                 size = m_size;
720                         if (p_size)
721                                 size = p_size;
722                 }
723         }
724
725         if (size == 0)
726                 dev_err(DEV, "Both nodes diskless!\n");
727
728         if (u_size) {
729                 if (u_size > size)
730                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
731                             (unsigned long)u_size>>1, (unsigned long)size>>1);
732                 else
733                         size = u_size;
734         }
735
736         return size;
737 }
738
739 /**
740  * drbd_check_al_size() - Ensures that the AL is of the right size
741  * @mdev:       DRBD device.
742  *
743  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
744  * failed, and 0 on success. You should call drbd_md_sync() after you called
745  * this function.
746  */
747 static int drbd_check_al_size(struct drbd_conf *mdev)
748 {
749         struct lru_cache *n, *t;
750         struct lc_element *e;
751         unsigned int in_use;
752         int i;
753
754         ERR_IF(mdev->sync_conf.al_extents < 7)
755                 mdev->sync_conf.al_extents = 127;
756
757         if (mdev->act_log &&
758             mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
759                 return 0;
760
761         in_use = 0;
762         t = mdev->act_log;
763         n = lc_create("act_log", drbd_al_ext_cache,
764                 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
765
766         if (n == NULL) {
767                 dev_err(DEV, "Cannot allocate act_log lru!\n");
768                 return -ENOMEM;
769         }
770         spin_lock_irq(&mdev->al_lock);
771         if (t) {
772                 for (i = 0; i < t->nr_elements; i++) {
773                         e = lc_element_by_index(t, i);
774                         if (e->refcnt)
775                                 dev_err(DEV, "refcnt(%d)==%d\n",
776                                     e->lc_number, e->refcnt);
777                         in_use += e->refcnt;
778                 }
779         }
780         if (!in_use)
781                 mdev->act_log = n;
782         spin_unlock_irq(&mdev->al_lock);
783         if (in_use) {
784                 dev_err(DEV, "Activity log still in use!\n");
785                 lc_destroy(n);
786                 return -EBUSY;
787         } else {
788                 if (t)
789                         lc_destroy(t);
790         }
791         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
792         return 0;
793 }
794
795 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
796 {
797         struct request_queue * const q = mdev->rq_queue;
798         int max_hw_sectors = max_bio_size >> 9;
799         int max_segments = 0;
800
801         if (get_ldev_if_state(mdev, D_ATTACHING)) {
802                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
803
804                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
805                 max_segments = mdev->ldev->dc.max_bio_bvecs;
806                 put_ldev(mdev);
807         }
808
809         blk_queue_logical_block_size(q, 512);
810         blk_queue_max_hw_sectors(q, max_hw_sectors);
811         /* This is the workaround for "bio would need to, but cannot, be split" */
812         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
813         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
814
815         if (get_ldev_if_state(mdev, D_ATTACHING)) {
816                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
817
818                 blk_queue_stack_limits(q, b);
819
820                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
821                         dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
822                                  q->backing_dev_info.ra_pages,
823                                  b->backing_dev_info.ra_pages);
824                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
825                 }
826                 put_ldev(mdev);
827         }
828 }
829
830 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
831 {
832         int now, new, local, peer;
833
834         now = queue_max_hw_sectors(mdev->rq_queue) << 9;
835         local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
836         peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
837
838         if (get_ldev_if_state(mdev, D_ATTACHING)) {
839                 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
840                 mdev->local_max_bio_size = local;
841                 put_ldev(mdev);
842         }
843
844         /* We may ignore peer limits if the peer is modern enough.
845            Because new from 8.3.8 onwards the peer can use multiple
846            BIOs for a single peer_request */
847         if (mdev->state.conn >= C_CONNECTED) {
848                 if (mdev->agreed_pro_version < 94)
849                         peer = mdev->peer_max_bio_size;
850                 else if (mdev->agreed_pro_version == 94)
851                         peer = DRBD_MAX_SIZE_H80_PACKET;
852                 else /* drbd 8.3.8 onwards */
853                         peer = DRBD_MAX_BIO_SIZE;
854         }
855
856         new = min_t(int, local, peer);
857
858         if (mdev->state.role == R_PRIMARY && new < now)
859                 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
860
861         if (new != now)
862                 dev_info(DEV, "max BIO size = %u\n", new);
863
864         drbd_setup_queue_param(mdev, new);
865 }
866
867 /* serialize deconfig (worker exiting, doing cleanup)
868  * and reconfig (drbdsetup disk, drbdsetup net)
869  *
870  * Wait for a potentially exiting worker, then restart it,
871  * or start a new one.  Flush any pending work, there may still be an
872  * after_state_change queued.
873  */
874 static void drbd_reconfig_start(struct drbd_conf *mdev)
875 {
876         wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
877         wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
878         drbd_thread_start(&mdev->worker);
879         drbd_flush_workqueue(mdev);
880 }
881
882 /* if still unconfigured, stops worker again.
883  * if configured now, clears CONFIG_PENDING.
884  * wakes potential waiters */
885 static void drbd_reconfig_done(struct drbd_conf *mdev)
886 {
887         spin_lock_irq(&mdev->req_lock);
888         if (mdev->state.disk == D_DISKLESS &&
889             mdev->state.conn == C_STANDALONE &&
890             mdev->state.role == R_SECONDARY) {
891                 set_bit(DEVICE_DYING, &mdev->flags);
892                 drbd_thread_stop_nowait(&mdev->worker);
893         } else
894                 clear_bit(CONFIG_PENDING, &mdev->flags);
895         spin_unlock_irq(&mdev->req_lock);
896         wake_up(&mdev->state_wait);
897 }
898
899 /* Make sure IO is suspended before calling this function(). */
900 static void drbd_suspend_al(struct drbd_conf *mdev)
901 {
902         int s = 0;
903
904         if (lc_try_lock(mdev->act_log)) {
905                 drbd_al_shrink(mdev);
906                 lc_unlock(mdev->act_log);
907         } else {
908                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
909                 return;
910         }
911
912         spin_lock_irq(&mdev->req_lock);
913         if (mdev->state.conn < C_CONNECTED)
914                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
915
916         spin_unlock_irq(&mdev->req_lock);
917
918         if (s)
919                 dev_info(DEV, "Suspended AL updates\n");
920 }
921
922 /* does always return 0;
923  * interesting return code is in reply->ret_code */
924 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
925                              struct drbd_nl_cfg_reply *reply)
926 {
927         enum drbd_ret_code retcode;
928         enum determine_dev_size dd;
929         sector_t max_possible_sectors;
930         sector_t min_md_device_sectors;
931         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
932         struct block_device *bdev;
933         struct lru_cache *resync_lru = NULL;
934         union drbd_state ns, os;
935         enum drbd_state_rv rv;
936         int cp_discovered = 0;
937         int logical_block_size;
938
939         drbd_reconfig_start(mdev);
940
941         /* if you want to reconfigure, please tear down first */
942         if (mdev->state.disk > D_DISKLESS) {
943                 retcode = ERR_DISK_CONFIGURED;
944                 goto fail;
945         }
946         /* It may just now have detached because of IO error.  Make sure
947          * drbd_ldev_destroy is done already, we may end up here very fast,
948          * e.g. if someone calls attach from the on-io-error handler,
949          * to realize a "hot spare" feature (not that I'd recommend that) */
950         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
951
952         /* allocation not in the IO path, cqueue thread context */
953         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
954         if (!nbc) {
955                 retcode = ERR_NOMEM;
956                 goto fail;
957         }
958
959         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
960         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
961         nbc->dc.fencing       = DRBD_FENCING_DEF;
962         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
963
964         if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
965                 retcode = ERR_MANDATORY_TAG;
966                 goto fail;
967         }
968
969         if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
970                 retcode = ERR_MD_IDX_INVALID;
971                 goto fail;
972         }
973
974         if (get_net_conf(mdev)) {
975                 int prot = mdev->net_conf->wire_protocol;
976                 put_net_conf(mdev);
977                 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
978                         retcode = ERR_STONITH_AND_PROT_A;
979                         goto fail;
980                 }
981         }
982
983         bdev = blkdev_get_by_path(nbc->dc.backing_dev,
984                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
985         if (IS_ERR(bdev)) {
986                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
987                         PTR_ERR(bdev));
988                 retcode = ERR_OPEN_DISK;
989                 goto fail;
990         }
991         nbc->backing_bdev = bdev;
992
993         /*
994          * meta_dev_idx >= 0: external fixed size, possibly multiple
995          * drbd sharing one meta device.  TODO in that case, paranoia
996          * check that [md_bdev, meta_dev_idx] is not yet used by some
997          * other drbd minor!  (if you use drbd.conf + drbdadm, that
998          * should check it for you already; but if you don't, or
999          * someone fooled it, we need to double check here)
1000          */
1001         bdev = blkdev_get_by_path(nbc->dc.meta_dev,
1002                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1003                                   (nbc->dc.meta_dev_idx < 0) ?
1004                                   (void *)mdev : (void *)drbd_m_holder);
1005         if (IS_ERR(bdev)) {
1006                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
1007                         PTR_ERR(bdev));
1008                 retcode = ERR_OPEN_MD_DISK;
1009                 goto fail;
1010         }
1011         nbc->md_bdev = bdev;
1012
1013         if ((nbc->backing_bdev == nbc->md_bdev) !=
1014             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1015              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1016                 retcode = ERR_MD_IDX_INVALID;
1017                 goto fail;
1018         }
1019
1020         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1021                         61, sizeof(struct bm_extent),
1022                         offsetof(struct bm_extent, lce));
1023         if (!resync_lru) {
1024                 retcode = ERR_NOMEM;
1025                 goto fail;
1026         }
1027
1028         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1029         drbd_md_set_sector_offsets(mdev, nbc);
1030
1031         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
1032                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1033                         (unsigned long long) drbd_get_max_capacity(nbc),
1034                         (unsigned long long) nbc->dc.disk_size);
1035                 retcode = ERR_DISK_TO_SMALL;
1036                 goto fail;
1037         }
1038
1039         if (nbc->dc.meta_dev_idx < 0) {
1040                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1041                 /* at least one MB, otherwise it does not make sense */
1042                 min_md_device_sectors = (2<<10);
1043         } else {
1044                 max_possible_sectors = DRBD_MAX_SECTORS;
1045                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
1046         }
1047
1048         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1049                 retcode = ERR_MD_DISK_TO_SMALL;
1050                 dev_warn(DEV, "refusing attach: md-device too small, "
1051                      "at least %llu sectors needed for this meta-disk type\n",
1052                      (unsigned long long) min_md_device_sectors);
1053                 goto fail;
1054         }
1055
1056         /* Make sure the new disk is big enough
1057          * (we may currently be R_PRIMARY with no local disk...) */
1058         if (drbd_get_max_capacity(nbc) <
1059             drbd_get_capacity(mdev->this_bdev)) {
1060                 retcode = ERR_DISK_TO_SMALL;
1061                 goto fail;
1062         }
1063
1064         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1065
1066         if (nbc->known_size > max_possible_sectors) {
1067                 dev_warn(DEV, "==> truncating very big lower level device "
1068                         "to currently maximum possible %llu sectors <==\n",
1069                         (unsigned long long) max_possible_sectors);
1070                 if (nbc->dc.meta_dev_idx >= 0)
1071                         dev_warn(DEV, "==>> using internal or flexible "
1072                                       "meta data may help <<==\n");
1073         }
1074
1075         drbd_suspend_io(mdev);
1076         /* also wait for the last barrier ack. */
1077         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state));
1078         /* and for any other previously queued work */
1079         drbd_flush_workqueue(mdev);
1080
1081         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1082         retcode = rv;  /* FIXME: Type mismatch. */
1083         drbd_resume_io(mdev);
1084         if (rv < SS_SUCCESS)
1085                 goto fail;
1086
1087         if (!get_ldev_if_state(mdev, D_ATTACHING))
1088                 goto force_diskless;
1089
1090         drbd_md_set_sector_offsets(mdev, nbc);
1091
1092         /* allocate a second IO page if logical_block_size != 512 */
1093         logical_block_size = bdev_logical_block_size(nbc->md_bdev);
1094         if (logical_block_size == 0)
1095                 logical_block_size = MD_SECTOR_SIZE;
1096
1097         if (logical_block_size != MD_SECTOR_SIZE) {
1098                 if (!mdev->md_io_tmpp) {
1099                         struct page *page = alloc_page(GFP_NOIO);
1100                         if (!page)
1101                                 goto force_diskless_dec;
1102
1103                         dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
1104                              logical_block_size, MD_SECTOR_SIZE);
1105                         dev_warn(DEV, "Workaround engaged (has performance impact).\n");
1106
1107                         mdev->md_io_tmpp = page;
1108                 }
1109         }
1110
1111         if (!mdev->bitmap) {
1112                 if (drbd_bm_init(mdev)) {
1113                         retcode = ERR_NOMEM;
1114                         goto force_diskless_dec;
1115                 }
1116         }
1117
1118         retcode = drbd_md_read(mdev, nbc);
1119         if (retcode != NO_ERROR)
1120                 goto force_diskless_dec;
1121
1122         if (mdev->state.conn < C_CONNECTED &&
1123             mdev->state.role == R_PRIMARY &&
1124             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1125                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1126                     (unsigned long long)mdev->ed_uuid);
1127                 retcode = ERR_DATA_NOT_CURRENT;
1128                 goto force_diskless_dec;
1129         }
1130
1131         /* Since we are diskless, fix the activity log first... */
1132         if (drbd_check_al_size(mdev)) {
1133                 retcode = ERR_NOMEM;
1134                 goto force_diskless_dec;
1135         }
1136
1137         /* Prevent shrinking of consistent devices ! */
1138         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1139             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1140                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1141                 retcode = ERR_DISK_TO_SMALL;
1142                 goto force_diskless_dec;
1143         }
1144
1145         if (!drbd_al_read_log(mdev, nbc)) {
1146                 retcode = ERR_IO_MD_DISK;
1147                 goto force_diskless_dec;
1148         }
1149
1150         /* Reset the "barriers don't work" bits here, then force meta data to
1151          * be written, to ensure we determine if barriers are supported. */
1152         if (nbc->dc.no_md_flush)
1153                 set_bit(MD_NO_FUA, &mdev->flags);
1154         else
1155                 clear_bit(MD_NO_FUA, &mdev->flags);
1156
1157         /* Point of no return reached.
1158          * Devices and memory are no longer released by error cleanup below.
1159          * now mdev takes over responsibility, and the state engine should
1160          * clean it up somewhere.  */
1161         D_ASSERT(mdev->ldev == NULL);
1162         mdev->ldev = nbc;
1163         mdev->resync = resync_lru;
1164         nbc = NULL;
1165         resync_lru = NULL;
1166
1167         mdev->write_ordering = WO_bdev_flush;
1168         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1169
1170         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1171                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1172         else
1173                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1174
1175         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1176             !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) {
1177                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1178                 cp_discovered = 1;
1179         }
1180
1181         mdev->send_cnt = 0;
1182         mdev->recv_cnt = 0;
1183         mdev->read_cnt = 0;
1184         mdev->writ_cnt = 0;
1185
1186         drbd_reconsider_max_bio_size(mdev);
1187
1188         /* If I am currently not R_PRIMARY,
1189          * but meta data primary indicator is set,
1190          * I just now recover from a hard crash,
1191          * and have been R_PRIMARY before that crash.
1192          *
1193          * Now, if I had no connection before that crash
1194          * (have been degraded R_PRIMARY), chances are that
1195          * I won't find my peer now either.
1196          *
1197          * In that case, and _only_ in that case,
1198          * we use the degr-wfc-timeout instead of the default,
1199          * so we can automatically recover from a crash of a
1200          * degraded but active "cluster" after a certain timeout.
1201          */
1202         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1203         if (mdev->state.role != R_PRIMARY &&
1204              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1205             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1206                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1207
1208         dd = drbd_determine_dev_size(mdev, 0);
1209         if (dd == dev_size_error) {
1210                 retcode = ERR_NOMEM_BITMAP;
1211                 goto force_diskless_dec;
1212         } else if (dd == grew)
1213                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1214
1215         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1216                 dev_info(DEV, "Assuming that all blocks are out of sync "
1217                      "(aka FullSync)\n");
1218                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1219                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1220                         retcode = ERR_IO_MD_DISK;
1221                         goto force_diskless_dec;
1222                 }
1223         } else {
1224                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1225                         "read from attaching", BM_LOCKED_MASK) < 0) {
1226                         retcode = ERR_IO_MD_DISK;
1227                         goto force_diskless_dec;
1228                 }
1229         }
1230
1231         if (cp_discovered) {
1232                 drbd_al_apply_to_bm(mdev);
1233                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1234                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1235                         retcode = ERR_IO_MD_DISK;
1236                         goto force_diskless_dec;
1237                 }
1238         }
1239
1240         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1241                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1242
1243         spin_lock_irq(&mdev->req_lock);
1244         os = mdev->state;
1245         ns.i = os.i;
1246         /* If MDF_CONSISTENT is not set go into inconsistent state,
1247            otherwise investigate MDF_WasUpToDate...
1248            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1249            otherwise into D_CONSISTENT state.
1250         */
1251         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1252                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1253                         ns.disk = D_CONSISTENT;
1254                 else
1255                         ns.disk = D_OUTDATED;
1256         } else {
1257                 ns.disk = D_INCONSISTENT;
1258         }
1259
1260         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1261                 ns.pdsk = D_OUTDATED;
1262
1263         if ( ns.disk == D_CONSISTENT &&
1264             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1265                 ns.disk = D_UP_TO_DATE;
1266
1267         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1268            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1269            this point, because drbd_request_state() modifies these
1270            flags. */
1271
1272         /* In case we are C_CONNECTED postpone any decision on the new disk
1273            state after the negotiation phase. */
1274         if (mdev->state.conn == C_CONNECTED) {
1275                 mdev->new_state_tmp.i = ns.i;
1276                 ns.i = os.i;
1277                 ns.disk = D_NEGOTIATING;
1278
1279                 /* We expect to receive up-to-date UUIDs soon.
1280                    To avoid a race in receive_state, free p_uuid while
1281                    holding req_lock. I.e. atomic with the state change */
1282                 kfree(mdev->p_uuid);
1283                 mdev->p_uuid = NULL;
1284         }
1285
1286         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1287         ns = mdev->state;
1288         spin_unlock_irq(&mdev->req_lock);
1289
1290         if (rv < SS_SUCCESS)
1291                 goto force_diskless_dec;
1292
1293         if (mdev->state.role == R_PRIMARY)
1294                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1295         else
1296                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1297
1298         drbd_md_mark_dirty(mdev);
1299         drbd_md_sync(mdev);
1300
1301         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1302         put_ldev(mdev);
1303         reply->ret_code = retcode;
1304         drbd_reconfig_done(mdev);
1305         return 0;
1306
1307  force_diskless_dec:
1308         put_ldev(mdev);
1309  force_diskless:
1310         drbd_force_state(mdev, NS(disk, D_FAILED));
1311         drbd_md_sync(mdev);
1312  fail:
1313         if (nbc) {
1314                 if (nbc->backing_bdev)
1315                         blkdev_put(nbc->backing_bdev,
1316                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1317                 if (nbc->md_bdev)
1318                         blkdev_put(nbc->md_bdev,
1319                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1320                 kfree(nbc);
1321         }
1322         lc_destroy(resync_lru);
1323
1324         reply->ret_code = retcode;
1325         drbd_reconfig_done(mdev);
1326         return 0;
1327 }
1328
1329 /* Detaching the disk is a process in multiple stages.  First we need to lock
1330  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1331  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1332  * internal references as well.
1333  * Only then we have finally detached. */
1334 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1335                           struct drbd_nl_cfg_reply *reply)
1336 {
1337         enum drbd_ret_code retcode;
1338         int ret;
1339         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1340         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1341         /* D_FAILED will transition to DISKLESS. */
1342         ret = wait_event_interruptible(mdev->misc_wait,
1343                         mdev->state.disk != D_FAILED);
1344         drbd_resume_io(mdev);
1345         if ((int)retcode == (int)SS_IS_DISKLESS)
1346                 retcode = SS_NOTHING_TO_DO;
1347         if (ret)
1348                 retcode = ERR_INTR;
1349         reply->ret_code = retcode;
1350         return 0;
1351 }
1352
1353 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1354                             struct drbd_nl_cfg_reply *reply)
1355 {
1356         int i, ns;
1357         enum drbd_ret_code retcode;
1358         struct net_conf *new_conf = NULL;
1359         struct crypto_hash *tfm = NULL;
1360         struct crypto_hash *integrity_w_tfm = NULL;
1361         struct crypto_hash *integrity_r_tfm = NULL;
1362         struct hlist_head *new_tl_hash = NULL;
1363         struct hlist_head *new_ee_hash = NULL;
1364         struct drbd_conf *odev;
1365         char hmac_name[CRYPTO_MAX_ALG_NAME];
1366         void *int_dig_out = NULL;
1367         void *int_dig_in = NULL;
1368         void *int_dig_vv = NULL;
1369         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1370
1371         drbd_reconfig_start(mdev);
1372
1373         if (mdev->state.conn > C_STANDALONE) {
1374                 retcode = ERR_NET_CONFIGURED;
1375                 goto fail;
1376         }
1377
1378         /* allocation not in the IO path, cqueue thread context */
1379         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1380         if (!new_conf) {
1381                 retcode = ERR_NOMEM;
1382                 goto fail;
1383         }
1384
1385         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1386         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1387         new_conf->ping_int         = DRBD_PING_INT_DEF;
1388         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1389         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1390         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1391         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1392         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1393         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1394         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1395         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1396         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1397         new_conf->want_lose        = 0;
1398         new_conf->two_primaries    = 0;
1399         new_conf->wire_protocol    = DRBD_PROT_C;
1400         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1401         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1402         new_conf->on_congestion    = DRBD_ON_CONGESTION_DEF;
1403         new_conf->cong_extents     = DRBD_CONG_EXTENTS_DEF;
1404
1405         if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1406                 retcode = ERR_MANDATORY_TAG;
1407                 goto fail;
1408         }
1409
1410         if (new_conf->two_primaries
1411             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1412                 retcode = ERR_NOT_PROTO_C;
1413                 goto fail;
1414         }
1415
1416         if (get_ldev(mdev)) {
1417                 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1418                 put_ldev(mdev);
1419                 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1420                         retcode = ERR_STONITH_AND_PROT_A;
1421                         goto fail;
1422                 }
1423         }
1424
1425         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1426                 retcode = ERR_CONG_NOT_PROTO_A;
1427                 goto fail;
1428         }
1429
1430         if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1431                 retcode = ERR_DISCARD;
1432                 goto fail;
1433         }
1434
1435         retcode = NO_ERROR;
1436
1437         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1438         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1439         for (i = 0; i < minor_count; i++) {
1440                 odev = minor_to_mdev(i);
1441                 if (!odev || odev == mdev)
1442                         continue;
1443                 if (get_net_conf(odev)) {
1444                         taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1445                         if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1446                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1447                                 retcode = ERR_LOCAL_ADDR;
1448
1449                         taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1450                         if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1451                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1452                                 retcode = ERR_PEER_ADDR;
1453
1454                         put_net_conf(odev);
1455                         if (retcode != NO_ERROR)
1456                                 goto fail;
1457                 }
1458         }
1459
1460         if (new_conf->cram_hmac_alg[0] != 0) {
1461                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1462                         new_conf->cram_hmac_alg);
1463                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1464                 if (IS_ERR(tfm)) {
1465                         tfm = NULL;
1466                         retcode = ERR_AUTH_ALG;
1467                         goto fail;
1468                 }
1469
1470                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1471                         retcode = ERR_AUTH_ALG_ND;
1472                         goto fail;
1473                 }
1474         }
1475
1476         if (new_conf->integrity_alg[0]) {
1477                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1478                 if (IS_ERR(integrity_w_tfm)) {
1479                         integrity_w_tfm = NULL;
1480                         retcode=ERR_INTEGRITY_ALG;
1481                         goto fail;
1482                 }
1483
1484                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1485                         retcode=ERR_INTEGRITY_ALG_ND;
1486                         goto fail;
1487                 }
1488
1489                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1490                 if (IS_ERR(integrity_r_tfm)) {
1491                         integrity_r_tfm = NULL;
1492                         retcode=ERR_INTEGRITY_ALG;
1493                         goto fail;
1494                 }
1495         }
1496
1497         ns = new_conf->max_epoch_size/8;
1498         if (mdev->tl_hash_s != ns) {
1499                 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1500                 if (!new_tl_hash) {
1501                         retcode = ERR_NOMEM;
1502                         goto fail;
1503                 }
1504         }
1505
1506         ns = new_conf->max_buffers/8;
1507         if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1508                 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1509                 if (!new_ee_hash) {
1510                         retcode = ERR_NOMEM;
1511                         goto fail;
1512                 }
1513         }
1514
1515         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1516
1517         if (integrity_w_tfm) {
1518                 i = crypto_hash_digestsize(integrity_w_tfm);
1519                 int_dig_out = kmalloc(i, GFP_KERNEL);
1520                 if (!int_dig_out) {
1521                         retcode = ERR_NOMEM;
1522                         goto fail;
1523                 }
1524                 int_dig_in = kmalloc(i, GFP_KERNEL);
1525                 if (!int_dig_in) {
1526                         retcode = ERR_NOMEM;
1527                         goto fail;
1528                 }
1529                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1530                 if (!int_dig_vv) {
1531                         retcode = ERR_NOMEM;
1532                         goto fail;
1533                 }
1534         }
1535
1536         if (!mdev->bitmap) {
1537                 if(drbd_bm_init(mdev)) {
1538                         retcode = ERR_NOMEM;
1539                         goto fail;
1540                 }
1541         }
1542
1543         drbd_flush_workqueue(mdev);
1544         spin_lock_irq(&mdev->req_lock);
1545         if (mdev->net_conf != NULL) {
1546                 retcode = ERR_NET_CONFIGURED;
1547                 spin_unlock_irq(&mdev->req_lock);
1548                 goto fail;
1549         }
1550         mdev->net_conf = new_conf;
1551
1552         mdev->send_cnt = 0;
1553         mdev->recv_cnt = 0;
1554
1555         if (new_tl_hash) {
1556                 kfree(mdev->tl_hash);
1557                 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1558                 mdev->tl_hash = new_tl_hash;
1559         }
1560
1561         if (new_ee_hash) {
1562                 kfree(mdev->ee_hash);
1563                 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1564                 mdev->ee_hash = new_ee_hash;
1565         }
1566
1567         crypto_free_hash(mdev->cram_hmac_tfm);
1568         mdev->cram_hmac_tfm = tfm;
1569
1570         crypto_free_hash(mdev->integrity_w_tfm);
1571         mdev->integrity_w_tfm = integrity_w_tfm;
1572
1573         crypto_free_hash(mdev->integrity_r_tfm);
1574         mdev->integrity_r_tfm = integrity_r_tfm;
1575
1576         kfree(mdev->int_dig_out);
1577         kfree(mdev->int_dig_in);
1578         kfree(mdev->int_dig_vv);
1579         mdev->int_dig_out=int_dig_out;
1580         mdev->int_dig_in=int_dig_in;
1581         mdev->int_dig_vv=int_dig_vv;
1582         retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL);
1583         spin_unlock_irq(&mdev->req_lock);
1584
1585         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1586         reply->ret_code = retcode;
1587         drbd_reconfig_done(mdev);
1588         return 0;
1589
1590 fail:
1591         kfree(int_dig_out);
1592         kfree(int_dig_in);
1593         kfree(int_dig_vv);
1594         crypto_free_hash(tfm);
1595         crypto_free_hash(integrity_w_tfm);
1596         crypto_free_hash(integrity_r_tfm);
1597         kfree(new_tl_hash);
1598         kfree(new_ee_hash);
1599         kfree(new_conf);
1600
1601         reply->ret_code = retcode;
1602         drbd_reconfig_done(mdev);
1603         return 0;
1604 }
1605
1606 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1607                               struct drbd_nl_cfg_reply *reply)
1608 {
1609         int retcode;
1610         struct disconnect dc;
1611
1612         memset(&dc, 0, sizeof(struct disconnect));
1613         if (!disconnect_from_tags(mdev, nlp->tag_list, &dc)) {
1614                 retcode = ERR_MANDATORY_TAG;
1615                 goto fail;
1616         }
1617
1618         if (dc.force) {
1619                 spin_lock_irq(&mdev->req_lock);
1620                 if (mdev->state.conn >= C_WF_CONNECTION)
1621                         _drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), CS_HARD, NULL);
1622                 spin_unlock_irq(&mdev->req_lock);
1623                 goto done;
1624         }
1625
1626         retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1627
1628         if (retcode == SS_NOTHING_TO_DO)
1629                 goto done;
1630         else if (retcode == SS_ALREADY_STANDALONE)
1631                 goto done;
1632         else if (retcode == SS_PRIMARY_NOP) {
1633                 /* Our statche checking code wants to see the peer outdated. */
1634                 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1635                                                       pdsk, D_OUTDATED));
1636         } else if (retcode == SS_CW_FAILED_BY_PEER) {
1637                 /* The peer probably wants to see us outdated. */
1638                 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1639                                                         disk, D_OUTDATED),
1640                                               CS_ORDERED);
1641                 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1642                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1643                         retcode = SS_SUCCESS;
1644                 }
1645         }
1646
1647         if (retcode < SS_SUCCESS)
1648                 goto fail;
1649
1650         if (wait_event_interruptible(mdev->state_wait,
1651                                      mdev->state.conn != C_DISCONNECTING)) {
1652                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1653                    someone else might connect us in the mean time! */
1654                 retcode = ERR_INTR;
1655                 goto fail;
1656         }
1657
1658  done:
1659         retcode = NO_ERROR;
1660  fail:
1661         drbd_md_sync(mdev);
1662         reply->ret_code = retcode;
1663         return 0;
1664 }
1665
1666 void resync_after_online_grow(struct drbd_conf *mdev)
1667 {
1668         int iass; /* I am sync source */
1669
1670         dev_info(DEV, "Resync of new storage after online grow\n");
1671         if (mdev->state.role != mdev->state.peer)
1672                 iass = (mdev->state.role == R_PRIMARY);
1673         else
1674                 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1675
1676         if (iass)
1677                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1678         else
1679                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1680 }
1681
1682 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1683                           struct drbd_nl_cfg_reply *reply)
1684 {
1685         struct resize rs;
1686         int retcode = NO_ERROR;
1687         enum determine_dev_size dd;
1688         enum dds_flags ddsf;
1689
1690         memset(&rs, 0, sizeof(struct resize));
1691         if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1692                 retcode = ERR_MANDATORY_TAG;
1693                 goto fail;
1694         }
1695
1696         if (mdev->state.conn > C_CONNECTED) {
1697                 retcode = ERR_RESIZE_RESYNC;
1698                 goto fail;
1699         }
1700
1701         if (mdev->state.role == R_SECONDARY &&
1702             mdev->state.peer == R_SECONDARY) {
1703                 retcode = ERR_NO_PRIMARY;
1704                 goto fail;
1705         }
1706
1707         if (!get_ldev(mdev)) {
1708                 retcode = ERR_NO_DISK;
1709                 goto fail;
1710         }
1711
1712         if (rs.no_resync && mdev->agreed_pro_version < 93) {
1713                 retcode = ERR_NEED_APV_93;
1714                 goto fail;
1715         }
1716
1717         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1718                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1719
1720         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1721         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1722         dd = drbd_determine_dev_size(mdev, ddsf);
1723         drbd_md_sync(mdev);
1724         put_ldev(mdev);
1725         if (dd == dev_size_error) {
1726                 retcode = ERR_NOMEM_BITMAP;
1727                 goto fail;
1728         }
1729
1730         if (mdev->state.conn == C_CONNECTED) {
1731                 if (dd == grew)
1732                         set_bit(RESIZE_PENDING, &mdev->flags);
1733
1734                 drbd_send_uuids(mdev);
1735                 drbd_send_sizes(mdev, 1, ddsf);
1736         }
1737
1738  fail:
1739         reply->ret_code = retcode;
1740         return 0;
1741 }
1742
1743 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1744                                struct drbd_nl_cfg_reply *reply)
1745 {
1746         int retcode = NO_ERROR;
1747         int err;
1748         int ovr; /* online verify running */
1749         int rsr; /* re-sync running */
1750         struct crypto_hash *verify_tfm = NULL;
1751         struct crypto_hash *csums_tfm = NULL;
1752         struct syncer_conf sc;
1753         cpumask_var_t new_cpu_mask;
1754         int *rs_plan_s = NULL;
1755         int fifo_size;
1756
1757         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1758                 retcode = ERR_NOMEM;
1759                 goto fail;
1760         }
1761
1762         if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1763                 memset(&sc, 0, sizeof(struct syncer_conf));
1764                 sc.rate       = DRBD_RATE_DEF;
1765                 sc.after      = DRBD_AFTER_DEF;
1766                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1767                 sc.on_no_data  = DRBD_ON_NO_DATA_DEF;
1768                 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1769                 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1770                 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1771                 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1772                 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1773         } else
1774                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1775
1776         if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1777                 retcode = ERR_MANDATORY_TAG;
1778                 goto fail;
1779         }
1780
1781         /* re-sync running */
1782         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1783                 mdev->state.conn == C_SYNC_TARGET ||
1784                 mdev->state.conn == C_PAUSED_SYNC_S ||
1785                 mdev->state.conn == C_PAUSED_SYNC_T );
1786
1787         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1788                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1789                 goto fail;
1790         }
1791
1792         if (!rsr && sc.csums_alg[0]) {
1793                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1794                 if (IS_ERR(csums_tfm)) {
1795                         csums_tfm = NULL;
1796                         retcode = ERR_CSUMS_ALG;
1797                         goto fail;
1798                 }
1799
1800                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1801                         retcode = ERR_CSUMS_ALG_ND;
1802                         goto fail;
1803                 }
1804         }
1805
1806         /* online verify running */
1807         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1808
1809         if (ovr) {
1810                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1811                         retcode = ERR_VERIFY_RUNNING;
1812                         goto fail;
1813                 }
1814         }
1815
1816         if (!ovr && sc.verify_alg[0]) {
1817                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1818                 if (IS_ERR(verify_tfm)) {
1819                         verify_tfm = NULL;
1820                         retcode = ERR_VERIFY_ALG;
1821                         goto fail;
1822                 }
1823
1824                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1825                         retcode = ERR_VERIFY_ALG_ND;
1826                         goto fail;
1827                 }
1828         }
1829
1830         /* silently ignore cpu mask on UP kernel */
1831         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1832                 err = bitmap_parse(sc.cpu_mask, 32,
1833                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1834                 if (err) {
1835                         dev_warn(DEV, "bitmap_parse() failed with %d\n", err);
1836                         retcode = ERR_CPU_MASK_PARSE;
1837                         goto fail;
1838                 }
1839         }
1840
1841         ERR_IF (sc.rate < 1) sc.rate = 1;
1842         ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1843 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1844         if (sc.al_extents > AL_MAX) {
1845                 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1846                 sc.al_extents = AL_MAX;
1847         }
1848 #undef AL_MAX
1849
1850         /* to avoid spurious errors when configuring minors before configuring
1851          * the minors they depend on: if necessary, first create the minor we
1852          * depend on */
1853         if (sc.after >= 0)
1854                 ensure_mdev(sc.after, 1);
1855
1856         /* most sanity checks done, try to assign the new sync-after
1857          * dependency.  need to hold the global lock in there,
1858          * to avoid a race in the dependency loop check. */
1859         retcode = drbd_alter_sa(mdev, sc.after);
1860         if (retcode != NO_ERROR)
1861                 goto fail;
1862
1863         fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1864         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1865                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1866                 if (!rs_plan_s) {
1867                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1868                         retcode = ERR_NOMEM;
1869                         goto fail;
1870                 }
1871         }
1872
1873         /* ok, assign the rest of it as well.
1874          * lock against receive_SyncParam() */
1875         spin_lock(&mdev->peer_seq_lock);
1876         mdev->sync_conf = sc;
1877
1878         if (!rsr) {
1879                 crypto_free_hash(mdev->csums_tfm);
1880                 mdev->csums_tfm = csums_tfm;
1881                 csums_tfm = NULL;
1882         }
1883
1884         if (!ovr) {
1885                 crypto_free_hash(mdev->verify_tfm);
1886                 mdev->verify_tfm = verify_tfm;
1887                 verify_tfm = NULL;
1888         }
1889
1890         if (fifo_size != mdev->rs_plan_s.size) {
1891                 kfree(mdev->rs_plan_s.values);
1892                 mdev->rs_plan_s.values = rs_plan_s;
1893                 mdev->rs_plan_s.size   = fifo_size;
1894                 mdev->rs_planed = 0;
1895                 rs_plan_s = NULL;
1896         }
1897
1898         spin_unlock(&mdev->peer_seq_lock);
1899
1900         if (get_ldev(mdev)) {
1901                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1902                 drbd_al_shrink(mdev);
1903                 err = drbd_check_al_size(mdev);
1904                 lc_unlock(mdev->act_log);
1905                 wake_up(&mdev->al_wait);
1906
1907                 put_ldev(mdev);
1908                 drbd_md_sync(mdev);
1909
1910                 if (err) {
1911                         retcode = ERR_NOMEM;
1912                         goto fail;
1913                 }
1914         }
1915
1916         if (mdev->state.conn >= C_CONNECTED)
1917                 drbd_send_sync_param(mdev, &sc);
1918
1919         if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1920                 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1921                 drbd_calc_cpu_mask(mdev);
1922                 mdev->receiver.reset_cpu_mask = 1;
1923                 mdev->asender.reset_cpu_mask = 1;
1924                 mdev->worker.reset_cpu_mask = 1;
1925         }
1926
1927         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1928 fail:
1929         kfree(rs_plan_s);
1930         free_cpumask_var(new_cpu_mask);
1931         crypto_free_hash(csums_tfm);
1932         crypto_free_hash(verify_tfm);
1933         reply->ret_code = retcode;
1934         return 0;
1935 }
1936
1937 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1938                               struct drbd_nl_cfg_reply *reply)
1939 {
1940         int retcode;
1941
1942         /* If there is still bitmap IO pending, probably because of a previous
1943          * resync just being finished, wait for it before requesting a new resync. */
1944         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1945
1946         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1947
1948         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1949                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1950
1951         while (retcode == SS_NEED_CONNECTION) {
1952                 spin_lock_irq(&mdev->req_lock);
1953                 if (mdev->state.conn < C_CONNECTED)
1954                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1955                 spin_unlock_irq(&mdev->req_lock);
1956
1957                 if (retcode != SS_NEED_CONNECTION)
1958                         break;
1959
1960                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1961         }
1962
1963         reply->ret_code = retcode;
1964         return 0;
1965 }
1966
1967 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
1968 {
1969         int rv;
1970
1971         rv = drbd_bmio_set_n_write(mdev);
1972         drbd_suspend_al(mdev);
1973         return rv;
1974 }
1975
1976 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1977                                    struct drbd_nl_cfg_reply *reply)
1978 {
1979         int retcode;
1980
1981         /* If there is still bitmap IO pending, probably because of a previous
1982          * resync just being finished, wait for it before requesting a new resync. */
1983         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1984
1985         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
1986
1987         if (retcode < SS_SUCCESS) {
1988                 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
1989                         /* The peer will get a resync upon connect anyways. Just make that
1990                            into a full resync. */
1991                         retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
1992                         if (retcode >= SS_SUCCESS) {
1993                                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
1994                                         "set_n_write from invalidate_peer",
1995                                         BM_LOCKED_SET_ALLOWED))
1996                                         retcode = ERR_IO_MD_DISK;
1997                         }
1998                 } else
1999                         retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
2000         }
2001
2002         reply->ret_code = retcode;
2003         return 0;
2004 }
2005
2006 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2007                               struct drbd_nl_cfg_reply *reply)
2008 {
2009         int retcode = NO_ERROR;
2010
2011         if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2012                 retcode = ERR_PAUSE_IS_SET;
2013
2014         reply->ret_code = retcode;
2015         return 0;
2016 }
2017
2018 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2019                                struct drbd_nl_cfg_reply *reply)
2020 {
2021         int retcode = NO_ERROR;
2022         union drbd_state s;
2023
2024         if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2025                 s = mdev->state;
2026                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2027                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2028                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2029                 } else {
2030                         retcode = ERR_PAUSE_IS_CLEAR;
2031                 }
2032         }
2033
2034         reply->ret_code = retcode;
2035         return 0;
2036 }
2037
2038 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2039                               struct drbd_nl_cfg_reply *reply)
2040 {
2041         reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
2042
2043         return 0;
2044 }
2045
2046 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2047                              struct drbd_nl_cfg_reply *reply)
2048 {
2049         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2050                 drbd_uuid_new_current(mdev);
2051                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2052         }
2053         drbd_suspend_io(mdev);
2054         reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2055         if (reply->ret_code == SS_SUCCESS) {
2056                 if (mdev->state.conn < C_CONNECTED)
2057                         tl_clear(mdev);
2058                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2059                         tl_restart(mdev, fail_frozen_disk_io);
2060         }
2061         drbd_resume_io(mdev);
2062
2063         return 0;
2064 }
2065
2066 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2067                            struct drbd_nl_cfg_reply *reply)
2068 {
2069         reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
2070         return 0;
2071 }
2072
2073 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2074                            struct drbd_nl_cfg_reply *reply)
2075 {
2076         unsigned short *tl;
2077
2078         tl = reply->tag_list;
2079
2080         if (get_ldev(mdev)) {
2081                 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
2082                 put_ldev(mdev);
2083         }
2084
2085         if (get_net_conf(mdev)) {
2086                 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
2087                 put_net_conf(mdev);
2088         }
2089         tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
2090
2091         put_unaligned(TT_END, tl++); /* Close the tag list */
2092
2093         return (int)((char *)tl - (char *)reply->tag_list);
2094 }
2095
2096 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2097                              struct drbd_nl_cfg_reply *reply)
2098 {
2099         unsigned short *tl = reply->tag_list;
2100         union drbd_state s = mdev->state;
2101         unsigned long rs_left;
2102         unsigned int res;
2103
2104         tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
2105
2106         /* no local ref, no bitmap, no syncer progress. */
2107         if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
2108                 if (get_ldev(mdev)) {
2109                         drbd_get_syncer_progress(mdev, &rs_left, &res);
2110                         tl = tl_add_int(tl, T_sync_progress, &res);
2111                         put_ldev(mdev);
2112                 }
2113         }
2114         put_unaligned(TT_END, tl++); /* Close the tag list */
2115
2116         return (int)((char *)tl - (char *)reply->tag_list);
2117 }
2118
2119 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2120                              struct drbd_nl_cfg_reply *reply)
2121 {
2122         unsigned short *tl;
2123
2124         tl = reply->tag_list;
2125
2126         if (get_ldev(mdev)) {
2127                 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
2128                 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
2129                 put_ldev(mdev);
2130         }
2131         put_unaligned(TT_END, tl++); /* Close the tag list */
2132
2133         return (int)((char *)tl - (char *)reply->tag_list);
2134 }
2135
2136 /**
2137  * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
2138  * @mdev:       DRBD device.
2139  * @nlp:        Netlink/connector packet from drbdsetup
2140  * @reply:      Reply packet for drbdsetup
2141  */
2142 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2143                                     struct drbd_nl_cfg_reply *reply)
2144 {
2145         unsigned short *tl;
2146         char rv;
2147
2148         tl = reply->tag_list;
2149
2150         rv = mdev->state.pdsk == D_OUTDATED        ? UT_PEER_OUTDATED :
2151           test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
2152
2153         tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
2154         put_unaligned(TT_END, tl++); /* Close the tag list */
2155
2156         return (int)((char *)tl - (char *)reply->tag_list);
2157 }
2158
2159 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2160                                     struct drbd_nl_cfg_reply *reply)
2161 {
2162         /* default to resume from last known position, if possible */
2163         struct start_ov args =
2164                 { .start_sector = mdev->ov_start_sector };
2165
2166         if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2167                 reply->ret_code = ERR_MANDATORY_TAG;
2168                 return 0;
2169         }
2170
2171         /* If there is still bitmap IO pending, e.g. previous resync or verify
2172          * just being finished, wait for it before requesting a new resync. */
2173         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2174
2175         /* w_make_ov_request expects position to be aligned */
2176         mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
2177         reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2178         return 0;
2179 }
2180
2181
2182 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2183                               struct drbd_nl_cfg_reply *reply)
2184 {
2185         int retcode = NO_ERROR;
2186         int skip_initial_sync = 0;
2187         int err;
2188
2189         struct new_c_uuid args;
2190
2191         memset(&args, 0, sizeof(struct new_c_uuid));
2192         if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2193                 reply->ret_code = ERR_MANDATORY_TAG;
2194                 return 0;
2195         }
2196
2197         mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
2198
2199         if (!get_ldev(mdev)) {
2200                 retcode = ERR_NO_DISK;
2201                 goto out;
2202         }
2203
2204         /* this is "skip initial sync", assume to be clean */
2205         if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
2206             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2207                 dev_info(DEV, "Preparing to skip initial sync\n");
2208                 skip_initial_sync = 1;
2209         } else if (mdev->state.conn != C_STANDALONE) {
2210                 retcode = ERR_CONNECTED;
2211                 goto out_dec;
2212         }
2213
2214         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2215         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2216
2217         if (args.clear_bm) {
2218                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2219                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2220                 if (err) {
2221                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2222                         retcode = ERR_IO_MD_DISK;
2223                 }
2224                 if (skip_initial_sync) {
2225                         drbd_send_uuids_skip_initial_sync(mdev);
2226                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2227                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2228                         spin_lock_irq(&mdev->req_lock);
2229                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2230                                         CS_VERBOSE, NULL);
2231                         spin_unlock_irq(&mdev->req_lock);
2232                 }
2233         }
2234
2235         drbd_md_sync(mdev);
2236 out_dec:
2237         put_ldev(mdev);
2238 out:
2239         mutex_unlock(&mdev->state_mutex);
2240
2241         reply->ret_code = retcode;
2242         return 0;
2243 }
2244
2245 struct cn_handler_struct {
2246         int (*function)(struct drbd_conf *,
2247                          struct drbd_nl_cfg_req *,
2248                          struct drbd_nl_cfg_reply *);
2249         int reply_body_size;
2250 };
2251
2252 static struct cn_handler_struct cnd_table[] = {
2253         [ P_primary ]           = { &drbd_nl_primary,           0 },
2254         [ P_secondary ]         = { &drbd_nl_secondary,         0 },
2255         [ P_disk_conf ]         = { &drbd_nl_disk_conf,         0 },
2256         [ P_detach ]            = { &drbd_nl_detach,            0 },
2257         [ P_net_conf ]          = { &drbd_nl_net_conf,          0 },
2258         [ P_disconnect ]        = { &drbd_nl_disconnect,        0 },
2259         [ P_resize ]            = { &drbd_nl_resize,            0 },
2260         [ P_syncer_conf ]       = { &drbd_nl_syncer_conf,       0 },
2261         [ P_invalidate ]        = { &drbd_nl_invalidate,        0 },
2262         [ P_invalidate_peer ]   = { &drbd_nl_invalidate_peer,   0 },
2263         [ P_pause_sync ]        = { &drbd_nl_pause_sync,        0 },
2264         [ P_resume_sync ]       = { &drbd_nl_resume_sync,       0 },
2265         [ P_suspend_io ]        = { &drbd_nl_suspend_io,        0 },
2266         [ P_resume_io ]         = { &drbd_nl_resume_io,         0 },
2267         [ P_outdate ]           = { &drbd_nl_outdate,           0 },
2268         [ P_get_config ]        = { &drbd_nl_get_config,
2269                                     sizeof(struct syncer_conf_tag_len_struct) +
2270                                     sizeof(struct disk_conf_tag_len_struct) +
2271                                     sizeof(struct net_conf_tag_len_struct) },
2272         [ P_get_state ]         = { &drbd_nl_get_state,
2273                                     sizeof(struct get_state_tag_len_struct) +
2274                                     sizeof(struct sync_progress_tag_len_struct) },
2275         [ P_get_uuids ]         = { &drbd_nl_get_uuids,
2276                                     sizeof(struct get_uuids_tag_len_struct) },
2277         [ P_get_timeout_flag ]  = { &drbd_nl_get_timeout_flag,
2278                                     sizeof(struct get_timeout_flag_tag_len_struct)},
2279         [ P_start_ov ]          = { &drbd_nl_start_ov,          0 },
2280         [ P_new_c_uuid ]        = { &drbd_nl_new_c_uuid,        0 },
2281 };
2282
2283 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2284 {
2285         struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2286         struct cn_handler_struct *cm;
2287         struct cn_msg *cn_reply;
2288         struct drbd_nl_cfg_reply *reply;
2289         struct drbd_conf *mdev;
2290         int retcode, rr;
2291         int reply_size = sizeof(struct cn_msg)
2292                 + sizeof(struct drbd_nl_cfg_reply)
2293                 + sizeof(short int);
2294
2295         if (!try_module_get(THIS_MODULE)) {
2296                 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2297                 return;
2298         }
2299
2300         if (!cap_raised(current_cap(), CAP_SYS_ADMIN)) {
2301                 retcode = ERR_PERM;
2302                 goto fail;
2303         }
2304
2305         mdev = ensure_mdev(nlp->drbd_minor,
2306                         (nlp->flags & DRBD_NL_CREATE_DEVICE));
2307         if (!mdev) {
2308                 retcode = ERR_MINOR_INVALID;
2309                 goto fail;
2310         }
2311
2312         if (nlp->packet_type >= P_nl_after_last_packet ||
2313             nlp->packet_type == P_return_code_only) {
2314                 retcode = ERR_PACKET_NR;
2315                 goto fail;
2316         }
2317
2318         cm = cnd_table + nlp->packet_type;
2319
2320         /* This may happen if packet number is 0: */
2321         if (cm->function == NULL) {
2322                 retcode = ERR_PACKET_NR;
2323                 goto fail;
2324         }
2325
2326         reply_size += cm->reply_body_size;
2327
2328         /* allocation not in the IO path, cqueue thread context */
2329         cn_reply = kzalloc(reply_size, GFP_KERNEL);
2330         if (!cn_reply) {
2331                 retcode = ERR_NOMEM;
2332                 goto fail;
2333         }
2334         reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2335
2336         reply->packet_type =
2337                 cm->reply_body_size ? nlp->packet_type : P_return_code_only;
2338         reply->minor = nlp->drbd_minor;
2339         reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2340         /* reply->tag_list; might be modified by cm->function. */
2341
2342         rr = cm->function(mdev, nlp, reply);
2343
2344         cn_reply->id = req->id;
2345         cn_reply->seq = req->seq;
2346         cn_reply->ack = req->ack  + 1;
2347         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2348         cn_reply->flags = 0;
2349
2350         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2351         if (rr && rr != -ESRCH)
2352                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2353
2354         kfree(cn_reply);
2355         module_put(THIS_MODULE);
2356         return;
2357  fail:
2358         drbd_nl_send_reply(req, retcode);
2359         module_put(THIS_MODULE);
2360 }
2361
2362 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2363
2364 static unsigned short *
2365 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2366         unsigned short len, int nul_terminated)
2367 {
2368         unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2369         len = (len < l) ? len :  l;
2370         put_unaligned(tag, tl++);
2371         put_unaligned(len, tl++);
2372         memcpy(tl, data, len);
2373         tl = (unsigned short*)((char*)tl + len);
2374         if (nul_terminated)
2375                 *((char*)tl - 1) = 0;
2376         return tl;
2377 }
2378
2379 static unsigned short *
2380 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2381 {
2382         return __tl_add_blob(tl, tag, data, len, 0);
2383 }
2384
2385 static unsigned short *
2386 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2387 {
2388         return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2389 }
2390
2391 static unsigned short *
2392 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2393 {
2394         put_unaligned(tag, tl++);
2395         switch(tag_type(tag)) {
2396         case TT_INTEGER:
2397                 put_unaligned(sizeof(int), tl++);
2398                 put_unaligned(*(int *)val, (int *)tl);
2399                 tl = (unsigned short*)((char*)tl+sizeof(int));
2400                 break;
2401         case TT_INT64:
2402                 put_unaligned(sizeof(u64), tl++);
2403                 put_unaligned(*(u64 *)val, (u64 *)tl);
2404                 tl = (unsigned short*)((char*)tl+sizeof(u64));
2405                 break;
2406         default:
2407                 /* someone did something stupid. */
2408                 ;
2409         }
2410         return tl;
2411 }
2412
2413 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2414 {
2415         char buffer[sizeof(struct cn_msg)+
2416                     sizeof(struct drbd_nl_cfg_reply)+
2417                     sizeof(struct get_state_tag_len_struct)+
2418                     sizeof(short int)];
2419         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2420         struct drbd_nl_cfg_reply *reply =
2421                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2422         unsigned short *tl = reply->tag_list;
2423
2424         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2425
2426         tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2427
2428         put_unaligned(TT_END, tl++); /* Close the tag list */
2429
2430         cn_reply->id.idx = CN_IDX_DRBD;
2431         cn_reply->id.val = CN_VAL_DRBD;
2432
2433         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2434         cn_reply->ack = 0; /* not used here. */
2435         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2436                 (int)((char *)tl - (char *)reply->tag_list);
2437         cn_reply->flags = 0;
2438
2439         reply->packet_type = P_get_state;
2440         reply->minor = mdev_to_minor(mdev);
2441         reply->ret_code = NO_ERROR;
2442
2443         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2444 }
2445
2446 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2447 {
2448         char buffer[sizeof(struct cn_msg)+
2449                     sizeof(struct drbd_nl_cfg_reply)+
2450                     sizeof(struct call_helper_tag_len_struct)+
2451                     sizeof(short int)];
2452         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2453         struct drbd_nl_cfg_reply *reply =
2454                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2455         unsigned short *tl = reply->tag_list;
2456
2457         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2458
2459         tl = tl_add_str(tl, T_helper, helper_name);
2460         put_unaligned(TT_END, tl++); /* Close the tag list */
2461
2462         cn_reply->id.idx = CN_IDX_DRBD;
2463         cn_reply->id.val = CN_VAL_DRBD;
2464
2465         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2466         cn_reply->ack = 0; /* not used here. */
2467         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2468                 (int)((char *)tl - (char *)reply->tag_list);
2469         cn_reply->flags = 0;
2470
2471         reply->packet_type = P_call_helper;
2472         reply->minor = mdev_to_minor(mdev);
2473         reply->ret_code = NO_ERROR;
2474
2475         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2476 }
2477
2478 void drbd_bcast_ee(struct drbd_conf *mdev,
2479                 const char *reason, const int dgs,
2480                 const char* seen_hash, const char* calc_hash,
2481                 const struct drbd_epoch_entry* e)
2482 {
2483         struct cn_msg *cn_reply;
2484         struct drbd_nl_cfg_reply *reply;
2485         unsigned short *tl;
2486         struct page *page;
2487         unsigned len;
2488
2489         if (!e)
2490                 return;
2491         if (!reason || !reason[0])
2492                 return;
2493
2494         /* apparently we have to memcpy twice, first to prepare the data for the
2495          * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2496          * netlink skb. */
2497         /* receiver thread context, which is not in the writeout path (of this node),
2498          * but may be in the writeout path of the _other_ node.
2499          * GFP_NOIO to avoid potential "distributed deadlock". */
2500         cn_reply = kzalloc(
2501                 sizeof(struct cn_msg)+
2502                 sizeof(struct drbd_nl_cfg_reply)+
2503                 sizeof(struct dump_ee_tag_len_struct)+
2504                 sizeof(short int),
2505                 GFP_NOIO);
2506
2507         if (!cn_reply) {
2508                 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2509                                 (unsigned long long)e->sector, e->size);
2510                 return;
2511         }
2512
2513         reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2514         tl = reply->tag_list;
2515
2516         tl = tl_add_str(tl, T_dump_ee_reason, reason);
2517         tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2518         tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2519         tl = tl_add_int(tl, T_ee_sector, &e->sector);
2520         tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2521
2522         /* dump the first 32k */
2523         len = min_t(unsigned, e->size, 32 << 10);
2524         put_unaligned(T_ee_data, tl++);
2525         put_unaligned(len, tl++);
2526
2527         page = e->pages;
2528         page_chain_for_each(page) {
2529                 void *d = kmap_atomic(page, KM_USER0);
2530                 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2531                 memcpy(tl, d, l);
2532                 kunmap_atomic(d, KM_USER0);
2533                 tl = (unsigned short*)((char*)tl + l);
2534                 len -= l;
2535                 if (len == 0)
2536                         break;
2537         }
2538         put_unaligned(TT_END, tl++); /* Close the tag list */
2539
2540         cn_reply->id.idx = CN_IDX_DRBD;
2541         cn_reply->id.val = CN_VAL_DRBD;
2542
2543         cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2544         cn_reply->ack = 0; // not used here.
2545         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2546                 (int)((char*)tl - (char*)reply->tag_list);
2547         cn_reply->flags = 0;
2548
2549         reply->packet_type = P_dump_ee;
2550         reply->minor = mdev_to_minor(mdev);
2551         reply->ret_code = NO_ERROR;
2552
2553         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2554         kfree(cn_reply);
2555 }
2556
2557 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2558 {
2559         char buffer[sizeof(struct cn_msg)+
2560                     sizeof(struct drbd_nl_cfg_reply)+
2561                     sizeof(struct sync_progress_tag_len_struct)+
2562                     sizeof(short int)];
2563         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2564         struct drbd_nl_cfg_reply *reply =
2565                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2566         unsigned short *tl = reply->tag_list;
2567         unsigned long rs_left;
2568         unsigned int res;
2569
2570         /* no local ref, no bitmap, no syncer progress, no broadcast. */
2571         if (!get_ldev(mdev))
2572                 return;
2573         drbd_get_syncer_progress(mdev, &rs_left, &res);
2574         put_ldev(mdev);
2575
2576         tl = tl_add_int(tl, T_sync_progress, &res);
2577         put_unaligned(TT_END, tl++); /* Close the tag list */
2578
2579         cn_reply->id.idx = CN_IDX_DRBD;
2580         cn_reply->id.val = CN_VAL_DRBD;
2581
2582         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2583         cn_reply->ack = 0; /* not used here. */
2584         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2585                 (int)((char *)tl - (char *)reply->tag_list);
2586         cn_reply->flags = 0;
2587
2588         reply->packet_type = P_sync_progress;
2589         reply->minor = mdev_to_minor(mdev);
2590         reply->ret_code = NO_ERROR;
2591
2592         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2593 }
2594
2595 int __init drbd_nl_init(void)
2596 {
2597         static struct cb_id cn_id_drbd;
2598         int err, try=10;
2599
2600         cn_id_drbd.val = CN_VAL_DRBD;
2601         do {
2602                 cn_id_drbd.idx = cn_idx;
2603                 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2604                 if (!err)
2605                         break;
2606                 cn_idx = (cn_idx + CN_IDX_STEP);
2607         } while (try--);
2608
2609         if (err) {
2610                 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2611                 return err;
2612         }
2613
2614         return 0;
2615 }
2616
2617 void drbd_nl_cleanup(void)
2618 {
2619         static struct cb_id cn_id_drbd;
2620
2621         cn_id_drbd.idx = cn_idx;
2622         cn_id_drbd.val = CN_VAL_DRBD;
2623
2624         cn_del_callback(&cn_id_drbd);
2625 }
2626
2627 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2628 {
2629         char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2630         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2631         struct drbd_nl_cfg_reply *reply =
2632                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2633         int rr;
2634
2635         memset(buffer, 0, sizeof(buffer));
2636         cn_reply->id = req->id;
2637
2638         cn_reply->seq = req->seq;
2639         cn_reply->ack = req->ack  + 1;
2640         cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2641         cn_reply->flags = 0;
2642
2643         reply->packet_type = P_return_code_only;
2644         reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2645         reply->ret_code = ret_code;
2646
2647         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2648         if (rr && rr != -ESRCH)
2649                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2650 }
2651