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         if (current == mdev->worker.task)
 151                 set_bit(CALLBACK_PENDING, &mdev->flags);
 152
 153         snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
 154
 155         if (get_net_conf(mdev)) {
 156                 switch (((struct sockaddr *)mdev->net_conf->peer_addr)->sa_family) {
 157                 case AF_INET6:
 158                         afs = "ipv6";
 159                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6",
 160                                  &((struct sockaddr_in6 *)mdev->net_conf->peer_addr)->sin6_addr);
 161                         break;
 162                 case AF_INET:
 163                         afs = "ipv4";
 164                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 165                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
 166                         break;
 167                 default:
 168                         afs = "ssocks";
 169                         snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
 170                                  &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
 171                 }
 172                 snprintf(af, 20, "DRBD_PEER_AF=%s", afs);
 173                 envp[3]=af;
 174                 envp[4]=ad;
 175                 put_net_conf(mdev);
 176         }
 177
 178         /* The helper may take some time.
 179          * write out any unsynced meta data changes now */
 180         drbd_md_sync(mdev);
 181
 182         dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
 183
 184         drbd_bcast_ev_helper(mdev, cmd);
 185         ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC);
 186         if (ret)
 187                 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 188                                 usermode_helper, cmd, mb,
 189                                 (ret >> 8) & 0xff, ret);
 190         else
 191                 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
 192                                 usermode_helper, cmd, mb,
 193                                 (ret >> 8) & 0xff, ret);
 194
 195         if (current == mdev->worker.task)
 196                 clear_bit(CALLBACK_PENDING, &mdev->flags);
 197
 198         if (ret < 0) /* Ignore any ERRNOs we got. */
 199                 ret = 0;
 200
 201         return ret;
 202 }
 203
 204 enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
 205 {
 206         char *ex_to_string;
 207         int r;
 208         enum drbd_disk_state nps;
 209         enum drbd_fencing_p fp;
 210
 211         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 212
 213         if (get_ldev_if_state(mdev, D_CONSISTENT)) {
 214                 fp = mdev->ldev->dc.fencing;
 215                 put_ldev(mdev);
 216         } else {
 217                 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n");
 218                 nps = mdev->state.pdsk;
 219                 goto out;
 220         }
 221
 222         r = drbd_khelper(mdev, "fence-peer");
 223
 224         switch ((r>>8) & 0xff) {
 225         case 3: /* peer is inconsistent */
 226                 ex_to_string = "peer is inconsistent or worse";
 227                 nps = D_INCONSISTENT;
 228                 break;
 229         case 4: /* peer got outdated, or was already outdated */
 230                 ex_to_string = "peer was fenced";
 231                 nps = D_OUTDATED;
 232                 break;
 233         case 5: /* peer was down */
 234                 if (mdev->state.disk == D_UP_TO_DATE) {
 235                         /* we will(have) create(d) a new UUID anyways... */
 236                         ex_to_string = "peer is unreachable, assumed to be dead";
 237                         nps = D_OUTDATED;
 238                 } else {
 239                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
 240                         nps = mdev->state.pdsk;
 241                 }
 242                 break;
 243         case 6: /* Peer is primary, voluntarily outdate myself.
 244                  * This is useful when an unconnected R_SECONDARY is asked to
 245                  * become R_PRIMARY, but finds the other peer being active. */
 246                 ex_to_string = "peer is active";
 247                 dev_warn(DEV, "Peer is primary, outdating myself.\n");
 248                 nps = D_UNKNOWN;
 249                 _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE);
 250                 break;
 251         case 7:
 252                 if (fp != FP_STONITH)
 253                         dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n");
 254                 ex_to_string = "peer was stonithed";
 255                 nps = D_OUTDATED;
 256                 break;
 257         default:
 258                 /* The script is broken ... */
 259                 nps = D_UNKNOWN;
 260                 dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
 261                 return nps;
 262         }
 263
 264         dev_info(DEV, "fence-peer helper returned %d (%s)\n",
 265                         (r>>8) & 0xff, ex_to_string);
 266
 267 out:
 268         if (mdev->state.susp_fen && nps >= D_UNKNOWN) {
 269                 /* The handler was not successful... unfreeze here, the
 270                    state engine can not unfreeze... */
 271                 _drbd_request_state(mdev, NS(susp_fen, 0), CS_VERBOSE);
 272         }
 273
 274         return nps;
 275 }
 276
 277 static int _try_outdate_peer_async(void *data)
 278 {
 279         struct drbd_conf *mdev = (struct drbd_conf *)data;
 280         enum drbd_disk_state nps;
 281         union drbd_state ns;
 282
 283         nps = drbd_try_outdate_peer(mdev);
 284
 285         /* Not using
 286            drbd_request_state(mdev, NS(pdsk, nps));
 287            here, because we might were able to re-establish the connection
 288            in the meantime. This can only partially be solved in the state's
 289            engine is_valid_state() and is_valid_state_transition()
 290            functions.
 291
 292            nps can be D_INCONSISTENT, D_OUTDATED or D_UNKNOWN.
 293            pdsk == D_INCONSISTENT while conn >= C_CONNECTED is valid,
 294            therefore we have to have the pre state change check here.
 295         */
 296         spin_lock_irq(&mdev->req_lock);
 297         ns = mdev->state;
 298         if (ns.conn < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &mdev->flags)) {
 299                 ns.pdsk = nps;
 300                 _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
 301         }
 302         spin_unlock_irq(&mdev->req_lock);
 303
 304         return 0;
 305 }
 306
 307 void drbd_try_outdate_peer_async(struct drbd_conf *mdev)
 308 {
 309         struct task_struct *opa;
 310
 311         opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev));
 312         if (IS_ERR(opa))
 313                 dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n");
 314 }
 315
 316 enum drbd_state_rv
 317 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
 318 {
 319         const int max_tries = 4;
 320         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
 321         int try = 0;
 322         int forced = 0;
 323         union drbd_state mask, val;
 324         enum drbd_disk_state nps;
 325
 326         if (new_role == R_PRIMARY)
 327                 request_ping(mdev); /* Detect a dead peer ASAP */
 328
 329         mutex_lock(&mdev->state_mutex);
 330
 331         mask.i = 0; mask.role = R_MASK;
 332         val.i  = 0; val.role  = new_role;
 333
 334         while (try++ < max_tries) {
 335                 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
 336
 337                 /* in case we first succeeded to outdate,
 338                  * but now suddenly could establish a connection */
 339                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
 340                         val.pdsk = 0;
 341                         mask.pdsk = 0;
 342                         continue;
 343                 }
 344
 345                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
 346                     (mdev->state.disk < D_UP_TO_DATE &&
 347                      mdev->state.disk >= D_INCONSISTENT)) {
 348                         mask.disk = D_MASK;
 349                         val.disk  = D_UP_TO_DATE;
 350                         forced = 1;
 351                         continue;
 352                 }
 353
 354                 if (rv == SS_NO_UP_TO_DATE_DISK &&
 355                     mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
 356                         D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
 357                         nps = drbd_try_outdate_peer(mdev);
 358
 359                         if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
 360                                 val.disk = D_UP_TO_DATE;
 361                                 mask.disk = D_MASK;
 362                         }
 363
 364                         val.pdsk = nps;
 365                         mask.pdsk = D_MASK;
 366
 367                         continue;
 368                 }
 369
 370                 if (rv == SS_NOTHING_TO_DO)
 371                         goto fail;
 372                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
 373                         nps = drbd_try_outdate_peer(mdev);
 374
 375                         if (force && nps > D_OUTDATED) {
 376                                 dev_warn(DEV, "Forced into split brain situation!\n");
 377                                 nps = D_OUTDATED;
 378                         }
 379
 380                         mask.pdsk = D_MASK;
 381                         val.pdsk  = nps;
 382
 383                         continue;
 384                 }
 385                 if (rv == SS_TWO_PRIMARIES) {
 386                         /* Maybe the peer is detected as dead very soon...
 387                            retry at most once more in this case. */
 388                         schedule_timeout_interruptible((mdev->net_conf->ping_timeo+1)*HZ/10);
 389                         if (try < max_tries)
 390                                 try = max_tries - 1;
 391                         continue;
 392                 }
 393                 if (rv < SS_SUCCESS) {
 394                         rv = _drbd_request_state(mdev, mask, val,
 395                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
 396                         if (rv < SS_SUCCESS)
 397                                 goto fail;
 398                 }
 399                 break;
 400         }
 401
 402         if (rv < SS_SUCCESS)
 403                 goto fail;
 404
 405         if (forced)
 406                 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
 407
 408         /* Wait until nothing is on the fly :) */
 409         wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
 410
 411         if (new_role == R_SECONDARY) {
 412                 set_disk_ro(mdev->vdisk, true);
 413                 if (get_ldev(mdev)) {
 414                         mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
 415                         put_ldev(mdev);
 416                 }
 417         } else {
 418                 if (get_net_conf(mdev)) {
 419                         mdev->net_conf->want_lose = 0;
 420                         put_net_conf(mdev);
 421                 }
 422                 set_disk_ro(mdev->vdisk, false);
 423                 if (get_ldev(mdev)) {
 424                         if (((mdev->state.conn < C_CONNECTED ||
 425                                mdev->state.pdsk <= D_FAILED)
 426                               && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
 427                                 drbd_uuid_new_current(mdev);
 428
 429                         mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
 430                         put_ldev(mdev);
 431                 }
 432         }
 433
 434         /* writeout of activity log covered areas of the bitmap
 435          * to stable storage done in after state change already */
 436
 437         if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
 438                 /* if this was forced, we should consider sync */
 439                 if (forced)
 440                         drbd_send_uuids(mdev);
 441                 drbd_send_current_state(mdev);
 442         }
 443
 444         drbd_md_sync(mdev);
 445
 446         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
 447  fail:
 448         mutex_unlock(&mdev->state_mutex);
 449         return rv;
 450 }
 451
 452 static struct drbd_conf *ensure_mdev(int minor, int create)
 453 {
 454         struct drbd_conf *mdev;
 455
 456         if (minor >= minor_count)
 457                 return NULL;
 458
 459         mdev = minor_to_mdev(minor);
 460
 461         if (!mdev && create) {
 462                 struct gendisk *disk = NULL;
 463                 mdev = drbd_new_device(minor);
 464
 465                 spin_lock_irq(&drbd_pp_lock);
 466                 if (minor_table[minor] == NULL) {
 467                         minor_table[minor] = mdev;
 468                         disk = mdev->vdisk;
 469                         mdev = NULL;
 470                 } /* else: we lost the race */
 471                 spin_unlock_irq(&drbd_pp_lock);
 472
 473                 if (disk) /* we won the race above */
 474                         /* in case we ever add a drbd_delete_device(),
 475                          * don't forget the del_gendisk! */
 476                         add_disk(disk);
 477                 else /* we lost the race above */
 478                         drbd_free_mdev(mdev);
 479
 480                 mdev = minor_to_mdev(minor);
 481         }
 482
 483         return mdev;
 484 }
 485
 486 static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 487                            struct drbd_nl_cfg_reply *reply)
 488 {
 489         struct primary primary_args;
 490
 491         memset(&primary_args, 0, sizeof(struct primary));
 492         if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) {
 493                 reply->ret_code = ERR_MANDATORY_TAG;
 494                 return 0;
 495         }
 496
 497         reply->ret_code =
 498                 drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force);
 499
 500         return 0;
 501 }
 502
 503 static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 504                              struct drbd_nl_cfg_reply *reply)
 505 {
 506         reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0);
 507
 508         return 0;
 509 }
 510
 511 /* initializes the md.*_offset members, so we are able to find
 512  * the on disk meta data */
 513 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 514                                        struct drbd_backing_dev *bdev)
 515 {
 516         sector_t md_size_sect = 0;
 517         switch (bdev->dc.meta_dev_idx) {
 518         default:
 519                 /* v07 style fixed size indexed meta data */
 520                 bdev->md.md_size_sect = MD_RESERVED_SECT;
 521                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 522                 bdev->md.al_offset = MD_AL_OFFSET;
 523                 bdev->md.bm_offset = MD_BM_OFFSET;
 524                 break;
 525         case DRBD_MD_INDEX_FLEX_EXT:
 526                 /* just occupy the full device; unit: sectors */
 527                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
 528                 bdev->md.md_offset = 0;
 529                 bdev->md.al_offset = MD_AL_OFFSET;
 530                 bdev->md.bm_offset = MD_BM_OFFSET;
 531                 break;
 532         case DRBD_MD_INDEX_INTERNAL:
 533         case DRBD_MD_INDEX_FLEX_INT:
 534                 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 535                 /* al size is still fixed */
 536                 bdev->md.al_offset = -MD_AL_MAX_SIZE;
 537                 /* we need (slightly less than) ~ this much bitmap sectors: */
 538                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 539                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
 540                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
 541                 md_size_sect = ALIGN(md_size_sect, 8);
 542
 543                 /* plus the "drbd meta data super block",
 544                  * and the activity log; */
 545                 md_size_sect += MD_BM_OFFSET;
 546
 547                 bdev->md.md_size_sect = md_size_sect;
 548                 /* bitmap offset is adjusted by 'super' block size */
 549                 bdev->md.bm_offset   = -md_size_sect + MD_AL_OFFSET;
 550                 break;
 551         }
 552 }
 553
 554 /* input size is expected to be in KB */
 555 char *ppsize(char *buf, unsigned long long size)
 556 {
 557         /* Needs 9 bytes at max including trailing NUL:
 558          * -1ULL ==> "16384 EB" */
 559         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
 560         int base = 0;
 561         while (size >= 10000 && base < sizeof(units)-1) {
 562                 /* shift + round */
 563                 size = (size >> 10) + !!(size & (1<<9));
 564                 base++;
 565         }
 566         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
 567
 568         return buf;
 569 }
 570
 571 /* there is still a theoretical deadlock when called from receiver
 572  * on an D_INCONSISTENT R_PRIMARY:
 573  *  remote READ does inc_ap_bio, receiver would need to receive answer
 574  *  packet from remote to dec_ap_bio again.
 575  *  receiver receive_sizes(), comes here,
 576  *  waits for ap_bio_cnt == 0. -> deadlock.
 577  * but this cannot happen, actually, because:
 578  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
 579  *  (not connected, or bad/no disk on peer):
 580  *  see drbd_fail_request_early, ap_bio_cnt is zero.
 581  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
 582  *  peer may not initiate a resize.
 583  */
 584 void drbd_suspend_io(struct drbd_conf *mdev)
 585 {
 586         set_bit(SUSPEND_IO, &mdev->flags);
 587         if (is_susp(mdev->state))
 588                 return;
 589         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
 590 }
 591
 592 void drbd_resume_io(struct drbd_conf *mdev)
 593 {
 594         clear_bit(SUSPEND_IO, &mdev->flags);
 595         wake_up(&mdev->misc_wait);
 596 }
 597
 598 /**
 599  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
 600  * @mdev:       DRBD device.
 601  *
 602  * Returns 0 on success, negative return values indicate errors.
 603  * You should call drbd_md_sync() after calling this function.
 604  */
 605 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
 606 {
 607         sector_t prev_first_sect, prev_size; /* previous meta location */
 608         sector_t la_size;
 609         sector_t size;
 610         char ppb[10];
 611
 612         int md_moved, la_size_changed;
 613         enum determine_dev_size rv = unchanged;
 614
 615         /* race:
 616          * application request passes inc_ap_bio,
 617          * but then cannot get an AL-reference.
 618          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
 619          *
 620          * to avoid that:
 621          * Suspend IO right here.
 622          * still lock the act_log to not trigger ASSERTs there.
 623          */
 624         drbd_suspend_io(mdev);
 625
 626         /* no wait necessary anymore, actually we could assert that */
 627         wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
 628
 629         prev_first_sect = drbd_md_first_sector(mdev->ldev);
 630         prev_size = mdev->ldev->md.md_size_sect;
 631         la_size = mdev->ldev->md.la_size_sect;
 632
 633         /* TODO: should only be some assert here, not (re)init... */
 634         drbd_md_set_sector_offsets(mdev, mdev->ldev);
 635
 636         size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
 637
 638         if (drbd_get_capacity(mdev->this_bdev) != size ||
 639             drbd_bm_capacity(mdev) != size) {
 640                 int err;
 641                 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
 642                 if (unlikely(err)) {
 643                         /* currently there is only one error: ENOMEM! */
 644                         size = drbd_bm_capacity(mdev)>>1;
 645                         if (size == 0) {
 646                                 dev_err(DEV, "OUT OF MEMORY! "
 647                                     "Could not allocate bitmap!\n");
 648                         } else {
 649                                 dev_err(DEV, "BM resizing failed. "
 650                                     "Leaving size unchanged at size = %lu KB\n",
 651                                     (unsigned long)size);
 652                         }
 653                         rv = dev_size_error;
 654                 }
 655                 /* racy, see comments above. */
 656                 drbd_set_my_capacity(mdev, size);
 657                 mdev->ldev->md.la_size_sect = size;
 658                 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
 659                      (unsigned long long)size>>1);
 660         }
 661         if (rv == dev_size_error)
 662                 goto out;
 663
 664         la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
 665
 666         md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
 667                 || prev_size       != mdev->ldev->md.md_size_sect;
 668
 669         if (la_size_changed || md_moved) {
 670                 int err;
 671
 672                 drbd_al_shrink(mdev); /* All extents inactive. */
 673                 dev_info(DEV, "Writing the whole bitmap, %s\n",
 674                          la_size_changed && md_moved ? "size changed and md moved" :
 675                          la_size_changed ? "size changed" : "md moved");
 676                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
 677                 err = drbd_bitmap_io(mdev, &drbd_bm_write,
 678                                 "size changed", BM_LOCKED_MASK);
 679                 if (err) {
 680                         rv = dev_size_error;
 681                         goto out;
 682                 }
 683                 drbd_md_mark_dirty(mdev);
 684         }
 685
 686         if (size > la_size)
 687                 rv = grew;
 688         if (size < la_size)
 689                 rv = shrunk;
 690 out:
 691         lc_unlock(mdev->act_log);
 692         wake_up(&mdev->al_wait);
 693         drbd_resume_io(mdev);
 694
 695         return rv;
 696 }
 697
 698 sector_t
 699 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
 700 {
 701         sector_t p_size = mdev->p_size;   /* partner's disk size. */
 702         sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
 703         sector_t m_size; /* my size */
 704         sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
 705         sector_t size = 0;
 706
 707         m_size = drbd_get_max_capacity(bdev);
 708
 709         if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
 710                 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
 711                 p_size = m_size;
 712         }
 713
 714         if (p_size && m_size) {
 715                 size = min_t(sector_t, p_size, m_size);
 716         } else {
 717                 if (la_size) {
 718                         size = la_size;
 719                         if (m_size && m_size < size)
 720                                 size = m_size;
 721                         if (p_size && p_size < size)
 722                                 size = p_size;
 723                 } else {
 724                         if (m_size)
 725                                 size = m_size;
 726                         if (p_size)
 727                                 size = p_size;
 728                 }
 729         }
 730
 731         if (size == 0)
 732                 dev_err(DEV, "Both nodes diskless!\n");
 733
 734         if (u_size) {
 735                 if (u_size > size)
 736                         dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
 737                             (unsigned long)u_size>>1, (unsigned long)size>>1);
 738                 else
 739                         size = u_size;
 740         }
 741
 742         return size;
 743 }
 744
 745 /**
 746  * drbd_check_al_size() - Ensures that the AL is of the right size
 747  * @mdev:       DRBD device.
 748  *
 749  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
 750  * failed, and 0 on success. You should call drbd_md_sync() after you called
 751  * this function.
 752  */
 753 static int drbd_check_al_size(struct drbd_conf *mdev)
 754 {
 755         struct lru_cache *n, *t;
 756         struct lc_element *e;
 757         unsigned int in_use;
 758         int i;
 759
 760         ERR_IF(mdev->sync_conf.al_extents < 7)
 761                 mdev->sync_conf.al_extents = 127;
 762
 763         if (mdev->act_log &&
 764             mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
 765                 return 0;
 766
 767         in_use = 0;
 768         t = mdev->act_log;
 769         n = lc_create("act_log", drbd_al_ext_cache,
 770                 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
 771
 772         if (n == NULL) {
 773                 dev_err(DEV, "Cannot allocate act_log lru!\n");
 774                 return -ENOMEM;
 775         }
 776         spin_lock_irq(&mdev->al_lock);
 777         if (t) {
 778                 for (i = 0; i < t->nr_elements; i++) {
 779                         e = lc_element_by_index(t, i);
 780                         if (e->refcnt)
 781                                 dev_err(DEV, "refcnt(%d)==%d\n",
 782                                     e->lc_number, e->refcnt);
 783                         in_use += e->refcnt;
 784                 }
 785         }
 786         if (!in_use)
 787                 mdev->act_log = n;
 788         spin_unlock_irq(&mdev->al_lock);
 789         if (in_use) {
 790                 dev_err(DEV, "Activity log still in use!\n");
 791                 lc_destroy(n);
 792                 return -EBUSY;
 793         } else {
 794                 if (t)
 795                         lc_destroy(t);
 796         }
 797         drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
 798         return 0;
 799 }
 800
 801 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
 802 {
 803         struct request_queue * const q = mdev->rq_queue;
 804         int max_hw_sectors = max_bio_size >> 9;
 805         int max_segments = 0;
 806
 807         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 808                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 809
 810                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
 811                 max_segments = mdev->ldev->dc.max_bio_bvecs;
 812                 put_ldev(mdev);
 813         }
 814
 815         blk_queue_logical_block_size(q, 512);
 816         blk_queue_max_hw_sectors(q, max_hw_sectors);
 817         /* This is the workaround for "bio would need to, but cannot, be split" */
 818         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
 819         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
 820
 821         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 822                 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
 823
 824                 blk_queue_stack_limits(q, b);
 825
 826                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
 827                         dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
 828                                  q->backing_dev_info.ra_pages,
 829                                  b->backing_dev_info.ra_pages);
 830                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
 831                 }
 832                 put_ldev(mdev);
 833         }
 834 }
 835
 836 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
 837 {
 838         int now, new, local, peer;
 839
 840         now = queue_max_hw_sectors(mdev->rq_queue) << 9;
 841         local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
 842         peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
 843
 844         if (get_ldev_if_state(mdev, D_ATTACHING)) {
 845                 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
 846                 mdev->local_max_bio_size = local;
 847                 put_ldev(mdev);
 848         }
 849
 850         /* We may ignore peer limits if the peer is modern enough.
 851            Because new from 8.3.8 onwards the peer can use multiple
 852            BIOs for a single peer_request */
 853         if (mdev->state.conn >= C_CONNECTED) {
 854                 if (mdev->agreed_pro_version < 94) {
 855                         peer = min_t(int, mdev->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
 856                         /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
 857                 } else if (mdev->agreed_pro_version == 94)
 858                         peer = DRBD_MAX_SIZE_H80_PACKET;
 859                 else /* drbd 8.3.8 onwards */
 860                         peer = DRBD_MAX_BIO_SIZE;
 861         }
 862
 863         new = min_t(int, local, peer);
 864
 865         if (mdev->state.role == R_PRIMARY && new < now)
 866                 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
 867
 868         if (new != now)
 869                 dev_info(DEV, "max BIO size = %u\n", new);
 870
 871         drbd_setup_queue_param(mdev, new);
 872 }
 873
 874 /* serialize deconfig (worker exiting, doing cleanup)
 875  * and reconfig (drbdsetup disk, drbdsetup net)
 876  *
 877  * Wait for a potentially exiting worker, then restart it,
 878  * or start a new one.  Flush any pending work, there may still be an
 879  * after_state_change queued.
 880  */
 881 static void drbd_reconfig_start(struct drbd_conf *mdev)
 882 {
 883         wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
 884         wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
 885         drbd_thread_start(&mdev->worker);
 886         drbd_flush_workqueue(mdev);
 887 }
 888
 889 /* if still unconfigured, stops worker again.
 890  * if configured now, clears CONFIG_PENDING.
 891  * wakes potential waiters */
 892 static void drbd_reconfig_done(struct drbd_conf *mdev)
 893 {
 894         spin_lock_irq(&mdev->req_lock);
 895         if (mdev->state.disk == D_DISKLESS &&
 896             mdev->state.conn == C_STANDALONE &&
 897             mdev->state.role == R_SECONDARY) {
 898                 set_bit(DEVICE_DYING, &mdev->flags);
 899                 drbd_thread_stop_nowait(&mdev->worker);
 900         } else
 901                 clear_bit(CONFIG_PENDING, &mdev->flags);
 902         spin_unlock_irq(&mdev->req_lock);
 903         wake_up(&mdev->state_wait);
 904 }
 905
 906 /* Make sure IO is suspended before calling this function(). */
 907 static void drbd_suspend_al(struct drbd_conf *mdev)
 908 {
 909         int s = 0;
 910
 911         if (lc_try_lock(mdev->act_log)) {
 912                 drbd_al_shrink(mdev);
 913                 lc_unlock(mdev->act_log);
 914         } else {
 915                 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
 916                 return;
 917         }
 918
 919         spin_lock_irq(&mdev->req_lock);
 920         if (mdev->state.conn < C_CONNECTED)
 921                 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
 922
 923         spin_unlock_irq(&mdev->req_lock);
 924
 925         if (s)
 926                 dev_info(DEV, "Suspended AL updates\n");
 927 }
 928
 929 /* does always return 0;
 930  * interesting return code is in reply->ret_code */
 931 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
 932                              struct drbd_nl_cfg_reply *reply)
 933 {
 934         enum drbd_ret_code retcode;
 935         enum determine_dev_size dd;
 936         sector_t max_possible_sectors;
 937         sector_t min_md_device_sectors;
 938         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
 939         struct block_device *bdev;
 940         struct lru_cache *resync_lru = NULL;
 941         union drbd_state ns, os;
 942         enum drbd_state_rv rv;
 943         int cp_discovered = 0;
 944         int logical_block_size;
 945
 946         drbd_reconfig_start(mdev);
 947
 948         /* if you want to reconfigure, please tear down first */
 949         if (mdev->state.disk > D_DISKLESS) {
 950                 retcode = ERR_DISK_CONFIGURED;
 951                 goto fail;
 952         }
 953         /* It may just now have detached because of IO error.  Make sure
 954          * drbd_ldev_destroy is done already, we may end up here very fast,
 955          * e.g. if someone calls attach from the on-io-error handler,
 956          * to realize a "hot spare" feature (not that I'd recommend that) */
 957         wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
 958
 959         /* make sure there is no leftover from previous force-detach attempts */
 960         clear_bit(FORCE_DETACH, &mdev->flags);
 961
 962         /* and no leftover from previously aborted resync or verify, either */
 963         mdev->rs_total = 0;
 964         mdev->rs_failed = 0;
 965         atomic_set(&mdev->rs_pending_cnt, 0);
 966
 967         /* allocation not in the IO path, cqueue thread context */
 968         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
 969         if (!nbc) {
 970                 retcode = ERR_NOMEM;
 971                 goto fail;
 972         }
 973
 974         nbc->dc.disk_size     = DRBD_DISK_SIZE_SECT_DEF;
 975         nbc->dc.on_io_error   = DRBD_ON_IO_ERROR_DEF;
 976         nbc->dc.fencing       = DRBD_FENCING_DEF;
 977         nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
 978
 979         if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
 980                 retcode = ERR_MANDATORY_TAG;
 981                 goto fail;
 982         }
 983
 984         if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
 985                 retcode = ERR_MD_IDX_INVALID;
 986                 goto fail;
 987         }
 988
 989         if (get_net_conf(mdev)) {
 990                 int prot = mdev->net_conf->wire_protocol;
 991                 put_net_conf(mdev);
 992                 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
 993                         retcode = ERR_STONITH_AND_PROT_A;
 994                         goto fail;
 995                 }
 996         }
 997
 998         bdev = blkdev_get_by_path(nbc->dc.backing_dev,
 999                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1000         if (IS_ERR(bdev)) {
1001                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
1002                         PTR_ERR(bdev));
1003                 retcode = ERR_OPEN_DISK;
1004                 goto fail;
1005         }
1006         nbc->backing_bdev = bdev;
1007
1008         /*
1009          * meta_dev_idx >= 0: external fixed size, possibly multiple
1010          * drbd sharing one meta device.  TODO in that case, paranoia
1011          * check that [md_bdev, meta_dev_idx] is not yet used by some
1012          * other drbd minor!  (if you use drbd.conf + drbdadm, that
1013          * should check it for you already; but if you don't, or
1014          * someone fooled it, we need to double check here)
1015          */
1016         bdev = blkdev_get_by_path(nbc->dc.meta_dev,
1017                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1018                                   (nbc->dc.meta_dev_idx < 0) ?
1019                                   (void *)mdev : (void *)drbd_m_holder);
1020         if (IS_ERR(bdev)) {
1021                 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
1022                         PTR_ERR(bdev));
1023                 retcode = ERR_OPEN_MD_DISK;
1024                 goto fail;
1025         }
1026         nbc->md_bdev = bdev;
1027
1028         if ((nbc->backing_bdev == nbc->md_bdev) !=
1029             (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1030              nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1031                 retcode = ERR_MD_IDX_INVALID;
1032                 goto fail;
1033         }
1034
1035         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1036                         61, sizeof(struct bm_extent),
1037                         offsetof(struct bm_extent, lce));
1038         if (!resync_lru) {
1039                 retcode = ERR_NOMEM;
1040                 goto fail;
1041         }
1042
1043         /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1044         drbd_md_set_sector_offsets(mdev, nbc);
1045
1046         if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
1047                 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1048                         (unsigned long long) drbd_get_max_capacity(nbc),
1049                         (unsigned long long) nbc->dc.disk_size);
1050                 retcode = ERR_DISK_TOO_SMALL;
1051                 goto fail;
1052         }
1053
1054         if (nbc->dc.meta_dev_idx < 0) {
1055                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1056                 /* at least one MB, otherwise it does not make sense */
1057                 min_md_device_sectors = (2<<10);
1058         } else {
1059                 max_possible_sectors = DRBD_MAX_SECTORS;
1060                 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
1061         }
1062
1063         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1064                 retcode = ERR_MD_DISK_TOO_SMALL;
1065                 dev_warn(DEV, "refusing attach: md-device too small, "
1066                      "at least %llu sectors needed for this meta-disk type\n",
1067                      (unsigned long long) min_md_device_sectors);
1068                 goto fail;
1069         }
1070
1071         /* Make sure the new disk is big enough
1072          * (we may currently be R_PRIMARY with no local disk...) */
1073         if (drbd_get_max_capacity(nbc) <
1074             drbd_get_capacity(mdev->this_bdev)) {
1075                 retcode = ERR_DISK_TOO_SMALL;
1076                 goto fail;
1077         }
1078
1079         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1080
1081         if (nbc->known_size > max_possible_sectors) {
1082                 dev_warn(DEV, "==> truncating very big lower level device "
1083                         "to currently maximum possible %llu sectors <==\n",
1084                         (unsigned long long) max_possible_sectors);
1085                 if (nbc->dc.meta_dev_idx >= 0)
1086                         dev_warn(DEV, "==>> using internal or flexible "
1087                                       "meta data may help <<==\n");
1088         }
1089
1090         drbd_suspend_io(mdev);
1091         /* also wait for the last barrier ack. */
1092         wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state));
1093         /* and for any other previously queued work */
1094         drbd_flush_workqueue(mdev);
1095
1096         rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1097         retcode = rv;  /* FIXME: Type mismatch. */
1098         drbd_resume_io(mdev);
1099         if (rv < SS_SUCCESS)
1100                 goto fail;
1101
1102         if (!get_ldev_if_state(mdev, D_ATTACHING))
1103                 goto force_diskless;
1104
1105         drbd_md_set_sector_offsets(mdev, nbc);
1106
1107         /* allocate a second IO page if logical_block_size != 512 */
1108         logical_block_size = bdev_logical_block_size(nbc->md_bdev);
1109         if (logical_block_size == 0)
1110                 logical_block_size = MD_SECTOR_SIZE;
1111
1112         if (logical_block_size != MD_SECTOR_SIZE) {
1113                 if (!mdev->md_io_tmpp) {
1114                         struct page *page = alloc_page(GFP_NOIO);
1115                         if (!page)
1116                                 goto force_diskless_dec;
1117
1118                         dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
1119                              logical_block_size, MD_SECTOR_SIZE);
1120                         dev_warn(DEV, "Workaround engaged (has performance impact).\n");
1121
1122                         mdev->md_io_tmpp = page;
1123                 }
1124         }
1125
1126         if (!mdev->bitmap) {
1127                 if (drbd_bm_init(mdev)) {
1128                         retcode = ERR_NOMEM;
1129                         goto force_diskless_dec;
1130                 }
1131         }
1132
1133         retcode = drbd_md_read(mdev, nbc);
1134         if (retcode != NO_ERROR)
1135                 goto force_diskless_dec;
1136
1137         if (mdev->state.conn < C_CONNECTED &&
1138             mdev->state.role == R_PRIMARY &&
1139             (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1140                 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1141                     (unsigned long long)mdev->ed_uuid);
1142                 retcode = ERR_DATA_NOT_CURRENT;
1143                 goto force_diskless_dec;
1144         }
1145
1146         /* Since we are diskless, fix the activity log first... */
1147         if (drbd_check_al_size(mdev)) {
1148                 retcode = ERR_NOMEM;
1149                 goto force_diskless_dec;
1150         }
1151
1152         /* Prevent shrinking of consistent devices ! */
1153         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1154             drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1155                 dev_warn(DEV, "refusing to truncate a consistent device\n");
1156                 retcode = ERR_DISK_TOO_SMALL;
1157                 goto force_diskless_dec;
1158         }
1159
1160         if (!drbd_al_read_log(mdev, nbc)) {
1161                 retcode = ERR_IO_MD_DISK;
1162                 goto force_diskless_dec;
1163         }
1164
1165         /* Reset the "barriers don't work" bits here, then force meta data to
1166          * be written, to ensure we determine if barriers are supported. */
1167         if (nbc->dc.no_md_flush)
1168                 set_bit(MD_NO_FUA, &mdev->flags);
1169         else
1170                 clear_bit(MD_NO_FUA, &mdev->flags);
1171
1172         /* Point of no return reached.
1173          * Devices and memory are no longer released by error cleanup below.
1174          * now mdev takes over responsibility, and the state engine should
1175          * clean it up somewhere.  */
1176         D_ASSERT(mdev->ldev == NULL);
1177         mdev->ldev = nbc;
1178         mdev->resync = resync_lru;
1179         nbc = NULL;
1180         resync_lru = NULL;
1181
1182         mdev->write_ordering = WO_bdev_flush;
1183         drbd_bump_write_ordering(mdev, WO_bdev_flush);
1184
1185         if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1186                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1187         else
1188                 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1189
1190         if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1191             !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) {
1192                 set_bit(CRASHED_PRIMARY, &mdev->flags);
1193                 cp_discovered = 1;
1194         }
1195
1196         mdev->send_cnt = 0;
1197         mdev->recv_cnt = 0;
1198         mdev->read_cnt = 0;
1199         mdev->writ_cnt = 0;
1200
1201         drbd_reconsider_max_bio_size(mdev);
1202
1203         /* If I am currently not R_PRIMARY,
1204          * but meta data primary indicator is set,
1205          * I just now recover from a hard crash,
1206          * and have been R_PRIMARY before that crash.
1207          *
1208          * Now, if I had no connection before that crash
1209          * (have been degraded R_PRIMARY), chances are that
1210          * I won't find my peer now either.
1211          *
1212          * In that case, and _only_ in that case,
1213          * we use the degr-wfc-timeout instead of the default,
1214          * so we can automatically recover from a crash of a
1215          * degraded but active "cluster" after a certain timeout.
1216          */
1217         clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1218         if (mdev->state.role != R_PRIMARY &&
1219              drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1220             !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1221                 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1222
1223         dd = drbd_determine_dev_size(mdev, 0);
1224         if (dd == dev_size_error) {
1225                 retcode = ERR_NOMEM_BITMAP;
1226                 goto force_diskless_dec;
1227         } else if (dd == grew)
1228                 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1229
1230         if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1231                 dev_info(DEV, "Assuming that all blocks are out of sync "
1232                      "(aka FullSync)\n");
1233                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1234                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1235                         retcode = ERR_IO_MD_DISK;
1236                         goto force_diskless_dec;
1237                 }
1238         } else {
1239                 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1240                         "read from attaching", BM_LOCKED_MASK) < 0) {
1241                         retcode = ERR_IO_MD_DISK;
1242                         goto force_diskless_dec;
1243                 }
1244         }
1245
1246         if (cp_discovered) {
1247                 drbd_al_apply_to_bm(mdev);
1248                 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1249                         "crashed primary apply AL", BM_LOCKED_MASK)) {
1250                         retcode = ERR_IO_MD_DISK;
1251                         goto force_diskless_dec;
1252                 }
1253         }
1254
1255         if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1256                 drbd_suspend_al(mdev); /* IO is still suspended here... */
1257
1258         spin_lock_irq(&mdev->req_lock);
1259         os = mdev->state;
1260         ns.i = os.i;
1261         /* If MDF_CONSISTENT is not set go into inconsistent state,
1262            otherwise investigate MDF_WasUpToDate...
1263            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1264            otherwise into D_CONSISTENT state.
1265         */
1266         if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1267                 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1268                         ns.disk = D_CONSISTENT;
1269                 else
1270                         ns.disk = D_OUTDATED;
1271         } else {
1272                 ns.disk = D_INCONSISTENT;
1273         }
1274
1275         if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1276                 ns.pdsk = D_OUTDATED;
1277
1278         if ( ns.disk == D_CONSISTENT &&
1279             (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1280                 ns.disk = D_UP_TO_DATE;
1281
1282         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1283            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1284            this point, because drbd_request_state() modifies these
1285            flags. */
1286
1287         /* In case we are C_CONNECTED postpone any decision on the new disk
1288            state after the negotiation phase. */
1289         if (mdev->state.conn == C_CONNECTED) {
1290                 mdev->new_state_tmp.i = ns.i;
1291                 ns.i = os.i;
1292                 ns.disk = D_NEGOTIATING;
1293
1294                 /* We expect to receive up-to-date UUIDs soon.
1295                    To avoid a race in receive_state, free p_uuid while
1296                    holding req_lock. I.e. atomic with the state change */
1297                 kfree(mdev->p_uuid);
1298                 mdev->p_uuid = NULL;
1299         }
1300
1301         rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1302         ns = mdev->state;
1303         spin_unlock_irq(&mdev->req_lock);
1304
1305         if (rv < SS_SUCCESS)
1306                 goto force_diskless_dec;
1307
1308         if (mdev->state.role == R_PRIMARY)
1309                 mdev->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1310         else
1311                 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1312
1313         drbd_md_mark_dirty(mdev);
1314         drbd_md_sync(mdev);
1315
1316         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1317         put_ldev(mdev);
1318         reply->ret_code = retcode;
1319         drbd_reconfig_done(mdev);
1320         return 0;
1321
1322  force_diskless_dec:
1323         put_ldev(mdev);
1324  force_diskless:
1325         drbd_force_state(mdev, NS(disk, D_FAILED));
1326         drbd_md_sync(mdev);
1327  fail:
1328         if (nbc) {
1329                 if (nbc->backing_bdev)
1330                         blkdev_put(nbc->backing_bdev,
1331                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1332                 if (nbc->md_bdev)
1333                         blkdev_put(nbc->md_bdev,
1334                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1335                 kfree(nbc);
1336         }
1337         lc_destroy(resync_lru);
1338
1339         reply->ret_code = retcode;
1340         drbd_reconfig_done(mdev);
1341         return 0;
1342 }
1343
1344 /* Detaching the disk is a process in multiple stages.  First we need to lock
1345  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1346  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1347  * internal references as well.
1348  * Only then we have finally detached. */
1349 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1350                           struct drbd_nl_cfg_reply *reply)
1351 {
1352         enum drbd_ret_code retcode;
1353         int ret;
1354         struct detach dt = {};
1355
1356         if (!detach_from_tags(mdev, nlp->tag_list, &dt)) {
1357                 reply->ret_code = ERR_MANDATORY_TAG;
1358                 goto out;
1359         }
1360
1361         if (dt.detach_force) {
1362                 set_bit(FORCE_DETACH, &mdev->flags);
1363                 drbd_force_state(mdev, NS(disk, D_FAILED));
1364                 reply->ret_code = SS_SUCCESS;
1365                 goto out;
1366         }
1367
1368         drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1369         drbd_md_get_buffer(mdev); /* make sure there is no in-flight meta-data IO */
1370         retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1371         drbd_md_put_buffer(mdev);
1372         /* D_FAILED will transition to DISKLESS. */
1373         ret = wait_event_interruptible(mdev->misc_wait,
1374                         mdev->state.disk != D_FAILED);
1375         drbd_resume_io(mdev);
1376
1377         if ((int)retcode == (int)SS_IS_DISKLESS)
1378                 retcode = SS_NOTHING_TO_DO;
1379         if (ret)
1380                 retcode = ERR_INTR;
1381         reply->ret_code = retcode;
1382 out:
1383         return 0;
1384 }
1385
1386 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1387                             struct drbd_nl_cfg_reply *reply)
1388 {
1389         int i, ns;
1390         enum drbd_ret_code retcode;
1391         struct net_conf *new_conf = NULL;
1392         struct crypto_hash *tfm = NULL;
1393         struct crypto_hash *integrity_w_tfm = NULL;
1394         struct crypto_hash *integrity_r_tfm = NULL;
1395         struct hlist_head *new_tl_hash = NULL;
1396         struct hlist_head *new_ee_hash = NULL;
1397         struct drbd_conf *odev;
1398         char hmac_name[CRYPTO_MAX_ALG_NAME];
1399         void *int_dig_out = NULL;
1400         void *int_dig_in = NULL;
1401         void *int_dig_vv = NULL;
1402         struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1403
1404         drbd_reconfig_start(mdev);
1405
1406         if (mdev->state.conn > C_STANDALONE) {
1407                 retcode = ERR_NET_CONFIGURED;
1408                 goto fail;
1409         }
1410
1411         /* allocation not in the IO path, cqueue thread context */
1412         new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1413         if (!new_conf) {
1414                 retcode = ERR_NOMEM;
1415                 goto fail;
1416         }
1417
1418         new_conf->timeout          = DRBD_TIMEOUT_DEF;
1419         new_conf->try_connect_int  = DRBD_CONNECT_INT_DEF;
1420         new_conf->ping_int         = DRBD_PING_INT_DEF;
1421         new_conf->max_epoch_size   = DRBD_MAX_EPOCH_SIZE_DEF;
1422         new_conf->max_buffers      = DRBD_MAX_BUFFERS_DEF;
1423         new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1424         new_conf->sndbuf_size      = DRBD_SNDBUF_SIZE_DEF;
1425         new_conf->rcvbuf_size      = DRBD_RCVBUF_SIZE_DEF;
1426         new_conf->ko_count         = DRBD_KO_COUNT_DEF;
1427         new_conf->after_sb_0p      = DRBD_AFTER_SB_0P_DEF;
1428         new_conf->after_sb_1p      = DRBD_AFTER_SB_1P_DEF;
1429         new_conf->after_sb_2p      = DRBD_AFTER_SB_2P_DEF;
1430         new_conf->want_lose        = 0;
1431         new_conf->two_primaries    = 0;
1432         new_conf->wire_protocol    = DRBD_PROT_C;
1433         new_conf->ping_timeo       = DRBD_PING_TIMEO_DEF;
1434         new_conf->rr_conflict      = DRBD_RR_CONFLICT_DEF;
1435         new_conf->on_congestion    = DRBD_ON_CONGESTION_DEF;
1436         new_conf->cong_extents     = DRBD_CONG_EXTENTS_DEF;
1437
1438         if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1439                 retcode = ERR_MANDATORY_TAG;
1440                 goto fail;
1441         }
1442
1443         if (new_conf->two_primaries
1444             && (new_conf->wire_protocol != DRBD_PROT_C)) {
1445                 retcode = ERR_NOT_PROTO_C;
1446                 goto fail;
1447         }
1448
1449         if (get_ldev(mdev)) {
1450                 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1451                 put_ldev(mdev);
1452                 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1453                         retcode = ERR_STONITH_AND_PROT_A;
1454                         goto fail;
1455                 }
1456         }
1457
1458         if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1459                 retcode = ERR_CONG_NOT_PROTO_A;
1460                 goto fail;
1461         }
1462
1463         if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1464                 retcode = ERR_DISCARD;
1465                 goto fail;
1466         }
1467
1468         retcode = NO_ERROR;
1469
1470         new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1471         new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1472         for (i = 0; i < minor_count; i++) {
1473                 odev = minor_to_mdev(i);
1474                 if (!odev || odev == mdev)
1475                         continue;
1476                 if (get_net_conf(odev)) {
1477                         taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1478                         if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1479                             !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1480                                 retcode = ERR_LOCAL_ADDR;
1481
1482                         taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1483                         if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1484                             !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1485                                 retcode = ERR_PEER_ADDR;
1486
1487                         put_net_conf(odev);
1488                         if (retcode != NO_ERROR)
1489                                 goto fail;
1490                 }
1491         }
1492
1493         if (new_conf->cram_hmac_alg[0] != 0) {
1494                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1495                         new_conf->cram_hmac_alg);
1496                 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1497                 if (IS_ERR(tfm)) {
1498                         tfm = NULL;
1499                         retcode = ERR_AUTH_ALG;
1500                         goto fail;
1501                 }
1502
1503                 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1504                         retcode = ERR_AUTH_ALG_ND;
1505                         goto fail;
1506                 }
1507         }
1508
1509         if (new_conf->integrity_alg[0]) {
1510                 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1511                 if (IS_ERR(integrity_w_tfm)) {
1512                         integrity_w_tfm = NULL;
1513                         retcode=ERR_INTEGRITY_ALG;
1514                         goto fail;
1515                 }
1516
1517                 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1518                         retcode=ERR_INTEGRITY_ALG_ND;
1519                         goto fail;
1520                 }
1521
1522                 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1523                 if (IS_ERR(integrity_r_tfm)) {
1524                         integrity_r_tfm = NULL;
1525                         retcode=ERR_INTEGRITY_ALG;
1526                         goto fail;
1527                 }
1528         }
1529
1530         ns = new_conf->max_epoch_size/8;
1531         if (mdev->tl_hash_s != ns) {
1532                 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1533                 if (!new_tl_hash) {
1534                         retcode = ERR_NOMEM;
1535                         goto fail;
1536                 }
1537         }
1538
1539         ns = new_conf->max_buffers/8;
1540         if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1541                 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1542                 if (!new_ee_hash) {
1543                         retcode = ERR_NOMEM;
1544                         goto fail;
1545                 }
1546         }
1547
1548         ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1549
1550         if (integrity_w_tfm) {
1551                 i = crypto_hash_digestsize(integrity_w_tfm);
1552                 int_dig_out = kmalloc(i, GFP_KERNEL);
1553                 if (!int_dig_out) {
1554                         retcode = ERR_NOMEM;
1555                         goto fail;
1556                 }
1557                 int_dig_in = kmalloc(i, GFP_KERNEL);
1558                 if (!int_dig_in) {
1559                         retcode = ERR_NOMEM;
1560                         goto fail;
1561                 }
1562                 int_dig_vv = kmalloc(i, GFP_KERNEL);
1563                 if (!int_dig_vv) {
1564                         retcode = ERR_NOMEM;
1565                         goto fail;
1566                 }
1567         }
1568
1569         if (!mdev->bitmap) {
1570                 if(drbd_bm_init(mdev)) {
1571                         retcode = ERR_NOMEM;
1572                         goto fail;
1573                 }
1574         }
1575
1576         drbd_flush_workqueue(mdev);
1577         spin_lock_irq(&mdev->req_lock);
1578         if (mdev->net_conf != NULL) {
1579                 retcode = ERR_NET_CONFIGURED;
1580                 spin_unlock_irq(&mdev->req_lock);
1581                 goto fail;
1582         }
1583         mdev->net_conf = new_conf;
1584
1585         mdev->send_cnt = 0;
1586         mdev->recv_cnt = 0;
1587
1588         if (new_tl_hash) {
1589                 kfree(mdev->tl_hash);
1590                 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1591                 mdev->tl_hash = new_tl_hash;
1592         }
1593
1594         if (new_ee_hash) {
1595                 kfree(mdev->ee_hash);
1596                 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1597                 mdev->ee_hash = new_ee_hash;
1598         }
1599
1600         crypto_free_hash(mdev->cram_hmac_tfm);
1601         mdev->cram_hmac_tfm = tfm;
1602
1603         crypto_free_hash(mdev->integrity_w_tfm);
1604         mdev->integrity_w_tfm = integrity_w_tfm;
1605
1606         crypto_free_hash(mdev->integrity_r_tfm);
1607         mdev->integrity_r_tfm = integrity_r_tfm;
1608
1609         kfree(mdev->int_dig_out);
1610         kfree(mdev->int_dig_in);
1611         kfree(mdev->int_dig_vv);
1612         mdev->int_dig_out=int_dig_out;
1613         mdev->int_dig_in=int_dig_in;
1614         mdev->int_dig_vv=int_dig_vv;
1615         retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL);
1616         spin_unlock_irq(&mdev->req_lock);
1617
1618         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1619         reply->ret_code = retcode;
1620         drbd_reconfig_done(mdev);
1621         return 0;
1622
1623 fail:
1624         kfree(int_dig_out);
1625         kfree(int_dig_in);
1626         kfree(int_dig_vv);
1627         crypto_free_hash(tfm);
1628         crypto_free_hash(integrity_w_tfm);
1629         crypto_free_hash(integrity_r_tfm);
1630         kfree(new_tl_hash);
1631         kfree(new_ee_hash);
1632         kfree(new_conf);
1633
1634         reply->ret_code = retcode;
1635         drbd_reconfig_done(mdev);
1636         return 0;
1637 }
1638
1639 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1640                               struct drbd_nl_cfg_reply *reply)
1641 {
1642         int retcode;
1643         struct disconnect dc;
1644
1645         memset(&dc, 0, sizeof(struct disconnect));
1646         if (!disconnect_from_tags(mdev, nlp->tag_list, &dc)) {
1647                 retcode = ERR_MANDATORY_TAG;
1648                 goto fail;
1649         }
1650
1651         if (dc.force) {
1652                 spin_lock_irq(&mdev->req_lock);
1653                 if (mdev->state.conn >= C_WF_CONNECTION)
1654                         _drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), CS_HARD, NULL);
1655                 spin_unlock_irq(&mdev->req_lock);
1656                 goto done;
1657         }
1658
1659         retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1660
1661         if (retcode == SS_NOTHING_TO_DO)
1662                 goto done;
1663         else if (retcode == SS_ALREADY_STANDALONE)
1664                 goto done;
1665         else if (retcode == SS_PRIMARY_NOP) {
1666                 /* Our statche checking code wants to see the peer outdated. */
1667                 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1668                                                       pdsk, D_OUTDATED));
1669         } else if (retcode == SS_CW_FAILED_BY_PEER) {
1670                 /* The peer probably wants to see us outdated. */
1671                 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1672                                                         disk, D_OUTDATED),
1673                                               CS_ORDERED);
1674                 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1675                         drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1676                         retcode = SS_SUCCESS;
1677                 }
1678         }
1679
1680         if (retcode < SS_SUCCESS)
1681                 goto fail;
1682
1683         if (wait_event_interruptible(mdev->state_wait,
1684                                      mdev->state.conn != C_DISCONNECTING)) {
1685                 /* Do not test for mdev->state.conn == C_STANDALONE, since
1686                    someone else might connect us in the mean time! */
1687                 retcode = ERR_INTR;
1688                 goto fail;
1689         }
1690
1691  done:
1692         retcode = NO_ERROR;
1693  fail:
1694         drbd_md_sync(mdev);
1695         reply->ret_code = retcode;
1696         return 0;
1697 }
1698
1699 void resync_after_online_grow(struct drbd_conf *mdev)
1700 {
1701         int iass; /* I am sync source */
1702
1703         dev_info(DEV, "Resync of new storage after online grow\n");
1704         if (mdev->state.role != mdev->state.peer)
1705                 iass = (mdev->state.role == R_PRIMARY);
1706         else
1707                 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1708
1709         if (iass)
1710                 drbd_start_resync(mdev, C_SYNC_SOURCE);
1711         else
1712                 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1713 }
1714
1715 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1716                           struct drbd_nl_cfg_reply *reply)
1717 {
1718         struct resize rs;
1719         int retcode = NO_ERROR;
1720         enum determine_dev_size dd;
1721         enum dds_flags ddsf;
1722
1723         memset(&rs, 0, sizeof(struct resize));
1724         if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1725                 retcode = ERR_MANDATORY_TAG;
1726                 goto fail;
1727         }
1728
1729         if (mdev->state.conn > C_CONNECTED) {
1730                 retcode = ERR_RESIZE_RESYNC;
1731                 goto fail;
1732         }
1733
1734         if (mdev->state.role == R_SECONDARY &&
1735             mdev->state.peer == R_SECONDARY) {
1736                 retcode = ERR_NO_PRIMARY;
1737                 goto fail;
1738         }
1739
1740         if (!get_ldev(mdev)) {
1741                 retcode = ERR_NO_DISK;
1742                 goto fail;
1743         }
1744
1745         if (rs.no_resync && mdev->agreed_pro_version < 93) {
1746                 retcode = ERR_NEED_APV_93;
1747                 goto fail_ldev;
1748         }
1749
1750         if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1751                 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1752
1753         mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1754         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1755         dd = drbd_determine_dev_size(mdev, ddsf);
1756         drbd_md_sync(mdev);
1757         put_ldev(mdev);
1758         if (dd == dev_size_error) {
1759                 retcode = ERR_NOMEM_BITMAP;
1760                 goto fail;
1761         }
1762
1763         if (mdev->state.conn == C_CONNECTED) {
1764                 if (dd == grew)
1765                         set_bit(RESIZE_PENDING, &mdev->flags);
1766
1767                 drbd_send_uuids(mdev);
1768                 drbd_send_sizes(mdev, 1, ddsf);
1769         }
1770
1771  fail:
1772         reply->ret_code = retcode;
1773         return 0;
1774
1775  fail_ldev:
1776         put_ldev(mdev);
1777         goto fail;
1778 }
1779
1780 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1781                                struct drbd_nl_cfg_reply *reply)
1782 {
1783         int retcode = NO_ERROR;
1784         int err;
1785         int ovr; /* online verify running */
1786         int rsr; /* re-sync running */
1787         struct crypto_hash *verify_tfm = NULL;
1788         struct crypto_hash *csums_tfm = NULL;
1789         struct syncer_conf sc;
1790         cpumask_var_t new_cpu_mask;
1791         int *rs_plan_s = NULL;
1792         int fifo_size;
1793
1794         if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1795                 retcode = ERR_NOMEM;
1796                 goto fail;
1797         }
1798
1799         if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1800                 memset(&sc, 0, sizeof(struct syncer_conf));
1801                 sc.rate       = DRBD_RATE_DEF;
1802                 sc.after      = DRBD_AFTER_DEF;
1803                 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1804                 sc.on_no_data  = DRBD_ON_NO_DATA_DEF;
1805                 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1806                 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1807                 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1808                 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1809                 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1810         } else
1811                 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1812
1813         if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1814                 retcode = ERR_MANDATORY_TAG;
1815                 goto fail;
1816         }
1817
1818         /* re-sync running */
1819         rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1820                 mdev->state.conn == C_SYNC_TARGET ||
1821                 mdev->state.conn == C_PAUSED_SYNC_S ||
1822                 mdev->state.conn == C_PAUSED_SYNC_T );
1823
1824         if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1825                 retcode = ERR_CSUMS_RESYNC_RUNNING;
1826                 goto fail;
1827         }
1828
1829         if (!rsr && sc.csums_alg[0]) {
1830                 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1831                 if (IS_ERR(csums_tfm)) {
1832                         csums_tfm = NULL;
1833                         retcode = ERR_CSUMS_ALG;
1834                         goto fail;
1835                 }
1836
1837                 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1838                         retcode = ERR_CSUMS_ALG_ND;
1839                         goto fail;
1840                 }
1841         }
1842
1843         /* online verify running */
1844         ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1845
1846         if (ovr) {
1847                 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1848                         retcode = ERR_VERIFY_RUNNING;
1849                         goto fail;
1850                 }
1851         }
1852
1853         if (!ovr && sc.verify_alg[0]) {
1854                 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1855                 if (IS_ERR(verify_tfm)) {
1856                         verify_tfm = NULL;
1857                         retcode = ERR_VERIFY_ALG;
1858                         goto fail;
1859                 }
1860
1861                 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1862                         retcode = ERR_VERIFY_ALG_ND;
1863                         goto fail;
1864                 }
1865         }
1866
1867         /* silently ignore cpu mask on UP kernel */
1868         if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1869                 err = bitmap_parse(sc.cpu_mask, 32,
1870                                 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1871                 if (err) {
1872                         dev_warn(DEV, "bitmap_parse() failed with %d\n", err);
1873                         retcode = ERR_CPU_MASK_PARSE;
1874                         goto fail;
1875                 }
1876         }
1877
1878         ERR_IF (sc.rate < 1) sc.rate = 1;
1879         ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1880 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1881         if (sc.al_extents > AL_MAX) {
1882                 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1883                 sc.al_extents = AL_MAX;
1884         }
1885 #undef AL_MAX
1886
1887         /* to avoid spurious errors when configuring minors before configuring
1888          * the minors they depend on: if necessary, first create the minor we
1889          * depend on */
1890         if (sc.after >= 0)
1891                 ensure_mdev(sc.after, 1);
1892
1893         /* most sanity checks done, try to assign the new sync-after
1894          * dependency.  need to hold the global lock in there,
1895          * to avoid a race in the dependency loop check. */
1896         retcode = drbd_alter_sa(mdev, sc.after);
1897         if (retcode != NO_ERROR)
1898                 goto fail;
1899
1900         fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1901         if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1902                 rs_plan_s   = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1903                 if (!rs_plan_s) {
1904                         dev_err(DEV, "kmalloc of fifo_buffer failed");
1905                         retcode = ERR_NOMEM;
1906                         goto fail;
1907                 }
1908         }
1909
1910         /* ok, assign the rest of it as well.
1911          * lock against receive_SyncParam() */
1912         spin_lock(&mdev->peer_seq_lock);
1913         mdev->sync_conf = sc;
1914
1915         if (!rsr) {
1916                 crypto_free_hash(mdev->csums_tfm);
1917                 mdev->csums_tfm = csums_tfm;
1918                 csums_tfm = NULL;
1919         }
1920
1921         if (!ovr) {
1922                 crypto_free_hash(mdev->verify_tfm);
1923                 mdev->verify_tfm = verify_tfm;
1924                 verify_tfm = NULL;
1925         }
1926
1927         if (fifo_size != mdev->rs_plan_s.size) {
1928                 kfree(mdev->rs_plan_s.values);
1929                 mdev->rs_plan_s.values = rs_plan_s;
1930                 mdev->rs_plan_s.size   = fifo_size;
1931                 mdev->rs_planed = 0;
1932                 rs_plan_s = NULL;
1933         }
1934
1935         spin_unlock(&mdev->peer_seq_lock);
1936
1937         if (get_ldev(mdev)) {
1938                 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1939                 drbd_al_shrink(mdev);
1940                 err = drbd_check_al_size(mdev);
1941                 lc_unlock(mdev->act_log);
1942                 wake_up(&mdev->al_wait);
1943
1944                 put_ldev(mdev);
1945                 drbd_md_sync(mdev);
1946
1947                 if (err) {
1948                         retcode = ERR_NOMEM;
1949                         goto fail;
1950                 }
1951         }
1952
1953         if (mdev->state.conn >= C_CONNECTED)
1954                 drbd_send_sync_param(mdev, &sc);
1955
1956         if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1957                 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1958                 drbd_calc_cpu_mask(mdev);
1959                 mdev->receiver.reset_cpu_mask = 1;
1960                 mdev->asender.reset_cpu_mask = 1;
1961                 mdev->worker.reset_cpu_mask = 1;
1962         }
1963
1964         kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1965 fail:
1966         kfree(rs_plan_s);
1967         free_cpumask_var(new_cpu_mask);
1968         crypto_free_hash(csums_tfm);
1969         crypto_free_hash(verify_tfm);
1970         reply->ret_code = retcode;
1971         return 0;
1972 }
1973
1974 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1975                               struct drbd_nl_cfg_reply *reply)
1976 {
1977         int retcode;
1978
1979         /* If there is still bitmap IO pending, probably because of a previous
1980          * resync just being finished, wait for it before requesting a new resync.
1981          * Also wait for it's after_state_ch(). */
1982         drbd_suspend_io(mdev);
1983         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1984         drbd_flush_workqueue(mdev);
1985
1986         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1987
1988         if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1989                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1990
1991         while (retcode == SS_NEED_CONNECTION) {
1992                 spin_lock_irq(&mdev->req_lock);
1993                 if (mdev->state.conn < C_CONNECTED)
1994                         retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1995                 spin_unlock_irq(&mdev->req_lock);
1996
1997                 if (retcode != SS_NEED_CONNECTION)
1998                         break;
1999
2000                 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2001         }
2002         drbd_resume_io(mdev);
2003
2004         reply->ret_code = retcode;
2005         return 0;
2006 }
2007
2008 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2009 {
2010         int rv;
2011
2012         rv = drbd_bmio_set_n_write(mdev);
2013         drbd_suspend_al(mdev);
2014         return rv;
2015 }
2016
2017 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2018                                    struct drbd_nl_cfg_reply *reply)
2019 {
2020         int retcode;
2021
2022         /* If there is still bitmap IO pending, probably because of a previous
2023          * resync just being finished, wait for it before requesting a new resync.
2024          * Also wait for it's after_state_ch(). */
2025         drbd_suspend_io(mdev);
2026         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2027         drbd_flush_workqueue(mdev);
2028
2029         retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
2030
2031         if (retcode < SS_SUCCESS) {
2032                 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
2033                         /* The peer will get a resync upon connect anyways. Just make that
2034                            into a full resync. */
2035                         retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
2036                         if (retcode >= SS_SUCCESS) {
2037                                 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
2038                                         "set_n_write from invalidate_peer",
2039                                         BM_LOCKED_SET_ALLOWED))
2040                                         retcode = ERR_IO_MD_DISK;
2041                         }
2042                 } else
2043                         retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
2044         }
2045         drbd_resume_io(mdev);
2046
2047         reply->ret_code = retcode;
2048         return 0;
2049 }
2050
2051 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2052                               struct drbd_nl_cfg_reply *reply)
2053 {
2054         int retcode = NO_ERROR;
2055
2056         if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2057                 retcode = ERR_PAUSE_IS_SET;
2058
2059         reply->ret_code = retcode;
2060         return 0;
2061 }
2062
2063 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2064                                struct drbd_nl_cfg_reply *reply)
2065 {
2066         int retcode = NO_ERROR;
2067         union drbd_state s;
2068
2069         if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2070                 s = mdev->state;
2071                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2072                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2073                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2074                 } else {
2075                         retcode = ERR_PAUSE_IS_CLEAR;
2076                 }
2077         }
2078
2079         reply->ret_code = retcode;
2080         return 0;
2081 }
2082
2083 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2084                               struct drbd_nl_cfg_reply *reply)
2085 {
2086         reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
2087
2088         return 0;
2089 }
2090
2091 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2092                              struct drbd_nl_cfg_reply *reply)
2093 {
2094         if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2095                 drbd_uuid_new_current(mdev);
2096                 clear_bit(NEW_CUR_UUID, &mdev->flags);
2097         }
2098         drbd_suspend_io(mdev);
2099         reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2100         if (reply->ret_code == SS_SUCCESS) {
2101                 if (mdev->state.conn < C_CONNECTED)
2102                         tl_clear(mdev);
2103                 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2104                         tl_restart(mdev, fail_frozen_disk_io);
2105         }
2106         drbd_resume_io(mdev);
2107
2108         return 0;
2109 }
2110
2111 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2112                            struct drbd_nl_cfg_reply *reply)
2113 {
2114         reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
2115         return 0;
2116 }
2117
2118 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2119                            struct drbd_nl_cfg_reply *reply)
2120 {
2121         unsigned short *tl;
2122
2123         tl = reply->tag_list;
2124
2125         if (get_ldev(mdev)) {
2126                 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
2127                 put_ldev(mdev);
2128         }
2129
2130         if (get_net_conf(mdev)) {
2131                 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
2132                 put_net_conf(mdev);
2133         }
2134         tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
2135
2136         put_unaligned(TT_END, tl++); /* Close the tag list */
2137
2138         return (int)((char *)tl - (char *)reply->tag_list);
2139 }
2140
2141 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2142                              struct drbd_nl_cfg_reply *reply)
2143 {
2144         unsigned short *tl = reply->tag_list;
2145         union drbd_state s = mdev->state;
2146         unsigned long rs_left;
2147         unsigned int res;
2148
2149         tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
2150
2151         /* no local ref, no bitmap, no syncer progress. */
2152         if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
2153                 if (get_ldev(mdev)) {
2154                         drbd_get_syncer_progress(mdev, &rs_left, &res);
2155                         tl = tl_add_int(tl, T_sync_progress, &res);
2156                         put_ldev(mdev);
2157                 }
2158         }
2159         put_unaligned(TT_END, tl++); /* Close the tag list */
2160
2161         return (int)((char *)tl - (char *)reply->tag_list);
2162 }
2163
2164 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2165                              struct drbd_nl_cfg_reply *reply)
2166 {
2167         unsigned short *tl;
2168
2169         tl = reply->tag_list;
2170
2171         if (get_ldev(mdev)) {
2172                 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
2173                 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
2174                 put_ldev(mdev);
2175         }
2176         put_unaligned(TT_END, tl++); /* Close the tag list */
2177
2178         return (int)((char *)tl - (char *)reply->tag_list);
2179 }
2180
2181 /**
2182  * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
2183  * @mdev:       DRBD device.
2184  * @nlp:        Netlink/connector packet from drbdsetup
2185  * @reply:      Reply packet for drbdsetup
2186  */
2187 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2188                                     struct drbd_nl_cfg_reply *reply)
2189 {
2190         unsigned short *tl;
2191         char rv;
2192
2193         tl = reply->tag_list;
2194
2195         rv = mdev->state.pdsk == D_OUTDATED        ? UT_PEER_OUTDATED :
2196           test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
2197
2198         tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
2199         put_unaligned(TT_END, tl++); /* Close the tag list */
2200
2201         return (int)((char *)tl - (char *)reply->tag_list);
2202 }
2203
2204 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2205                                     struct drbd_nl_cfg_reply *reply)
2206 {
2207         /* default to resume from last known position, if possible */
2208         struct start_ov args =
2209                 { .start_sector = mdev->ov_start_sector };
2210
2211         if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2212                 reply->ret_code = ERR_MANDATORY_TAG;
2213                 return 0;
2214         }
2215
2216         /* If there is still bitmap IO pending, e.g. previous resync or verify
2217          * just being finished, wait for it before requesting a new resync. */
2218         drbd_suspend_io(mdev);
2219         wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2220
2221         /* w_make_ov_request expects position to be aligned */
2222         mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
2223         reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2224         drbd_resume_io(mdev);
2225         return 0;
2226 }
2227
2228
2229 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2230                               struct drbd_nl_cfg_reply *reply)
2231 {
2232         int retcode = NO_ERROR;
2233         int skip_initial_sync = 0;
2234         int err;
2235
2236         struct new_c_uuid args;
2237
2238         memset(&args, 0, sizeof(struct new_c_uuid));
2239         if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2240                 reply->ret_code = ERR_MANDATORY_TAG;
2241                 return 0;
2242         }
2243
2244         mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
2245
2246         if (!get_ldev(mdev)) {
2247                 retcode = ERR_NO_DISK;
2248                 goto out;
2249         }
2250
2251         /* this is "skip initial sync", assume to be clean */
2252         if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
2253             mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2254                 dev_info(DEV, "Preparing to skip initial sync\n");
2255                 skip_initial_sync = 1;
2256         } else if (mdev->state.conn != C_STANDALONE) {
2257                 retcode = ERR_CONNECTED;
2258                 goto out_dec;
2259         }
2260
2261         drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2262         drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2263
2264         if (args.clear_bm) {
2265                 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2266                         "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2267                 if (err) {
2268                         dev_err(DEV, "Writing bitmap failed with %d\n",err);
2269                         retcode = ERR_IO_MD_DISK;
2270                 }
2271                 if (skip_initial_sync) {
2272                         drbd_send_uuids_skip_initial_sync(mdev);
2273                         _drbd_uuid_set(mdev, UI_BITMAP, 0);
2274                         drbd_print_uuids(mdev, "cleared bitmap UUID");
2275                         spin_lock_irq(&mdev->req_lock);
2276                         _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2277                                         CS_VERBOSE, NULL);
2278                         spin_unlock_irq(&mdev->req_lock);
2279                 }
2280         }
2281
2282         drbd_md_sync(mdev);
2283 out_dec:
2284         put_ldev(mdev);
2285 out:
2286         mutex_unlock(&mdev->state_mutex);
2287
2288         reply->ret_code = retcode;
2289         return 0;
2290 }
2291
2292 struct cn_handler_struct {
2293         int (*function)(struct drbd_conf *,
2294                          struct drbd_nl_cfg_req *,
2295                          struct drbd_nl_cfg_reply *);
2296         int reply_body_size;
2297 };
2298
2299 static struct cn_handler_struct cnd_table[] = {
2300         [ P_primary ]           = { &drbd_nl_primary,           0 },
2301         [ P_secondary ]         = { &drbd_nl_secondary,         0 },
2302         [ P_disk_conf ]         = { &drbd_nl_disk_conf,         0 },
2303         [ P_detach ]            = { &drbd_nl_detach,            0 },
2304         [ P_net_conf ]          = { &drbd_nl_net_conf,          0 },
2305         [ P_disconnect ]        = { &drbd_nl_disconnect,        0 },
2306         [ P_resize ]            = { &drbd_nl_resize,            0 },
2307         [ P_syncer_conf ]       = { &drbd_nl_syncer_conf,       0 },
2308         [ P_invalidate ]        = { &drbd_nl_invalidate,        0 },
2309         [ P_invalidate_peer ]   = { &drbd_nl_invalidate_peer,   0 },
2310         [ P_pause_sync ]        = { &drbd_nl_pause_sync,        0 },
2311         [ P_resume_sync ]       = { &drbd_nl_resume_sync,       0 },
2312         [ P_suspend_io ]        = { &drbd_nl_suspend_io,        0 },
2313         [ P_resume_io ]         = { &drbd_nl_resume_io,         0 },
2314         [ P_outdate ]           = { &drbd_nl_outdate,           0 },
2315         [ P_get_config ]        = { &drbd_nl_get_config,
2316                                     sizeof(struct syncer_conf_tag_len_struct) +
2317                                     sizeof(struct disk_conf_tag_len_struct) +
2318                                     sizeof(struct net_conf_tag_len_struct) },
2319         [ P_get_state ]         = { &drbd_nl_get_state,
2320                                     sizeof(struct get_state_tag_len_struct) +
2321                                     sizeof(struct sync_progress_tag_len_struct) },
2322         [ P_get_uuids ]         = { &drbd_nl_get_uuids,
2323                                     sizeof(struct get_uuids_tag_len_struct) },
2324         [ P_get_timeout_flag ]  = { &drbd_nl_get_timeout_flag,
2325                                     sizeof(struct get_timeout_flag_tag_len_struct)},
2326         [ P_start_ov ]          = { &drbd_nl_start_ov,          0 },
2327         [ P_new_c_uuid ]        = { &drbd_nl_new_c_uuid,        0 },
2328 };
2329
2330 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2331 {
2332         struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2333         struct cn_handler_struct *cm;
2334         struct cn_msg *cn_reply;
2335         struct drbd_nl_cfg_reply *reply;
2336         struct drbd_conf *mdev;
2337         int retcode, rr;
2338         int reply_size = sizeof(struct cn_msg)
2339                 + sizeof(struct drbd_nl_cfg_reply)
2340                 + sizeof(short int);
2341
2342         if (!try_module_get(THIS_MODULE)) {
2343                 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2344                 return;
2345         }
2346
2347         if (!capable(CAP_SYS_ADMIN)) {
2348                 retcode = ERR_PERM;
2349                 goto fail;
2350         }
2351
2352         mdev = ensure_mdev(nlp->drbd_minor,
2353                         (nlp->flags & DRBD_NL_CREATE_DEVICE));
2354         if (!mdev) {
2355                 retcode = ERR_MINOR_INVALID;
2356                 goto fail;
2357         }
2358
2359         if (nlp->packet_type >= P_nl_after_last_packet ||
2360             nlp->packet_type == P_return_code_only) {
2361                 retcode = ERR_PACKET_NR;
2362                 goto fail;
2363         }
2364
2365         cm = cnd_table + nlp->packet_type;
2366
2367         /* This may happen if packet number is 0: */
2368         if (cm->function == NULL) {
2369                 retcode = ERR_PACKET_NR;
2370                 goto fail;
2371         }
2372
2373         reply_size += cm->reply_body_size;
2374
2375         /* allocation not in the IO path, cqueue thread context */
2376         cn_reply = kzalloc(reply_size, GFP_KERNEL);
2377         if (!cn_reply) {
2378                 retcode = ERR_NOMEM;
2379                 goto fail;
2380         }
2381         reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2382
2383         reply->packet_type =
2384                 cm->reply_body_size ? nlp->packet_type : P_return_code_only;
2385         reply->minor = nlp->drbd_minor;
2386         reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2387         /* reply->tag_list; might be modified by cm->function. */
2388
2389         rr = cm->function(mdev, nlp, reply);
2390
2391         cn_reply->id = req->id;
2392         cn_reply->seq = req->seq;
2393         cn_reply->ack = req->ack  + 1;
2394         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2395         cn_reply->flags = 0;
2396
2397         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2398         if (rr && rr != -ESRCH)
2399                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2400
2401         kfree(cn_reply);
2402         module_put(THIS_MODULE);
2403         return;
2404  fail:
2405         drbd_nl_send_reply(req, retcode);
2406         module_put(THIS_MODULE);
2407 }
2408
2409 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2410
2411 static unsigned short *
2412 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2413         unsigned short len, int nul_terminated)
2414 {
2415         unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2416         len = (len < l) ? len :  l;
2417         put_unaligned(tag, tl++);
2418         put_unaligned(len, tl++);
2419         memcpy(tl, data, len);
2420         tl = (unsigned short*)((char*)tl + len);
2421         if (nul_terminated)
2422                 *((char*)tl - 1) = 0;
2423         return tl;
2424 }
2425
2426 static unsigned short *
2427 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2428 {
2429         return __tl_add_blob(tl, tag, data, len, 0);
2430 }
2431
2432 static unsigned short *
2433 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2434 {
2435         return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2436 }
2437
2438 static unsigned short *
2439 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2440 {
2441         put_unaligned(tag, tl++);
2442         switch(tag_type(tag)) {
2443         case TT_INTEGER:
2444                 put_unaligned(sizeof(int), tl++);
2445                 put_unaligned(*(int *)val, (int *)tl);
2446                 tl = (unsigned short*)((char*)tl+sizeof(int));
2447                 break;
2448         case TT_INT64:
2449                 put_unaligned(sizeof(u64), tl++);
2450                 put_unaligned(*(u64 *)val, (u64 *)tl);
2451                 tl = (unsigned short*)((char*)tl+sizeof(u64));
2452                 break;
2453         default:
2454                 /* someone did something stupid. */
2455                 ;
2456         }
2457         return tl;
2458 }
2459
2460 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2461 {
2462         char buffer[sizeof(struct cn_msg)+
2463                     sizeof(struct drbd_nl_cfg_reply)+
2464                     sizeof(struct get_state_tag_len_struct)+
2465                     sizeof(short int)];
2466         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2467         struct drbd_nl_cfg_reply *reply =
2468                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2469         unsigned short *tl = reply->tag_list;
2470
2471         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2472
2473         tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2474
2475         put_unaligned(TT_END, tl++); /* Close the tag list */
2476
2477         cn_reply->id.idx = CN_IDX_DRBD;
2478         cn_reply->id.val = CN_VAL_DRBD;
2479
2480         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2481         cn_reply->ack = 0; /* not used here. */
2482         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2483                 (int)((char *)tl - (char *)reply->tag_list);
2484         cn_reply->flags = 0;
2485
2486         reply->packet_type = P_get_state;
2487         reply->minor = mdev_to_minor(mdev);
2488         reply->ret_code = NO_ERROR;
2489
2490         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2491 }
2492
2493 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2494 {
2495         char buffer[sizeof(struct cn_msg)+
2496                     sizeof(struct drbd_nl_cfg_reply)+
2497                     sizeof(struct call_helper_tag_len_struct)+
2498                     sizeof(short int)];
2499         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2500         struct drbd_nl_cfg_reply *reply =
2501                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2502         unsigned short *tl = reply->tag_list;
2503
2504         /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2505
2506         tl = tl_add_str(tl, T_helper, helper_name);
2507         put_unaligned(TT_END, tl++); /* Close the tag list */
2508
2509         cn_reply->id.idx = CN_IDX_DRBD;
2510         cn_reply->id.val = CN_VAL_DRBD;
2511
2512         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2513         cn_reply->ack = 0; /* not used here. */
2514         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2515                 (int)((char *)tl - (char *)reply->tag_list);
2516         cn_reply->flags = 0;
2517
2518         reply->packet_type = P_call_helper;
2519         reply->minor = mdev_to_minor(mdev);
2520         reply->ret_code = NO_ERROR;
2521
2522         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2523 }
2524
2525 void drbd_bcast_ee(struct drbd_conf *mdev,
2526                 const char *reason, const int dgs,
2527                 const char* seen_hash, const char* calc_hash,
2528                 const struct drbd_epoch_entry* e)
2529 {
2530         struct cn_msg *cn_reply;
2531         struct drbd_nl_cfg_reply *reply;
2532         unsigned short *tl;
2533         struct page *page;
2534         unsigned len;
2535
2536         if (!e)
2537                 return;
2538         if (!reason || !reason[0])
2539                 return;
2540
2541         /* apparently we have to memcpy twice, first to prepare the data for the
2542          * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2543          * netlink skb. */
2544         /* receiver thread context, which is not in the writeout path (of this node),
2545          * but may be in the writeout path of the _other_ node.
2546          * GFP_NOIO to avoid potential "distributed deadlock". */
2547         cn_reply = kzalloc(
2548                 sizeof(struct cn_msg)+
2549                 sizeof(struct drbd_nl_cfg_reply)+
2550                 sizeof(struct dump_ee_tag_len_struct)+
2551                 sizeof(short int),
2552                 GFP_NOIO);
2553
2554         if (!cn_reply) {
2555                 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2556                                 (unsigned long long)e->sector, e->size);
2557                 return;
2558         }
2559
2560         reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2561         tl = reply->tag_list;
2562
2563         tl = tl_add_str(tl, T_dump_ee_reason, reason);
2564         tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2565         tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2566         tl = tl_add_int(tl, T_ee_sector, &e->sector);
2567         tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2568
2569         /* dump the first 32k */
2570         len = min_t(unsigned, e->size, 32 << 10);
2571         put_unaligned(T_ee_data, tl++);
2572         put_unaligned(len, tl++);
2573
2574         page = e->pages;
2575         page_chain_for_each(page) {
2576                 void *d = kmap_atomic(page);
2577                 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2578                 memcpy(tl, d, l);
2579                 kunmap_atomic(d);
2580                 tl = (unsigned short*)((char*)tl + l);
2581                 len -= l;
2582                 if (len == 0)
2583                         break;
2584         }
2585         put_unaligned(TT_END, tl++); /* Close the tag list */
2586
2587         cn_reply->id.idx = CN_IDX_DRBD;
2588         cn_reply->id.val = CN_VAL_DRBD;
2589
2590         cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2591         cn_reply->ack = 0; // not used here.
2592         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2593                 (int)((char*)tl - (char*)reply->tag_list);
2594         cn_reply->flags = 0;
2595
2596         reply->packet_type = P_dump_ee;
2597         reply->minor = mdev_to_minor(mdev);
2598         reply->ret_code = NO_ERROR;
2599
2600         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2601         kfree(cn_reply);
2602 }
2603
2604 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2605 {
2606         char buffer[sizeof(struct cn_msg)+
2607                     sizeof(struct drbd_nl_cfg_reply)+
2608                     sizeof(struct sync_progress_tag_len_struct)+
2609                     sizeof(short int)];
2610         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2611         struct drbd_nl_cfg_reply *reply =
2612                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2613         unsigned short *tl = reply->tag_list;
2614         unsigned long rs_left;
2615         unsigned int res;
2616
2617         /* no local ref, no bitmap, no syncer progress, no broadcast. */
2618         if (!get_ldev(mdev))
2619                 return;
2620         drbd_get_syncer_progress(mdev, &rs_left, &res);
2621         put_ldev(mdev);
2622
2623         tl = tl_add_int(tl, T_sync_progress, &res);
2624         put_unaligned(TT_END, tl++); /* Close the tag list */
2625
2626         cn_reply->id.idx = CN_IDX_DRBD;
2627         cn_reply->id.val = CN_VAL_DRBD;
2628
2629         cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2630         cn_reply->ack = 0; /* not used here. */
2631         cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2632                 (int)((char *)tl - (char *)reply->tag_list);
2633         cn_reply->flags = 0;
2634
2635         reply->packet_type = P_sync_progress;
2636         reply->minor = mdev_to_minor(mdev);
2637         reply->ret_code = NO_ERROR;
2638
2639         cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2640 }
2641
2642 int __init drbd_nl_init(void)
2643 {
2644         static struct cb_id cn_id_drbd;
2645         int err, try=10;
2646
2647         cn_id_drbd.val = CN_VAL_DRBD;
2648         do {
2649                 cn_id_drbd.idx = cn_idx;
2650                 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2651                 if (!err)
2652                         break;
2653                 cn_idx = (cn_idx + CN_IDX_STEP);
2654         } while (try--);
2655
2656         if (err) {
2657                 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2658                 return err;
2659         }
2660
2661         return 0;
2662 }
2663
2664 void drbd_nl_cleanup(void)
2665 {
2666         static struct cb_id cn_id_drbd;
2667
2668         cn_id_drbd.idx = cn_idx;
2669         cn_id_drbd.val = CN_VAL_DRBD;
2670
2671         cn_del_callback(&cn_id_drbd);
2672 }
2673
2674 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2675 {
2676         char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2677         struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2678         struct drbd_nl_cfg_reply *reply =
2679                 (struct drbd_nl_cfg_reply *)cn_reply->data;
2680         int rr;
2681
2682         memset(buffer, 0, sizeof(buffer));
2683         cn_reply->id = req->id;
2684
2685         cn_reply->seq = req->seq;
2686         cn_reply->ack = req->ack  + 1;
2687         cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2688         cn_reply->flags = 0;
2689
2690         reply->packet_type = P_return_code_only;
2691         reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2692         reply->ret_code = ret_code;
2693
2694         rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2695         if (rr && rr != -ESRCH)
2696                 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2697 }
2698